table of contents
        
      
      
    | CK_EPOCH_BARRIER(3) | Library Functions Manual | CK_EPOCH_BARRIER(3) | 
NAME¶
ck_epoch_barrier —
    block until a grace period and all callbacks have been
    dispatched
LIBRARY¶
Concurrency Kit (libck, -lck)
SYNOPSIS¶
#include
    <ck_epoch.h>
void
  
  ck_epoch_barrier(ck_epoch_record_t
    *record);
DESCRIPTION¶
The
    ck_epoch_barrier(3)
    function will block the caller until a grace period has been detected,
    according to the semantics of epoch reclamation. Any objects requiring safe
    memory reclamation which are logically deleted are safe for physical
    deletion following a call to
    ck_epoch_barrier(3). This
    function will also dispatch all callbacks associated with
    epoch that were previously scheduled via
    ck_epoch_call(3).
EXAMPLE¶
#include <ck_epoch.h>
#include <ck_stack.h>
#include <stdlib.h>
/*
 * epoch was previously initialized with ck_epoch_init.
 * stack was previously initialized with ck_stack_init.
 */
ck_epoch_t *epoch;
ck_stack_t *stack;
void
function(void)
{
	ck_epoch_record_t *record;
	ck_stack_entry_t *s;
	record = malloc(sizeof *record);
	ck_epoch_register(&epoch, record, NULL);
	/*
	 * We are using an epoch section here to guarantee no
	 * nodes in the stack are deleted while we are dereferencing
	 * them. This is needed here because there are multiple writers.
	 * If there was only one thread popping from the this stack,
	 * then there is no need to ck_epoch_begin/ck_epoch_end.
	 */
	ck_epoch_begin(record);
	/* Logically delete an object. */
	s = ck_stack_pop_upmc(stack);
	ck_epoch_end(record);
	/*
	 * Wait until no threads could possibly have a reference to the
	 * object we just popped (assume all threads are simply executing
	 * ck_stack_pop_upmc).
	 */
	ck_epoch_barrier(record);
	/* It is now safe to physically delete the object. */
	free(s);
	return;
}
RETURN VALUES¶
This function has no return value.
ERRORS¶
Behavior is undefined if the object pointed to by
    epoch is not a valid epoch object. The object pointed
    to by record must have been previously registered via
    ck_epoch_register(3).
SEE ALSO¶
ck_epoch_init(3), ck_epoch_register(3), ck_epoch_unregister(3), ck_epoch_recycle(3), ck_epoch_poll(3), ck_epoch_synchronize(3), ck_epoch_reclaim(3), ck_epoch_call(3), ck_epoch_begin(3), ck_epoch_end(3)
Additional information available at http://concurrencykit.org/
| September 2, 2012 |