Claude review: drm/loongson: clean up KMS polling on probe failure

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

Patch Review

**The bug is real, the fix is suboptimal.**

The commit message correctly identifies that `drm_kms_helper_poll_init()` at line 296 of `lsdc_drv.c` starts a polling timer that is never cleaned up on failure or removal. However:

1. **Prefer `drmm_kms_helper_poll_init()` instead.** The driver already uses managed DRM resources (`devm_drm_dev_alloc`, `drmm_mode_config_init`, `drmm_add_action_or_reset` for TTM/PLL/I2C). The idiomatic fix is a one-liner — change:

```c
drm_kms_helper_poll_init(ddev);
```
to:
```c
drmm_kms_helper_poll_init(ddev);
```

This automatically handles cleanup on both error paths and device removal, with no `goto` labels or manual `poll_fini` calls needed. The managed variant exists specifically for this purpose (see `drm_probe_helper.c:963`).

2. **Remove-path ordering is wrong.** The patch adds:
```c
+	drm_kms_helper_poll_fini(ddev);
 	drm_dev_unregister(ddev);
```
Calling `poll_fini` before `drm_dev_unregister` means the polling work is cancelled while the device is still registered with userspace. The conventional order (used by verisilicon and others) is `drm_dev_unregister` first, then `drm_kms_helper_poll_fini`. With the `drmm_` approach this ordering question goes away entirely.

3. **The `err_poll_fini` label itself is fine** in terms of correctness for the error paths — all three failure points (vblank init, IRQ request, dev register) happen after `poll_init`, so jumping to the cleanup label is correct. But again, this whole error-path dance is unnecessary with the managed API.

**Recommendation:** Request a v2 that simply switches from `drm_kms_helper_poll_init()` to `drmm_kms_helper_poll_init()` and drops the `err_poll_fini` label and the explicit `poll_fini` in remove. This is a smaller, cleaner, and more correct fix.

---
Generated by Claude Code Patch Reviewer