ck_hp: hp_acquire patterns require strict fences under TSO.

Several counter-examples were found which break in the
presence of store-to-load re-ordering. Strict fence
semantics are necessary.

Thanks to Paul McKenney for helpful discussions.

include/ck_hp_fifo.h

@@ -81,7 +81,7 @@ ck_hp_fifo_enqueue_mpmc(ck_hp_record_t *record,
 	for (;;) {
 		tail = ck_pr_load_ptr(&fifo->tail);
 		ck_hp_set(record, 0, tail);
-		ck_pr_fence_memory();
+		ck_pr_fence_strict_memory();
 		if (tail != ck_pr_load_ptr(&fifo->tail))
 			continue;
 
@@ -112,7 +112,7 @@ ck_hp_fifo_tryenqueue_mpmc(ck_hp_record_t *record,
 
 	tail = ck_pr_load_ptr(&fifo->tail);
 	ck_hp_set(record, 0, tail);
-	ck_pr_fence_memory();
+	ck_pr_fence_strict_memory();
 	if (tail != ck_pr_load_ptr(&fifo->tail))
 		return false;
 
@@ -140,13 +140,13 @@ ck_hp_fifo_dequeue_mpmc(ck_hp_record_t *record,
 		ck_pr_fence_load();
 		tail = ck_pr_load_ptr(&fifo->tail);
 		ck_hp_set(record, 0, head);
-		ck_pr_fence_memory();
+		ck_pr_fence_strict_memory();
 		if (head != ck_pr_load_ptr(&fifo->head))
 			continue;
 
 		next = ck_pr_load_ptr(&head->next);
 		ck_hp_set(record, 1, next);
-		ck_pr_fence_memory();
+		ck_pr_fence_strict_memory();
 		if (head != ck_pr_load_ptr(&fifo->head))
 			continue;
 
@@ -175,13 +175,13 @@ ck_hp_fifo_trydequeue_mpmc(ck_hp_record_t *record,
 	ck_pr_fence_load();
 	tail = ck_pr_load_ptr(&fifo->tail);
 	ck_hp_set(record, 0, head);
-	ck_pr_fence_memory();
+	ck_pr_fence_strict_memory();
 	if (head != ck_pr_load_ptr(&fifo->head))
 		return NULL;
 
 	next = ck_pr_load_ptr(&head->next);
 	ck_hp_set(record, 1, next);
-	ck_pr_fence_memory();
+	ck_pr_fence_strict_memory();
 	if (head != ck_pr_load_ptr(&fifo->head))
 		return NULL;
 

include/ck_hp_stack.h

@@ -62,7 +62,7 @@ ck_hp_stack_pop_mpmc(ck_hp_record_t *record, struct ck_stack *target)
 			return (NULL);
 
 		ck_hp_set(record, 0, entry);
-		ck_pr_fence_memory();
+		ck_pr_fence_strict_memory();
 	} while (entry != ck_pr_load_ptr(&target->head));
 
 	while (ck_pr_cas_ptr_value(&target->head, entry, entry->next, &entry) == false) {
@@ -70,11 +70,11 @@ ck_hp_stack_pop_mpmc(ck_hp_record_t *record, struct ck_stack *target)
 			return (NULL);
 
 		ck_hp_set(record, 0, entry);
-		ck_pr_fence_memory();
+		ck_pr_fence_strict_memory();
 		update = ck_pr_load_ptr(&target->head);
 		while (entry != update) {
 			ck_hp_set(record, 0, update);
-			ck_pr_fence_memory();
+			ck_pr_fence_strict_memory();
 			entry = update;
 			update = ck_pr_load_ptr(&target->head);
 			if (update == NULL)
@@ -95,7 +95,7 @@ ck_hp_stack_trypop_mpmc(ck_hp_record_t *record, struct ck_stack *target, struct
 		return false;
 
 	ck_hp_set(record, 0, entry);
-	ck_pr_fence_memory();
+	ck_pr_fence_strict_memory();
 	if (entry != ck_pr_load_ptr(&target->head))
 		goto leave;