X-Git-Url: https://git.liburcu.org/?p=urcu.git;a=blobdiff_plain;f=rculfhash.c;h=46513dfd40562bb7c1e0bb24f39fee3656e378a2;hp=72abe40f9ae7eb49dc8586c5cde4396e7f4aacba;hb=14360f1c5d88854a94caa9e6a933c430b38731f4;hpb=743f9143d0dce4b1385531577ebf702f119576ef

diff --git a/rculfhash.c b/rculfhash.c
index 72abe40..46513df 100644
--- a/rculfhash.c
+++ b/rculfhash.c
@@ -53,7 +53,7 @@
  *   operation.
  * - The resize operation for larger tables (and available through an
  *   API) allows both expanding and shrinking the hash table.
- * - Per-CPU Split-counters are used to keep track of the number of
+ * - Split-counters are used to keep track of the number of
  *   nodes within the hash table for automatic resize triggering.
  * - Resize operation initiated by long chain detection is executed by a
  *   call_rcu thread, which keeps lock-freedom of add and remove.
@@ -91,12 +91,32 @@
  *   the "dummy node" tables.
  * - There is one dummy node table per hash index order. The size of
  *   each dummy node table is half the number of hashes contained in
- *   this order.
- * - call_rcu is used to garbage-collect the old order table.
+ *   this order (except for order 0).
+ * - synchronzie_rcu is used to garbage-collect the old dummy node table.
  * - The per-order dummy node tables contain a compact version of the
  *   hash table nodes. These tables are invariant after they are
  *   populated into the hash table.
- * 
+ *
+ * Dummy node tables:
+ *
+ * hash table	hash table	the last	all dummy node tables
+ * order	size		dummy node	0   1   2   3   4   5   6(index)
+ * 				table size
+ * 0		1		1		1
+ * 1		2		1		1   1
+ * 2		4		2		1   1   2
+ * 3		8		4		1   1   2   4
+ * 4		16		8		1   1   2   4   8
+ * 5		32		16		1   1   2   4   8  16
+ * 6		64		32		1   1   2   4   8  16  32
+ *
+ * When growing/shrinking, we only focus on the last dummy node table
+ * which size is (!order ? 1 : (1 << (order -1))).
+ *
+ * Example for growing/shrinking:
+ * grow hash table from order 5 to 6: init the index=6 dummy node table
+ * shrink hash table from order 6 to 5: fini the index=6 dummy node table
+ *
  * A bit of ascii art explanation:
  * 
  * Order index is the off-by-one compare to the actual power of 2 because 
@@ -119,12 +139,9 @@
  * 
  * order              bits       reverse
  * 0               0  000        000
- *                 |
- * 1               |  1  001        100       <-    <-
- *                 |  |                        |     |
- * 2               |  |  2  010        010     |     |
+ * 1               |  1  001        100             <-
+ * 2               |  |  2  010        010    <-     |
  *                 |  |  |  3  011        110  | <-  |
- *                 |  |  |  |                  |  |  |
  * 3               -> |  |  |  4  100        001  |  |
  *                    -> |  |     5  101        101  |
  *                       -> |        6  110        011
@@ -156,7 +173,7 @@
 #endif
 
 /*
- * Per-CPU split-counters lazily update the global counter each 1024
+ * Split-counters lazily update the global counter each 1024
  * addition/removal. It automatically keeps track of resize required.
  * We use the bucket length as indicator for need to expand for small
  * tables and machines lacking per-cpu data suppport.
@@ -224,6 +241,8 @@ struct cds_lfht {
 	struct rcu_table t;
 	cds_lfht_hash_fct hash_fct;
 	cds_lfht_compare_fct compare_fct;
+	unsigned long min_alloc_order;
+	unsigned long min_alloc_size;
 	unsigned long hash_seed;
 	int flags;
 	/*
@@ -246,7 +265,7 @@ struct cds_lfht {
 	void (*cds_lfht_rcu_unregister_thread)(void);
 	pthread_attr_t *resize_attr;	/* Resize threads attributes */
 	long count;			/* global approximate item count */
-	struct ht_items_count *percpu_count;	/* per-cpu item count */
+	struct ht_items_count *split_count;	/* split item count */
 };
 
 struct rcu_resize_work {
@@ -484,79 +503,89 @@ void cds_lfht_resize_lazy(struct cds_lfht *ht, unsigned long size, int growth);
  * In the unfortunate event the number of CPUs reported would be
  * inaccurate, we use modulo arithmetic on the number of CPUs we got.
  */
-#if defined(HAVE_SCHED_GETCPU) && defined(HAVE_SYSCONF)
+#if defined(HAVE_SYSCONF)
 
 static
 void cds_lfht_resize_lazy_count(struct cds_lfht *ht, unsigned long size,
 				unsigned long count);
 
 static long nr_cpus_mask = -1;
+static long split_count_mask = -1;
+
+static void ht_init_nr_cpus_mask(void)
+{
+	long maxcpus;
+
+	maxcpus = sysconf(_SC_NPROCESSORS_CONF);
+	if (maxcpus <= 0) {
+		nr_cpus_mask = -2;
+		return;
+	}
+	/*
+	 * round up number of CPUs to next power of two, so we
+	 * can use & for modulo.
+	 */
+	maxcpus = 1UL << get_count_order_ulong(maxcpus);
+	nr_cpus_mask = maxcpus - 1;
+}
 
 static
-struct ht_items_count *alloc_per_cpu_items_count(void)
+struct ht_items_count *alloc_split_items_count(void)
 {
 	struct ht_items_count *count;
 
-	switch (nr_cpus_mask) {
-	case -2:
-		return NULL;
-	case -1:
-	{
-		long maxcpus;
-
-		maxcpus = sysconf(_SC_NPROCESSORS_CONF);
-		if (maxcpus <= 0) {
-			nr_cpus_mask = -2;
-			return NULL;
-		}
-		/*
-		 * round up number of CPUs to next power of two, so we
-		 * can use & for modulo.
-		 */
-		maxcpus = 1UL << get_count_order_ulong(maxcpus);
-		nr_cpus_mask = maxcpus - 1;
-	}
-		/* Fall-through */
-	default:
-		return calloc(nr_cpus_mask + 1, sizeof(*count));
+	if (nr_cpus_mask == -1)	{
+		ht_init_nr_cpus_mask();
+		split_count_mask = nr_cpus_mask;
 	}
+
+	if (split_count_mask < 0)
+		return NULL;
+	else
+		return calloc(split_count_mask + 1, sizeof(*count));
 }
 
 static
-void free_per_cpu_items_count(struct ht_items_count *count)
+void free_split_items_count(struct ht_items_count *count)
 {
 	poison_free(count);
 }
 
+#if defined(HAVE_SCHED_GETCPU)
 static
-int ht_get_cpu(void)
+int ht_get_split_count_index(unsigned long hash)
 {
 	int cpu;
 
-	assert(nr_cpus_mask >= 0);
+	assert(split_count_mask >= 0);
 	cpu = sched_getcpu();
 	if (unlikely(cpu < 0))
-		return cpu;
+		return hash & split_count_mask;
 	else
-		return cpu & nr_cpus_mask;
+		return cpu & split_count_mask;
 }
+#else /* #if defined(HAVE_SCHED_GETCPU) */
+static
+int ht_get_split_count_index(unsigned long hash)
+{
+	return hash & split_count_mask;
+}
+#endif /* #else #if defined(HAVE_SCHED_GETCPU) */
 
 static
-void ht_count_add(struct cds_lfht *ht, unsigned long size)
+void ht_count_add(struct cds_lfht *ht, unsigned long size, unsigned long hash)
 {
-	unsigned long percpu_count;
-	int cpu;
+	unsigned long split_count;
+	int index;
 
-	if (unlikely(!ht->percpu_count))
+	if (unlikely(!ht->split_count))
 		return;
-	cpu = ht_get_cpu();
-	if (unlikely(cpu < 0))
-		return;
-	percpu_count = uatomic_add_return(&ht->percpu_count[cpu].add, 1);
-	if (unlikely(!(percpu_count & ((1UL << COUNT_COMMIT_ORDER) - 1)))) {
+	index = ht_get_split_count_index(hash);
+	split_count = uatomic_add_return(&ht->split_count[index].add, 1);
+	if (unlikely(!(split_count & ((1UL << COUNT_COMMIT_ORDER) - 1)))) {
 		long count;
 
-		dbg_printf("add percpu %lu\n", percpu_count);
+		dbg_printf("add split count %lu\n", split_count);
 		count = uatomic_add_return(&ht->count,
 					   1UL << COUNT_COMMIT_ORDER);
 		/* If power of 2 */
@@ -571,21 +600,19 @@ void ht_count_add(struct cds_lfht *ht, unsigned long size)
 }
 
 static
-void ht_count_del(struct cds_lfht *ht, unsigned long size)
+void ht_count_del(struct cds_lfht *ht, unsigned long size, unsigned long hash)
 {
-	unsigned long percpu_count;
-	int cpu;
+	unsigned long split_count;
+	int index;
 
-	if (unlikely(!ht->percpu_count))
-		return;
-	cpu = ht_get_cpu();
-	if (unlikely(cpu < 0))
+	if (unlikely(!ht->split_count))
 		return;
-	percpu_count = uatomic_add_return(&ht->percpu_count[cpu].del, 1);
-	if (unlikely(!(percpu_count & ((1UL << COUNT_COMMIT_ORDER) - 1)))) {
+	index = ht_get_split_count_index(hash);
+	split_count = uatomic_add_return(&ht->split_count[index].del, 1);
+	if (unlikely(!(split_count & ((1UL << COUNT_COMMIT_ORDER) - 1)))) {
 		long count;
 
-		dbg_printf("del percpu %lu\n", percpu_count);
+		dbg_printf("del split count %lu\n", split_count);
 		count = uatomic_add_return(&ht->count,
 					   -(1UL << COUNT_COMMIT_ORDER));
 		/* If power of 2 */
@@ -597,7 +624,7 @@ void ht_count_del(struct cds_lfht *ht, unsigned long size)
 			 * Don't shrink table if the number of nodes is below a
 			 * certain threshold.
 			 */
-			if (count < (1UL << COUNT_COMMIT_ORDER) * (nr_cpus_mask + 1))
+			if (count < (1UL << COUNT_COMMIT_ORDER) * (split_count_mask + 1))
 				return;
 			cds_lfht_resize_lazy_count(ht, size,
 				count >> (CHAIN_LEN_TARGET - 1));
@@ -605,32 +632,33 @@ void ht_count_del(struct cds_lfht *ht, unsigned long size)
 	}
 }
 
-#else /* #if defined(HAVE_SCHED_GETCPU) && defined(HAVE_SYSCONF) */
+#else /* #if defined(HAVE_SYSCONF) */
 
 static const long nr_cpus_mask = -2;
+static const long split_count_mask = -2;
 
 static
-struct ht_items_count *alloc_per_cpu_items_count(void)
+struct ht_items_count *alloc_split_items_count(void)
 {
 	return NULL;
 }
 
 static
-void free_per_cpu_items_count(struct ht_items_count *count)
+void free_split_items_count(struct ht_items_count *count)
 {
 }
 
 static
-void ht_count_add(struct cds_lfht *ht, unsigned long size)
+void ht_count_add(struct cds_lfht *ht, unsigned long size, unsigned long hash)
 {
 }
 
 static
-void ht_count_del(struct cds_lfht *ht, unsigned long size)
+void ht_count_del(struct cds_lfht *ht, unsigned long size, unsigned long hash)
 {
 }
 
-#endif /* #else #if defined(HAVE_SCHED_GETCPU) && defined(HAVE_SYSCONF) */
+#endif /* #else #if defined(HAVE_SYSCONF) */
 
 
 static
@@ -719,17 +747,21 @@ struct _cds_lfht_node *lookup_bucket(struct cds_lfht *ht, unsigned long size,
 
 	assert(size > 0);
 	index = hash & (size - 1);
+
+	if (index < ht->min_alloc_size) {
+		dbg_printf("lookup hash %lu index %lu order 0 aridx 0\n",
+			   hash, index);
+		return &ht->t.tbl[0]->nodes[index];
+	}
 	/*
 	 * 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))];
+		   hash, index, order, index & ((1UL << (order - 1)) - 1));
+	return &ht->t.tbl[order]->nodes[index & ((1UL << (order - 1)) - 1)];
 }
 
 /*
@@ -854,12 +886,6 @@ void _cds_lfht_add(struct cds_lfht *ht,
 
 	assert(!is_dummy(node));
 	assert(!is_removed(node));
-	if (!size) {
-		assert(dummy);
-		assert(!unique_ret);
-		node->p.next = flag_dummy(get_end());
-		return;		/* Initial first add (head) */
-	}
 	lookup = lookup_bucket(ht, size, bit_reverse_ulong(node->p.reverse_hash));
 	for (;;) {
 		uint32_t chain_len = 0;
@@ -877,22 +903,38 @@ void _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;
+
+			/* uniquely add */
 			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)) {
-				unique_ret->node = clear_flag(iter);
-				unique_ret->next = next;
+			    && clear_flag(iter)->p.reverse_hash == node->p.reverse_hash) {
+				struct cds_lfht_iter d_iter = { .node = node, .next = iter, };
+
+				/*
+				 * uniquely adding inserts the node as the first
+				 * node of the identical-hash-value node chain.
+				 *
+				 * This semantic ensures no duplicated keys
+				 * should ever be observable in the table
+				 * (including observe one node by one node
+				 * by forward iterations)
+				 */
+				cds_lfht_next_duplicate(ht, &d_iter);
+				if (!d_iter.node)
+					goto insert;
+
+				*unique_ret = d_iter;
 				return;
 			}
+
 			/* Only account for identical reverse hash once */
 			if (iter_prev->p.reverse_hash != clear_flag(iter)->p.reverse_hash
 			    && !is_dummy(next))
@@ -1051,16 +1093,17 @@ void init_table_populate_partition(struct cds_lfht *ht, unsigned long i,
 {
 	unsigned long j;
 
+	assert(i > ht->min_alloc_order);
 	ht->cds_lfht_rcu_read_lock();
 	for (j = start; j < start + len; j++) {
 		struct cds_lfht_node *new_node =
 			(struct cds_lfht_node *) &ht->t.tbl[i]->nodes[j];
 
 		dbg_printf("init populate: i %lu j %lu hash %lu\n",
-			   i, j, !i ? 0 : (1UL << (i - 1)) + j);
+			   i, j, (1UL << (i - 1)) + j);
 		new_node->p.reverse_hash =
-			bit_reverse_ulong(!i ? 0 : (1UL << (i - 1)) + j);
-		_cds_lfht_add(ht, !i ? 0 : (1UL << (i - 1)),
+				bit_reverse_ulong((1UL << (i - 1)) + j);
+		_cds_lfht_add(ht, 1UL << (i - 1),
 				new_node, NULL, 1);
 	}
 	ht->cds_lfht_rcu_read_unlock();
@@ -1082,21 +1125,21 @@ void init_table_populate(struct cds_lfht *ht, unsigned long i,
 
 static
 void init_table(struct cds_lfht *ht,
-		unsigned long first_order, unsigned long len_order)
+		unsigned long first_order, unsigned long last_order)
 {
-	unsigned long i, end_order;
+	unsigned long i;
 
-	dbg_printf("init table: first_order %lu end_order %lu\n",
-		   first_order, first_order + len_order);
-	end_order = first_order + len_order;
-	for (i = first_order; i < end_order; i++) {
+	dbg_printf("init table: first_order %lu last_order %lu\n",
+		   first_order, last_order);
+	assert(first_order > ht->min_alloc_order);
+	for (i = first_order; i <= last_order; i++) {
 		unsigned long len;
 
-		len = !i ? 1 : 1UL << (i - 1);
+		len = 1UL << (i - 1);
 		dbg_printf("init order %lu len: %lu\n", i, len);
 
 		/* Stop expand if the resize target changes under us */
-		if (CMM_LOAD_SHARED(ht->t.resize_target) < (!i ? 1 : (1UL << i)))
+		if (CMM_LOAD_SHARED(ht->t.resize_target) < (1UL << i))
 			break;
 
 		ht->t.tbl[i] = calloc(1, len * sizeof(struct _cds_lfht_node));
@@ -1112,9 +1155,9 @@ void init_table(struct cds_lfht *ht,
 		 * Update table size.
 		 */
 		cmm_smp_wmb();	/* populate data before RCU size */
-		CMM_STORE_SHARED(ht->t.size, !i ? 1 : (1UL << i));
+		CMM_STORE_SHARED(ht->t.size, 1UL << i);
 
-		dbg_printf("init new size: %lu\n", !i ? 1 : (1UL << i));
+		dbg_printf("init new size: %lu\n", 1UL << i);
 		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
 			break;
 	}
@@ -1151,17 +1194,17 @@ void remove_table_partition(struct cds_lfht *ht, unsigned long i,
 {
 	unsigned long j;
 
+	assert(i > ht->min_alloc_order);
 	ht->cds_lfht_rcu_read_lock();
 	for (j = start; j < start + len; j++) {
 		struct cds_lfht_node *fini_node =
 			(struct cds_lfht_node *) &ht->t.tbl[i]->nodes[j];
 
 		dbg_printf("remove entry: i %lu j %lu hash %lu\n",
-			   i, j, !i ? 0 : (1UL << (i - 1)) + j);
+			   i, j, (1UL << (i - 1)) + j);
 		fini_node->p.reverse_hash =
-			bit_reverse_ulong(!i ? 0 : (1UL << (i - 1)) + j);
-		(void) _cds_lfht_del(ht, !i ? 0 : (1UL << (i - 1)),
-				fini_node, 1);
+			bit_reverse_ulong((1UL << (i - 1)) + j);
+		(void) _cds_lfht_del(ht, 1UL << (i - 1), fini_node, 1);
 	}
 	ht->cds_lfht_rcu_read_unlock();
 }
@@ -1182,19 +1225,18 @@ void remove_table(struct cds_lfht *ht, unsigned long i, unsigned long len)
 
 static
 void fini_table(struct cds_lfht *ht,
-		unsigned long first_order, unsigned long len_order)
+		unsigned long first_order, unsigned long last_order)
 {
-	long i, end_order;
+	long i;
 	void *free_by_rcu = NULL;
 
-	dbg_printf("fini table: first_order %lu end_order %lu\n",
-		   first_order, first_order + len_order);
-	end_order = first_order + len_order;
-	assert(first_order > 0);
-	for (i = end_order - 1; i >= first_order; i--) {
+	dbg_printf("fini table: first_order %lu last_order %lu\n",
+		   first_order, last_order);
+	assert(first_order > ht->min_alloc_order);
+	for (i = last_order; i >= first_order; i--) {
 		unsigned long len;
 
-		len = !i ? 1 : 1UL << (i - 1);
+		len = 1UL << (i - 1);
 		dbg_printf("fini order %lu len: %lu\n", i, len);
 
 		/* Stop shrink if the resize target changes under us */
@@ -1235,10 +1277,54 @@ void fini_table(struct cds_lfht *ht,
 	}
 }
 
+static
+void cds_lfht_create_dummy(struct cds_lfht *ht, unsigned long size)
+{
+	struct _cds_lfht_node *prev, *node;
+	unsigned long order, len, i, j;
+
+	ht->t.tbl[0] = calloc(1, ht->min_alloc_size * sizeof(struct _cds_lfht_node));
+	assert(ht->t.tbl[0]);
+
+	dbg_printf("create dummy: order %lu index %lu hash %lu\n", 0, 0, 0);
+	ht->t.tbl[0]->nodes[0].next = flag_dummy(get_end());
+	ht->t.tbl[0]->nodes[0].reverse_hash = 0;
+
+	for (order = 1; order < get_count_order_ulong(size) + 1; order++) {
+		len = 1UL << (order - 1);
+		if (order <= ht->min_alloc_order) {
+			ht->t.tbl[order] = (struct rcu_level *) (ht->t.tbl[0]->nodes + len);
+		} else {
+			ht->t.tbl[order] = calloc(1, len * sizeof(struct _cds_lfht_node));
+			assert(ht->t.tbl[order]);
+		}
+
+		i = 0;
+		prev = ht->t.tbl[i]->nodes;
+		for (j = 0; j < len; j++) {
+			if (j & (j - 1)) {	/* Between power of 2 */
+				prev++;
+			} else if (j) {		/* At each power of 2 */
+				i++;
+				prev = ht->t.tbl[i]->nodes;
+			}
+
+			node = &ht->t.tbl[order]->nodes[j];
+			dbg_printf("create dummy: order %lu index %lu hash %lu\n",
+				   order, j, j + len);
+			node->next = prev->next;
+			assert(is_dummy(node->next));
+			node->reverse_hash = bit_reverse_ulong(j + len);
+			prev->next = flag_dummy((struct cds_lfht_node *)node);
+		}
+	}
+}
+
 struct cds_lfht *_cds_lfht_new(cds_lfht_hash_fct hash_fct,
 			cds_lfht_compare_fct compare_fct,
 			unsigned long hash_seed,
 			unsigned long init_size,
+			unsigned long min_alloc_size,
 			int flags,
 			void (*cds_lfht_call_rcu)(struct rcu_head *head,
 					void (*func)(struct rcu_head *head)),
@@ -1254,9 +1340,14 @@ struct cds_lfht *_cds_lfht_new(cds_lfht_hash_fct hash_fct,
 	struct cds_lfht *ht;
 	unsigned long order;
 
+	/* min_alloc_size must be power of two */
+	if (!min_alloc_size || (min_alloc_size & (min_alloc_size - 1)))
+		return NULL;
 	/* init_size must be power of two */
-	if (init_size && (init_size & (init_size - 1)))
+	if (!init_size || (init_size & (init_size - 1)))
 		return NULL;
+	min_alloc_size = max(min_alloc_size, MIN_TABLE_SIZE);
+	init_size = max(init_size, min_alloc_size);
 	ht = calloc(1, sizeof(struct cds_lfht));
 	assert(ht);
 	ht->hash_fct = hash_fct;
@@ -1271,17 +1362,16 @@ struct cds_lfht *_cds_lfht_new(cds_lfht_hash_fct hash_fct,
 	ht->cds_lfht_rcu_register_thread = cds_lfht_rcu_register_thread;
 	ht->cds_lfht_rcu_unregister_thread = cds_lfht_rcu_unregister_thread;
 	ht->resize_attr = attr;
-	ht->percpu_count = alloc_per_cpu_items_count();
+	ht->split_count = alloc_split_items_count();
 	/* this mutex should not nest in read-side C.S. */
 	pthread_mutex_init(&ht->resize_mutex, NULL);
-	order = get_count_order_ulong(max(init_size, MIN_TABLE_SIZE)) + 1;
 	ht->flags = flags;
-	ht->cds_lfht_rcu_thread_offline();
-	pthread_mutex_lock(&ht->resize_mutex);
-	ht->t.resize_target = 1UL << (order - 1);
-	init_table(ht, 0, order);
-	pthread_mutex_unlock(&ht->resize_mutex);
-	ht->cds_lfht_rcu_thread_online();
+	order = get_count_order_ulong(init_size);
+	ht->t.resize_target = 1UL << order;
+	cds_lfht_create_dummy(ht, 1UL << order);
+	ht->t.size = 1UL << order;
+	ht->min_alloc_size = min_alloc_size;
+	ht->min_alloc_order = get_count_order_ulong(min_alloc_size);
 	return ht;
 }
 
@@ -1404,7 +1494,7 @@ void cds_lfht_add(struct cds_lfht *ht, struct cds_lfht_node *node)
 
 	size = rcu_dereference(ht->t.size);
 	_cds_lfht_add(ht, size, node, NULL, 0);
-	ht_count_add(ht, size);
+	ht_count_add(ht, size, hash);
 }
 
 struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht,
@@ -1419,7 +1509,7 @@ struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht,
 	size = rcu_dereference(ht->t.size);
 	_cds_lfht_add(ht, size, node, &iter, 0);
 	if (iter.node == node)
-		ht_count_add(ht, size);
+		ht_count_add(ht, size, hash);
 	return iter.node;
 }
 
@@ -1436,7 +1526,7 @@ struct cds_lfht_node *cds_lfht_add_replace(struct cds_lfht *ht,
 	for (;;) {
 		_cds_lfht_add(ht, size, node, &iter, 0);
 		if (iter.node == node) {
-			ht_count_add(ht, size);
+			ht_count_add(ht, size, hash);
 			return NULL;
 		}
 
@@ -1457,13 +1547,15 @@ int cds_lfht_replace(struct cds_lfht *ht, struct cds_lfht_iter *old_iter,
 
 int cds_lfht_del(struct cds_lfht *ht, struct cds_lfht_iter *iter)
 {
-	unsigned long size;
+	unsigned long size, hash;
 	int ret;
 
 	size = rcu_dereference(ht->t.size);
 	ret = _cds_lfht_del(ht, size, iter->node, 0);
-	if (!ret)
-		ht_count_del(ht, size);
+	if (!ret) {
+		hash = bit_reverse_ulong(iter->node->p.reverse_hash);
+		ht_count_del(ht, size, hash);
+	}
 	return ret;
 }
 
@@ -1499,7 +1591,12 @@ int cds_lfht_delete_dummy(struct cds_lfht *ht)
 				bit_reverse_ulong(ht->t.tbl[order]->nodes[i].reverse_hash));
 			assert(is_dummy(ht->t.tbl[order]->nodes[i].next));
 		}
-		poison_free(ht->t.tbl[order]);
+
+		if (order == ht->min_alloc_order)
+			poison_free(ht->t.tbl[0]);
+		else if (order > ht->min_alloc_order)
+			poison_free(ht->t.tbl[order]);
+		/* Nothing to delete for order < ht->min_alloc_order */
 	}
 	return 0;
 }
@@ -1520,7 +1617,7 @@ int cds_lfht_destroy(struct cds_lfht *ht, pthread_attr_t **attr)
 	ret = cds_lfht_delete_dummy(ht);
 	if (ret)
 		return ret;
-	free_per_cpu_items_count(ht->percpu_count);
+	free_split_items_count(ht->split_count);
 	if (attr)
 		*attr = ht->resize_attr;
 	poison_free(ht);
@@ -1538,12 +1635,12 @@ void cds_lfht_count_nodes(struct cds_lfht *ht,
 	unsigned long nr_dummy = 0;
 
 	*approx_before = 0;
-	if (nr_cpus_mask >= 0) {
+	if (split_count_mask >= 0) {
 		int i;
 
-		for (i = 0; i < nr_cpus_mask + 1; i++) {
-			*approx_before += uatomic_read(&ht->percpu_count[i].add);
-			*approx_before -= uatomic_read(&ht->percpu_count[i].del);
+		for (i = 0; i < split_count_mask + 1; i++) {
+			*approx_before += uatomic_read(&ht->split_count[i].add);
+			*approx_before -= uatomic_read(&ht->split_count[i].del);
 		}
 	}
 
@@ -1568,12 +1665,12 @@ void cds_lfht_count_nodes(struct cds_lfht *ht,
 	} while (!is_end(node));
 	dbg_printf("number of dummy nodes: %lu\n", nr_dummy);
 	*approx_after = 0;
-	if (nr_cpus_mask >= 0) {
+	if (split_count_mask >= 0) {
 		int i;
 
-		for (i = 0; i < nr_cpus_mask + 1; i++) {
-			*approx_after += uatomic_read(&ht->percpu_count[i].add);
-			*approx_after -= uatomic_read(&ht->percpu_count[i].del);
+		for (i = 0; i < split_count_mask + 1; i++) {
+			*approx_after += uatomic_read(&ht->split_count[i].add);
+			*approx_after -= uatomic_read(&ht->split_count[i].del);
 		}
 	}
 }
@@ -1585,12 +1682,12 @@ void _do_cds_lfht_grow(struct cds_lfht *ht,
 {
 	unsigned long old_order, new_order;
 
-	old_order = get_count_order_ulong(old_size) + 1;
-	new_order = get_count_order_ulong(new_size) + 1;
+	old_order = get_count_order_ulong(old_size);
+	new_order = get_count_order_ulong(new_size);
 	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);
+	init_table(ht, old_order + 1, new_order);
 }
 
 /* called with resize mutex held */
@@ -1600,15 +1697,15 @@ void _do_cds_lfht_shrink(struct cds_lfht *ht,
 {
 	unsigned long old_order, new_order;
 
-	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;
+	new_size = max(new_size, ht->min_alloc_size);
+	old_order = get_count_order_ulong(old_size);
+	new_order = get_count_order_ulong(new_size);
 	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. */
-	fini_table(ht, new_order, old_order - new_order);
+	fini_table(ht, new_order + 1, old_order);
 }
 
 
@@ -1650,7 +1747,7 @@ static
 void resize_target_update_count(struct cds_lfht *ht,
 				unsigned long count)
 {
-	count = max(count, MIN_TABLE_SIZE);
+	count = max(count, ht->min_alloc_size);
 	uatomic_set(&ht->t.resize_target, count);
 }
 
@@ -1705,7 +1802,7 @@ void cds_lfht_resize_lazy(struct cds_lfht *ht, unsigned long size, int growth)
 	}
 }
 
-#if defined(HAVE_SCHED_GETCPU) && defined(HAVE_SYSCONF)
+#if defined(HAVE_SYSCONF)
 
 static
 void cds_lfht_resize_lazy_count(struct cds_lfht *ht, unsigned long size,