From mboxrd@z Thu Jan  1 00:00:00 1970
From: Claude Code Review Bot <claude-review@example.com>
To: dri-devel-reviews@example.com
Subject: Claude review: rust: pci: make Bar lifetime-parameterized
Date: Mon, 25 May 2026 19:29:14 +1000
Message-ID: <review-patch17-20260521233501.1191842-18-dakr@kernel.org>
In-Reply-To: <20260521233501.1191842-18-dakr@kernel.org>
References: <20260521233501.1191842-1-dakr@kernel.org> <20260521233501.1191842-18-dakr@kernel.org>
X-Mailer: Claude Code Patch Reviewer
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0

Patch Review

Converts `Bar<SIZE>` to `Bar<'a, SIZE>`, storing `&'a Device<Bound>` instea=
d of `ARef<Device>`. The `iomap_region_sized()` method now returns `Result<=
Bar<'a, SIZE>>` directly instead of `impl PinInit<Devres<Bar<SIZE>>, Error>=
`.

The notable addition is `Bar::into_devres()`:

```rust
pub fn into_devres(self) -> Result<Devres<Bar<'static, SIZE>>> {
    let bar: Bar<'static, SIZE> =3D unsafe { core::mem::transmute(self) };
    let pdev =3D bar.pdev;
    Devres::new(pdev.as_ref(), bar)
}
```

The transmute to `'static` is justified by the Devres invariant: access is =
revoked on unbind, so the Bar never actually outlives the device. The safet=
y comment is clear.

**Minor observation:** After the transmute, `bar.pdev` is read as a `&'stat=
ic Device<Bound>`. While this is fine because the Devres owns the Bar and t=
he device outlives the Devres, the code derives `pdev` from the already-tra=
nsmuted `bar` =E2=80=94 this means `pdev` is a `&'static` reference being p=
assed to `Devres::new()`. Since Devres ties the resource to the device and =
the device *does* outlive the Devres, this should be sound, but worth notin=
g for reviewers.

---

---
Generated by Claude Code Patch Reviewer