InterruptAttach(), InterruptAttach_r()

Attach an interrupt handler to an interrupt source

Synopsis:

#include <sys/neutrino.h>

int InterruptAttach( int intr,
       const struct sigevent * (* handler)(void *, int),
       const void * area,
       int size,
       unsigned flags );

int InterruptAttach_r( int intr,
       const struct sigevent * (* handler)(void *, int),
       const void * area,
       int size,
       unsigned flags );

Arguments:

intr

The interrupt that you want to attach a handler to; see “Interrupt vector numbers,” below.

handler

A pointer to the handler function; see “Interrupt handler function,” below.

area

A pointer to a communications area in your process, or NULL if you don't want a communications area.

size

The size of the communications area; this should be 0 if area is NULL. InterruptAttach() currently ignores this argument.

flags

Flags that specify how you want to attach the interrupt handler; a bitwise OR of zero or more of the following:

_NTO_INTR_FLAGS_END
_NTO_INTR_FLAGS_PROCESS
_NTO_INTR_FLAGS_TRK_MSK

For more information, see “Flags,” below.

Library:

libc

Use the -l c option to qcc to link against this library. This library is usually included automatically.

The InterruptAttach() and InterruptAttach_r() kernel calls attach the interrupt function handler to the hardware interrupt specified by intr. They automatically enable (i.e., unmask) the interrupt level.

These functions are identical except in the way they indicate errors. See the Returns section for details.

Before calling either of these functions, the thread must obtain I/O privileges by successfully calling:

ThreadCtl( _NTO_TCTL_IO, 0 );

If the thread doesn't do this, the attachment fails with an error code of EPERM.

On a multicore system, the interrupt handler runs on the CPU that takes the interrupt.

Interrupt vector numbers

The interrupt values for intr are logical interrupt vector numbers grouped into related “interrupt classes” that generally correspond to a particular interrupt controller on the CPU. The following interrupt classes are present on all QNX Neutrino systems:

_NTO_INTR_CLASS_EXTERNAL

Normal external interrupts (such as the ones generated by the INTR pin on x86 CPUs).

_NTO_INTR_CLASS_SYNTHETIC

Synthetic, kernel-generated interrupts:

_NTO_HOOK_IDLE — used when a processor becomes idle; see InterruptHookIdle().
_NTO_HOOK_TRACE — used by the instrumented kernel to notify the data-capture program that event data is available; see InterruptHookTrace().
_NTO_INTR_SPARE — an interrupt that's guaranteed not to match any valid logical interrupt vector number. It's like a NULL pointer; you use it (for example) to indicate that a board doesn't have a cascade vector.

_NTO_INTR_SPARE is usually the only _NTO_INTR_CLASS_SYNTHETIC interrupt you'll use directly.

There can be additional interrupt classes defined for specific CPUs or embedded systems. For the interrupt assignments for specific boards, see:

some *intr.h files in the platform-specific directories under ${QNX_TARGET}/usr/include
the sample buildfiles in ${QNX_TARGET}/x86/boot/build
the buildfiles in the Board Support Package for your board

The mapping of logical interrupt vector numbers is completely dependent on the implementer of the startup code. Device drivers should generally allow runtime configuration of interrupt numbers.

Interrupt handler function

The function to call is specified by the handler argument. Its prototype is:

const struct sigevent* handler( void* area, int id );

The arguments are:

area: The same as the area passed to InterruptAttach(). This pointer— which could be NULL—and the size argument to InterruptAttach() define a communications area in your process. This typically is a structure containing buffers and information that the handler and the process need to share.
id: The ID returned by InterruptAttach().

The handler function runs in the environment of your process.

Follow the following guidelines when writing your handler:

A temporary interrupt stack of limited depth is provided at interrupt time, so avoid placing large arrays or structures on the stack frame of the handler, and avoid recursion. It's safe to assume that about 200 bytes of stack are available.
The interrupt handler runs asynchronously with the threads in the process. Any variables modified by the handler should be declared with the volatile keyword and modified with interrupts disabled or using atomic operations in any thread and ISR.
The interrupt handler should be kept as short as possible. If a significant amount of work needs to be done, the handler should deliver an event to awaken a thread to do the work.
The handler can't call library routines that contain kernel calls except for InterruptDisable(), InterruptEnable(), InterruptLock(), InterruptMask(), InterruptUnlock(), and InterruptUnmask(). For a list of the functions that you can call from the interrupt handler, see the Summary of Safety Information appendix.
The handler can call TraceEvent(), but not all of its commands are valid; see the Caveats in its entry.
It isn't safe to use floating-point operations in Interrupt Service Routines.

The handler function should return NULL or a pointer to a valid sigevent structure that the kernel delivers. These events are defined in <signal.h>. Consider the following when choosing the event type:

Message-driven processes that block in a receive loop using MsgReceivev() should consider using SIGEV_PULSE to trigger a pulse.
Threads that block at a particular point in their code and don't go back to a common receive point should consider using SIGEV_INTR as the event notification type and InterruptWait() as the blocking call.

The thread that calls InterruptWait() must be the one that called InterruptAttach().
Using SIGEV_THREAD is very inefficient, because that would create a new thread for every interrupt.
For SIGEV_SIGNAL, SIGEV_SIGNAL_CODE, and SIGEV_SIGNAL_THREAD, handling a signal using a signal handler is noticeably inefficient as compared to using pulses. But, delivering a signal to a thread that's blocked on sigwaitinfo() is actually more efficient than the pulse choice, though less efficient than the InterruptWait() choice. The advantage is in flexibility, allowing the thread to be woken up both by interrupts and by other threads in the process as needed. Architecturally, this lets you create a thread that's dedicated to handling hardware.
On a multicore system, the thread that receives the event returned by the interrupt handler runs on any CPU, limited only by the scheduler and the runmask. If you know which CPU the handler will run on and which thread will be handling the interrupt, you may wish to consider using the runmask to make sure they're the same CPU, in order to reduce cache-coherency operations on data shared between the interrupt handler and the thread.

Flags

The flags argument is a bitwise OR of the following values, or 0:

_NTO_INTR_FLAGS_END

Put the new handler at the end of the list of existing handlers (for shared interrupts) instead of the start.

The interrupt structure allows hardware interrupts to be shared. For example, if two processes take over the same physical interrupt, both handlers are invoked consecutively. When a handler attaches, it's placed in front of any existing handlers for that interrupt and is called first. You can change this behavior by setting _NTO_INTR_FLAGS_END. Although the Neutrino microkernel allows full interrupt sharing, your hardware might not. For example, the ISA bus doesn't allow interrupt sharing, while the PCI bus does.

Processor interrupts are enabled during the execution of the handler. Don't attempt to talk to the interrupt controller chip. The operating system issues the end-of-interrupt command to the chip after processing all handlers at a given level.

The first process to attach to an interrupt unmasks the interrupt. When the last process detaches from an interrupt, the system masks it.

If the thread that attached the interrupt handler terminates without detaching the handler, the kernel does it automatically.

_NTO_INTR_FLAGS_PROCESS

Associate the handler with the process instead of the attaching thread. The interrupt handler is removed when the process exits, instead of when the attaching thread exits.

_NTO_INTR_FLAGS_TRK_MSK

Track calls to InterruptMask() and InterruptUnmask() to make detaching the interrupt handler safer.

The _NTO_INTR_FLAGS_TRK_MSK flag and the id argument to InterruptMask() and InterruptUnmask() let the kernel track the number of times a particular interrupt handler or event has been masked. Then, when an application detaches from the interrupt, the kernel can perform the proper number of unmasks to ensure that the interrupt functions normally. This is important for shared interrupt levels.

You should always set _NTO_INTR_FLAGS_TRK_MSK.

Blocking states

These calls don't block.

Returns:

An interrupt function ID. If an error occurs:

InterruptAttach() returns -1 and sets errno.
InterruptAttach_r() returns the negative of a value from the Errors section and doesn't set errno.

Use the function ID with the InterruptDetach() function to detach this interrupt handler.

Errors:

EAGAIN: All kernel interrupt entries are in use.
EFAULT: A fault occurred when the kernel tried to access the buffers provided.
EINVAL: The value of intr isn't a valid interrupt number.
EPERM: The process doesn't have I/O privileges.

Classification:

QNX Neutrino

Safety:
Cancellation point	No
Interrupt handler	No
Signal handler	Yes
Thread	Yes

Caveats:

If you're writing a multithreaded resource manager using the thread_pool_*() functions, and the thread that has called InterruptAttach() is also joining the thread pool by calling thread_pool_start() with POOL_FLAG_USE_SELF set, then that thread must also set _NTO_INTR_FLAGS_PROCESS in the flags argument when calling InterruptAttach(). If it doesn't, the thread may exit as part of the thread pool's dynamic threading behavior, and the interrupt handling will then be lost.