Re: [PATCH v2] media: i2c: thp7312: use fwnode_for_each_child_node()

[Laurent Pinchart]

On Tue, Apr 08, 2025 at 11:08:10AM +0000, Sakari Ailus wrote:
> Moi,
> 
> On Tue, Apr 08, 2025 at 01:42:12PM +0300, Matti Vaittinen wrote:
> > On 08/04/2025 13:36, Sakari Ailus wrote:
> > > Hei Laurent, Matti,
> > > 
> > > On Tue, Apr 08, 2025 at 01:26:42PM +0300, Matti Vaittinen wrote:
> > > > On 08/04/2025 13:12, Laurent Pinchart wrote:
> > > > > Hi Sakari,
> > > > > 
> > > > > On Tue, Apr 08, 2025 at 08:48:45AM +0000, Sakari Ailus wrote:
> > > > > > On Fri, Mar 21, 2025 at 12:41:00PM +0200, Laurent Pinchart wrote:
> > > > > > > On Fri, Mar 21, 2025 at 10:58:27AM +0200, Matti Vaittinen wrote:
> > > > > > > > When fwnode_for_each_available_child_node() is used on the device-tree
> > > > > > > > backed systems, it renders to same operation as the
> > > > > > > > fwnode_for_each_child_node(), because the fwnode_for_each_child_node()
> > > > > > > > does only iterate through those device-tree nodes which are available.
> > > > > > > 
> > > > > > > This makes me wonder why the OF backend implements
> > > > > > > fwnode_for_each_child_node() as fwnode_for_each_available_child_node().
> > > > > > > Is that on purpose, or is it a bug ?
> > > > > > 
> > > > > > I discussed this with Rafael and he didn't recall why the original
> > > > > > implementation was like that. The general direction later on has been not
> > > > > > to present unavailable nodes over the fwnode interface.
> > > > > > 
> > > > > > So I'd say:
> > > > > > 
> > > > > > Reviewed-by: Sakari Ailus <sakari.ailus@xxxxxxxxxxxxxxx>
> > > > > > 
> > > > > > We should also change the documentation of the fwnode API accordingly.
> > > > > 
> > > > > Does that also mean that the fwnode_for_each_available_child_node()
> > > > > function will be dropped ? It's used by few drivers (5 in addition to
> > > > > the thp7312 driver, plus 3 call sites in drivers/base/core.c), so a
> > > > > patch series to drop it should be easy.
> > > > > 
> > > > 
> > > > I assume the fwnode_for_each_available_child_node() still makes sense for
> > > > ACPI backed users, no?
> > > 
> > > Not really (see my earlier explanation in
> > > <Z9mQPJwnKAkPHriT@kekkonen.localdomain>).
> > 
> > I capture that the _named_ available nodes don't have value as ACPI names
> > aren't really what is expected by the _named_ callers. What I didn't pick is
> > that the fwnode_for_each_available_child_node() - which should iterate all
> > available child nodes ignoring the name - wouldn't be useful.
> 
> Fair enough. I don't think we need to support enumerating unavailable ACPI
> device nodes in this API. I'd indeed change the behaviour so that only
> available nodes are enumerated. I can post a patch for that.

Unless there's a specific reason against it that I wouldn't be aware of,
I would also very much favour merging the fwnode_for_each_child_node()
and fwnode_for_each_available_child_node() functions into a single one
that only enumerates available nodes, with a consistent behaviour across
all backend. Having fwnode_for_each_child_node() return unavailable ACPI
nodes but not unavailable DT nodes would be really confusing.

> > > I think all the *available* stuff
> > > should be removed from include/linux/property.h, apart from
> > > fwnode_device_is_availble(), which should be turned to work on struct
> > > device to signal its availability for device nodes only.
> > 
> > I am not saying I have any understanding of the uses of the 'unavailable'
> > nodes. As such I am not arguing over what you say here :)

-- 
Regards,

Laurent Pinchart