Claude review: drm/mediatek: Set dedicated DMA device and drop custom GEM callbacks

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

Patch Review

This is the main payoff patch, deleting the entire `mtk_gem.c` (231 lines) and `mtk_gem.h` (17 lines) files. The key changes:

1. Replaces the per-private-data `dma_dev` loop with a single `drm_dev_set_dma_dev(drm, dma_dev)` call:
```c
-	for (i = 0; i < private->data->mmsys_dev_num; i++)
-		private->all_drm_private[i]->dma_dev = dma_dev;
+	drm_dev_set_dma_dev(drm, dma_dev);
```

2. Replaces custom driver ops with `DRM_GEM_DMA_DRIVER_OPS`:
```c
-	.dumb_create = mtk_gem_dumb_create,
+	DRM_GEM_DMA_DRIVER_OPS,
 	DRM_FBDEV_DMA_DRIVER_OPS,
-
-	.gem_prime_import = mtk_gem_prime_import,
-	.gem_prime_import_sg_table = mtk_gem_prime_import_sg_table,
```

**Minor observation**: The old `mtk_gem_dumb_create` had a manual pitch calculation with `DIV_ROUND_UP(args->width * args->bpp, 8)` and explicit overflow-safe size computation (`args->size = args->pitch; args->size *= args->height`). The new `drm_gem_dma_dumb_create` uses `drm_mode_size_dumb()` which should handle this equivalently. This is fine.

**Minor observation**: As the commit message notes, the old mmap helper set `VM_IO | VM_DONTDUMP` flags that the standard GEM DMA mmap helper does not set. The commit message correctly argues that `VM_IO` doesn't make sense for `dma_alloc_attrs()` buffers. `VM_DONTDUMP` absence means these mappings will now appear in core dumps, which is a very minor behavioral change but acceptable.

No blocking issues.

---
Generated by Claude Code Patch Reviewer