Claude review: drm/panfrost: Explicitly enable MMU interrupts at device init

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

Patch Review

**Verdict: Correct, follows naturally from patch 6.**

```c
int panfrost_mmu_init(struct panfrost_device *pfdev)
{
    ...
    panfrost_mmu_enable_interrupts(pfdev);
    return 0;
}
```

Since patch 6 ensures the device starts active (with PM get/put around probe), there's a window between `drm_dev_register()` and when autosuspend triggers the first `runtime_suspend`/`runtime_resume` cycle. During this window, jobs could be submitted but MMU interrupts weren't enabled (they're only enabled in `panfrost_mmu_reset()`, called from `panfrost_device_reset()`, called from `runtime_resume()`). This patch closes that gap.

The refactoring of `mmu_write` calls into `panfrost_mmu_enable_interrupts()` is clean and used in both `panfrost_mmu_reset()` and `panfrost_mmu_init()`.

---

## SUMMARY OF ISSUES

1. **Patch 5 — Bug:** `ret = EAGAIN` must be `ret = -EAGAIN` (line ~988 in the mbox)
2. **Patch 5 — Bug:** `err_disable` error path leaks the AS reference from `panfrost_mmu_as_get()` — needs `panfrost_mmu_as_put()` before falling through to `err_vunmap`
3. **Patch 4 — Note:** Register write ordering changed in `disable_locked` (enable regs vs config reg) — verify HW is order-agnostic
4. **Patch 5 — Design:** `panfrost_perfcnt_reset()` writes GPU registers without locking — acceptable given the subsequent soft reset, but should be explicitly noted

---
Generated by Claude Code Patch Reviewer