Claude review: drm: Set dev->registered back to false in case of register failure

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

Patch Review

**Correctness: Good.** The patch places `dev->registered = false` at the `err_minors` label, which is reached by both error paths (`err_unload` falls through to `err_minors`). This means the flag is cleared regardless of whether the failure came from `driver->load()` or `drm_modeset_register_all()`.

Critically, `err_minors` is also the target for failures *before* `dev->registered = true` is set (lines 1075–1089), so setting it to `false` there is a no-op in those cases (it was already `false`). This is harmless and correct.

**Placement within `err_minors`:** The patch places `dev->registered = false` at the top of `err_minors`, before `remove_compat_control_link()`:

```c
err_minors:
+	dev->registered = false;
	remove_compat_control_link(dev);
```

This mirrors `drm_dev_unregister()` which sets `dev->registered = false` as its first action (line 1148). The placement is consistent and correct — clearing the flag first ensures that any code triggered during teardown (e.g., `drm_minor_unregister` → eventual mode object cleanup) sees the device as unregistered.

**Commit message quality:** The commit message is well-written. It explains the history of how the bug was introduced (the later addition of `drm_modeset_register_all()` error checking), describes the concrete symptom (spurious `WARN_ON` in `drm_mode_object_register()`), and is clear about the fix.

**One minor note on the commit message:** The message says the WARN_ON is in `drm_mode_object_register()`, but looking at the current tree, `drm_mode_object_register()` doesn't contain a WARN_ON — it's a simple `idr_replace()` under a mutex. The WARN_ON the author is likely referring to may be elsewhere (possibly in `drm_mode_object_unregister()` or in connector/framebuffer registration paths that check `dev->registered`). This doesn't affect the correctness of the fix, but the commit message could be slightly more precise about which WARN_ON is triggered.

**Missing cleanup consideration:** Looking at the `err_unload` path: when `drm_modeset_register_all()` fails, the error path calls `driver->unload()` and then falls through to `err_minors`. However, `drm_modeset_register_all()` success would be followed by `drm_panic_register()` and `drm_client_sysrq_register()` on the happy path. In the `err_unload` path, `drm_modeset_unregister_all()` is *not* called to undo the partial modeset registration — this is a pre-existing issue and out of scope for this patch, but worth noting.

**Verdict: The patch is correct and should be applied.** It's a clear, minimal fix for a real inconsistency in the error handling.

---
Generated by Claude Code Patch Reviewer