Re: [V9, 1/2] media: dt-bindings: media: i2c: Document OV02A10 bindings

[Dongchun Zhu <dongchun.zhu@mediatek.com>]

Hi Sakari,

On Tue, 2020-06-02 at 12:56 +0300, Sakari Ailus wrote:
> Hi Dongchun,
> 
> On Tue, Jun 02, 2020 at 02:15:01PM +0800, Dongchun Zhu wrote:
> > Hi Tomasz, Sakari,
> > 
> > On Mon, 2020-06-01 at 20:18 +0200, Tomasz Figa wrote:
> > > On Mon, Jun 1, 2020 at 4:35 AM Dongchun Zhu <dongchun.zhu@mediatek.com> wrote:
> > > >
> > > > Hi Tomasz,
> > > >
> > > > On Fri, 2020-05-29 at 15:43 +0200, Tomasz Figa wrote:
> > > > > On Thu, May 28, 2020 at 10:06 AM Dongchun Zhu <dongchun.zhu@mediatek.com> wrote:
> > > > > >
> > > > > > Hi Sakari,
> > > > > >
> > > > > > On Thu, 2020-05-28 at 10:23 +0300, Sakari Ailus wrote:
> > > > > > > Hi Dongchun,
> > > > > > >
> > > > > > > On Thu, May 28, 2020 at 11:34:42AM +0800, Dongchun Zhu wrote:
> > > > > > > > Hi Sakari, Rob,
> > > > > > > >
> > > > > > > > On Thu, 2020-05-28 at 00:16 +0300, Sakari Ailus wrote:
> > > > > > > > > Hi Rob, Dongchun,
> > > > > > > > >
> > > > > > > > > On Wed, May 27, 2020 at 09:27:22AM -0600, Rob Herring wrote:
> > > > > > > > > > > > > +    properties:
> > > > > > > > > > > > > +      endpoint:
> > > > > > > > > > > > > +        type: object
> > > > > > > > > > > > > +        additionalProperties: false
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +        properties:
> > > > > > > > > > >
> > > > > > > > > > > Actually I wonder whether we need to declare 'clock-lanes' here?
> > > > > > > > > >
> > > > > > > > > > Yes, if you are using it.
> > > > > > > > >
> > > > > > > > > Dongchun, can you confirm the chip has a single data and a single clock
> > > > > > > > > lane and that it does not support lane reordering?
> > > > > > > > >
> > > > > > > >
> > > > > > > > From the datasheet, 'MIPI inside the OV02A10 provides one single
> > > > > > > > uni-directional clock lane and one bi-directional data lane solution for
> > > > > > > > communication links between components inside a mobile device.
> > > > > > > > The data lane has full support for HS(uni-directional) and
> > > > > > > > LP(bi-directional) data transfer mode.'
> > > > > > > >
> > > > > > > > The sensor doesn't support lane reordering, so 'clock-lanes' property
> > > > > > > > would not be added in next release.
> > > > > > > >
> > > > > > > > > So if there's nothing to convey to the driver, also the data-lanes should
> > > > > > > > > be removed IMO.
> > > > > > > > >
> > > > > > > >
> > > > > > > > However, 'data-lanes' property may still be required.
> > > > > > > > It is known that either data-lanes or clock-lanes is an array of
> > > > > > > > physical data lane indexes. Position of an entry determines the logical
> > > > > > > > lane number, while the value of an entry indicates physical lane, e.g.,
> > > > > > > > for 1-lane MIPI CSI-2 bus we could have "data-lanes = <1>;", assuming
> > > > > > > > the clock lane is on hardware lane 0.
> > > > > > > >
> > > > > > > > As mentioned earlier, the OV02A10 sensor supports only 1C1D and does not
> > > > > > > > support lane reordering, so here we shall use 'data-lanes = <1>' as
> > > > > > > > there is only a clock lane for OV02A10.
> > > > > > > >
> > > > > > > > Reminder:
> > > > > > > > If 'data-lanes' property is not present, the driver would assume
> > > > > > > > four-lane operation. This means for one-lane or two-lane operation, this
> > > > > > > > property must be present and set to the right physical lane indexes.
> > > > > > > > If the hardware does not support lane reordering, monotonically
> > > > > > > > incremented values shall be used from 0 or 1 onwards, depending on
> > > > > > > > whether or not there is also a clock lane.
> > > > > > >
> > > > > > > How can the driver use four lanes, considering the device only supports a
> > > > > > > single lane??
> > > > > > >
> > > > > >
> > > > > > I understood your meaning.
> > > > > > If we omit the property 'data-lanes', the sensor should work still.
> > > > > > But then what's the meaning of the existence of 'data-lanes'?
> > > > > > If this property 'data-lanes' is always optional, then why dt-bindings
> > > > > > provide the interface?
> > > > > >
> > > > > > In the meantime, if omitting 'data-lanes' for one sensor(transmitter)
> > > > > > that has only one physical data lane, MIPI receiver(e.g., MIPI CSI-2)
> > > > > > shall enable four-lane configuration, which may increase consumption of
> > > > > > both power and resource in the process of IIC communication.
> > > > >
> > > > > Wouldn't the receiver still have the data-lanes property under its
> > > > > endpoint node, telling it how many lanes and in which order should be
> > > > > used?
> > > > >
> > > >
> > > > The MIPI receiver(RX) shall use
> > > > v4l2_async_notifier_add_fwnode_remote_subdev() API to parse the property
> > > > "data-lanes" under sensor output port.
> > > 
> > > That's not true. The MIPI receiver driver parses its own port node
> > > corresponding to the sensor. Also quoting the documentation [1]:
> > > 
> > > "An endpoint subnode of a device contains all properties needed for
> > > _configuration of this device_ for data exchange with other device. In most
> > > cases properties at the peer 'endpoint' nodes will be identical, however they
> > > might need to be different when there is any signal modifications on the bus
> > > between two devices, e.g. there are logic signal inverters on the lines."
> > > 
> > > In this case, there is such a signal modification if the sensor has a
> > > 1-lane bus and the receiver more lines, so the data-lanes properties
> > > would be different on both sides.
> > > 
> > > [1] https://elixir.bootlin.com/linux/v5.7/source/Documentation/devicetree/bindings/media/video-interfaces.txt
> > > 
> > 
> > Sorry for the misunderstanding.
> > After doing some experiments about the data-lanes property under sensor
> > i2c node, we found the API
> > v4l2_async_notifier_add_fwnode_remote_subdev() that MIPI receiver driver
> > used indeed parses the data-lanes under its own port node.
> > 
> > Sorry make a mistake for the use case of sensor data-lanes previously.
> > Now We may encounter one new question for this patch.
> > In practice we haven't used the data-lanes under sensor i2c node
> > anywhere, if sensor driver itself doesn't parse that.
> > 
> > But there is still one reason to keep the exactly right data-lanes in
> > DT. That is, the data-lanes under sensor i2c node could be used as a
> > reference for MIPI receiver driver.
> > Just as Tomasz said, 'The MIPI receiver driver parses its own port node
> > corresponding to the sensor'.
> > 
> > Sakari, Tomasz, what's your opinions about the present of data-lanes
> > under sensor node or not?
> 
> The receiver driver doesn't parse the properties in the sensor
> (transmitter) device's endpoint. If that property provides no information
> to the receiver, as is the case here, it should be omitted.
> 

Understood.
Fixed in next release.