X-Git-Url: https://git.liburcu.org/?p=urcu.git;a=blobdiff_plain;f=rculfhash.c;h=a7e751bd29c34b04a2f9523a4718c433fff6cd89;hp=93514d429a5d4fe242fec8b102288101e9a1d78f;hb=b4cb483f1a9fd9e1a1c465ab6f5fc44176252d7c;hpb=59290e9dc90ce01920f1a18a8699dd6872c45ffc

diff --git a/rculfhash.c b/rculfhash.c
index 93514d4..a7e751b 100644
--- a/rculfhash.c
+++ b/rculfhash.c
@@ -209,7 +209,7 @@ struct ht_items_count {
 } __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 
 struct rcu_level {
-	struct rcu_head head;
+	/* Note: manually update allocation length when adding a field */
 	struct _cds_lfht_node nodes[0];
 };
 
@@ -255,24 +255,19 @@ struct rcu_resize_work {
 };
 
 struct partition_resize_work {
-	struct rcu_head head;
+	pthread_t thread_id;
 	struct cds_lfht *ht;
 	unsigned long i, start, len;
 	void (*fct)(struct cds_lfht *ht, unsigned long i,
 		    unsigned long start, unsigned long len);
 };
 
-enum add_mode {
-	ADD_DEFAULT = 0,
-	ADD_UNIQUE = 1,
-	ADD_REPLACE = 2,
-};
-
 static
-struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
-				unsigned long size,
-				struct cds_lfht_node *node,
-				enum add_mode mode, int dummy);
+void _cds_lfht_add(struct cds_lfht *ht,
+		unsigned long size,
+		struct cds_lfht_node *node,
+		struct cds_lfht_iter *unique_ret,
+		int dummy);
 
 /*
  * Algorithm to reverse bits in a word by lookup table, extended to
@@ -446,24 +441,28 @@ unsigned int fls_ulong(unsigned long x)
 #endif
 }
 
+/*
+ * Return the minimum order for which x <= (1UL << order).
+ * Return -1 if x is 0.
+ */
 int get_count_order_u32(uint32_t x)
 {
-	int order;
+	if (!x)
+		return -1;
 
-	order = fls_u32(x) - 1;
-	if (x & (x - 1))
-		order++;
-	return order;
+	return fls_u32(x - 1);
 }
 
+/*
+ * Return the minimum order for which x <= (1UL << order).
+ * Return -1 if x is 0.
+ */
 int get_count_order_ulong(unsigned long x)
 {
-	int order;
+	if (!x)
+		return -1;
 
-	order = fls_ulong(x) - 1;
-	if (x & (x - 1))
-		order++;
-	return order;
+	return fls_ulong(x - 1);
 }
 
 #ifdef POISON_FREE
@@ -713,11 +712,24 @@ unsigned long _uatomic_max(unsigned long *ptr, unsigned long v)
 }
 
 static
-void cds_lfht_free_level(struct rcu_head *head)
+struct _cds_lfht_node *lookup_bucket(struct cds_lfht *ht, unsigned long size,
+		unsigned long hash)
 {
-	struct rcu_level *l =
-		caa_container_of(head, struct rcu_level, head);
-	poison_free(l);
+	unsigned long index, order;
+
+	assert(size > 0);
+	index = hash & (size - 1);
+	/*
+	 * equivalent to get_count_order_ulong(index + 1), but optimizes
+	 * away the non-existing 0 special-case for
+	 * get_count_order_ulong.
+	 */
+	order = fls_ulong(index);
+
+	dbg_printf("lookup hash %lu index %lu order %lu aridx %lu\n",
+		   hash, index, order, index & (!order ? 0 : ((1UL << (order - 1)) - 1)));
+
+	return &ht->t.tbl[order]->nodes[index & (!order ? 0 : ((1UL << (order - 1)) - 1))];
 }
 
 /*
@@ -736,6 +748,7 @@ void _cds_lfht_gc_bucket(struct cds_lfht_node *dummy, struct cds_lfht_node *node
 		iter_prev = dummy;
 		/* We can always skip the dummy node initially */
 		iter = rcu_dereference(iter_prev->p.next);
+		assert(!is_removed(iter));
 		assert(iter_prev->p.reverse_hash <= node->p.reverse_hash);
 		/*
 		 * We should never be called with dummy (start of chain)
@@ -760,8 +773,6 @@ void _cds_lfht_gc_bucket(struct cds_lfht_node *dummy, struct cds_lfht_node *node
 			new_next = flag_dummy(clear_flag(next));
 		else
 			new_next = clear_flag(next);
-		if (is_removed(iter))
-			new_next = flag_removed(new_next);
 		(void) uatomic_cmpxchg(&iter_prev->p.next, iter, new_next);
 	}
 	return;
@@ -770,13 +781,12 @@ void _cds_lfht_gc_bucket(struct cds_lfht_node *dummy, struct cds_lfht_node *node
 static
 int _cds_lfht_replace(struct cds_lfht *ht, unsigned long size,
 		struct cds_lfht_node *old_node,
-		struct cds_lfht_node *ret_next,
+		struct cds_lfht_node *old_next,
 		struct cds_lfht_node *new_node)
 {
-	struct cds_lfht_node *dummy, *old_next;
+	struct cds_lfht_node *dummy, *ret_next;
 	struct _cds_lfht_node *lookup;
 	int flagged = 0;
-	unsigned long hash, index, order;
 
 	if (!old_node)	/* Return -ENOENT if asked to replace NULL node */
 		goto end;
@@ -786,9 +796,8 @@ int _cds_lfht_replace(struct cds_lfht *ht, unsigned long size,
 	assert(!is_removed(new_node));
 	assert(!is_dummy(new_node));
 	assert(new_node != old_node);
-	do {
+	for (;;) {
 		/* Insert after node to be replaced */
-		old_next = ret_next;
 		if (is_removed(old_next)) {
 			/*
 			 * Too late, the old node has been removed under us
@@ -811,7 +820,10 @@ int _cds_lfht_replace(struct cds_lfht *ht, unsigned long size,
 		 */
 		ret_next = uatomic_cmpxchg(&old_node->p.next,
 			      old_next, flag_removed(new_node));
-	} while (ret_next != old_next);
+		if (ret_next == old_next)
+			break;
+		old_next = ret_next;
+	}
 
 	/* We performed the replacement. */
 	flagged = 1;
@@ -821,11 +833,7 @@ int _cds_lfht_replace(struct cds_lfht *ht, unsigned long size,
 	 * lookup for the node, and remove it (along with any other
 	 * logically removed node) if found.
 	 */
-	hash = bit_reverse_ulong(old_node->p.reverse_hash);
-	assert(size > 0);
-	index = hash & (size - 1);
-	order = get_count_order_ulong(index + 1);
-	lookup = &ht->t.tbl[order]->nodes[index & (!order ? 0 : ((1UL << (order - 1)) - 1))];
+	lookup = lookup_bucket(ht, size, bit_reverse_ulong(old_node->p.reverse_hash));
 	dummy = (struct cds_lfht_node *) lookup;
 	_cds_lfht_gc_bucket(dummy, new_node);
 end:
@@ -841,25 +849,30 @@ end:
 	}
 }
 
+/*
+ * A non-NULL unique_ret pointer uses the "add unique" (or uniquify) add
+ * mode. A NULL unique_ret allows creation of duplicate keys.
+ */
 static
-struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
-				unsigned long size,
-				struct cds_lfht_node *node,
-				enum add_mode mode, int dummy)
+void _cds_lfht_add(struct cds_lfht *ht,
+		unsigned long size,
+		struct cds_lfht_node *node,
+		struct cds_lfht_iter *unique_ret,
+		int dummy)
 {
 	struct cds_lfht_node *iter_prev, *iter, *next, *new_node, *new_next,
-			*dummy_node, *return_node;
+			*return_node;
 	struct _cds_lfht_node *lookup;
-	unsigned long hash, index, order;
 
 	assert(!is_dummy(node));
 	assert(!is_removed(node));
 	if (!size) {
 		assert(dummy);
+		assert(!unique_ret);
 		node->p.next = flag_dummy(get_end());
-		return node;	/* Initial first add (head) */
+		return;		/* Initial first add (head) */
 	}
-	hash = bit_reverse_ulong(node->p.reverse_hash);
+	lookup = lookup_bucket(ht, size, bit_reverse_ulong(node->p.reverse_hash));
 	for (;;) {
 		uint32_t chain_len = 0;
 
@@ -867,9 +880,6 @@ struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 		 * iter_prev points to the non-removed node prior to the
 		 * insert location.
 		 */
-		index = hash & (size - 1);
-		order = get_count_order_ulong(index + 1);
-		lookup = &ht->t.tbl[order]->nodes[index & ((!order ? 0 : (1UL << (order - 1))) - 1)];
 		iter_prev = (struct cds_lfht_node *) lookup;
 		/* We can always skip the dummy node initially */
 		iter = rcu_dereference(iter_prev->p.next);
@@ -879,18 +889,21 @@ struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 				goto insert;
 			if (likely(clear_flag(iter)->p.reverse_hash > node->p.reverse_hash))
 				goto insert;
+			/* dummy node is the first node of the identical-hash-value chain */
+			if (dummy && clear_flag(iter)->p.reverse_hash == node->p.reverse_hash)
+				goto insert;
 			next = rcu_dereference(clear_flag(iter)->p.next);
 			if (unlikely(is_removed(next)))
 				goto gc_node;
-			if ((mode == ADD_UNIQUE || mode == ADD_REPLACE)
+			if (unique_ret
 			    && !is_dummy(next)
+			    && clear_flag(iter)->p.reverse_hash == node->p.reverse_hash
 			    && !ht->compare_fct(node->key, node->key_len,
 						clear_flag(iter)->key,
 						clear_flag(iter)->key_len)) {
-				if (mode == ADD_UNIQUE)
-					return clear_flag(iter);
-				else /* mode == ADD_REPLACE */
-					goto replace;
+				unique_ret->node = clear_flag(iter);
+				unique_ret->next = next;
+				return;
 			}
 			/* Only account for identical reverse hash once */
 			if (iter_prev->p.reverse_hash != clear_flag(iter)->p.reverse_hash
@@ -917,21 +930,8 @@ struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 				    new_node) != iter) {
 			continue;	/* retry */
 		} else {
-			if (mode == ADD_REPLACE)
-				return_node = NULL;
-			else	/* ADD_DEFAULT and ADD_UNIQUE */
-				return_node = node;
-			goto gc_end;
-		}
-
-	replace:
-
-		if (!_cds_lfht_replace(ht, size, clear_flag(iter), next,
-				    node)) {
-			return_node = clear_flag(iter);
-			goto end;	/* gc already done */
-		} else {
-			continue;	/* retry */
+			return_node = node;
+			goto end;
 		}
 
 	gc_node:
@@ -943,15 +943,11 @@ struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 		(void) uatomic_cmpxchg(&iter_prev->p.next, iter, new_next);
 		/* retry */
 	}
-gc_end:
-	/* Garbage collect logically removed nodes in the bucket */
-	index = hash & (size - 1);
-	order = get_count_order_ulong(index + 1);
-	lookup = &ht->t.tbl[order]->nodes[index & (!order ? 0 : ((1UL << (order - 1)) - 1))];
-	dummy_node = (struct cds_lfht_node *) lookup;
-	_cds_lfht_gc_bucket(dummy_node, node);
 end:
-	return return_node;
+	if (unique_ret) {
+		unique_ret->node = return_node;
+		/* unique_ret->next left unset, never used. */
+	}
 }
 
 static
@@ -962,7 +958,6 @@ int _cds_lfht_del(struct cds_lfht *ht, unsigned long size,
 	struct cds_lfht_node *dummy, *next, *old;
 	struct _cds_lfht_node *lookup;
 	int flagged = 0;
-	unsigned long hash, index, order;
 
 	if (!node)	/* Return -ENOENT if asked to delete NULL node */
 		goto end;
@@ -993,11 +988,7 @@ int _cds_lfht_del(struct cds_lfht *ht, unsigned long size,
 	 * the node, and remove it (along with any other logically removed node)
 	 * if found.
 	 */
-	hash = bit_reverse_ulong(node->p.reverse_hash);
-	assert(size > 0);
-	index = hash & (size - 1);
-	order = get_count_order_ulong(index + 1);
-	lookup = &ht->t.tbl[order]->nodes[index & (!order ? 0 : ((1UL << (order - 1)) - 1))];
+	lookup = lookup_bucket(ht, size, bit_reverse_ulong(node->p.reverse_hash));
 	dummy = (struct cds_lfht_node *) lookup;
 	_cds_lfht_gc_bucket(dummy, node);
 end:
@@ -1034,7 +1025,6 @@ void partition_resize_helper(struct cds_lfht *ht, unsigned long i,
 	struct partition_resize_work *work;
 	int thread, ret;
 	unsigned long nr_threads;
-	pthread_t *thread_id;
 
 	/*
 	 * Note: nr_cpus_mask + 1 is always power of 2.
@@ -1049,7 +1039,6 @@ void partition_resize_helper(struct cds_lfht *ht, unsigned long i,
 	}
 	partition_len = len >> get_count_order_ulong(nr_threads);
 	work = calloc(nr_threads, sizeof(*work));
-	thread_id = calloc(nr_threads, sizeof(*thread_id));
 	assert(work);
 	for (thread = 0; thread < nr_threads; thread++) {
 		work[thread].ht = ht;
@@ -1057,16 +1046,15 @@ void partition_resize_helper(struct cds_lfht *ht, unsigned long i,
 		work[thread].len = partition_len;
 		work[thread].start = thread * partition_len;
 		work[thread].fct = fct;
-		ret = pthread_create(&thread_id[thread], ht->resize_attr,
+		ret = pthread_create(&(work[thread].thread_id), ht->resize_attr,
 			partition_resize_thread, &work[thread]);
 		assert(!ret);
 	}
 	for (thread = 0; thread < nr_threads; thread++) {
-		ret = pthread_join(thread_id[thread], NULL);
+		ret = pthread_join(work[thread].thread_id, NULL);
 		assert(!ret);
 	}
 	free(work);
-	free(thread_id);
 }
 
 /*
@@ -1095,10 +1083,8 @@ void init_table_populate_partition(struct cds_lfht *ht, unsigned long i,
 			   i, j, !i ? 0 : (1UL << (i - 1)) + j);
 		new_node->p.reverse_hash =
 			bit_reverse_ulong(!i ? 0 : (1UL << (i - 1)) + j);
-		(void) _cds_lfht_add(ht, !i ? 0 : (1UL << (i - 1)),
-				new_node, ADD_DEFAULT, 1);
-		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
-			break;
+		_cds_lfht_add(ht, !i ? 0 : (1UL << (i - 1)),
+				new_node, NULL, 1);
 	}
 	ht->cds_lfht_rcu_read_unlock();
 }
@@ -1136,8 +1122,7 @@ void init_table(struct cds_lfht *ht,
 		if (CMM_LOAD_SHARED(ht->t.resize_target) < (!i ? 1 : (1UL << i)))
 			break;
 
-		ht->t.tbl[i] = calloc(1, sizeof(struct rcu_level)
-				+ (len * sizeof(struct _cds_lfht_node)));
+		ht->t.tbl[i] = calloc(1, len * sizeof(struct _cds_lfht_node));
 		assert(ht->t.tbl[i]);
 
 		/*
@@ -1200,8 +1185,6 @@ void remove_table_partition(struct cds_lfht *ht, unsigned long i,
 			bit_reverse_ulong(!i ? 0 : (1UL << (i - 1)) + j);
 		(void) _cds_lfht_del(ht, !i ? 0 : (1UL << (i - 1)),
 				fini_node, 1);
-		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
-			break;
 	}
 	ht->cds_lfht_rcu_read_unlock();
 }
@@ -1225,6 +1208,7 @@ void fini_table(struct cds_lfht *ht,
 		unsigned long first_order, unsigned long len_order)
 {
 	long i, end_order;
+	void *free_by_rcu = NULL;
 
 	dbg_printf("fini table: first_order %lu end_order %lu\n",
 		   first_order, first_order + len_order);
@@ -1250,6 +1234,8 @@ void fini_table(struct cds_lfht *ht,
 		 * return a logically removed node as insert position.
 		 */
 		ht->cds_lfht_synchronize_rcu();
+		if (free_by_rcu)
+			free(free_by_rcu);
 
 		/*
 		 * Set "removed" flag in dummy nodes about to be removed.
@@ -1259,12 +1245,17 @@ void fini_table(struct cds_lfht *ht,
 		 */
 		remove_table(ht, i, len);
 
-		ht->cds_lfht_call_rcu(&ht->t.tbl[i]->head, cds_lfht_free_level);
+		free_by_rcu = ht->t.tbl[i];
 
 		dbg_printf("fini new size: %lu\n", 1UL << i);
 		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
 			break;
 	}
+
+	if (free_by_rcu) {
+		ht->cds_lfht_synchronize_rcu();
+		free(free_by_rcu);
+	}
 }
 
 struct cds_lfht *_cds_lfht_new(cds_lfht_hash_fct hash_fct,
@@ -1322,17 +1313,13 @@ void cds_lfht_lookup(struct cds_lfht *ht, void *key, size_t key_len,
 {
 	struct cds_lfht_node *node, *next, *dummy_node;
 	struct _cds_lfht_node *lookup;
-	unsigned long hash, reverse_hash, index, order, size;
+	unsigned long hash, reverse_hash, size;
 
 	hash = ht->hash_fct(key, key_len, ht->hash_seed);
 	reverse_hash = bit_reverse_ulong(hash);
 
 	size = rcu_dereference(ht->t.size);
-	index = hash & (size - 1);
-	order = get_count_order_ulong(index + 1);
-	lookup = &ht->t.tbl[order]->nodes[index & (!order ? 0 : ((1UL << (order - 1))) - 1)];
-	dbg_printf("lookup hash %lu index %lu order %lu aridx %lu\n",
-		   hash, index, order, index & (!order ? 0 : ((1UL << (order - 1)) - 1)));
+	lookup = lookup_bucket(ht, size, hash);
 	dummy_node = (struct cds_lfht_node *) lookup;
 	/* We can always skip the dummy node initially */
 	node = rcu_dereference(dummy_node->p.next);
@@ -1349,6 +1336,7 @@ void cds_lfht_lookup(struct cds_lfht *ht, void *key, size_t key_len,
 		next = rcu_dereference(node->p.next);
 		if (likely(!is_removed(next))
 		    && !is_dummy(next)
+		    && clear_flag(node)->p.reverse_hash == reverse_hash
 		    && likely(!ht->compare_fct(node->key, node->key_len, key, key_len))) {
 				break;
 		}
@@ -1438,7 +1426,7 @@ void cds_lfht_add(struct cds_lfht *ht, struct cds_lfht_node *node)
 	node->p.reverse_hash = bit_reverse_ulong((unsigned long) hash);
 
 	size = rcu_dereference(ht->t.size);
-	(void) _cds_lfht_add(ht, size, node, ADD_DEFAULT, 0);
+	_cds_lfht_add(ht, size, node, NULL, 0);
 	ht_count_add(ht, size);
 }
 
@@ -1446,32 +1434,38 @@ struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht,
 				struct cds_lfht_node *node)
 {
 	unsigned long hash, size;
-	struct cds_lfht_node *ret;
+	struct cds_lfht_iter iter;
 
 	hash = ht->hash_fct(node->key, node->key_len, ht->hash_seed);
 	node->p.reverse_hash = bit_reverse_ulong((unsigned long) hash);
 
 	size = rcu_dereference(ht->t.size);
-	ret = _cds_lfht_add(ht, size, node, ADD_UNIQUE, 0);
-	if (ret == node)
+	_cds_lfht_add(ht, size, node, &iter, 0);
+	if (iter.node == node)
 		ht_count_add(ht, size);
-	return ret;
+	return iter.node;
 }
 
 struct cds_lfht_node *cds_lfht_add_replace(struct cds_lfht *ht,
 				struct cds_lfht_node *node)
 {
 	unsigned long hash, size;
-	struct cds_lfht_node *ret;
+	struct cds_lfht_iter iter;
 
 	hash = ht->hash_fct(node->key, node->key_len, ht->hash_seed);
 	node->p.reverse_hash = bit_reverse_ulong((unsigned long) hash);
 
 	size = rcu_dereference(ht->t.size);
-	ret = _cds_lfht_add(ht, size, node, ADD_REPLACE, 0);
-	if (ret == NULL)
-		ht_count_add(ht, size);
-	return ret;
+	for (;;) {
+		_cds_lfht_add(ht, size, node, &iter, 0);
+		if (iter.node == node) {
+			ht_count_add(ht, size);
+			return NULL;
+		}
+
+		if (!_cds_lfht_replace(ht, size, iter.node, iter.next, node))
+			return iter.node;
+	}
 }
 
 int cds_lfht_replace(struct cds_lfht *ht, struct cds_lfht_iter *old_iter,
@@ -1616,8 +1610,8 @@ void _do_cds_lfht_grow(struct cds_lfht *ht,
 
 	old_order = get_count_order_ulong(old_size) + 1;
 	new_order = get_count_order_ulong(new_size) + 1;
-	printf("resize from %lu (order %lu) to %lu (order %lu) buckets\n",
-	       old_size, old_order, new_size, new_order);
+	dbg_printf("resize from %lu (order %lu) to %lu (order %lu) buckets\n",
+		   old_size, old_order, new_size, new_order);
 	assert(new_size > old_size);
 	init_table(ht, old_order, new_order - old_order);
 }
@@ -1632,8 +1626,8 @@ void _do_cds_lfht_shrink(struct cds_lfht *ht,
 	new_size = max(new_size, MIN_TABLE_SIZE);
 	old_order = get_count_order_ulong(old_size) + 1;
 	new_order = get_count_order_ulong(new_size) + 1;
-	printf("resize from %lu (order %lu) to %lu (order %lu) buckets\n",
-	       old_size, old_order, new_size, new_order);
+	dbg_printf("resize from %lu (order %lu) to %lu (order %lu) buckets\n",
+		   old_size, old_order, new_size, new_order);
 	assert(new_size < old_size);
 
 	/* Remove and unlink all dummy nodes to remove. */
@@ -1651,6 +1645,9 @@ void _do_cds_lfht_resize(struct cds_lfht *ht)
 	 * Resize table, re-do if the target size has changed under us.
 	 */
 	do {
+		assert(uatomic_read(&ht->in_progress_resize));
+		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
+			break;
 		ht->t.resize_initiated = 1;
 		old_size = ht->t.size;
 		new_size = CMM_LOAD_SHARED(ht->t.resize_target);
@@ -1720,8 +1717,10 @@ void cds_lfht_resize_lazy(struct cds_lfht *ht, unsigned long size, int growth)
 	if (!CMM_LOAD_SHARED(ht->t.resize_initiated) && size < target_size) {
 		uatomic_inc(&ht->in_progress_resize);
 		cmm_smp_mb();	/* increment resize count before load destroy */
-		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
+		if (CMM_LOAD_SHARED(ht->in_progress_destroy)) {
+			uatomic_dec(&ht->in_progress_resize);
 			return;
+		}
 		work = malloc(sizeof(*work));
 		work->ht = ht;
 		ht->cds_lfht_call_rcu(&work->head, do_resize_cb);
@@ -1745,8 +1744,10 @@ void cds_lfht_resize_lazy_count(struct cds_lfht *ht, unsigned long size,
 	if (!CMM_LOAD_SHARED(ht->t.resize_initiated)) {
 		uatomic_inc(&ht->in_progress_resize);
 		cmm_smp_mb();	/* increment resize count before load destroy */
-		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
+		if (CMM_LOAD_SHARED(ht->in_progress_destroy)) {
+			uatomic_dec(&ht->in_progress_resize);
 			return;
+		}
 		work = malloc(sizeof(*work));
 		work->ht = ht;
 		ht->cds_lfht_call_rcu(&work->head, do_resize_cb);