Merge struct rcu_table into struct cds_lfht
authorLai Jiangshan <laijs@cn.fujitsu.com>
Mon, 28 Nov 2011 13:24:27 +0000 (08:24 -0500)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mon, 28 Nov 2011 13:24:27 +0000 (08:24 -0500)
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
rculfhash.c

index 18cead30cce6d015064ab965e96a552b6237fb9b..8dc9c59903dd28f85129eeb5c2e82990c0503689 100644 (file)
@@ -239,37 +239,15 @@ struct ht_items_count {
        unsigned long add, del;
 } __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 
-/*
- * rcu_table: Contains the size and desired new size if a resize
- * operation is in progress, as well as the statically-sized array of
- * bucket table pointers.
- */
-struct rcu_table {
-       unsigned long size;     /* always a power of 2, shared (RCU) */
-       unsigned long resize_target;
-       int resize_initiated;
-
-       /*
-        * Contains the per order-index-level bucket node table. The size
-        * of each bucket node table is half the number of hashes contained
-        * in this order (except for order 0). The minimum allocation size
-        * parameter allows combining the bucket node arrays of the lowermost
-        * levels to improve cache locality for small index orders.
-        */
-       struct cds_lfht_node *tbl[MAX_TABLE_ORDER];
-};
-
 /*
  * cds_lfht: Top-level data structure representing a lock-free hash
  * table. Defined in the implementation file to make it be an opaque
  * cookie to users.
  */
 struct cds_lfht {
-       struct rcu_table t;
-       unsigned long min_alloc_buckets_order;
-       unsigned long min_nr_alloc_buckets;
-       unsigned long max_nr_buckets;
+       unsigned long size;     /* always a power of 2, shared (RCU) */
        int flags;
+
        /*
         * We need to put the work threads offline (QSBR) when taking this
         * mutex, because we use synchronize_rcu within this mutex critical
@@ -278,11 +256,26 @@ struct cds_lfht {
         * completion.
         */
        pthread_mutex_t resize_mutex;   /* resize mutex: add/del mutex */
+       pthread_attr_t *resize_attr;    /* Resize threads attributes */
        unsigned int in_progress_resize, in_progress_destroy;
+       unsigned long resize_target;
+       int resize_initiated;
        const struct rcu_flavor_struct *flavor;
-       pthread_attr_t *resize_attr;    /* Resize threads attributes */
+
        long count;                     /* global approximate item count */
        struct ht_items_count *split_count;     /* split item count */
+
+       unsigned long min_alloc_buckets_order;
+       unsigned long min_nr_alloc_buckets;
+       unsigned long max_nr_buckets;
+       /*
+        * Contains the per order-index-level bucket node table. The size
+        * of each bucket node table is half the number of hashes contained
+        * in this order (except for order 0). The minimum allocation size
+        * parameter allows combining the bucket node arrays of the lowermost
+        * levels to improve cache locality for small index orders.
+        */
+       struct cds_lfht_node *tbl[MAX_TABLE_ORDER];
 };
 
 /*
@@ -751,13 +744,13 @@ static
 void cds_lfht_alloc_bucket_table(struct cds_lfht *ht, unsigned long order)
 {
        if (order == 0) {
-               ht->t.tbl[0] = calloc(ht->min_nr_alloc_buckets,
+               ht->tbl[0] = calloc(ht->min_nr_alloc_buckets,
                        sizeof(struct cds_lfht_node));
-               assert(ht->t.tbl[0]);
+               assert(ht->tbl[0]);
        } else if (order > ht->min_alloc_buckets_order) {
-               ht->t.tbl[order] = calloc(1UL << (order -1),
+               ht->tbl[order] = calloc(1UL << (order -1),
                        sizeof(struct cds_lfht_node));
-               assert(ht->t.tbl[order]);
+               assert(ht->tbl[order]);
        }
        /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */
 }
@@ -771,9 +764,9 @@ static
 void cds_lfht_free_bucket_table(struct cds_lfht *ht, unsigned long order)
 {
        if (order == 0)
-               poison_free(ht->t.tbl[0]);
+               poison_free(ht->tbl[0]);
        else if (order > ht->min_alloc_buckets_order)
-               poison_free(ht->t.tbl[order]);
+               poison_free(ht->tbl[order]);
        /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */
 }
 
@@ -785,7 +778,7 @@ struct cds_lfht_node *bucket_at(struct cds_lfht *ht, unsigned long index)
        if ((__builtin_constant_p(index) && index == 0)
                        || index < ht->min_nr_alloc_buckets) {
                dbg_printf("bucket index %lu order 0 aridx 0\n", index);
-               return &ht->t.tbl[0][index];
+               return &ht->tbl[0][index];
        }
        /*
         * equivalent to get_count_order_ulong(index + 1), but optimizes
@@ -795,7 +788,7 @@ struct cds_lfht_node *bucket_at(struct cds_lfht *ht, unsigned long index)
        order = fls_ulong(index);
        dbg_printf("bucket index %lu order %lu aridx %lu\n",
                   index, order, index & ((1UL << (order - 1)) - 1));
-       return &ht->t.tbl[order][index & ((1UL << (order - 1)) - 1)];
+       return &ht->tbl[order][index & ((1UL << (order - 1)) - 1)];
 }
 
 static inline
@@ -1177,7 +1170,7 @@ void init_table(struct cds_lfht *ht,
                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) < (1UL << i))
+               if (CMM_LOAD_SHARED(ht->resize_target) < (1UL << i))
                        break;
 
                cds_lfht_alloc_bucket_table(ht, i);
@@ -1192,7 +1185,7 @@ void init_table(struct cds_lfht *ht,
                 * Update table size.
                 */
                cmm_smp_wmb();  /* populate data before RCU size */
-               CMM_STORE_SHARED(ht->t.size, 1UL << i);
+               CMM_STORE_SHARED(ht->size, 1UL << i);
 
                dbg_printf("init new size: %lu\n", 1UL << i);
                if (CMM_LOAD_SHARED(ht->in_progress_destroy))
@@ -1276,11 +1269,11 @@ void fini_table(struct cds_lfht *ht,
                dbg_printf("fini order %lu len: %lu\n", i, len);
 
                /* Stop shrink if the resize target changes under us */
-               if (CMM_LOAD_SHARED(ht->t.resize_target) > (1UL << (i - 1)))
+               if (CMM_LOAD_SHARED(ht->resize_target) > (1UL << (i - 1)))
                        break;
 
                cmm_smp_wmb();  /* populate data before RCU size */
-               CMM_STORE_SHARED(ht->t.size, 1UL << (i - 1));
+               CMM_STORE_SHARED(ht->size, 1UL << (i - 1));
 
                /*
                 * We need to wait for all add operations to reach Q.S. (and
@@ -1395,12 +1388,12 @@ struct cds_lfht *_cds_lfht_new(unsigned long init_size,
        /* this mutex should not nest in read-side C.S. */
        pthread_mutex_init(&ht->resize_mutex, NULL);
        order = get_count_order_ulong(init_size);
-       ht->t.resize_target = 1UL << order;
+       ht->resize_target = 1UL << order;
        ht->min_nr_alloc_buckets = min_nr_alloc_buckets;
        ht->min_alloc_buckets_order = get_count_order_ulong(min_nr_alloc_buckets);
        ht->max_nr_buckets = max_nr_buckets;
        cds_lfht_create_bucket(ht, 1UL << order);
-       ht->t.size = 1UL << order;
+       ht->size = 1UL << order;
        return ht;
 }
 
@@ -1413,7 +1406,7 @@ void cds_lfht_lookup(struct cds_lfht *ht, unsigned long hash,
 
        reverse_hash = bit_reverse_ulong(hash);
 
-       size = rcu_dereference(ht->t.size);
+       size = rcu_dereference(ht->size);
        bucket = lookup_bucket(ht, size, hash);
        /* We can always skip the bucket node initially */
        node = rcu_dereference(bucket->next);
@@ -1513,7 +1506,7 @@ void cds_lfht_add(struct cds_lfht *ht, unsigned long hash,
        unsigned long size;
 
        node->reverse_hash = bit_reverse_ulong((unsigned long) hash);
-       size = rcu_dereference(ht->t.size);
+       size = rcu_dereference(ht->size);
        _cds_lfht_add(ht, NULL, NULL, size, node, NULL, 0);
        ht_count_add(ht, size, hash);
 }
@@ -1528,7 +1521,7 @@ struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht,
        struct cds_lfht_iter iter;
 
        node->reverse_hash = bit_reverse_ulong((unsigned long) hash);
-       size = rcu_dereference(ht->t.size);
+       size = rcu_dereference(ht->size);
        _cds_lfht_add(ht, match, key, size, node, &iter, 0);
        if (iter.node == node)
                ht_count_add(ht, size, hash);
@@ -1545,7 +1538,7 @@ struct cds_lfht_node *cds_lfht_add_replace(struct cds_lfht *ht,
        struct cds_lfht_iter iter;
 
        node->reverse_hash = bit_reverse_ulong((unsigned long) hash);
-       size = rcu_dereference(ht->t.size);
+       size = rcu_dereference(ht->size);
        for (;;) {
                _cds_lfht_add(ht, match, key, size, node, &iter, 0);
                if (iter.node == node) {
@@ -1563,7 +1556,7 @@ int cds_lfht_replace(struct cds_lfht *ht, struct cds_lfht_iter *old_iter,
 {
        unsigned long size;
 
-       size = rcu_dereference(ht->t.size);
+       size = rcu_dereference(ht->size);
        return _cds_lfht_replace(ht, size, old_iter->node, old_iter->next,
                        new_node);
 }
@@ -1573,7 +1566,7 @@ int cds_lfht_del(struct cds_lfht *ht, struct cds_lfht_iter *iter)
        unsigned long size, hash;
        int ret;
 
-       size = rcu_dereference(ht->t.size);
+       size = rcu_dereference(ht->size);
        ret = _cds_lfht_del(ht, size, iter->node, 0);
        if (!ret) {
                hash = bit_reverse_ulong(iter->node->reverse_hash);
@@ -1600,7 +1593,7 @@ int cds_lfht_delete_bucket(struct cds_lfht *ht)
         * size accessed without rcu_dereference because hash table is
         * being destroyed.
         */
-       size = ht->t.size;
+       size = ht->size;
        /* Internal sanity check: all nodes left should be bucket */
        for (i = 0; i < size; i++) {
                node = bucket_at(ht, i);
@@ -1734,23 +1727,23 @@ void _do_cds_lfht_resize(struct cds_lfht *ht)
                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);
+               ht->resize_initiated = 1;
+               old_size = ht->size;
+               new_size = CMM_LOAD_SHARED(ht->resize_target);
                if (old_size < new_size)
                        _do_cds_lfht_grow(ht, old_size, new_size);
                else if (old_size > new_size)
                        _do_cds_lfht_shrink(ht, old_size, new_size);
-               ht->t.resize_initiated = 0;
+               ht->resize_initiated = 0;
                /* write resize_initiated before read resize_target */
                cmm_smp_mb();
-       } while (ht->t.size != CMM_LOAD_SHARED(ht->t.resize_target));
+       } while (ht->size != CMM_LOAD_SHARED(ht->resize_target));
 }
 
 static
 unsigned long resize_target_grow(struct cds_lfht *ht, unsigned long new_size)
 {
-       return _uatomic_xchg_monotonic_increase(&ht->t.resize_target, new_size);
+       return _uatomic_xchg_monotonic_increase(&ht->resize_target, new_size);
 }
 
 static
@@ -1759,13 +1752,13 @@ void resize_target_update_count(struct cds_lfht *ht,
 {
        count = max(count, MIN_TABLE_SIZE);
        count = min(count, ht->max_nr_buckets);
-       uatomic_set(&ht->t.resize_target, count);
+       uatomic_set(&ht->resize_target, count);
 }
 
 void cds_lfht_resize(struct cds_lfht *ht, unsigned long new_size)
 {
        resize_target_update_count(ht, new_size);
-       CMM_STORE_SHARED(ht->t.resize_initiated, 1);
+       CMM_STORE_SHARED(ht->resize_initiated, 1);
        ht->flavor->thread_offline();
        pthread_mutex_lock(&ht->resize_mutex);
        _do_cds_lfht_resize(ht);
@@ -1797,7 +1790,7 @@ void __cds_lfht_resize_lazy_launch(struct cds_lfht *ht)
 
        /* Store resize_target before read resize_initiated */
        cmm_smp_mb();
-       if (!CMM_LOAD_SHARED(ht->t.resize_initiated)) {
+       if (!CMM_LOAD_SHARED(ht->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)) {
@@ -1807,7 +1800,7 @@ void __cds_lfht_resize_lazy_launch(struct cds_lfht *ht)
                work = malloc(sizeof(*work));
                work->ht = ht;
                ht->flavor->update_call_rcu(&work->head, do_resize_cb);
-               CMM_STORE_SHARED(ht->t.resize_initiated, 1);
+               CMM_STORE_SHARED(ht->resize_initiated, 1);
        }
 }
 
@@ -1845,7 +1838,7 @@ void cds_lfht_resize_lazy_count(struct cds_lfht *ht, unsigned long size,
                for (;;) {
                        unsigned long s;
 
-                       s = uatomic_cmpxchg(&ht->t.resize_target, size, count);
+                       s = uatomic_cmpxchg(&ht->resize_target, size, count);
                        if (s == size)
                                break;  /* no resize needed */
                        if (s > size)
This page took 0.030167 seconds and 4 git commands to generate.