Claude review: drm/panthor: Add a new guard for our custom resume_and_get() PM helper

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

Patch Review

This patch has the most concerning changes:

**API signature change:**
```c
-static inline int panthor_device_resume_and_get(struct panthor_device *ptdev)
+static inline int panthor_device_resume_and_get(struct device *dev)
```
The function now takes `struct device *` instead of `struct panthor_device *`, to match the signature expected by the `DEFINE_GUARD_COND` mechanism (which passes through the type from the base guard `pm_runtime_active_auto`, which operates on `struct device *`). This is the right call for making the guard work, but it changes the API for all callers. The patch does update the callsites accordingly.

**Guard definition:**
```c
DEFINE_GUARD_COND(pm_runtime_active_auto, _try_enabled_or_suspend,
          panthor_device_resume_and_get(_T), _RET == 0)
```
This overloads the existing `pm_runtime_active_auto` guard with a new `_try_enabled_or_suspend` variant. The new variant calls `panthor_device_resume_and_get()` instead of the standard `pm_runtime_resume_and_get()`. On exit, it calls `pm_runtime_put_autosuspend()` (inherited from the base guard).

**Concern: `pm_runtime_mark_last_busy()` is dropped:**
The original `tick_work()` called:
```c
pm_runtime_mark_last_busy(ptdev->base.dev);
pm_runtime_put_autosuspend(ptdev->base.dev);
```
The guard only calls `pm_runtime_put_autosuspend()` on cleanup. The `pm_runtime_mark_last_busy()` call is **removed and not replaced**. This means the autosuspend timer won't be reset when the tick completes, potentially causing earlier-than-expected device suspension. The same issue applies to `queue_run_job()`. **This is a behavioral change that could cause power management regressions** — the autosuspend delay effectively starts from the last `mark_last_busy` call, and without it, the timeout may expire sooner than intended.

The commit message says "Overload the already existing pm_runtime_active_auto_try_enabled guard" but it's actually creating a new variant (`_try_enabled_or_suspend`), not overloading an existing one. The name is also panthor-specific (referencing the "or suspend" recovery behavior), but it's defined via the generic pm_runtime guard infrastructure. **This is a panthor-specific function injected into a generic guard namespace**, which may not sit well with PM maintainers. Consider defining a separate `panthor_pm` guard instead.

**Verbosity of the ACQUIRE pattern:**
```c
ACQUIRE(pm_runtime_active_auto_try_enabled_or_suspend, pm_guard)(ptdev->base.dev);
ret = ACQUIRE_ERR(pm_runtime_active_auto_try_enabled_or_suspend, &pm_guard);
```
This is significantly less readable than the original `ret = panthor_device_resume_and_get(ptdev);`. The guard name is 46 characters long. While functionally correct, this is a readability regression at the callsites. A shorter alias or a panthor-specific guard would help.

**Typo in commit message:** "force the state to suspened" should be "suspended".

---
Generated by Claude Code Patch Reviewer