Using Raw Input from C# to handle multiple keyboards



Sample Image - rawinput.jpg

Support for different devices
Using the code
Implementing a Windows API Raw Input handler



There was a time when you were lucky if a PC had so much as a mouse, but today, it is common to have a wide variety of Human Interface Devices (HIDs) ranging from game controllers to touch screens. In particular, users can connect more than one keyboard to their PCs. However, the usual keyboard programming methods in the .NET Framework offer no way to differentiate the input from different keyboards. Any application handling KeyPress events will receive the input from all connected keyboards as if they were a single device.

Windows XP and above now support a "raw input" API which allows programs to handle the input from any connected human interface devices directly. Intercepting this information and filtering it for keyboards enables an application to identify which device triggered the message. For example, this could allow two different windows to respond to input from different keyboards.

This article and the enclosed code demonstrate how to handle raw input in order to process keystrokes and identify which device they come from. The InputDevice.cs file in the attached zip contains the raw input API wrapper; copy this file to your own project and follow the instructions in "Using the code" if you want to use the class without running the sample application.


I recently published an article on implementing a low-level keyboard hook in C#[^] using the SetWindowsHookEx and related methods from user32.dll. While looking for a solution to handle multiple keyboards, Steve Messer[^] came across my article and we discussed whether my code could be adapted to his needs. In fact, it turned out that it couldn't, and that the Raw Input API was the solution.

Unfortunately, there are very few keyboard-related Raw Input samples online, so when Steve had finished a working sample of his code, I offered to write this article so that future .NET developers faced with this problem wouldn't have to look far to find the solution. While I have made minor adjustments to the code, it is primarily Steve's work and I thank him for sharing it. Note: as of March 2007, you can also download Steve's WPF sample illustrating the use of WndProc in Windows Vista. However, this article only describes the Windows XP source code.

Please note that this will only work on Windows XP or later in a non-Terminal Server environment, and the attached sample projects are for Visual Studio 2005.

Support for different devices

The attached code is a generic solution that mostly mirrors the sample code given on MSDN. Different devices will work in different ways, and you may need to amend the code to suit the keyboards you are using. Unfortunately, we won't always be able to help with device-specific queries, as we won't have the same devices you have. Steve Messer has tested the code with different keyboards, however, and is confident that it will work with most devices provided they are correctly installed.

Using the code

All the code related to raw input handling is encapsulated in the InputDevice class, and using it is a matter of implementing three simple steps:

1. Instantiate an InputDevice object

The InputDevice class's constructor takes one argument, which is the handle to the current window.

InputDevice id = new InputDevice( Handle );

The handle is required to ensure that the window will continue to listen for events even when it doesn't have the focus.

2. Handle the KeyPressed event

When a key is pressed, the InputDevice class raises a custom KeyPressed event containing some KeyControlEventArgs. This needs to be handled by a method of the type DeviceEventHandler, which can be set up as follows:

id.KeyPressed += new InputDevice.DeviceEventHandler( m_KeyPressed );

The method that handles the event can then perform whatever actions are required based on the contents of the KeyControlEventArgs argument. The sample application attached to this article simply uses the values to populate a dialog box.

3. Override the WndProc method

In its present form, the InputDevice class works by intercepting messages to the window in order to process the WM_INPUT messages containing raw input data. The window listening for raw input will therefore need to override its own Windows procedure and pass all its messages to the instantiated InputDevice object.

protected override void WndProc( ref Message message )
    if( id != null )
    id.ProcessMessage( message );
    base.WndProc( ref message );

After writing the code used in this article, Steve decided that the InputDevice class could be truly independent from the application using it by inheriting from NativeWindow[^]. However, as the purpose of this article is primarily to illustrate the use of the Raw Input API, we decided to keep the code in its original form.

The rest of this article describes how to handle "raw input" from a C# application, as illustrated by the InputDevice class in the sample application.

Implementing a Windows API Raw Input handler

MSDN identifies "raw input" [^] as being the raw data supplied by an interface device. In the case of a keyboard, this data is normally intercepted by Windows and translated into the information provided by Key events in the .NET Framework. For example, the Windows manager translates the device-specific data about keystrokes into virtual keys.

However, the normal Windows manager doesn't provide any information about which device received the keystroke; it just bundles events from all keyboards into one category and behaves as if there were just one keyboard.

This is where the Raw Input API is useful. It allows an application to receive data directly from the device, with minimal intervention from Windows. Part of the information it provides is the identity of the device that triggered the event.

The user32.dll in Windows XP and Vista contains the following methods for handling raw input:

  • RegisterRawInputDevices allows the application to register the input devices it wants to monitor.
  • GetRawInputData retrieves the data from the input device.
  • GetRawInputDeviceList retrieves the list of input devices attached to the system.
  • GetRawInputDeviceInfo retrieves information on a device.

The following sections give an overview of how these four methods are used to process raw data from keyboards.

Registering raw input devices

By default, no application receives raw input. The first step is therefore to register the input devices that will be providing the desired raw data, and associate them with the window that will be handling this data.

To do this, the RegisterRawInputDevices method is imported from user32.dll:

    extern static bool RegisterRawInputDevices(
    RAWINPUTDEVICE[] pRawInputDevice,
    uint uiNumDevices, uint cbSize);

To determine which devices should be registered, the method accepts an array of RAWINPUTDEVICE structures. The other two arguments are the number of items in the array, and the number of bytes in a RAWINPUTDEVICE structure.

The RAWINPUTDEVICE structure is defined in Windows.h for C++ projects, but as this file isn't used in C#, the structure has been redefined as a member of the InputDevice class.

    internal struct RAWINPUTDEVICE
    public ushort usUsagePage;
    public ushort usUsage;
    public int dwFlags;
    public IntPtr hwndTarget;

Each RAWINPUTDEVICE structure added to the array contains information on a type of device which interests the application. For example, it is possible to register keyboards and telephony devices. The structure uses the following information:

  • Usage Page: The top level HID "usage page". For most HIDs, including the keyboard, this is 0x01.
  • Usage ID: A number indicating which precise type of device should be monitored. For the keyboard, this is 0x06. (A list of Usage Page and Usage ID values can be found in this MSDN article on HIDs[^])
  • Flags: These determine how the data should be handled, and whether some types should be ignored. A list of possible values is given in the MSDN article[^], and the constants they represent are defined in Windows.h (there's a copy of it here[^] if you don't already have one).
  • Target Handle: The handle of the window which will be monitoring data from this particular type of device.

In this case, we are only interested in keyboards, so the array only has one member and is set up as follows:

    rid[0].usUsagePage  = 0x01;
    rid[0].usUsage      = 0x06;
    rid[0].dwFlags      = RIDEV_INPUTSINK;
    rid[0].hwndTarget   = hwnd;

Here, the code only defines the RIDEV_INPUTSINK flag, which means that the window will always receive the input messages, even if it is no longer has the focus. This will enable two windows to respond to events from different keyboards, even though at least one of them won't be active.

With the array ready to be used, the method can be called to register the window's interest in any devices which identify themselves as keyboards:

RegisterRawInputDevices(rid, (uint)rid.Length,

Once the type of device has been registered this way, the application can begin to process the data using the GetRawInputData method described in the next section.

Retrieving and processing raw input

When the type of device is registered, the application begins to receive raw input. Whenever a registered device is used, Windows generates a WM_INPUT message containing the unprocessed data from the device.

Each window whose handle is associated with a registered device as described in the previous section must therefore check the messages it receives and take appropriate action when a WM_INPUT one is detected. In the sample application, the InputDevice class takes care of checking for WM_INPUT messages, so all the main window does is override its base WndProc method to get access to the messages, and pass any valid ones to the InputDevice object:

protected override void WndProc( ref Message message ) {
    if( id != null ) {
    id.ProcessMessage( message );
    base.WndProc( ref message );

The ProcessMessage method in InputDevice filters the messages, calling ProcessInputCommand whenever a WM_INPUT is received. Any other type of message will fall through to the call to the base WndProc, so the application will respond to other events normally.

public void ProcessMessage( Message message ) {
    switch( message.Msg ) {
    case WM_INPUT: {
    ProcessInputCommand( message );

ProcessInputCommand then uses the GetRawInputData method to retrieve the contents of the message and translate it into meaningful information.

Retrieving the information from the message

In order to process the data in WM_INPUT messages, the GetRawInputData method is imported from user32.dll:

    extern static uint GetRawInputData(IntPtr hRawInput, uint uiCommand,
    IntPtr pData, ref uint pcbSize, uint cbSizeHeader);

The method uses the following parameters:

  • hRawInput
    The handle to the RAWINPUT structure containing the data, as provided by the lParam in a WM_INPUT message.
  • uiCommand
    A flag which sets whether to retrieve the input data or the header information from the RAWINPUT structure. Possible values are RID_INPUT (0x10000003) or RID_HEADER (0x10000005) respectively.
  • pData:
    Depending on the desired result, this can be one of two things:
    • If pData is set to IntrPtr.Zero, the size of the buffer required to contain the data is returned in the pcbSize variable.
    • Otherwise, pData must be a pointer to allocated memory that can hold the RAWINPUT structure provided by the WM_INPUT message. When the method call returns, the contents of the allocated memory will be either the message's header information or input data, depending on the value of uiCommand.
  • pcbSize
    A variable that returns or specifies the size of the data pointed to by pData.
  • cbSizeHeader
    The size of a RAWINPUTHEADER structure.

In order to ensure that enough memory is allocated to store the desired information, the GetRawInputData method should first be called with pData set to IntPtr.Zero.

uint dwSize = 0;
    GetRawInputData( message.LParam, RID_INPUT,
    IntPtr.Zero, ref dwSize,
    (uint)Marshal.SizeOf( typeof( RAWINPUTHEADER )));

Following this call, the value of dwSize will correspond to the number of bytes needed to store the raw input data (as indicated by the use of the RID_INPUT flag).

It is then possible to allocate the right amount of memory; in this case, the pointer is stored in a variable called buffer.

IntPtr buffer = Marshal.AllocHGlobal( (int)dwSize );

Now that buffer points to a suitable location, GetRawInputData can be called again to populate the allocated memory with the RAWINPUT structure from the current message. If it succeeds, the method returns the size of the data it retrieved, so it is worth checking that this matches the result of the previous call before continuing.

if( GetRawInputData( message.LParam, RID_INPUT,
    buffer, ref dwSize, (uint)Marshal.SizeOf( typeof( RAWINPUTHEADER )))
    == dwSize )
    //do something with the data

Once this has been done, the contents pointed to by buffer can be marshaled into a RAWINPUT structure, which gives easy access to the data's various members, as illustrated in the following section.

RAWINPUT raw = (RAWINPUT)Marshal.PtrToStructure(
    buffer, typeof( RAWINPUT ));

Processing the data

As mentioned above, the WM_INPUT message contains raw data encapsulated in a RAWINPUT structure. As with the RAWINPUTDEVICE structure described in the previous section, this structure is redefined in the InputDevice class as follows.

    internal struct RAWINPUT
    public RAWINPUTHEADER header;
    public RAWMOUSE mouse;
    public RAWKEYBOARD keyboard;
    public RAWHID hid;

Following the second call to GetRawInputData (see previous section), the raw structure will contain the following information:

A RAWINPUTHEADER structure called header, which contains information on the message and the device that triggered it.

A second structure of type RAWKEYBOARD called keyboard. This could also be a RAWMOUSE or RAWHID structure called mouse or hid, depending on the type of device.

The RAWINPUTHEADER structure is laid out as follows:

    internal struct RAWINPUTHEADER
    public int dwType;
    public int dwSize;
    public IntPtr hDevice;
    public int wParam;

Its members return the following information:

  • dwType
    The type of raw input the message represents. The values can be RIM_TYPEHID (2), RIM_TYPEKEYBOARD (1), or RIM_TYPEMOUSE (0).
  • dwSize
    The size of all the information in the message (header and input data included).
  • hDevice
    The handle of the device which triggered the message.
  • wParam
    The wParam data from the WM_INPUT message.

The second structure will be a RAWMOUSE, a RAWKEYBOARD, or a RAWHID type. For the sake of completeness, the InputDevice class does contain definitions for RAWMOUSE and RAWHID, though it is only designed to process keyboard information.

The keyboard information is provided by a RAWKEYBOARD structure, laid out as follows.

    internal struct RAWKEYBOARD
    public ushort MakeCode;
    public ushort Flags;
    public ushort Reserved;
    public ushort VKey;
    public uint Message;
    public uint ExtraInformation;

Since the InputDevice class is only interested in keyboard input, the ProcessInputCommand method starts by checking the header to make sure this is a keyboard message before proceeding:

if( raw.header.dwType == RIM_TYPEKEYBOARD )

The next step is to filter the message to see if it is a key down event. This could just as easily be a check for a key up event; the point here is to filter the messages so that the same keystroke isn't processed for both key down and key up events.

private const int WM_KEYDOWN     = 0x0100;
    private const int WM_SYSKEYDOWN  = 0x0104;
    if (raw.keyboard.Message == WM_KEYDOWN ||
    raw.keyboard.Message == WM_SYSKEYDOWN)
    //Do something like...
      int vkey = raw.keyboard.vkey;

At this point, the InputDevice class retrieves further information about the message and the device that triggered it, and raises its custom KeyPressed event. The following sections describe how to get information on the devices.

Retrieving the list of input devices

Although this step isn't required to handle raw input, the list of input devices can be useful. The sample application retrieves a list of devices, filters it for keyboards, and then returns the number of keyboards. This is part of the information returned by the KeyControlEventArgs in the InputDevice class's KeyPressed event.

The first step is to import the necessary method from user32.dll:

    extern static uint GetRawInputDeviceList(IntPtr pRawInputDeviceList,
    ref uint uiNumDevices, uint cbSize);

The method's arguments are as follows:

  • pRawInputDeviceList: Depending on the desired result, this can be one of two things:
    • IntPtr.Zero if the purpose is only to retrieve the number of devices.
    • A pointer to an array of RAWINPUTDEVICELIST structures if the purpose of the method call is to retrieve the complete list of devices.
  • uiNumDevices: A reference to an unsigned integer to store the number of devices.
    • If the pRawInputDeviceList argument is IntPtr.Zero, then this variable will return the number of devices.
    • If the pRawInputDeviceList argument is a pointer to an array, then this variable must contain the size of the array. This allows the method to allocate memory appropriately. If uiNumDevices is less than the size of the array in this case, the method will return the size of the array, but an "insufficient buffer" error will occur and the method will fail.
  • cbSize: The size of a RAWINPUTDEVICELIST structure.

In order to ensure that the first and second arguments are correctly configured when the list of devices is required, the method should be set up in three stages.

First, it should be called with pRawInputDeviceList set to IntPtr.Zero. This will ensure that the variable in the second argument (deviceCount here) is filled with the correct number of devices. The result of this call should be checked, as an error means that the code can proceed no further.

uint deviceCount = 0;
    int dwSize = (Marshal.SizeOf( typeof( RAWINPUTDEVICELIST )));
    if( GetRawInputDeviceList( IntPtr.Zero, ref deviceCount, (uint)dwSize )
    == 0 )
    //continue retrieving the information (see below)
    //handle the error or throw an exception

Once the deviceCount variable contains the right value, the correct amount of memory can be allocated and associated with a pointer:

IntPtr pRawInputDeviceList =
    Marshal.AllocHGlobal((int)(dwSize * deviceCount ));

And the method can be called again, this time to fill the allocated memory with an array of RAWINPUTDEVICELIST structures:

GetRawInputDeviceList( pRawInputDeviceList, ref deviceCount, (uint)dwSize );

The pRawInputDeviceList data can then be converted into individual RAWINPUTDEVICELIST structures. In the example below, a for loop has been used to iterate through the devices, so i represents the position of the current device in the array.

for( int i = 0; i < deviceCount; i++ )
    new IntPtr(( pRawInputDeviceList.ToInt32() + ( dwSize * i ))),
    //do something with the information (see section on GetRawInputDeviceInfo)

When any subsequent processing is completed, the memory should be deallocated.

Marshal.FreeHGlobal( pRawInputDeviceList );

Getting information on specific devices

Once GetRawInputDeviceList has been used to retrieve an array of RAWINPUTDEVICELIST structures as well as the number of items in the array, it is possible to use GetRawInputDeviceInfo to retrieve specific information on each device.

First, the method is imported from user32.dll:

    extern static uint GetRawInputDeviceInfo(IntPtr hDevice,
    uint uiCommand, IntPtr pData, ref uint pcbSize);

Its arguments are as follows:

  • hDevice
    The device handle returned in the corresponding RAWINPUTDEVICELIST structure.
  • uiCommandA flag to set what type of data will be returned in pData. Possible values are RIDI_PREPARSEDDATA (0x20000005 - returns previously parsed data), RIDI_DEVICENAME (0x20000007 - a string containing the device name), or RIDI_DEVICEINFO (0x2000000b - an RIDI_DEVICE_INFO structure)
  • pData: Depending on the desired result, this can be one of two things:
    • If pData is set to IntrPtr.Zero, the size of the buffer required to contain the data is returned in the pcbSize variable.
    • Otherwise, pData must be a pointer to allocated memory that can hold the type of data specified by uiCommand.
      (Note: if uiCommand is set to RIDI_DEVICEINFO, then the cbSize member of the RIDI_DEVICE_INFO structure must be set to the size of the structure)
  • pcbSize
    A variable that returns or specifies the size of the data pointed to by pData. If uiCommand is RIDI_DEVICENAME, pcbSize will indicate the number of characters in the string. Otherwise, it indicates the number of bytes in the data.

The example code uses a for loop to iterate through the available devices as indicated by the deviceCount variable. At the start of each loop, a RAWINPUTDEVICELIST structure called rid is filled with the information on the current device (see GetRawInputDeviceList section above).

In order to ensure that enough memory is allocated to store the desired information, the GetRawInputDeviceInfo method should first be called with pData set to IntPtr.Zero. The handle in the hDevice parameter is provided by the rid structure containing information on the current device in the loop.

uint pcbSize = 0;
    GetRawInputDeviceInfo( rid.hDevice, RIDI_DEVICENAME, IntPtr.Zero,
    ref pcbSize );

In this example, the purpose is to find out the device name, which will be used to look up information on the device in the Registry.

Following this call, the value of pcbSize will correspond to the number of characters needed to store the device name. Once the code has checked that pcbSize is greater than 0, the appropriate amount of memory can be allocated.

IntPtr pData = Marshal.AllocHGlobal( (int)pcbSize );

And the method can be called again, this time to fill the allocated memory with the device name. The data can then be converted into a C# string for ease of use.

string deviceName;
    GetRawInputDeviceInfo( rid.hDevice, RIDI_DEVICENAME, pData, ref pcbSize );
    deviceName = (string)Marshal.PtrToStringAnsi( pData );

The list will also include "root" keyboard and mouse devices that are used for Terminal Services or Remote Desktop connections. As these don't interest us here, the following code will skip those when they are encountered in the loop.

if (deviceName.ToUpper().Contains("ROOT"))
    continue;  //Drop into next iteration of the loop

The next stage is to identify whether the enumerated device is a keyboard.

if( deviceType.Equals( "KEYBOARD" ) || deviceType.Equals( "HID" ))
    //It's a keyboard – or a USB device that could be a keyboard
      //Do something

The rest of the code then retrieves information about the device and checks the Registry to see whether the device is really a keyboard.

Reading device information from the Registry

Following the above code, deviceName will have a value similar to the following:


This string mirrors the device's entry in the Registry; parsing it therefore allows us to find the relevant Registry key, which contains further information on the device. So the first step is to break down the relevant part of the string:

// remove the \??\
    item = item.Substring( 4 );
    string[] split = item.Split( '#' );
    string id_01 = split[0];    // ACPI (Class code)
    string id_02 = split[1];    // PNP0303 (SubClass code)
    string id_03 = split[2];    // 3&13c0b0c5&0 (Protocol code)
    // The final part is the class GUID and is not needed here

The Class code, SubClass code and Protocol retrieved this way correspond to the device's path under HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet, so the next stage is to open that key:

RegistryKey OurKey = Registry.LocalMachine;
    string findme = string.Format(
    id_01, id_02, id_03 );

The information we are interested in is the device's friendly description, and its class, as this latter will tell us if it's a keyboard:

string deviceDesc  = (string)OurKey.GetValue( "DeviceDesc" );
    string deviceClass = (string)OurKey.GetValue( "Class" );
    if( deviceClass.ToUpper().Equals( "KEYBOARD" )){
    isKeyboard = true;
    isKeyboard = false;

All that is left then is to deallocate any allocated memory and do something with the data that has been retrieved.


Although the .NET Framework offers methods for most common purposes, the Raw Input API offers a more flexible approach to device data. The enclosed code and the explanations in this article will hopefully prove a useful starting point for anyone looking to handle multiple keyboards in an XP or Vista based application.


This article gives an overview of the different steps required to implement the Raw Input API. For further information on handling raw input:


  • March 2007 - Added WPF sample in response to user request
  • January 2007 - Original version

posted on 2008-10-25 11:25  starspace  阅读(5602)  评论(0编辑  收藏  举报