Claude review: drm/rockchip: vop2: Support setting custom background color

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

Patch Review

**Status: Mostly good, one concern**

The register programming using `FIELD_PREP`/`FIELD_MODIFY` with the BPCS (shift-based) helpers is clean:

```c
+	val = FIELD_PREP(RK3568_VP_DSP_BG__DSP_BG_RED, DRM_ARGB64_GETR_BPCS(bgcolor, 10));
+	FIELD_MODIFY(RK3568_VP_DSP_BG__DSP_BG_GREEN, &val, DRM_ARGB64_GETG_BPCS(bgcolor, 10));
+	FIELD_MODIFY(RK3568_VP_DSP_BG__DSP_BG_BLUE, &val, DRM_ARGB64_GETB_BPCS(bgcolor, 10));
```

The register field definitions (three 10-bit fields in bits 29-0) look correct for VOP2 documentation. Using BPCS (shift instead of divide) is justified for a hot path like `vop2_post_config()`.

**Concern — alpha-blending rejection check:**

```c
+	if ((cstate->background_color << 16) &&
+	    (fb->format->has_alpha || pstate->alpha != 0xffff)) {
+		drm_dbg_kms(vop2->drm,
+			    "Alpha-blending with background color is unsupported\n");
+		return -EINVAL;
+	}
```

The `<< 16` trick to strip the alpha and check if RGB is non-zero is clever but somewhat obscure — a named helper or a mask like `bgcolor & 0x0000FFFFFFFFFFFF` would be clearer.

More substantively, this check may be **overly conservative**: it rejects any plane that has alpha (format or property) whenever the background is non-black, regardless of whether that plane's pixels actually interact with the background. For example, if plane A is a fully opaque bottom layer covering the entire screen, and plane B above it has `has_alpha`, the check would reject plane B even though plane B blends with plane A, not the background. I understand this is a hardware limitation being enforced conservatively, but it could break valid compositing scenarios with non-black backgrounds. Is the hardware truly incapable of handling this, or is it just that the background-to-first-plane blending is problematic? If the latter, this could be refined to only reject when the lowest z-order plane has alpha.

The state dump addition is a nice touch for debugging:

```c
+	seq_printf(s, "\tbackground color (10bpc): r=0x%x g=0x%x b=0x%x\n",
```

R-b from Andy Yan (Rockchip maintainer) is appropriate.

---
Generated by Claude Code Patch Reviewer