by Ralf Brown
Updated 6/88, 4/90, 9/92
What is an interrupt?
An interrupt is a hardware signal that tells the CPU to
temporarily stop what it is doing and go do something else.
Without interrupts, the CPU would have to constantly check for
external events; with interrupts, the CPU can work on
something else and still respond to an event as soon as it
CPUs typically have an instruction to disable interrupts for
use when a section of code has to run without being disturbed
by external events. Because of this, most CPUs also have a
special interrupt called a Non-Maskable Interrupt (NMI), which
is responded to even when all other interrupts are disabled.
The NMI is used to signal calamities such as memory failure or
imminent power loss.
Why so many different interrupts?
The 8086 family of processors has the ability to recognize 256
different interrupts. They also have the ability to let a
program invoke any of these interrupts with a special
instruction, known as a software interrupt (as opposed to a
hardware interrupt which is signalled from outside the
processor). Software interrupts are treated just like
hardware interrupts, except that they are never disabled and
do not result in an acknowledgement to other chips in the
computer. The software interrupt instruction on the 8086
family is called INT, and is given the number of the
interrupt. Thus an INT 21h instruction invokes interrupt
number 33 decimal.
Other processors also have software interrupts, though they
often use different names, such as the Motorola 68000 family
TRAP instruction, the Intel 8080 RST (ReSTart) instruction, or
many mainframes' SVC (SuperVisor Call).
Since a program can invoke an interrupt by number rather than
by its address (as it has to in calling subroutines),
interrupts are a convenient way of providing services without
having to recompile a program whenever the address of the code
providing the service changes. This also allows a user
program to enhance the services provided by directing the
interrupt to itself. These enhanced services can then be made
available to other programs.
How does an interrupt work?
The 8086 reserves the lowest 1024 bytes of memory for a table
containing the addresses for each of the 256 possible
interrupts. When an interrupt occurs (hardware or software),
the processor multiplies its number by 4 and looks at the
resulting memory location to find the address of the piece of
code which handles the interrupt. It then places the current
address in the program and the processor flags on the stack,
and jumps to the beginning of the interrupt handler.
When the interrupt handler finishes, it invokes a special
instruction to return from the interrupt. This instruction
takes the previously saved flags and program address off of
the stack and places them back in the appropriate registers in
The interrupt handler has to be careful to preserve any
registers that it uses which are not used to communicate
results to the program that invoked the interrupt. If the
interrupt can be triggered by a hardware interrupt (only
certain ones can on IBM PC's, XT's, and AT's), then the
interrupt handler has to preserve ALL registers, since the
interrupt could have happened anywhere.
API (Application Program[ming] Interface)
An API is the set of function calls and services that a
program makes available to other processes (applications).
Each function or service has a set format which specifies
the values to be supplied by the caller and the values
which are returned. Because of this interface
specification, the underlying organization of the
function or service can be changed without affecting the
applications which use it. For example, the DOS INT 21h
file access functions remained unchanged between DOS 2.x
and DOS 3.x, even though the internal data structures and
code organization changed significantly.
NMI (Non-Maskable Interrupt)
Most external (hardware) interrupts can be disabled by the
CLI (CLear Interrupt enable flag) instruction when the CPU
is executing critical code that should not be interrupted,
such as switching from one stack to another. However, there
are some situations so dire that the CPU must act on them
immediately no matter what else it is doing, even if it has
disabled interrupts. The Non-Maskable Interrupt serves
precisely this purpose, as it cannot be disabled (masked) by