Re: [PATCH] driver core: platform: allow attaching software nodes when creating devices

[Dmitry Torokhov]

On Sat, Feb 14, 2026 at 12:25:02AM +0100, Danilo Krummrich wrote:
> On Fri Feb 13, 2026 at 11:36 PM CET, Dmitry Torokhov wrote:
> > Extend platform_device_info structure with an optional pointer to a
> > software node to be used as a secondary firmware node for the device
> > being created. If software node has not been registered yet it will be
> > automatically registered.
> >
> > This reduces boilerplate needed when switching legacy board code to
> > static device properties/GPIO references.
> 
> Are there any patches for this already? If so, how do you plan to land them?

I have a couple, the affect different boards/subsystems. Ideally if we
could have an immutable branch for everyone to pull in that would be
better. Otherwise we'll have to hold the patches for a bit...

> 
> > Signed-off-by: Dmitry Torokhov <dmitry.torokhov@xxxxxxxxx>
> > ---
> >  drivers/base/platform.c         | 15 ++++++++++-----
> >  include/linux/platform_device.h | 23 ++++++++++++-----------
> >  2 files changed, 22 insertions(+), 16 deletions(-)
> >
> > diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> > index b45d41b018ca..bef69ad68e02 100644
> > --- a/drivers/base/platform.c
> > +++ b/drivers/base/platform.c
> > @@ -850,6 +850,9 @@ struct platform_device *platform_device_register_full(
> >  	int ret;
> >  	struct platform_device *pdev;
> >  
> > +	if (pdevinfo->swnode && pdevinfo->properties)
> > +		return ERR_PTR(-EINVAL);
> 
> Unfortunate that we can't catch this at build time.
> 
> > +
> >  	pdev = platform_device_alloc(pdevinfo->name, pdevinfo->id);
> >  	if (!pdev)
> >  		return ERR_PTR(-ENOMEM);
> > @@ -865,17 +868,19 @@ struct platform_device *platform_device_register_full(
> >  		pdev->dev.coherent_dma_mask = pdevinfo->dma_mask;
> >  	}
> >  
> > -	ret = platform_device_add_resources(pdev,
> > -			pdevinfo->res, pdevinfo->num_res);
> 
> While the formatting is indeed not great, I prefer not to change this in this
> patch. The same goes for the other unrelated formatting changes in this patch.

I'll split into several patches.

> 
> > +	ret = platform_device_add_resources(pdev, pdevinfo->res, pdevinfo->num_res);
> >  	if (ret)
> >  		goto err;
> >  
> > -	ret = platform_device_add_data(pdev,
> > -			pdevinfo->data, pdevinfo->size_data);
> > +	ret = platform_device_add_data(pdev, pdevinfo->data, pdevinfo->size_data);
> >  	if (ret)
> >  		goto err;
> >  
> > -	if (pdevinfo->properties) {
> > +	if (pdevinfo->swnode) {
> > +		ret = device_add_software_node(&pdev->dev, pdevinfo->swnode);
> > +		if (ret)
> > +			goto err;
> > +	} else if (pdevinfo->properties) {
> >  		ret = device_create_managed_software_node(&pdev->dev,
> >  							  pdevinfo->properties, NULL);
> >  		if (ret)
> > diff --git a/include/linux/platform_device.h b/include/linux/platform_device.h
> > index 813da101b5bf..3d96edfe0ee6 100644
> > --- a/include/linux/platform_device.h
> > +++ b/include/linux/platform_device.h
> > @@ -119,21 +119,22 @@ extern int platform_get_irq_byname_optional(struct platform_device *dev,
> >  extern int platform_add_devices(struct platform_device **, int);
> >  
> >  struct platform_device_info {
> > -		struct device *parent;
> > -		struct fwnode_handle *fwnode;
> > -		bool of_node_reused;
> > +	struct device *parent;
> > +	struct fwnode_handle *fwnode;
> > +	bool of_node_reused;
> >  
> > -		const char *name;
> > -		int id;
> > +	const char *name;
> > +	int id;
> >  
> > -		const struct resource *res;
> > -		unsigned int num_res;
> > +	const struct resource *res;
> > +	unsigned int num_res;
> >  
> > -		const void *data;
> > -		size_t size_data;
> > -		u64 dma_mask;
> > +	const void *data;
> > +	size_t size_data;
> > +	u64 dma_mask;
> >  
> > -		const struct property_entry *properties;
> > +	const struct software_node *swnode;
> > +	const struct property_entry *properties;
> 
> It would be nice to document that the two are mutually exclusive.

I'll document platform_device_info.

Thanks.

-- 
Dmitry