Hi,

On Wed, Nov 21, 2018 at 11:56:50PM +0200, Sakari Ailus wrote:
> On Thu, Nov 15, 2018 at 08:04:24PM +0100, Maxime Ripard wrote:
> > On Tue, Nov 13, 2018 at 10:38:55AM +0200, Sakari Ailus wrote:
> > > > +  - allwinner,has-isp: Whether the CSI controller has an ISP
> > > > +                       associated to it or not
> > > 
> > > Is the ISP a part of the same device? It sounds like that this is actually
> > > a different device if it contains an ISP as well, and that should be
> > > apparent from the compatible string. What do you think?
> > 
> > I guess we can see it as both. It seems to be the exact same register
> > set, except for the fact that the first instance has that ISP in
> > addition, and several channels, as you pointed out in your other mail.
> 
> I'm simply referring to existing practices as far as I know them. If it's a
> different device, it should have a different compatible string, not a
> vendor-specific property to tell it's somehow different.

The line is blurrier than that. Different devices are indeed
represented using different compatibles, but different features set
can be represented using properties.

> Many SoCs also separate the DMA and the CSI-2 receivers, and thus they have
> separate drivers. I don't know about your case, but the ISP requiring a
> different clock is a hint.

In this particular case, the datasheet represents the ISP as part of
the DMA, so it looks like a feature. And since we don't have any
source code for this, we can only do guesswork.

> > What these channels are is not exactly clear. It looks like it's
> > related to the BT656 interface where you could interleave channel
> > bytes over the bus. I haven't really looked into it, and it doesn't
> > look like we have any code (or hardware) able to do that though.
> > 
> > > > +
> > > > +If allwinner,has-isp is set, an additional "isp" clock is needed,
> > > > +being a phandle to the clock driving the ISP.
> > > > +
> > > > +The CSI node should contain one 'port' child node with one child
> > > > +'endpoint' node, according to the bindings defined in
> > > > +Documentation/devicetree/bindings/media/video-interfaces.txt. The
> > > > +endpoint's bus type must be parallel or BT656.
> > > > +
> > > > +Endpoint node properties for CSI
> > > > +---------------------------------
> > > > +
> > > > +- remote-endpoint	: (required) a phandle to the bus receiver's endpoint
> > > > +			   node
> > > 
> > > Rob's opinion has been (AFAIU) that this is not needed as it's already a
> > > part of the graph bindings. Unless you want to say that it's required, that
> > > is --- the graph bindings document it as optional.
> > 
> > Ok, I'll remove it.
> > 
> > > > +- bus-width:		: (required) must be 8
> > > 
> > > If this is the only value the hardware supports, I don't see why you should
> > > specify it here.
> > 
> > Ditto :)
> > 
> > > > +- pclk-sample		: (optional) (default: sample on falling edge)
> > > > +- hsync-active		: (only required for parallel)
> > > > +- vsync-active		: (only required for parallel)
> > > > +
> > > > +Example:
> > > > +
> > > > +csi0: csi@1c09000 {
> > > > +	compatible = "allwinner,sun7i-a20-csi",
> > > > +		     "allwinner,sun4i-a10-csi";
> > > > +	reg = <0x01c09000 0x1000>;
> > > > +	interrupts = <GIC_SPI 42 IRQ_TYPE_LEVEL_HIGH>;
> > > > +	clocks = <&ccu CLK_AHB_CSI0>, <&ccu CLK_CSI0>,
> > > > +		 <&ccu CLK_CSI_SCLK>, <&ccu CLK_DRAM_CSI0>;
> > > > +	clock-names = "ahb", "mod", "isp", "ram";
> > > > +	resets = <&ccu RST_CSI0>;
> > > > +	allwinner,csi-channels = <4>;
> > > > +	allwinner,has-isp;
> > > > +	
> > > > +	port {
> > > > +		csi_from_ov5640: endpoint {
> > > > +                        remote-endpoint = <&ov5640_to_csi>;
> > > > +                        bus-width = <8>;
> > > > +                        data-shift = <2>;
> > > 
> > > data-shift needs to be documented above if it's relevant for the device.
> > 
> > It's not really related to the CSI device in that case but the sensor,
> > so I'll just leave it out.
> 
> Hmm. data-shift should only be relevant for the receiver, shoudn't it?

Sorry, I mispoke. We're not using it anywhere in either drivers, so
it's totally useless.

Maxime

-- 
Maxime Ripard, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com