X-Git-Url: https://git.liburcu.org/?p=urcu.git;a=blobdiff_plain;f=rculfhash.c;h=6c648f17da8a939d22894067eb1cc8c24620a8e0;hp=41b774adfdd3257f53e7acb548f147d43fe0dc10;hb=c0b8a8653e108be201ae8d1ca9918461c97d45ed;hpb=1ee8f000286db8180c95cbc740566554f403f933 diff --git a/rculfhash.c b/rculfhash.c index 41b774a..6c648f1 100644 --- a/rculfhash.c +++ b/rculfhash.c @@ -4,6 +4,7 @@ * Userspace RCU library - Lock-Free Resizable RCU Hash Table * * Copyright 2010-2011 - Mathieu Desnoyers + * Copyright 2011 - Lai Jiangshan * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public @@ -159,19 +160,15 @@ #include "config.h" #include #include +#include #include #include #include #include +#include #include #include -#ifdef DEBUG -#define dbg_printf(fmt, args...) printf("[debug rculfhash] " fmt, ## args) -#else -#define dbg_printf(fmt, args...) -#endif - /* * Split-counters lazily update the global counter each 1024 * addition/removal. It automatically keeps track of resize required. @@ -186,13 +183,8 @@ /* * Define the minimum table size. */ -#define MIN_TABLE_SIZE 1 - -#if (CAA_BITS_PER_LONG == 32) -#define MAX_TABLE_ORDER 32 -#else -#define MAX_TABLE_ORDER 64 -#endif +#define MIN_TABLE_ORDER 0 +#define MIN_TABLE_SIZE (1UL << MIN_TABLE_ORDER) /* * Minimum number of bucket nodes to touch per thread to parallelize grow/shrink. @@ -200,14 +192,6 @@ #define MIN_PARTITION_PER_THREAD_ORDER 12 #define MIN_PARTITION_PER_THREAD (1UL << MIN_PARTITION_PER_THREAD_ORDER) -#ifndef min -#define min(a, b) ((a) < (b) ? (a) : (b)) -#endif - -#ifndef max -#define max(a, b) ((a) > (b) ? (a) : (b)) -#endif - /* * The removed flag needs to be updated atomically with the pointer. * It indicates that no node must attach to the node scheduled for @@ -236,63 +220,6 @@ struct ht_items_count { unsigned long add, del; } __attribute__((aligned(CAA_CACHE_LINE_SIZE))); -/* - * rcu_level: 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 rcu_level { - /* Note: manually update allocation length when adding a field */ - struct cds_lfht_node nodes[0]; -}; - -/* - * rcu_table: Contains the size and desired new size if a resize - * operation is in progress, as well as the statically-sized array of - * rcu_level pointers. - */ -struct rcu_table { - unsigned long size; /* always a power of 2, shared (RCU) */ - unsigned long resize_target; - int resize_initiated; - struct rcu_level *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_order; - unsigned long min_alloc_size; - int flags; - /* - * We need to put the work threads offline (QSBR) when taking this - * mutex, because we use synchronize_rcu within this mutex critical - * section, which waits on read-side critical sections, and could - * therefore cause grace-period deadlock if we hold off RCU G.P. - * completion. - */ - pthread_mutex_t resize_mutex; /* resize mutex: add/del mutex */ - unsigned int in_progress_resize, in_progress_destroy; - void (*cds_lfht_call_rcu)(struct rcu_head *head, - void (*func)(struct rcu_head *head)); - void (*cds_lfht_synchronize_rcu)(void); - void (*cds_lfht_rcu_read_lock)(void); - void (*cds_lfht_rcu_read_unlock)(void); - void (*cds_lfht_rcu_thread_offline)(void); - void (*cds_lfht_rcu_thread_online)(void); - void (*cds_lfht_rcu_register_thread)(void); - 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 *split_count; /* split item count */ -}; - /* * rcu_resize_work: Contains arguments passed to RCU worker thread * responsible for performing lazy resize. @@ -315,15 +242,6 @@ struct partition_resize_work { unsigned long start, unsigned long len); }; -static -void _cds_lfht_add(struct cds_lfht *ht, - cds_lfht_match_fct match, - void *key, - unsigned long size, - struct cds_lfht_node *node, - struct cds_lfht_iter *unique_ret, - int bucket); - /* * Algorithm to reverse bits in a word by lookup table, extended to * 64-bit words. @@ -520,18 +438,6 @@ int get_count_order_ulong(unsigned long x) return fls_ulong(x - 1); } -#ifdef POISON_FREE -#define poison_free(ptr) \ - do { \ - if (ptr) { \ - memset(ptr, 0x42, sizeof(*(ptr))); \ - free(ptr); \ - } \ - } while (0) -#else -#define poison_free(ptr) free(ptr) -#endif - static void cds_lfht_resize_lazy_grow(struct cds_lfht *ht, unsigned long size, int growth); @@ -621,26 +527,28 @@ void ht_count_add(struct cds_lfht *ht, unsigned long size, unsigned long hash) { unsigned long split_count; int index; + long count; if (caa_unlikely(!ht->split_count)) return; index = ht_get_split_count_index(hash); split_count = uatomic_add_return(&ht->split_count[index].add, 1); - if (caa_unlikely(!(split_count & ((1UL << COUNT_COMMIT_ORDER) - 1)))) { - long count; - - dbg_printf("add split count %lu\n", split_count); - count = uatomic_add_return(&ht->count, - 1UL << COUNT_COMMIT_ORDER); - /* If power of 2 */ - if (!(count & (count - 1))) { - if ((count >> CHAIN_LEN_RESIZE_THRESHOLD) < size) - return; - dbg_printf("add set global %ld\n", count); - cds_lfht_resize_lazy_count(ht, size, - count >> (CHAIN_LEN_TARGET - 1)); - } - } + if (caa_likely(split_count & ((1UL << COUNT_COMMIT_ORDER) - 1))) + return; + /* Only if number of add multiple of 1UL << COUNT_COMMIT_ORDER */ + + dbg_printf("add split count %lu\n", split_count); + count = uatomic_add_return(&ht->count, + 1UL << COUNT_COMMIT_ORDER); + if (caa_likely(count & (count - 1))) + return; + /* Only if global count is power of 2 */ + + if ((count >> CHAIN_LEN_RESIZE_THRESHOLD) < size) + return; + dbg_printf("add set global %ld\n", count); + cds_lfht_resize_lazy_count(ht, size, + count >> (CHAIN_LEN_TARGET - 1)); } static @@ -648,32 +556,34 @@ void ht_count_del(struct cds_lfht *ht, unsigned long size, unsigned long hash) { unsigned long split_count; int index; + long count; if (caa_unlikely(!ht->split_count)) return; index = ht_get_split_count_index(hash); split_count = uatomic_add_return(&ht->split_count[index].del, 1); - if (caa_unlikely(!(split_count & ((1UL << COUNT_COMMIT_ORDER) - 1)))) { - long count; - - dbg_printf("del split count %lu\n", split_count); - count = uatomic_add_return(&ht->count, - -(1UL << COUNT_COMMIT_ORDER)); - /* If power of 2 */ - if (!(count & (count - 1))) { - if ((count >> CHAIN_LEN_RESIZE_THRESHOLD) >= size) - return; - dbg_printf("del set global %ld\n", count); - /* - * Don't shrink table if the number of nodes is below a - * certain threshold. - */ - if (count < (1UL << COUNT_COMMIT_ORDER) * (split_count_mask + 1)) - return; - cds_lfht_resize_lazy_count(ht, size, - count >> (CHAIN_LEN_TARGET - 1)); - } - } + if (caa_likely(split_count & ((1UL << COUNT_COMMIT_ORDER) - 1))) + return; + /* Only if number of deletes multiple of 1UL << COUNT_COMMIT_ORDER */ + + dbg_printf("del split count %lu\n", split_count); + count = uatomic_add_return(&ht->count, + -(1UL << COUNT_COMMIT_ORDER)); + if (caa_likely(count & (count - 1))) + return; + /* Only if global count is power of 2 */ + + if ((count >> CHAIN_LEN_RESIZE_THRESHOLD) >= size) + return; + dbg_printf("del set global %ld\n", count); + /* + * Don't shrink table if the number of nodes is below a + * certain threshold. + */ + if (count < (1UL << COUNT_COMMIT_ORDER) * (split_count_mask + 1)) + return; + cds_lfht_resize_lazy_count(ht, size, + count >> (CHAIN_LEN_TARGET - 1)); } static @@ -756,28 +666,34 @@ unsigned long _uatomic_xchg_monotonic_increase(unsigned long *ptr, } static +void cds_lfht_alloc_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + return ht->mm->alloc_bucket_table(ht, order); +} + +/* + * cds_lfht_free_bucket_table() should be called with decreasing order. + * When cds_lfht_free_bucket_table(0) is called, it means the whole + * lfht is destroyed. + */ +static +void cds_lfht_free_bucket_table(struct cds_lfht *ht, unsigned long order) +{ + return ht->mm->free_bucket_table(ht, order); +} + +static inline +struct cds_lfht_node *bucket_at(struct cds_lfht *ht, unsigned long index) +{ + return ht->bucket_at(ht, index); +} + +static inline struct cds_lfht_node *lookup_bucket(struct cds_lfht *ht, unsigned long size, unsigned long hash) { - unsigned long index, order; - 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 & ((1UL << (order - 1)) - 1)); - return &ht->t.tbl[order]->nodes[index & ((1UL << (order - 1)) - 1)]; + return bucket_at(ht, hash & (size - 1)); } /* @@ -890,7 +806,7 @@ int _cds_lfht_replace(struct cds_lfht *ht, unsigned long size, static void _cds_lfht_add(struct cds_lfht *ht, cds_lfht_match_fct match, - void *key, + const void *key, unsigned long size, struct cds_lfht_node *node, struct cds_lfht_iter *unique_ret, @@ -1042,9 +958,9 @@ void *partition_resize_thread(void *arg) { struct partition_resize_work *work = arg; - work->ht->cds_lfht_rcu_register_thread(); + work->ht->flavor->register_thread(); work->fct(work->ht, work->i, work->start, work->len); - work->ht->cds_lfht_rcu_unregister_thread(); + work->ht->flavor->unregister_thread(); return NULL; } @@ -1105,21 +1021,20 @@ static void init_table_populate_partition(struct cds_lfht *ht, unsigned long i, unsigned long start, unsigned long len) { - unsigned long j; + unsigned long j, size = 1UL << (i - 1); - 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 = &ht->t.tbl[i]->nodes[j]; + assert(i > MIN_TABLE_ORDER); + ht->flavor->read_lock(); + for (j = size + start; j < size + start + len; j++) { + struct cds_lfht_node *new_node = bucket_at(ht, j); - dbg_printf("init populate: i %lu j %lu hash %lu\n", - i, j, (1UL << (i - 1)) + j); - new_node->reverse_hash = - bit_reverse_ulong((1UL << (i - 1)) + j); - _cds_lfht_add(ht, NULL, NULL, 1UL << (i - 1), - new_node, NULL, 1); + assert(j >= size && j < (size << 1)); + dbg_printf("init populate: order %lu index %lu hash %lu\n", + i, j, j); + new_node->reverse_hash = bit_reverse_ulong(j); + _cds_lfht_add(ht, NULL, NULL, size, new_node, NULL, 1); } - ht->cds_lfht_rcu_read_unlock(); + ht->flavor->read_unlock(); } static @@ -1128,9 +1043,9 @@ void init_table_populate(struct cds_lfht *ht, unsigned long i, { assert(nr_cpus_mask != -1); if (nr_cpus_mask < 0 || len < 2 * MIN_PARTITION_PER_THREAD) { - ht->cds_lfht_rcu_thread_online(); + ht->flavor->thread_online(); init_table_populate_partition(ht, i, 0, len); - ht->cds_lfht_rcu_thread_offline(); + ht->flavor->thread_offline(); return; } partition_resize_helper(ht, i, len, init_table_populate_partition); @@ -1144,7 +1059,7 @@ void init_table(struct cds_lfht *ht, dbg_printf("init table: first_order %lu last_order %lu\n", first_order, last_order); - assert(first_order > ht->min_alloc_order); + assert(first_order > MIN_TABLE_ORDER); for (i = first_order; i <= last_order; i++) { unsigned long len; @@ -1152,11 +1067,10 @@ 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; - ht->t.tbl[i] = calloc(1, len * sizeof(struct cds_lfht_node)); - assert(ht->t.tbl[i]); + cds_lfht_alloc_bucket_table(ht, i); /* * Set all bucket nodes reverse hash values for a level and @@ -1168,7 +1082,7 @@ 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)) @@ -1205,20 +1119,20 @@ static void remove_table_partition(struct cds_lfht *ht, unsigned long i, unsigned long start, unsigned long len) { - unsigned long j; + unsigned long j, size = 1UL << (i - 1); - 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 = &ht->t.tbl[i]->nodes[j]; + assert(i > MIN_TABLE_ORDER); + ht->flavor->read_lock(); + for (j = size + start; j < size + start + len; j++) { + struct cds_lfht_node *fini_node = bucket_at(ht, j); - dbg_printf("remove entry: i %lu j %lu hash %lu\n", - i, j, (1UL << (i - 1)) + j); - fini_node->reverse_hash = - bit_reverse_ulong((1UL << (i - 1)) + j); - (void) _cds_lfht_del(ht, 1UL << (i - 1), fini_node, 1); + assert(j >= size && j < (size << 1)); + dbg_printf("remove entry: order %lu index %lu hash %lu\n", + i, j, j); + fini_node->reverse_hash = bit_reverse_ulong(j); + (void) _cds_lfht_del(ht, size, fini_node, 1); } - ht->cds_lfht_rcu_read_unlock(); + ht->flavor->read_unlock(); } static @@ -1227,24 +1141,29 @@ void remove_table(struct cds_lfht *ht, unsigned long i, unsigned long len) assert(nr_cpus_mask != -1); if (nr_cpus_mask < 0 || len < 2 * MIN_PARTITION_PER_THREAD) { - ht->cds_lfht_rcu_thread_online(); + ht->flavor->thread_online(); remove_table_partition(ht, i, 0, len); - ht->cds_lfht_rcu_thread_offline(); + ht->flavor->thread_offline(); return; } partition_resize_helper(ht, i, len, remove_table_partition); } +/* + * fini_table() is never called for first_order == 0, which is why + * free_by_rcu_order == 0 can be used as criterion to know if free must + * be called. + */ static void fini_table(struct cds_lfht *ht, unsigned long first_order, unsigned long last_order) { long i; - void *free_by_rcu = NULL; + unsigned long free_by_rcu_order = 0; dbg_printf("fini table: first_order %lu last_order %lu\n", first_order, last_order); - assert(first_order > ht->min_alloc_order); + assert(first_order > MIN_TABLE_ORDER); for (i = last_order; i >= first_order; i--) { unsigned long len; @@ -1252,11 +1171,11 @@ 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 @@ -1264,9 +1183,9 @@ void fini_table(struct cds_lfht *ht, * releasing the old bucket nodes. Otherwise their lookup will * return a logically removed node as insert position. */ - ht->cds_lfht_synchronize_rcu(); - if (free_by_rcu) - free(free_by_rcu); + ht->flavor->update_synchronize_rcu(); + if (free_by_rcu_order) + cds_lfht_free_bucket_table(ht, free_by_rcu_order); /* * Set "removed" flag in bucket nodes about to be removed. @@ -1276,16 +1195,16 @@ void fini_table(struct cds_lfht *ht, */ remove_table(ht, i, len); - free_by_rcu = ht->t.tbl[i]; + free_by_rcu_order = 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); + if (free_by_rcu_order) { + ht->flavor->update_synchronize_rcu(); + cds_lfht_free_bucket_table(ht, free_by_rcu_order); } } @@ -1293,103 +1212,106 @@ static void cds_lfht_create_bucket(struct cds_lfht *ht, unsigned long size) { struct cds_lfht_node *prev, *node; - unsigned long order, len, i, j; + unsigned long order, len, i; - ht->t.tbl[0] = calloc(1, ht->min_alloc_size * sizeof(struct cds_lfht_node)); - assert(ht->t.tbl[0]); + cds_lfht_alloc_bucket_table(ht, 0); - dbg_printf("create bucket: order %lu index %lu hash %lu\n", 0, 0, 0); - ht->t.tbl[0]->nodes[0].next = flag_bucket(get_end()); - ht->t.tbl[0]->nodes[0].reverse_hash = 0; + dbg_printf("create bucket: order 0 index 0 hash 0\n"); + node = bucket_at(ht, 0); + node->next = flag_bucket(get_end()); + node->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]); - } + cds_lfht_alloc_bucket_table(ht, 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; - } + for (i = 0; i < len; i++) { + /* + * Now, we are trying to init the node with the + * hash=(len+i) (which is also a bucket with the + * index=(len+i)) and insert it into the hash table, + * so this node has to be inserted after the bucket + * with the index=(len+i)&(len-1)=i. And because there + * is no other non-bucket node nor bucket node with + * larger index/hash inserted, so the bucket node + * being inserted should be inserted directly linked + * after the bucket node with index=i. + */ + prev = bucket_at(ht, i); + node = bucket_at(ht, len + i); - node = &ht->t.tbl[order]->nodes[j]; dbg_printf("create bucket: order %lu index %lu hash %lu\n", - order, j, j + len); + order, len + i, len + i); + node->reverse_hash = bit_reverse_ulong(len + i); + + /* insert after prev */ + assert(is_bucket(prev->next)); node->next = prev->next; - assert(is_bucket(node->next)); - node->reverse_hash = bit_reverse_ulong(j + len); prev->next = flag_bucket(node); } } } struct cds_lfht *_cds_lfht_new(unsigned long init_size, - unsigned long min_alloc_size, + unsigned long min_nr_alloc_buckets, + unsigned long max_nr_buckets, int flags, - void (*cds_lfht_call_rcu)(struct rcu_head *head, - void (*func)(struct rcu_head *head)), - void (*cds_lfht_synchronize_rcu)(void), - void (*cds_lfht_rcu_read_lock)(void), - void (*cds_lfht_rcu_read_unlock)(void), - void (*cds_lfht_rcu_thread_offline)(void), - void (*cds_lfht_rcu_thread_online)(void), - void (*cds_lfht_rcu_register_thread)(void), - void (*cds_lfht_rcu_unregister_thread)(void), + const struct cds_lfht_mm_type *mm, + const struct rcu_flavor_struct *flavor, pthread_attr_t *attr) { 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))) + /* min_nr_alloc_buckets must be power of two */ + if (!min_nr_alloc_buckets || (min_nr_alloc_buckets & (min_nr_alloc_buckets - 1))) return NULL; + /* init_size must be power of two */ 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)); + + /* max_nr_buckets == 0 for order based mm means infinite */ + if (mm == &cds_lfht_mm_order && !max_nr_buckets) + max_nr_buckets = 1UL << (MAX_TABLE_ORDER - 1); + + /* max_nr_buckets must be power of two */ + if (!max_nr_buckets || (max_nr_buckets & (max_nr_buckets - 1))) + return NULL; + + min_nr_alloc_buckets = max(min_nr_alloc_buckets, MIN_TABLE_SIZE); + init_size = max(init_size, MIN_TABLE_SIZE); + max_nr_buckets = max(max_nr_buckets, min_nr_alloc_buckets); + init_size = min(init_size, max_nr_buckets); + + ht = mm->alloc_cds_lfht(min_nr_alloc_buckets, max_nr_buckets); assert(ht); + assert(ht->mm == mm); + assert(ht->bucket_at == mm->bucket_at); + ht->flags = flags; - ht->cds_lfht_call_rcu = cds_lfht_call_rcu; - ht->cds_lfht_synchronize_rcu = cds_lfht_synchronize_rcu; - ht->cds_lfht_rcu_read_lock = cds_lfht_rcu_read_lock; - ht->cds_lfht_rcu_read_unlock = cds_lfht_rcu_read_unlock; - ht->cds_lfht_rcu_thread_offline = cds_lfht_rcu_thread_offline; - ht->cds_lfht_rcu_thread_online = cds_lfht_rcu_thread_online; - ht->cds_lfht_rcu_register_thread = cds_lfht_rcu_register_thread; - ht->cds_lfht_rcu_unregister_thread = cds_lfht_rcu_unregister_thread; + ht->flavor = flavor; ht->resize_attr = attr; alloc_split_items_count(ht); /* 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->min_alloc_size = min_alloc_size; - ht->min_alloc_order = get_count_order_ulong(min_alloc_size); + ht->resize_target = 1UL << order; cds_lfht_create_bucket(ht, 1UL << order); - ht->t.size = 1UL << order; + ht->size = 1UL << order; return ht; } -void cds_lfht_lookup(struct cds_lfht *ht, cds_lfht_match_fct match, - unsigned long hash, void *key, struct cds_lfht_iter *iter) +void cds_lfht_lookup(struct cds_lfht *ht, unsigned long hash, + cds_lfht_match_fct match, const void *key, + struct cds_lfht_iter *iter) { struct cds_lfht_node *node, *next, *bucket; unsigned long reverse_hash, size; 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); @@ -1419,7 +1341,7 @@ void cds_lfht_lookup(struct cds_lfht *ht, cds_lfht_match_fct match, } void cds_lfht_next_duplicate(struct cds_lfht *ht, cds_lfht_match_fct match, - void *key, struct cds_lfht_iter *iter) + const void *key, struct cds_lfht_iter *iter) { struct cds_lfht_node *node, *next; unsigned long reverse_hash; @@ -1475,14 +1397,11 @@ void cds_lfht_next(struct cds_lfht *ht, struct cds_lfht_iter *iter) void cds_lfht_first(struct cds_lfht *ht, struct cds_lfht_iter *iter) { - struct cds_lfht_node *lookup; - /* * Get next after first bucket node. The first bucket node is the * first node of the linked list. */ - lookup = &ht->t.tbl[0]->nodes[0]; - iter->next = lookup->next; + iter->next = bucket_at(ht, 0)->next; cds_lfht_next(ht, iter); } @@ -1492,22 +1411,22 @@ 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); } struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht, - cds_lfht_match_fct match, - void *key, unsigned long hash, + cds_lfht_match_fct match, + const void *key, struct cds_lfht_node *node) { unsigned long size; 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); @@ -1515,16 +1434,16 @@ struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht, } struct cds_lfht_node *cds_lfht_add_replace(struct cds_lfht *ht, - cds_lfht_match_fct match, - void *key, unsigned long hash, + cds_lfht_match_fct match, + const void *key, struct cds_lfht_node *node) { unsigned long size; 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) { @@ -1542,7 +1461,7 @@ 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); } @@ -1552,7 +1471,7 @@ 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); @@ -1568,7 +1487,7 @@ int cds_lfht_delete_bucket(struct cds_lfht *ht) unsigned long order, i, size; /* Check that the table is empty */ - node = &ht->t.tbl[0]->nodes[0]; + node = bucket_at(ht, 0); do { node = clear_flag(node)->next; if (!is_bucket(node)) @@ -1579,25 +1498,18 @@ 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 (order = 0; order < get_count_order_ulong(size) + 1; order++) { - unsigned long len; + for (i = 0; i < size; i++) { + node = bucket_at(ht, i); + dbg_printf("delete bucket: index %lu expected hash %lu hash %lu\n", + i, i, bit_reverse_ulong(node->reverse_hash)); + assert(is_bucket(node->next)); + } - len = !order ? 1 : 1UL << (order - 1); - for (i = 0; i < len; i++) { - dbg_printf("delete order %lu i %lu hash %lu\n", - order, i, - bit_reverse_ulong(ht->t.tbl[order]->nodes[i].reverse_hash)); - assert(is_bucket(ht->t.tbl[order]->nodes[i].next)); - } + for (order = get_count_order_ulong(size); (long)order >= 0; order--) + cds_lfht_free_bucket_table(ht, 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; } @@ -1647,7 +1559,7 @@ void cds_lfht_count_nodes(struct cds_lfht *ht, *removed = 0; /* Count non-bucket nodes in the table */ - node = &ht->t.tbl[0]->nodes[0]; + node = bucket_at(ht, 0); do { next = rcu_dereference(node->next); if (is_removed(next)) { @@ -1695,7 +1607,7 @@ void _do_cds_lfht_shrink(struct cds_lfht *ht, { unsigned long old_order, new_order; - new_size = max(new_size, ht->min_alloc_size); + new_size = max(new_size, MIN_TABLE_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", @@ -1720,42 +1632,43 @@ 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 void resize_target_update_count(struct cds_lfht *ht, unsigned long count) { - count = max(count, ht->min_alloc_size); - uatomic_set(&ht->t.resize_target, count); + count = max(count, MIN_TABLE_SIZE); + count = min(count, ht->max_nr_buckets); + 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); - ht->cds_lfht_rcu_thread_offline(); + CMM_STORE_SHARED(ht->resize_initiated, 1); + ht->flavor->thread_offline(); pthread_mutex_lock(&ht->resize_mutex); _do_cds_lfht_resize(ht); pthread_mutex_unlock(&ht->resize_mutex); - ht->cds_lfht_rcu_thread_online(); + ht->flavor->thread_online(); } static @@ -1765,11 +1678,11 @@ void do_resize_cb(struct rcu_head *head) caa_container_of(head, struct rcu_resize_work, head); struct cds_lfht *ht = work->ht; - ht->cds_lfht_rcu_thread_offline(); + ht->flavor->thread_offline(); pthread_mutex_lock(&ht->resize_mutex); _do_cds_lfht_resize(ht); pthread_mutex_unlock(&ht->resize_mutex); - ht->cds_lfht_rcu_thread_online(); + ht->flavor->thread_online(); poison_free(work); cmm_smp_mb(); /* finish resize before decrement */ uatomic_dec(&ht->in_progress_resize); @@ -1782,7 +1695,7 @@ 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)) { @@ -1791,8 +1704,8 @@ void __cds_lfht_resize_lazy_launch(struct cds_lfht *ht) } work = malloc(sizeof(*work)); work->ht = ht; - ht->cds_lfht_call_rcu(&work->head, do_resize_cb); - CMM_STORE_SHARED(ht->t.resize_initiated, 1); + ht->flavor->update_call_rcu(&work->head, do_resize_cb); + CMM_STORE_SHARED(ht->resize_initiated, 1); } } @@ -1801,6 +1714,7 @@ void cds_lfht_resize_lazy_grow(struct cds_lfht *ht, unsigned long size, int grow { unsigned long target_size = size << growth; + target_size = min(target_size, ht->max_nr_buckets); if (resize_target_grow(ht, target_size) >= target_size) return; @@ -1818,7 +1732,8 @@ void cds_lfht_resize_lazy_count(struct cds_lfht *ht, unsigned long size, { if (!(ht->flags & CDS_LFHT_AUTO_RESIZE)) return; - count = max(count, ht->min_alloc_size); + count = max(count, MIN_TABLE_SIZE); + count = min(count, ht->max_nr_buckets); if (count == size) return; /* Already the right size, no resize needed */ if (count > size) { /* lazy grow */ @@ -1828,7 +1743,7 @@ 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)