Claude review: drm/msm/adreno: set cx_misc_mmio regardless of if platform has LLCC

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

Patch Review

**Status: Mostly good, one observation**

This is the key architectural fix. Previously, `a6xx_llc_slices_init()` did three things:
1. Detected MMU500
2. Mapped `cx_misc_mmio`
3. Got LLCC slices
4. Set `cx_misc_mmio = ERR_PTR(-EINVAL)` if no LLCC slices available

Step 4 was the problem: on platforms like Milos that have no LLCC, `cx_misc_mmio` was clobbered with an error pointer, causing crashes when the GMU later read TCM retention or software fuses via `cx_misc_read()`.

The fix moves the MMU500 detection and `cx_misc_mmio` mapping into `a6xx_gpu_init()`, after `a6xx_llc_slices_init()`, so the MMIO mapping is independent of LLCC availability. The early-return guards in `a6xx_llc_activate()` and `a7xx_llc_activate()` are changed from checking `IS_ERR(cx_misc_mmio)` to checking whether both LLCC slices are unavailable — which is the actual condition that matters for LLC activation.

```c
-	if (IS_ERR(a6xx_gpu->cx_misc_mmio))
+	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
 		return;
```

This is correct — if there are no LLCC slices to activate, there's nothing for the LLC activation functions to do. The `cx_misc_mmio` validity is now a separate concern.

**Minor observation**: After this patch, `cx_misc_mmio` could potentially be an `ERR_PTR` if `msm_ioremap()` fails, but the LLC activate functions no longer check for that before the LLCC slice activate calls (which don't use `cx_misc_mmio`). In `a6xx_llc_activate`, the `cx_misc_write/rmw` calls happen later only in the `!have_mmu500` path, and `a7xx_llc_activate` doesn't use cx_misc at all. For A8xx, `a8xx_llc_activate` also doesn't use cx_misc. So this is safe, but callers of `a6xx_cx_misc_read/write` elsewhere (GMU code for TCM retention, fuses, speedbin) should be careful — a failed ioremap would cause a crash. In practice, if `cx_mem` is missing from the DT, the GPU would likely fail earlier, so this is acceptable.

Has `Reviewed-by` from Konrad Dybcio and Akhil P Oommen.

---

---
Generated by Claude Code Patch Reviewer