Re: [RFC PATCH 1/3] i2c: core: Follow i2c-parent when retrieving an adapter from node

[Herve Codina <herve.codina@bootlin.com>]

Hi Wolfram,

On Thu, 3 Apr 2025 11:03:27 +0200
Wolfram Sang <wsa+renesas@sang-engineering.com> wrote:

> > Extend i2c_find_adapter_by_fwnode() to perform the walking from the
> > given fwnode through i2c-parent references up to the adapter.  
> 
> Even with the review of the schema going on, here are some comments
> already.

Yes. Of course, depending on this review, things could be changed in the
implementation but every things already discussed here make the topic
moving forward. Thanks for that!

...

> > +
> > +	/* Walk extension busses (through i2c-parent) up to the adapter node */
> > +	while (fwnode_property_present(adap_fwnode, "i2c-parent")) {
> > +		/*
> > +		 * A specific case exists for the i2c demux pinctrl. The i2c bus
> > +		 * node related this component (the i2c demux pinctrl node
> > +		 * itself) has an i2c-parent property set. This property is used
> > +		 * by the i2c demux pinctrl component for the demuxing purpose
> > +		 * and is not related to the extension bus feature.
> > +		 *
> > +		 * In this current i2c-parent walking, the i2c demux pinctrl
> > +		 * node has to be considered as an adapter node and so, if
> > +		 * the adap_fwnode node is an i2c demux pinctrl node, simply
> > +		 * stop the i2c-parent walking.
> > +		 */
> > +		if (fwnode_property_match_string(adap_fwnode, "compatible",
> > +						 "i2c-demux-pinctrl") >= 0)
> > +			break;  
> 
> I understand the unlikeliness of another demux driver showing up, yet
> relying on compatible-values here is too easy to get stale. What about
> checking if the i2c-parent property has more than one entry? That should
> be only true for demuxers.

Indeed, this is better.
I will stop the walking based on this number of entries in the i2c-parent
property.

> 
> > +
> > +		/*
> > +		 * i2c-parent property available in a i2c bus node means that
> > +		 * this node is an extension bus node. In that case,
> > +		 * continue i2c-parent walking up to the adapter node.
> > +		 */
> > +		err = fwnode_property_get_reference_args(adap_fwnode, "i2c-parent",
> > +							 NULL, 0, 0, &args);
> > +		if (err)
> > +			break;
> > +
> > +		pr_debug("Find adapter for %pfw, use parent: %pfw\n", fwnode,
> > +			 args.fwnode);  
> 
> Is this useful when creating the overlays? I tend to ask you to remove
> it one RFC phase is over. If it is useful, it should be at least
> dev_dbg?

Using dev_dbg could lead to issues. Indeed, dev is not given as an argument
of i2c_find_adapter_by_fwnode() and there is no reason to add it (except
for this debug message).

Without a dev given as argument we have to retrieve it from the given
fw_node argument. This given fw_node may have its dev field not already set.
Indeed, the dev instanciation could be done later when the bus this fw_node
is connected to will probe().

For instance
  https://elixir.bootlin.com/linux/v6.14-rc6/source/drivers/gpu/drm/panel/panel-simple.c#L606
The panel driver can call of_find_i2c_adapter_by_node() which in turn call
i2c_find_adapter_by_fwnode() without having the I2C controller related to the
adapter already present. the panel driver will return a legit EPROBE_DEFER.

So back to our debug message in i2c_find_adapter_by_fwnode(), either I keep
pr_debug() or I fully remove the message but I don't thing I should change
the i2c_find_adapter_by_fwnode() prototype and update all the callers just
for this debug message.

The debug message can be interesting when things went wrong and we want
to investigate potential issue with i2c-parent chain from the last device
up to the adapter.

I don't have a strong opinion about the need of this message and I can
simply remove it.

What is your preference ?

Best regards,
Hervé