Claude review: drm/amdgpu: Switch private_obj initialization to atomic_create_state

[Claude Code Review Bot <claude-review@example.com>]

Patch Review

**Commit message**: Clear and well-written.

**Code review**:

The new `dm_atomic_create_state()` function looks correct:

```c
static struct drm_private_state *
dm_atomic_create_state(struct drm_private_obj *obj)
{
	struct amdgpu_device *adev = drm_to_adev(obj->dev);
	struct dm_atomic_state *dm_state;
	struct dc_state *context;

	dm_state = kzalloc_obj(*dm_state);
	if (!dm_state)
		return ERR_PTR(-ENOMEM);

	context = dc_state_create_current_copy(adev->dm.dc);
	if (!context) {
		kfree(dm_state);
		return ERR_PTR(-ENOMEM);
	}

	__drm_atomic_helper_private_obj_create_state(obj, &dm_state->base);
	dm_state->context = context;

	return &dm_state->base;
}
```

This is well-structured with proper error cleanup if `dc_state_create_current_copy()` fails.

**Issue: Missing return value check.** The call to `drm_atomic_private_obj_init()` now passes `NULL` as the state, which means it will internally call `atomic_create_state` — which can fail (returning `ERR_PTR(-ENOMEM)`). The return value is not checked:

```c
	drm_atomic_private_obj_init(adev_to_drm(adev),
				    &adev->dm.atomic_obj,
				    NULL,
				    &dm_atomic_state_funcs);
```

This should be something like:
```c
	ret = drm_atomic_private_obj_init(adev_to_drm(adev),
				    &adev->dm.atomic_obj,
				    NULL,
				    &dm_atomic_state_funcs);
	if (ret)
		return ret;
```

**Positive note**: The error path simplification is a nice side-effect. The old code was arguably buggy — it manually freed state on error after `drm_atomic_private_obj_init()` had already registered the object, which could lead to a use-after-free if cleanup later called `drm_atomic_private_obj_fini()`. The new code avoids that class of bug.

**Minor note (pre-existing)**: `dm_atomic_state_funcs` is not `const`, unlike the omap and tegra equivalents. Not introduced by this patch.

---

---
Generated by Claude Code Patch Reviewer