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: drm/bridge: cdns-dsi: Replace deprecated UNIVERSAL_DEV_PM_OPS()
Date: Sun, 12 Apr 2026 13:22:18 +1000
Message-ID: <review-patch1-20260407144142.1420354-2-ivitro@gmail.com>
In-Reply-To: <20260407144142.1420354-2-ivitro@gmail.com>
References: <20260407144142.1420354-2-ivitro@gmail.com> <20260407144142.1420354-2-ivitro@gmail.com>
X-Mailer: Claude Code Patch Reviewer
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0

Patch Review

**Correctness: Good**

The core change is replacing:
```c
static UNIVERSAL_DEV_PM_OPS(cdns_dsi_pm_ops, cdns_dsi_suspend, cdns_dsi_res=
ume, NULL);
```
with:
```c
static const struct dev_pm_ops cdns_dsi_pm_ops =3D {
    RUNTIME_PM_OPS(cdns_dsi_suspend, cdns_dsi_resume, NULL)
};
```

This is correct because:

1. `UNIVERSAL_DEV_PM_OPS` (pm.h:458-462) expands to both `SET_SYSTEM_SLEEP_=
PM_OPS` and `SET_RUNTIME_PM_OPS` using the *same* callbacks, which is exact=
ly what causes the double-disable bug described in the commit message.

2. The bridge's `atomic_pre_enable` calls `pm_runtime_get_sync()` (line 736=
) and `atomic_post_disable` calls `pm_runtime_put()` (line 631). This means=
 the DRM framework already handles bringing the device in/out of runtime PM=
 around display usage. During system suspend, the DRM core will call `atomi=
c_post_disable` which triggers `pm_runtime_put` =E2=86=92 runtime suspend =
=E2=86=92 `cdns_dsi_suspend`. There is no need for separate system sleep PM=
 ops.

3. The `__maybe_unused` removal from `cdns_dsi_resume` and `cdns_dsi_suspen=
d` is correct. The old `SET_RUNTIME_PM_OPS` macro was conditionally compile=
d (guarded by `#ifdef CONFIG_PM`), which could leave the functions unrefere=
nced when PM was disabled, requiring `__maybe_unused`. The new `RUNTIME_PM_=
OPS` macro (pm.h:338-341) unconditionally expands regardless of CONFIG_PM, =
so the functions are always referenced. The `pm_ptr()` wrapper on the drive=
r side handles the compile-out when CONFIG_PM is disabled.

4. The change from `.pm =3D &cdns_dsi_pm_ops` to `.pm =3D pm_ptr(&cdns_dsi_=
pm_ops)` is the correct modern pattern =E2=80=94 it compiles out the PM ops=
 pointer when `CONFIG_PM` is not enabled, allowing the compiler to dead-cod=
e-eliminate the entire `cdns_dsi_pm_ops` structure and the suspend/resume f=
unctions.

**Minor note on commit message:** The commit message text says "SET_RUNTIME=
_PM_OPS()" but the actual code uses `RUNTIME_PM_OPS()`. These are functiona=
lly equivalent (`SET_RUNTIME_PM_OPS` is just an alias per pm.h:365-366), bu=
t ideally the commit message would match the code. This is a very minor nit=
 and not worth respinning the patch for, especially since it already has a =
Reviewed-by.

**No issues found. Patch is ready to merge.**

---
Generated by Claude Code Patch Reviewer