Claude review: drm/sun4i: layers: Use drm_fb_dma_get_gem_addr() to get display memory

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

Patch Review

**UI layer (`sun8i_ui_layer.c`) — Clean conversion:**

The old code:
```c
gem = drm_fb_dma_get_gem_obj(fb, 0);
bpp = fb->format->cpp[0];
dma_addr = gem->dma_addr + fb->offsets[0];
dma_addr += (state->src.x1 >> 16) * bpp;
dma_addr += (state->src.y1 >> 16) * fb->pitches[0];
```

is replaced by:
```c
dma_addr = drm_fb_dma_get_gem_addr(fb, state, 0);
```

For plane 0 with standard (non-block) formats, the helper computes `obj->dma_addr + offsets[0] + sample_x * char_per_block[0] + sample_y * pitches[0]`, which is equivalent. This conversion is correct.

**VI layer (`sun8i_vi_layer.c`) — Subtle behavioral difference:**

The old code explicitly aligns source coordinates to chroma subsampling boundaries before computing offsets:
```c
src_x = (state->src.x1 >> 16) & ~(format->hsub - 1);
src_y = (state->src.y1 >> 16) & ~(format->vsub - 1);
```

The helper does not perform this alignment — it just divides by hsub/vsub directly:
```c
sample_x = (state->src_x >> 16) / h_div;
sample_y = (state->src_y >> 16) / v_div;
```

For **plane > 0** (chroma planes), these produce the same result: integer division by hsub has the same effect as masking then dividing when the coordinates are valid (aligned to subsampling boundaries). The DRM atomic check framework should validate that source coordinates are properly aligned for subsampled formats.

For **plane 0** (luma), the old code applied the alignment mask before using the coordinates (`dx = src_x` without division), effectively rounding the luma offset down to a chroma-aligned boundary. The helper does not do this alignment for plane 0 (h_div=v_div=1). In practice, for valid state, source coordinates should already be aligned, so this is a no-op difference. If anything, the old behavior of rounding down luma coordinates was slightly wrong — it could silently shift the displayed region.

**Minor nits:**
- The `DRM_DEBUG_DRIVER("Using GEM @ %pad\n", ...)` debug prints are removed. This is fine — they add noise and the helper is well-established across many drivers. Other sun4i files (`sun4i_backend.c`, `sun4i_frontend.c`) already use this helper without such debug prints.
- The unused variable `format` (alias for `fb->format`) is left in `sun8i_vi_layer_update_buffer` — it's still used for `format->num_planes` in the loop, so that's fine.

**Verdict:** The patch is correct. The alignment difference in the VI layer is not a real-world concern since DRM enforces valid coordinates.

Reviewed-by: AI Review (informational, non-binding)

---
Generated by Claude Code Patch Reviewer