Claude review: dma: coherent: register to coherent heap

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

Patch Review

Adds a static array `rmem_coherent_areas[MAX_COHERENT_REGIONS]` in `kernel/dma/coherent.c` to record reserved regions during early boot, then exposes them via `dma_coherent_get_reserved_region()`.

**Issues:**

1. **Fixed-size static array of 64 entries is a design concern.** The `MAX_COHERENT_REGIONS` of 64 is hardcoded with no Kconfig tunability:
```c
+#define MAX_COHERENT_REGIONS 64
+static struct reserved_mem *rmem_coherent_areas[MAX_COHERENT_REGIONS];
```
This wastes memory when unused (most systems have 0-2 coherent regions) and is an arbitrary limit. A dynamic list or leveraging the existing reserved_mem infrastructure would be cleaner. This is similar to how `cma_heap.c` uses `MAX_CMA_AREAS`, but `MAX_CMA_AREAS` is a well-known kernel-wide constant, while `MAX_COHERENT_REGIONS` is newly invented here.

2. **`rmem_coherent_insert_area()` returns `-EINVAL` on overflow** but `-ENOSPC` or `-ENOMEM` would be more semantically correct.

3. **The array is not `__initdata`.** Unlike the CMA heap's `dma_areas` which is `__initdata` (freed after init), `rmem_coherent_areas` persists forever. Since `dma_coherent_get_reserved_region()` is called from `module_init` (which could be called after `__init` section is freed if built as a module), this is actually necessary, but worth noting the permanent memory cost.

4. **`IS_ENABLED(CONFIG_DMABUF_HEAPS_COHERENT)` guard in `rmem_dma_setup()`:**
```c
+	if (IS_ENABLED(CONFIG_DMABUF_HEAPS_COHERENT)) {
```
This is `__init` code. If the coherent heap is built as a module (`=m`), `IS_ENABLED()` returns true, so regions are collected. Good -- this is the correct behavior to support the modular case.

---
Generated by Claude Code Patch Reviewer