X-Git-Url: https://git.liburcu.org/?a=blobdiff_plain;f=rculfhash.c;h=91958dd108b38f8177fc49432ffa1a139ddf1505;hb=34e01c64a78d067648d14c96ddd519b69abbeaeb;hp=8dc9c59903dd28f85129eeb5c2e82990c0503689;hpb=7b3893e462472cd25a19ac6f5ebc09aeb031ea60;p=urcu.git diff --git a/rculfhash.c b/rculfhash.c index 8dc9c59..91958dd 100644 --- a/rculfhash.c +++ b/rculfhash.c @@ -165,15 +165,10 @@ #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. @@ -191,26 +186,12 @@ #define MIN_TABLE_ORDER 0 #define MIN_TABLE_SIZE (1UL << MIN_TABLE_ORDER) -#if (CAA_BITS_PER_LONG == 32) -#define MAX_TABLE_ORDER 32 -#else -#define MAX_TABLE_ORDER 64 -#endif - /* * Minimum number of bucket nodes to touch per thread to parallelize grow/shrink. */ #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 @@ -239,45 +220,6 @@ struct ht_items_count { unsigned long add, del; } __attribute__((aligned(CAA_CACHE_LINE_SIZE))); -/* - * 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 { - 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 - * 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 */ - 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; - - 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]; -}; - /* * rcu_resize_work: Contains arguments passed to RCU worker thread * responsible for performing lazy resize. @@ -300,15 +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, - const 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. @@ -472,7 +405,7 @@ unsigned int fls_u32(uint32_t x) } #endif -unsigned int fls_ulong(unsigned long x) +unsigned int cds_lfht_fls_ulong(unsigned long x) { #if (CAA_BITS_PER_LONG == 32) return fls_u32(x); @@ -485,7 +418,7 @@ unsigned int fls_ulong(unsigned long x) * Return the minimum order for which x <= (1UL << order). * Return -1 if x is 0. */ -int get_count_order_u32(uint32_t x) +int cds_lfht_get_count_order_u32(uint32_t x) { if (!x) return -1; @@ -497,26 +430,14 @@ int get_count_order_u32(uint32_t x) * Return the minimum order for which x <= (1UL << order). * Return -1 if x is 0. */ -int get_count_order_ulong(unsigned long x) +int cds_lfht_get_count_order_ulong(unsigned long x) { if (!x) return -1; - return fls_ulong(x - 1); + return cds_lfht_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); @@ -541,7 +462,7 @@ static void ht_init_nr_cpus_mask(void) * round up number of CPUs to next power of two, so we * can use & for modulo. */ - maxcpus = 1UL << get_count_order_ulong(maxcpus); + maxcpus = 1UL << cds_lfht_get_count_order_ulong(maxcpus); nr_cpus_mask = maxcpus - 1; } #else /* #if defined(HAVE_SYSCONF) */ @@ -606,26 +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 @@ -633,32 +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 @@ -680,7 +605,7 @@ void check_resize(struct cds_lfht *ht, unsigned long size, uint32_t chain_len) chain_len); if (chain_len >= CHAIN_LEN_RESIZE_THRESHOLD) cds_lfht_resize_lazy_grow(ht, size, - get_count_order_u32(chain_len - (CHAIN_LEN_TARGET - 1))); + cds_lfht_get_count_order_u32(chain_len - (CHAIN_LEN_TARGET - 1))); } static @@ -743,16 +668,7 @@ unsigned long _uatomic_xchg_monotonic_increase(unsigned long *ptr, static void cds_lfht_alloc_bucket_table(struct cds_lfht *ht, unsigned long order) { - if (order == 0) { - ht->tbl[0] = calloc(ht->min_nr_alloc_buckets, - sizeof(struct cds_lfht_node)); - assert(ht->tbl[0]); - } else if (order > ht->min_alloc_buckets_order) { - ht->tbl[order] = calloc(1UL << (order -1), - sizeof(struct cds_lfht_node)); - assert(ht->tbl[order]); - } - /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */ + return ht->mm->alloc_bucket_table(ht, order); } /* @@ -763,32 +679,13 @@ void cds_lfht_alloc_bucket_table(struct cds_lfht *ht, unsigned long order) static void cds_lfht_free_bucket_table(struct cds_lfht *ht, unsigned long order) { - if (order == 0) - poison_free(ht->tbl[0]); - else if (order > ht->min_alloc_buckets_order) - poison_free(ht->tbl[order]); - /* Nothing to do for 0 < order && order <= ht->min_alloc_buckets_order */ + return ht->mm->free_bucket_table(ht, order); } static inline struct cds_lfht_node *bucket_at(struct cds_lfht *ht, unsigned long index) { - unsigned long order; - - 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->tbl[0][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("bucket index %lu order %lu aridx %lu\n", - index, order, index & ((1UL << (order - 1)) - 1)); - return &ht->tbl[order][index & ((1UL << (order - 1)) - 1)]; + return ht->bucket_at(ht, index); } static inline @@ -1089,7 +986,7 @@ void partition_resize_helper(struct cds_lfht *ht, unsigned long i, } else { nr_threads = 1; } - partition_len = len >> get_count_order_ulong(nr_threads); + partition_len = len >> cds_lfht_get_count_order_ulong(nr_threads); work = calloc(nr_threads, sizeof(*work)); assert(work); for (thread = 0; thread < nr_threads; thread++) { @@ -1232,7 +1129,6 @@ void remove_table_partition(struct cds_lfht *ht, unsigned long i, 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->flavor->read_unlock(); @@ -1252,6 +1148,11 @@ void remove_table(struct cds_lfht *ht, unsigned long i, unsigned long len) 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) @@ -1319,7 +1220,7 @@ void cds_lfht_create_bucket(struct cds_lfht *ht, unsigned long size) node->next = flag_bucket(get_end()); node->reverse_hash = 0; - for (order = 1; order < get_count_order_ulong(size) + 1; order++) { + for (order = 1; order < cds_lfht_get_count_order_ulong(size) + 1; order++) { len = 1UL << (order - 1); cds_lfht_alloc_bucket_table(ht, order); @@ -1354,6 +1255,7 @@ struct cds_lfht *_cds_lfht_new(unsigned long init_size, unsigned long min_nr_alloc_buckets, unsigned long max_nr_buckets, int flags, + const struct cds_lfht_mm_type *mm, const struct rcu_flavor_struct *flavor, pthread_attr_t *attr) { @@ -1368,7 +1270,32 @@ struct cds_lfht *_cds_lfht_new(unsigned long init_size, if (!init_size || (init_size & (init_size - 1))) return NULL; - if (!max_nr_buckets) + /* + * Memory management plugin default. + */ + if (!mm) { + if (CAA_BITS_PER_LONG > 32 + && max_nr_buckets + && max_nr_buckets <= (1ULL << 32)) { + /* + * For 64-bit architectures, with max number of + * buckets small enough not to use the entire + * 64-bit memory mapping space (and allowing a + * fair number of hash table instances), use the + * mmap allocator, which is faster than the + * order allocator. + */ + mm = &cds_lfht_mm_mmap; + } else { + /* + * The fallback is to use the order allocator. + */ + mm = &cds_lfht_mm_order; + } + } + + /* 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 */ @@ -1379,19 +1306,20 @@ struct cds_lfht *_cds_lfht_new(unsigned long init_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 = calloc(1, sizeof(struct cds_lfht)); + + 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->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); + order = cds_lfht_get_count_order_ulong(init_size); 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->size = 1UL << order; return ht; @@ -1602,7 +1530,7 @@ int cds_lfht_delete_bucket(struct cds_lfht *ht) assert(is_bucket(node->next)); } - for (order = get_count_order_ulong(size); (long)order >= 0; order--) + for (order = cds_lfht_get_count_order_ulong(size); (long)order >= 0; order--) cds_lfht_free_bucket_table(ht, order); return 0; @@ -1687,8 +1615,8 @@ void _do_cds_lfht_grow(struct cds_lfht *ht, { unsigned long old_order, new_order; - old_order = get_count_order_ulong(old_size); - new_order = get_count_order_ulong(new_size); + old_order = cds_lfht_get_count_order_ulong(old_size); + new_order = cds_lfht_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); @@ -1703,8 +1631,8 @@ 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); - new_order = get_count_order_ulong(new_size); + old_order = cds_lfht_get_count_order_ulong(old_size); + new_order = cds_lfht_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);