Re: [PATCH v2 2/2] gpiolib: match secondary fwnode too in gpio_device_find_by_fwnode()

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Wed, Feb 25, 2026 at 09:39:12AM +0200, Sakari Ailus wrote:
> On Tue, Feb 24, 2026 at 11:43:39AM +0200, Andy Shevchenko wrote:
> > On Tue, Feb 24, 2026 at 10:56:16AM +0200, Sakari Ailus wrote:
> > > On Tue, Feb 24, 2026 at 09:47:57AM +0100, Bartosz Golaszewski wrote:
> > > > On Mon, Feb 23, 2026 at 11:07 PM Sakari Ailus
> > > > <sakari.ailus@linux.intel.com> wrote:

...

> > > > > > > Could device_match_fwnode() match secondary fwnode as well?
> > > > > >
> > > > > > In the previous discussion on this, Andy was against doing that due to
> > > > > > the concern that it might introduce subtle bugs, which I agree with.
> > > > >
> > > > > Could you elaborate or provide an example?
> > 
> > I believe you ask me. Okay, the sophisticated case I have in mind is the
> > intel_quark_i2c_gpio.c which provides a GPIO device with a list of children.
> > 
> > First of all, it seems broken as it rewrites the secondary link for the
> > I²C device. (Which makes me think that we need to have a copy of the
> > [primary] fwnode in the children devices of MFD, but I don't know how
> > to refcount that properly). The gpiolib-acpi-core.c has a matching function
> > via ACPI_HANDLE(). So it might be not affected by this.
> > 
> > What I don't know is USB Type-C and USB DWC3 code where it's much more
> > complicated. And I'm not in a position to state that the change won't
> > affect those.
> 
> Any idea who has the hardware in these cases? There aren't that many users
> of this function out there and I think at some point we do need to fix
> this.

Ask Heikki?

> What we could also do is that we add another function that only cares about
> the very fwnode you have at hand, switch the dubious cases to use that and
> have the proper function test both available fwnodes. That'd get us on the
> right path to fix this eventually, if not now.
> 
> > > > > The function has some 27 users although few are individual drivers.
> > > > >
> > > > > My understanding is that we only have the secondary fwnode for being able
> > > > > to attach objects from different backend to the same node. The fwnode API
> > > > > in the meantime generally tries to hide the existence of the secondary
> > > > > fwnode; a rewrite (which ideally would have happened perhaps a few years
> > > > > ago?) would probably make the fwnode a linked list instead so we'd lose
> > > > > that secondary pointer in the process.
> > > > 
> > > > It already is a (singly) linked list. Ideally it would be a
> > > 
> > > With two entries at most.
> > 
> > There is no technical limitation based on the data type.
> 
> There aren't any, no, but the current implementation assumes this, and I
> wouldn't change this without changing the data structure as well.

How does it assume? A caller may crawl via the list pretending that each of
fwnode is "the head of the single linked list".

I would agree with you if the struct fwnode_handle was opaque, but it doesn't.

> > > > doubly-linked list moved into struct device with struct fwnode_handle
> > > > having no concept of primary and secondary nodes.
> > > 
> > > I'd think we had that list in struct fwnode_handle, which will still
> > > represent nodes. But let's see the details when someone gets to implement
> > > it. :-)
> > 
> > In the case above single or double linked list doesn't solve the issue of
> > the corrupted (parent) fwnode. We need also to have a siblings list so it
> > looks more like a tree.

-- 
With Best Regards,
Andy Shevchenko