ADC Home > Reference Library > Technical Notes > Legacy Documents > Hardware & Drivers >

Legacy Documentclose button

Important: This document is part of the Legacy section of the ADC Reference Library. This information should not be used for new development.

Current information on this Reference Library topic can be found here:

ADB - The Untold Story: Space Aliens Ate My Mouse


This Technical Note explains a number of esoteric and unknown issues concerning the Apple Desktop Bus (ADB). It is intended to detail information concerning all levels of the ADB, from the hardware to the application usage level. This note is supplementary to the information in The Guide To Macintosh Family Hardware and in the Apple Desktop Bus Specification. The information in the specification is the most accurate source of data, and unless it is specifically refuted, it should be treated as the authoritative source.

[Sep 28 1998]

The ADB Hardware

The ADB is a simple serial bus with collision detection. A common implementation platform for the protocol is a simple microcontroller: a 2-MHz Motorola 68HC11 has been used with success, although any number of microcontrollers would do a more than adequate job. The 2-MHz clock seemed to be close to minimal for this application; the bus has a relatively low bandwidth, and, given the packet structure, the theoretical maximum bandwidth is on the order of 100 - 200 bytes/second. In practice, the bus is not suitable for data- xtransfer applications; it serves well as a general purpose input device bus, but high data rates will not work well.

In general, the timing for the ADB is fairly tolerant of small variances; however, many Macintoshes are more demanding than their predecessors. For this reason, devices should be tested on a wide range of machines; the best test for compliance is testing. Specifically, the Tlt (stop to start time) parameter has become much less tolerant; fairly close adherence to its 200-us timing is important. In particular, devices which respond more rapidly than the 140-us minimum delay will often fail to initialize properly.

It is important that your device be fairly tolerant of problems on the bus; if a command packet begins but never seems to complete, your device should time out rather than hang. The design of the bus and its connectors means that there can be occasional glitches in the connection with the host, and you should try to be as tolerant of these as possible.

Because the ADB bus is open collector, collisions can be detected when a device is attempting to drive the bus high and another device pulls it low. This means that whenever the device is driving the bus high, it should be watching to make sure the bus is actually high; if the bus goes low, some other device is sending at the same time. When a device detects a collision, it should immediately stop transmitting: this means that if two devices are colliding, one of them will detect the collision, while one will not. This occurs because a device can only detect the collision if it is driving the bus high and another device drives it low. The device driving it low has no way to tell that there was a collision, as the bus follows it. Since the detecting device immediately stops transmitting, the other device will not detect the collision. Thus, if there are a number of devices transmitting on the bus, only one of them will complete its transmission without detecting the collision, unless the unlikely occurrence of more than one device transmitting exactly the same data with the same timing occurs, and neither detects the collision.

The ADB is not particularly tolerant of devices being connected and disconnected while the bus is live. There isn't any software architecture to detect the presence of new devices or the absence of old ones; furthermore, on some CPUs, the motherboard is not well protected from voltage transients on the ADB connector. Plugging in a device while the Macintosh is powered on could cause damage to the ADB transceiver or other portions of the Macintosh's circuitry.

On some portable Macintoshes, the voltage characteristics do not adhere to the ADB specification. Clearly, power is very valuable on all portable computers, so if your device is targeted towards portable use, you should be extremely careful to keep your power consumption as low as possible. While the ADB can supply the full voltage stated in the specification, drawing this much power will lead to much more rapid battery draining. On the PowerBook 140 and 170, there was a specific problem which caused the low-level input voltage to go above the specified maximum of 0.8 volts: it commonly came close to 1.5 volts. This has caused problems for some third-party devices, and it has been corrected on subsequent machines. (There is a recommended service procedure for repairing this problem available from Apple Service representatives, should this be a problem for a user.)

Soft Power

On Macintoshes with software power control, the machine can be turned on by an ADB device. This is accomplished by momentarily connecting pin 2 on the ADB connector (reserved) to pin 4 (ground). Pin 2 should be kept grounded until pin 3 (+5 volts) comes up to power, letting you know that the machine has actually come on. Some Macintoshes do not have soft-power capabilities; on these machines, grounding pin 2 will have no effect.

Back to top

The ADB Protocol


Each ADB device has four logical registers; the host can ask the device to talk to or listen on each one of these registers. A talk command asks the device to output the stored value from that register; a listen commands asks the device to accept a new value for that register. A register can contain between two and eight bytes. Some of the registers have predefined functions: register 0 is used as the primary data transfer register for most devices; it is this register which is polled by the input mechanism (as described below). Register 1 has no specified use; it is available for any use the device might require. Register 2 has no specified use for most devices; the ADB specification does define an "extended address device" protocol for register 2 on devices at address 1, but this is unused by most developers. Register 3 is used to identify devices and to separate devices which occupy the same address, as discussed in "Address Resolution" below.

Default addresses and handler IDs

Each ADB device identifies its software interface with two constants: the default address and the handler ID. This pair uniquely identifies a device's software interface; the default address usually specifies a device's general type (such as relative pointing device or keyboard), while the handler ID specified the particular data protocol this device uses for communication. The default address categories are as follows:

Device Type


Security and Dongles




Relative pointing devices (mice)


Absolute pointing devices (tablets)


Low speed data transfer devices


Reserved (PowerBook Duo charger)


Appliances (Miscellaneous catch-all)

Table 1 -Default address categories

The default address is only a guide; there's no real reason a mouse can't be at address 7, but default addresses are assigned on a category basis to try to avoid the case where a user has more than one device at a particular address. By putting all the relative devices at address 3, collisions will be avoided at address 3 for all users who have only one relative pointing device. While the bus is robust with respect to separating devices which are at the same address, the ADBS driver-loading mechanism, which is described below, made it useful to try to avoid having several dissimilar devices at the same address.

Default addresses and handler IDs are assigned by Apple Software Licensing when an ADB license contract is completed. Default address 0 is used by the ADB host; addresses 8 through 15 are used as locations for locating devices dynamically.

Some devices support more than one data protocol. An example is the extended keyboard, which can be asked to send separate key codes for the left and right shift keys. This change is accomplished by changing its handler ID to 3; the new handler ID reflects the new data protocol. If your device receives a request to change handler IDs (via a listen register 3 command), it should only obey the request if it knows how to speak the protocol specified by the new handler ID. For example, the extended keyboard, when receiving a request to change handler IDs, should accept the change if it is going to ID 3 or some other ID it knows about, and should thereafter report that ID as its own in response to talk register 3 commands. If it receives a request to change to handler ID $52, it should ignore it, as it does not know what handler ID $52 implies, and continue to report its original handler ID in response to talk register 3 commands.

A special case is devices which emulate Apple device protocols. For example, you may be constructing a 17-button mouse for use by specially trained squid; this device has a special software protocol to allow it to convey the state of all 17 mouse buttons. However, you may wish to emulate the Apple mouse protocol so your device can be used as a one-button mouse on machines which don't have your driver software installed. Due to the software design of the ADB Manager, if your device is at default address 3, it will have the default mouse driver installed as its driver at startup regardless of its handler ID. When your software loads, you can install a new driver for your device and tell it to begin talking the 17-button protocol. You have two options:

  1. You can have your device power up with the default Apple mouse handler ID ($01) or with your own handler ID as assigned by Apple Software Licensing. If you power up with the $01 handler ID, your software will have to locate your device by trying to change the handler ID of each device which started out at address 3 to you custom ID, then reading the handler ID of that device back. If the change sticks, then you know you're talking to your device, and can install your custom driver. In addition, the command to change handler IDs told the device to begin using your custom protocol.
  2. Alternatively, you could have your device power up with your custom handler ID; this would make identifying your device somewhat easier, as your device could be identified by its special handler ID. However, you would have to use some special command to tell the device that you have installed the new driver and that it can begin speaking the custom protocol. For example, you could use a talk register 1 to tell the device that you're ready for the 17-button data protocol.

Both approaches work well, but the first one is recommended. It ensures that the current protocol can always be determined by looking at the current handler ID; however, it relies on all devices at that default address handling commands asking them to change to a different handler ID properly, as a device which incorrectly changes its handler ID to your assigned ID would fool you into thinking it was your device. This doesn't seem to be a large problem, but there could be some obscure devices with this bug.

Address Resolution

Address resolution is the process the ADB Manager uses to separate devices that share the same default address so they power up shadowing each other at a particular address. It relies on devices using collision detection to determine when there is more than one device at a particular address.

An ADB device's register 3 is 2 bytes long, and includes 4 bits in which the address is stored. When the device receives a listen register 3 command, it should take its address from this 4-bit field. When it receives a talk register 3 command, it would be redundant to put the device's address in that field: the device's address is already uniquely determined by the fact that the device is responding to the talk register 3 command, which was sent to a specific address. Instead, a random 4-bit value should be returned in this field; this makes it easier to detect collisions between two devices responding to a talk register 3.

When a device receives a talk register 3 command, it should send back all of register 3, including the random field, and it should pay careful attention to collision detection. Should a device detect a collision when responding to a talk register 3 request, there is a special provision in the ADB protocol which says that the device should ignore the next listen register 3, which asks it to change address. The next time the device receives a listen register 3 command, it should check to see if the handler ID field is set to $FE. This is a reserved handler ID used to distinguish this type of listen request. If the handler ID field is $FE, the device should only change address if it has safely transmitted a complete packet since its last collision. Of course, the device should not adopt the handler ID of $FE.

Here is a summary of a typical sequence where the host is attempting to separate two devices. There are two devices of the same type, which I will call Fred and Wilma, at address 3.

  1. The host will send a talk register 3 command to address 3.
  2. Both Fred and Wilma will receive the command and begin to respond.
  3. Even though they are sending the same data, and they begin sending at the same time, they select different random numbers to insert into the "device address" field of register 3, bits 7-10.
  4. Wilma detects a collision while trying to send bit 7 and immediately stops sending.
  5. Fred does not detect a collision and completes his transmission of register 3.
  6. The host, seeing that a device has responded to its talk register 3, sends a listen register 3 to address 3, asking the device there to move to address $F. The handler ID field is $FE, indicating that a device should not move if it has detected a collision.
  7. Both Fred and Wilma receive the request.
  8. Fred moves to address $F; however, Wilma does not, as she has been locked out of moving because she detected a collision while responding to her last talk register 3 command.
  9. The host sends a talk register 3 to address $F to ensure that a device moved there. Fred responds, assuring the host he has moved.
  10. The host repeats the separation procedure, sending a talk register 3 to address 3, followed by a listen register 3 to address 3, asking the device there to move to address $E.
  11. This time, Wilma does not detect a collision, as she is the only device remaining at address 3, so she moves to address $E.
  12. The host sends yet another talk register 3 to address 3, but does not receive a response, as there are no more devices remaining at address 3.
  13. The host relocates Wilma from address $E back to address 3, with a listen register 3 command sent to address $E.
  14. The host moves on and repeats the process for address 4.

Each and every ADB controller isn't guaranteed to follow this procedure precisely, but it gives a feeling for the principle behind the address resolution process. You may see some implementations moving devices many more times than is necessary. This is done because some devices have been manufactured to tolerances close enough that not only do they send their bits at exactly the same time and so cannot detect collisions with each other, but they select the same random numbers to transmit. We recommend that you include some low-tolerance device (such as a capacitor on your reset line) to ensure that various devices will respond differently and be able to detect their collisions with each other.


Autopolling is the primary method by which the host fetches data from your device. The host repeatedly issues talk register 0 commands to your device; if your device responds with data, it is passed to your device's driver, which should act on it as new data.

This implies that register 0 should be the primary data transfer register for most devices; registers 1 and 2 are usually only used for supplementary data to configure the device. Most device drivers have no need to issue commands to their device, as all necessary data has been transferred within register 0.

When a device wishes to transmit data, it should wait until a command is issued to it or some other device. If, when this transmission is completing, it still wishes to transfer data (if the command was sent to our device, it might have fetched the data already), it should assert SRQ after the data portion of the command by holding the bus low for 300 us after the stop bit. This will alert the host that some device wishes to transmit data. It will then begin polling those addresses which it knows hold devices. If a device does not have any data, the host will move on to the next address, asking each device in turn, until SRQ is no longer asserted, indicating that all pending data has been fetched.

When an SRQ is not asserted, the host will continually poll the last device to send it data, sending it talk register 0 commands periodically. This is done under the assumption that this is likely to be the next place the user interacts; if the user types a character, she is very likely to type another soon. On current hosts, this can happen up to 150 times a second, although it can happen much less frequently in some cases. If the device responds with any data, the host will call the device's driver with the data.

Your device should only respond with data when sent a talk register 0 command if it has new data. If the status of the device has not changed since the last talk register 0, then it should not respond at all, allowing the bus to time out. This is useful for two reasons. First, it tends to reduce the demands on the host, as your driver need not be called when your device has nothing useful to say. Second, in some ADB implementations, the host can get "hung up" on your device if you always respond. For example, say that you have a device at address 4 which will always respond regardless of whether it has new data; there is another device at address 7. Moreover, the system is currently autopolling address 2. If the device at address 7 asserts SRQ, the system will begin looking through the addresses for the device which has data to send. When it reaches your device at address 4, your device will respond, although it has nothing new to say. On some implementations, this will cause the host to repeatedly ask your device for more data, and your device will continue responding. Meanwhile, device 7 is withering away at the end of the bus, and will never get serviced. What your device should do is not respond to the talk register 0 (since it has no new data); this will allow all host implementations to pass you by and reach the device at address 7, which needs the host's attention.

A useful summary of a reasonable algorithm is:

  • Wait for a command to begin.
  • If the command is directed to my address, handle it. If it is a talk register 0 command, only respond if there is new data waiting to be sent.
  • At the end of the command, if there is data waiting to be sent, assert SRQ, regardless of whom the original command was directed toward.

This simple behavior will produce the appropriate responses and SRQ generation for proper bus functioning. It shouldn't be necessary for your device to have any explicit knowledge of whether it is the "active" device or not; a robust basic behavior will eliminate any need for such information.

As an optimization, all recent versions of the ADB Manager will not automatically poll a device which does not have a driver service routine installed. In this case, they will switch to autopolling some other device, even if the manager has not been recently communicated with that device. However, the host may poll a device, even if it does not have a service request, in order to try to clear an SRQ on the bus.

Bus Initialization

Bus initialization doesn't work exactly as it might seem from looking at some documentation. A SendReset command is never sent to individual devices; rather, when a bus reset is requested, the ADB Manager sends a SendReset command which is broadcast to all devices, causing them to reset themselves and go to their default addresses and handler IDs. The relocation and driver loading procedure will follow immediately after the reset command is sent.

Back to top

ADB Drivers

Driver Installation

In the past, the recommended way to install an ADB driver was to install a resource of type 'ADBS' with the same resource ID as your device's default address in the system file (which held code to install the driver for the ADB driver, along with the driver itself). When the system was booted or when the bus was reinitialized, the system would get the 'ADBS' resource and execute it, which would let your code set up the driver for that device. Unfortunately, this system is not well-designed for the wide variety of devices available; the most prominent flaw is that the 'ADBS'resources are indexed by default address only. Thus, there is no way to use the 'ADBS' mechanism to load drivers for two different devices at address 7, since there can be only one resource with ID 7. A further flaw is that it requires installation into the system file, which is not currently recommended for a number of reasons. In any case, it has always required a special installation and deinstallation program, which has been more than a little confusing for users.

Currently, the recommended method is to supply the user with a system extension which will load your driver. This can either be a simple extension, or it can be contained within a control panel, should your device require some user interface for configuration. Your code, when loaded, should look for your device and install your driver for it. If your device is at a default address which is not shared with standard Apple devices, you don't have to concerned with what driver is installed for you by default--your device can just power up at its standard address with its handler ID. Your extension can then locate your device's current address by indexing through all the known devices with the ADB Manager call GetADBInfo. When you find a device whose initial address and handler ID match your device's, you can call SetADBInfo to install your driver's completion routine to handle autopolled data from your device.

Because your ADB driver is in a system extension which will not load until well into the system startup process, you might need to provide standard system functionality before your driver loads in order to allow the user to interact with the system during the startup process if your device is a standard one (a pointing device or keyboard). This arrangement also allows the user to use your device to control their machine even if they don't have the software installed, such as when they are booting from a floppy. In this case, your device will need to be able to emulate the Apple protocol for mouse or keyboard devices until your software driver loads, as discussed above in "Default Addresses and Handler IDs". Until your driver is installed, your device will be serviced by the default driver for this address. Even if it cannot emulate an Apple device, a device at address 2 or 3 must supply harmless data in register 0 until your driver loads and is installed, as the Apple ADB driver for that address may inadvertently receive the contents of your register 0 and attempt to use it as input data. If the data in your field caused effects such as the mouse button or shift key sticking down, this could cause problems for the user.

If you use the recommended procedure for handler IDs in this case (powering up with the appropriate Apple handler ID and switching to your custom ID when your driver loads), you will need to use ADB commands to find your device at startup. You should index through the connected ADB devices with GetADBInfo, and if you find a device that has an original address and handler ID which indicates that it might be your device, attempt to switch it to your handler ID by using ADBOp to send it a listen register 3 to change its handler ID. (You will first need to read its register 3 with a talk register 3 command so you can correctly copy the various flag bits in the register you send to the device.) You should then issue a talk register 3 command to the device and examine the response to see if the new handler ID was accepted and reported back by the device. If so, you can be certain that this device is yours, and you can then call SetADBInfo to install your driver as the handler for that device.

In the original method using 'ADBS' resources, the system could automatically reload your driver any time the bus was reinitialized. If your driver is loaded via a system extension, there isn't any way for the system to find your device and driver, and reconnect them after the bus is reinitialized and devices are relocated. Thus, you must manually reconnect your driver to your device each time the bus is reinitialized. Fortunately, there is a system provision for you to be notified each time the bus is reinitialized. There is a low-memory global called jADBProc (at address $6B8) which is a pointer to a procedure to be called just before and just after the bus is initialized. When the procedure is called before the bus is initialized, register D0 is set to 0; afterword, it is set to 1. When your extension loads and installs your driver, you should remember the value in the jADBProc global and then install a pointer to a procedure of your own in that location. When your procedure is called and has finished its processing, it should then call through to the previous procedure whose address was in jADBProc before you replaced it. (If jADBProc was equal to 0 before you installed, just return from your procedure.) Don't forget to preserve register D0 so subsequent procedures can tell if they are being called before or after the reinitialization. The actions you take in the post-initialization case should basically duplicate your original installation procedure, including looking through all the devices for devices of your type and calling SetADBInfo to install your driver for your devices.

For history buffs, this is the second time the recommended procedure for loading ADB drivers has changed. Initially, it was recommended that ADB drivers be loaded with INITs; however, at the time, jADBProc did not exist, so when the Macintosh Portable came out, INIT-handled devices had the fatal flaw of not reloading their drivers when the Macintosh was put to sleep and reawakened, since this procedure involved resetting the bus. This caused us to begin recommending the use of the 'ADBS' resource, since in this way the driver could be reloaded from the System file. However, the 'ADBS' approach has so many major flaws that we have stopped recommending it now that the jADBProc global has been introduced. The jADBProc global has been available since System 6.0.4.

For true history buffs, or possibly for specialized applications, here is the description of the 'ADBS' resource functionality. The 'ADBS' resource is loaded into the system heap by the system and detached with DetachResource. The resource is then called by JSRing to the first byte of the resource. At this time, the registers are set up as follows: A0 holds the address of the 'ADBS' resource data (although it is no longer in a resource handle, thanks to the DetachResource call); if a handle to the resource is needed, RecoverHandle can be called to retrieve it. The low byte of register D0 holds the ADB device's current address (due to relocation, this might not be the same as the default address) and the low byte of register D1 holds the handler ID of the device in question.

Back to top

The Cursor Device Manager

In order to be able to manage an expanding set of relative movement devices, Apple has created the Cursor Device Manager, which is a software architecture which provides a standard interface to devices of widely varying resolutions and capabilities. This also allows better management of multiple relative devices on a single bus. In the old architecture, all connected devices shared a single button state and acceleration curve, which became a problem for Apple and for third-party device manufacturers. The Cursor Device Manager provides a number of calls for finding, configuring, and manipulating relative devices connected to the bus. It also supports the new extended mouse protocol, which is described below in the "Apple Devices" section.

Cursor Device Manager types

The Cursor Device Manager treats each relative or absolute device as a Cursor Device. Each one is specified by a CursorDeviceRec which is defined as follows:

typedef struct CursorDevice {
    struct CursorDevice    *nextCursorDevice;    // pointer to next record in
                                                 // linked list
    CursorData             *whichCursor;         // pointer to data for target cursor
    long                   refCon;               // application-defined
    long                   unused;               // reserved for future
    OSType                 devID;                // device identifier (from ADB reg 1)
    Fixed                  resolution;           // units/inch (orig. from ADB reg 1)
    UInt8                  devClass;             // device class (from ADB reg 1)
    UInt8                  cntButton;            // number of buttons (from ADB reg 1)
    UInt8                  filler1;              // reserved for future
    UInt8                  buttons;              // state of all buttons
    UInt8                  buttonOp[8];          // action performed per button
    unsigned long          buttonTicks[8];       // ticks when button last went
                                                 // up (for debounce)
    long                   buttonData[8];        // data for the button operation
    unsigned long          doubleClickTime;      // device-specific double click speed
    Fixed                  acceleration;         // current acceleration
    short                  privateFields[15];    // fields used internally to CDM

The cursor controlled by this cursor device is described with a CursorDataRec:

typedef struct CursorData {
    struct CursorData    *nextCursorData;      // next in global list
    Ptr                  displayInfo;          // unused (reserved for future)
    Fixed                whereX;               // horizontal position
    Fixed                whereY;               // vertical position
    Point                where;                // the pixel position
    Boolean              isAbs;                // has been stuffed with absolute coords
    UInt8                buttonCount;          // number of buttons currently pressed
    long                 screenRes;            // pixels per inch on the current display
    short                privateFields[22];    // fields use internally by CDM

Most of the fields are fairly self-explanatory. The fields labeled as private at the end of the CursorDeviceRec are used to manage cursor acceleration, and shouldn't be modified by your software. Some of the usage of the other fields will be explained below in the description of the Cursor Device Manager routines.

Cursor Device Manager Routines


pascal OSErr CursorDeviceNextDevice(CursorDevicePtr *ourDevice);

CursorDeviceNextDevice can be used to index through the various devices the Cursor Device Manager is aware of. You pass it a CursorDevicePtr; initialize this variable to nil to get the first device in the list, then call CursorDeviceNextDevice repeatedly, each time passing the CursorDevicePtr as it was last modified; when you have reached the end of the device list, the pointer returned in the curDevice parameter will be 0.


pascal OSErr CursorDeviceNewDevice(CursorDevicePtr *ourDevice);

Call CursorDeviceNewDevice to create a new cursor device and link it into the device chain. The new device record will be initialized with values representing a standard one-button mouse; you should call CursorDeviceSetAcceleration for the device after creating it. A pointer to the created device is returned in the ourDevice variable.

New cursor devices are created for all ADB devices with a type 3 or 4 handler ID. This routine should only be needed by devices which are connected though some other method, such as the serial port.


pascal OSErr CursorDeviceDisposeDevice(CursorDevicePtr ourDevice);

This routine disposes of a cursor device and unlinks it from the device chain. This isn't needed by most developers, but could be useful for non-ADB devices which might be connected and disconnected.


pascal OSErr CursorDeviceMove(CursorDevicePtr ourDevice,long deltaX,long deltaY);

CursorDeviceMove accumulates the deltaX and deltaY values into the recorded movement of the device; the next time the cursor position is calculated, these deltas will be fed through the acceleration algorithm and used to move the cursor. This routine should be called by a relative device driver with the data it receives from its device, even if the deltas are both zero. This lets the acceleration algorithm properly calculate the appropriate motion. This routine is automatically called by the default driver for all devices with a type 3 or 4 handler ID. You would only need to call it if you were using an ADB device that isn't handled by the default driver, or a non-ADB device.


pascal OSErr CursorDeviceMoveTo(CursorDevicePtr ourDevice,long absX,long absY);

CursorDeviceMoveTo sets the absolute position of the cursor to (absX, absY). The next time the cursor position is calculated, it will be moved to this absolute location. This would normally be used by a driver for an absolute pointing device to position the cursor.


pascal OSErr CursorDeviceFlush(CursorDevicePtr ourDevice);

CursorDeviceFlush causes the acceleration and motion algorithms to flush out all their error collection and motion deltas into the cursor position: it indicates that your device is done moving temporarily. This may be useful for devices which can tell when they become idle (such as a stylus pointing device); if they call this routine when they become idle, it ensures that all unused motion data will be worked into the cursor position at the next time it is calculated.


pascal OSErr CursorDeviceButtons(CursorDevicePtr ourDevice,short buttons);

CursorDeviceButtons handles posting mouseUp and mouseDown events and also deals with debouncing mouse clicks. Pass the current button status in the buttons parameter, going from bit 0 is button 0 to bit 7 representing button 7. For each button, a one-bit represents down, a zero-bit represents up. This routine debounces mouse clicks to keep them from looking like double clicks; if the button goes down less than 2 ticks after coming up, then the mouseDown will be ignored. Button up events are never ignored to avoid problems such as continuous scrolling, which are confusing and difficult for the user to deal with. A device driver should call this routine any time it gets a data packet; this routine deals with keeping track of whether the button state has changed. This routine automatically calls routines installed with CursorDeviceButtonOp.


pascal OSErr CursorDeviceButtonDown(CursorDevicePtr ourDevice);

CursorDeviceButtonDown posts a mouseDown event if this is the first button to go down for this device. It is called by the standard button operation routines, and you should need to call it only if you use a custom button operation routine.


pascal OSErr CursorDeviceButtonUp(CursorDevicePtr ourDevice);

CursorDeviceButtonUp posts a mouseUp event if this is the last button to come up for this device. It is called by the standard button operation routines; you should need to call it only if you use a custom button operation routine.


pascal OSErr CursorDeviceButtonOp(CursorDevicePtr ourDevice,short buttonNumber,ButtonOpcode opcode,long data);

CursorDeviceButtonOp sets a new operation to be associated with a particular button. btnNo may range from 0 to 7, and opCode specifies what operation to use. The data field specifies a parameter for the operation you are setting for the kButtonCustom operation. The opCode parameter may have one of the following values:

kButtonNoOp No action
kButtonSingleClick Normal mouse button
kButtonDoubleClickkButton Click, release, and click again when pressed
kButtonClickLock Click on press, release on next press
Custom Call a custom procedure; data holds its address

Using the btnCustom operation will cause a procedure whose address is passed in data to be called whenever this button changes state. The procedure takes the following parameters: the address of the CursorDeviceRec record for its device is passed in register A2; the button being pressed or released is in register D3. The new state of the button will already have been filled into the buttons field in the CursorDeviceRec, so you may use that flag to determine if the button is being clicked or released. Your routine does not need to preserve registers D0, D2, A0, and A1; it must preserve all others.

The kButtonCharStroke and kButtonAppleScript operations are currently unimplemented and will simply cause the button press to be ignored.


pascal OSErr CursorDeviceSetButtons(CursorDevicePtr ourDevice,short numberOfButtons);

CursorDeviceSetButtons allows you to set the number of buttons on the device specified by ourDevice to numButtons.


pascal OSErr CursorDeviceSetAcceleration(CursorDevicePtr ourDevice,Fixed acceleration);

CursorDeviceSetAcceleration lets you set the acceleration for the device specified by ourDevice to the value specified by acceleration, where 0 <= acceleration <= 1. The Cursor Device Manager will build an acceleration table for the device based on its device ID or device class and the desired acceleration value. For details on acceleration resources, see "Acceleration Tables" below. The acceleration table set by CursorDeviceSetAcceleration is found by interpolating the tables stored in the appropriate acceleration resource. All 'accl' resources in the resource chain or in the ROM are searched looking for the one applying to the specified device, so a control panel which called CursorDeviceSetAcceleration could implicitly use 'accl' resources stored within the control panel's resource fork.


pascal OSErr CursorDeviceDoubleTime(CursorDevicePtr ourDevice;

duration: LongInt);

CursorDeviceDoubleTime lets you set the double-click time associated with a particular device. The duration parameter specifies the time, in ticks, to use as the double-click time for this device. An application could be written to check the double-click time for the particular device in checking for a double-click.


pascal OSErr CursorDeviceUnitsPerInch(CursorDevicePtr ourDevice;

resolution: Fixed);

CursorDeviceUnitsPerInch lets you set the resolution of a particular device to its physical resolution, in units per inch. For devices adhering to the Apple extended mouse protocol, this call shouldn't be needed, as the resolution can be read from the device's register 1; however, this call might be made by a driver for a device which doesn't use the ADB.

Acceleration Tables

Acceleration tables are stored in resources of type 'accl', which have the following Rez description:

type 'accl' {
    literal longInt /* Device identifier or device class */
    classAbsolute, /* A flat-response device */
    classMouse, /* Mechanical or optical mouse */
    classTrackball; /* Trackball */
    integer = $$CountOf(AcclTable);
    /* Number of tables for this device */
    array AcclTable
    { /* Entries sorted by first value; must have */ /* at least 0.0 and 1.0 tables */
        unsigned hex longint;
        /* Acceleration provided by this table (Fixed) */
        integer = $$CountOf(AcclPoint); /* Number of control points for this device */
        wide array AcclPoint { /* Entries sorted by first value; implicit */
            /* first entry (0.0, 0.0); at least one more */
            /* entry required. */
            unsigned hex longint; /* Device speed (inches per second) (Fixed) */
            unsigned hex longint; /* Cursor speed (inches per second) (Fixed) */

The identifier for this 'accl' resource is stored in the first long word; this is either an OSType four-character device identifier or an integer value specifying the device class. In either case, the device's identity is generally read from the device's register 1, as described below in the section "Extended Apple Mouse Protocol." The Cursor Device Manager first tries to match against the specific device identifier, then against the more general class. Each 'accl' resource can contain a number of acceleration tables for different acceleration values; each table contains a number of entries that match a particular device speed to a particular cursor speed. An 'accl' resource must contain at least two acceleration tables, one for an acceleration of 0.0, and one for an acceleration value of 1.0. When an acceleration value is set for a particular device, the acceleration table is calculated by interpolating between the two nearest tables from the 'accl' resource. There is an implicit entry in each acceleration table of (0.0, 0.0), which indicates that the cursor should not move if the device does not. At least one additional entry is required; the Cursor Device Manager will use the table entries to figure the cursor movement by using the device movement to interpolate based on the specified movement control points.

Availability Of The Cursor Device Manager

The Cursor Device Manager was introduced in the ROMs of Macintoshes introduced in February, 1993. It may be installed via software on any Macintosh. To check to see if the Cursor Device Manager is available, you should use the standard TrapAvailable routine to check to see if its trap is implemented. The Cursor Device Manager trap is $AADB, making it a toolbox trap, trap number $2DB.

#include <Traps.h>
#include <OSUtils.h>
static Boolean CursorDeviceManagerAvailable(void)
    return GetToolboxTrapAddress(_CursorDeviceDispatch) !=


When the Cursor Device Manager is installed, all Apple mouse drivers use its interface to move the cursor; this means that the low memory globals such as Mouse and RawMouse are no longer used. While a compatibility mode keeps drivers which still modify these globals continue to work, the cursor position can no longer be read from these globals.

Back to top

Using the CursorDevice Manager in Native Code

With the original release of the CursorDevice Manager, a problem was introduced in the native InterfaceLib code such that the glue code for the following CrsrDev calls was broken.

extern pascal OSErr CrsrDevButtons(CrsrDevicePtr ourDevice);

extern pascal OSErr CrsrDevButtonOp(CrsrDevicePtr ourDevice);

extern pascal OSErr CrsrDevSetButtons(CrsrDevicePtr ourDevice);

extern pascal OSErr CrsrDevSetAcceleration(CrsrDevicePtr ourDevice);

extern pascal OSErr CrsrDevDoubleTime(CrsrDevicePtr ourDevice);

extern pascal OSErr CrsrDevUnitsPerInch(CrsrDevicePtr ourDevice);

To address this problem, an updated set of CursorDevices interfaces were released in which the calls were modified by changing the call prefix to CursorDevice.... This change introduced a new problem. One could no longer link CursorDevice Manager calls with the InterfaceLib stub library. To address this new issue, the "CursorDevices.Glue.c" file was made available. The downside of using this code was that it statically linked to the 68K ADB trap code. If in a future release of MacOS, we were to fix a problem with the CursorDevice Manager, code linked to "CursorDevices.Glue.c" might not benefit from these fixes.

Since the release of ETO #24, there is now the "CursorDevicesGlue.o" object library which all CursorDevice Manager processes should link with. This library will properly detect for an updated CursorDevice Manager, else use the existing 68K ADB trap code. This library expects one to use the CursorDevice... calls as prototyped in the CursorDevices interface files.

Back to top

Apple Devices

Classic Apple Mouse Protocol

The original Apple mouse protocol allows for mice with a resolution of 100 or 200 units per inch with 7-bit accumulation of relative movement with one or two buttons. Handler ID 1 was used to indicate 100 cpi operation; handler 2 to indicate 200 cpi operation. All data is transferred through register 0 in the following format:

Classic Mouse Register 0 Format

Figure 1 - Classic Mouse Register 0 Format

A relative motion in each axis is accumulated until fetched with a talk register 0 command sent by the host. The X value represents left-to-right motion. The Y value holds accumulated forward-to-back motion.

Extended Apple Mouse Protocol

In order to take advantage of the higher movement resolution support available with the CursorDevice Manager, an extended mouse protocol has been defined. The extended mouse protocol allows a device to communicate up to 16 bits of movement data with each polling, and allows a standard device to communicate the state of up to 8 buttons. In addition, the new format allows each device to communicate its exact type, resolution, and class. This new format applies to all relative devices at default address 3, handler ID 4.

During startup or after a reset of the ADB bus, devices which power up at default address 3 with handler ID 1 will be switched to handler ID 4 with a listen register 3 command. If the device accepts the switch and reports the new handler ID of 4 back in response to a talk register 3 command, all subsequent communication with the device will be assumed to follow the extended mouse protocol. Currently, the ADB Manager makes an additional check, making sure that a talk register 1 returns 8 bytes (the format of register 1 is specified below). If a device accepts a handler ID change but does not return 8 bytes from register 1, it is assumed to not actually be an extended mouse protocol device and it is switched back to its original handler ID.

All movement and button data still passes through register 0, which can now hold between 2 and 5 bytes, where 2 bytes provides the device with the data transfer capability of the original mouse protocol and additional bytes allow added resolution and buttons.

Bytes 0 and 1 of register 0 have the same format as they did in the classic mouse protocol, communicating the state of the first two buttons and the low order 7 bits of the accumulated motion in the X and Y axes. There can be up to 3 bytes of additional information: each one of these bytes can communicate the state of 2 more buttons and add 3 higher order bits of resolution to each of the X and Y axes. The format of each additional byte is:

Format Of Additional Register 0 Bytes For
         Extended Mouse Protocol

Figure 2 - Format Of Additional Register 0 Bytes For Extended Mouse Protocol

As a specific example, in a maximal 5-byte transmission, the data would be transferred in this format:

Bit #:









Byte 0:









Byte 1:









Byte 2:









Byte 3:









Byte 4:









Table 2 - Format of a 5-Byte Register 0 Data Transfer

Where bn indicates the status of button n and xaa or ybb indicates the value of bit aa or bb of the X or Y movement.

In addition, register 1 is used in the extended mouse protocol to provide some general information about the device. Register 1 is 8 bytes long and is formatted in this way:

Byte Range



Unique device identifier


Device resolution in units/inch


Device class (Mouse, trackball, etc.)


Number of buttons (0-8)

Table 3 - Format Of Extended Mouse Protocol Register 1

The unique device identifier is intended to be a four-character ASCII identifier similar to the OSType identifiers used as types and creators in the Macintosh file system; they can be registered using the same mechanism used to register creator types. A developer should only use a device identifier in this field if they have obtained a registration for that identifier's use as a creator from Apple.

The device class is a value which is used to identify the type of device and to control the acceleration curve used for that device. The currently defined constants include:

Device Constant

Device Type


Tablet device (absolutely positioned)





Table 4 - Currently defined device classes

There currently isn't any mechanism for developers to create or register device classes; if a developer needs a device class not available from Apple, the only alternative available currently is not to use the handler ID 4 extended mouse protocol. Instead they should use a custom handler ID and custom driver software.

Apple Keyboard Protocol

The Apple keyboards have a simple data transfer protocol. Register 0 is used to inform the host as keys are depressed and released; register 2 is used to communicate the state of the modifier keys and to control the LED indicators on the extended keyboards.

The format of register 0 is:

Keyboard Register 0 Format

Figure 3 - Keyboard Register 0 Format

Register 0 can communicate up to two key transitions at once. Each transition consists of a key code and a key released bit, which is 0 for key depressions and 1 for key releases. The key codes are described in Inside Macintosh volume V, pages 191-192. The special case is the reset key, which returns the value $7F7F in register 0 when it is depressed and $FFFF when released; thus, it uses both key code positions within register 0.

The format of register 2 is:

Keyboard Register 2 Format

Figure 4 - Keyboard Register 2 Format

The current state of the keys listed in figure 4 is available in bits 6-14 of register 2, if those keys exist on the keyboard being examined. Bits 0-2 hold the current state of the LEDs on the extended keyboard; the states of these LEDs can be changed by sending the keyboard a listen register 2 command. Note that key transition events are generated in register 0 for modifier keys, as they are for all other keys; these keys are available in register 2 in addition to their status being transmitted through register 0.

Back to top


The Apple Desktop Bus is patented. In order to build an ADB device, you will need to get a license from Apple.

Contact our Software Licensing department at or 800-793-9378 or 512-919-2645.

The license is available for a nominal fee in most cases, and the licensing package includes the latest version of the ADB specification, which is the definitive reference to the bus.

Back to top


The Apple Desktop Bus specification, revision F.

Inside Macintosh, Volume V, chapter 20, "The Apple Desktop Bus."

Guide to the Macintosh Family Hardware, Second Edition, Chapter 8, "Apple Desktop Bus."

Back to top

Change History


New information appears throughout the document. In particular, the Cursor Device Manager is discussed in this note for the first time. The information regarding bugs fixed in System 6.0.4 was omitted.


The Cursor Device Manager API has been updated for Universal Headers 3.1 and some issues with ROM bugs have been documented.

Back to top


Acrobat gif

Acrobat version of this Note (252K)


Back to top

Did this document help you?
Yes: Tell us what works for you.
It’s good, but: Report typos, inaccuracies, and so forth.
It wasn’t helpful: Tell us what would have helped.