sigevent

Structure that describes an event

Synopsis:

#include <sys/siginfo.h>

union __sigval32 {
        int                     sival_int;
        _Uintptr32t     sival_ptr;
};

union __sigval64 {
        int                     sival_int;
        _Uintptr64t     sival_ptr;
};

union sigval {
    int         sival_int;
    void       *sival_ptr;
};

struct __sigevent32 {
        int                             sigev_notify;
        union {
                int                     __sigev_signo;
                int                     __sigev_coid;
                int                     __sigev_id;
                _Uintptr32t             __sigev_notify_function;
                _Uintptr32t             __sigev_addr;
                unsigned                __sigev_handle;
        } __sigev_un1;
        union __sigval32 sigev_value;
        union {
                struct {
                        short           __sigev_code;
                        short           __sigev_priority;
                } __st;
                _Uintptr32t             __sigev_notify_attributes;
                int                     __sigev_memop;
        } __sigev_un2;
};

struct __sigevent64 {
        int                             sigev_notify;
        union {
                int                     __sigev_signo;
                int                     __sigev_coid;
                int                     __sigev_id;
                _Uintptr64t             __sigev_notify_function;
                _Uintptr64t             __sigev_addr;
                unsigned                __sigev_handle;
        } __sigev_un1;
        union __sigval64 sigev_value;
        union {
                struct {
                        short           __sigev_code;
                        short           __sigev_priority;
                } __st;
                _Uintptr64t             __sigev_notify_attributes;
                int                     __sigev_memop;
        } __sigev_un2;
};

struct sigevent {
        int                             sigev_notify;
        union {
                int                     __sigev_signo;
                int                     __sigev_coid;
                int                     __sigev_id;
                void                    (*__sigev_notify_function)(union sigval);
                volatile unsigned       *__sigev_addr;
                unsigned                __sigev_handle;
        } __sigev_un1;
        union sigval    sigev_value;
        union {
                struct {
                        short           __sigev_code;
                        short           __sigev_priority;
                } __st;
                pthread_attr_t          *__sigev_notify_attributes;
                int                     __sigev_memop;
        } __sigev_un2;
};

# define sigev_signo               __sigev_un1.__sigev_signo
# define sigev_coid                __sigev_un1.__sigev_coid
# define sigev_id                  __sigev_un1.__sigev_id
# define sigev_notify_function     __sigev_un1.__sigev_notify_function
# define sigev_addr                __sigev_un1.__sigev_addr
# define sigev_handle              __sigev_un1.__sigev_handle

# define sigev_notify_attributes   __sigev_un2.__sigev_notify_attributes
# define sigev_code                __sigev_un2.__st.__sigev_code
# define sigev_priority            __sigev_un2.__st.__sigev_priority
# define sigev_memop               __sigev_un2.__sigev_memop

Description:

The sigevent structure describes an event. POSIX specifies the basic structure and allows a lot of leeway in extending it. It effectively includes these members (although they aren't all usable at the same time):

Member Classification
int sigev_notify POSIX
int sigev_signo POSIX
int sigev_coid QNX Neutrino
int sigev_id (not currently used) QNX Neutrino
void sigev_notify_function POSIX
volatile unsigned *sigev_addr QNX Neutrino
unsigned sigev_handle QNX Neutrino
union sigval sigev_value POSIX
short sigev_code QNX Neutrino
short sigev_priority QNX Neutrino
pthread_attr_t *sigev_notify_attributes POSIX
int sigev_memop QNX Neutrino

The value of sigev_notify indicates how the notification is to occur and which of the other members are used:

Note: The library uses some of the extra bits in sigev_notify for other purposes (see Hidden bits,” below). After initializing an event, don't set sigev_notify directly; instead, use the SIGEV_SET_TYPE() macro. For example:
SIGEV_SET_TYPE(&my_event, SIGEV_PULSE)

If you want to test the value of this member, use the SIGEV_GET_TYPE() macro. For example, instead of:

if( my_event.sigev_notify == SIGEV_PULSE) 

use:

if( SIGEV_GET_TYPE(&my_event) == SIGEV_PULSE) 

(QNX Neutrino 7.0.1 or later) The kernel uses the sigev_handle field for registered events, which are more secure than regular sigevents. For more information, see the entry for MsgRegisterEvent(), as well as Events in the “Interprocess Communication (IPC)” chapter of the System Architecture guide.

The <sys/siginfo.h> file also defines some macros to make initializing the sigevent structure easier. All the macros take a pointer to a sigevent structure as their event argument and set the sigev_notify member to the appropriate value. These macros are QNX Neutrino extensions and are described below.

SIGEV_INTR (QNX Neutrino extension)

Send an interrupt notification to a specific thread, depending on the situation:

No other fields in the structure are used.

The initialization macro is:

SIGEV_INTR_INIT( &event )

SIGEV_MEMORY (QNX Neutrino extension)

(QNX Neutrino 6.6 or later) Directly manipulate a location in memory. The following fields are used:

volatile unsigned *sigev_addr
The address of the data that you want to manipulate.
int sigev_value.sival_int
The value to use in the operation.
int sigev_memop
The operation to perform; one of the following:
  • SIGEV_MEM_ASSIGN — set the location in memory to the given value.
  • SIGEV_MEM_ADD — use atomic_add() to add the value to the location in memory.
  • SIGEV_MEM_SUB — use atomic_sub() to subtract the value from the location in memory.
  • SIGEV_MEM_BITSET — use atomic_set() to set the bits in the location in memory.
  • SIGEV_MEM_BITCLR — use atomic_clr() to clear the bits in the location in memory.
  • SIGEV_MEM_BITTOGGLE — use atomic_toggle() to toggle the bits in the location in memory.

For the SIGEV_MEM_BIT* operations, the bits that are set in the value indicate which bits to set, clear, or toggle in the location in memory.

The initialization macro is:

SIGEV_MEMORY_INIT( &event, addr, value, operation )

One use for this type of event is to indicate when a local cache has been invalidated. For example, confstr(CS_TIMEZONE, ...) is called a lot, but the time zone hardly ever changes. A client can register an event to be delivered when the time zone configuration value changes, like this:
procmgr_event_notify_add( PROCMGR_EVENT_CONFSTR, my_sigevent );
and then call confstr(CS_TIMEZONE, ...) once and cache the result. The process manager will notify the client by setting a memory location when the time zone string might have changed. If somebody wants to get the time zone, and the memory location hasn't been set, the local cached value is still current. If the memory location is set, they need to call confstr(CS_TIMEZONE, ...) again.
Note: If a client program replaces its process image (e.g., by calling exec*()) with a setuid or setgid binary, the kernel sets the _NTO_COF_INSECURE flag on the channel. If this flag is set, calls to MsgDeliverEvent() with an event type of SIGEV_MEMORY or SIGEV_THREAD fail with an error of EACCES. In QNX Neutrino 7.0 and later, only the kernel can clear this flag; currently it stays set until the connection is detached.

SIGEV_NONE (POSIX)

Don't send any notification. No other fields in the structure are used.

The initialization macro is:

SIGEV_NONE_INIT( &event )

SIGEV_PULSE (QNX Neutrino extension)

Send a pulse. The following fields are used:

int sigev_coid
The connection ID. This should be attached to the channel with which the pulse will be received.
short sigev_priority
The priority of the pulse, or SIGEV_PULSE_PRIO_INHERIT if you want the thread that receives the pulse to run at the initial priority of the process.

The priority of the pulse must be in the range for the target process, or (in QNX Neutrino 6.6 or later) that process must have the PROCMGR_AID_PRIORITY ability enabled (see procmgr_ability()). In QNX Neutrino 6.6 or later, if procnto was started with an “s” appended to the -P option, then out-of-range priority requests use the maximum allowed value instead of resulting in an error.

short sigev_code
A code to be interpreted by the pulse handler. Although sigev_code can be any 8-bit signed value, you should avoid sigev_code values less than zero in order to avoid conflict with kernel or pulse codes generated by a QNX Neutrino manager. These codes all start with _PULSE_CODE_ and are defined in <sys/neutrino.h>; for more information, see the documentation for the _pulse structure. A safe range of pulse values is _PULSE_CODE_MINAVAIL to _PULSE_CODE_MAXAVAIL.
int sigev_value.sival_int or void *sigev_value.sival_ptr
A 32-bit value or a pointer to be interpreted by the pulse handler.

The initialization macros are:

SIGEV_SIGNAL (POSIX)

Send a signal to a process. The following fields are used:

int sigev_signo
The signal to raise. This must be in the range from 1 through NSIG − 1.

The initialization macro is:

SIGEV_SIGNAL_INIT( &event, signal )

If you need to set the sigev_value for a SIGEV_SIGNAL event (for example if SA_SIGINFO is set), you can use one of these macros instead of SIGEV_SIGNAL_INIT():

SIGEV_SIGNAL_CODE (QNX Neutrino extension)

This is similar to SIGEV_SIGNAL, except that SIGEV_SIGNAL_CODE also includes a value and a code. The following fields are used:

int sigev_signo
The signal to raise. This must be in the range from 1 through NSIG − 1.
short sigev_code
A code to be interpreted by the signal handler. This must be in the range from SI_MINAVAIL through SI_MAXAVAIL.
int sigev_value.sival_int or void *sigev_value.sival_ptr
A 32-bit value or a pointer to be interpreted by the signal handler.

The initialization macros are:

These macros store the code in sigev_code.

SIGEV_SIGNAL_THREAD (QNX Neutrino extension)

Send a signal to a specific thread, depending on the situation:

In the case of timers, SyncMutexEvent(), and interrupts, if the thread dies before the event gets delivered, the kernel sends the signal to a random thread in the same process.

The following fields are used:

int sigev_signo
The signal to raise. This must be in the range from 1 through NSIG − 1.
short sigev_code
A code to be interpreted by the signal handler. This must be in the range from SI_MINAVAIL through SI_MAXAVAIL.
int sigev_value.sival_int
A 32-bit value to be interpreted by the signal handler.

The initialization macro is:

SIGEV_SIGNAL_THREAD_INIT( &event, signal, value, code )

SIGEV_THREAD (POSIX)

Create a new thread.

Note: We don't recommend using this type of event. Pulses are more efficient.

(QNX Neutrino 7.0.4 or later) In order to register or use an event of type SIGEV_THREAD, your process must have the PROCMGR_AID_SIGEV_THREAD ability enabled. For more information, see procmgr_ability().

The following fields are used:

void (*sigev_notify_function) (union sigval)
A pointer to the function to be notified.
pthread_attr *sigev_notify_attributes
A pointer to thread attributes. This must be NULL, or point to a structure initialized by pthread_attr_init() at the time of delivery.
void *sigev_value.sival_ptr
A value that's to be passed to the notification function.

The initialization macro is:

SIGEV_THREAD_INIT( &event, fn, value, attr )

The sigval union is defined as follows:

union sigval {
    int        sival_int;
    void     * sival_ptr;
};
Note: (QNX Neutrino 6.6 or later) If a client program replaces its process image (e.g., by calling exec*()) with a setuid or setgid binary, the kernel sets the _NTO_COF_INSECURE flag on the channel. If this flag is set, calls to MsgDeliverEvent() with an event type of SIGEV_MEMORY or SIGEV_THREAD fail with an error of EACCES. In QNX Neutrino 7.0 and later, only the kernel can clear this flag.

SIGEV_UNBLOCK (QNX Neutrino extension)

Force a thread to become unblocked. No other fields in the structure are used.

The initialization macro is:

SIGEV_UNBLOCK_INIT( &event )

Hidden bits (QNX Neutrino extension)

After you've set up the sigevent structure as appropriate, you can use some extra bits in sigev_notify for other purposes. The bits and their associated macros include:

SIGEV_FLAG_SIVAL_INT
(QNX Neutrino 7.0 or later) The sigevent was initialized specifically to hold an integer with one of the SIGEV_*_INT_INIT() macros.
SIGEV_FLAG_CODE_UPDATEABLE
(QNX Neutrino 7.0.4 or later) Allow the server to update the sigev_code member before the event is delivered. The actual update depends on the server.
To determine if the bit is set, the server should just test it directly:
if (event.sigev_notify & SIGEV_FLAG_CODE_UPDATEABLE)
{ // may update event before delivering }
else
{ // should not update event before delivering } 
SIGEV_FLAG_UPDATEABLE
(QNX Neutrino 6.6 or later) Allow the server to update the sigev_value member before the event is delivered:
  • SIGEV_MAKE_UPDATEABLE( &event )
  • SIGEV_CLEAR_UPDATEABLE( &event )

The actual update depends on the server. For information about the updated information for kernel events, see procmgr_event_notify_add().

To determine if the bit is set, the server should just test it directly:
if (event.sigev_notify & SIGEV_FLAG_UPDATEABLE)
{ // may update event before delivering }
else
{ // should not update event before delivering } 

32- and 64-bit architectures

You'll typically use struct sigevent, unless your program needs to work with sigevents from 32- and 64-bit programs, in which case you'll also have to work with struct __sigevent32 and struct __sigevent64.

Since a sigevent might be passed between 32- and 64-bit programs, we've defined a SIGEV_64BIT flag that indicates which type of sigevent it is. We OR this flag into the sigev_notify types. For example, in addition to SIGEV_SIGNAL, we have SIGEV_SIGNAL32 and SIGEV_SIGNAL64:

Some programs use SIGEV_NONE to identify an empty sigevent. For compatibility, SIGEV_NONE is always the same as SIGEV_NONE32, no matter which architecture you compile for.

If your program stores sigevents from 32- and 64-bit programs, you should replace struct sigevent with a union:

union {
   struct sigevent ev;
   struct __sigevent32 ev32;
   struct __sigevent64 ev64;
};

Your code can then check ev.sigev_notify, determine whether the sigevent is from a 32- or 64-bit program, and then use the ev32 or ev64 structure, as appropriate.

Classification:

POSIX 1003.1, with QNX Neutrino extensions