Claude review: drm/msm/dp: turn link_ready into plugged

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

Patch Review

Replaces the public `link_ready` flag with a private `plugged` boolean protected by `plugged_lock` mutex. Also removes `msm_dp_bridge_atomic_check()` which returned `-ENOTCONN` when link was not ready.

**Issues:**

1. **msm_dp_bridge_get_modes logic broken:** The patch changes:
```c
-	if (dp->link_ready) {
-		rc = msm_dp_display_get_modes(dp);
-		if (rc <= 0) {
-			DRM_ERROR("failed to get DP sink modes, rc=%d\n", rc);
-			return rc;
-		}
-	} else {
-		drm_dbg_dp(connector->dev, "No sink connected\n");
-	}
+	rc = msm_dp_display_get_modes(dp);
+	if (rc <= 0) {
+		DRM_ERROR("failed to get DP sink modes, rc=%d\n", rc);
+		return rc;
+	} else {
+		drm_dbg_dp(connector->dev, "No sink connected\n");
+	}
```
The `else` branch is now unreachable — if `rc <= 0` the function returns, and if `rc > 0` it falls into the `else` printing "No sink connected" which is wrong (it means modes were successfully found). This should be `if (rc > 0)` to reach the else, or the else should be removed.

2. **plug_handle unconditional pm_runtime_get:** After removing the `link_ready` guard:
```c
-	if (dp->msm_dp_display.link_ready)
-		return 0;
+	mutex_lock(&dp->plugged_lock);
```
Now `msm_dp_hpd_plug_handle()` always calls `pm_runtime_resume_and_get()` even if already plugged. On repeated plug events, this will increment the pm_runtime refcount without matching puts. The old code had `if (link_ready) return 0` which prevented this.

3. **Error handling regression in plug_handle:** After the rework:
```c
	ret = msm_dp_display_process_hpd_high(dp);
	...
	dp->plugged = true;
	mutex_unlock(&dp->plugged_lock);
	return ret;
```
Even on failure (`ret != 0`), `dp->plugged` is set to `true`. This means a failed plug will leave the device in a state where it thinks it's plugged, preventing future plug attempts from working and leaving a stale pm_runtime ref.

4. **Removal of atomic_check:** Removing `msm_dp_bridge_atomic_check()` that returned `-ENOTCONN` when link was down means the DRM core can now attempt to enable a disconnected display. The commit message mentions "After removing the link_ready flag, DP bridge doesn't have a distinction between link trained and display plugged in" but doesn't explain how the "display enabled while disconnected" case is handled. The `sink_count == 0` check in `atomic_enable()` returns early but doesn't prevent the mode set.

5. **detect() leaves device powered if plugged:** The new detect logic calls `pm_runtime_resume_and_get()` and only calls `pm_runtime_put_sync()` if `!priv->plugged`. This means every detect() call that finds the device plugged increments pm_runtime without a put. Multiple detect() calls would stack refs. The comment says "leave the controller on so that it doesn't lose the state" but this ref stacking is problematic.

6. **Typo:** "loose" should be "lose" in:
```c
+	 * If we detected the DPRX, leave the controller on so that it doesn't
+	 * loose the state.
```

---
Generated by Claude Code Patch Reviewer