The mask 0x1000 is incorrect for matching VBE_DISPI_CFG_ID values. The VBE_DISPI_CFG_ID constants range from 0x0000 to 0x0004, so masking with 0x1000 will always result in 0x0000 or 0x1000, never matching cases VBE_DISPI_CFG_ID_VRAM_SIZE (0x0001), VBE_DISPI_CFG_ID_3D (0x0002), VBE_DISPI_CFG_ID_VMSVGA (0x0003), or VBE_DISPI_CFG_ID_VMSVGA_DX (0x0004).

The mask should likely be 0xFFFF or a value that preserves the lower bits. Looking at VirtualBox's implementation, this should read the value without masking, or mask with 0xFFFF to get the full 16-bit value.

This 0x1000 comes from VBE_DISPI_CFG_MASK_ID as is defined here in the VBox source. It isn't correct, either way, it should be 0x0FFF (I got 0x1000 from

#define VBE_DISPI_CFG_MASK_SUPPORT      0x1000 /* Query whether the identifier is supported. */

By mistake)

The return value should be cast to uint16_t or the function should handle potential overflow. The dev->vram_size is a uint32_t (can be up to 512 MB based on config options), but ret is a uint16_t. This will truncate the value, returning incorrect VRAM size information.

For example, 512 MB (0x20000000) truncated to uint16_t becomes 0x0000. The correct approach would be to either:

1. Return the full 32-bit value if the protocol supports it, or
2. Return VRAM size in a different unit (e.g., MB or 64K blocks) that fits in 16 bits

VBox Doesn't do this in this register.

Then either VBox has a bug or there's some other difference going on here.

VirtualBox does VRAM size return in the dword handler, so you need an inl() handler in 86Box - 86Box uses separate handlers for byte, word, and dword.

Indeed I did overlook what Copilot said before,
ret in 86Box is uint16_t, while val in VBox is uint32_t
Both of the ->vram_size in both of VBox & 86Box are both uint32_t
Will fix as soon as I can.

The VBE_DISPI_ID_HGSMI (0xBE01) constant is defined at line 77 but is missing from the ID validation check. The condition checks for VBE_DISPI_ID_VBOX_VIDEO (0xBE00), VBE_DISPI_ID_ANYX (0xBE02), and VBE_DISPI_ID_CFG (0xBE03), but omits VBE_DISPI_ID_HGSMI which is part of the VirtualBox VGA device ID sequence.

Add || (val == VBE_DISPI_ID_HGSMI) to the condition to allow this ID to be set.

Not implemented, adding it for XP makes it BSOD.

The VBE_DISPI_INDEX_CFG register needs a write handler to store the configuration ID value. Currently, writes to this register fall through to the default case which does nothing. The guest driver writes the config ID (e.g., VBE_DISPI_CFG_ID_VRAM_SIZE) to this register, then reads it back to get the corresponding value.

Add a case handler:

case VBE_DISPI_INDEX_CFG:
    dev->vbe_regs[dev->vbe_index] = val;
    break;

Fixed it by adding all registers to the write list:

            default:
                dev->vbe_regs[dev->vbe_index] = val;
                break;

[nitpick] These comments have extra indentation that is inconsistent with the rest of the code. The comments should be aligned with the code at the same level (no leading spaces before the /*). This appears to be a formatting error introduced during the refactor.

clang-format issue

[nitpick] Binary literal 0b11100011 should be written as 0xE3 for consistency. Binary literals are not commonly used in this codebase, and using hexadecimal notation is more conventional for bit masks in hardware emulation code.

[nitpick] Missing space around the * operator. Should be CVT_M_FACTOR * CVT_K_FACTOR / 256 for consistency with the spacing in the line below and general C style conventions.

clang-format issue