Re: [PATCH v16 2/4] rust: types: add `ForeignOwnable::PointedTo`

[Tamir Duberstein <tamird@xxxxxxxxx>]

On Mon, Feb 17, 2025 at 9:15 AM Danilo Krummrich <dakr@xxxxxxxxxx> wrote:
>
> On Mon, Feb 17, 2025 at 09:02:12AM -0500, Tamir Duberstein wrote:
> > > > diff --git a/rust/kernel/pci.rs b/rust/kernel/pci.rs
> > > > index 6c3bc14b42ad..eb25fabbff9c 100644
> > > > --- a/rust/kernel/pci.rs
> > > > +++ b/rust/kernel/pci.rs
> > > > @@ -73,6 +73,7 @@ extern "C" fn probe_callback(
> > > >          match T::probe(&mut pdev, info) {
> > > >              Ok(data) => {
> > > >                  let data = data.into_foreign();
> > > > +                let data = data.cast();
> > >
> > > Same here and below, see also [2].
> >
> > You're the maintainer,
>
> This isn't true. I'm the original author, but I'm not an official maintainer of
> this code. :)
>
> > so I'll do what you ask here as well. I did it
> > this way because it avoids shadowing the git history with this change,
> > which I thought was the dominant preference.
>
> As mentioned in [2], if you do it the other way around first the "rust: types:
> add `ForeignOwnable::PointedTo`" patch and then the conversion to cast() it's
> even cleaner and less code to change.

This is true for the two instances of `as _`, but not for all the
other instances where currently there's no cast, but one is now
needed.

> >
> > > I understand you like this style and I'm not saying it's wrong or forbidden and
> > > for code that you maintain such nits are entirely up to you as far as I'm
> > > concerned.
> > >
> > > But I also don't think there is a necessity to convert things to your preference
> > > wherever you touch existing code.
> >
> > This isn't a conversion, it's a choice made specifically to avoid
> > touching code that doesn't need to be touched (in this instance).
>
> See above.

This doesn't address my point. I claim that

@@ -246,6 +248,7 @@ impl<T: MiscDevice> VtableHelper<T> {
 ) -> c_int {
     // SAFETY: The release call of a file owns the private data.
     let private = unsafe { (*file).private_data };
+    let private = private.cast();
     // SAFETY: The release call of a file owns the private data.
     let ptr = unsafe { <T::Ptr as ForeignOwnable>::from_foreign(private) };

is a better diff than

@@ -245,7 +245,7 @@ impl<T: MiscDevice> VtableHelper<T> {
     file: *mut bindings::file,
 ) -> c_int {
     // SAFETY: The release call of a file owns the private data.
-    let private = unsafe { (*file).private_data };
+    let private = unsafe { (*file).private_data }.cast();
     // SAFETY: The release call of a file owns the private data.
     let ptr = unsafe { <T::Ptr as ForeignOwnable>::from_foreign(private) };

because it doesn't acquire the git blame on the existing line.

> >
> > > I already explicitly asked you not to do so in [3] and yet you did so while
> > > keeping my ACK. :(
> > >
> > > (Only saying the latter for reference, no need to send a new version of [3],
> > > otherwise I would have replied.)
> > >
> > > [2] https://lore.kernel.org/rust-for-linux/Z7MYNQgo28sr_4RS@cassiopeiae/
> > > [3] https://lore.kernel.org/rust-for-linux/20250213-aligned-alloc-v7-1-d2a2d0be164b@xxxxxxxxx/
> >
> > I will drop [2] and leave the `as _` casts in place to minimize
> > controversy here.
>
> As mentioned I think the conversion to cast() is great, just do it after this
> one and keep it a single line -- no controversy. :)

The code compiles either way, so I'll leave it untouched rather than
risk being scolded for sneaking unrelated changes.