The commit 407d1a51921e ("PCI: Create device tree node for bridge")
creates of_node for PCI devices. The newly created of_node is attached
to an existing device. This is done setting directly pdev->dev.of_node
in the code.

Even if pdev->dev.of_node cannot be previously set, this doesn't handle
the fwnode field of the struct device. Indeed, this field needs to be
set if it hasn't already been set.

device_{add,remove}_of_node() have been introduced to handle this case.

Use them instead of the direct setting.

Signed-off-by: Herve Codina <herve.codina@bootlin.com>
Reviewed-by: Rob Herring (Arm) <robh@kernel.org>
---
 drivers/pci/of.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/pci/of.c b/drivers/pci/of.c
index 7a806f5c0d20..fb5e6da1ead0 100644
--- a/drivers/pci/of.c
+++ b/drivers/pci/of.c
@@ -653,8 +653,8 @@ void of_pci_remove_node(struct pci_dev *pdev)
 	np = pci_device_to_OF_node(pdev);
 	if (!np || !of_node_check_flag(np, OF_DYNAMIC))
 		return;
-	pdev->dev.of_node = NULL;
 
+	device_remove_of_node(&pdev->dev);
 	of_changeset_revert(np->data);
 	of_changeset_destroy(np->data);
 	of_node_put(np);
@@ -711,11 +711,18 @@ void of_pci_make_dev_node(struct pci_dev *pdev)
 		goto out_free_node;
 
 	np->data = cset;
-	pdev->dev.of_node = np;
+
+	ret = device_add_of_node(&pdev->dev, np);
+	if (ret)
+		goto out_revert_cset;
+
 	kfree(name);
 
 	return;
 
+out_revert_cset:
+	np->data = NULL;
+	of_changeset_revert(cset);
 out_free_node:
 	of_node_put(np);
 out_destroy_cset:
-- 
2.48.1

On Mon, Feb 24, 2025 at 03:13:52PM +0100, Herve Codina wrote:
> The commit 407d1a51921e ("PCI: Create device tree node for bridge")
> creates of_node for PCI devices. The newly created of_node is attached
> to an existing device. This is done setting directly pdev->dev.of_node
> in the code.
> 
> Even if pdev->dev.of_node cannot be previously set, this doesn't handle
> the fwnode field of the struct device. Indeed, this field needs to be
> set if it hasn't already been set.
> 
> device_{add,remove}_of_node() have been introduced to handle this case.

I guess another way to say this is:

  - If dev->of_node has already been set, it is an error and we want
    to do nothing.  The error is impossible in this case because
    of_pci_make_dev_node() returns early if dev->of_node has been set.

  - Otherwise, we want to set dev->of_node (just as we previously
    did), and

  - if dev->fwnode has not been set, we want to set that too.

So the whole point of this is to set dev->fwnode, which we didn't do
before.  But has np->fwnode been set to anything?  Maybe it's buried
somewhere inside of_changeset_create_node(), but I didn't see it.

I can't tell if this actually *does* anything.  And if it does, all
the above is basically C code translated into English, so it doesn't
tell us *why* this is useful, which is what I try to put in the merge
commit log.

> Use them instead of the direct setting.
> 
> Signed-off-by: Herve Codina <herve.codina@bootlin.com>
> Reviewed-by: Rob Herring (Arm) <robh@kernel.org>
> ---
>  drivers/pci/of.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/pci/of.c b/drivers/pci/of.c
> index 7a806f5c0d20..fb5e6da1ead0 100644
> --- a/drivers/pci/of.c
> +++ b/drivers/pci/of.c
> @@ -653,8 +653,8 @@ void of_pci_remove_node(struct pci_dev *pdev)
>  	np = pci_device_to_OF_node(pdev);
>  	if (!np || !of_node_check_flag(np, OF_DYNAMIC))
>  		return;
> -	pdev->dev.of_node = NULL;
>  
> +	device_remove_of_node(&pdev->dev);
>  	of_changeset_revert(np->data);
>  	of_changeset_destroy(np->data);
>  	of_node_put(np);
> @@ -711,11 +711,18 @@ void of_pci_make_dev_node(struct pci_dev *pdev)
>  		goto out_free_node;
>  
>  	np->data = cset;
> -	pdev->dev.of_node = np;
> +
> +	ret = device_add_of_node(&pdev->dev, np);
> +	if (ret)
> +		goto out_revert_cset;
> +
>  	kfree(name);
>  
>  	return;
>  
> +out_revert_cset:
> +	np->data = NULL;
> +	of_changeset_revert(cset);
>  out_free_node:
>  	of_node_put(np);
>  out_destroy_cset:
> -- 
> 2.48.1
>

On Fri, 28 Feb 2025 14:58:55 -0600
Bjorn Helgaas <helgaas@kernel.org> wrote:

> On Mon, Feb 24, 2025 at 03:13:52PM +0100, Herve Codina wrote:
> > The commit 407d1a51921e ("PCI: Create device tree node for bridge")
> > creates of_node for PCI devices. The newly created of_node is attached
> > to an existing device. This is done setting directly pdev->dev.of_node
> > in the code.
> > 
> > Even if pdev->dev.of_node cannot be previously set, this doesn't handle
> > the fwnode field of the struct device. Indeed, this field needs to be
> > set if it hasn't already been set.
> > 
> > device_{add,remove}_of_node() have been introduced to handle this case.  
> 
> I guess another way to say this is:
> 
>   - If dev->of_node has already been set, it is an error and we want
>     to do nothing.  The error is impossible in this case because
>     of_pci_make_dev_node() returns early if dev->of_node has been set.
> 
>   - Otherwise, we want to set dev->of_node (just as we previously
>     did), and
> 
>   - if dev->fwnode has not been set, we want to set that too.
> 
> So the whole point of this is to set dev->fwnode, which we didn't do
> before.  But has np->fwnode been set to anything?  Maybe it's buried
> somewhere inside of_changeset_create_node(), but I didn't see it.

np->fwnode can be set by ACPI. We are at the frontier between ACPI and
device-tree.

The ofnode is created and filled from an already existing device. This
device can be created from information provided by the ACPI world.
In that case, np->fwnode is set to and ACPI fwnode.

Best regards,
Hervé