Claude review: driver core: count references of the platform device's fwnode, not OF node

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

Overall Series Review

Subject: driver core: count references of the platform device's fwnode, not OF node
Author: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
Patches: 26
Reviewed: 2026-05-25T20:33:06.580208

---

This is a well-structured 23-patch series from Bartosz Golaszewski that addresses a real architectural gap in platform device firmware node refcounting. The series follows a sound methodology:

1. **Patches 1-5**: Fix existing refcount bugs where `of_node_get()` was missing
2. **Patches 6-8**: Introduce three new helper functions (`platform_device_set_of_node()`, `platform_device_set_fwnode()`, `platform_device_set_of_node_from_dev()`)
3. **Patches 9-22**: Convert all in-tree callers to use the new helpers
4. **Patch 23**: Switch from OF-only refcounting to generic fwnode refcounting

The overall design is correct. The key insight is that `platform_device_release()` currently only calls `of_node_put()`, so non-OF fwnode handles (software nodes, ACPI nodes) were never refcounted through the platform device lifecycle. By funneling all assignments through helpers first, patch 23 can atomically switch to `fwnode_handle_get()`/`fwnode_handle_put()` without breaking any caller.

The series is a prerequisite for unifying the release path for dynamic platform devices with unmanaged software nodes, which is a worthwhile goal.

**One concern**: The series depends on a function `dev_assign_of_node_reused()` (used in patch 23's `platform_device_register_full()` change) that does not exist in the current drm-next tree. This is presumably introduced in a companion series or different base. This should be clarified.

**DRM-specific impact**: Only patch 18 touches DRM code (`drivers/gpu/drm/xe/xe_i2c.c`), and it's a trivial mechanical conversion. Low risk for the DRM subsystem.

---
Generated by Claude Code Patch Reviewer