Re: [PATCH 2/4] rust: devres: replace Devres::new_foreign_owned()

[Danilo Krummrich <dakr@kernel.org>]

On Sat, Jun 21, 2025 at 11:10:14PM +0200, Benno Lossin wrote:
> On Thu Jun 12, 2025 at 4:51 PM CEST, Danilo Krummrich wrote:
> > diff --git a/rust/kernel/cpufreq.rs b/rust/kernel/cpufreq.rs
> > index b0a9c6182aec..f20636079f7a 100644
> > --- a/rust/kernel/cpufreq.rs
> > +++ b/rust/kernel/cpufreq.rs
> > @@ -12,7 +12,7 @@
> >      clk::Hertz,
> >      cpumask,
> >      device::{Bound, Device},
> > -    devres::Devres,
> > +    devres,
> >      error::{code::*, from_err_ptr, from_result, to_result, Result, VTABLE_DEFAULT_ERROR},
> >      ffi::{c_char, c_ulong},
> >      prelude::*,
> > @@ -910,7 +910,7 @@ unsafe impl<T: Driver> Sync for Registration<T> {}
> >  /// thread.
> >  unsafe impl<T: Driver> Send for Registration<T> {}
> >  
> > -impl<T: Driver> Registration<T> {
> > +impl<T: Driver + 'static> Registration<T> {
> 
> This change should probably be its own patch? If not, then it should be
> mentioned in the commit message.

It's a consequence of register_foreign_boxed() requiring T: 'static.

> >      const VTABLE: bindings::cpufreq_driver = bindings::cpufreq_driver {
> >          name: Self::copy_name(T::NAME),
> >          boost_enabled: T::BOOST_ENABLED,
> > @@ -1044,10 +1044,10 @@ pub fn new() -> Result<Self> {
> >  
> >      /// Same as [`Registration::new`], but does not return a [`Registration`] instance.
> >      ///
> > -    /// Instead the [`Registration`] is owned by [`Devres`] and will be revoked / dropped, once the
> > +    /// Instead the [`Registration`] is owned by [`kernel::devres`] and will be dropped, once the
> >      /// device is detached.
> >      pub fn new_foreign_owned(dev: &Device<Bound>) -> Result {
> 
> I think we can improve the names here. How about `new_attached`? See
> more below.

I feel like the name pretty much nails it: it's a new instance that is not
owned, by the Rust side, but by the C devres implementation (i.e. foreign
owned), which automatically drops it when the device is unbound.

Maybe Registration::new_devres_owned() instead?

> > -        Devres::new_foreign_owned(dev, Self::new()?, GFP_KERNEL)
> > +        devres::register_foreign_boxed(dev, Self::new()?, GFP_KERNEL)
> >      }
> >  }
> 
> > +/// Encapsulate `data` in a [`KBox`] and [`Drop::drop`] `data` once `dev` is unbound.
> > +///
> > +/// # Examples
> > +///
> > +/// ```no_run
> > +/// use kernel::{device::{Bound, Device}, devres};
> > +///
> > +/// struct Registration;
> > +///
> > +/// impl Registration {
> > +///     fn new() -> Self {
> > +///         // register (e.g. class device, IRQ, etc.)
> > +///
> > +///         Self
> > +///     }
> > +/// }
> > +///
> > +/// impl Drop for Registration {
> > +///     fn drop(&mut self) {
> > +///        // unregister
> > +///     }
> > +/// }
> > +///
> > +/// fn from_bound_context(dev: &Device<Bound>) -> Result {
> > +///     devres::register_foreign_boxed(dev, Registration::new(), GFP_KERNEL)
> > +/// }
> > +/// ```
> > +pub fn register_foreign_boxed<T, E>(
> 
> I don't really get the name of this function. The data isn't really
> foreign and that the user also shouldn't really care about the fact that
> you use `KBox` under the hood.
> 
> How about we call this something like `attach_data`?

Hm, I think attach_data() doesn't quite hit the point. Maybe just
devres::register_owned() instead. I agree that 'boxed' is an unnecessary
implementation detail.

> ---
> Cheers,
> Benno
> 
> > +    dev: &Device<Bound>,
> > +    data: impl PinInit<T, E>,
> > +    flags: Flags,
> > +) -> Result
> > +where
> > +    T: 'static,
> > +    Error: From<E>,
> > +{
> > +    let data = KBox::pin_init(data, flags)?;
> > +
> > +    register_foreign(dev, data)
> > +}