Copyright ©1995 by NeXT Computer, Inc.  All Rights Reserved.

IODirectDevice



Inherits From: IODevice : Object
Declared In: driverkit/IODirectDevice.h
driverkit/architecture/directDevice.h
driverkit/architecture/IOPCIDirectDevice.h
driverkit/architecture/IOPCMCIADirectDevice.h



Class Description

IODirectDevice is a device-independent abstract class that is the superclass of all direct device driver classes. Most of the functionality of IODirectDevice is provided by device-dependent categories, which are described in detail below. IODirectDevice provides:

An implementation of the deviceStyle IODevice class method, so IODirectDevice subclasses don't have to override it
Methods for getting and setting IODirectDevice information, such as the interrupt port and the IODeviceDescription
A default I/O thread that listens for messages to the interrupt port
An efficient way to receive messages, to be used by drivers that provide their own I/O thread (see the waitForInterrupt: method description)

To use the default I/O thread, subclasses invoke one of the startIOThread... methods and implement one or more of the following methods:

interruptOccurred or interruptOccurredAt:
timeoutOccurred
commandRequestOccurred
otherOccurred:
receiveMsg

Each of these methods is invoked when the I/O thread receives a corresponding Mach message on its interrupt port. For example, when the kernel sends an IO_DEVICE_INTERRUPT_MSG Mach message to the interrupt port, the I/O thread receives it and invokes interruptOccurred. The documentation for startIOThread describes in detail how the I/O thread listens for Mach messages and which methods it invokes in response to which Mach messages.

Interrupt messages are the only Mach messages that the kernel automatically sends. If you want to receive other types of Mach messages, your driver or some other module it works with must explicitly send them. For example, if you want your driver's timeoutOccurred method to be invoked by the I/O thread, you must ensure that your driver sends an IO_TIMEOUT_MSG at some point. Some classes, such as IOEthernet, have this functionality built in. Others, such as IOSCSIController, don't. See the IOSCSIController class description for an example of how to send a message.




ISA and EISA IODirectDevices

The IOEISADirectDevice category of IODirectDevice defined in the header file driverkit/i386/directDevice.h provides the following additional functionality for IODirectDevices that control hardware on ISA or EISA Intel-based computers:

Reserving and releasing ranges of I/O ports
Reserving, releasing, enabling, and disabling interrupts (also known as IRQs)
A way of providing an interrupt handler, if interrupt messages aren't sufficient
Mapping device memory into virtual memory
Reserving and releasing DMA channels
Starting DMA and dealing with DMA buffers
Determining whether the computer has EISA slots

Note:  The ISA/EISA category works for all hardware attached to ISA and EISA computers--ISA slots, EISA slots, VL-Bus, and so on. Remember that EISA computers can have ISA slots, but ISA computers don't have EISA slots.

I/O ports, interrupts, device memory ranges, and DMA channels are collectively known as resources.




PCI IODirectDevices

The IOPCIDirectDevice category of IODirectDevice defined in the header file driverkit/i386/IOPCIDirectDevice.h provides the following additional functionality for IODirectDevices that control hardware on PCI Intel-based computers:

Indicating whether the PCI bus is enabled or not
Reading and writing the device's configuration space

The PCI configuration space is memory available for configuation information for each device. A 256-byte portion is available for each device, addressed by the PCI anchor, which consists of three fields:

Device number between 0 and 31
Function number between 0 and 7
Bus number between 0 and 255

Methods can either read or write the entire configuraion space or access individual 32-bit pieces, accessing it by a register address--a byte address into the 256-byte portion.




PCMCIA IODirectDevices

The IOPCMCIADirectDevice category of IODirectDevice defined in the header file driverkit/i386/IOPCMCIADirectDevice.h provides the following additional functionality for IODirectDevices that control hardware on PCMCIA Intel-based computers:

Mapping and unmapping attribute memory

Attribute memory resides on the PCMCIA card and contains tuples, i.e., configuration information that's stored on the card. To access attribute memory, you must map the memory using the mapping method; when you've completed your access, you must unmap it with the method provided. If you attempt to map the memory and it's already mapped, the mapping method returns failure status.



Local Equivalents of Resources

The ISA/EISA category refers to resources not by their actual numbers or addresses, but by their local equivalent. The local equivalent of a resource is the position (starting at 0) of that resource in the configuration list of all resources of that type.

For example, if a device is configured to have one DMA channel (DMA channel 6, for example), the local equivalent of that channel is 0. If a device is configured to have two DMA channels (specified in order as 4 and 6, for example), then channel 4 has the local equivalent of 0, and channel 6 has the local equivalent of 1.

Similarly, the first range of I/O ports in a device's configuration has the local equivalent of 0, the second range is 1, and so on.

The local equivalent is used in all ISA/EISA methods that refer to DMA channels, specific interrupts, I/O ports, and memory ranges. For example, to enable the first DMA channel in a device's configuration, a driver sends an enableChannel: message to self, specifying 0 as the channel.

See Chapter 4 and Chapter 5, "Configuation Keys" in "Other Features" for information on configuration files.



Implementing a Subclass

The IODirectDevice methods you must implement in a subclass depend on your driver's capabilities. To start with, you must implement all the methods that IODevice requires, except for deviceStyle, which is implemented by IODirectDevice. You must also implement initFromDeviceDescription: to perform any driver- or device-specific initialization.

If your device performs DMA, you must implement startDMAForBuffer:channel:.

If your device can interrupt, you generally need to implement either interruptOccurred (if your device uses only one interrupt) or interruptOccurredAt:. If your driver needs to handle some interrupts directly, instead of receiving interrupt notification by Mach messages, you must implement getHandler:level:argument:forInterrupt:.

If your driver uses other Mach messages, you might also need to implement timeoutOccurred, commandRequestOccurred, otherOccurred:, or receiveMsg.

Most drivers need an I/O thread, as discussed in Chapter 1. All Driver Kit subclasses of IODirectDevice (such as IOEthernet) provide an I/O thread for you, if necessary. However, if your class is a direct subclass of IODirectDevice, you need to provide your own I/O thread. You can do so by invoking one of the startIOThread... methods.



Instance Variables

None declared in this class.



Method Types (Architecture-Independent)

Freeing instances
free
Registering the class + deviceStyle
Getting and setting the interrupt port
attachInterruptPort
interruptPort
Handling messages to the interrupt port
commandRequestOccurred
interruptOccurred
interruptOccurredAt:
receiveMsg
timeoutOccurred
waitForInterrupt:
Running an I/O thread startIOThread
startIOThreadWithPriority:
startIOThreadWithFixedPriority:


Getting and setting the IODeviceDescription
deviceDescription
setDeviceDescription:



Method Types (ISA/EISA Architecture)

Initializing instances initFromDeviceDescription:
Reserving I/O ports reservePortRange:
releasePortRange:
Dealing with interrupts enableAllInterrupts
disableAllInterrupts
reserveInterrupt:
releaseInterrupt:
enableInterrupt:
disableInterrupt:
getHandler:level:argument:forInterrupt:
Mapping memory mapMemoryRange:to:findSpace:cache:
unmapMemoryRange:from:
Dealing with DMA channels enableChannel:
disableChannel:
reserveChannel:
releaseChannel:
Dealing with DMA buffers startDMAForBuffer:channel:
createDMABufferFor:length:read:needsLowMemory:
limitSize:
freeDMABuffer:
abortDMABuffer:
Setting the DMA mode setTransferMode:forChannel:
setAutoinitialize:forChannel:
setIncrementMode:forChannel:
Using the EISA extended mode register
setDMATransferWidth:forChannel:
setDMATiming:forChannel:
setEOPAsOutput:forChannel:
setStopRegisterMode:forChannel:
Getting a DMA channel's status currentAddressForChannel:
currentCountForChannel:
getDMATransferWidth:forChannel:
isDMADone:
Optional DMA locking reserveDMALock
releaseDMALock
Getting information about EISA slots
isEISAPresent
getEISAId:forSlot:



Method Types (PCI Architecture)

Determining if PCI bus support is enabled
+ isPCIPresent
isPCIPresent
Reading and writing the entire configuration space
+ getPCIConfigSpace:withDeviceDescription:
+ setPCIConfigSpace:withDeviceDescription:
getPCIConfigSpace:withDeviceDescription:
setPCIConfigSpace:withDeviceDescription:
Reading and writing the configuration space
+ getPCIConfigData:atRegister:withDeviceDescription:
+ setPCIConfigData:atRegister:withDeviceDescription:
getPCIConfigData:atRegister:withDeviceDescription:
setPCIConfigData:atRegister:withDeviceDescription:



Method Types (PCMCIA Architecture)

Managing attribute memory mapAttributeMemoryTo:findSpace:
unmapAttributeMemory:



Class Methods (Architecture-Independent)

deviceStyle
+ (IODeviceStyle)deviceStyle

Reports the basic style of driver as IO_DirectDevice. Because IODirectDevice implements this method, its subclasses don't have to.

See also:  + deviceStyle (IODevice)



Instance Methods (Architecture-Independent)

attachInterruptPort
(IOReturn)attachInterruptPort

Creates the interrupt port, if none exists already, and requests that the interrupt port receive all interrupt messages for the device's reserved interrupts. This method is invoked whenever an interrupt is enabled. Returns IO_R_SUCCESS if successful; otherwise, returns IO_R_NOT_ATTACHED.

See also:  interruptPort, enableAllInterrupts ("Instance Methods (ISA/EISA Architecture)")



commandRequestOccurred
(void)commandRequestOccurred

Does nothing; subclasses can implement this method if desired. This method is invoked by the default I/O thread (implemented by startIOThread...) whenever it receives a bodyless message with ID IO_COMMAND_MSG. The part of a driver that handles user requests can use this message to notify the I/O thread that it should execute a command that's been placed in global data.

See also:  startIOThread



deviceDescription
deviceDescription

Returns the IODeviceDescription associated with this instance.

See also:  setDeviceDescription:



free
free

Deallocates the IODirectDevice's memory and its interrupt port, if one exists. Returns nil.



interruptOccurred
(void)interruptOccurred

Invokes interruptOccurredAt: with an argument of zero. This method is invoked by the default I/O thread (implemented by startIOThread...) whenever it receives a bodyless Mach message with the ID IO_DEVICE_INTERRUPT_MSG. Subclasses that support only one interrupt should implement this method so that it processes the hardware interrupt, as described in Chapter 1 and 2.

See also:  interruptOccurredAt:, startIOThread



interruptOccurredAt:
(void)interruptOccurredAt:(int)localInterrupt

Does nothing; subclasses that need to handle interrupts should implement this method so that it processes the hardware interrupt, as described in Chapter 1. This method is invoked by the default I/O thread (implemented by startIOThread...) whenever it receives a bodyless Mach message with an ID between IO_DEVICE_INTERRUPT_MSG_FIRST and IO_DEVICE_INTERRUPT_MSG_LAST (excluding IO_DEVICE_INTERRUPT_MSG).

See also:  interruptOccurred, startIOThread



interruptPort
(port_t)interruptPort

Returns the Mach port on which the IODirectDevice should receive interrupt messages. The returned port_t is in the context of the kernel I/O task.

See also:  attachInterruptPort:



otherOccurred:
(void)otherOccurred:(int)msgID

Does nothing; subclasses can implement this method if desired. This method is invoked by the default I/O thread (implemented by startIOThread...) whenever it receives a bodyless message with an unrecognized ID. The ID is given in msgID.

See also:  receiveMsg, startIOThread



receiveMsg
(void)receiveMsg

Dequeues the next Mach message from the interrupt port and throws it away; subclasses can implement this method if desired to handle custom messages. This method is invoked by the default I/O thread (implemented by startIOThread...) whenever it tries to receive a message that has a body. To implement this message, you need to call msg_receive() on the interrupt port. In this sample implementation, fill in the italicized text between angle brackets, that is << >>, with device-specific code:

- (void)receiveMsg
{
IOReturn        result;
port_t          inPort;
MyMsg           myMsg;
kern_return_t   result;

inPort = [self interruptPort];
if (inPort == PORT_NULL) {
<< React to having no interrupt port. >>
return;
}

myMsg.header.msg_size = sizeof (myMsg);
myMsg.header.msg_local_port = inPort;

result = msg_receive(&myMsg.header, (msg_option_t)RCV_TIMEOUT, 0);

if (result != RCV_SUCCESS) {
IOLog("%s receiveMsg:  msg_receive returns %d\n", result);
return;
}
else {
switch (myMsg.header.msg_id) {
case MyMsg1:
[self handleMsg1];
break;

case MyMsg2:
[self handleMsg2];
break;
.
.
.
}
}
}

See also:  otherOccurred:, startIOThread



setDeviceDescription:
(void)setDeviceDescription:deviceDescription

Records deviceDescription as the IODeviceDescription associated with this instance. ISA/EISA-architecture devices don't need to invoke this method because initFromDeviceDescription: already does so.

See also:  deviceDescription



startIOThread
(IOReturn)startIOThread

Invokes attachInterruptPort and, if attaching the interrupt port was successful, forks a thread to serve as the instance's I/O thread. This thread, which is appropriate for most drivers, sits in an endless loop that does the following:

Waits for a Mach message on the interrupt port by invoking waitForInterrupt:
If the message couldn't be dequeued because it was too large, invokes receiveMsg so that the subclass can dequeue and handle the message itself
If the message is dequeued successfully, invokes one of five methods, depending on the message ID:

Message ID Method Invoked
IO_TIMEOUT_MSG timeoutOccurred
IO_COMMAND_MSG commandRequestOccurred
IO_DEVICE_INTERRUPT_MSG interruptOccurred
IO_DEVICE_INTERRUPT_MSG_FIRST interruptOccurredAt:
to IO_DEVICE_INTERRUPT_MSG_LAST
(anything else) otherOccurred:

Returns the value returned by attachInterruptPort.

See also:  startIOThreadWithFixedPriority:, startIOThreadWithPriority:



startIOThreadWithFixedPriority:
(IOReturn)startIOThreadWithFixedPriority:(int)priority

The same as startIOThreadWithPriority:, except that the I/O thread's priority never lessens due to aging. This method lets you do performance tuning by disabling priority aging.

For more information about scheduling policies and priorities, see Chapter 1 of the NEXTSTEP Operating System Software manual.

See also:  startIOThread, startIOThreadWithPriority:



startIOThreadWithPriority:
(IOReturn)startIOThreadWithPriority:(int)priority

The same as startIOThread, except that the I/O thread runs at the specified priority. This method lets you do performance tuning by raising or lowering the thread's scheduling priority. By default, kernel I/O threads start with a priority equal to the maximum user priority (currently 18).

For more information about priorities, see Chapter 1 of the NEXTSTEP Operating System Software manual.

See also:  startIOThread, startIOThreadWithFixedPriority:



timeoutOccurred
(void)timeoutOccurred

Does nothing; subclasses that support timeouts can implement this method. See the IOEthernet class for an example of implementing this method as part of timeout support. This method is invoked by the default I/O thread (implemented by startIOThread...) whenever it receives a bodyless Mach message with an ID of IO_TIMEOUT_MSG. See the IOSCSIController class for an example of sending Mach messages.

See also:  startIOThread



waitForInterrupt:
(IOReturn)waitForInterrupt:(int *)msgID

Listens to the interrupt port until it detects a Mach message; dequeues the message if possible. This method should be invoked by the I/O thread whenever the thread needs to listen to the interrupt port. The default I/O thread provided by IODirectDevice invokes this message as described under startIOThread.

If the interrupt port hasn't been set, this message returns IO_R_NO_INTERRUPT. If the message has a body, this method leaves the message on the queue and returns IO_R_MSG_TOO_LARGE. If the message couldn't be dequeued due to another reason, this method returns IO_R_IPC_FAILURE and logs an error message.

If a message is already on the queue when this method is invoked, this method dequeues the message and then attempts to give up the processor before returning. Without this precaution, a thread with many messages queued could prevent other kernel threads from being executed.

If this method successfully detects and dequeues a message, it sets msgId to the message's ID and returns IO_R_SUCCESS.

See also:  startIOThread



Instance Methods (ISA/EISA Architecture)

abortDMABuffer:
(void)abortDMABuffer:(IOEISADMABuffer)buffer

Frees the memory allocated to buffer. If a read transfer is in progress, the data read is lost.

See also:  freeDMABuffer:



createDMABufferFor:length:read:needsLowMemory:limitSize:
(IOEISADMABuffer)createDMABufferFor:(unsigned int *)physicalAddress length:(unsigned int)numBytes
read:(BOOL)isRead
needsLowMemory:(BOOL)lowerMem
limitSize:(BOOL)limitSize

Returns a DMA buffer for the contents of physical memory starting at physicalAddress and continuing for numBytes bytes. You should specify YES for isRead if the data will be read from the device; if the data will be written to the device, specify NO. lowerMem should be YES if the transfer must be from or to the first 16MB of physical memory (as required by some ISA devices); otherwise, it should be NO. To limit the size of the transfer to 64KB, specify limitSize as YES; otherwise, limitSize should be NO.

This method changes the physical address if necessary to accommodate the ISA bus. When the physical address is changed, the data is copied to the new physical address (if the transfer is a write), and the new physical address is returned in physicalAddress.

Returns NULL if kernel memory for the buffer couldn't be allocated.

See also:  freeDMABuffer:



currentAddressForChannel:
(unsigned int)currentAddressForChannel:(unsigned int)localChannel

Returns the physical address currently in the address register of the specified DMA channel. This method can be invoked at any time--even when DMA is in progress. This method is often used along with autoinitialize mode. It's also used to help diagnose errors when a device or channel aborts a DMA transfer.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  currentCountForChannel:, setAutoinitialize:forChannel:



currentCountForChannel:
(unsigned int)currentCountForChannel:(unsigned int)localChannel

Returns the number of bytes remaining to be transferred on the specified channel. The maximum number returned is equal to the length of the DMA buffer currently being handled by the channel. This method is often used along with autoinitialize mode. It's also used to help diagnose errors when a device or channel aborts a DMA transfer.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  currentAddressForChannel:, setAutoinitialize:forChannel:



disableAllInterrupts
(void)disableAllInterrupts

Disables all interrupts associated with this IODirectDevice, so that no interrupts can be generated by the hardware. Returns IO_R_NO_INTERRUPT if no interrupt port is attached; otherwise, returns IO_R_SUCCESS.

Note:  Even after invoking disableAllInterrupts: successfully, your driver may still receive interrupt messages for interrupts that occurred before they were disabled.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  enableAllInterrupts, disableInterrupt:



disableChannel:
(void)disableChannel:(unsigned int)localChannel

If the DMA channel corresponding to localChannel is reserved by this device, this method disables the channel. You typically disable the channel just before changing its setting. You need to invoke enableChannel: once the channel is set up so that transfers can occur.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  enableChannel:



disableInterrupt:
(void)disableInterrupt:(unsigned int)localInterrupt

Disables the interrupt corresponding to localInterrupt.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  disableAllInterrupts, enableInterrupt:



enableAllInterrupts
(IOReturn)enableAllInterrupts

Creates and attaches an interrupt port, if one isn't already attached, and enables all interrupts associated with this IODirectDevice. Returns IO_R_NO_INTERRUPT if the interrupt port couldn't be attached; otherwise, returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  attachInterruptPort, disableAllInterrupts, enableInterrupt:



enableChannel:
(IOReturn)enableChannel:(unsigned int)localChannel

Enables transfers on the DMA channel corresponding to localChannel. Returns IO_R_NOT_ATTACHED if localChannel doesn't correspond to a DMA channel or if the DMA channel isn't reserved by this device. Otherwise, returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  disableChannel:, startDMAForBuffer:channel:



enableInterrupt:
(IOReturn)enableInterrupt:(unsigned int)localInterrupt

Invokes attachInterruptPort and, if attachInterruptPort succeeds, enables the interrupt corresponding to localInterrupt and returns IO_R_SUCCESS. If attachInterruptPort doesn't succeed, returns IO_R_NOT_ATTACHED.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  disableInterrupt:, enableAllInterrupts



freeDMABuffer:
(void)freeDMABuffer:(IOEISADMABuffer)buffer

Completes the transfer associated with buffer and frees the buffer. buffer should be a value returned by createDMABufferFor:.... If createDMABufferFor:... changed the physical address and the transfer is a read, this method moves the data from the new physical address to the old one. In other words, any data that's read appears at the address passed to createDMABufferFor:... in the physicalAddress argument, not at the address returned in physicalAddress.

See also:  abortDMABuffer:, createDMABufferFor:length:read:needsLowMemory:limitSize:



getDMATransferWidth:forChannel:
(IOReturn)getDMATransferWidth:(IOEISADMATransferWidth *)width forChannel:(unsigned int)localChannel

Returns in width the width currently used for DMA transfers on the specified channel. The width can be 8-bit (IO_8Bit), 16-bit (IO_16BitByteCount), or 32-bit (IO_32Bit). On EISA systems, you can set the width using setDMATransferWidth:forChannel:.

If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG. If the DMA channel isn't reserved by this device, this method does nothing and returns IO_R_NOT_ATTACHED. Otherwise, this method returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  setDMATransferWidth:forChannel:



getEISAId:forSlot:
(BOOL)getEISAId:(unsigned int *)id forSlot:(int)slotNumber

Returns in id the EISA id for the specified slot. Returns YES if the slot is a valid EISA slot; otherwise, returns NO. You can use this method to loop through the computer's slots, testing each slot for whether it contains a particular card. For example, the following code is executed in the QVision display driver's initFromDeviceDescription: method to determine whether QVision hardware is present in the system.

adapter = UnknownAdapter;
for (slot = 1; slot <= 0xF; slot++) {
if ([self getEISAId:&product_id forSlot:slot] == YES) {
switch (product_id) {
case QVISION_EISA_ID:
adapter = QVisionAdapter;
break;
case ORION_EISA_ID:
adapter = OrionAdapter;
break;
case ORION12_EISA_ID:
adapter = Orion12Adapter;
break;
case QVISION_ISA_ID:
case ORION_ISA_ID:
case ORION12_ISA_ID:
IOLog("%s: Sorry, ISA cards are not supported.\n",
[self name]);
break;
}
break;
}
}

See also:  isEISAPresent



getHandler:level:argument:forInterrupt:
(BOOL)getHandler:(IOInterruptHandler *)handler
level:(unsigned int *)ipl
argument:(unsigned int *)arg
forInterrupt:(unsigned int)localInterrupt

Does nothing and returns NO. Subclasses can implement this method to specify a function to directly handle the interrupt specified by localInterrupt. This method is invoked every time an interrupt is enabled.

If this method returns YES, interrupts from the device result directly in a call to handler, with the driver-dependent argument arg, at interrupt level ipl. Otherwise, interrupts result in a Mach message to the instance's interrupt port.

If you implement this method, you should use interrupt level 3 (IPLDEVICE, as defined in kernserv/i386/spl.h) unless a higher interrupt level is absolutely necessary. Using interrupt levels greater than 3 requires great care and a good grasp of NeXT kernel internals.

Note:  The interrupt level is different from the interrupt number (which is also known as the IRQ). The kernel handles interrupts on each of the 15 IRQs at an interrupt level between 0 and 7; the default is 3. The interrupt level determines which devices can interrupt; specifically, only devices with an interrupt level higher than the current interrupt level can interrupt. For example, a device that interrupts using IRQ 9 might have a direct interrupt handler that runs at interrupt level 3. While this interrupt handler is running, other devices with handlers that run at interrupt level 3 can't interrupt the CPU.

Here's a typical implementation of this method:

- (BOOL) getHandler:(IOEISAInterruptHandler *)handler
level:(unsigned int *) ipl
argument:(unsigned int *) arg
forInterrupt:(unsigned int) localInterrupt
{
*handler = myIntHandler;
*ipl = IPLDEVICE;
*arg = 0;
return YES;
}

In the example above, myIntHandler is the function that handles the interrupt. It might be implemented as follows (fill in the italicized text between angle brackets, that is << >>, with device-specific code):



static void myIntHandler(void *identity, void *state,
unsigned int arg)
{
<< . . . Do what we must at interrupt level . . . >>
if (<< I/O thread doesn't need to know about this interrupt >>)
return;

/* Forward this to the I/O thread for further handling. */
IOSendInterrupt(identity, state, IO_DEVICE_INTERRUPT_MSG);
}

See also:  IOSendInterrupt()



initFromDeviceDescription:
initFromDeviceDescription:deviceDescription

Initializes and returns the IODirectDevice instance. Records deviceDescription as the IODeviceDescription corresponding to this IODirectDevice. Reserves all the interrupts, DMA channels, and I/O ports specified in deviceDescription. If any resources can't be reserved, releases all resources and returns nil.

This method must be invoked before any methods that require local equivalents of resources can be used. For example, mapMemoryRange:... requires that you specify the local equivalent of a memory range. However, IODirectDevices don't know what memory ranges they can use until initFromDeviceDescription: has been invoked. This means, for example, that subclass implementations of initFromDeviceDescription: must invoke the superclass's implementation of initFromDeviceDescription: before they can map any memory ranges or do anything else that requires access to resources.




isDMADone:
(BOOL)isDMADone:(unsigned int)localChannel

Returns YES if DMA has completed on the specified channel; otherwise, returns NO. If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.



isEISAPresent
(BOOL)isEISAPresent

Returns YES if the computer conforms to the EISA specification; otherwise, returns NO.

See also:  getEISAId:forSlot:



mapMemoryRange:to:findSpace:cache:
(IOReturn)mapMemoryRange:(unsigned int)localMemoryRange to:(vm_address_t *)destinationAddress
findSpace:(BOOL)findSpace
cache:(IOCache)caching

Maps the device memory corresponding to localMemoryRange into the calling task's address space. localMemoryRange is the local range number in the device description.

If findSpace is TRUE, this method ignores the destinationAddress and determines where the mapped memory should go, returning the value in destinationAddress. If findSpace is FALSE, this method truncates destinationAddress to the nearest page boundary, maps the memory to the truncated address, and returns the truncated address.

The caching argument determines how the memory is cached. Usually, it should be IO_WriteThrough. However, if caching seems to be causing problems, try using IO_CacheOff instead.

If localMemoryRange doesn't correspond to one of this device's memory ranges, IO_R_INVALID_ARG is returned. There must also be more than one I/O port range associated with the device (i.e. [deviceDescription numPortRanges] > 1); otherwise IO_R_INVALID_ARG is returned. If the mapping couldn't be performed for another reason, IO_R_NO_SPACE is returned. If the mapping was successful, returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  unmapMemoryRange:from:



releaseChannel:
(void)releaseChannel:(unsigned int)localChannel

Releases the DMA channel corresponding to localChannel so that another device can use the channel.

See also:  reserveChannel:



releaseDMALock
(void)releaseDMALock

Releases the lock associated with DMA. This method panics if this IODirectDevice doesn't hold the DMA lock.

Most drivers don't need to use DMA locking. However, the floppy drive (and possibly other devices) tends to have DMA underruns when the bus is saturated. As a result, the floppy driver and drivers for devices that tend to saturate the bus use DMA locking to avoid performing I/O at the same time. DMA locking is ignored by all other device drivers.

You don't have to use DMA locking unless your device is having DMA underruns or is causing another device to have underruns. Sometimes these underruns occur on ISA computers, but not EISA ones. If your device is causing the floppy drive to have underruns, you'll see the following error on the console while your device is performing I/O:

fd0: DMA Over/underrun

See also:  reserveDMALock



releaseInterrupt:
(void)releaseInterrupt:(unsigned int)localInterrupt

Releases the interrupt corresponding to localInterrupt so that another device can use the interrupt.

See also:  reserveInterrupt:



releasePortRange:
(void)releasePortRange:(unsigned int)localPortRange

Releases the range of I/O ports corresponding to localPortRange.

See also:  reservePortRange:



reserveChannel:
(IOReturn)reserveChannel:(unsigned int)localChannel

Reserves the DMA channel corresponding to localChannel so that no other device can use the channel. Returns IO_R_NOT_ATTACHED if localChannel doesn't correspond to a DMA channel or if the DMA channel is reserved by another device. Otherwise, returns IO_R_SUCCESS.

You don't normally have to invoke this method, since initFromDeviceDescription: reserves all the device's DMA channels.

See also:  releaseChannel:



reserveDMALock
(void)reserveDMALock

Reserves the lock associated with DMA. See releaseDMALock for information on DMA locking.



reserveInterrupt:
(IOReturn)reserveInterrupt:(unsigned int)localInterrupt

Reserves the interrupt corresponding to localInterrupt so that no other device can use it. Returns IO_R_NOT_ATTACHED if localInterrupt doesn't correspond to an interrupt or if another device has already reserved the interrupt. Otherwise, returns IO_R_SUCCESS.

You don't normally have to invoke this method, since initFromDeviceDescription: reserves all the device's interrupts.

See also:  releaseInterrupt:



reservePortRange:
(IOReturn)reservePortRange:(unsigned int)localPortRange

Releases the range of I/O ports corresponding to localPortRange and returns IO_R_SUCCESS.

You don't normally have to invoke this method, since initFromDeviceDescription: reserves all the device's I/O ports.

See also:  releasePortRange:



setAutoinitialize:forChannel:
(IOReturn)setAutoinitialize:(BOOL)flag forChannel:(unsigned int)localChannel

Sets the specified channel's autoinitialize DMA mode to on if flag is YES; otherwise, sets it off. The new autoinitialize mode stays in effect until this method is invoked again or the computer is rebooted. By default, autoinitialize mode is disabled.

If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG. If the DMA channel isn't reserved by this device, this method does nothing and returns IO_R_NOT_ATTACHED. Otherwise, this method returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  setIncrementMode:forChannel:, setTransferMode:forChannel:



setDMATiming:forChannel:
(IOReturn)setDMATiming:(IOEISADMATiming)timing forChannel:(unsigned int)localChannel

Makes the specified channel use the specified DMA bus cycle--ISA-compatible (IO_Compatible), Type A (IO_TypeA), Type B (IO_TypeB), or burst (IO_Burst), which is also known as Type C. This method is valid only on EISA systems.

If the system is ISA-based, this method does nothing and returns IO_R_UNSUPPORTED. If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG. If the DMA channel isn't reserved by this device, this method does nothing and returns IO_R_NOT_ATTACHED. Otherwise, this method returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.



setDMATransferWidth:forChannel:
(IOReturn)setDMATransferWidth:(IOEISADMATransferWidth)width forChannel:(unsigned int)localChannel

Makes the specified channel use the specified width for DMA transfers. The width can be 8-bit (IO_8Bit), 16-bit (IO_16BitByteCount), or 32-bit (IO_32Bit). The 16-bit mode requires byte counting, not word counting (which is unsupported). This method is valid only on EISA systems.

If the system is ISA-based, this method does nothing and returns IO_R_UNSUPPORTED. If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG. If the DMA channel isn't reserved by this device, this method does nothing and returns IO_R_NOT_ATTACHED. Otherwise, this method returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.



setEOPAsOutput:forChannel:
(IOReturn)setEOPAsOutput:(BOOL)flag forChannel:(unsigned int)localChannel

Selects whether the specified channel's EOP pin is an output signal (the default) or an input signal. This method is valid only on EISA systems.

If the system is ISA-based, this method does nothing and returns IO_R_UNSUPPORTED. If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG. If the DMA channel isn't reserved by this device, this method does nothing and returns IO_R_NOT_ATTACHED. Otherwise, this method returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.



setIncrementMode:forChannel:
(IOReturn)setIncrementMode:(IOIncrementMode)mode forChannel:(unsigned int)localChannel

This method lets the driver specify how the start address and length of its DMA buffers should be interpreted. By default, the increment mode is IO_Increment, so each DMA buffer is interpreted so that if the start address is n and the length is m, the data in addresses n through n + m 1 are transferred. By setting the increment mode to IO_Decrement, however, the driver specifies that the affected addresses should be n through n m + 1. The new increment mode is in effect until this method is invoked again or until the computer is rebooted.

If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG. If the DMA channel isn't reserved by this device, this method does nothing and returns IO_R_NOT_ATTACHED. Otherwise, this method returns IO_R_SUCCESS.

Note:  IO_Decrement mode is not currently supported.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  setAutoinitialize:forChannel:, setTransferMode:forChannel:



setStopRegisterMode:forChannel:
(IOReturn)setStopRegisterMode:(IOEISAStopRegisterMode)mode forChannel:(unsigned int)localChannel

Enables or disables the specified channel's Stop register. By default, the Stop register is disabled. You can enable it by specifying mode to be IO_StopRegisterEnable. This method is valid only on EISA systems.

Note:  Enabling the Stop register isn't currently supported.

If the system is ISA-based or mode is IO_StopRegisterEnable, this method does nothing and returns IO_R_UNSUPPORTED. If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG. If the DMA channel isn't reserved by this device, this method does nothing and returns IO_R_NOT_ATTACHED. Otherwise, this method returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.



setTransferMode:forChannel:
(IOReturn)setTransferMode:(IODMATransferMode)mode forChannel:(unsigned int)localChannel

Sets the specified channel's transfer mode to mode. The new transfer mode stays in effect until this method is invoked again or the computer is rebooted.

If localChannel doesn't correspond to a DMA channel, this method does nothing and returns IO_R_INVALID_ARG. If the DMA channel isn't reserved by this device, this method does nothing and returns IO_R_NOT_ATTACHED. Otherwise, this method returns IO_R_SUCCESS.

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.

See also:  setAutoinitialize:forChannel:, setIncrementMode:forChannel:



startDMAForBuffer:channel:
(IOReturn)startDMAForBuffer:(IOEISADMABuffer)buffer channel:(unsigned int)localChannel

Begins DMA with buffer on the DMA channel specified by localChannel, and returns IO_R_SUCCESS. DMA isn't started if localChannel doesn't correspond to a DMA channel (in which case IO_R_INVALID_ARG is returned), if the DMA channel isn't assigned, or if no DMA frames could be allocated (IO_R_NO_FRAMES is returned).

Because this method uses a local equivalent of a resource, it can't be invoked until after this category's implementation of initFromDeviceDescription: is invoked.



unmapMemoryRange:from:
(void)unmapMemoryRange:(unsigned int)localMemoryRange
from:(vm_address_t)address

Unmaps the device memory corresponding to localMemoryRange from the calling task's address space. The value of address must be the same as the value returned by the destinationAddress argument of mapMemoryRange:to:findSpace:cache: for the same localMemoryRange.

See also:  mapMemoryRange:to:findSpace:cache:



Class Methods (PCI Architecture)

getPCIConfigData:atRegister:withDeviceDescription:
+ (IOReturn)getPCIConfigData:(unsigned long *)data
atRegister:(unsigned char)address
withDeviceDescription:description

Reads from the device's configuration space at the byte address address using the IOPCIDeviceDescription description. All accesses are 32 bits wide and address must be aligned as such.



getPCIConfigSpace:withDeviceDescription:
+ (IOReturn)getPCIConfigSpace:(IOPCIConfigSpace *)configurationSpace withDeviceDescription:description

Reads the device's entire configuration space using the IOPCIDeviceDescription description. Returns IO_R_SUCCESS if successful. If this method fails, the driver should make no assumptions about the state of the data returned in the IOPCIConfigSpace struct.



isPCIPresent
+ (BOOL)isPCIPresent

Returns YES if PCI Bus support is enabled. Returns NO otherwise.



setPCIConfigData:atRegister:withDeviceDescription:
+ (IOReturn)setPCIConfigData:(unsigned long)data
atRegister:(unsigned char)address
withDeviceDescription:description

Writes to the device's configuration space at the byte address address using the IOPCIDeviceDescription description. All accesses are 32 bits wide and address must be aligned as such.



setPCIConfigSpace:withDeviceDescription:
+ (IOReturn)setPCIConfigSpace:(IOPCIConfigSpace *)configurationSpace withDeviceDescription:description

Writes the device's entire configuration space using the IOPCIDeviceDescription description. Returns IO_R_SUCCESS if successful. If this method fails, the driver should make no assumptions about the state of the device's configuration space.



Instance Methods (PCI Architecture)

getPCIConfigData:atRegister:
(IOReturn)getPCIConfigData:(unsigned long *)data
atRegister:(unsigned char)address

Reads from the device's configuration space at the byte address address. All accesses are 32 bits wide and address must be aligned as such.



getPCIConfigSpace:
(IOReturn)getPCIConfigSpace:(IOPCIConfigSpace *)configurationSpace

Reads the device's entire configuration space. Returns IO_R_SUCCESS if successful. If this method fails, the driver should make no assumptions about the state of the data returned in the IOPCIConfigSpace struct.



isPCIPresent
(BOOL)isPCIPresent

Returns YES if PCI Bus support is enabled. Returns NO otherwise.



setPCIConfigData:atRegister:
(IOReturn)setPCIConfigData:(unsigned long)data
atRegister:(unsigned char)address

Writes to the device's configuration space at the byte address address. All accesses are 32 bits wide and address must be aligned as such.



setPCIConfigSpace:
(IOReturn)setPCIConfigSpace:(IOPCIConfigSpace *)configurationSpace

Writes the device's entire configuration space. Returns IO_R_SUCCESS if successful. If this method fails, the driver should make no assumptions about the state of the device's configuration space.



Instance Methods (PCMCIA Architecture)

mapAttributeMemoryTo:findSpace:
(IOReturn)mapAttributeMemoryTo:(vm_address_t *)destinationAddress findSpace:(BOOL)findSpace

Maps attribute memory to destinationAddress in findSpace.

See also:  unmapAttributeMemory:



unmapAttributeMemory:
(void)unmapAttributeMemory

Unmaps attribute memory.

See also:  mapAttributeMemoryTo:findSpace: