Claude review: drm/panthor: Use a local iomem base for MMU AS registers

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

Patch Review

**Status: Good. Acked-by Boris Brezillon.**

Adds `void __iomem *iomem` to `struct panthor_mmu`, set to `ptdev->iomem + MMU_AS_BASE` (0x2400):

```c
+	mmu->iomem = ptdev->iomem + MMU_AS_BASE;
```

The `MMU_AS()` macro is rebased:
```c
-#define MMU_AS(as)			(MMU_BASE + ((as) << MMU_AS_SHIFT))
+#define MMU_AS(as)			((as) << MMU_AS_SHIFT)
```

Where `MMU_BASE` is renamed to `MMU_AS_BASE` (0x2400) and the AS-relative defines like `AS_TRANSTAB_LO`, `AS_COMMAND` etc. remain unchanged since they were already relative to their address space slot.

The MMU_INT IRQ iomem is set to `ptdev->iomem + MMU_INT_BASE` (0x2000). Correct.

No issues.

---

## SUMMARY

| Patch | Verdict | Notes |
|-------|---------|-------|
| 1/8 | Clean | Mechanical refactor of helper signatures |
| 2/8 | Clean | Header split, no logic changes |
| 3/8 | Clean | Good encapsulation of cross-component accesses |
| 4/8 | Clean | Minor: generic `INT_*` names could use prefix |
| 5/8 | Clean | GPU iomem localization |
| 6/8 | Clean | Pre-existing u64→u32 truncation in `panthor_pwr_l2_power_on()` |
| 7/8 | Clean | FW iomem localization |
| 8/8 | Clean | MMU AS iomem localization |

**Reviewed-by recommendation:** The series is sound. No functional regressions introduced. The register arithmetic is verified correct across all patches. The two observations (generic `INT_*` names, pre-existing u32 truncation) are non-blocking.

---
Generated by Claude Code Patch Reviewer