AnyPS5/docs/dev/debt/libSceMouse.md

libSceMouse technical debt

Unknown function info

• sceMouseGetDeviceInfo, sceMouseSetHandType, sceMouseSetPointerSpeed and the extension calls (libSceMouse) - argument lists and checks follow @BunnyFl's firmware 13.42 disassembly of libSceMouse.sprx (SHA-256 e07cd04766ea48f5216e40fd4397758f640a7200cfa7edb6a00255326bd33c61), not a console. Past their checks they throw:

  • GetDeviceInfo(int32, void*), SetHandType(handle, uint32), SetPointerSpeed(handle, uint32), ExtensionSetCursor(handle, int32) and ExtensionGetCursor(handle, void*) end in a mouse driver ioctl, which isn't modelled. GetDeviceInfo's first check, a closed device fd, is taken as not initialised here.
  • ExtensionRead(handle, void*, uint32 size, int32 num) and ExtensionReadState(handle, void*, uint32 size) run sceMouseRead's checks with the extension flag, after their size checks (0x30 and 0x40) and ReadState's 64-byte clear; the 0x30-byte extension record and the state the slot keeps at +0x100 are unknown.

• sceMouseDisconnectAll, sceMouseVirtualDeviceAddDevice, sceMouseVirtualDeviceDeleteDevice and sceMouseVirtualDeviceGetCursor - the names and NIDs are recorded in the 13.60 export inventory. Signatures and behavior remain unknown; the parameterless int declarations only resolve imports and every call throws. Not measured on a console.

• sceMouseConnectPort, sceMouseDebugGetDeviceId, sceMouseDeviceOpen, sceMouseDisconnectDevice, sceMouseDisconnectPort, sceMouseMbusInit and sceMouseSetProcessPrivilege (libSceMouse) - names and NIDs are present in the PS5 3.20 export inventory; the no-argument prototypes follow shadPS4's PS4 stubs, not measured on a console. These declarations have no parameters solely to resolve known imports and make no ABI claim.

Functional

• sceMouseGetHandle, sceMouseExtensionSetConfig and sceMouseExtensionScalePosition (libSceMouse) - follow the same 13.42 disassembly, not a console; none reaches the driver.

  • GetHandle(userId, type, index) returns the handle sceMouseOpen gave that user and type, or 0x80df0008. The console skips slots whose +0xc8 field is 1 when index is non-zero; that field's meaning is unknown, and no slot here has it.
  • ExtensionSetConfig(handle, config) refuses a null or a non-zero byte in 5..0x1f, then stores the u32 at 0 and the byte at 4; nothing here reads them yet.
  • ExtensionScalePosition(uint16* position, uint16 scale) rounds *position * (scale - 1) / 65535 up. The console divides with a multiply by 0x80008001 and a shift by 47, which is the floor, and adds 1 when the remainder isn't 0, through a compare of -(product * 65537) with 0x10002 (@BunnyFl's emulation).

• Publish (libSceMouse) - consecutive motion-only reports with the same buttons are summed into the newest unread entry, so a title that reads a few entries per frame gets current motion instead of a full queue of old reports. Reports with no motion and no button, wheel, tilt or connection change are dropped. Per-report granularity and the console's own report rate are not reproduced

• sceMouseResetAbsolutePosition, sceMouseSetAbsoluteVirtualScreenSize, sceMouseSetAbsoluteHighPrecisionMode, sceMouseReadAbsolutePosition (libSceMouse) - follow firmware 13.42 run natively with the driver stubbed: checks and order, screen sizes and rescale, the mode byte, the curve with its fraction, the 40-byte record. Host choices, after XatiX's title-tested version: each host event is one sample with unscaled counts, in its own queue, so sceMouseRead's motion merging does not feed the curve. A read takes all pending samples; past num, same-button neighbours share a record, each still moving the pointer, and the last num records are kept. Speed is taken as 4, the hand swap and flagged bit never apply, and a disconnected mouse stands in for a zero driver status