Re: [PATCH v8 4/6] rust: implement Class for ww_class support

["Onur Özkan" <work@onurozkan.dev>]

On Wed, 3 Dec 2025 13:10:44 +0000
Alice Ryhl <aliceryhl@google.com> wrote:

> On Mon, Dec 01, 2025 at 01:28:53PM +0300, Onur Özkan wrote:
> > Adds the Class type, the first step in supporting
> > ww_mutex in Rust. Class represents ww_class, used
> > for deadlock avoidance for supporting both wait-die
> > and wound-wait semantics.
> > 
> > Also adds the define_class macro for safely declaring
> > static instances.
> 
> > +impl Class {
> > +    /// Creates an unpinned [`Class`].
> > +    ///
> > +    /// # Safety
> > +    ///
> > +    /// Caller must guarantee that the returned value is not moved
> > after creation.
> 
> The value is moved when you return it. Perhaps you meant that it must
> be pinned before first use?
> 

Yes, that was my point.

> > +                            // TODO: Replace with
> > `bindings::lock_class_key::default()` once
> > +                            // stabilized for `const`.
> > +                            //
> > +                            // SAFETY: This is always
> > zero-initialized when defined with
> > +                            // `DEFINE_WD_CLASS` globally on C
> > side.
> > +                            //
> > +                            // For reference, see
> > __WW_CLASS_INITIALIZER() in
> > +                            // "include/linux/ww_mutex.h".
> > +                            acquire_key: core::mem::zeroed(),
> > +                            mutex_key: core::mem::zeroed(),
> 
> For global lock class keys, this is fine, but this constructor seems
> to be for the non-global case. In that case, lockdep_register_key()
> must be called.
> 
> Is a ww_class ever created as a non-global? I don't really think so.
> If not, then let's not support it.
> 
> Alice

No, it's always used as global on the C side. Good point, will make the
change in the next version.

-Onur