Claude review: drm/xe/i2c: use device_create_managed_software_node()

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

Patch Review

**Clean and correct.** This converts xe_i2c from manually creating a software node as the primary fwnode to using `device_create_managed_software_node()`, which ties the software node's lifecycle to the device.

Key changes:
- `fwnode_create_software_node()` + `pdev->dev.fwnode = fwnode` is replaced by `device_create_managed_software_node(&pdev->dev, ...)` called after `platform_device_alloc()` but before `platform_device_add()`.
- The explicit `fwnode_remove_software_node()` in `xe_i2c_unregister_adapter()` is correctly removed since the managed node self-cleans.
- The `adapter_node` field is correctly removed from `struct xe_i2c`.

The comment explaining why `platform_device_register_full()` isn't used (need the pdev handle before registration for `xe_i2c_notifier()`) is retained, which is good.

Semantically, `device_create_managed_software_node()` called when `dev->fwnode` is NULL will set the software node as the primary fwnode (via `set_secondary_fwnode()` logic), matching the old behavior.

The error path is also improved: on failure of `device_create_managed_software_node()`, we jump to `err_pdev_put` which calls `platform_device_put()`. The managed node will be properly cleaned up through the release path (especially after patch 4 unifies it).

**Verdict**: Ack recommended for the DRM side.

---

---
Generated by Claude Code Patch Reviewer