select(9) [netbsd man page]

SELECT(9)						   BSD Kernel Developer's Manual						 SELECT(9)

NAME

     seldestroy, selinit, selrecord, selnotify -- select and poll subsystem

SYNOPSIS

     #include <sys/param.h>
     #include <sys/select.h>

     void
     seldestroy(struct selinfo *sip);

     void
     selinit(struct selinfo *sip);

     void
     selrecord(struct lwp *selector, struct selinfo *sip);

     void
     selnotify(struct selinfo *sip, int events, long knhint);

DESCRIPTION

     selinit() and seldestroy() functions must be used to initialize and destroy the struct selinfo.  The seldestroy() function may block.

     selrecord() and selnotify() are used by device drivers to coordinate with the kernel implementation of select(2) and poll(2).  Each object
     that can be polled contains a selinfo record.  Device drivers provide locking for the selinfo record.

     selrecord() records that the calling thread is interested in events related to a given object.  selrecord() should only be called when the
     poll routine determines that the object is not ready for I/O: there are no events of interest pending.  The check for pending I/O and call to
     selrecord() must be atomic.  Atomicity can be provided by holding the object's lock across the test and call to selrecord().  For non-MPSAFE
     drivers, the global kernel_lock is enough to provide atomicity.

     selnotify() is called by the underlying object handling code in order to notify any waiting threads that an event of interest has occurred.
     The same lock held across the poll method and call to selrecord() must be held across the call to selnotify().  The lock prevents an event of
     interest being signalled while a thread is in the process of recording its interest.

     The events indicates which event happen.  Zero may be used if unknown.

     selnotify() also calls KNOTE() passing knhint as an argument.

CODE REFERENCES

     The core of the select and poll subsystem implementation is in sys/kern/sys_select.c.  Data structures and function prototypes are located in
     sys/sys/select.h, sys/sys/poll.h and sys/sys/selinfo.h.

SEE ALSO

     poll(2), select(2), knote(9)

BSD
								   May 13, 2008 							       BSD

chpoll(9E)							Driver Entry Points							chpoll(9E)

NAME

       chpoll - poll entry point for a non-STREAMS character driver

SYNOPSIS

       #include <sys/types.h>
       #include <sys/poll.h>
       #include <sys/ddi.h>
       #include <sys/sunddi.h>

       int prefixchpoll(dev_t dev, short events, int anyyet, short *reventsp, struct pollhead **phpp);

INTERFACE LEVEL

       This entry point is optional. Architecture independent level 1 (DDI/DKI).

PARAMETERS

	dev	       The device number for the device to be polled.

       events	       The events that may occur. Valid events are:

		       POLLIN

			   Data other than high priority data may be read without blocking.

		       POLLOUT

			   Normal data may be written without blocking.

		       POLLPRI

			   High priority data may be received without blocking.

		       POLLHUP

			   A device hangup has occurred.

		       POLLERR

			   An error has occurred on the device.

		       POLLRDNORM

			   Normal data (priority band = 0) may be read without blocking.

		       POLLRDBAND

			   Data from a non-zero priority band may be read without blocking

		       POLLWRNORM

			   The same as POLLOUT.

		       POLLWRBAND

			   Priority data (priority band > 0) may be written.

       anyyet	       A  flag	that  is  non-zero  if any other file descriptors in the pollfd array have events pending. The poll(2) system call
		       takes a pointer to an array of pollfd structures as one of its arguments. See the poll(2) reference page for more details.

       reventsp        A pointer to a bitmask of the returned events satisfied.

       phpp	       A pointer to a pointer to a pollhead structure.

DESCRIPTION

       The chpoll() entry point routine is used by non-STREAMS character device drivers that wish to support polling. The  driver  must  implement
       the polling discipline itself. The following rules must be followed when implementing the polling discipline:

       1.  Implement the following algorithm when the chpoll() entry point is called:

	   if (events_are_satisfied_now) {
		 *reventsp = mask_of_satisfied_events;
	   } else {
		 *reventsp = 0;
		 if (!anyyet)
		       *phpp = &my_local_pollhead_structure;
	   }
	   return(0);

       2.  Allocate  an  instance  of the pollhead structure. This instance may be tied to the per-minor data structure defined by the driver. The
	   pollhead structure should be treated as a ``black box'' by the driver. None of its fields should be referenced.  However, the  size	of
	   this structure is guaranteed to remain the same across releases.

       3.  Call the pollwakeup() function whenever an event of type events listed above occur. This function can be called with multiple events at
	   one time. The driver must not hold any mutex across the call to pollwakeup(9F) that is acquired in its chpoll() entry point, or a dead-
	   lock may result.

RETURN VALUES

       chpoll() should return 0 for success, or the appropriate error number.

SEE ALSO

       poll(2), nochpoll(9F), pollwakeup(9F)

       Writing Device Drivers

SunOS 5.10							    8 Nov 2001								chpoll(9E)