[v2,10/11] staging: typec: fusb302: Hook up mux support using tcpc_gen_mux support

Message ID 20170905164221.11266-11-hdegoede@redhat.com
State Under Review
Headers show
Series
  • Untitled series #1609
Related show

Commit Message

Hans de Goede Sept. 5, 2017, 4:42 p.m.
Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.

Also document the mux-names used by the generic tcpc_mux_dev code in
our devicetree bindings.

Cc: Rob Herring <robh+dt@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: devicetree@vger.kernel.org
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
---
 Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
 drivers/staging/typec/fusb302/fusb302.c               | 11 ++++++++++-
 2 files changed, 13 insertions(+), 1 deletion(-)

Comments

Rob Herring Sept. 12, 2017, 10:20 p.m. | #1
On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
> 
> Also document the mux-names used by the generic tcpc_mux_dev code in
> our devicetree bindings.
> 
> Cc: Rob Herring <robh+dt@kernel.org>
> Cc: Mark Rutland <mark.rutland@arm.com>
> Cc: devicetree@vger.kernel.org
> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
> ---
>  Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>  drivers/staging/typec/fusb302/fusb302.c               | 11 ++++++++++-
>  2 files changed, 13 insertions(+), 1 deletion(-)
> 
> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> index 472facfa5a71..63d639eadacd 100644
> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> @@ -6,6 +6,9 @@ Required properties :
>  - interrupts             : Interrupt specifier
>  
>  Optional properties :
> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or 2 muxes
> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
> +                           "type-c-mode-mux", "usb-role-mux" when using 2 muxes

I'm not sure this is the right place for this. The mux is outside the 
FUSB302. In a type-C connector node or USB phy node would make more 
sense to me.

>  - fcs,max-sink-microvolt : Maximum voltage to negotiate when configured as sink
>  - fcs,max-sink-microamp  : Maximum current to negotiate when configured as sink
>  - fcs,max-sink-microwatt : Maximum power to negotiate when configured as sink
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Hans de Goede Sept. 13, 2017, 8:56 a.m. | #2
Hi,

On 13-09-17 00:20, Rob Herring wrote:
> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>
>> Also document the mux-names used by the generic tcpc_mux_dev code in
>> our devicetree bindings.
>>
>> Cc: Rob Herring <robh+dt@kernel.org>
>> Cc: Mark Rutland <mark.rutland@arm.com>
>> Cc: devicetree@vger.kernel.org
>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>> ---
>>   Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>   drivers/staging/typec/fusb302/fusb302.c               | 11 ++++++++++-
>>   2 files changed, 13 insertions(+), 1 deletion(-)
>>
>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>> index 472facfa5a71..63d639eadacd 100644
>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>> @@ -6,6 +6,9 @@ Required properties :
>>   - interrupts             : Interrupt specifier
>>   
>>   Optional properties :
>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or 2 muxes
>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>> +                           "type-c-mode-mux", "usb-role-mux" when using 2 muxes
> 
> I'm not sure this is the right place for this. The mux is outside the
> FUSB302. In a type-C connector node or USB phy node would make more
> sense to me.

The mux certainly does not belong in the USB phy node, it sits between the USB PHY
and the Type-C connector and can for example also mux the Type-C pins to a Display
Port PHY.

As for putting it in a type-C connector node, currently we do not have such a node,
the closest thing we do have to a node describing it is actually the fusb302 node
itself. E.g. it may also contain a regulator to turn Vbus on / off (already there
in the code, but I forgot to document this when I added the missing DT bindings
doc for the fusb302 with a previous patch).

Also these properties:

>>   - fcs,max-sink-microvolt : Maximum voltage to negotiate when configured as sink
>>   - fcs,max-sink-microamp  : Maximum current to negotiate when configured as sink
>>   - fcs,max-sink-microwatt : Maximum power to negotiate when configured as sink

Have more to do with the charger-IC used (which determines the limits) then with
the fusb302 itself, but the fusb302 needs to know these as it negotiates the limits.

Likewise the fusb302 negotiates how the data pins will be used and thus to which pins
on the SoC the mux should mux the data pins.

TL;DR: The fusb302 does all the negotiation and ties all the Type-C connected
ICs together, so this seems like the right place for it (it certainly is the
natural place to put these from a driver code pov).

Regards,

Hans





--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Rob Herring Sept. 13, 2017, 1:38 p.m. | #3
On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com> wrote:
> Hi,
>
>
> On 13-09-17 00:20, Rob Herring wrote:
>>
>> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>>>
>>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>>
>>> Also document the mux-names used by the generic tcpc_mux_dev code in
>>> our devicetree bindings.
>>>
>>> Cc: Rob Herring <robh+dt@kernel.org>
>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>> Cc: devicetree@vger.kernel.org
>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>> ---
>>>   Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>>   drivers/staging/typec/fusb302/fusb302.c               | 11 ++++++++++-
>>>   2 files changed, 13 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>> b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>> index 472facfa5a71..63d639eadacd 100644
>>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>> @@ -6,6 +6,9 @@ Required properties :
>>>   - interrupts             : Interrupt specifier
>>>     Optional properties :
>>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or 2
>>> muxes
>>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>>> +                           "type-c-mode-mux", "usb-role-mux" when using
>>> 2 muxes
>>
>>
>> I'm not sure this is the right place for this. The mux is outside the
>> FUSB302. In a type-C connector node or USB phy node would make more
>> sense to me.
>
>
> The mux certainly does not belong in the USB phy node, it sits between the
> USB PHY
> and the Type-C connector and can for example also mux the Type-C pins to a
> Display
> Port PHY.

Thinking about this some more, the mux(es) should be its own node(s).
Then the question is where to put the nodes.

> As for putting it in a type-C connector node, currently we do not have such
> a node,

Well, you should. Type-C connectors are certainly complicated enough
that we'll need one. Plus we already require connector nodes for
display outputs, so what do we do once you add display muxing?

> the closest thing we do have to a node describing it is actually the fusb302
> node
> itself. E.g. it may also contain a regulator to turn Vbus on / off (already
> there
> in the code, but I forgot to document this when I added the missing DT
> bindings
> doc for the fusb302 with a previous patch).

Either you can have a vbus-supply property in connector node or it can
be implied that the controller chip provides that. For example, HDMI
connectors have a hpd-gpios property if HPD is connected to GPIO or
they have nothing and it's implicit that the HDMI encoder handles HPD.


> Also these properties:
>
>>>   - fcs,max-sink-microvolt : Maximum voltage to negotiate when configured
>>> as sink
>>>   - fcs,max-sink-microamp  : Maximum current to negotiate when configured
>>> as sink
>>>   - fcs,max-sink-microwatt : Maximum power to negotiate when configured
>>> as sink
>
>
> Have more to do with the charger-IC used (which determines the limits) then
> with
> the fusb302 itself, but the fusb302 needs to know these as it negotiates the
> limits.

Those should probably be elsewhere and not be fusb302 specific. I did
ack that, but it was a single node for a single component which is
fine. But once we start adding more external pieces we need to pay
more attention to the overall structure.

> Likewise the fusb302 negotiates how the data pins will be used and thus to
> which pins
> on the SoC the mux should mux the data pins.
>
> TL;DR: The fusb302 does all the negotiation and ties all the Type-C
> connected
> ICs together, so this seems like the right place for it (it certainly is the
> natural place to put these from a driver code pov).

Things in DT should follow what the h/w design looks like which is not
necessarily aligned with the driver structure. If the USB PD chip
needs information from the charger, then we need a kernel interface
for that.

My concern here is not so much this binding in particular, but rather
that we handle Type-C connectors in a common way and not adhoc with
each platform doing things their own way. Otherwise, we end up with a
mess of platform specific bindings like charger/battery bindings
(though there's some work improving those).

Rob
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Hans de Goede Sept. 13, 2017, 2:06 p.m. | #4
Hi,

On 13-09-17 15:38, Rob Herring wrote:
> On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com> wrote:
>> Hi,
>>
>>
>> On 13-09-17 00:20, Rob Herring wrote:
>>>
>>> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>>>>
>>>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>>>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>>>
>>>> Also document the mux-names used by the generic tcpc_mux_dev code in
>>>> our devicetree bindings.
>>>>
>>>> Cc: Rob Herring <robh+dt@kernel.org>
>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>> Cc: devicetree@vger.kernel.org
>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>> ---
>>>>    Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>>>    drivers/staging/typec/fusb302/fusb302.c               | 11 ++++++++++-
>>>>    2 files changed, 13 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>> b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>> index 472facfa5a71..63d639eadacd 100644
>>>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>> @@ -6,6 +6,9 @@ Required properties :
>>>>    - interrupts             : Interrupt specifier
>>>>      Optional properties :
>>>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or 2
>>>> muxes
>>>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>>>> +                           "type-c-mode-mux", "usb-role-mux" when using
>>>> 2 muxes
>>>
>>>
>>> I'm not sure this is the right place for this. The mux is outside the
>>> FUSB302. In a type-C connector node or USB phy node would make more
>>> sense to me.
>>
>>
>> The mux certainly does not belong in the USB phy node, it sits between the
>> USB PHY
>> and the Type-C connector and can for example also mux the Type-C pins to a
>> Display
>> Port PHY.
> 
> Thinking about this some more, the mux(es) should be its own node(s).
> Then the question is where to put the nodes.

Right, the mux will be its own node per Documentation/devicetree/bindings/mux/mux-controller.txt
the bindings bit this patch is adding is only adding phandles pointing
to that mux-node as the fusb302 "consumes" the mux functionality.

So as such (the fusb302 is the component which should control the mux)
I do believe that the bindings this patch adds are correct.

>> As for putting it in a type-C connector node, currently we do not have such
>> a node,
> 
> Well, you should. Type-C connectors are certainly complicated enough
> that we'll need one. Plus we already require connector nodes for
> display outputs, so what do we do once you add display muxing?

An interesting question, I'm working on this for x86 + ACPI boards actually,
not a board using DT I've been adding DT bindings docs for device-properties
I use because that seems like the right thing to do where the binding is obvious
(which I believe it is in this case as the fusb302 is the mux consumer) and
because the device-property code should work the same on x86 + ACPI
(where some platform-specific drivers attach the device properties) and
on e.g. ARM + DT.

The rest should probably be left to be figured out when an actual DT
using device using the fusb302 or another Type-C controller shows up.

>> the closest thing we do have to a node describing it is actually the fusb302
>> node
>> itself. E.g. it may also contain a regulator to turn Vbus on / off (already
>> there
>> in the code, but I forgot to document this when I added the missing DT
>> bindings
>> doc for the fusb302 with a previous patch).
> 
> Either you can have a vbus-supply property in connector node or it can
> be implied that the controller chip provides that. For example, HDMI
> connectors have a hpd-gpios property if HPD is connected to GPIO or
> they have nothing and it's implicit that the HDMI encoder handles HPD.
> 
> 
>> Also these properties:
>>
>>>>    - fcs,max-sink-microvolt : Maximum voltage to negotiate when configured
>>>> as sink
>>>>    - fcs,max-sink-microamp  : Maximum current to negotiate when configured
>>>> as sink
>>>>    - fcs,max-sink-microwatt : Maximum power to negotiate when configured
>>>> as sink
>>
>>
>> Have more to do with the charger-IC used (which determines the limits) then
>> with
>> the fusb302 itself, but the fusb302 needs to know these as it negotiates the
>> limits.
> 
> Those should probably be elsewhere and not be fusb302 specific. I did
> ack that, but it was a single node for a single component which is
> fine. But once we start adding more external pieces we need to pay
> more attention to the overall structure.
> 
>> Likewise the fusb302 negotiates how the data pins will be used and thus to
>> which pins
>> on the SoC the mux should mux the data pins.
>>
>> TL;DR: The fusb302 does all the negotiation and ties all the Type-C
>> connected
>> ICs together, so this seems like the right place for it (it certainly is the
>> natural place to put these from a driver code pov).
> 
> Things in DT should follow what the h/w design looks like which is not
> necessarily aligned with the driver structure. If the USB PD chip
> needs information from the charger, then we need a kernel interface
> for that.

Well this really is board specific data, the charger IC may be handle
for example up to 17V, but if the board is only designed for 12V then
we should only negotiate up to 12V because the board may have voltage
overprotection circuitry which trips at say 14V. This is actually the
case with the 2 x86 boards with a Type-C connector and fusb302 Type-C
controller I have, the charger on there can handle upto 17V according
to its datasheet but Windows never negotiates more then 12V, even when
attaching a Type-C charger which can do 5V, 9V or 14V, Windows choses 9V.

So it is the Type-C controller which does the negotiating and
the limits may be stricter then the maximum ratings of the
charger IC, so I guess my earlier remark of this coming from the
charger IC was not accurate and having this info in the Type-C controller
node is the right thing to do, sorry about that.

> My concern here is not so much this binding in particular, but rather
> that we handle Type-C connectors in a common way and not adhoc with
> each platform doing things their own way. Otherwise, we end up with a
> mess of platform specific bindings like charger/battery bindings
> (though there's some work improving those).

I understand, but see my remark about me working on this on
X86 / ACPI boards. One advantage of this, is that the device-properties
are being set by platform drivers living under drivers/platform/x86,
so if the need arises they can be changed without breaking any ABI as
in my use-case they are 100% kernel internal stuff.

Regards,

Hans
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Rob Herring Sept. 13, 2017, 3:07 p.m. | #5
On Wed, Sep 13, 2017 at 9:06 AM, Hans de Goede <hdegoede@redhat.com> wrote:
> Hi,
>
>
> On 13-09-17 15:38, Rob Herring wrote:
>>
>> On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com>
>> wrote:
>>>
>>> Hi,
>>>
>>>
>>> On 13-09-17 00:20, Rob Herring wrote:
>>>>
>>>>
>>>> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>>>>>
>>>>>
>>>>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>>>>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>>>>
>>>>> Also document the mux-names used by the generic tcpc_mux_dev code in
>>>>> our devicetree bindings.
>>>>>
>>>>> Cc: Rob Herring <robh+dt@kernel.org>
>>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>>> Cc: devicetree@vger.kernel.org
>>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>>> ---
>>>>>    Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>>>>    drivers/staging/typec/fusb302/fusb302.c               | 11
>>>>> ++++++++++-
>>>>>    2 files changed, 13 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>> b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>> index 472facfa5a71..63d639eadacd 100644
>>>>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>> @@ -6,6 +6,9 @@ Required properties :
>>>>>    - interrupts             : Interrupt specifier
>>>>>      Optional properties :
>>>>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or
>>>>> 2
>>>>> muxes
>>>>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>>>>> +                           "type-c-mode-mux", "usb-role-mux" when
>>>>> using
>>>>> 2 muxes
>>>>
>>>>
>>>>
>>>> I'm not sure this is the right place for this. The mux is outside the
>>>> FUSB302. In a type-C connector node or USB phy node would make more
>>>> sense to me.
>>>
>>>
>>>
>>> The mux certainly does not belong in the USB phy node, it sits between
>>> the
>>> USB PHY
>>> and the Type-C connector and can for example also mux the Type-C pins to
>>> a
>>> Display
>>> Port PHY.
>>
>>
>> Thinking about this some more, the mux(es) should be its own node(s).
>> Then the question is where to put the nodes.
>
>
> Right, the mux will be its own node per
> Documentation/devicetree/bindings/mux/mux-controller.txt
> the bindings bit this patch is adding is only adding phandles pointing
> to that mux-node as the fusb302 "consumes" the mux functionality.
>
> So as such (the fusb302 is the component which should control the mux)
> I do believe that the bindings this patch adds are correct.

Humm, that's not how the mux binding works. The mux controller is what
drives the mux select lines and is the provider. The consumer is the
mux device itself. What decides the mux state is determined by what
you are muxing, not which node has mux-controls property.

By putting mux-controls in fusb302 node, you are saying fusb302 is a
mux (or contains a mux).


>>> As for putting it in a type-C connector node, currently we do not have
>>> such
>>> a node,
>>
>>
>> Well, you should. Type-C connectors are certainly complicated enough
>> that we'll need one. Plus we already require connector nodes for
>> display outputs, so what do we do once you add display muxing?
>
>
> An interesting question, I'm working on this for x86 + ACPI boards actually,
> not a board using DT I've been adding DT bindings docs for device-properties
> I use because that seems like the right thing to do where the binding is
> obvious
> (which I believe it is in this case as the fusb302 is the mux consumer) and
> because the device-property code should work the same on x86 + ACPI
> (where some platform-specific drivers attach the device properties) and
> on e.g. ARM + DT.
>
> The rest should probably be left to be figured out when an actual DT
> using device using the fusb302 or another Type-C controller shows up.

Well this is a new one (maybe, I suppose others have sneaked by). If
ACPI folks want to use DT bindings, then what do I care. But I have no
interest in reviewing ACPI properties. The whole notion of sharing
bindings between DT and ACPI beyond anything trivial is flawed IMO.
The ptifalls have been discussed multiple times before, so I'm not
going to repeat them here.


>>> the closest thing we do have to a node describing it is actually the
>>> fusb302
>>> node
>>> itself. E.g. it may also contain a regulator to turn Vbus on / off
>>> (already
>>> there
>>> in the code, but I forgot to document this when I added the missing DT
>>> bindings
>>> doc for the fusb302 with a previous patch).
>>
>>
>> Either you can have a vbus-supply property in connector node or it can
>> be implied that the controller chip provides that. For example, HDMI
>> connectors have a hpd-gpios property if HPD is connected to GPIO or
>> they have nothing and it's implicit that the HDMI encoder handles HPD.
>>
>>
>>> Also these properties:
>>>
>>>>>    - fcs,max-sink-microvolt : Maximum voltage to negotiate when
>>>>> configured
>>>>> as sink
>>>>>    - fcs,max-sink-microamp  : Maximum current to negotiate when
>>>>> configured
>>>>> as sink
>>>>>    - fcs,max-sink-microwatt : Maximum power to negotiate when
>>>>> configured
>>>>> as sink
>>>
>>>
>>>
>>> Have more to do with the charger-IC used (which determines the limits)
>>> then
>>> with
>>> the fusb302 itself, but the fusb302 needs to know these as it negotiates
>>> the
>>> limits.
>>
>>
>> Those should probably be elsewhere and not be fusb302 specific. I did
>> ack that, but it was a single node for a single component which is
>> fine. But once we start adding more external pieces we need to pay
>> more attention to the overall structure.
>>
>>> Likewise the fusb302 negotiates how the data pins will be used and thus
>>> to
>>> which pins
>>> on the SoC the mux should mux the data pins.
>>>
>>> TL;DR: The fusb302 does all the negotiation and ties all the Type-C
>>> connected
>>> ICs together, so this seems like the right place for it (it certainly is
>>> the
>>> natural place to put these from a driver code pov).
>>
>>
>> Things in DT should follow what the h/w design looks like which is not
>> necessarily aligned with the driver structure. If the USB PD chip
>> needs information from the charger, then we need a kernel interface
>> for that.
>
>
> Well this really is board specific data, the charger IC may be handle
> for example up to 17V, but if the board is only designed for 12V then
> we should only negotiate up to 12V because the board may have voltage
> overprotection circuitry which trips at say 14V. This is actually the
> case with the 2 x86 boards with a Type-C connector and fusb302 Type-C
> controller I have, the charger on there can handle upto 17V according
> to its datasheet but Windows never negotiates more then 12V, even when
> attaching a Type-C charger which can do 5V, 9V or 14V, Windows choses 9V.

Just as the fusb302 can have board specific properties, so can a charger IC.

> So it is the Type-C controller which does the negotiating and
> the limits may be stricter then the maximum ratings of the
> charger IC, so I guess my earlier remark of this coming from the
> charger IC was not accurate and having this info in the Type-C controller
> node is the right thing to do, sorry about that.
>
>> My concern here is not so much this binding in particular, but rather
>> that we handle Type-C connectors in a common way and not adhoc with
>> each platform doing things their own way. Otherwise, we end up with a
>> mess of platform specific bindings like charger/battery bindings
>> (though there's some work improving those).
>
>
> I understand, but see my remark about me working on this on
> X86 / ACPI boards. One advantage of this, is that the device-properties
> are being set by platform drivers living under drivers/platform/x86,
> so if the need arises they can be changed without breaking any ABI as
> in my use-case they are 100% kernel internal stuff.

Then don't document this stuff as DT bindings when it is not really.

Rob
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Hans de Goede Sept. 13, 2017, 3:48 p.m. | #6
Hi,

On 13-09-17 17:07, Rob Herring wrote:
> On Wed, Sep 13, 2017 at 9:06 AM, Hans de Goede <hdegoede@redhat.com> wrote:
>> Hi,
>>
>>
>> On 13-09-17 15:38, Rob Herring wrote:
>>>
>>> On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com>
>>> wrote:
>>>>
>>>> Hi,
>>>>
>>>>
>>>> On 13-09-17 00:20, Rob Herring wrote:
>>>>>
>>>>>
>>>>> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>>>>>>
>>>>>>
>>>>>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>>>>>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>>>>>
>>>>>> Also document the mux-names used by the generic tcpc_mux_dev code in
>>>>>> our devicetree bindings.
>>>>>>
>>>>>> Cc: Rob Herring <robh+dt@kernel.org>
>>>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>>>> Cc: devicetree@vger.kernel.org
>>>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>>>> ---
>>>>>>     Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>>>>>     drivers/staging/typec/fusb302/fusb302.c               | 11
>>>>>> ++++++++++-
>>>>>>     2 files changed, 13 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>> b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>> index 472facfa5a71..63d639eadacd 100644
>>>>>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>> @@ -6,6 +6,9 @@ Required properties :
>>>>>>     - interrupts             : Interrupt specifier
>>>>>>       Optional properties :
>>>>>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or
>>>>>> 2
>>>>>> muxes
>>>>>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>>>>>> +                           "type-c-mode-mux", "usb-role-mux" when
>>>>>> using
>>>>>> 2 muxes
>>>>>
>>>>>
>>>>>
>>>>> I'm not sure this is the right place for this. The mux is outside the
>>>>> FUSB302. In a type-C connector node or USB phy node would make more
>>>>> sense to me.
>>>>
>>>>
>>>>
>>>> The mux certainly does not belong in the USB phy node, it sits between
>>>> the
>>>> USB PHY
>>>> and the Type-C connector and can for example also mux the Type-C pins to
>>>> a
>>>> Display
>>>> Port PHY.
>>>
>>>
>>> Thinking about this some more, the mux(es) should be its own node(s).
>>> Then the question is where to put the nodes.
>>
>>
>> Right, the mux will be its own node per
>> Documentation/devicetree/bindings/mux/mux-controller.txt
>> the bindings bit this patch is adding is only adding phandles pointing
>> to that mux-node as the fusb302 "consumes" the mux functionality.
>>
>> So as such (the fusb302 is the component which should control the mux)
>> I do believe that the bindings this patch adds are correct.
> 
> Humm, that's not how the mux binding works. The mux controller is what
> drives the mux select lines and is the provider. The consumer is the
> mux device itself. What decides the mux state is determined by what
> you are muxing, not which node has mux-controls property.
> 
> By putting mux-controls in fusb302 node, you are saying fusb302 is a
> mux (or contains a mux).
> 
> 
>>>> As for putting it in a type-C connector node, currently we do not have
>>>> such
>>>> a node,
>>>
>>>
>>> Well, you should. Type-C connectors are certainly complicated enough
>>> that we'll need one. Plus we already require connector nodes for
>>> display outputs, so what do we do once you add display muxing?
>>
>>
>> An interesting question, I'm working on this for x86 + ACPI boards actually,
>> not a board using DT I've been adding DT bindings docs for device-properties
>> I use because that seems like the right thing to do where the binding is
>> obvious
>> (which I believe it is in this case as the fusb302 is the mux consumer) and
>> because the device-property code should work the same on x86 + ACPI
>> (where some platform-specific drivers attach the device properties) and
>> on e.g. ARM + DT.
>>
>> The rest should probably be left to be figured out when an actual DT
>> using device using the fusb302 or another Type-C controller shows up.
> 
> Well this is a new one (maybe, I suppose others have sneaked by). If
> ACPI folks want to use DT bindings, then what do I care. But I have no
> interest in reviewing ACPI properties. The whole notion of sharing
> bindings between DT and ACPI beyond anything trivial is flawed IMO.
> The ptifalls have been discussed multiple times before, so I'm not
> going to repeat them here.

Ok, so shall I just drop the Documentation/devicetree/bindings/usb/fcs,fusb302.txt
part of this patch then ?

Regards,

Hans
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Guenter Roeck Sept. 13, 2017, 4:17 p.m. | #7
On Wed, Sep 13, 2017 at 05:48:25PM +0200, Hans de Goede wrote:
> Hi,
> 
> On 13-09-17 17:07, Rob Herring wrote:
> >On Wed, Sep 13, 2017 at 9:06 AM, Hans de Goede <hdegoede@redhat.com> wrote:
> >>Hi,
> >>
> >>
> >>On 13-09-17 15:38, Rob Herring wrote:
> >>>
> >>>On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com>
> >>>wrote:
> >>>>
> >>>>Hi,
> >>>>
> >>>>
> >>>>On 13-09-17 00:20, Rob Herring wrote:
> >>>>>
> >>>>>
> >>>>>On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
> >>>>>>
> >>>>>>
> >>>>>>Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
> >>>>>>to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
> >>>>>>
> >>>>>>Also document the mux-names used by the generic tcpc_mux_dev code in
> >>>>>>our devicetree bindings.
> >>>>>>
> >>>>>>Cc: Rob Herring <robh+dt@kernel.org>
> >>>>>>Cc: Mark Rutland <mark.rutland@arm.com>
> >>>>>>Cc: devicetree@vger.kernel.org
> >>>>>>Signed-off-by: Hans de Goede <hdegoede@redhat.com>
> >>>>>>---
> >>>>>>    Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
> >>>>>>    drivers/staging/typec/fusb302/fusb302.c               | 11
> >>>>>>++++++++++-
> >>>>>>    2 files changed, 13 insertions(+), 1 deletion(-)
> >>>>>>
> >>>>>>diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> >>>>>>b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> >>>>>>index 472facfa5a71..63d639eadacd 100644
> >>>>>>--- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> >>>>>>+++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> >>>>>>@@ -6,6 +6,9 @@ Required properties :
> >>>>>>    - interrupts             : Interrupt specifier
> >>>>>>      Optional properties :
> >>>>>>+- mux-controls           : List of mux-ctrl-specifiers containing 1 or
> >>>>>>2
> >>>>>>muxes
> >>>>>>+- mux-names              : "type-c-mode-mux" when using 1 mux, or
> >>>>>>+                           "type-c-mode-mux", "usb-role-mux" when
> >>>>>>using
> >>>>>>2 muxes
> >>>>>
> >>>>>
> >>>>>
> >>>>>I'm not sure this is the right place for this. The mux is outside the
> >>>>>FUSB302. In a type-C connector node or USB phy node would make more
> >>>>>sense to me.
> >>>>
> >>>>
> >>>>
> >>>>The mux certainly does not belong in the USB phy node, it sits between
> >>>>the
> >>>>USB PHY
> >>>>and the Type-C connector and can for example also mux the Type-C pins to
> >>>>a
> >>>>Display
> >>>>Port PHY.
> >>>
> >>>
> >>>Thinking about this some more, the mux(es) should be its own node(s).
> >>>Then the question is where to put the nodes.
> >>
> >>
> >>Right, the mux will be its own node per
> >>Documentation/devicetree/bindings/mux/mux-controller.txt
> >>the bindings bit this patch is adding is only adding phandles pointing
> >>to that mux-node as the fusb302 "consumes" the mux functionality.
> >>
> >>So as such (the fusb302 is the component which should control the mux)
> >>I do believe that the bindings this patch adds are correct.
> >
> >Humm, that's not how the mux binding works. The mux controller is what
> >drives the mux select lines and is the provider. The consumer is the
> >mux device itself. What decides the mux state is determined by what
> >you are muxing, not which node has mux-controls property.
> >
> >By putting mux-controls in fusb302 node, you are saying fusb302 is a
> >mux (or contains a mux).
> >
> >
> >>>>As for putting it in a type-C connector node, currently we do not have
> >>>>such
> >>>>a node,
> >>>
> >>>
> >>>Well, you should. Type-C connectors are certainly complicated enough
> >>>that we'll need one. Plus we already require connector nodes for
> >>>display outputs, so what do we do once you add display muxing?
> >>
> >>
> >>An interesting question, I'm working on this for x86 + ACPI boards actually,
> >>not a board using DT I've been adding DT bindings docs for device-properties
> >>I use because that seems like the right thing to do where the binding is
> >>obvious
> >>(which I believe it is in this case as the fusb302 is the mux consumer) and
> >>because the device-property code should work the same on x86 + ACPI
> >>(where some platform-specific drivers attach the device properties) and
> >>on e.g. ARM + DT.
> >>
> >>The rest should probably be left to be figured out when an actual DT
> >>using device using the fusb302 or another Type-C controller shows up.
> >
> >Well this is a new one (maybe, I suppose others have sneaked by). If
> >ACPI folks want to use DT bindings, then what do I care. But I have no
> >interest in reviewing ACPI properties. The whole notion of sharing
> >bindings between DT and ACPI beyond anything trivial is flawed IMO.
> >The ptifalls have been discussed multiple times before, so I'm not
> >going to repeat them here.
> 
> Ok, so shall I just drop the Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> part of this patch then ?
> 
FWIW, Android (where the fusb302 driver is coming from) does use dt.
On the other side I assume they won't jump on the new mux handling
immediately. I won't have time to look into it myself, so whatever
is done here may not match the "real" dt use case. Given that, maybe
it does make sense to drop the bindings part and revisit once it
becomes relevant.

Guenter
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin Sept. 25, 2017, 10:34 a.m. | #8
On 2017-09-13 17:48, Hans de Goede wrote:
> Hi,
> 
> On 13-09-17 17:07, Rob Herring wrote:
>> On Wed, Sep 13, 2017 at 9:06 AM, Hans de Goede <hdegoede@redhat.com> wrote:
>>> Hi,
>>>
>>>
>>> On 13-09-17 15:38, Rob Herring wrote:
>>>>
>>>> On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com>
>>>> wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>>
>>>>> On 13-09-17 00:20, Rob Herring wrote:
>>>>>>
>>>>>>
>>>>>> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>>>>>>>
>>>>>>>
>>>>>>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>>>>>>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>>>>>>
>>>>>>> Also document the mux-names used by the generic tcpc_mux_dev code in
>>>>>>> our devicetree bindings.
>>>>>>>
>>>>>>> Cc: Rob Herring <robh+dt@kernel.org>
>>>>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>>>>> Cc: devicetree@vger.kernel.org
>>>>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>>>>> ---
>>>>>>>     Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>>>>>>     drivers/staging/typec/fusb302/fusb302.c               | 11
>>>>>>> ++++++++++-
>>>>>>>     2 files changed, 13 insertions(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>> b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>> index 472facfa5a71..63d639eadacd 100644
>>>>>>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>> @@ -6,6 +6,9 @@ Required properties :
>>>>>>>     - interrupts             : Interrupt specifier
>>>>>>>       Optional properties :
>>>>>>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or
>>>>>>> 2
>>>>>>> muxes
>>>>>>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>>>>>>> +                           "type-c-mode-mux", "usb-role-mux" when
>>>>>>> using
>>>>>>> 2 muxes
>>>>>>
>>>>>>
>>>>>>
>>>>>> I'm not sure this is the right place for this. The mux is outside the
>>>>>> FUSB302. In a type-C connector node or USB phy node would make more
>>>>>> sense to me.
>>>>>
>>>>>
>>>>>
>>>>> The mux certainly does not belong in the USB phy node, it sits between
>>>>> the
>>>>> USB PHY
>>>>> and the Type-C connector and can for example also mux the Type-C pins to
>>>>> a
>>>>> Display
>>>>> Port PHY.
>>>>
>>>>
>>>> Thinking about this some more, the mux(es) should be its own node(s).
>>>> Then the question is where to put the nodes.
>>>
>>>
>>> Right, the mux will be its own node per
>>> Documentation/devicetree/bindings/mux/mux-controller.txt
>>> the bindings bit this patch is adding is only adding phandles pointing
>>> to that mux-node as the fusb302 "consumes" the mux functionality.
>>>
>>> So as such (the fusb302 is the component which should control the mux)
>>> I do believe that the bindings this patch adds are correct.
>>
>> Humm, that's not how the mux binding works. The mux controller is what
>> drives the mux select lines and is the provider. The consumer is the
>> mux device itself. What decides the mux state is determined by what
>> you are muxing, not which node has mux-controls property.
>>
>> By putting mux-controls in fusb302 node, you are saying fusb302 is a
>> mux (or contains a mux).
>>
>>
>>>>> As for putting it in a type-C connector node, currently we do not have
>>>>> such
>>>>> a node,
>>>>
>>>>
>>>> Well, you should. Type-C connectors are certainly complicated enough
>>>> that we'll need one. Plus we already require connector nodes for
>>>> display outputs, so what do we do once you add display muxing?
>>>
>>>
>>> An interesting question, I'm working on this for x86 + ACPI boards actually,
>>> not a board using DT I've been adding DT bindings docs for device-properties
>>> I use because that seems like the right thing to do where the binding is
>>> obvious
>>> (which I believe it is in this case as the fusb302 is the mux consumer) and
>>> because the device-property code should work the same on x86 + ACPI
>>> (where some platform-specific drivers attach the device properties) and
>>> on e.g. ARM + DT.
>>>
>>> The rest should probably be left to be figured out when an actual DT
>>> using device using the fusb302 or another Type-C controller shows up.
>>
>> Well this is a new one (maybe, I suppose others have sneaked by). If
>> ACPI folks want to use DT bindings, then what do I care. But I have no
>> interest in reviewing ACPI properties. The whole notion of sharing
>> bindings between DT and ACPI beyond anything trivial is flawed IMO.
>> The ptifalls have been discussed multiple times before, so I'm not
>> going to repeat them here.
> 
> Ok, so shall I just drop the Documentation/devicetree/bindings/usb/fcs,fusb302.txt
> part of this patch then ?

I totally agree with the concern that Rob expressed about handling USB C
muxes in a non-adhoc way. And this makes this series scary. I don't know
enough details about USB C muxes and PD (I just have a very superficial
mental model) to tell if this series is going down the right path. Or
how terrible it will be to fix things up if not?

The series extends the mux subsystem with muxes that pins semantics
to specific states. That is new, and shows up exactly here when Rob is
not happy about the binding. And if Rob does not want this in the
DT bindings then I'm not so sure it is wise to do it at all? This
problem doesn't go away just because you remove the binding. I think
I would feel much better if there was a path forward on how to
represent USB C muxes in DT and how that would fit with the driver
structure.

If you compare to the i2c-muxes, there is a relatively new i2c-mux-gpmux
driver that uses some general purpose mux from the mux subsystem to
implement an i2c-mux. If USB C muxes where to be done similarly, I'd
imagine there should be a general abstraction of what USB C muxes provide
somewhere outside of the mux subsystem, and a bunch of implementations
of that abstraction. One of those implementations could be to use "raw"
muxes from the mux subsystem. Of course, this is not what this series is
doing.

Also, muxes that are not general purpose such as the ones added to the
mux subsystem by this series could perhaps be repurposed for some other
application, but since the interface implemented does not really obey
the rules (the provided mux controller interacts with different sets of
signals depending on the state) this will not be possible.

These issues are what has caused me to do a lot of thinking and to sit
silent, sorry about that, but I would like input from someone more
experienced. If possible. But I'm not sure where to turn?

As a crazy example, why is it not possible to hook up one signal pair
from the USB C mux, not to DP, but instead to some I2C controller? Then,
if done right, i2c-mux-gpmux could be hooked up with the relevant mux
controller and use the signal pair for I2C, with the mux controller
acting as a gate. So, maybe a bit crazy, but something like that is how
I think it should work from the mux subsystem point of view. And while
maybe crazy and while it might not be technically possible to do I2C
over a USB C connector for some reason, I do think that whatever
abstraction you come up with for USB C muxes, it has to deal with and
broker requests from both the USB subsystem and whatever other
subsystems deals with the alt pairs. Be it graphics for DP signals, or
whatever. IIUC, the alt signals need not be graphics, and it would be
sad to implement the USB C mux it in a way that makes it hard to use
the alt pairs for something else.

[maybe my understanding of USB C is just wrong]

Cheers,
Peter
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Hans de Goede Sept. 25, 2017, 11:35 a.m. | #9
Hi,

On 25-09-17 12:34, Peter Rosin wrote:
> On 2017-09-13 17:48, Hans de Goede wrote:
>> Hi,
>>
>> On 13-09-17 17:07, Rob Herring wrote:
>>> On Wed, Sep 13, 2017 at 9:06 AM, Hans de Goede <hdegoede@redhat.com> wrote:
>>>> Hi,
>>>>
>>>>
>>>> On 13-09-17 15:38, Rob Herring wrote:
>>>>>
>>>>> On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com>
>>>>> wrote:
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>>
>>>>>> On 13-09-17 00:20, Rob Herring wrote:
>>>>>>>
>>>>>>>
>>>>>>> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>>>>>>>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>>>>>>>
>>>>>>>> Also document the mux-names used by the generic tcpc_mux_dev code in
>>>>>>>> our devicetree bindings.
>>>>>>>>
>>>>>>>> Cc: Rob Herring <robh+dt@kernel.org>
>>>>>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>>>>>> Cc: devicetree@vger.kernel.org
>>>>>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>>>>>> ---
>>>>>>>>      Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>>>>>>>      drivers/staging/typec/fusb302/fusb302.c               | 11
>>>>>>>> ++++++++++-
>>>>>>>>      2 files changed, 13 insertions(+), 1 deletion(-)
>>>>>>>>
>>>>>>>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>> b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>> index 472facfa5a71..63d639eadacd 100644
>>>>>>>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>> @@ -6,6 +6,9 @@ Required properties :
>>>>>>>>      - interrupts             : Interrupt specifier
>>>>>>>>        Optional properties :
>>>>>>>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or
>>>>>>>> 2
>>>>>>>> muxes
>>>>>>>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>>>>>>>> +                           "type-c-mode-mux", "usb-role-mux" when
>>>>>>>> using
>>>>>>>> 2 muxes
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> I'm not sure this is the right place for this. The mux is outside the
>>>>>>> FUSB302. In a type-C connector node or USB phy node would make more
>>>>>>> sense to me.
>>>>>>
>>>>>>
>>>>>>
>>>>>> The mux certainly does not belong in the USB phy node, it sits between
>>>>>> the
>>>>>> USB PHY
>>>>>> and the Type-C connector and can for example also mux the Type-C pins to
>>>>>> a
>>>>>> Display
>>>>>> Port PHY.
>>>>>
>>>>>
>>>>> Thinking about this some more, the mux(es) should be its own node(s).
>>>>> Then the question is where to put the nodes.
>>>>
>>>>
>>>> Right, the mux will be its own node per
>>>> Documentation/devicetree/bindings/mux/mux-controller.txt
>>>> the bindings bit this patch is adding is only adding phandles pointing
>>>> to that mux-node as the fusb302 "consumes" the mux functionality.
>>>>
>>>> So as such (the fusb302 is the component which should control the mux)
>>>> I do believe that the bindings this patch adds are correct.
>>>
>>> Humm, that's not how the mux binding works. The mux controller is what
>>> drives the mux select lines and is the provider. The consumer is the
>>> mux device itself. What decides the mux state is determined by what
>>> you are muxing, not which node has mux-controls property.
>>>
>>> By putting mux-controls in fusb302 node, you are saying fusb302 is a
>>> mux (or contains a mux).
>>>
>>>
>>>>>> As for putting it in a type-C connector node, currently we do not have
>>>>>> such
>>>>>> a node,
>>>>>
>>>>>
>>>>> Well, you should. Type-C connectors are certainly complicated enough
>>>>> that we'll need one. Plus we already require connector nodes for
>>>>> display outputs, so what do we do once you add display muxing?
>>>>
>>>>
>>>> An interesting question, I'm working on this for x86 + ACPI boards actually,
>>>> not a board using DT I've been adding DT bindings docs for device-properties
>>>> I use because that seems like the right thing to do where the binding is
>>>> obvious
>>>> (which I believe it is in this case as the fusb302 is the mux consumer) and
>>>> because the device-property code should work the same on x86 + ACPI
>>>> (where some platform-specific drivers attach the device properties) and
>>>> on e.g. ARM + DT.
>>>>
>>>> The rest should probably be left to be figured out when an actual DT
>>>> using device using the fusb302 or another Type-C controller shows up.
>>>
>>> Well this is a new one (maybe, I suppose others have sneaked by). If
>>> ACPI folks want to use DT bindings, then what do I care. But I have no
>>> interest in reviewing ACPI properties. The whole notion of sharing
>>> bindings between DT and ACPI beyond anything trivial is flawed IMO.
>>> The ptifalls have been discussed multiple times before, so I'm not
>>> going to repeat them here.
>>
>> Ok, so shall I just drop the Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>> part of this patch then ?
> 
> I totally agree with the concern that Rob expressed about handling USB C
> muxes in a non-adhoc way. And this makes this series scary. I don't know
> enough details about USB C muxes and PD (I just have a very superficial
> mental model) to tell if this series is going down the right path. Or
> how terrible it will be to fix things up if not?
> 
> The series extends the mux subsystem with muxes that pins semantics
> to specific states. That is new, and shows up exactly here when Rob is
> not happy about the binding. And if Rob does not want this in the
> DT bindings then I'm not so sure it is wise to do it at all? This
> problem doesn't go away just because you remove the binding. I think
> I would feel much better if there was a path forward on how to
> represent USB C muxes in DT and how that would fit with the driver
> structure.
> 
> If you compare to the i2c-muxes, there is a relatively new i2c-mux-gpmux
> driver that uses some general purpose mux from the mux subsystem to
> implement an i2c-mux. If USB C muxes where to be done similarly, I'd
> imagine there should be a general abstraction of what USB C muxes provide
> somewhere outside of the mux subsystem, and a bunch of implementations
> of that abstraction. One of those implementations could be to use "raw"
> muxes from the mux subsystem. Of course, this is not what this series is
> doing.
> 
> Also, muxes that are not general purpose such as the ones added to the
> mux subsystem by this series could perhaps be repurposed for some other
> application, but since the interface implemented does not really obey
> the rules (the provided mux controller interacts with different sets of
> signals depending on the state) this will not be possible.
> 
> These issues are what has caused me to do a lot of thinking and to sit
> silent, sorry about that, but I would like input from someone more
> experienced. If possible. But I'm not sure where to turn?
> 
> As a crazy example, why is it not possible to hook up one signal pair
> from the USB C mux, not to DP, but instead to some I2C controller? Then,
> if done right, i2c-mux-gpmux could be hooked up with the relevant mux
> controller and use the signal pair for I2C, with the mux controller
> acting as a gate. So, maybe a bit crazy, but something like that is how
> I think it should work from the mux subsystem point of view. And while
> maybe crazy and while it might not be technically possible to do I2C
> over a USB C connector for some reason, I do think that whatever
> abstraction you come up with for USB C muxes, it has to deal with and
> broker requests from both the USB subsystem and whatever other
> subsystems deals with the alt pairs. Be it graphics for DP signals, or
> whatever. IIUC, the alt signals need not be graphics, and it would be
> sad to implement the USB C mux it in a way that makes it hard to use
> the alt pairs for something else.
> 
> [maybe my understanding of USB C is just wrong]

So 2 things:

1) The Type-C subsys does actually abstract the mux outside of the
Type-C port-manager (TCPM) core in the form of a tcpc_mux_dev, so for
DT based platforms we could instantiate a tcpc_mux_dev from a node
which then represents the mux as usual.

2) What is very different from how the mux subsys currently is used,
is who is in control of the mux. Currently a subsys which wants to
route data through the mux selects the mux in the right mode, and
then deselects it when done. This assumes that the mux can mux
the data paths to the requested destination at all times, just not
to multiple destinations at once. With Type-C and any moment in
time, there really is only one correct setting as the mux is
connect to a Type-C device on the other side which typically
only has one config. So what happens with Type-C is that the TCPM
negotiates with the externally connected Type-C device, and then
sets the mux to the one and only correct setting for that device
to be fully functional.

So your i2c example, for example, will not work with the normal
way where the i2c subsys asks the mux to mux a data-pair to the i2c
controller, as that only makes sense when your theoretical Type-C
device which talks i2c over a pair is connected externally, where
as when something else is connected then trying to talk i2c might
even damage the connected device. So with Type-C it is the TCPM
and only the TCPM which controls the mux.

Regards,

Hans
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin Sept. 25, 2017, 1:45 p.m. | #10
On 2017-09-25 13:35, Hans de Goede wrote:
> Hi,
> 
> On 25-09-17 12:34, Peter Rosin wrote:
>> On 2017-09-13 17:48, Hans de Goede wrote:
>>> Hi,
>>>
>>> On 13-09-17 17:07, Rob Herring wrote:
>>>> On Wed, Sep 13, 2017 at 9:06 AM, Hans de Goede <hdegoede@redhat.com> wrote:
>>>>> Hi,
>>>>>
>>>>>
>>>>> On 13-09-17 15:38, Rob Herring wrote:
>>>>>>
>>>>>> On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com>
>>>>>> wrote:
>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>>
>>>>>>> On 13-09-17 00:20, Rob Herring wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>>>>>>>>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>>>>>>>>
>>>>>>>>> Also document the mux-names used by the generic tcpc_mux_dev code in
>>>>>>>>> our devicetree bindings.
>>>>>>>>>
>>>>>>>>> Cc: Rob Herring <robh+dt@kernel.org>
>>>>>>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>>>>>>> Cc: devicetree@vger.kernel.org
>>>>>>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>>>>>>> ---
>>>>>>>>>      Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>>>>>>>>      drivers/staging/typec/fusb302/fusb302.c               | 11
>>>>>>>>> ++++++++++-
>>>>>>>>>      2 files changed, 13 insertions(+), 1 deletion(-)
>>>>>>>>>
>>>>>>>>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>>> b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>>> index 472facfa5a71..63d639eadacd 100644
>>>>>>>>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>>> @@ -6,6 +6,9 @@ Required properties :
>>>>>>>>>      - interrupts             : Interrupt specifier
>>>>>>>>>        Optional properties :
>>>>>>>>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or
>>>>>>>>> 2
>>>>>>>>> muxes
>>>>>>>>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>>>>>>>>> +                           "type-c-mode-mux", "usb-role-mux" when
>>>>>>>>> using
>>>>>>>>> 2 muxes
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> I'm not sure this is the right place for this. The mux is outside the
>>>>>>>> FUSB302. In a type-C connector node or USB phy node would make more
>>>>>>>> sense to me.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> The mux certainly does not belong in the USB phy node, it sits between
>>>>>>> the
>>>>>>> USB PHY
>>>>>>> and the Type-C connector and can for example also mux the Type-C pins to
>>>>>>> a
>>>>>>> Display
>>>>>>> Port PHY.
>>>>>>
>>>>>>
>>>>>> Thinking about this some more, the mux(es) should be its own node(s).
>>>>>> Then the question is where to put the nodes.
>>>>>
>>>>>
>>>>> Right, the mux will be its own node per
>>>>> Documentation/devicetree/bindings/mux/mux-controller.txt
>>>>> the bindings bit this patch is adding is only adding phandles pointing
>>>>> to that mux-node as the fusb302 "consumes" the mux functionality.
>>>>>
>>>>> So as such (the fusb302 is the component which should control the mux)
>>>>> I do believe that the bindings this patch adds are correct.
>>>>
>>>> Humm, that's not how the mux binding works. The mux controller is what
>>>> drives the mux select lines and is the provider. The consumer is the
>>>> mux device itself. What decides the mux state is determined by what
>>>> you are muxing, not which node has mux-controls property.
>>>>
>>>> By putting mux-controls in fusb302 node, you are saying fusb302 is a
>>>> mux (or contains a mux).
>>>>
>>>>
>>>>>>> As for putting it in a type-C connector node, currently we do not have
>>>>>>> such
>>>>>>> a node,
>>>>>>
>>>>>>
>>>>>> Well, you should. Type-C connectors are certainly complicated enough
>>>>>> that we'll need one. Plus we already require connector nodes for
>>>>>> display outputs, so what do we do once you add display muxing?
>>>>>
>>>>>
>>>>> An interesting question, I'm working on this for x86 + ACPI boards actually,
>>>>> not a board using DT I've been adding DT bindings docs for device-properties
>>>>> I use because that seems like the right thing to do where the binding is
>>>>> obvious
>>>>> (which I believe it is in this case as the fusb302 is the mux consumer) and
>>>>> because the device-property code should work the same on x86 + ACPI
>>>>> (where some platform-specific drivers attach the device properties) and
>>>>> on e.g. ARM + DT.
>>>>>
>>>>> The rest should probably be left to be figured out when an actual DT
>>>>> using device using the fusb302 or another Type-C controller shows up.
>>>>
>>>> Well this is a new one (maybe, I suppose others have sneaked by). If
>>>> ACPI folks want to use DT bindings, then what do I care. But I have no
>>>> interest in reviewing ACPI properties. The whole notion of sharing
>>>> bindings between DT and ACPI beyond anything trivial is flawed IMO.
>>>> The ptifalls have been discussed multiple times before, so I'm not
>>>> going to repeat them here.
>>>
>>> Ok, so shall I just drop the Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>> part of this patch then ?
>>
>> I totally agree with the concern that Rob expressed about handling USB C
>> muxes in a non-adhoc way. And this makes this series scary. I don't know
>> enough details about USB C muxes and PD (I just have a very superficial
>> mental model) to tell if this series is going down the right path. Or
>> how terrible it will be to fix things up if not?
>>
>> The series extends the mux subsystem with muxes that pins semantics
>> to specific states. That is new, and shows up exactly here when Rob is
>> not happy about the binding. And if Rob does not want this in the
>> DT bindings then I'm not so sure it is wise to do it at all? This
>> problem doesn't go away just because you remove the binding. I think
>> I would feel much better if there was a path forward on how to
>> represent USB C muxes in DT and how that would fit with the driver
>> structure.
>>
>> If you compare to the i2c-muxes, there is a relatively new i2c-mux-gpmux
>> driver that uses some general purpose mux from the mux subsystem to
>> implement an i2c-mux. If USB C muxes where to be done similarly, I'd
>> imagine there should be a general abstraction of what USB C muxes provide
>> somewhere outside of the mux subsystem, and a bunch of implementations
>> of that abstraction. One of those implementations could be to use "raw"
>> muxes from the mux subsystem. Of course, this is not what this series is
>> doing.
>>
>> Also, muxes that are not general purpose such as the ones added to the
>> mux subsystem by this series could perhaps be repurposed for some other
>> application, but since the interface implemented does not really obey
>> the rules (the provided mux controller interacts with different sets of
>> signals depending on the state) this will not be possible.
>>
>> These issues are what has caused me to do a lot of thinking and to sit
>> silent, sorry about that, but I would like input from someone more
>> experienced. If possible. But I'm not sure where to turn?
>>
>> As a crazy example, why is it not possible to hook up one signal pair
>> from the USB C mux, not to DP, but instead to some I2C controller? Then,
>> if done right, i2c-mux-gpmux could be hooked up with the relevant mux
>> controller and use the signal pair for I2C, with the mux controller
>> acting as a gate. So, maybe a bit crazy, but something like that is how
>> I think it should work from the mux subsystem point of view. And while
>> maybe crazy and while it might not be technically possible to do I2C
>> over a USB C connector for some reason, I do think that whatever
>> abstraction you come up with for USB C muxes, it has to deal with and
>> broker requests from both the USB subsystem and whatever other
>> subsystems deals with the alt pairs. Be it graphics for DP signals, or
>> whatever. IIUC, the alt signals need not be graphics, and it would be
>> sad to implement the USB C mux it in a way that makes it hard to use
>> the alt pairs for something else.
>>
>> [maybe my understanding of USB C is just wrong]
> 
> So 2 things:
> 
> 1) The Type-C subsys does actually abstract the mux outside of the
> Type-C port-manager (TCPM) core in the form of a tcpc_mux_dev, so for
> DT based platforms we could instantiate a tcpc_mux_dev from a node
> which then represents the mux as usual.
> 
> 2) What is very different from how the mux subsys currently is used,
> is who is in control of the mux. Currently a subsys which wants to
> route data through the mux selects the mux in the right mode, and
> then deselects it when done. This assumes that the mux can mux
> the data paths to the requested destination at all times, just not
> to multiple destinations at once. With Type-C and any moment in
> time, there really is only one correct setting as the mux is
> connect to a Type-C device on the other side which typically
> only has one config. So what happens with Type-C is that the TCPM
> negotiates with the externally connected Type-C device, and then
> sets the mux to the one and only correct setting for that device
> to be fully functional.
> 
> So your i2c example, for example, will not work with the normal
> way where the i2c subsys asks the mux to mux a data-pair to the i2c
> controller, as that only makes sense when your theoretical Type-C
> device which talks i2c over a pair is connected externally, where
> as when something else is connected then trying to talk i2c might
> even damage the connected device. So with Type-C it is the TCPM
> and only the TCPM which controls the mux.

If I get this correctly, some code (the TCPM?) negotiates with the
other side over PD, and after that the way to set up the USB C mux
is known and will not change until the cable is unplugged (unless
we go wild and renegotiate, but let's assume that will not happen). 
What I don't get is why do you want to squeeze what the TCPM(?) is
doing with its specialized mux into the existing mux subsystem
interface?

I mean, the mux interface isn't really a perfect fit with its
locking and support for multiple consumers etc that really just gets
in the way of setting a register in some chip to some value. Which
is all that is really needed when you know that the TCPM is the only
one accessing the chip.

And from the point of view of the mux subsystem, the USB C muxes
like the Pericom driver will be in a class of their own. Muxes of
that class can really only be used by one thing -- the TCPM code.

So, I don't see the benefit of doing this through the mux subsystem.
My hope is that the TCPM code is generic, and that by putting the
USB C mux inside the mux subsystem, you can keep the TCPM code
generic? That might be good for the TCPM code, but it does create a
new class of devices in the mux subsystem, and opens the door for
other classes later on.

Why not just add a function pointer that the generic TCPM code calls
if it is set, and then add some code somewhere more local to the TCPM
code to back that call, and completely avoid the mux subsystem?

Cheers,
Peter
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Hans de Goede Sept. 25, 2017, 2:17 p.m. | #11
Hi,

On 25-09-17 15:45, Peter Rosin wrote:
> On 2017-09-25 13:35, Hans de Goede wrote:
>> Hi,
>>
>> On 25-09-17 12:34, Peter Rosin wrote:
>>> On 2017-09-13 17:48, Hans de Goede wrote:
>>>> Hi,
>>>>
>>>> On 13-09-17 17:07, Rob Herring wrote:
>>>>> On Wed, Sep 13, 2017 at 9:06 AM, Hans de Goede <hdegoede@redhat.com> wrote:
>>>>>> Hi,
>>>>>>
>>>>>>
>>>>>> On 13-09-17 15:38, Rob Herring wrote:
>>>>>>>
>>>>>>> On Wed, Sep 13, 2017 at 3:56 AM, Hans de Goede <hdegoede@redhat.com>
>>>>>>> wrote:
>>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>>
>>>>>>>> On 13-09-17 00:20, Rob Herring wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On Tue, Sep 05, 2017 at 06:42:20PM +0200, Hans de Goede wrote:
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Add mux support to the fusb302 driver, call devm_tcpc_gen_mux_create()
>>>>>>>>>> to let the generic tcpc_mux_dev code create a tcpc_mux_dev for us.
>>>>>>>>>>
>>>>>>>>>> Also document the mux-names used by the generic tcpc_mux_dev code in
>>>>>>>>>> our devicetree bindings.
>>>>>>>>>>
>>>>>>>>>> Cc: Rob Herring <robh+dt@kernel.org>
>>>>>>>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>>>>>>>> Cc: devicetree@vger.kernel.org
>>>>>>>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>>>>>>>> ---
>>>>>>>>>>       Documentation/devicetree/bindings/usb/fcs,fusb302.txt |  3 +++
>>>>>>>>>>       drivers/staging/typec/fusb302/fusb302.c               | 11
>>>>>>>>>> ++++++++++-
>>>>>>>>>>       2 files changed, 13 insertions(+), 1 deletion(-)
>>>>>>>>>>
>>>>>>>>>> diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>>>> b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>>>> index 472facfa5a71..63d639eadacd 100644
>>>>>>>>>> --- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>>>> +++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>>>>>>>> @@ -6,6 +6,9 @@ Required properties :
>>>>>>>>>>       - interrupts             : Interrupt specifier
>>>>>>>>>>         Optional properties :
>>>>>>>>>> +- mux-controls           : List of mux-ctrl-specifiers containing 1 or
>>>>>>>>>> 2
>>>>>>>>>> muxes
>>>>>>>>>> +- mux-names              : "type-c-mode-mux" when using 1 mux, or
>>>>>>>>>> +                           "type-c-mode-mux", "usb-role-mux" when
>>>>>>>>>> using
>>>>>>>>>> 2 muxes
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> I'm not sure this is the right place for this. The mux is outside the
>>>>>>>>> FUSB302. In a type-C connector node or USB phy node would make more
>>>>>>>>> sense to me.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> The mux certainly does not belong in the USB phy node, it sits between
>>>>>>>> the
>>>>>>>> USB PHY
>>>>>>>> and the Type-C connector and can for example also mux the Type-C pins to
>>>>>>>> a
>>>>>>>> Display
>>>>>>>> Port PHY.
>>>>>>>
>>>>>>>
>>>>>>> Thinking about this some more, the mux(es) should be its own node(s).
>>>>>>> Then the question is where to put the nodes.
>>>>>>
>>>>>>
>>>>>> Right, the mux will be its own node per
>>>>>> Documentation/devicetree/bindings/mux/mux-controller.txt
>>>>>> the bindings bit this patch is adding is only adding phandles pointing
>>>>>> to that mux-node as the fusb302 "consumes" the mux functionality.
>>>>>>
>>>>>> So as such (the fusb302 is the component which should control the mux)
>>>>>> I do believe that the bindings this patch adds are correct.
>>>>>
>>>>> Humm, that's not how the mux binding works. The mux controller is what
>>>>> drives the mux select lines and is the provider. The consumer is the
>>>>> mux device itself. What decides the mux state is determined by what
>>>>> you are muxing, not which node has mux-controls property.
>>>>>
>>>>> By putting mux-controls in fusb302 node, you are saying fusb302 is a
>>>>> mux (or contains a mux).
>>>>>
>>>>>
>>>>>>>> As for putting it in a type-C connector node, currently we do not have
>>>>>>>> such
>>>>>>>> a node,
>>>>>>>
>>>>>>>
>>>>>>> Well, you should. Type-C connectors are certainly complicated enough
>>>>>>> that we'll need one. Plus we already require connector nodes for
>>>>>>> display outputs, so what do we do once you add display muxing?
>>>>>>
>>>>>>
>>>>>> An interesting question, I'm working on this for x86 + ACPI boards actually,
>>>>>> not a board using DT I've been adding DT bindings docs for device-properties
>>>>>> I use because that seems like the right thing to do where the binding is
>>>>>> obvious
>>>>>> (which I believe it is in this case as the fusb302 is the mux consumer) and
>>>>>> because the device-property code should work the same on x86 + ACPI
>>>>>> (where some platform-specific drivers attach the device properties) and
>>>>>> on e.g. ARM + DT.
>>>>>>
>>>>>> The rest should probably be left to be figured out when an actual DT
>>>>>> using device using the fusb302 or another Type-C controller shows up.
>>>>>
>>>>> Well this is a new one (maybe, I suppose others have sneaked by). If
>>>>> ACPI folks want to use DT bindings, then what do I care. But I have no
>>>>> interest in reviewing ACPI properties. The whole notion of sharing
>>>>> bindings between DT and ACPI beyond anything trivial is flawed IMO.
>>>>> The ptifalls have been discussed multiple times before, so I'm not
>>>>> going to repeat them here.
>>>>
>>>> Ok, so shall I just drop the Documentation/devicetree/bindings/usb/fcs,fusb302.txt
>>>> part of this patch then ?
>>>
>>> I totally agree with the concern that Rob expressed about handling USB C
>>> muxes in a non-adhoc way. And this makes this series scary. I don't know
>>> enough details about USB C muxes and PD (I just have a very superficial
>>> mental model) to tell if this series is going down the right path. Or
>>> how terrible it will be to fix things up if not?
>>>
>>> The series extends the mux subsystem with muxes that pins semantics
>>> to specific states. That is new, and shows up exactly here when Rob is
>>> not happy about the binding. And if Rob does not want this in the
>>> DT bindings then I'm not so sure it is wise to do it at all? This
>>> problem doesn't go away just because you remove the binding. I think
>>> I would feel much better if there was a path forward on how to
>>> represent USB C muxes in DT and how that would fit with the driver
>>> structure.
>>>
>>> If you compare to the i2c-muxes, there is a relatively new i2c-mux-gpmux
>>> driver that uses some general purpose mux from the mux subsystem to
>>> implement an i2c-mux. If USB C muxes where to be done similarly, I'd
>>> imagine there should be a general abstraction of what USB C muxes provide
>>> somewhere outside of the mux subsystem, and a bunch of implementations
>>> of that abstraction. One of those implementations could be to use "raw"
>>> muxes from the mux subsystem. Of course, this is not what this series is
>>> doing.
>>>
>>> Also, muxes that are not general purpose such as the ones added to the
>>> mux subsystem by this series could perhaps be repurposed for some other
>>> application, but since the interface implemented does not really obey
>>> the rules (the provided mux controller interacts with different sets of
>>> signals depending on the state) this will not be possible.
>>>
>>> These issues are what has caused me to do a lot of thinking and to sit
>>> silent, sorry about that, but I would like input from someone more
>>> experienced. If possible. But I'm not sure where to turn?
>>>
>>> As a crazy example, why is it not possible to hook up one signal pair
>>> from the USB C mux, not to DP, but instead to some I2C controller? Then,
>>> if done right, i2c-mux-gpmux could be hooked up with the relevant mux
>>> controller and use the signal pair for I2C, with the mux controller
>>> acting as a gate. So, maybe a bit crazy, but something like that is how
>>> I think it should work from the mux subsystem point of view. And while
>>> maybe crazy and while it might not be technically possible to do I2C
>>> over a USB C connector for some reason, I do think that whatever
>>> abstraction you come up with for USB C muxes, it has to deal with and
>>> broker requests from both the USB subsystem and whatever other
>>> subsystems deals with the alt pairs. Be it graphics for DP signals, or
>>> whatever. IIUC, the alt signals need not be graphics, and it would be
>>> sad to implement the USB C mux it in a way that makes it hard to use
>>> the alt pairs for something else.
>>>
>>> [maybe my understanding of USB C is just wrong]
>>
>> So 2 things:
>>
>> 1) The Type-C subsys does actually abstract the mux outside of the
>> Type-C port-manager (TCPM) core in the form of a tcpc_mux_dev, so for
>> DT based platforms we could instantiate a tcpc_mux_dev from a node
>> which then represents the mux as usual.
>>
>> 2) What is very different from how the mux subsys currently is used,
>> is who is in control of the mux. Currently a subsys which wants to
>> route data through the mux selects the mux in the right mode, and
>> then deselects it when done. This assumes that the mux can mux
>> the data paths to the requested destination at all times, just not
>> to multiple destinations at once. With Type-C and any moment in
>> time, there really is only one correct setting as the mux is
>> connect to a Type-C device on the other side which typically
>> only has one config. So what happens with Type-C is that the TCPM
>> negotiates with the externally connected Type-C device, and then
>> sets the mux to the one and only correct setting for that device
>> to be fully functional.
>>
>> So your i2c example, for example, will not work with the normal
>> way where the i2c subsys asks the mux to mux a data-pair to the i2c
>> controller, as that only makes sense when your theoretical Type-C
>> device which talks i2c over a pair is connected externally, where
>> as when something else is connected then trying to talk i2c might
>> even damage the connected device. So with Type-C it is the TCPM
>> and only the TCPM which controls the mux.
> 
> If I get this correctly, some code (the TCPM?) negotiates with the
> other side over PD, and after that the way to set up the USB C mux
> is known and will not change until the cable is unplugged (unless
> we go wild and renegotiate, but let's assume that will not happen).

Correct.

> What I don't get is why do you want to squeeze what the TCPM(?) is
> doing with its specialized mux into the existing mux subsystem
> interface?

That is a good question, this really started with the micro-usb
otg stuff, where we have more or less the same problem in a much
simpler way:

1) We have a cable plugged in which determined if we should mux the
usb2 data lines on a micro-usb to an usb-host or an usb-device/gadget
controller.

2) Some way to detect the type of cable plugged in

3) A hardware block or dedicated chip which is independent from 2,
which needs to be controlled by the code implementing 2.

So we need some glue layer with an abstract (device independent)
interface between 2 and 3 so that we can plug random combinations
of 2 and 3 together. In the beginning I tried using the extcon
framework for that, but that is not really a good fit.

Then someone else did some patches for the otg-mux found on the
Intel Cherry Trail block where using the mux subsys for this and
that seemed like a good idea to me.

And sofar I must say that from my pov the mux subsys does work
quite well for this, but now that I better understand the initial
design of the mux subsys I can understand that you are reluctant
about my use case.


> I mean, the mux interface isn't really a perfect fit with its
> locking and support for multiple consumers etc that really just gets
> in the way of setting a register in some chip to some value. Which
> is all that is really needed when you know that the TCPM is the only
> one accessing the chip.
> 
> And from the point of view of the mux subsystem, the USB C muxes
> like the Pericom driver will be in a class of their own. Muxes of
> that class can really only be used by one thing -- the TCPM code.
> 
> So, I don't see the benefit of doing this through the mux subsystem.
> My hope is that the TCPM code is generic, and that by putting the
> USB C mux inside the mux subsystem, you can keep the TCPM code
> generic?

Correct, the TCPM code is generic and for example the Cherry Trail
SoC OTG mux needs to be controller by either:

A) Something which is detecting the type of cable connected to a
micro-usb (either a PMIC or a gpio reading the ID pin) on devices
with a micro-usb connecter; or

B) The TCPM code on devices with a Type-C connector.

So this is another example where the mux-subsys more or less is
a pretty good fit.

As for the locking, you're right that the locking is not necessary,
but the Type-C connector and its muxes have a clearly defined idle
state, so the deselect functionality also is a decent fit there.

> That might be good for the TCPM code, but it does create a
> new class of devices in the mux subsystem, and opens the door for
> other classes later on.
> 
> Why not just add a function pointer that the generic TCPM code calls
> if it is set, and then add some code somewhere more local to the TCPM
> code to back that call, and completely avoid the mux subsystem?

See the above example about 2 completely different ways how we
need to control the Cherry Trail SoC OTG mux. We really need some
sort of abstraction layer here, and then some way to hook the TCPM
and the Type-C and OTG-mux together, either through pnode handles
in DT on DT platforms or through something like the lookup table
code one of my patches add on x86.

For me using the mux subsys for this works well. But as said I
can understand you being reluctant. Alternatively we could add a
new usb_mux subsys for this I guess.

Regards,

Hans
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
index 472facfa5a71..63d639eadacd 100644
--- a/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
+++ b/Documentation/devicetree/bindings/usb/fcs,fusb302.txt
@@ -6,6 +6,9 @@  Required properties :
 - interrupts             : Interrupt specifier
 
 Optional properties :
+- mux-controls           : List of mux-ctrl-specifiers containing 1 or 2 muxes
+- mux-names              : "type-c-mode-mux" when using 1 mux, or
+                           "type-c-mode-mux", "usb-role-mux" when using 2 muxes
 - fcs,max-sink-microvolt : Maximum voltage to negotiate when configured as sink
 - fcs,max-sink-microamp  : Maximum current to negotiate when configured as sink
 - fcs,max-sink-microwatt : Maximum power to negotiate when configured as sink
diff --git a/drivers/staging/typec/fusb302/fusb302.c b/drivers/staging/typec/fusb302/fusb302.c
index cf6355f59cd9..00d045d0246b 100644
--- a/drivers/staging/typec/fusb302/fusb302.c
+++ b/drivers/staging/typec/fusb302/fusb302.c
@@ -1259,7 +1259,6 @@  static void init_tcpc_dev(struct tcpc_dev *fusb302_tcpc_dev)
 	fusb302_tcpc_dev->set_roles = tcpm_set_roles;
 	fusb302_tcpc_dev->start_drp_toggling = tcpm_start_drp_toggling;
 	fusb302_tcpc_dev->pd_transmit = tcpm_pd_transmit;
-	fusb302_tcpc_dev->mux = NULL;
 }
 
 static const char * const cc_polarity_name[] = {
@@ -1817,6 +1816,16 @@  static int fusb302_probe(struct i2c_client *client,
 			return -EPROBE_DEFER;
 	}
 
+	chip->tcpc_dev.mux = devm_tcpc_gen_mux_create(dev);
+	if (IS_ERR(chip->tcpc_dev.mux)) {
+		ret = PTR_ERR(chip->tcpc_dev.mux);
+		/* Use of a mux is optional (for now?), ignore -ENODEV errors */
+		if (ret == -ENODEV)
+			chip->tcpc_dev.mux = NULL;
+		else
+			return ret;
+	}
+
 	cfg.drv_data = chip;
 	chip->psy = devm_power_supply_register(dev, &fusb302_psy_desc, &cfg);
 	if (IS_ERR(chip->psy)) {