Claude review: drm: verisilicon: add support for cursor planes

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

Patch Review

**Bug: Missing `FIELD_PREP`/shift for X_OFF and Y_OFF values**

When writing negative offsets to the CONFIG register:

```c
regmap_update_bits(dc->regs, VSDC_CURSOR_CONFIG(output),
                   VSDC_CURSOR_CONFIG_X_OFF_MASK,
                   -state->crtc_x);
```

The X_OFF field is at bits [20:16] (`GENMASK(20, 16)`), but the raw value `-state->crtc_x` is not shifted. It should use the defined `VSDC_CURSOR_CONFIG_X_OFF(v)` macro:

```c
VSDC_CURSOR_CONFIG_X_OFF(-state->crtc_x)
```

Same issue for Y_OFF at bits [12:8]:

```c
regmap_update_bits(dc->regs, VSDC_CURSOR_CONFIG(output),
                   VSDC_CURSOR_CONFIG_Y_OFF_MASK,
                   -state->crtc_y);
```

Should be `VSDC_CURSOR_CONFIG_Y_OFF(-state->crtc_y)`. Without these shifts, the offset value lands in the wrong bit positions (bits [4:0] instead of [20:16] for X, and [4:0] instead of [12:8] for Y), corrupting other fields in the CONFIG register.

**Issue: `mode_config.cursor_width/height` not set**

The DRM core exposes cursor size limits to userspace via `drm_mode_config.cursor_width` and `cursor_height`. These are not set anywhere in the driver (checked `vs_drm.c`). Without this, userspace won't know the maximum cursor size. This should be set in the mode_config init, e.g.:

```c
drm_dev->mode_config.cursor_width = dc->identity.max_cursor_size;
drm_dev->mode_config.cursor_height = dc->identity.max_cursor_size;
```

**Nit: Typos in register header**

```c
#define VSDC_CURSOR_BACKGRUOND_DEFAULT  0x00FFFFFF
#define VSDC_CURSOR_FOREGRUOND_DEFAULT  0x00AAAAAA
```

`BACKGRUOND` and `FOREGRUOND` should be `BACKGROUND` and `FOREGROUND`. These macros are currently unused but should be fixed before they get used.

**Nit: Comment typo**

```c
 * To for positive left-top  coordinates the
```

Should read "For positive left-top coordinates the" (remove "To", fix double space).

**Nit: `atomic_update` calls `atomic_disable` directly**

```c
if (!state->visible) {
    vs_cursor_plane_atomic_disable(plane, atomic_state);
    return;
}
```

This works but is unusual — normally `atomic_disable` is only called by the DRM core. The DRM atomic helper should call `atomic_disable` when the plane becomes invisible, so this explicit call may be redundant. Worth verifying that the DRM helpers handle this transition correctly for cursor planes, or if this is an intentional workaround.

**Style: `ERR_PTR(PTR_ERR(primary))` vs `ERR_CAST`**

In the existing code at `vs_crtc.c:175`:
```c
return ERR_PTR(PTR_ERR(primary));
```

The new cursor error path correctly uses `ERR_CAST(cursor)`. The pre-existing primary path should also use `ERR_CAST` for consistency, though that's outside the scope of this series.

---
Generated by Claude Code Patch Reviewer