Re: [PATCH v1 1/1] PCI: of: Propagate firmware node

[Bjorn Helgaas <helgaas@kernel.org>]

On Thu, Apr 13, 2023 at 07:00:26PM +0300, Andy Shevchenko wrote:
> On Wed, Apr 12, 2023 at 11:02:53AM -0500, Bjorn Helgaas wrote:
> > On Wed, Apr 12, 2023 at 04:15:20PM +0300, Andy Shevchenko wrote:
> > > Propagate firmware node by using a specific API call, i.e. device_set_node().
> > 
> > Can you add a line or two about *why* we should do this, e.g., is this
> > headed toward some goal?
> 
> Because dereferencing the fwnode in struct device is preventing us from
> modifications of how fwnode looks like in the future.

How do you want to express this in the commit log?  Something like
this?

  Insulate pci_set_of_node() and pci_set_bus_of_node() from possible
  changes to fwnode_handle implementation by using device_set_node()
  instead of open-coding dev->dev.fwnode assignments.

> > Is it a simplification that's 100%
> > equivalent (doesn't seem so, see below)?
> 
> To me it's an equivalent, I'll explain below.
> 
> > Seems like there's an underlying long-term effort to unify things from
> > OF and ACPI, which seems like a good thing, but at the moment it's a
> > little confusing to follow.  For instance pci_set_of_node() seems like
> > it ought to be sort of analogous to pci_set_acpi_fwnode(), but they
> > look nothing alike.
> 
> Unification to some extent, but here is not a point of this change.
> 
> ...
> 
> > > +	struct device_node *node;
> > > +
> > >  	if (!dev->bus->dev.of_node)
> > >  		return;
> > > -	dev->dev.of_node = of_pci_find_child_device(dev->bus->dev.of_node,
> > > -						    dev->devfn);
> > > -	if (dev->dev.of_node)
> > > -		dev->dev.fwnode = &dev->dev.of_node->fwnode;
> > > +	node = of_pci_find_child_device(dev->bus->dev.of_node, dev->devfn);
> > > +	device_set_node(&dev->dev, of_fwnode_handle(node));
> > 
> > This doesn't seem 100% equivalent.  If of_pci_find_child_device()
> > returns NULL, the previous code doesn't set dev->dev.fwnode, but the
> > new code does.
> 
> Yes and this is not a problem. We create device with pci_alloc_dev() in both
> callers of the pci_setup_device() and the field is NULL anyway. So, the last
> condition there is a simple micro-optimisation.

OK, makes sense, thanks.

Bjorn