Re: [PATCH v5 4/9] rust: device: Enable printing fwnode name and path

[Danilo Krummrich <dakr@kernel.org>]

On Thu, May 22, 2025 at 06:49:15AM +0200, Greg Kroah-Hartman wrote:
> On Wed, May 21, 2025 at 09:13:08PM +0200, Danilo Krummrich wrote:
> > On Wed, May 21, 2025 at 08:36:10PM +0200, Remo Senekowitsch wrote:
> > > On Wed May 21, 2025 at 6:58 PM CEST, Greg Kroah-Hartman wrote:
> > > > On Wed, May 21, 2025 at 03:03:07PM +0200, Remo Senekowitsch wrote:
> > > >> On Wed May 21, 2025 at 2:02 PM CEST, Greg Kroah-Hartman wrote:
> > > >> > On Tue, May 20, 2025 at 10:00:19PM +0200, Remo Senekowitsch wrote:
> > > >> >> Add two new public methods `display_name` and `display_path` to
> > > >> >> `FwNode`. They can be used by driver authors for logging purposes. In
> > > >> >> addition, they will be used by core property abstractions for automatic
> > > >> >> logging, for example when a driver attempts to read a required but
> > > >> >> missing property.
> > > >> >> 
> > > >> >> Signed-off-by: Remo Senekowitsch <remo@buenzli.dev>
> > > >> >> ---
> > > >> >>  rust/kernel/device/property.rs | 72 ++++++++++++++++++++++++++++++++++
> > > >> >>  1 file changed, 72 insertions(+)
> > > >> >> 
> > > >> >> diff --git a/rust/kernel/device/property.rs b/rust/kernel/device/property.rs
> > > >> >> index 70593343bd811..6ccc7947f9c31 100644
> > > >> >> --- a/rust/kernel/device/property.rs
> > > >> >> +++ b/rust/kernel/device/property.rs
> > > >> >> @@ -32,6 +32,78 @@ pub(crate) fn as_raw(&self) -> *mut bindings::fwnode_handle {
> > > >> >>          self.0.get()
> > > >> >>      }
> > > >> >>  
> > > >> >> +    /// Returns an object that implements [`Display`](core::fmt::Display) for
> > > >> >> +    /// printing the name of a node.
> > > >> >> +    pub fn display_name(&self) -> impl core::fmt::Display + '_ {
> > > >> >> +        struct FwNodeDisplayName<'a>(&'a FwNode);
> > > >> >> +
> > > >> >> +        impl core::fmt::Display for FwNodeDisplayName<'_> {
> > > >> >> +            fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
> > > >> >> +                // SAFETY: self is valid by its type invariant
> > > >> >> +                let name = unsafe { bindings::fwnode_get_name(self.0.as_raw()) };
> > > >> >> +                if name.is_null() {
> > > >> >> +                    return Ok(());
> > > >> >
> > > >> > So if there is no name, you are returning Ok()?  Are you sure that's ok
> > > >> > to do?  What will the result of the string look like then?
> > > >> 
> > > >> In that case we're not writing anything to the formatter, which is
> > > >> equivalent to an empty string. `Ok(())` means that writing succeeded.
> > > >> 
> > > >> I assumed that a valid node would always have a name. And we're
> > > >> guaranteed to have a valid node. So I assumed this case would never
> > > >> happen and didn't think too hard about it. But even if a valid node has
> > > >> not name, empty string is probably the correct thing, right?
> > > >
> > > > I don't know what this "name" is used for.  An empty string might not be
> > > > what you want to use here, given that you could be naming something
> > > > based on it, right?  fwnode_get_name() is used for many things,
> > > > including the detection if a name is not present at all, and if not,
> > > > then the code needs to clean up and abort.
> > > >
> > > > So what exactly are you going to be using this for?
> > > 
> > > Valid question... I'm not using it for anything.
> > 
> > You're using this in PropertyGuard::required_by(), where you use display_path()
> > and hence display_name() to print the node path in the error case.
> > 
> > So, currently, this is only used in the error case when a property of a given
> > node that is required by a driver can't be found.
> 
> And in that case, the "normal" dev_err() output should be all that is
> needed here, not a "pretty printer" that might fail to document it at
> all :)

That's what the code does; PropertyGuard::required_by() uses dev_err!() to print
the string.

It's just that the functions from the C side that provide the strings are
abstracted in an implementation of core::fmt::Display for the FwNode, which is
the idiomatic way to abstract a string representations of a type.

And since Display::fmt() is fallible, we need to decide if NULL is a valid value
for fwnode_get_name() to return. The interface, i.e. struct fwnode_operations,
does not document this, but to me it seems that fwnode_get_name() returning a
NULL pointer is an error case.