diff options
Diffstat (limited to 'lib/librte_table')
| -rw-r--r-- | lib/librte_table/Makefile | 7 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash.h | 305 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_cuckoo.c | 205 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_ext.c | 417 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_key16.c | 750 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_key32.c | 436 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_key8.c | 716 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_lru.c | 516 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_version.map | 26 |
9 files changed, 1152 insertions, 2226 deletions
diff --git a/lib/librte_table/Makefile b/lib/librte_table/Makefile index 8ddc8804..2e32fbf1 100644 --- a/lib/librte_table/Makefile +++ b/lib/librte_table/Makefile @@ -38,10 +38,15 @@ LIB = librte_table.a CFLAGS += -O3 CFLAGS += $(WERROR_FLAGS) +LDLIBS += -lrte_eal -lrte_mempool -lrte_mbuf -lrte_port +LDLIBS += -lrte_lpm -lrte_hash +ifeq ($(CONFIG_RTE_LIBRTE_ACL),y) +LDLIBS += -lrte_acl +endif EXPORT_MAP := rte_table_version.map -LIBABIVER := 2 +LIBABIVER := 3 # # all source are stored in SRCS-y diff --git a/lib/librte_table/rte_table_hash.h b/lib/librte_table/rte_table_hash.h index 57505a6f..15f1902b 100644 --- a/lib/librte_table/rte_table_hash.h +++ b/lib/librte_table/rte_table_hash.h @@ -1,7 +1,7 @@ /*- * BSD LICENSE * - * Copyright(c) 2010-2016 Intel Corporation. All rights reserved. + * Copyright(c) 2010-2017 Intel Corporation. All rights reserved. * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -45,8 +45,6 @@ extern "C" { * These tables use the exact match criterion to uniquely associate data to * lookup keys. * - * Use-cases: Flow classification table, Address Resolution Protocol (ARP) table - * * Hash table types: * 1. Entry add strategy on bucket full: * a. Least Recently Used (LRU): One of the existing keys in the bucket is @@ -59,7 +57,7 @@ extern "C" { * to the bucket, it also becomes the new MRU key. When a key needs to * be picked and dropped, the most likely candidate for drop, i.e. the * current LRU key, is always picked. The LRU logic requires maintaining - * specific data structures per each bucket. + * specific data structures per each bucket. Use-cases: flow cache, etc. * b. Extendible bucket (ext): The bucket is extended with space for 4 more * keys. This is done by allocating additional memory at table init time, * which is used to create a pool of free keys (the size of this pool is @@ -73,20 +71,8 @@ extern "C" { * first group of 4 keys, the search continues beyond the first group of * 4 keys, potentially until all keys in this bucket are examined. The * extendible bucket logic requires maintaining specific data structures - * per table and per each bucket. - * 2. Key signature computation: - * a. Pre-computed key signature: The key lookup operation is split between - * two CPU cores. The first CPU core (typically the CPU core performing - * packet RX) extracts the key from the input packet, computes the key - * signature and saves both the key and the key signature in the packet - * buffer as packet meta-data. The second CPU core reads both the key and - * the key signature from the packet meta-data and performs the bucket - * search step of the key lookup operation. - * b. Key signature computed on lookup (do-sig): The same CPU core reads - * the key from the packet meta-data, uses it to compute the key - * signature and also performs the bucket search step of the key lookup - * operation. - * 3. Key size: + * per table and per each bucket. Use-cases: flow table, etc. + * 2. Key size: * a. Configurable key size * b. Single key size (8-byte, 16-byte or 32-byte key size) * @@ -98,59 +84,28 @@ extern "C" { /** Hash function */ typedef uint64_t (*rte_table_hash_op_hash)( void *key, + void *key_mask, uint32_t key_size, uint64_t seed); -/** - * Hash tables with configurable key size - * - */ -/** Extendible bucket hash table parameters */ -struct rte_table_hash_ext_params { +/** Hash table parameters */ +struct rte_table_hash_params { + /** Name */ + const char *name; + /** Key size (number of bytes) */ uint32_t key_size; - /** Maximum number of keys */ - uint32_t n_keys; - - /** Number of hash table buckets. Each bucket stores up to 4 keys. */ - uint32_t n_buckets; - - /** Number of hash table bucket extensions. Each bucket extension has - space for 4 keys and each bucket can have 0, 1 or more extensions. */ - uint32_t n_buckets_ext; - - /** Hash function */ - rte_table_hash_op_hash f_hash; - - /** Seed value for the hash function */ - uint64_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - /** Byte offset within packet meta-data where the key is located */ uint32_t key_offset; -}; -/** Extendible bucket hash table operations for pre-computed key signature */ -extern struct rte_table_ops rte_table_hash_ext_ops; - -/** Extendible bucket hash table operations for key signature computed on - lookup ("do-sig") */ -extern struct rte_table_ops rte_table_hash_ext_dosig_ops; - -/** LRU hash table parameters */ -struct rte_table_hash_lru_params { - /** Key size (number of bytes) */ - uint32_t key_size; + /** Key mask */ + uint8_t *key_mask; - /** Maximum number of keys */ + /** Number of keys */ uint32_t n_keys; - /** Number of hash table buckets. Each bucket stores up to 4 keys. */ + /** Number of buckets */ uint32_t n_buckets; /** Hash function */ @@ -158,239 +113,23 @@ struct rte_table_hash_lru_params { /** Seed value for the hash function */ uint64_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - - /** Byte offset within packet meta-data where the key is located */ - uint32_t key_offset; -}; - -/** LRU hash table operations for pre-computed key signature */ -extern struct rte_table_ops rte_table_hash_lru_ops; - -/** LRU hash table operations for key signature computed on lookup ("do-sig") */ -extern struct rte_table_ops rte_table_hash_lru_dosig_ops; - -/** - * 8-byte key hash tables - * - */ -/** LRU hash table parameters */ -struct rte_table_hash_key8_lru_params { - /** Maximum number of entries (and keys) in the table */ - uint32_t n_entries; - - /** Hash function */ - rte_table_hash_op_hash f_hash; - - /** Seed for the hash function */ - uint64_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - - /** Byte offset within packet meta-data where the key is located */ - uint32_t key_offset; - - /** Bit-mask to be AND-ed to the key on lookup */ - uint8_t *key_mask; }; -/** LRU hash table operations for pre-computed key signature */ -extern struct rte_table_ops rte_table_hash_key8_lru_ops; - -/** LRU hash table operations for key signature computed on lookup ("do-sig") */ -extern struct rte_table_ops rte_table_hash_key8_lru_dosig_ops; - -/** Extendible bucket hash table parameters */ -struct rte_table_hash_key8_ext_params { - /** Maximum number of entries (and keys) in the table */ - uint32_t n_entries; - - /** Number of entries (and keys) for hash table bucket extensions. Each - bucket is extended in increments of 4 keys. */ - uint32_t n_entries_ext; - - /** Hash function */ - rte_table_hash_op_hash f_hash; - - /** Seed for the hash function */ - uint64_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - - /** Byte offset within packet meta-data where the key is located */ - uint32_t key_offset; - - /** Bit-mask to be AND-ed to the key on lookup */ - uint8_t *key_mask; -}; - -/** Extendible bucket hash table operations for pre-computed key signature */ +/** Extendible bucket hash table operations */ +extern struct rte_table_ops rte_table_hash_ext_ops; extern struct rte_table_ops rte_table_hash_key8_ext_ops; - -/** Extendible bucket hash table operations for key signature computed on - lookup ("do-sig") */ -extern struct rte_table_ops rte_table_hash_key8_ext_dosig_ops; - -/** - * 16-byte key hash tables - * - */ -/** LRU hash table parameters */ -struct rte_table_hash_key16_lru_params { - /** Maximum number of entries (and keys) in the table */ - uint32_t n_entries; - - /** Hash function */ - rte_table_hash_op_hash f_hash; - - /** Seed for the hash function */ - uint64_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - - /** Byte offset within packet meta-data where the key is located */ - uint32_t key_offset; - - /** Bit-mask to be AND-ed to the key on lookup */ - uint8_t *key_mask; -}; - -/** LRU hash table operations for pre-computed key signature */ -extern struct rte_table_ops rte_table_hash_key16_lru_ops; - -/** LRU hash table operations for key signature computed on lookup - ("do-sig") */ -extern struct rte_table_ops rte_table_hash_key16_lru_dosig_ops; - -/** Extendible bucket hash table parameters */ -struct rte_table_hash_key16_ext_params { - /** Maximum number of entries (and keys) in the table */ - uint32_t n_entries; - - /** Number of entries (and keys) for hash table bucket extensions. Each - bucket is extended in increments of 4 keys. */ - uint32_t n_entries_ext; - - /** Hash function */ - rte_table_hash_op_hash f_hash; - - /** Seed for the hash function */ - uint64_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - - /** Byte offset within packet meta-data where the key is located */ - uint32_t key_offset; - - /** Bit-mask to be AND-ed to the key on lookup */ - uint8_t *key_mask; -}; - -/** Extendible bucket operations for pre-computed key signature */ extern struct rte_table_ops rte_table_hash_key16_ext_ops; - -/** Extendible bucket hash table operations for key signature computed on - lookup ("do-sig") */ -extern struct rte_table_ops rte_table_hash_key16_ext_dosig_ops; - -/** - * 32-byte key hash tables - * - */ -/** LRU hash table parameters */ -struct rte_table_hash_key32_lru_params { - /** Maximum number of entries (and keys) in the table */ - uint32_t n_entries; - - /** Hash function */ - rte_table_hash_op_hash f_hash; - - /** Seed for the hash function */ - uint64_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - - /** Byte offset within packet meta-data where the key is located */ - uint32_t key_offset; -}; - -/** LRU hash table operations for pre-computed key signature */ -extern struct rte_table_ops rte_table_hash_key32_lru_ops; - -/** Extendible bucket hash table parameters */ -struct rte_table_hash_key32_ext_params { - /** Maximum number of entries (and keys) in the table */ - uint32_t n_entries; - - /** Number of entries (and keys) for hash table bucket extensions. Each - bucket is extended in increments of 4 keys. */ - uint32_t n_entries_ext; - - /** Hash function */ - rte_table_hash_op_hash f_hash; - - /** Seed for the hash function */ - uint64_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - - /** Byte offset within packet meta-data where the key is located */ - uint32_t key_offset; -}; - -/** Extendible bucket hash table operations */ extern struct rte_table_ops rte_table_hash_key32_ext_ops; -/** Cuckoo hash table parameters */ -struct rte_table_hash_cuckoo_params { - /** Key size (number of bytes */ - uint32_t key_size; - - /** Maximum number of hash table entries */ - uint32_t n_keys; - - /** Hash function used to calculate hash */ - rte_table_hash_op_hash f_hash; - - /** Seed value or Init value used by f_hash */ - uint32_t seed; - - /** Byte offset within packet meta-data where the 4-byte key signature - is located. Valid for pre-computed key signature tables, ignored for - do-sig tables. */ - uint32_t signature_offset; - - /** Byte offset within packet meta-data where the key is located */ - uint32_t key_offset; +/** LRU hash table operations */ +extern struct rte_table_ops rte_table_hash_lru_ops; - /** Hash table name */ - const char *name; -}; +extern struct rte_table_ops rte_table_hash_key8_lru_ops; +extern struct rte_table_ops rte_table_hash_key16_lru_ops; +extern struct rte_table_ops rte_table_hash_key32_lru_ops; /** Cuckoo hash table operations */ -extern struct rte_table_ops rte_table_hash_cuckoo_dosig_ops; +extern struct rte_table_ops rte_table_hash_cuckoo_ops; #ifdef __cplusplus } diff --git a/lib/librte_table/rte_table_hash_cuckoo.c b/lib/librte_table/rte_table_hash_cuckoo.c index da1597fa..f3845c75 100644 --- a/lib/librte_table/rte_table_hash_cuckoo.c +++ b/lib/librte_table/rte_table_hash_cuckoo.c @@ -1,34 +1,34 @@ /*- - * BSD LICENSE + * BSD LICENSE * - * Copyright(c) 2016 Intel Corporation. All rights reserved. - * All rights reserved. + * Copyright(c) 2010-2017 Intel Corporation. All rights reserved. + * All rights reserved. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Intel Corporation nor the names of its - * contributors may be used to endorse or promote products derived - * from this software without specific prior written permission. + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include <string.h> #include <stdio.h> @@ -66,25 +66,28 @@ struct rte_table_hash { uint32_t n_keys; rte_table_hash_op_hash f_hash; uint32_t seed; - uint32_t signature_offset; uint32_t key_offset; - const char *name; /* cuckoo hash table object */ struct rte_hash *h_table; /* Lookup table */ - uint8_t memory[0] __rte_cache_aligned; }; + uint8_t memory[0] __rte_cache_aligned; +}; static int -check_params_create_hash_cuckoo(const struct -rte_table_hash_cuckoo_params *params) { - /* Check for valid parameters */ +check_params_create_hash_cuckoo(struct rte_table_hash_params *params) +{ if (params == NULL) { RTE_LOG(ERR, TABLE, "NULL Input Parameters.\n"); return -EINVAL; } + if (params->name == NULL) { + RTE_LOG(ERR, TABLE, "Table name is NULL.\n"); + return -EINVAL; + } + if (params->key_size == 0) { RTE_LOG(ERR, TABLE, "Invalid key_size.\n"); return -EINVAL; @@ -100,11 +103,6 @@ rte_table_hash_cuckoo_params *params) { return -EINVAL; } - if (params->name == NULL) { - RTE_LOG(ERR, TABLE, "Table name is NULL.\n"); - return -EINVAL; - } - return 0; } @@ -113,34 +111,24 @@ rte_table_hash_cuckoo_create(void *params, int socket_id, uint32_t entry_size) { - struct rte_hash *rte_hash_handle; + struct rte_table_hash_params *p = params; + struct rte_hash *h_table; struct rte_table_hash *t; - uint32_t total_size, total_cl_size; + uint32_t total_size; /* Check input parameters */ - struct rte_table_hash_cuckoo_params *p = - (struct rte_table_hash_cuckoo_params *) params; - if (check_params_create_hash_cuckoo(params)) return NULL; /* Memory allocation */ - total_cl_size = - (sizeof(struct rte_table_hash) + - RTE_CACHE_LINE_SIZE) / RTE_CACHE_LINE_SIZE; - total_cl_size += (p->n_keys * entry_size + - RTE_CACHE_LINE_SIZE) / RTE_CACHE_LINE_SIZE; - total_size = total_cl_size * RTE_CACHE_LINE_SIZE; - - t = rte_zmalloc_socket("TABLE", - total_size, - RTE_CACHE_LINE_SIZE, - socket_id); + total_size = sizeof(struct rte_table_hash) + + RTE_CACHE_LINE_ROUNDUP(p->n_keys * entry_size); + + t = rte_zmalloc_socket(p->name, total_size, RTE_CACHE_LINE_SIZE, socket_id); if (t == NULL) { RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for Cuckoo hash table\n", - __func__, - (uint32_t)sizeof(struct rte_table_hash)); + "%s: Cannot allocate %u bytes for cuckoo hash table %s\n", + __func__, total_size, p->name); return NULL; } @@ -154,13 +142,13 @@ rte_table_hash_cuckoo_create(void *params, .name = p->name }; - rte_hash_handle = rte_hash_find_existing(p->name); - if (rte_hash_handle == NULL) { - rte_hash_handle = rte_hash_create(&hash_cuckoo_params); - if (NULL == rte_hash_handle) { + h_table = rte_hash_find_existing(p->name); + if (h_table == NULL) { + h_table = rte_hash_create(&hash_cuckoo_params); + if (h_table == NULL) { RTE_LOG(ERR, TABLE, - "%s: failed to create cuckoo hash table. keysize: %u", - __func__, hash_cuckoo_params.key_len); + "%s: failed to create cuckoo hash table %s\n", + __func__, p->name); rte_free(t); return NULL; } @@ -172,26 +160,22 @@ rte_table_hash_cuckoo_create(void *params, t->n_keys = p->n_keys; t->f_hash = p->f_hash; t->seed = p->seed; - t->signature_offset = p->signature_offset; t->key_offset = p->key_offset; - t->name = p->name; - t->h_table = rte_hash_handle; + t->h_table = h_table; RTE_LOG(INFO, TABLE, - "%s: Cuckoo Hash table memory footprint is %u bytes\n", - __func__, total_size); + "%s: Cuckoo hash table %s memory footprint is %u bytes\n", + __func__, p->name, total_size); return t; } static int rte_table_hash_cuckoo_free(void *table) { - if (table == NULL) { - RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); - return -EINVAL; - } - struct rte_table_hash *t = table; + if (table == NULL) + return -EINVAL; + rte_hash_free(t->h_table); rte_free(t); @@ -200,25 +184,18 @@ rte_table_hash_cuckoo_free(void *table) { static int rte_table_hash_cuckoo_entry_add(void *table, void *key, void *entry, - int *key_found, void **entry_ptr) { + int *key_found, void **entry_ptr) +{ + struct rte_table_hash *t = table; int pos = 0; - if (table == NULL) { - RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); - return -EINVAL; - } - - if (key == NULL) { - RTE_LOG(ERR, TABLE, "%s: key parameter is NULL\n", __func__); - return -EINVAL; - } - - if (entry == NULL) { - RTE_LOG(ERR, TABLE, "%s: entry parameter is NULL\n", __func__); + /* Check input parameters */ + if ((table == NULL) || + (key == NULL) || + (entry == NULL) || + (key_found == NULL) || + (entry_ptr == NULL)) return -EINVAL; - } - - struct rte_table_hash *t = table; /* Find Existing entries */ pos = rte_hash_lookup(t->h_table, key); @@ -231,17 +208,15 @@ rte_table_hash_cuckoo_entry_add(void *table, void *key, void *entry, *entry_ptr = existing_entry; return 0; -} else if (pos == -ENOENT) { - /* Entry not found. Adding new entry */ + } + + if (pos == -ENOENT) { + /* Entry not found. Adding new entry */ uint8_t *new_entry; pos = rte_hash_add_key(t->h_table, key); - if (pos < 0) { - RTE_LOG(ERR, TABLE, - "%s: Entry not added, status : %u\n", - __func__, pos); + if (pos < 0) return pos; - } new_entry = &t->memory[pos * t->entry_size]; memcpy(new_entry, entry, t->entry_size); @@ -250,25 +225,22 @@ rte_table_hash_cuckoo_entry_add(void *table, void *key, void *entry, *entry_ptr = new_entry; return 0; } + return pos; } static int rte_table_hash_cuckoo_entry_delete(void *table, void *key, - int *key_found, __rte_unused void *entry) { + int *key_found, void *entry) +{ + struct rte_table_hash *t = table; int pos = 0; - if (table == NULL) { - RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); - return -EINVAL; - } - - if (key == NULL) { - RTE_LOG(ERR, TABLE, "%s: key parameter is NULL\n", __func__); + /* Check input parameters */ + if ((table == NULL) || + (key == NULL) || + (key_found == NULL)) return -EINVAL; - } - - struct rte_table_hash *t = table; pos = rte_hash_del_key(t->h_table, key); if (pos >= 0) { @@ -279,20 +251,21 @@ rte_table_hash_cuckoo_entry_delete(void *table, void *key, memcpy(entry, entry_ptr, t->entry_size); memset(&t->memory[pos * t->entry_size], 0, t->entry_size); + return 0; } + *key_found = 0; return pos; } - static int -rte_table_hash_cuckoo_lookup_dosig(void *table, +rte_table_hash_cuckoo_lookup(void *table, struct rte_mbuf **pkts, uint64_t pkts_mask, uint64_t *lookup_hit_mask, void **entries) { - struct rte_table_hash *t = (struct rte_table_hash *)table; + struct rte_table_hash *t = table; uint64_t pkts_mask_out = 0; uint32_t i; @@ -301,20 +274,19 @@ rte_table_hash_cuckoo_lookup_dosig(void *table, RTE_TABLE_HASH_CUCKOO_STATS_PKTS_IN_ADD(t, n_pkts_in); if ((pkts_mask & (pkts_mask + 1)) == 0) { - const uint8_t *keys[64]; - int32_t positions[64], status; + const uint8_t *keys[RTE_PORT_IN_BURST_SIZE_MAX]; + int32_t positions[RTE_PORT_IN_BURST_SIZE_MAX], status; /* Keys for bulk lookup */ for (i = 0; i < n_pkts_in; i++) keys[i] = RTE_MBUF_METADATA_UINT8_PTR(pkts[i], - t->key_offset); + t->key_offset); /* Bulk Lookup */ status = rte_hash_lookup_bulk(t->h_table, (const void **) keys, n_pkts_in, positions); - if (status == 0) { for (i = 0; i < n_pkts_in; i++) { if (likely(positions[i] >= 0)) { @@ -326,7 +298,7 @@ rte_table_hash_cuckoo_lookup_dosig(void *table, } } } - } else { + } else for (i = 0; i < (uint32_t)(RTE_PORT_IN_BURST_SIZE_MAX - __builtin_clzll(pkts_mask)); i++) { uint64_t pkt_mask = 1LLU << i; @@ -345,7 +317,6 @@ rte_table_hash_cuckoo_lookup_dosig(void *table, } } } - } *lookup_hit_mask = pkts_mask_out; RTE_TABLE_HASH_CUCKOO_STATS_PKTS_LOOKUP_MISS(t, @@ -370,13 +341,13 @@ rte_table_hash_cuckoo_stats_read(void *table, struct rte_table_stats *stats, return 0; } -struct rte_table_ops rte_table_hash_cuckoo_dosig_ops = { +struct rte_table_ops rte_table_hash_cuckoo_ops = { .f_create = rte_table_hash_cuckoo_create, .f_free = rte_table_hash_cuckoo_free, .f_add = rte_table_hash_cuckoo_entry_add, .f_delete = rte_table_hash_cuckoo_entry_delete, .f_add_bulk = NULL, .f_delete_bulk = NULL, - .f_lookup = rte_table_hash_cuckoo_lookup_dosig, + .f_lookup = rte_table_hash_cuckoo_lookup, .f_stats = rte_table_hash_cuckoo_stats_read, }; diff --git a/lib/librte_table/rte_table_hash_ext.c b/lib/librte_table/rte_table_hash_ext.c index e7181026..3af1bcab 100644 --- a/lib/librte_table/rte_table_hash_ext.c +++ b/lib/librte_table/rte_table_hash_ext.c @@ -1,34 +1,34 @@ /*- - * BSD LICENSE + * BSD LICENSE * - * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. - * All rights reserved. + * Copyright(c) 2010-2017 Intel Corporation. All rights reserved. + * All rights reserved. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Intel Corporation nor the names of its - * contributors may be used to endorse or promote products derived - * from this software without specific prior written permission. + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include <string.h> @@ -106,7 +106,6 @@ struct rte_table_hash { uint32_t n_buckets_ext; rte_table_hash_op_hash f_hash; uint64_t seed; - uint32_t signature_offset; uint32_t key_offset; /* Internal */ @@ -120,6 +119,7 @@ struct rte_table_hash { struct grinder grinders[RTE_PORT_IN_BURST_SIZE_MAX]; /* Tables */ + uint64_t *key_mask; struct bucket *buckets; struct bucket *buckets_ext; uint8_t *key_mem; @@ -132,29 +132,53 @@ struct rte_table_hash { }; static int -check_params_create(struct rte_table_hash_ext_params *params) +keycmp(void *a, void *b, void *b_mask, uint32_t n_bytes) { - uint32_t n_buckets_min; + uint64_t *a64 = a, *b64 = b, *b_mask64 = b_mask; + uint32_t i; + + for (i = 0; i < n_bytes / sizeof(uint64_t); i++) + if (a64[i] != (b64[i] & b_mask64[i])) + return 1; + + return 0; +} + +static void +keycpy(void *dst, void *src, void *src_mask, uint32_t n_bytes) +{ + uint64_t *dst64 = dst, *src64 = src, *src_mask64 = src_mask; + uint32_t i; + + for (i = 0; i < n_bytes / sizeof(uint64_t); i++) + dst64[i] = src64[i] & src_mask64[i]; +} + +static int +check_params_create(struct rte_table_hash_params *params) +{ + /* name */ + if (params->name == NULL) { + RTE_LOG(ERR, TABLE, "%s: name invalid value\n", __func__); + return -EINVAL; + } /* key_size */ - if ((params->key_size == 0) || + if ((params->key_size < sizeof(uint64_t)) || (!rte_is_power_of_2(params->key_size))) { RTE_LOG(ERR, TABLE, "%s: key_size invalid value\n", __func__); return -EINVAL; } /* n_keys */ - if ((params->n_keys == 0) || - (!rte_is_power_of_2(params->n_keys))) { + if (params->n_keys == 0) { RTE_LOG(ERR, TABLE, "%s: n_keys invalid value\n", __func__); return -EINVAL; } /* n_buckets */ - n_buckets_min = (params->n_keys + KEYS_PER_BUCKET - 1) / params->n_keys; if ((params->n_buckets == 0) || - (!rte_is_power_of_2(params->n_keys)) || - (params->n_buckets < n_buckets_min)) { + (!rte_is_power_of_2(params->n_buckets))) { RTE_LOG(ERR, TABLE, "%s: n_buckets invalid value\n", __func__); return -EINVAL; } @@ -171,15 +195,13 @@ check_params_create(struct rte_table_hash_ext_params *params) static void * rte_table_hash_ext_create(void *params, int socket_id, uint32_t entry_size) { - struct rte_table_hash_ext_params *p = - params; + struct rte_table_hash_params *p = params; struct rte_table_hash *t; - uint32_t total_size, table_meta_sz; - uint32_t bucket_sz, bucket_ext_sz, key_sz; - uint32_t key_stack_sz, bkt_ext_stack_sz, data_sz; - uint32_t bucket_offset, bucket_ext_offset, key_offset; - uint32_t key_stack_offset, bkt_ext_stack_offset, data_offset; - uint32_t i; + uint64_t table_meta_sz, key_mask_sz, bucket_sz, bucket_ext_sz, key_sz; + uint64_t key_stack_sz, bkt_ext_stack_sz, data_sz, total_size; + uint64_t key_mask_offset, bucket_offset, bucket_ext_offset, key_offset; + uint64_t key_stack_offset, bkt_ext_stack_offset, data_offset; + uint32_t n_buckets_ext, i; /* Check input parameters */ if ((check_params_create(p) != 0) || @@ -188,38 +210,66 @@ rte_table_hash_ext_create(void *params, int socket_id, uint32_t entry_size) (sizeof(struct bucket) != (RTE_CACHE_LINE_SIZE / 2))) return NULL; + /* + * Table dimensioning + * + * Objective: Pick the number of bucket extensions (n_buckets_ext) so that + * it is guaranteed that n_keys keys can be stored in the table at any time. + * + * The worst case scenario takes place when all the n_keys keys fall into + * the same bucket. Actually, due to the KEYS_PER_BUCKET scheme, the worst + * case takes place when (n_keys - KEYS_PER_BUCKET + 1) keys fall into the + * same bucket, while the remaining (KEYS_PER_BUCKET - 1) keys each fall + * into a different bucket. This case defeats the purpose of the hash table. + * It indicates unsuitable f_hash or n_keys to n_buckets ratio. + * + * n_buckets_ext = n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1 + */ + n_buckets_ext = p->n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1; + /* Memory allocation */ table_meta_sz = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_table_hash)); + key_mask_sz = RTE_CACHE_LINE_ROUNDUP(p->key_size); bucket_sz = RTE_CACHE_LINE_ROUNDUP(p->n_buckets * sizeof(struct bucket)); bucket_ext_sz = - RTE_CACHE_LINE_ROUNDUP(p->n_buckets_ext * sizeof(struct bucket)); + RTE_CACHE_LINE_ROUNDUP(n_buckets_ext * sizeof(struct bucket)); key_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * p->key_size); key_stack_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * sizeof(uint32_t)); bkt_ext_stack_sz = - RTE_CACHE_LINE_ROUNDUP(p->n_buckets_ext * sizeof(uint32_t)); + RTE_CACHE_LINE_ROUNDUP(n_buckets_ext * sizeof(uint32_t)); data_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * entry_size); - total_size = table_meta_sz + bucket_sz + bucket_ext_sz + key_sz + - key_stack_sz + bkt_ext_stack_sz + data_sz; + total_size = table_meta_sz + key_mask_sz + bucket_sz + bucket_ext_sz + + key_sz + key_stack_sz + bkt_ext_stack_sz + data_sz; - t = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + if (total_size > SIZE_MAX) { + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes" + " for hash table %s\n", + __func__, total_size, p->name); + return NULL; + } + + t = rte_zmalloc_socket(p->name, + (size_t)total_size, + RTE_CACHE_LINE_SIZE, + socket_id); if (t == NULL) { - RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for hash table\n", - __func__, total_size); + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes" + " for hash table %s\n", + __func__, total_size, p->name); return NULL; } - RTE_LOG(INFO, TABLE, "%s (%u-byte key): Hash table memory footprint is " - "%u bytes\n", __func__, p->key_size, total_size); + RTE_LOG(INFO, TABLE, "%s (%u-byte key): Hash table %s memory " + "footprint is %" PRIu64 " bytes\n", + __func__, p->key_size, p->name, total_size); /* Memory initialization */ t->key_size = p->key_size; t->entry_size = entry_size; t->n_keys = p->n_keys; t->n_buckets = p->n_buckets; - t->n_buckets_ext = p->n_buckets_ext; + t->n_buckets_ext = n_buckets_ext; t->f_hash = p->f_hash; t->seed = p->seed; - t->signature_offset = p->signature_offset; t->key_offset = p->key_offset; /* Internal */ @@ -228,13 +278,15 @@ rte_table_hash_ext_create(void *params, int socket_id, uint32_t entry_size) t->data_size_shl = __builtin_ctzl(entry_size); /* Tables */ - bucket_offset = 0; + key_mask_offset = 0; + bucket_offset = key_mask_offset + key_mask_sz; bucket_ext_offset = bucket_offset + bucket_sz; key_offset = bucket_ext_offset + bucket_ext_sz; key_stack_offset = key_offset + key_sz; bkt_ext_stack_offset = key_stack_offset + key_stack_sz; data_offset = bkt_ext_stack_offset + bkt_ext_stack_sz; + t->key_mask = (uint64_t *) &t->memory[key_mask_offset]; t->buckets = (struct bucket *) &t->memory[bucket_offset]; t->buckets_ext = (struct bucket *) &t->memory[bucket_ext_offset]; t->key_mem = &t->memory[key_offset]; @@ -242,6 +294,12 @@ rte_table_hash_ext_create(void *params, int socket_id, uint32_t entry_size) t->bkt_ext_stack = (uint32_t *) &t->memory[bkt_ext_stack_offset]; t->data_mem = &t->memory[data_offset]; + /* Key mask */ + if (p->key_mask == NULL) + memset(t->key_mask, 0xFF, p->key_size); + else + memcpy(t->key_mask, p->key_mask, p->key_size); + /* Key stack */ for (i = 0; i < t->n_keys; i++) t->key_stack[i] = t->n_keys - 1 - i; @@ -277,7 +335,7 @@ rte_table_hash_ext_entry_add(void *table, void *key, void *entry, uint64_t sig; uint32_t bkt_index, i; - sig = t->f_hash(key, t->key_size, t->seed); + sig = t->f_hash(key, t->key_mask, t->key_size, t->seed); bkt_index = sig & t->bucket_mask; bkt0 = &t->buckets[bkt_index]; sig = (sig >> 16) | 1LLU; @@ -290,7 +348,7 @@ rte_table_hash_ext_entry_add(void *table, void *key, void *entry, uint8_t *bkt_key = &t->key_mem[bkt_key_index << t->key_size_shl]; - if ((sig == bkt_sig) && (memcmp(key, bkt_key, + if ((sig == bkt_sig) && (keycmp(bkt_key, key, t->key_mask, t->key_size) == 0)) { uint8_t *data = &t->data_mem[bkt_key_index << t->data_size_shl]; @@ -327,7 +385,7 @@ rte_table_hash_ext_entry_add(void *table, void *key, void *entry, bkt->sig[i] = (uint16_t) sig; bkt->key_pos[i] = bkt_key_index; - memcpy(bkt_key, key, t->key_size); + keycpy(bkt_key, key, t->key_mask, t->key_size); memcpy(data, entry, t->entry_size); *key_found = 0; @@ -358,7 +416,7 @@ rte_table_hash_ext_entry_add(void *table, void *key, void *entry, /* Install new key into bucket */ bkt->sig[0] = (uint16_t) sig; bkt->key_pos[0] = bkt_key_index; - memcpy(bkt_key, key, t->key_size); + keycpy(bkt_key, key, t->key_mask, t->key_size); memcpy(data, entry, t->entry_size); *key_found = 0; @@ -378,7 +436,7 @@ void *entry) uint64_t sig; uint32_t bkt_index, i; - sig = t->f_hash(key, t->key_size, t->seed); + sig = t->f_hash(key, t->key_mask, t->key_size, t->seed); bkt_index = sig & t->bucket_mask; bkt0 = &t->buckets[bkt_index]; sig = (sig >> 16) | 1LLU; @@ -392,7 +450,7 @@ void *entry) uint8_t *bkt_key = &t->key_mem[bkt_key_index << t->key_size_shl]; - if ((sig == bkt_sig) && (memcmp(key, bkt_key, + if ((sig == bkt_sig) && (keycmp(bkt_key, key, t->key_mask, t->key_size) == 0)) { uint8_t *data = &t->data_mem[bkt_key_index << t->data_size_shl]; @@ -437,8 +495,7 @@ static int rte_table_hash_ext_lookup_unoptimized( struct rte_mbuf **pkts, uint64_t pkts_mask, uint64_t *lookup_hit_mask, - void **entries, - int dosig) + void **entries) { struct rte_table_hash *t = (struct rte_table_hash *) table; uint64_t pkts_mask_out = 0; @@ -458,11 +515,7 @@ static int rte_table_hash_ext_lookup_unoptimized( pkt = pkts[pkt_index]; key = RTE_MBUF_METADATA_UINT8_PTR(pkt, t->key_offset); - if (dosig) - sig = (uint64_t) t->f_hash(key, t->key_size, t->seed); - else - sig = RTE_MBUF_METADATA_UINT32(pkt, - t->signature_offset); + sig = (uint64_t) t->f_hash(key, t->key_mask, t->key_size, t->seed); bkt_index = sig & t->bucket_mask; bkt0 = &t->buckets[bkt_index]; @@ -476,8 +529,8 @@ static int rte_table_hash_ext_lookup_unoptimized( uint8_t *bkt_key = &t->key_mem[bkt_key_index << t->key_size_shl]; - if ((sig == bkt_sig) && (memcmp(key, bkt_key, - t->key_size) == 0)) { + if ((sig == bkt_sig) && (keycmp(bkt_key, key, + t->key_mask, t->key_size) == 0)) { uint8_t *data = &t->data_mem[ bkt_key_index << t->data_size_shl]; @@ -576,11 +629,12 @@ static int rte_table_hash_ext_lookup_unoptimized( { \ uint64_t *pkt_key = RTE_MBUF_METADATA_UINT64_PTR(mbuf, f->key_offset);\ uint64_t *bkt_key = (uint64_t *) key; \ + uint64_t *key_mask = f->key_mask; \ \ switch (f->key_size) { \ case 8: \ { \ - uint64_t xor = pkt_key[0] ^ bkt_key[0]; \ + uint64_t xor = (pkt_key[0] & key_mask[0]) ^ bkt_key[0]; \ match_key = 0; \ if (xor == 0) \ match_key = 1; \ @@ -591,8 +645,8 @@ static int rte_table_hash_ext_lookup_unoptimized( { \ uint64_t xor[2], or; \ \ - xor[0] = pkt_key[0] ^ bkt_key[0]; \ - xor[1] = pkt_key[1] ^ bkt_key[1]; \ + xor[0] = (pkt_key[0] & key_mask[0]) ^ bkt_key[0]; \ + xor[1] = (pkt_key[1] & key_mask[1]) ^ bkt_key[1]; \ or = xor[0] | xor[1]; \ match_key = 0; \ if (or == 0) \ @@ -604,10 +658,10 @@ static int rte_table_hash_ext_lookup_unoptimized( { \ uint64_t xor[4], or; \ \ - xor[0] = pkt_key[0] ^ bkt_key[0]; \ - xor[1] = pkt_key[1] ^ bkt_key[1]; \ - xor[2] = pkt_key[2] ^ bkt_key[2]; \ - xor[3] = pkt_key[3] ^ bkt_key[3]; \ + xor[0] = (pkt_key[0] & key_mask[0]) ^ bkt_key[0]; \ + xor[1] = (pkt_key[1] & key_mask[1]) ^ bkt_key[1]; \ + xor[2] = (pkt_key[2] & key_mask[2]) ^ bkt_key[2]; \ + xor[3] = (pkt_key[3] & key_mask[3]) ^ bkt_key[3]; \ or = xor[0] | xor[1] | xor[2] | xor[3]; \ match_key = 0; \ if (or == 0) \ @@ -619,14 +673,14 @@ static int rte_table_hash_ext_lookup_unoptimized( { \ uint64_t xor[8], or; \ \ - xor[0] = pkt_key[0] ^ bkt_key[0]; \ - xor[1] = pkt_key[1] ^ bkt_key[1]; \ - xor[2] = pkt_key[2] ^ bkt_key[2]; \ - xor[3] = pkt_key[3] ^ bkt_key[3]; \ - xor[4] = pkt_key[4] ^ bkt_key[4]; \ - xor[5] = pkt_key[5] ^ bkt_key[5]; \ - xor[6] = pkt_key[6] ^ bkt_key[6]; \ - xor[7] = pkt_key[7] ^ bkt_key[7]; \ + xor[0] = (pkt_key[0] & key_mask[0]) ^ bkt_key[0]; \ + xor[1] = (pkt_key[1] & key_mask[1]) ^ bkt_key[1]; \ + xor[2] = (pkt_key[2] & key_mask[2]) ^ bkt_key[2]; \ + xor[3] = (pkt_key[3] & key_mask[3]) ^ bkt_key[3]; \ + xor[4] = (pkt_key[4] & key_mask[4]) ^ bkt_key[4]; \ + xor[5] = (pkt_key[5] & key_mask[5]) ^ bkt_key[5]; \ + xor[6] = (pkt_key[6] & key_mask[6]) ^ bkt_key[6]; \ + xor[7] = (pkt_key[7] & key_mask[7]) ^ bkt_key[7]; \ or = xor[0] | xor[1] | xor[2] | xor[3] | \ xor[4] | xor[5] | xor[6] | xor[7]; \ match_key = 0; \ @@ -637,7 +691,7 @@ static int rte_table_hash_ext_lookup_unoptimized( \ default: \ match_key = 0; \ - if (memcmp(pkt_key, bkt_key, f->key_size) == 0) \ + if (keycmp(bkt_key, pkt_key, key_mask, f->key_size) == 0) \ match_key = 1; \ } \ } @@ -685,38 +739,7 @@ static int rte_table_hash_ext_lookup_unoptimized( rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));\ } -#define lookup2_stage1(t, g, pkts, pkt10_index, pkt11_index) \ -{ \ - struct grinder *g10, *g11; \ - uint64_t sig10, sig11, bkt10_index, bkt11_index; \ - struct rte_mbuf *mbuf10, *mbuf11; \ - struct bucket *bkt10, *bkt11, *buckets = t->buckets; \ - uint64_t bucket_mask = t->bucket_mask; \ - uint32_t signature_offset = t->signature_offset; \ - \ - mbuf10 = pkts[pkt10_index]; \ - sig10 = (uint64_t) RTE_MBUF_METADATA_UINT32(mbuf10, signature_offset);\ - bkt10_index = sig10 & bucket_mask; \ - bkt10 = &buckets[bkt10_index]; \ - \ - mbuf11 = pkts[pkt11_index]; \ - sig11 = (uint64_t) RTE_MBUF_METADATA_UINT32(mbuf11, signature_offset);\ - bkt11_index = sig11 & bucket_mask; \ - bkt11 = &buckets[bkt11_index]; \ - \ - rte_prefetch0(bkt10); \ - rte_prefetch0(bkt11); \ - \ - g10 = &g[pkt10_index]; \ - g10->sig = sig10; \ - g10->bkt = bkt10; \ - \ - g11 = &g[pkt11_index]; \ - g11->sig = sig11; \ - g11->bkt = bkt11; \ -} - -#define lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index) \ +#define lookup2_stage1(t, g, pkts, pkt10_index, pkt11_index) \ { \ struct grinder *g10, *g11; \ uint64_t sig10, sig11, bkt10_index, bkt11_index; \ @@ -731,13 +754,13 @@ static int rte_table_hash_ext_lookup_unoptimized( \ mbuf10 = pkts[pkt10_index]; \ key10 = RTE_MBUF_METADATA_UINT8_PTR(mbuf10, key_offset); \ - sig10 = (uint64_t) f_hash(key10, key_size, seed); \ + sig10 = (uint64_t) f_hash(key10, t->key_mask, key_size, seed); \ bkt10_index = sig10 & bucket_mask; \ bkt10 = &buckets[bkt10_index]; \ \ mbuf11 = pkts[pkt11_index]; \ key11 = RTE_MBUF_METADATA_UINT8_PTR(mbuf11, key_offset); \ - sig11 = (uint64_t) f_hash(key11, key_size, seed); \ + sig11 = (uint64_t) f_hash(key11, t->key_mask, key_size, seed); \ bkt11_index = sig11 & bucket_mask; \ bkt11 = &buckets[bkt11_index]; \ \ @@ -874,7 +897,7 @@ static int rte_table_hash_ext_lookup( /* Cannot run the pipeline with less than 7 packets */ if (__builtin_popcountll(pkts_mask) < 7) { status = rte_table_hash_ext_lookup_unoptimized(table, pkts, - pkts_mask, lookup_hit_mask, entries, 0); + pkts_mask, lookup_hit_mask, entries); RTE_TABLE_HASH_EXT_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(*lookup_hit_mask)); return status; @@ -982,144 +1005,7 @@ static int rte_table_hash_ext_lookup( uint64_t pkts_mask_out_slow = 0; status = rte_table_hash_ext_lookup_unoptimized(table, pkts, - pkts_mask_match_many, &pkts_mask_out_slow, entries, 0); - pkts_mask_out |= pkts_mask_out_slow; - } - - *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_EXT_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(pkts_mask_out)); - return status; -} - -static int rte_table_hash_ext_lookup_dosig( - void *table, - struct rte_mbuf **pkts, - uint64_t pkts_mask, - uint64_t *lookup_hit_mask, - void **entries) -{ - struct rte_table_hash *t = (struct rte_table_hash *) table; - struct grinder *g = t->grinders; - uint64_t pkt00_index, pkt01_index, pkt10_index, pkt11_index; - uint64_t pkt20_index, pkt21_index, pkt30_index, pkt31_index; - uint64_t pkts_mask_out = 0, pkts_mask_match_many = 0; - int status = 0; - - __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); - RTE_TABLE_HASH_EXT_STATS_PKTS_IN_ADD(t, n_pkts_in); - - /* Cannot run the pipeline with less than 7 packets */ - if (__builtin_popcountll(pkts_mask) < 7) { - status = rte_table_hash_ext_lookup_unoptimized(table, pkts, - pkts_mask, lookup_hit_mask, entries, 1); - RTE_TABLE_HASH_EXT_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - - __builtin_popcountll(*lookup_hit_mask)); - return status; - } - - /* Pipeline stage 0 */ - lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); - - /* Pipeline feed */ - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); - - /* Pipeline feed */ - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); - - /* Pipeline stage 2 */ - lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); - - /* - * Pipeline run - * - */ - for ( ; pkts_mask; ) { - /* Pipeline feed */ - pkt30_index = pkt20_index; - pkt31_index = pkt21_index; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0_with_odd_support(t, g, pkts, pkts_mask, - pkt00_index, pkt01_index); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); - - /* Pipeline stage 2 */ - lookup2_stage2(t, g, pkt20_index, pkt21_index, - pkts_mask_match_many); - - /* Pipeline stage 3 */ - lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, - pkts_mask_out, entries); - } - - /* Pipeline feed */ - pkt30_index = pkt20_index; - pkt31_index = pkt21_index; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); - - /* Pipeline stage 2 */ - lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); - - /* Pipeline stage 3 */ - lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, - entries); - - /* Pipeline feed */ - pkt30_index = pkt20_index; - pkt31_index = pkt21_index; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - - /* Pipeline stage 2 */ - lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); - - /* Pipeline stage 3 */ - lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, - entries); - - /* Pipeline feed */ - pkt30_index = pkt20_index; - pkt31_index = pkt21_index; - - /* Pipeline stage 3 */ - lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, - entries); - - /* Slow path */ - pkts_mask_match_many &= ~pkts_mask_out; - if (pkts_mask_match_many) { - uint64_t pkts_mask_out_slow = 0; - - status = rte_table_hash_ext_lookup_unoptimized(table, pkts, - pkts_mask_match_many, &pkts_mask_out_slow, entries, 1); + pkts_mask_match_many, &pkts_mask_out_slow, entries); pkts_mask_out |= pkts_mask_out_slow; } @@ -1152,14 +1038,3 @@ struct rte_table_ops rte_table_hash_ext_ops = { .f_lookup = rte_table_hash_ext_lookup, .f_stats = rte_table_hash_ext_stats_read, }; - -struct rte_table_ops rte_table_hash_ext_dosig_ops = { - .f_create = rte_table_hash_ext_create, - .f_free = rte_table_hash_ext_free, - .f_add = rte_table_hash_ext_entry_add, - .f_delete = rte_table_hash_ext_entry_delete, - .f_add_bulk = NULL, - .f_delete_bulk = NULL, - .f_lookup = rte_table_hash_ext_lookup_dosig, - .f_stats = rte_table_hash_ext_stats_read, -}; diff --git a/lib/librte_table/rte_table_hash_key16.c b/lib/librte_table/rte_table_hash_key16.c index ce057b78..b541735c 100644 --- a/lib/librte_table/rte_table_hash_key16.c +++ b/lib/librte_table/rte_table_hash_key16.c @@ -1,34 +1,34 @@ /*- - * BSD LICENSE + * BSD LICENSE * - * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. - * All rights reserved. + * Copyright(c) 2010-2017 Intel Corporation. All rights reserved. + * All rights reserved. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Intel Corporation nor the names of its - * contributors may be used to endorse or promote products derived - * from this software without specific prior written permission. + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include <string.h> #include <stdio.h> @@ -42,7 +42,9 @@ #include "rte_table_hash.h" #include "rte_lru.h" -#define RTE_TABLE_HASH_KEY_SIZE 16 +#define KEY_SIZE 16 + +#define KEYS_PER_BUCKET 4 #define RTE_BUCKET_ENTRY_VALID 0x1LLU @@ -79,11 +81,9 @@ struct rte_table_hash { /* Input parameters */ uint32_t n_buckets; - uint32_t n_entries_per_bucket; uint32_t key_size; uint32_t entry_size; uint32_t bucket_size; - uint32_t signature_offset; uint32_t key_offset; uint64_t key_mask[2]; rte_table_hash_op_hash f_hash; @@ -99,17 +99,55 @@ struct rte_table_hash { }; static int -check_params_create_lru(struct rte_table_hash_key16_lru_params *params) { - /* n_entries */ - if (params->n_entries == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries is zero\n", __func__); +keycmp(void *a, void *b, void *b_mask) +{ + uint64_t *a64 = a, *b64 = b, *b_mask64 = b_mask; + + return (a64[0] != (b64[0] & b_mask64[0])) || + (a64[1] != (b64[1] & b_mask64[1])); +} + +static void +keycpy(void *dst, void *src, void *src_mask) +{ + uint64_t *dst64 = dst, *src64 = src, *src_mask64 = src_mask; + + dst64[0] = src64[0] & src_mask64[0]; + dst64[1] = src64[1] & src_mask64[1]; +} + +static int +check_params_create(struct rte_table_hash_params *params) +{ + /* name */ + if (params->name == NULL) { + RTE_LOG(ERR, TABLE, "%s: name invalid value\n", __func__); + return -EINVAL; + } + + /* key_size */ + if (params->key_size != KEY_SIZE) { + RTE_LOG(ERR, TABLE, "%s: key_size invalid value\n", __func__); + return -EINVAL; + } + + /* n_keys */ + if (params->n_keys == 0) { + RTE_LOG(ERR, TABLE, "%s: n_keys is zero\n", __func__); + return -EINVAL; + } + + /* n_buckets */ + if ((params->n_buckets == 0) || + (!rte_is_power_of_2(params->n_buckets))) { + RTE_LOG(ERR, TABLE, "%s: n_buckets invalid value\n", __func__); return -EINVAL; } /* f_hash */ if (params->f_hash == NULL) { - RTE_LOG(ERR, TABLE, - "%s: f_hash function pointer is NULL\n", __func__); + RTE_LOG(ERR, TABLE, "%s: f_hash function pointer is NULL\n", + __func__); return -EINVAL; } @@ -121,46 +159,67 @@ rte_table_hash_create_key16_lru(void *params, int socket_id, uint32_t entry_size) { - struct rte_table_hash_key16_lru_params *p = - (struct rte_table_hash_key16_lru_params *) params; + struct rte_table_hash_params *p = params; struct rte_table_hash *f; - uint32_t n_buckets, n_entries_per_bucket, - key_size, bucket_size_cl, total_size, i; + uint64_t bucket_size, total_size; + uint32_t n_buckets, i; /* Check input parameters */ - if ((check_params_create_lru(p) != 0) || + if ((check_params_create(p) != 0) || ((sizeof(struct rte_table_hash) % RTE_CACHE_LINE_SIZE) != 0) || ((sizeof(struct rte_bucket_4_16) % 64) != 0)) return NULL; - n_entries_per_bucket = 4; - key_size = 16; + + /* + * Table dimensioning + * + * Objective: Pick the number of buckets (n_buckets) so that there a chance + * to store n_keys keys in the table. + * + * Note: Since the buckets do not get extended, it is not possible to + * guarantee that n_keys keys can be stored in the table at any time. In the + * worst case scenario when all the n_keys fall into the same bucket, only + * a maximum of KEYS_PER_BUCKET keys will be stored in the table. This case + * defeats the purpose of the hash table. It indicates unsuitable f_hash or + * n_keys to n_buckets ratio. + * + * MIN(n_buckets) = (n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET + */ + n_buckets = rte_align32pow2( + (p->n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET); + n_buckets = RTE_MAX(n_buckets, p->n_buckets); /* Memory allocation */ - n_buckets = rte_align32pow2((p->n_entries + n_entries_per_bucket - 1) / - n_entries_per_bucket); - bucket_size_cl = (sizeof(struct rte_bucket_4_16) + n_entries_per_bucket - * entry_size + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; - total_size = sizeof(struct rte_table_hash) + n_buckets * - bucket_size_cl * RTE_CACHE_LINE_SIZE; - - f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + bucket_size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_bucket_4_16) + + KEYS_PER_BUCKET * entry_size); + total_size = sizeof(struct rte_table_hash) + n_buckets * bucket_size; + + if (total_size > SIZE_MAX) { + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); + return NULL; + } + + f = rte_zmalloc_socket(p->name, + (size_t)total_size, + RTE_CACHE_LINE_SIZE, + socket_id); if (f == NULL) { - RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for hash table\n", - __func__, total_size); + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); return NULL; } - RTE_LOG(INFO, TABLE, - "%s: Hash table memory footprint is %u bytes\n", - __func__, total_size); + RTE_LOG(INFO, TABLE, "%s: Hash table %s memory footprint " + "is %" PRIu64 " bytes\n", + __func__, p->name, total_size); /* Memory initialization */ f->n_buckets = n_buckets; - f->n_entries_per_bucket = n_entries_per_bucket; - f->key_size = key_size; + f->key_size = KEY_SIZE; f->entry_size = entry_size; - f->bucket_size = bucket_size_cl * RTE_CACHE_LINE_SIZE; - f->signature_offset = p->signature_offset; + f->bucket_size = bucket_size; f->key_offset = p->key_offset; f->f_hash = p->f_hash; f->seed = p->seed; @@ -212,19 +271,19 @@ rte_table_hash_entry_add_key16_lru( uint64_t signature, pos; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket = (struct rte_bucket_4_16 *) - &f->memory[bucket_index * f->bucket_size]; + &f->memory[bucket_index * f->bucket_size]; signature |= RTE_BUCKET_ENTRY_VALID; /* Key is present in the bucket */ for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if ((bucket_signature == signature) && - (memcmp(key, bucket_key, f->key_size) == 0)) { + (keycmp(bucket_key, key, f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; memcpy(bucket_data, entry, f->entry_size); @@ -238,13 +297,13 @@ rte_table_hash_entry_add_key16_lru( /* Key is not present in the bucket */ for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if (bucket_signature == 0) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature[i] = signature; - memcpy(bucket_key, key, f->key_size); + keycpy(bucket_key, key, f->key_mask); memcpy(bucket_data, entry, f->entry_size); lru_update(bucket, i); *key_found = 0; @@ -257,7 +316,7 @@ rte_table_hash_entry_add_key16_lru( /* Bucket full: replace LRU entry */ pos = lru_pos(bucket); bucket->signature[pos] = signature; - memcpy(bucket->key[pos], key, f->key_size); + keycpy(&bucket->key[pos], key, f->key_mask); memcpy(&bucket->data[pos * f->entry_size], entry, f->entry_size); lru_update(bucket, pos); *key_found = 0; @@ -278,19 +337,19 @@ rte_table_hash_entry_delete_key16_lru( uint64_t signature; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket = (struct rte_bucket_4_16 *) - &f->memory[bucket_index * f->bucket_size]; + &f->memory[bucket_index * f->bucket_size]; signature |= RTE_BUCKET_ENTRY_VALID; /* Key is present in the bucket */ for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if ((bucket_signature == signature) && - (memcmp(key, bucket_key, f->key_size) == 0)) { + (keycmp(bucket_key, key, f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature[i] = 0; @@ -306,81 +365,71 @@ rte_table_hash_entry_delete_key16_lru( return 0; } -static int -check_params_create_ext(struct rte_table_hash_key16_ext_params *params) { - /* n_entries */ - if (params->n_entries == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries is zero\n", __func__); - return -EINVAL; - } - - /* n_entries_ext */ - if (params->n_entries_ext == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries_ext is zero\n", __func__); - return -EINVAL; - } - - /* f_hash */ - if (params->f_hash == NULL) { - RTE_LOG(ERR, TABLE, - "%s: f_hash function pointer is NULL\n", __func__); - return -EINVAL; - } - - return 0; -} - static void * rte_table_hash_create_key16_ext(void *params, int socket_id, uint32_t entry_size) { - struct rte_table_hash_key16_ext_params *p = - (struct rte_table_hash_key16_ext_params *) params; + struct rte_table_hash_params *p = params; struct rte_table_hash *f; - uint32_t n_buckets, n_buckets_ext, n_entries_per_bucket, key_size, - bucket_size_cl, stack_size_cl, total_size, i; + uint64_t bucket_size, stack_size, total_size; + uint32_t n_buckets_ext, i; /* Check input parameters */ - if ((check_params_create_ext(p) != 0) || + if ((check_params_create(p) != 0) || ((sizeof(struct rte_table_hash) % RTE_CACHE_LINE_SIZE) != 0) || ((sizeof(struct rte_bucket_4_16) % 64) != 0)) return NULL; - n_entries_per_bucket = 4; - key_size = 16; + /* + * Table dimensioning + * + * Objective: Pick the number of bucket extensions (n_buckets_ext) so that + * it is guaranteed that n_keys keys can be stored in the table at any time. + * + * The worst case scenario takes place when all the n_keys keys fall into + * the same bucket. Actually, due to the KEYS_PER_BUCKET scheme, the worst + * case takes place when (n_keys - KEYS_PER_BUCKET + 1) keys fall into the + * same bucket, while the remaining (KEYS_PER_BUCKET - 1) keys each fall + * into a different bucket. This case defeats the purpose of the hash table. + * It indicates unsuitable f_hash or n_keys to n_buckets ratio. + * + * n_buckets_ext = n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1 + */ + n_buckets_ext = p->n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1; /* Memory allocation */ - n_buckets = rte_align32pow2((p->n_entries + n_entries_per_bucket - 1) / - n_entries_per_bucket); - n_buckets_ext = (p->n_entries_ext + n_entries_per_bucket - 1) / - n_entries_per_bucket; - bucket_size_cl = (sizeof(struct rte_bucket_4_16) + n_entries_per_bucket - * entry_size + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; - stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + RTE_CACHE_LINE_SIZE - 1) - / RTE_CACHE_LINE_SIZE; + bucket_size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_bucket_4_16) + + KEYS_PER_BUCKET * entry_size); + stack_size = RTE_CACHE_LINE_ROUNDUP(n_buckets_ext * sizeof(uint32_t)); total_size = sizeof(struct rte_table_hash) + - ((n_buckets + n_buckets_ext) * bucket_size_cl + stack_size_cl) * - RTE_CACHE_LINE_SIZE; + (p->n_buckets + n_buckets_ext) * bucket_size + stack_size; + if (total_size > SIZE_MAX) { + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); + return NULL; + } - f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + f = rte_zmalloc_socket(p->name, + (size_t)total_size, + RTE_CACHE_LINE_SIZE, + socket_id); if (f == NULL) { - RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for hash table\n", - __func__, total_size); + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); return NULL; } - RTE_LOG(INFO, TABLE, - "%s: Hash table memory footprint is %u bytes\n", - __func__, total_size); + RTE_LOG(INFO, TABLE, "%s: Hash table %s memory footprint " + "is %" PRIu64 " bytes\n", + __func__, p->name, total_size); /* Memory initialization */ - f->n_buckets = n_buckets; - f->n_entries_per_bucket = n_entries_per_bucket; - f->key_size = key_size; + f->n_buckets = p->n_buckets; + f->key_size = KEY_SIZE; f->entry_size = entry_size; - f->bucket_size = bucket_size_cl * RTE_CACHE_LINE_SIZE; - f->signature_offset = p->signature_offset; + f->bucket_size = bucket_size; f->key_offset = p->key_offset; f->f_hash = p->f_hash; f->seed = p->seed; @@ -388,10 +437,7 @@ rte_table_hash_create_key16_ext(void *params, f->n_buckets_ext = n_buckets_ext; f->stack_pos = n_buckets_ext; f->stack = (uint32_t *) - &f->memory[(n_buckets + n_buckets_ext) * f->bucket_size]; - - for (i = 0; i < n_buckets_ext; i++) - f->stack[i] = i; + &f->memory[(p->n_buckets + n_buckets_ext) * f->bucket_size]; if (p->key_mask != NULL) { f->key_mask[0] = (((uint64_t *)p->key_mask)[0]); @@ -401,6 +447,9 @@ rte_table_hash_create_key16_ext(void *params, f->key_mask[1] = 0xFFFFFFFFFFFFFFFFLLU; } + for (i = 0; i < n_buckets_ext; i++) + f->stack[i] = i; + return f; } @@ -432,20 +481,20 @@ rte_table_hash_entry_add_key16_ext( uint64_t signature; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket0 = (struct rte_bucket_4_16 *) - &f->memory[bucket_index * f->bucket_size]; + &f->memory[bucket_index * f->bucket_size]; signature |= RTE_BUCKET_ENTRY_VALID; /* Key is present in the bucket */ for (bucket = bucket0; bucket != NULL; bucket = bucket->next) for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if ((bucket_signature == signature) && - (memcmp(key, bucket_key, f->key_size) == 0)) { + (keycmp(bucket_key, key, f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; @@ -458,17 +507,17 @@ rte_table_hash_entry_add_key16_ext( /* Key is not present in the bucket */ for (bucket_prev = NULL, bucket = bucket0; bucket != NULL; - bucket_prev = bucket, bucket = bucket->next) + bucket_prev = bucket, bucket = bucket->next) for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if (bucket_signature == 0) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature[i] = signature; - memcpy(bucket_key, key, f->key_size); + keycpy(bucket_key, key, f->key_mask); memcpy(bucket_data, entry, f->entry_size); *key_found = 0; *entry_ptr = (void *) bucket_data; @@ -487,7 +536,7 @@ rte_table_hash_entry_add_key16_ext( bucket_prev->next_valid = 1; bucket->signature[0] = signature; - memcpy(bucket->key[0], key, f->key_size); + keycpy(&bucket->key[0], key, f->key_mask); memcpy(&bucket->data[0], entry, f->entry_size); *key_found = 0; *entry_ptr = (void *) &bucket->data[0]; @@ -509,7 +558,7 @@ rte_table_hash_entry_delete_key16_ext( uint64_t signature; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket0 = (struct rte_bucket_4_16 *) &f->memory[bucket_index * f->bucket_size]; @@ -520,18 +569,17 @@ rte_table_hash_entry_delete_key16_ext( bucket_prev = bucket, bucket = bucket->next) for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if ((bucket_signature == signature) && - (memcmp(key, bucket_key, f->key_size) == 0)) { + (keycmp(bucket_key, key, f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature[i] = 0; *key_found = 1; if (entry) - memcpy(entry, bucket_data, - f->entry_size); + memcpy(entry, bucket_data, f->entry_size); if ((bucket->signature[0] == 0) && (bucket->signature[1] == 0) && @@ -558,26 +606,28 @@ rte_table_hash_entry_delete_key16_ext( return 0; } -#define lookup_key16_cmp(key_in, bucket, pos) \ +#define lookup_key16_cmp(key_in, bucket, pos, f) \ { \ - uint64_t xor[4][2], or[4], signature[4]; \ + uint64_t xor[4][2], or[4], signature[4], k[2]; \ \ + k[0] = key_in[0] & f->key_mask[0]; \ + k[1] = key_in[1] & f->key_mask[1]; \ signature[0] = (~bucket->signature[0]) & 1; \ signature[1] = (~bucket->signature[1]) & 1; \ signature[2] = (~bucket->signature[2]) & 1; \ signature[3] = (~bucket->signature[3]) & 1; \ \ - xor[0][0] = key_in[0] ^ bucket->key[0][0]; \ - xor[0][1] = key_in[1] ^ bucket->key[0][1]; \ + xor[0][0] = k[0] ^ bucket->key[0][0]; \ + xor[0][1] = k[1] ^ bucket->key[0][1]; \ \ - xor[1][0] = key_in[0] ^ bucket->key[1][0]; \ - xor[1][1] = key_in[1] ^ bucket->key[1][1]; \ + xor[1][0] = k[0] ^ bucket->key[1][0]; \ + xor[1][1] = k[1] ^ bucket->key[1][1]; \ \ - xor[2][0] = key_in[0] ^ bucket->key[2][0]; \ - xor[2][1] = key_in[1] ^ bucket->key[2][1]; \ + xor[2][0] = k[0] ^ bucket->key[2][0]; \ + xor[2][1] = k[1] ^ bucket->key[2][1]; \ \ - xor[3][0] = key_in[0] ^ bucket->key[3][0]; \ - xor[3][1] = key_in[1] ^ bucket->key[3][1]; \ + xor[3][0] = k[0] ^ bucket->key[3][0]; \ + xor[3][1] = k[1] ^ bucket->key[3][1]; \ \ or[0] = xor[0][0] | xor[0][1] | signature[0]; \ or[1] = xor[1][0] | xor[1][1] | signature[1]; \ @@ -610,30 +660,12 @@ rte_table_hash_entry_delete_key16_ext( #define lookup1_stage1(mbuf1, bucket1, f) \ { \ - uint64_t signature; \ - uint32_t bucket_index; \ - \ - signature = RTE_MBUF_METADATA_UINT32(mbuf1, f->signature_offset);\ - bucket_index = signature & (f->n_buckets - 1); \ - bucket1 = (struct rte_bucket_4_16 *) \ - &f->memory[bucket_index * f->bucket_size]; \ - rte_prefetch0(bucket1); \ - rte_prefetch0((void *)(((uintptr_t) bucket1) + RTE_CACHE_LINE_SIZE));\ -} - -#define lookup1_stage1_dosig(mbuf1, bucket1, f) \ -{ \ uint64_t *key; \ uint64_t signature = 0; \ uint32_t bucket_index; \ - uint64_t hash_key_buffer[2]; \ \ key = RTE_MBUF_METADATA_UINT64_PTR(mbuf1, f->key_offset);\ - \ - hash_key_buffer[0] = key[0] & f->key_mask[0]; \ - hash_key_buffer[1] = key[1] & f->key_mask[1]; \ - signature = f->f_hash(hash_key_buffer, \ - RTE_TABLE_HASH_KEY_SIZE, f->seed); \ + signature = f->f_hash(key, f->key_mask, KEY_SIZE, f->seed); \ \ bucket_index = signature & (f->n_buckets - 1); \ bucket1 = (struct rte_bucket_4_16 *) \ @@ -648,14 +680,10 @@ rte_table_hash_entry_delete_key16_ext( void *a; \ uint64_t pkt_mask; \ uint64_t *key; \ - uint64_t hash_key_buffer[2]; \ uint32_t pos; \ \ key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ - hash_key_buffer[0] = key[0] & f->key_mask[0]; \ - hash_key_buffer[1] = key[1] & f->key_mask[1]; \ - \ - lookup_key16_cmp(hash_key_buffer, bucket2, pos); \ + lookup_key16_cmp(key, bucket2, pos, f); \ \ pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\ pkts_mask_out |= pkt_mask; \ @@ -673,14 +701,10 @@ rte_table_hash_entry_delete_key16_ext( void *a; \ uint64_t pkt_mask, bucket_mask; \ uint64_t *key; \ - uint64_t hash_key_buffer[2]; \ uint32_t pos; \ \ key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ - hash_key_buffer[0] = key[0] & f->key_mask[0]; \ - hash_key_buffer[1] = key[1] & f->key_mask[1]; \ - \ - lookup_key16_cmp(hash_key_buffer, bucket2, pos); \ + lookup_key16_cmp(key, bucket2, pos, f); \ \ pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\ pkts_mask_out |= pkt_mask; \ @@ -703,15 +727,11 @@ rte_table_hash_entry_delete_key16_ext( void *a; \ uint64_t pkt_mask, bucket_mask; \ uint64_t *key; \ - uint64_t hash_key_buffer[2]; \ uint32_t pos; \ \ bucket = buckets[pkt_index]; \ key = keys[pkt_index]; \ - hash_key_buffer[0] = key[0] & f->key_mask[0]; \ - hash_key_buffer[1] = key[1] & f->key_mask[1]; \ - \ - lookup_key16_cmp(hash_key_buffer, bucket, pos); \ + lookup_key16_cmp(key, bucket, pos, f); \ \ pkt_mask = (bucket->signature[pos] & 1LLU) << pkt_index;\ pkts_mask_out |= pkt_mask; \ @@ -775,36 +795,12 @@ rte_table_hash_entry_delete_key16_ext( #define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f) \ { \ - uint64_t signature10, signature11; \ - uint32_t bucket10_index, bucket11_index; \ - \ - signature10 = RTE_MBUF_METADATA_UINT32(mbuf10, f->signature_offset);\ - bucket10_index = signature10 & (f->n_buckets - 1); \ - bucket10 = (struct rte_bucket_4_16 *) \ - &f->memory[bucket10_index * f->bucket_size]; \ - rte_prefetch0(bucket10); \ - rte_prefetch0((void *)(((uintptr_t) bucket10) + RTE_CACHE_LINE_SIZE));\ - \ - signature11 = RTE_MBUF_METADATA_UINT32(mbuf11, f->signature_offset);\ - bucket11_index = signature11 & (f->n_buckets - 1); \ - bucket11 = (struct rte_bucket_4_16 *) \ - &f->memory[bucket11_index * f->bucket_size]; \ - rte_prefetch0(bucket11); \ - rte_prefetch0((void *)(((uintptr_t) bucket11) + RTE_CACHE_LINE_SIZE));\ -} - -#define lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f) \ -{ \ uint64_t *key10, *key11; \ - uint64_t hash_offset_buffer[2]; \ uint64_t signature10, signature11; \ uint32_t bucket10_index, bucket11_index; \ \ key10 = RTE_MBUF_METADATA_UINT64_PTR(mbuf10, f->key_offset);\ - hash_offset_buffer[0] = key10[0] & f->key_mask[0]; \ - hash_offset_buffer[1] = key10[1] & f->key_mask[1]; \ - signature10 = f->f_hash(hash_offset_buffer, \ - RTE_TABLE_HASH_KEY_SIZE, f->seed);\ + signature10 = f->f_hash(key10, f->key_mask, KEY_SIZE, f->seed);\ bucket10_index = signature10 & (f->n_buckets - 1); \ bucket10 = (struct rte_bucket_4_16 *) \ &f->memory[bucket10_index * f->bucket_size]; \ @@ -812,10 +808,7 @@ rte_table_hash_entry_delete_key16_ext( rte_prefetch0((void *)(((uintptr_t) bucket10) + RTE_CACHE_LINE_SIZE));\ \ key11 = RTE_MBUF_METADATA_UINT64_PTR(mbuf11, f->key_offset);\ - hash_offset_buffer[0] = key11[0] & f->key_mask[0]; \ - hash_offset_buffer[1] = key11[1] & f->key_mask[1]; \ - signature11 = f->f_hash(hash_offset_buffer, \ - RTE_TABLE_HASH_KEY_SIZE, f->seed);\ + signature11 = f->f_hash(key11, f->key_mask, KEY_SIZE, f->seed);\ bucket11_index = signature11 & (f->n_buckets - 1); \ bucket11 = (struct rte_bucket_4_16 *) \ &f->memory[bucket11_index * f->bucket_size]; \ @@ -829,19 +822,13 @@ rte_table_hash_entry_delete_key16_ext( void *a20, *a21; \ uint64_t pkt20_mask, pkt21_mask; \ uint64_t *key20, *key21; \ - uint64_t hash_key_buffer20[2]; \ - uint64_t hash_key_buffer21[2]; \ uint32_t pos20, pos21; \ \ key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ - hash_key_buffer20[0] = key20[0] & f->key_mask[0]; \ - hash_key_buffer20[1] = key20[1] & f->key_mask[1]; \ - hash_key_buffer21[0] = key21[0] & f->key_mask[0]; \ - hash_key_buffer21[1] = key21[1] & f->key_mask[1]; \ \ - lookup_key16_cmp(hash_key_buffer20, bucket20, pos20); \ - lookup_key16_cmp(hash_key_buffer21, bucket21, pos21); \ + lookup_key16_cmp(key20, bucket20, pos20, f); \ + lookup_key16_cmp(key21, bucket21, pos21, f); \ \ pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\ pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\ @@ -864,19 +851,13 @@ rte_table_hash_entry_delete_key16_ext( void *a20, *a21; \ uint64_t pkt20_mask, pkt21_mask, bucket20_mask, bucket21_mask;\ uint64_t *key20, *key21; \ - uint64_t hash_key_buffer20[2]; \ - uint64_t hash_key_buffer21[2]; \ uint32_t pos20, pos21; \ \ key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ - hash_key_buffer20[0] = key20[0] & f->key_mask[0]; \ - hash_key_buffer20[1] = key20[1] & f->key_mask[1]; \ - hash_key_buffer21[0] = key21[0] & f->key_mask[0]; \ - hash_key_buffer21[1] = key21[1] & f->key_mask[1]; \ \ - lookup_key16_cmp(hash_key_buffer20, bucket20, pos20); \ - lookup_key16_cmp(hash_key_buffer21, bucket21, pos21); \ + lookup_key16_cmp(key20, bucket20, pos20, f); \ + lookup_key16_cmp(key21, bucket21, pos21, f); \ \ pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\ pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\ @@ -916,6 +897,7 @@ rte_table_hash_lookup_key16_lru( uint64_t pkts_mask_out = 0; __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(f, n_pkts_in); /* Cannot run the pipeline with less than 5 packets */ @@ -932,8 +914,8 @@ rte_table_hash_lookup_key16_lru( } *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, - n_pkts_in - __builtin_popcountll(pkts_mask_out)); + RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - + __builtin_popcountll(pkts_mask_out)); return 0; } @@ -1026,136 +1008,7 @@ rte_table_hash_lookup_key16_lru( RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); return 0; -} /* rte_table_hash_lookup_key16_lru() */ - -static int -rte_table_hash_lookup_key16_lru_dosig( - void *table, - struct rte_mbuf **pkts, - uint64_t pkts_mask, - uint64_t *lookup_hit_mask, - void **entries) -{ - struct rte_table_hash *f = (struct rte_table_hash *) table; - struct rte_bucket_4_16 *bucket10, *bucket11, *bucket20, *bucket21; - struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; - uint32_t pkt00_index, pkt01_index, pkt10_index; - uint32_t pkt11_index, pkt20_index, pkt21_index; - uint64_t pkts_mask_out = 0; - - __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); - - RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(f, n_pkts_in); - - /* Cannot run the pipeline with less than 5 packets */ - if (__builtin_popcountll(pkts_mask) < 5) { - for ( ; pkts_mask; ) { - struct rte_bucket_4_16 *bucket; - struct rte_mbuf *mbuf; - uint32_t pkt_index; - - lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); - lookup1_stage1_dosig(mbuf, bucket, f); - lookup1_stage2_lru(pkt_index, mbuf, bucket, - pkts_mask_out, entries, f); - } - - *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - - __builtin_popcountll(pkts_mask_out)); - return 0; - } - - /* - * Pipeline fill - * - */ - /* Pipeline stage 0 */ - lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, - pkts_mask, f); - - /* Pipeline feed */ - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, - pkts_mask, f); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* - * Pipeline run - * - */ - for ( ; pkts_mask; ) { - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, - mbuf00, mbuf01, pkts, pkts_mask, f); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* Pipeline stage 2 */ - lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, f); - } - - /* - * Pipeline flush - * - */ - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* Pipeline stage 2 */ - lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, f); - - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - - /* Pipeline stage 2 */ - lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, f); - - *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - - __builtin_popcountll(pkts_mask_out)); - return 0; -} /* rte_table_hash_lookup_key16_lru_dosig() */ +} /* lookup LRU */ static int rte_table_hash_lookup_key16_ext( @@ -1175,6 +1028,7 @@ rte_table_hash_lookup_key16_ext( uint64_t *keys[RTE_PORT_IN_BURST_SIZE_MAX]; __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(f, n_pkts_in); /* Cannot run the pipeline with less than 5 packets */ @@ -1306,159 +1160,7 @@ grind_next_buckets: RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); return 0; -} /* rte_table_hash_lookup_key16_ext() */ - -static int -rte_table_hash_lookup_key16_ext_dosig( - void *table, - struct rte_mbuf **pkts, - uint64_t pkts_mask, - uint64_t *lookup_hit_mask, - void **entries) -{ - struct rte_table_hash *f = (struct rte_table_hash *) table; - struct rte_bucket_4_16 *bucket10, *bucket11, *bucket20, *bucket21; - struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; - uint32_t pkt00_index, pkt01_index, pkt10_index; - uint32_t pkt11_index, pkt20_index, pkt21_index; - uint64_t pkts_mask_out = 0, buckets_mask = 0; - struct rte_bucket_4_16 *buckets[RTE_PORT_IN_BURST_SIZE_MAX]; - uint64_t *keys[RTE_PORT_IN_BURST_SIZE_MAX]; - - __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); - - RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(f, n_pkts_in); - - /* Cannot run the pipeline with less than 5 packets */ - if (__builtin_popcountll(pkts_mask) < 5) { - for ( ; pkts_mask; ) { - struct rte_bucket_4_16 *bucket; - struct rte_mbuf *mbuf; - uint32_t pkt_index; - - lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); - lookup1_stage1_dosig(mbuf, bucket, f); - lookup1_stage2_ext(pkt_index, mbuf, bucket, - pkts_mask_out, entries, buckets_mask, - buckets, keys, f); - } - - goto grind_next_buckets; - } - - /* - * Pipeline fill - * - */ - /* Pipeline stage 0 */ - lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, - pkts_mask, f); - - /* Pipeline feed */ - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, - pkts_mask, f); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* - * Pipeline run - * - */ - for ( ; pkts_mask; ) { - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, - mbuf00, mbuf01, pkts, pkts_mask, f); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* Pipeline stage 2 */ - lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, - buckets_mask, buckets, keys, f); - } - - /* - * Pipeline flush - * - */ - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* Pipeline stage 2 */ - lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, - buckets_mask, buckets, keys, f); - - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - - /* Pipeline stage 2 */ - lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, - buckets_mask, buckets, keys, f); - -grind_next_buckets: - /* Grind next buckets */ - for ( ; buckets_mask; ) { - uint64_t buckets_mask_next = 0; - - for ( ; buckets_mask; ) { - uint64_t pkt_mask; - uint32_t pkt_index; - - pkt_index = __builtin_ctzll(buckets_mask); - pkt_mask = 1LLU << pkt_index; - buckets_mask &= ~pkt_mask; - - lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, - entries, buckets_mask_next, f); - } - - buckets_mask = buckets_mask_next; - } - - *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - - __builtin_popcountll(pkts_mask_out)); - return 0; -} /* rte_table_hash_lookup_key16_ext_dosig() */ +} /* lookup EXT */ static int rte_table_hash_key16_stats_read(void *table, struct rte_table_stats *stats, int clear) @@ -1485,15 +1187,6 @@ struct rte_table_ops rte_table_hash_key16_lru_ops = { .f_stats = rte_table_hash_key16_stats_read, }; -struct rte_table_ops rte_table_hash_key16_lru_dosig_ops = { - .f_create = rte_table_hash_create_key16_lru, - .f_free = rte_table_hash_free_key16_lru, - .f_add = rte_table_hash_entry_add_key16_lru, - .f_delete = rte_table_hash_entry_delete_key16_lru, - .f_lookup = rte_table_hash_lookup_key16_lru_dosig, - .f_stats = rte_table_hash_key16_stats_read, -}; - struct rte_table_ops rte_table_hash_key16_ext_ops = { .f_create = rte_table_hash_create_key16_ext, .f_free = rte_table_hash_free_key16_ext, @@ -1504,12 +1197,3 @@ struct rte_table_ops rte_table_hash_key16_ext_ops = { .f_lookup = rte_table_hash_lookup_key16_ext, .f_stats = rte_table_hash_key16_stats_read, }; - -struct rte_table_ops rte_table_hash_key16_ext_dosig_ops = { - .f_create = rte_table_hash_create_key16_ext, - .f_free = rte_table_hash_free_key16_ext, - .f_add = rte_table_hash_entry_add_key16_ext, - .f_delete = rte_table_hash_entry_delete_key16_ext, - .f_lookup = rte_table_hash_lookup_key16_ext_dosig, - .f_stats = rte_table_hash_key16_stats_read, -}; diff --git a/lib/librte_table/rte_table_hash_key32.c b/lib/librte_table/rte_table_hash_key32.c index 31fe6fda..d4364d62 100644 --- a/lib/librte_table/rte_table_hash_key32.c +++ b/lib/librte_table/rte_table_hash_key32.c @@ -1,34 +1,34 @@ /*- - * BSD LICENSE + * BSD LICENSE * - * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. - * All rights reserved. + * Copyright(c) 2010-2017 Intel Corporation. All rights reserved. + * All rights reserved. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Intel Corporation nor the names of its - * contributors may be used to endorse or promote products derived - * from this software without specific prior written permission. + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include <string.h> #include <stdio.h> @@ -42,7 +42,9 @@ #include "rte_table_hash.h" #include "rte_lru.h" -#define RTE_TABLE_HASH_KEY_SIZE 32 +#define KEY_SIZE 32 + +#define KEYS_PER_BUCKET 4 #define RTE_BUCKET_ENTRY_VALID 0x1LLU @@ -79,12 +81,11 @@ struct rte_table_hash { /* Input parameters */ uint32_t n_buckets; - uint32_t n_entries_per_bucket; uint32_t key_size; uint32_t entry_size; uint32_t bucket_size; - uint32_t signature_offset; uint32_t key_offset; + uint64_t key_mask[4]; rte_table_hash_op_hash f_hash; uint64_t seed; @@ -98,10 +99,52 @@ struct rte_table_hash { }; static int -check_params_create_lru(struct rte_table_hash_key32_lru_params *params) { - /* n_entries */ - if (params->n_entries == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries is zero\n", __func__); +keycmp(void *a, void *b, void *b_mask) +{ + uint64_t *a64 = a, *b64 = b, *b_mask64 = b_mask; + + return (a64[0] != (b64[0] & b_mask64[0])) || + (a64[1] != (b64[1] & b_mask64[1])) || + (a64[2] != (b64[2] & b_mask64[2])) || + (a64[3] != (b64[3] & b_mask64[3])); +} + +static void +keycpy(void *dst, void *src, void *src_mask) +{ + uint64_t *dst64 = dst, *src64 = src, *src_mask64 = src_mask; + + dst64[0] = src64[0] & src_mask64[0]; + dst64[1] = src64[1] & src_mask64[1]; + dst64[2] = src64[2] & src_mask64[2]; + dst64[3] = src64[3] & src_mask64[3]; +} + +static int +check_params_create(struct rte_table_hash_params *params) +{ + /* name */ + if (params->name == NULL) { + RTE_LOG(ERR, TABLE, "%s: name invalid value\n", __func__); + return -EINVAL; + } + + /* key_size */ + if (params->key_size != KEY_SIZE) { + RTE_LOG(ERR, TABLE, "%s: key_size invalid value\n", __func__); + return -EINVAL; + } + + /* n_keys */ + if (params->n_keys == 0) { + RTE_LOG(ERR, TABLE, "%s: n_keys is zero\n", __func__); + return -EINVAL; + } + + /* n_buckets */ + if ((params->n_buckets == 0) || + (!rte_is_power_of_2(params->n_buckets))) { + RTE_LOG(ERR, TABLE, "%s: n_buckets invalid value\n", __func__); return -EINVAL; } @@ -120,51 +163,83 @@ rte_table_hash_create_key32_lru(void *params, int socket_id, uint32_t entry_size) { - struct rte_table_hash_key32_lru_params *p = - (struct rte_table_hash_key32_lru_params *) params; + struct rte_table_hash_params *p = params; struct rte_table_hash *f; - uint32_t n_buckets, n_entries_per_bucket, key_size, bucket_size_cl; - uint32_t total_size, i; + uint64_t bucket_size, total_size; + uint32_t n_buckets, i; /* Check input parameters */ - if ((check_params_create_lru(p) != 0) || + if ((check_params_create(p) != 0) || ((sizeof(struct rte_table_hash) % RTE_CACHE_LINE_SIZE) != 0) || - ((sizeof(struct rte_bucket_4_32) % 64) != 0)) { + ((sizeof(struct rte_bucket_4_32) % 64) != 0)) return NULL; - } - n_entries_per_bucket = 4; - key_size = 32; + + /* + * Table dimensioning + * + * Objective: Pick the number of buckets (n_buckets) so that there a chance + * to store n_keys keys in the table. + * + * Note: Since the buckets do not get extended, it is not possible to + * guarantee that n_keys keys can be stored in the table at any time. In the + * worst case scenario when all the n_keys fall into the same bucket, only + * a maximum of KEYS_PER_BUCKET keys will be stored in the table. This case + * defeats the purpose of the hash table. It indicates unsuitable f_hash or + * n_keys to n_buckets ratio. + * + * MIN(n_buckets) = (n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET + */ + n_buckets = rte_align32pow2( + (p->n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET); + n_buckets = RTE_MAX(n_buckets, p->n_buckets); /* Memory allocation */ - n_buckets = rte_align32pow2((p->n_entries + n_entries_per_bucket - 1) / - n_entries_per_bucket); - bucket_size_cl = (sizeof(struct rte_bucket_4_32) + n_entries_per_bucket - * entry_size + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; - total_size = sizeof(struct rte_table_hash) + n_buckets * - bucket_size_cl * RTE_CACHE_LINE_SIZE; - - f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + bucket_size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_bucket_4_32) + + KEYS_PER_BUCKET * entry_size); + total_size = sizeof(struct rte_table_hash) + n_buckets * bucket_size; + if (total_size > SIZE_MAX) { + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); + return NULL; + } + + f = rte_zmalloc_socket(p->name, + (size_t)total_size, + RTE_CACHE_LINE_SIZE, + socket_id); if (f == NULL) { - RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for hash table\n", - __func__, total_size); + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); return NULL; } RTE_LOG(INFO, TABLE, - "%s: Hash table memory footprint is %u bytes\n", __func__, - total_size); + "%s: Hash table %s memory footprint " + "is %" PRIu64 " bytes\n", + __func__, p->name, total_size); /* Memory initialization */ f->n_buckets = n_buckets; - f->n_entries_per_bucket = n_entries_per_bucket; - f->key_size = key_size; + f->key_size = KEY_SIZE; f->entry_size = entry_size; - f->bucket_size = bucket_size_cl * RTE_CACHE_LINE_SIZE; - f->signature_offset = p->signature_offset; + f->bucket_size = bucket_size; f->key_offset = p->key_offset; f->f_hash = p->f_hash; f->seed = p->seed; + if (p->key_mask != NULL) { + f->key_mask[0] = ((uint64_t *)p->key_mask)[0]; + f->key_mask[1] = ((uint64_t *)p->key_mask)[1]; + f->key_mask[2] = ((uint64_t *)p->key_mask)[2]; + f->key_mask[3] = ((uint64_t *)p->key_mask)[3]; + } else { + f->key_mask[0] = 0xFFFFFFFFFFFFFFFFLLU; + f->key_mask[1] = 0xFFFFFFFFFFFFFFFFLLU; + f->key_mask[2] = 0xFFFFFFFFFFFFFFFFLLU; + f->key_mask[3] = 0xFFFFFFFFFFFFFFFFLLU; + } + for (i = 0; i < n_buckets; i++) { struct rte_bucket_4_32 *bucket; @@ -204,7 +279,7 @@ rte_table_hash_entry_add_key32_lru( uint64_t signature, pos; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket = (struct rte_bucket_4_32 *) &f->memory[bucket_index * f->bucket_size]; @@ -213,10 +288,10 @@ rte_table_hash_entry_add_key32_lru( /* Key is present in the bucket */ for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if ((bucket_signature == signature) && - (memcmp(key, bucket_key, f->key_size) == 0)) { + (keycmp(bucket_key, key, f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; memcpy(bucket_data, entry, f->entry_size); @@ -230,13 +305,13 @@ rte_table_hash_entry_add_key32_lru( /* Key is not present in the bucket */ for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if (bucket_signature == 0) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature[i] = signature; - memcpy(bucket_key, key, f->key_size); + keycpy(bucket_key, key, f->key_mask); memcpy(bucket_data, entry, f->entry_size); lru_update(bucket, i); *key_found = 0; @@ -249,10 +324,10 @@ rte_table_hash_entry_add_key32_lru( /* Bucket full: replace LRU entry */ pos = lru_pos(bucket); bucket->signature[pos] = signature; - memcpy(bucket->key[pos], key, f->key_size); + keycpy(&bucket->key[pos], key, f->key_mask); memcpy(&bucket->data[pos * f->entry_size], entry, f->entry_size); lru_update(bucket, pos); - *key_found = 0; + *key_found = 0; *entry_ptr = (void *) &bucket->data[pos * f->entry_size]; return 0; @@ -270,7 +345,7 @@ rte_table_hash_entry_delete_key32_lru( uint64_t signature; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket = (struct rte_bucket_4_32 *) &f->memory[bucket_index * f->bucket_size]; @@ -279,10 +354,10 @@ rte_table_hash_entry_delete_key32_lru( /* Key is present in the bucket */ for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if ((bucket_signature == signature) && - (memcmp(key, bucket_key, f->key_size) == 0)) { + (keycmp(bucket_key, key, f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature[i] = 0; @@ -299,81 +374,72 @@ rte_table_hash_entry_delete_key32_lru( return 0; } -static int -check_params_create_ext(struct rte_table_hash_key32_ext_params *params) { - /* n_entries */ - if (params->n_entries == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries is zero\n", __func__); - return -EINVAL; - } - - /* n_entries_ext */ - if (params->n_entries_ext == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries_ext is zero\n", __func__); - return -EINVAL; - } - - /* f_hash */ - if (params->f_hash == NULL) { - RTE_LOG(ERR, TABLE, "%s: f_hash function pointer is NULL\n", - __func__); - return -EINVAL; - } - - return 0; -} - static void * rte_table_hash_create_key32_ext(void *params, int socket_id, uint32_t entry_size) { - struct rte_table_hash_key32_ext_params *p = - params; + struct rte_table_hash_params *p = params; struct rte_table_hash *f; - uint32_t n_buckets, n_buckets_ext, n_entries_per_bucket; - uint32_t key_size, bucket_size_cl, stack_size_cl, total_size, i; + uint64_t bucket_size, stack_size, total_size; + uint32_t n_buckets_ext, i; /* Check input parameters */ - if ((check_params_create_ext(p) != 0) || + if ((check_params_create(p) != 0) || ((sizeof(struct rte_table_hash) % RTE_CACHE_LINE_SIZE) != 0) || ((sizeof(struct rte_bucket_4_32) % 64) != 0)) return NULL; - n_entries_per_bucket = 4; - key_size = 32; + /* + * Table dimensioning + * + * Objective: Pick the number of bucket extensions (n_buckets_ext) so that + * it is guaranteed that n_keys keys can be stored in the table at any time. + * + * The worst case scenario takes place when all the n_keys keys fall into + * the same bucket. Actually, due to the KEYS_PER_BUCKET scheme, the worst + * case takes place when (n_keys - KEYS_PER_BUCKET + 1) keys fall into the + * same bucket, while the remaining (KEYS_PER_BUCKET - 1) keys each fall + * into a different bucket. This case defeats the purpose of the hash table. + * It indicates unsuitable f_hash or n_keys to n_buckets ratio. + * + * n_buckets_ext = n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1 + */ + n_buckets_ext = p->n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1; /* Memory allocation */ - n_buckets = rte_align32pow2((p->n_entries + n_entries_per_bucket - 1) / - n_entries_per_bucket); - n_buckets_ext = (p->n_entries_ext + n_entries_per_bucket - 1) / - n_entries_per_bucket; - bucket_size_cl = (sizeof(struct rte_bucket_4_32) + n_entries_per_bucket - * entry_size + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; - stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + RTE_CACHE_LINE_SIZE - 1) - / RTE_CACHE_LINE_SIZE; + bucket_size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_bucket_4_32) + + KEYS_PER_BUCKET * entry_size); + stack_size = RTE_CACHE_LINE_ROUNDUP(n_buckets_ext * sizeof(uint32_t)); total_size = sizeof(struct rte_table_hash) + - ((n_buckets + n_buckets_ext) * bucket_size_cl + stack_size_cl) * - RTE_CACHE_LINE_SIZE; + (p->n_buckets + n_buckets_ext) * bucket_size + stack_size; + if (total_size > SIZE_MAX) { + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); + return NULL; + } - f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + f = rte_zmalloc_socket(p->name, + (size_t)total_size, + RTE_CACHE_LINE_SIZE, + socket_id); if (f == NULL) { - RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for hash table\n", - __func__, total_size); + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); return NULL; } RTE_LOG(INFO, TABLE, - "%s: Hash table memory footprint is %u bytes\n", __func__, - total_size); + "%s: Hash table %s memory footprint " + "is %" PRIu64" bytes\n", + __func__, p->name, total_size); /* Memory initialization */ - f->n_buckets = n_buckets; - f->n_entries_per_bucket = n_entries_per_bucket; - f->key_size = key_size; + f->n_buckets = p->n_buckets; + f->key_size = KEY_SIZE; f->entry_size = entry_size; - f->bucket_size = bucket_size_cl * RTE_CACHE_LINE_SIZE; - f->signature_offset = p->signature_offset; + f->bucket_size = bucket_size; f->key_offset = p->key_offset; f->f_hash = p->f_hash; f->seed = p->seed; @@ -381,7 +447,19 @@ rte_table_hash_create_key32_ext(void *params, f->n_buckets_ext = n_buckets_ext; f->stack_pos = n_buckets_ext; f->stack = (uint32_t *) - &f->memory[(n_buckets + n_buckets_ext) * f->bucket_size]; + &f->memory[(p->n_buckets + n_buckets_ext) * f->bucket_size]; + + if (p->key_mask != NULL) { + f->key_mask[0] = (((uint64_t *)p->key_mask)[0]); + f->key_mask[1] = (((uint64_t *)p->key_mask)[1]); + f->key_mask[2] = (((uint64_t *)p->key_mask)[2]); + f->key_mask[3] = (((uint64_t *)p->key_mask)[3]); + } else { + f->key_mask[0] = 0xFFFFFFFFFFFFFFFFLLU; + f->key_mask[1] = 0xFFFFFFFFFFFFFFFFLLU; + f->key_mask[2] = 0xFFFFFFFFFFFFFFFFLLU; + f->key_mask[3] = 0xFFFFFFFFFFFFFFFFLLU; + } for (i = 0; i < n_buckets_ext; i++) f->stack[i] = i; @@ -417,7 +495,7 @@ rte_table_hash_entry_add_key32_ext( uint64_t signature; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket0 = (struct rte_bucket_4_32 *) &f->memory[bucket_index * f->bucket_size]; @@ -427,10 +505,10 @@ rte_table_hash_entry_add_key32_ext( for (bucket = bucket0; bucket != NULL; bucket = bucket->next) { for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if ((bucket_signature == signature) && - (memcmp(key, bucket_key, f->key_size) == 0)) { + (keycmp(bucket_key, key, f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; @@ -448,14 +526,14 @@ rte_table_hash_entry_add_key32_ext( bucket_prev = bucket, bucket = bucket->next) for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if (bucket_signature == 0) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature[i] = signature; - memcpy(bucket_key, key, f->key_size); + keycpy(bucket_key, key, f->key_mask); memcpy(bucket_data, entry, f->entry_size); *key_found = 0; *entry_ptr = (void *) bucket_data; @@ -475,7 +553,7 @@ rte_table_hash_entry_add_key32_ext( bucket_prev->next_valid = 1; bucket->signature[0] = signature; - memcpy(bucket->key[0], key, f->key_size); + keycpy(&bucket->key[0], key, f->key_mask); memcpy(&bucket->data[0], entry, f->entry_size); *key_found = 0; *entry_ptr = (void *) &bucket->data[0]; @@ -497,7 +575,7 @@ rte_table_hash_entry_delete_key32_ext( uint64_t signature; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket0 = (struct rte_bucket_4_32 *) &f->memory[bucket_index * f->bucket_size]; @@ -508,24 +586,23 @@ rte_table_hash_entry_delete_key32_ext( bucket_prev = bucket, bucket = bucket->next) for (i = 0; i < 4; i++) { uint64_t bucket_signature = bucket->signature[i]; - uint8_t *bucket_key = (uint8_t *) bucket->key[i]; + uint8_t *bucket_key = (uint8_t *) &bucket->key[i]; if ((bucket_signature == signature) && - (memcmp(key, bucket_key, f->key_size) == 0)) { + (keycmp(bucket_key, key, f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature[i] = 0; *key_found = 1; if (entry) - memcpy(entry, bucket_data, - f->entry_size); + memcpy(entry, bucket_data, f->entry_size); if ((bucket->signature[0] == 0) && - (bucket->signature[1] == 0) && - (bucket->signature[2] == 0) && - (bucket->signature[3] == 0) && - (bucket_prev != NULL)) { + (bucket->signature[1] == 0) && + (bucket->signature[2] == 0) && + (bucket->signature[3] == 0) && + (bucket_prev != NULL)) { bucket_prev->next = bucket->next; bucket_prev->next_valid = bucket->next_valid; @@ -546,34 +623,39 @@ rte_table_hash_entry_delete_key32_ext( return 0; } -#define lookup_key32_cmp(key_in, bucket, pos) \ +#define lookup_key32_cmp(key_in, bucket, pos, f) \ { \ - uint64_t xor[4][4], or[4], signature[4]; \ + uint64_t xor[4][4], or[4], signature[4], k[4]; \ + \ + k[0] = key_in[0] & f->key_mask[0]; \ + k[1] = key_in[1] & f->key_mask[1]; \ + k[2] = key_in[2] & f->key_mask[2]; \ + k[3] = key_in[3] & f->key_mask[3]; \ \ signature[0] = ((~bucket->signature[0]) & 1); \ signature[1] = ((~bucket->signature[1]) & 1); \ signature[2] = ((~bucket->signature[2]) & 1); \ signature[3] = ((~bucket->signature[3]) & 1); \ \ - xor[0][0] = key_in[0] ^ bucket->key[0][0]; \ - xor[0][1] = key_in[1] ^ bucket->key[0][1]; \ - xor[0][2] = key_in[2] ^ bucket->key[0][2]; \ - xor[0][3] = key_in[3] ^ bucket->key[0][3]; \ + xor[0][0] = k[0] ^ bucket->key[0][0]; \ + xor[0][1] = k[1] ^ bucket->key[0][1]; \ + xor[0][2] = k[2] ^ bucket->key[0][2]; \ + xor[0][3] = k[3] ^ bucket->key[0][3]; \ \ - xor[1][0] = key_in[0] ^ bucket->key[1][0]; \ - xor[1][1] = key_in[1] ^ bucket->key[1][1]; \ - xor[1][2] = key_in[2] ^ bucket->key[1][2]; \ - xor[1][3] = key_in[3] ^ bucket->key[1][3]; \ + xor[1][0] = k[0] ^ bucket->key[1][0]; \ + xor[1][1] = k[1] ^ bucket->key[1][1]; \ + xor[1][2] = k[2] ^ bucket->key[1][2]; \ + xor[1][3] = k[3] ^ bucket->key[1][3]; \ \ - xor[2][0] = key_in[0] ^ bucket->key[2][0]; \ - xor[2][1] = key_in[1] ^ bucket->key[2][1]; \ - xor[2][2] = key_in[2] ^ bucket->key[2][2]; \ - xor[2][3] = key_in[3] ^ bucket->key[2][3]; \ + xor[2][0] = k[0] ^ bucket->key[2][0]; \ + xor[2][1] = k[1] ^ bucket->key[2][1]; \ + xor[2][2] = k[2] ^ bucket->key[2][2]; \ + xor[2][3] = k[3] ^ bucket->key[2][3]; \ \ - xor[3][0] = key_in[0] ^ bucket->key[3][0]; \ - xor[3][1] = key_in[1] ^ bucket->key[3][1]; \ - xor[3][2] = key_in[2] ^ bucket->key[3][2]; \ - xor[3][3] = key_in[3] ^ bucket->key[3][3]; \ + xor[3][0] = k[0] ^ bucket->key[3][0]; \ + xor[3][1] = k[1] ^ bucket->key[3][1]; \ + xor[3][2] = k[2] ^ bucket->key[3][2]; \ + xor[3][3] = k[3] ^ bucket->key[3][3]; \ \ or[0] = xor[0][0] | xor[0][1] | xor[0][2] | xor[0][3] | signature[0];\ or[1] = xor[1][0] | xor[1][1] | xor[1][2] | xor[1][3] | signature[1];\ @@ -604,12 +686,15 @@ rte_table_hash_entry_delete_key32_ext( rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf0, key_offset));\ } -#define lookup1_stage1(mbuf1, bucket1, f) \ +#define lookup1_stage1(mbuf1, bucket1, f) \ { \ + uint64_t *key; \ uint64_t signature; \ uint32_t bucket_index; \ \ - signature = RTE_MBUF_METADATA_UINT32(mbuf1, f->signature_offset);\ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf1, f->key_offset); \ + signature = f->f_hash(key, f->key_mask, KEY_SIZE, f->seed); \ + \ bucket_index = signature & (f->n_buckets - 1); \ bucket1 = (struct rte_bucket_4_32 *) \ &f->memory[bucket_index * f->bucket_size]; \ @@ -627,8 +712,7 @@ rte_table_hash_entry_delete_key32_ext( uint32_t pos; \ \ key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ - \ - lookup_key32_cmp(key, bucket2, pos); \ + lookup_key32_cmp(key, bucket2, pos, f); \ \ pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\ pkts_mask_out |= pkt_mask; \ @@ -649,8 +733,7 @@ rte_table_hash_entry_delete_key32_ext( uint32_t pos; \ \ key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ - \ - lookup_key32_cmp(key, bucket2, pos); \ + lookup_key32_cmp(key, bucket2, pos, f); \ \ pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\ pkts_mask_out |= pkt_mask; \ @@ -678,7 +761,7 @@ rte_table_hash_entry_delete_key32_ext( bucket = buckets[pkt_index]; \ key = keys[pkt_index]; \ \ - lookup_key32_cmp(key, bucket, pos); \ + lookup_key32_cmp(key, bucket, pos, f); \ \ pkt_mask = (bucket->signature[pos] & 1LLU) << pkt_index;\ pkts_mask_out |= pkt_mask; \ @@ -745,22 +828,27 @@ rte_table_hash_entry_delete_key32_ext( #define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f) \ { \ - uint64_t signature10, signature11; \ - uint32_t bucket10_index, bucket11_index; \ + uint64_t *key10, *key11; \ + uint64_t signature10, signature11; \ + uint32_t bucket10_index, bucket11_index; \ \ - signature10 = RTE_MBUF_METADATA_UINT32(mbuf10, f->signature_offset);\ - bucket10_index = signature10 & (f->n_buckets - 1); \ + key10 = RTE_MBUF_METADATA_UINT64_PTR(mbuf10, f->key_offset); \ + signature10 = f->f_hash(key10, f->key_mask, KEY_SIZE, f->seed); \ + \ + bucket10_index = signature10 & (f->n_buckets - 1); \ bucket10 = (struct rte_bucket_4_32 *) \ &f->memory[bucket10_index * f->bucket_size]; \ - rte_prefetch0(bucket10); \ + rte_prefetch0(bucket10); \ rte_prefetch0((void *)(((uintptr_t) bucket10) + RTE_CACHE_LINE_SIZE));\ rte_prefetch0((void *)(((uintptr_t) bucket10) + 2 * RTE_CACHE_LINE_SIZE));\ \ - signature11 = RTE_MBUF_METADATA_UINT32(mbuf11, f->signature_offset);\ - bucket11_index = signature11 & (f->n_buckets - 1); \ + key11 = RTE_MBUF_METADATA_UINT64_PTR(mbuf11, f->key_offset); \ + signature11 = f->f_hash(key11, f->key_mask, KEY_SIZE, f->seed);\ + \ + bucket11_index = signature11 & (f->n_buckets - 1); \ bucket11 = (struct rte_bucket_4_32 *) \ &f->memory[bucket11_index * f->bucket_size]; \ - rte_prefetch0(bucket11); \ + rte_prefetch0(bucket11); \ rte_prefetch0((void *)(((uintptr_t) bucket11) + RTE_CACHE_LINE_SIZE));\ rte_prefetch0((void *)(((uintptr_t) bucket11) + 2 * RTE_CACHE_LINE_SIZE));\ } @@ -776,8 +864,8 @@ rte_table_hash_entry_delete_key32_ext( key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ \ - lookup_key32_cmp(key20, bucket20, pos20); \ - lookup_key32_cmp(key21, bucket21, pos21); \ + lookup_key32_cmp(key20, bucket20, pos20, f); \ + lookup_key32_cmp(key21, bucket21, pos21, f); \ \ pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\ pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\ @@ -805,8 +893,8 @@ rte_table_hash_entry_delete_key32_ext( key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ \ - lookup_key32_cmp(key20, bucket20, pos20); \ - lookup_key32_cmp(key21, bucket21, pos21); \ + lookup_key32_cmp(key20, bucket20, pos20, f); \ + lookup_key32_cmp(key21, bucket21, pos21, f); \ \ pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\ pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\ diff --git a/lib/librte_table/rte_table_hash_key8.c b/lib/librte_table/rte_table_hash_key8.c index 5f0c6566..94373043 100644 --- a/lib/librte_table/rte_table_hash_key8.c +++ b/lib/librte_table/rte_table_hash_key8.c @@ -1,34 +1,34 @@ /*- - * BSD LICENSE + * BSD LICENSE * - * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. - * All rights reserved. + * Copyright(c) 2010-2017 Intel Corporation. All rights reserved. + * All rights reserved. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Intel Corporation nor the names of its - * contributors may be used to endorse or promote products derived - * from this software without specific prior written permission. + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include <string.h> #include <stdio.h> @@ -42,7 +42,9 @@ #include "rte_table_hash.h" #include "rte_lru.h" -#define RTE_TABLE_HASH_KEY_SIZE 8 +#define KEY_SIZE 8 + +#define KEYS_PER_BUCKET 4 #ifdef RTE_TABLE_STATS_COLLECT @@ -76,11 +78,9 @@ struct rte_table_hash { /* Input parameters */ uint32_t n_buckets; - uint32_t n_entries_per_bucket; uint32_t key_size; uint32_t entry_size; uint32_t bucket_size; - uint32_t signature_offset; uint32_t key_offset; uint64_t key_mask; rte_table_hash_op_hash f_hash; @@ -96,10 +96,46 @@ struct rte_table_hash { }; static int -check_params_create_lru(struct rte_table_hash_key8_lru_params *params) { - /* n_entries */ - if (params->n_entries == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries is zero\n", __func__); +keycmp(void *a, void *b, void *b_mask) +{ + uint64_t *a64 = a, *b64 = b, *b_mask64 = b_mask; + + return a64[0] != (b64[0] & b_mask64[0]); +} + +static void +keycpy(void *dst, void *src, void *src_mask) +{ + uint64_t *dst64 = dst, *src64 = src, *src_mask64 = src_mask; + + dst64[0] = src64[0] & src_mask64[0]; +} + +static int +check_params_create(struct rte_table_hash_params *params) +{ + /* name */ + if (params->name == NULL) { + RTE_LOG(ERR, TABLE, "%s: name invalid value\n", __func__); + return -EINVAL; + } + + /* key_size */ + if (params->key_size != KEY_SIZE) { + RTE_LOG(ERR, TABLE, "%s: key_size invalid value\n", __func__); + return -EINVAL; + } + + /* n_keys */ + if (params->n_keys == 0) { + RTE_LOG(ERR, TABLE, "%s: n_keys is zero\n", __func__); + return -EINVAL; + } + + /* n_buckets */ + if ((params->n_buckets == 0) || + (!rte_is_power_of_2(params->n_buckets))) { + RTE_LOG(ERR, TABLE, "%s: n_buckets invalid value\n", __func__); return -EINVAL; } @@ -116,47 +152,68 @@ check_params_create_lru(struct rte_table_hash_key8_lru_params *params) { static void * rte_table_hash_create_key8_lru(void *params, int socket_id, uint32_t entry_size) { - struct rte_table_hash_key8_lru_params *p = - (struct rte_table_hash_key8_lru_params *) params; + struct rte_table_hash_params *p = params; struct rte_table_hash *f; - uint32_t n_buckets, n_entries_per_bucket, key_size, bucket_size_cl; - uint32_t total_size, i; + uint64_t bucket_size, total_size; + uint32_t n_buckets, i; /* Check input parameters */ - if ((check_params_create_lru(p) != 0) || + if ((check_params_create(p) != 0) || ((sizeof(struct rte_table_hash) % RTE_CACHE_LINE_SIZE) != 0) || - ((sizeof(struct rte_bucket_4_8) % 64) != 0)) { + ((sizeof(struct rte_bucket_4_8) % 64) != 0)) return NULL; - } - n_entries_per_bucket = 4; - key_size = 8; + + /* + * Table dimensioning + * + * Objective: Pick the number of buckets (n_buckets) so that there a chance + * to store n_keys keys in the table. + * + * Note: Since the buckets do not get extended, it is not possible to + * guarantee that n_keys keys can be stored in the table at any time. In the + * worst case scenario when all the n_keys fall into the same bucket, only + * a maximum of KEYS_PER_BUCKET keys will be stored in the table. This case + * defeats the purpose of the hash table. It indicates unsuitable f_hash or + * n_keys to n_buckets ratio. + * + * MIN(n_buckets) = (n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET + */ + n_buckets = rte_align32pow2( + (p->n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET); + n_buckets = RTE_MAX(n_buckets, p->n_buckets); /* Memory allocation */ - n_buckets = rte_align32pow2((p->n_entries + n_entries_per_bucket - 1) / - n_entries_per_bucket); - bucket_size_cl = (sizeof(struct rte_bucket_4_8) + n_entries_per_bucket * - entry_size + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; - total_size = sizeof(struct rte_table_hash) + n_buckets * - bucket_size_cl * RTE_CACHE_LINE_SIZE; - - f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + bucket_size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_bucket_4_8) + + KEYS_PER_BUCKET * entry_size); + total_size = sizeof(struct rte_table_hash) + n_buckets * bucket_size; + + if (total_size > SIZE_MAX) { + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes" + " for hash table %s\n", + __func__, total_size, p->name); + return NULL; + } + + f = rte_zmalloc_socket(p->name, + (size_t)total_size, + RTE_CACHE_LINE_SIZE, + socket_id); if (f == NULL) { - RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for hash table\n", - __func__, total_size); + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes" + " for hash table %s\n", + __func__, total_size, p->name); return NULL; } - RTE_LOG(INFO, TABLE, - "%s: Hash table memory footprint is %u bytes\n", - __func__, total_size); + + RTE_LOG(INFO, TABLE, "%s: Hash table %s memory footprint " + "is %" PRIu64 " bytes\n", + __func__, p->name, total_size); /* Memory initialization */ f->n_buckets = n_buckets; - f->n_entries_per_bucket = n_entries_per_bucket; - f->key_size = key_size; + f->key_size = KEY_SIZE; f->entry_size = entry_size; - f->bucket_size = bucket_size_cl * RTE_CACHE_LINE_SIZE; - f->signature_offset = p->signature_offset; + f->bucket_size = bucket_size; f->key_offset = p->key_offset; f->f_hash = p->f_hash; f->seed = p->seed; @@ -205,7 +262,7 @@ rte_table_hash_entry_add_key8_lru( uint64_t signature, mask, pos; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, &f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket = (struct rte_bucket_4_8 *) &f->memory[bucket_index * f->bucket_size]; @@ -213,10 +270,10 @@ rte_table_hash_entry_add_key8_lru( /* Key is present in the bucket */ for (i = 0, mask = 1LLU; i < 4; i++, mask <<= 1) { uint64_t bucket_signature = bucket->signature; - uint64_t bucket_key = bucket->key[i]; + uint64_t *bucket_key = &bucket->key[i]; if ((bucket_signature & mask) && - (*((uint64_t *) key) == bucket_key)) { + (keycmp(bucket_key, key, &f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; memcpy(bucket_data, entry, f->entry_size); @@ -235,7 +292,7 @@ rte_table_hash_entry_add_key8_lru( uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature |= mask; - bucket->key[i] = *((uint64_t *) key); + keycpy(&bucket->key[i], key, &f->key_mask); memcpy(bucket_data, entry, f->entry_size); lru_update(bucket, i); *key_found = 0; @@ -247,10 +304,10 @@ rte_table_hash_entry_add_key8_lru( /* Bucket full: replace LRU entry */ pos = lru_pos(bucket); - bucket->key[pos] = *((uint64_t *) key); + keycpy(&bucket->key[pos], key, &f->key_mask); memcpy(&bucket->data[pos * f->entry_size], entry, f->entry_size); lru_update(bucket, pos); - *key_found = 0; + *key_found = 0; *entry_ptr = (void *) &bucket->data[pos * f->entry_size]; return 0; @@ -268,7 +325,7 @@ rte_table_hash_entry_delete_key8_lru( uint64_t signature, mask; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, &f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket = (struct rte_bucket_4_8 *) &f->memory[bucket_index * f->bucket_size]; @@ -276,10 +333,10 @@ rte_table_hash_entry_delete_key8_lru( /* Key is present in the bucket */ for (i = 0, mask = 1LLU; i < 4; i++, mask <<= 1) { uint64_t bucket_signature = bucket->signature; - uint64_t bucket_key = bucket->key[i]; + uint64_t *bucket_key = &bucket->key[i]; if ((bucket_signature & mask) && - (*((uint64_t *) key) == bucket_key)) { + (keycmp(bucket_key, key, &f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; bucket->signature &= ~mask; @@ -296,79 +353,71 @@ rte_table_hash_entry_delete_key8_lru( return 0; } -static int -check_params_create_ext(struct rte_table_hash_key8_ext_params *params) { - /* n_entries */ - if (params->n_entries == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries is zero\n", __func__); - return -EINVAL; - } - - /* n_entries_ext */ - if (params->n_entries_ext == 0) { - RTE_LOG(ERR, TABLE, "%s: n_entries_ext is zero\n", __func__); - return -EINVAL; - } - - /* f_hash */ - if (params->f_hash == NULL) { - RTE_LOG(ERR, TABLE, "%s: f_hash function pointer is NULL\n", - __func__); - return -EINVAL; - } - - return 0; -} - static void * rte_table_hash_create_key8_ext(void *params, int socket_id, uint32_t entry_size) { - struct rte_table_hash_key8_ext_params *p = - (struct rte_table_hash_key8_ext_params *) params; + struct rte_table_hash_params *p = params; struct rte_table_hash *f; - uint32_t n_buckets, n_buckets_ext, n_entries_per_bucket, key_size; - uint32_t bucket_size_cl, stack_size_cl, total_size, i; + uint64_t bucket_size, stack_size, total_size; + uint32_t n_buckets_ext, i; /* Check input parameters */ - if ((check_params_create_ext(p) != 0) || + if ((check_params_create(p) != 0) || ((sizeof(struct rte_table_hash) % RTE_CACHE_LINE_SIZE) != 0) || ((sizeof(struct rte_bucket_4_8) % 64) != 0)) return NULL; - n_entries_per_bucket = 4; - key_size = 8; + /* + * Table dimensioning + * + * Objective: Pick the number of bucket extensions (n_buckets_ext) so that + * it is guaranteed that n_keys keys can be stored in the table at any time. + * + * The worst case scenario takes place when all the n_keys keys fall into + * the same bucket. Actually, due to the KEYS_PER_BUCKET scheme, the worst + * case takes place when (n_keys - KEYS_PER_BUCKET + 1) keys fall into the + * same bucket, while the remaining (KEYS_PER_BUCKET - 1) keys each fall + * into a different bucket. This case defeats the purpose of the hash table. + * It indicates unsuitable f_hash or n_keys to n_buckets ratio. + * + * n_buckets_ext = n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1 + */ + n_buckets_ext = p->n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1; /* Memory allocation */ - n_buckets = rte_align32pow2((p->n_entries + n_entries_per_bucket - 1) / - n_entries_per_bucket); - n_buckets_ext = (p->n_entries_ext + n_entries_per_bucket - 1) / - n_entries_per_bucket; - bucket_size_cl = (sizeof(struct rte_bucket_4_8) + n_entries_per_bucket * - entry_size + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; - stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + RTE_CACHE_LINE_SIZE - 1) - / RTE_CACHE_LINE_SIZE; - total_size = sizeof(struct rte_table_hash) + ((n_buckets + - n_buckets_ext) * bucket_size_cl + stack_size_cl) * - RTE_CACHE_LINE_SIZE; - - f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + bucket_size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_bucket_4_8) + + KEYS_PER_BUCKET * entry_size); + stack_size = RTE_CACHE_LINE_ROUNDUP(n_buckets_ext * sizeof(uint32_t)); + total_size = sizeof(struct rte_table_hash) + + (p->n_buckets + n_buckets_ext) * bucket_size + stack_size; + + if (total_size > SIZE_MAX) { + RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); + return NULL; + } + + f = rte_zmalloc_socket(p->name, + (size_t)total_size, + RTE_CACHE_LINE_SIZE, + socket_id); if (f == NULL) { RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for hash table\n", - __func__, total_size); + "%s: Cannot allocate %" PRIu64 " bytes " + "for hash table %s\n", + __func__, total_size, p->name); return NULL; } - RTE_LOG(INFO, TABLE, - "%s: Hash table memory footprint is %u bytes\n", - __func__, total_size); + RTE_LOG(INFO, TABLE, "%s: Hash table %s memory footprint " + "is %" PRIu64 " bytes\n", + __func__, p->name, total_size); /* Memory initialization */ - f->n_buckets = n_buckets; - f->n_entries_per_bucket = n_entries_per_bucket; - f->key_size = key_size; + f->n_buckets = p->n_buckets; + f->key_size = KEY_SIZE; f->entry_size = entry_size; - f->bucket_size = bucket_size_cl * RTE_CACHE_LINE_SIZE; - f->signature_offset = p->signature_offset; + f->bucket_size = bucket_size; f->key_offset = p->key_offset; f->f_hash = p->f_hash; f->seed = p->seed; @@ -376,7 +425,7 @@ rte_table_hash_create_key8_ext(void *params, int socket_id, uint32_t entry_size) f->n_buckets_ext = n_buckets_ext; f->stack_pos = n_buckets_ext; f->stack = (uint32_t *) - &f->memory[(n_buckets + n_buckets_ext) * f->bucket_size]; + &f->memory[(p->n_buckets + n_buckets_ext) * f->bucket_size]; if (p->key_mask != NULL) f->key_mask = ((uint64_t *)p->key_mask)[0]; @@ -417,7 +466,7 @@ rte_table_hash_entry_add_key8_ext( uint64_t signature; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, &f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket0 = (struct rte_bucket_4_8 *) &f->memory[bucket_index * f->bucket_size]; @@ -428,10 +477,10 @@ rte_table_hash_entry_add_key8_ext( for (i = 0, mask = 1LLU; i < 4; i++, mask <<= 1) { uint64_t bucket_signature = bucket->signature; - uint64_t bucket_key = bucket->key[i]; + uint64_t *bucket_key = &bucket->key[i]; if ((bucket_signature & mask) && - (*((uint64_t *) key) == bucket_key)) { + (keycmp(bucket_key, key, &f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; @@ -456,7 +505,7 @@ rte_table_hash_entry_add_key8_ext( f->entry_size]; bucket->signature |= mask; - bucket->key[i] = *((uint64_t *) key); + keycpy(&bucket->key[i], key, &f->key_mask); memcpy(bucket_data, entry, f->entry_size); *key_found = 0; *entry_ptr = (void *) bucket_data; @@ -476,7 +525,7 @@ rte_table_hash_entry_add_key8_ext( bucket_prev->next_valid = 1; bucket->signature = 1; - bucket->key[0] = *((uint64_t *) key); + keycpy(&bucket->key[0], key, &f->key_mask); memcpy(&bucket->data[0], entry, f->entry_size); *key_found = 0; *entry_ptr = (void *) &bucket->data[0]; @@ -498,7 +547,7 @@ rte_table_hash_entry_delete_key8_ext( uint64_t signature; uint32_t bucket_index, i; - signature = f->f_hash(key, f->key_size, f->seed); + signature = f->f_hash(key, &f->key_mask, f->key_size, f->seed); bucket_index = signature & (f->n_buckets - 1); bucket0 = (struct rte_bucket_4_8 *) &f->memory[bucket_index * f->bucket_size]; @@ -510,10 +559,10 @@ rte_table_hash_entry_delete_key8_ext( for (i = 0, mask = 1LLU; i < 4; i++, mask <<= 1) { uint64_t bucket_signature = bucket->signature; - uint64_t bucket_key = bucket->key[i]; + uint64_t *bucket_key = &bucket->key[i]; if ((bucket_signature & mask) && - (*((uint64_t *) key) == bucket_key)) { + (keycmp(bucket_key, key, &f->key_mask) == 0)) { uint8_t *bucket_data = &bucket->data[i * f->entry_size]; @@ -546,16 +595,17 @@ rte_table_hash_entry_delete_key8_ext( return 0; } -#define lookup_key8_cmp(key_in, bucket, pos) \ +#define lookup_key8_cmp(key_in, bucket, pos, f) \ { \ - uint64_t xor[4], signature; \ + uint64_t xor[4], signature, k; \ \ signature = ~bucket->signature; \ \ - xor[0] = (key_in[0] ^ bucket->key[0]) | (signature & 1);\ - xor[1] = (key_in[0] ^ bucket->key[1]) | (signature & 2);\ - xor[2] = (key_in[0] ^ bucket->key[2]) | (signature & 4);\ - xor[3] = (key_in[0] ^ bucket->key[3]) | (signature & 8);\ + k = key_in[0] & f->key_mask; \ + xor[0] = (k ^ bucket->key[0]) | (signature & 1); \ + xor[1] = (k ^ bucket->key[1]) | (signature & 2); \ + xor[2] = (k ^ bucket->key[2]) | (signature & 4); \ + xor[3] = (k ^ bucket->key[3]) | (signature & 8); \ \ pos = 4; \ if (xor[0] == 0) \ @@ -583,27 +633,12 @@ rte_table_hash_entry_delete_key8_ext( #define lookup1_stage1(mbuf1, bucket1, f) \ { \ - uint64_t signature; \ - uint32_t bucket_index; \ - \ - signature = RTE_MBUF_METADATA_UINT32(mbuf1, f->signature_offset);\ - bucket_index = signature & (f->n_buckets - 1); \ - bucket1 = (struct rte_bucket_4_8 *) \ - &f->memory[bucket_index * f->bucket_size]; \ - rte_prefetch0(bucket1); \ -} - -#define lookup1_stage1_dosig(mbuf1, bucket1, f) \ -{ \ uint64_t *key; \ uint64_t signature; \ uint32_t bucket_index; \ - uint64_t hash_key_buffer; \ \ key = RTE_MBUF_METADATA_UINT64_PTR(mbuf1, f->key_offset);\ - hash_key_buffer = *key & f->key_mask; \ - signature = f->f_hash(&hash_key_buffer, \ - RTE_TABLE_HASH_KEY_SIZE, f->seed); \ + signature = f->f_hash(key, &f->key_mask, KEY_SIZE, f->seed); \ bucket_index = signature & (f->n_buckets - 1); \ bucket1 = (struct rte_bucket_4_8 *) \ &f->memory[bucket_index * f->bucket_size]; \ @@ -617,12 +652,9 @@ rte_table_hash_entry_delete_key8_ext( uint64_t pkt_mask; \ uint64_t *key; \ uint32_t pos; \ - uint64_t hash_key_buffer; \ \ key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ - hash_key_buffer = key[0] & f->key_mask; \ - \ - lookup_key8_cmp((&hash_key_buffer), bucket2, pos); \ + lookup_key8_cmp(key, bucket2, pos, f); \ \ pkt_mask = ((bucket2->signature >> pos) & 1LLU) << pkt2_index;\ pkts_mask_out |= pkt_mask; \ @@ -641,12 +673,9 @@ rte_table_hash_entry_delete_key8_ext( uint64_t pkt_mask, bucket_mask; \ uint64_t *key; \ uint32_t pos; \ - uint64_t hash_key_buffer; \ \ key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ - hash_key_buffer = *key & f->key_mask; \ - \ - lookup_key8_cmp((&hash_key_buffer), bucket2, pos); \ + lookup_key8_cmp(key, bucket2, pos, f); \ \ pkt_mask = ((bucket2->signature >> pos) & 1LLU) << pkt2_index;\ pkts_mask_out |= pkt_mask; \ @@ -670,13 +699,10 @@ rte_table_hash_entry_delete_key8_ext( uint64_t pkt_mask, bucket_mask; \ uint64_t *key; \ uint32_t pos; \ - uint64_t hash_key_buffer; \ \ bucket = buckets[pkt_index]; \ key = keys[pkt_index]; \ - hash_key_buffer = (*key) & f->key_mask; \ - \ - lookup_key8_cmp((&hash_key_buffer), bucket, pos); \ + lookup_key8_cmp(key, bucket, pos, f); \ \ pkt_mask = ((bucket->signature >> pos) & 1LLU) << pkt_index;\ pkts_mask_out |= pkt_mask; \ @@ -738,29 +764,9 @@ rte_table_hash_entry_delete_key8_ext( rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));\ } -#define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f) \ -{ \ - uint64_t signature10, signature11; \ - uint32_t bucket10_index, bucket11_index; \ - \ - signature10 = RTE_MBUF_METADATA_UINT32(mbuf10, f->signature_offset);\ - bucket10_index = signature10 & (f->n_buckets - 1); \ - bucket10 = (struct rte_bucket_4_8 *) \ - &f->memory[bucket10_index * f->bucket_size]; \ - rte_prefetch0(bucket10); \ - \ - signature11 = RTE_MBUF_METADATA_UINT32(mbuf11, f->signature_offset);\ - bucket11_index = signature11 & (f->n_buckets - 1); \ - bucket11 = (struct rte_bucket_4_8 *) \ - &f->memory[bucket11_index * f->bucket_size]; \ - rte_prefetch0(bucket11); \ -} - -#define lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f)\ +#define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f)\ { \ uint64_t *key10, *key11; \ - uint64_t hash_offset_buffer10; \ - uint64_t hash_offset_buffer11; \ uint64_t signature10, signature11; \ uint32_t bucket10_index, bucket11_index; \ rte_table_hash_op_hash f_hash = f->f_hash; \ @@ -769,18 +775,14 @@ rte_table_hash_entry_delete_key8_ext( \ key10 = RTE_MBUF_METADATA_UINT64_PTR(mbuf10, key_offset);\ key11 = RTE_MBUF_METADATA_UINT64_PTR(mbuf11, key_offset);\ - hash_offset_buffer10 = *key10 & f->key_mask; \ - hash_offset_buffer11 = *key11 & f->key_mask; \ \ - signature10 = f_hash(&hash_offset_buffer10, \ - RTE_TABLE_HASH_KEY_SIZE, seed); \ + signature10 = f_hash(key10, &f->key_mask, KEY_SIZE, seed); \ bucket10_index = signature10 & (f->n_buckets - 1); \ bucket10 = (struct rte_bucket_4_8 *) \ &f->memory[bucket10_index * f->bucket_size]; \ rte_prefetch0(bucket10); \ \ - signature11 = f_hash(&hash_offset_buffer11, \ - RTE_TABLE_HASH_KEY_SIZE, seed); \ + signature11 = f_hash(key11, &f->key_mask, KEY_SIZE, seed); \ bucket11_index = signature11 & (f->n_buckets - 1); \ bucket11 = (struct rte_bucket_4_8 *) \ &f->memory[bucket11_index * f->bucket_size]; \ @@ -793,17 +795,13 @@ rte_table_hash_entry_delete_key8_ext( void *a20, *a21; \ uint64_t pkt20_mask, pkt21_mask; \ uint64_t *key20, *key21; \ - uint64_t hash_offset_buffer20; \ - uint64_t hash_offset_buffer21; \ uint32_t pos20, pos21; \ \ key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ - hash_offset_buffer20 = *key20 & f->key_mask; \ - hash_offset_buffer21 = *key21 & f->key_mask; \ \ - lookup_key8_cmp((&hash_offset_buffer20), bucket20, pos20);\ - lookup_key8_cmp((&hash_offset_buffer21), bucket21, pos21);\ + lookup_key8_cmp(key20, bucket20, pos20, f); \ + lookup_key8_cmp(key21, bucket21, pos21, f); \ \ pkt20_mask = ((bucket20->signature >> pos20) & 1LLU) << pkt20_index;\ pkt21_mask = ((bucket21->signature >> pos21) & 1LLU) << pkt21_index;\ @@ -826,17 +824,13 @@ rte_table_hash_entry_delete_key8_ext( void *a20, *a21; \ uint64_t pkt20_mask, pkt21_mask, bucket20_mask, bucket21_mask;\ uint64_t *key20, *key21; \ - uint64_t hash_offset_buffer20; \ - uint64_t hash_offset_buffer21; \ uint32_t pos20, pos21; \ \ key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ - hash_offset_buffer20 = *key20 & f->key_mask; \ - hash_offset_buffer21 = *key21 & f->key_mask; \ \ - lookup_key8_cmp((&hash_offset_buffer20), bucket20, pos20);\ - lookup_key8_cmp((&hash_offset_buffer21), bucket21, pos21);\ + lookup_key8_cmp(key20, bucket20, pos20, f); \ + lookup_key8_cmp(key21, bucket21, pos21, f); \ \ pkt20_mask = ((bucket20->signature >> pos20) & 1LLU) << pkt20_index;\ pkt21_mask = ((bucket21->signature >> pos21) & 1LLU) << pkt21_index;\ @@ -871,8 +865,8 @@ rte_table_hash_lookup_key8_lru( struct rte_table_hash *f = (struct rte_table_hash *) table; struct rte_bucket_4_8 *bucket10, *bucket11, *bucket20, *bucket21; struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; - uint32_t pkt00_index, pkt01_index, pkt10_index, - pkt11_index, pkt20_index, pkt21_index; + uint32_t pkt00_index, pkt01_index, pkt10_index; + uint32_t pkt11_index, pkt20_index, pkt21_index; uint64_t pkts_mask_out = 0; __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); @@ -888,7 +882,7 @@ rte_table_hash_lookup_key8_lru( lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); lookup1_stage1(mbuf, bucket, f); lookup1_stage2_lru(pkt_index, mbuf, bucket, - pkts_mask_out, entries, f); + pkts_mask_out, entries, f); } *lookup_hit_mask = pkts_mask_out; @@ -984,133 +978,7 @@ rte_table_hash_lookup_key8_lru( *lookup_hit_mask = pkts_mask_out; RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); return 0; -} /* rte_table_hash_lookup_key8_lru() */ - -static int -rte_table_hash_lookup_key8_lru_dosig( - void *table, - struct rte_mbuf **pkts, - uint64_t pkts_mask, - uint64_t *lookup_hit_mask, - void **entries) -{ - struct rte_table_hash *f = (struct rte_table_hash *) table; - struct rte_bucket_4_8 *bucket10, *bucket11, *bucket20, *bucket21; - struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; - uint32_t pkt00_index, pkt01_index, pkt10_index; - uint32_t pkt11_index, pkt20_index, pkt21_index; - uint64_t pkts_mask_out = 0; - - __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); - RTE_TABLE_HASH_KEY8_STATS_PKTS_IN_ADD(f, n_pkts_in); - - /* Cannot run the pipeline with less than 5 packets */ - if (__builtin_popcountll(pkts_mask) < 5) { - for ( ; pkts_mask; ) { - struct rte_bucket_4_8 *bucket; - struct rte_mbuf *mbuf; - uint32_t pkt_index; - - lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); - lookup1_stage1_dosig(mbuf, bucket, f); - lookup1_stage2_lru(pkt_index, mbuf, bucket, - pkts_mask_out, entries, f); - } - - *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); - return 0; - } - - /* - * Pipeline fill - * - */ - /* Pipeline stage 0 */ - lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, - pkts_mask, f); - - /* Pipeline feed */ - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, - pkts_mask, f); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* - * Pipeline run - * - */ - for ( ; pkts_mask; ) { - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, - mbuf00, mbuf01, pkts, pkts_mask, f); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* Pipeline stage 2 */ - lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, f); - } - - /* - * Pipeline flush - * - */ - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* Pipeline stage 2 */ - lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, f); - - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - - /* Pipeline stage 2 */ - lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, f); - - *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); - return 0; -} /* rte_table_hash_lookup_key8_lru_dosig() */ +} /* lookup LRU */ static int rte_table_hash_lookup_key8_ext( @@ -1142,8 +1010,8 @@ rte_table_hash_lookup_key8_ext( lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); lookup1_stage1(mbuf, bucket, f); lookup1_stage2_ext(pkt_index, mbuf, bucket, - pkts_mask_out, entries, buckets_mask, buckets, - keys, f); + pkts_mask_out, entries, buckets_mask, + buckets, keys, f); } goto grind_next_buckets; @@ -1260,157 +1128,7 @@ grind_next_buckets: *lookup_hit_mask = pkts_mask_out; RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); return 0; -} /* rte_table_hash_lookup_key8_ext() */ - -static int -rte_table_hash_lookup_key8_ext_dosig( - void *table, - struct rte_mbuf **pkts, - uint64_t pkts_mask, - uint64_t *lookup_hit_mask, - void **entries) -{ - struct rte_table_hash *f = (struct rte_table_hash *) table; - struct rte_bucket_4_8 *bucket10, *bucket11, *bucket20, *bucket21; - struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; - uint32_t pkt00_index, pkt01_index, pkt10_index; - uint32_t pkt11_index, pkt20_index, pkt21_index; - uint64_t pkts_mask_out = 0, buckets_mask = 0; - struct rte_bucket_4_8 *buckets[RTE_PORT_IN_BURST_SIZE_MAX]; - uint64_t *keys[RTE_PORT_IN_BURST_SIZE_MAX]; - - __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); - RTE_TABLE_HASH_KEY8_STATS_PKTS_IN_ADD(f, n_pkts_in); - - /* Cannot run the pipeline with less than 5 packets */ - if (__builtin_popcountll(pkts_mask) < 5) { - for ( ; pkts_mask; ) { - struct rte_bucket_4_8 *bucket; - struct rte_mbuf *mbuf; - uint32_t pkt_index; - - lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); - lookup1_stage1_dosig(mbuf, bucket, f); - lookup1_stage2_ext(pkt_index, mbuf, bucket, - pkts_mask_out, entries, buckets_mask, - buckets, keys, f); - } - - goto grind_next_buckets; - } - - /* - * Pipeline fill - * - */ - /* Pipeline stage 0 */ - lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, - pkts_mask, f); - - /* Pipeline feed */ - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, - pkts_mask, f); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* - * Pipeline run - * - */ - for ( ; pkts_mask; ) { - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, - mbuf00, mbuf01, pkts, pkts_mask, f); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* Pipeline stage 2 */ - lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, - buckets_mask, buckets, keys, f); - } - - /* - * Pipeline flush - * - */ - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - mbuf10 = mbuf00; - mbuf11 = mbuf01; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); - - /* Pipeline stage 2 */ - lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, - buckets_mask, buckets, keys, f); - - /* Pipeline feed */ - bucket20 = bucket10; - bucket21 = bucket11; - mbuf20 = mbuf10; - mbuf21 = mbuf11; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - - /* Pipeline stage 2 */ - lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, - bucket20, bucket21, pkts_mask_out, entries, - buckets_mask, buckets, keys, f); - -grind_next_buckets: - /* Grind next buckets */ - for ( ; buckets_mask; ) { - uint64_t buckets_mask_next = 0; - - for ( ; buckets_mask; ) { - uint64_t pkt_mask; - uint32_t pkt_index; - - pkt_index = __builtin_ctzll(buckets_mask); - pkt_mask = 1LLU << pkt_index; - buckets_mask &= ~pkt_mask; - - lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, - entries, buckets_mask_next, f); - } - - buckets_mask = buckets_mask_next; - } - - *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); - return 0; -} /* rte_table_hash_lookup_key8_dosig_ext() */ +} /* lookup EXT */ static int rte_table_hash_key8_stats_read(void *table, struct rte_table_stats *stats, int clear) @@ -1437,17 +1155,6 @@ struct rte_table_ops rte_table_hash_key8_lru_ops = { .f_stats = rte_table_hash_key8_stats_read, }; -struct rte_table_ops rte_table_hash_key8_lru_dosig_ops = { - .f_create = rte_table_hash_create_key8_lru, - .f_free = rte_table_hash_free_key8_lru, - .f_add = rte_table_hash_entry_add_key8_lru, - .f_delete = rte_table_hash_entry_delete_key8_lru, - .f_add_bulk = NULL, - .f_delete_bulk = NULL, - .f_lookup = rte_table_hash_lookup_key8_lru_dosig, - .f_stats = rte_table_hash_key8_stats_read, -}; - struct rte_table_ops rte_table_hash_key8_ext_ops = { .f_create = rte_table_hash_create_key8_ext, .f_free = rte_table_hash_free_key8_ext, @@ -1458,14 +1165,3 @@ struct rte_table_ops rte_table_hash_key8_ext_ops = { .f_lookup = rte_table_hash_lookup_key8_ext, .f_stats = rte_table_hash_key8_stats_read, }; - -struct rte_table_ops rte_table_hash_key8_ext_dosig_ops = { - .f_create = rte_table_hash_create_key8_ext, - .f_free = rte_table_hash_free_key8_ext, - .f_add = rte_table_hash_entry_add_key8_ext, - .f_delete = rte_table_hash_entry_delete_key8_ext, - .f_add_bulk = NULL, - .f_delete_bulk = NULL, - .f_lookup = rte_table_hash_lookup_key8_ext_dosig, - .f_stats = rte_table_hash_key8_stats_read, -}; diff --git a/lib/librte_table/rte_table_hash_lru.c b/lib/librte_table/rte_table_hash_lru.c index 5a4864e2..a07392fd 100644 --- a/lib/librte_table/rte_table_hash_lru.c +++ b/lib/librte_table/rte_table_hash_lru.c @@ -1,34 +1,34 @@ /*- - * BSD LICENSE + * BSD LICENSE * - * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. - * All rights reserved. + * Copyright(c) 2010-2017 Intel Corporation. All rights reserved. + * All rights reserved. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Intel Corporation nor the names of its - * contributors may be used to endorse or promote products derived - * from this software without specific prior written permission. + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include <string.h> @@ -86,7 +86,6 @@ struct rte_table_hash { uint32_t n_buckets; rte_table_hash_op_hash f_hash; uint64_t seed; - uint32_t signature_offset; uint32_t key_offset; /* Internal */ @@ -99,6 +98,7 @@ struct rte_table_hash { struct grinder grinders[RTE_PORT_IN_BURST_SIZE_MAX]; /* Tables */ + uint64_t *key_mask; struct bucket *buckets; uint8_t *key_mem; uint8_t *data_mem; @@ -109,29 +109,53 @@ struct rte_table_hash { }; static int -check_params_create(struct rte_table_hash_lru_params *params) +keycmp(void *a, void *b, void *b_mask, uint32_t n_bytes) { - uint32_t n_buckets_min; + uint64_t *a64 = a, *b64 = b, *b_mask64 = b_mask; + uint32_t i; + + for (i = 0; i < n_bytes / sizeof(uint64_t); i++) + if (a64[i] != (b64[i] & b_mask64[i])) + return 1; + + return 0; +} + +static void +keycpy(void *dst, void *src, void *src_mask, uint32_t n_bytes) +{ + uint64_t *dst64 = dst, *src64 = src, *src_mask64 = src_mask; + uint32_t i; + + for (i = 0; i < n_bytes / sizeof(uint64_t); i++) + dst64[i] = src64[i] & src_mask64[i]; +} + +static int +check_params_create(struct rte_table_hash_params *params) +{ + /* name */ + if (params->name == NULL) { + RTE_LOG(ERR, TABLE, "%s: name invalid value\n", __func__); + return -EINVAL; + } /* key_size */ - if ((params->key_size == 0) || + if ((params->key_size < sizeof(uint64_t)) || (!rte_is_power_of_2(params->key_size))) { RTE_LOG(ERR, TABLE, "%s: key_size invalid value\n", __func__); return -EINVAL; } /* n_keys */ - if ((params->n_keys == 0) || - (!rte_is_power_of_2(params->n_keys))) { + if (params->n_keys == 0) { RTE_LOG(ERR, TABLE, "%s: n_keys invalid value\n", __func__); return -EINVAL; } /* n_buckets */ - n_buckets_min = (params->n_keys + KEYS_PER_BUCKET - 1) / params->n_keys; if ((params->n_buckets == 0) || - (!rte_is_power_of_2(params->n_keys)) || - (params->n_buckets < n_buckets_min)) { + (!rte_is_power_of_2(params->n_buckets))) { RTE_LOG(ERR, TABLE, "%s: n_buckets invalid value\n", __func__); return -EINVAL; } @@ -148,13 +172,13 @@ check_params_create(struct rte_table_hash_lru_params *params) static void * rte_table_hash_lru_create(void *params, int socket_id, uint32_t entry_size) { - struct rte_table_hash_lru_params *p = - params; + struct rte_table_hash_params *p = params; struct rte_table_hash *t; - uint32_t total_size, table_meta_sz; - uint32_t bucket_sz, key_sz, key_stack_sz, data_sz; - uint32_t bucket_offset, key_offset, key_stack_offset, data_offset; - uint32_t i; + uint64_t table_meta_sz, key_mask_sz, bucket_sz, key_sz, key_stack_sz; + uint64_t data_sz, total_size; + uint64_t key_mask_offset, bucket_offset, key_offset, key_stack_offset; + uint64_t data_offset; + uint32_t n_buckets, i; /* Check input parameters */ if ((check_params_create(p) != 0) || @@ -164,33 +188,65 @@ rte_table_hash_lru_create(void *params, int socket_id, uint32_t entry_size) return NULL; } + /* + * Table dimensioning + * + * Objective: Pick the number of buckets (n_buckets) so that there a chance + * to store n_keys keys in the table. + * + * Note: Since the buckets do not get extended, it is not possible to + * guarantee that n_keys keys can be stored in the table at any time. In the + * worst case scenario when all the n_keys fall into the same bucket, only + * a maximum of KEYS_PER_BUCKET keys will be stored in the table. This case + * defeats the purpose of the hash table. It indicates unsuitable f_hash or + * n_keys to n_buckets ratio. + * + * MIN(n_buckets) = (n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET + */ + n_buckets = rte_align32pow2( + (p->n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET); + n_buckets = RTE_MAX(n_buckets, p->n_buckets); + /* Memory allocation */ table_meta_sz = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_table_hash)); - bucket_sz = RTE_CACHE_LINE_ROUNDUP(p->n_buckets * sizeof(struct bucket)); + key_mask_sz = RTE_CACHE_LINE_ROUNDUP(p->key_size); + bucket_sz = RTE_CACHE_LINE_ROUNDUP(n_buckets * sizeof(struct bucket)); key_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * p->key_size); key_stack_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * sizeof(uint32_t)); data_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * entry_size); - total_size = table_meta_sz + bucket_sz + key_sz + key_stack_sz + - data_sz; + total_size = table_meta_sz + key_mask_sz + bucket_sz + key_sz + + key_stack_sz + data_sz; + + if (total_size > SIZE_MAX) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %" PRIu64 " bytes for hash " + "table %s\n", + __func__, total_size, p->name); + return NULL; + } - t = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + t = rte_zmalloc_socket(p->name, + (size_t)total_size, + RTE_CACHE_LINE_SIZE, + socket_id); if (t == NULL) { RTE_LOG(ERR, TABLE, - "%s: Cannot allocate %u bytes for hash table\n", - __func__, total_size); + "%s: Cannot allocate %" PRIu64 " bytes for hash " + "table %s\n", + __func__, total_size, p->name); return NULL; } - RTE_LOG(INFO, TABLE, "%s (%u-byte key): Hash table memory footprint is " - "%u bytes\n", __func__, p->key_size, total_size); + RTE_LOG(INFO, TABLE, "%s (%u-byte key): Hash table %s memory footprint" + " is %" PRIu64 " bytes\n", + __func__, p->key_size, p->name, total_size); /* Memory initialization */ t->key_size = p->key_size; t->entry_size = entry_size; t->n_keys = p->n_keys; - t->n_buckets = p->n_buckets; + t->n_buckets = n_buckets; t->f_hash = p->f_hash; t->seed = p->seed; - t->signature_offset = p->signature_offset; t->key_offset = p->key_offset; /* Internal */ @@ -199,16 +255,24 @@ rte_table_hash_lru_create(void *params, int socket_id, uint32_t entry_size) t->data_size_shl = __builtin_ctzl(entry_size); /* Tables */ - bucket_offset = 0; + key_mask_offset = 0; + bucket_offset = key_mask_offset + key_mask_sz; key_offset = bucket_offset + bucket_sz; key_stack_offset = key_offset + key_sz; data_offset = key_stack_offset + key_stack_sz; + t->key_mask = (uint64_t *) &t->memory[key_mask_offset]; t->buckets = (struct bucket *) &t->memory[bucket_offset]; t->key_mem = &t->memory[key_offset]; t->key_stack = (uint32_t *) &t->memory[key_stack_offset]; t->data_mem = &t->memory[data_offset]; + /* Key mask */ + if (p->key_mask == NULL) + memset(t->key_mask, 0xFF, p->key_size); + else + memcpy(t->key_mask, p->key_mask, p->key_size); + /* Key stack */ for (i = 0; i < t->n_keys; i++) t->key_stack[i] = t->n_keys - 1 - i; @@ -246,7 +310,7 @@ rte_table_hash_lru_entry_add(void *table, void *key, void *entry, uint64_t sig; uint32_t bkt_index, i; - sig = t->f_hash(key, t->key_size, t->seed); + sig = t->f_hash(key, t->key_mask, t->key_size, t->seed); bkt_index = sig & t->bucket_mask; bkt = &t->buckets[bkt_index]; sig = (sig >> 16) | 1LLU; @@ -258,8 +322,8 @@ rte_table_hash_lru_entry_add(void *table, void *key, void *entry, uint8_t *bkt_key = &t->key_mem[bkt_key_index << t->key_size_shl]; - if ((sig == bkt_sig) && (memcmp(key, bkt_key, t->key_size) - == 0)) { + if ((sig == bkt_sig) && (keycmp(bkt_key, key, t->key_mask, + t->key_size) == 0)) { uint8_t *data = &t->data_mem[bkt_key_index << t->data_size_shl]; @@ -292,7 +356,7 @@ rte_table_hash_lru_entry_add(void *table, void *key, void *entry, bkt->sig[i] = (uint16_t) sig; bkt->key_pos[i] = bkt_key_index; - memcpy(bkt_key, key, t->key_size); + keycpy(bkt_key, key, t->key_mask, t->key_size); memcpy(data, entry, t->entry_size); lru_update(bkt, i); @@ -311,7 +375,7 @@ rte_table_hash_lru_entry_add(void *table, void *key, void *entry, uint8_t *data = &t->data_mem[bkt_key_index << t->data_size_shl]; bkt->sig[pos] = (uint16_t) sig; - memcpy(bkt_key, key, t->key_size); + keycpy(bkt_key, key, t->key_mask, t->key_size); memcpy(data, entry, t->entry_size); lru_update(bkt, pos); @@ -330,7 +394,7 @@ rte_table_hash_lru_entry_delete(void *table, void *key, int *key_found, uint64_t sig; uint32_t bkt_index, i; - sig = t->f_hash(key, t->key_size, t->seed); + sig = t->f_hash(key, t->key_mask, t->key_size, t->seed); bkt_index = sig & t->bucket_mask; bkt = &t->buckets[bkt_index]; sig = (sig >> 16) | 1LLU; @@ -343,14 +407,15 @@ rte_table_hash_lru_entry_delete(void *table, void *key, int *key_found, t->key_size_shl]; if ((sig == bkt_sig) && - (memcmp(key, bkt_key, t->key_size) == 0)) { + (keycmp(bkt_key, key, t->key_mask, t->key_size) == 0)) { uint8_t *data = &t->data_mem[bkt_key_index << t->data_size_shl]; bkt->sig[i] = 0; t->key_stack[t->key_stack_tos++] = bkt_key_index; *key_found = 1; - memcpy(entry, data, t->entry_size); + if (entry) + memcpy(entry, data, t->entry_size); return 0; } } @@ -365,8 +430,7 @@ static int rte_table_hash_lru_lookup_unoptimized( struct rte_mbuf **pkts, uint64_t pkts_mask, uint64_t *lookup_hit_mask, - void **entries, - int dosig) + void **entries) { struct rte_table_hash *t = (struct rte_table_hash *) table; uint64_t pkts_mask_out = 0; @@ -387,11 +451,7 @@ static int rte_table_hash_lru_lookup_unoptimized( pkt = pkts[pkt_index]; key = RTE_MBUF_METADATA_UINT8_PTR(pkt, t->key_offset); - if (dosig) - sig = (uint64_t) t->f_hash(key, t->key_size, t->seed); - else - sig = RTE_MBUF_METADATA_UINT32(pkt, - t->signature_offset); + sig = (uint64_t) t->f_hash(key, t->key_mask, t->key_size, t->seed); bkt_index = sig & t->bucket_mask; bkt = &t->buckets[bkt_index]; @@ -404,7 +464,7 @@ static int rte_table_hash_lru_lookup_unoptimized( uint8_t *bkt_key = &t->key_mem[bkt_key_index << t->key_size_shl]; - if ((sig == bkt_sig) && (memcmp(key, bkt_key, + if ((sig == bkt_sig) && (keycmp(bkt_key, key, t->key_mask, t->key_size) == 0)) { uint8_t *data = &t->data_mem[bkt_key_index << t->data_size_shl]; @@ -502,74 +562,75 @@ static int rte_table_hash_lru_lookup_unoptimized( match_pos = (LUT_MATCH_POS >> (mask_all << 1)) & 3; \ } -#define lookup_cmp_key(mbuf, key, match_key, f) \ -{ \ +#define lookup_cmp_key(mbuf, key, match_key, f) \ +{ \ uint64_t *pkt_key = RTE_MBUF_METADATA_UINT64_PTR(mbuf, f->key_offset);\ - uint64_t *bkt_key = (uint64_t *) key; \ - \ - switch (f->key_size) { \ - case 8: \ - { \ - uint64_t xor = pkt_key[0] ^ bkt_key[0]; \ - match_key = 0; \ - if (xor == 0) \ - match_key = 1; \ - } \ - break; \ - \ - case 16: \ - { \ - uint64_t xor[2], or; \ - \ - xor[0] = pkt_key[0] ^ bkt_key[0]; \ - xor[1] = pkt_key[1] ^ bkt_key[1]; \ - or = xor[0] | xor[1]; \ - match_key = 0; \ - if (or == 0) \ - match_key = 1; \ - } \ - break; \ - \ - case 32: \ - { \ - uint64_t xor[4], or; \ - \ - xor[0] = pkt_key[0] ^ bkt_key[0]; \ - xor[1] = pkt_key[1] ^ bkt_key[1]; \ - xor[2] = pkt_key[2] ^ bkt_key[2]; \ - xor[3] = pkt_key[3] ^ bkt_key[3]; \ - or = xor[0] | xor[1] | xor[2] | xor[3]; \ - match_key = 0; \ - if (or == 0) \ - match_key = 1; \ - } \ - break; \ - \ - case 64: \ - { \ - uint64_t xor[8], or; \ - \ - xor[0] = pkt_key[0] ^ bkt_key[0]; \ - xor[1] = pkt_key[1] ^ bkt_key[1]; \ - xor[2] = pkt_key[2] ^ bkt_key[2]; \ - xor[3] = pkt_key[3] ^ bkt_key[3]; \ - xor[4] = pkt_key[4] ^ bkt_key[4]; \ - xor[5] = pkt_key[5] ^ bkt_key[5]; \ - xor[6] = pkt_key[6] ^ bkt_key[6]; \ - xor[7] = pkt_key[7] ^ bkt_key[7]; \ - or = xor[0] | xor[1] | xor[2] | xor[3] | \ - xor[4] | xor[5] | xor[6] | xor[7]; \ - match_key = 0; \ - if (or == 0) \ - match_key = 1; \ - } \ - break; \ - \ - default: \ - match_key = 0; \ - if (memcmp(pkt_key, bkt_key, f->key_size) == 0) \ - match_key = 1; \ - } \ + uint64_t *bkt_key = (uint64_t *) key; \ + uint64_t *key_mask = f->key_mask; \ + \ + switch (f->key_size) { \ + case 8: \ + { \ + uint64_t xor = (pkt_key[0] & key_mask[0]) ^ bkt_key[0]; \ + match_key = 0; \ + if (xor == 0) \ + match_key = 1; \ + } \ + break; \ + \ + case 16: \ + { \ + uint64_t xor[2], or; \ + \ + xor[0] = (pkt_key[0] & key_mask[0]) ^ bkt_key[0]; \ + xor[1] = (pkt_key[1] & key_mask[1]) ^ bkt_key[1]; \ + or = xor[0] | xor[1]; \ + match_key = 0; \ + if (or == 0) \ + match_key = 1; \ + } \ + break; \ + \ + case 32: \ + { \ + uint64_t xor[4], or; \ + \ + xor[0] = (pkt_key[0] & key_mask[0]) ^ bkt_key[0]; \ + xor[1] = (pkt_key[1] & key_mask[1]) ^ bkt_key[1]; \ + xor[2] = (pkt_key[2] & key_mask[2]) ^ bkt_key[2]; \ + xor[3] = (pkt_key[3] & key_mask[3]) ^ bkt_key[3]; \ + or = xor[0] | xor[1] | xor[2] | xor[3]; \ + match_key = 0; \ + if (or == 0) \ + match_key = 1; \ + } \ + break; \ + \ + case 64: \ + { \ + uint64_t xor[8], or; \ + \ + xor[0] = (pkt_key[0] & key_mask[0]) ^ bkt_key[0]; \ + xor[1] = (pkt_key[1] & key_mask[1]) ^ bkt_key[1]; \ + xor[2] = (pkt_key[2] & key_mask[2]) ^ bkt_key[2]; \ + xor[3] = (pkt_key[3] & key_mask[3]) ^ bkt_key[3]; \ + xor[4] = (pkt_key[4] & key_mask[4]) ^ bkt_key[4]; \ + xor[5] = (pkt_key[5] & key_mask[5]) ^ bkt_key[5]; \ + xor[6] = (pkt_key[6] & key_mask[6]) ^ bkt_key[6]; \ + xor[7] = (pkt_key[7] & key_mask[7]) ^ bkt_key[7]; \ + or = xor[0] | xor[1] | xor[2] | xor[3] | \ + xor[4] | xor[5] | xor[6] | xor[7]; \ + match_key = 0; \ + if (or == 0) \ + match_key = 1; \ + } \ + break; \ + \ + default: \ + match_key = 0; \ + if (keycmp(bkt_key, pkt_key, key_mask, f->key_size) == 0) \ + match_key = 1; \ + } \ } #define lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index)\ @@ -616,38 +677,7 @@ static int rte_table_hash_lru_lookup_unoptimized( rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));\ } -#define lookup2_stage1(t, g, pkts, pkt10_index, pkt11_index) \ -{ \ - struct grinder *g10, *g11; \ - uint64_t sig10, sig11, bkt10_index, bkt11_index; \ - struct rte_mbuf *mbuf10, *mbuf11; \ - struct bucket *bkt10, *bkt11, *buckets = t->buckets; \ - uint64_t bucket_mask = t->bucket_mask; \ - uint32_t signature_offset = t->signature_offset; \ - \ - mbuf10 = pkts[pkt10_index]; \ - sig10 = (uint64_t) RTE_MBUF_METADATA_UINT32(mbuf10, signature_offset);\ - bkt10_index = sig10 & bucket_mask; \ - bkt10 = &buckets[bkt10_index]; \ - \ - mbuf11 = pkts[pkt11_index]; \ - sig11 = (uint64_t) RTE_MBUF_METADATA_UINT32(mbuf11, signature_offset);\ - bkt11_index = sig11 & bucket_mask; \ - bkt11 = &buckets[bkt11_index]; \ - \ - rte_prefetch0(bkt10); \ - rte_prefetch0(bkt11); \ - \ - g10 = &g[pkt10_index]; \ - g10->sig = sig10; \ - g10->bkt = bkt10; \ - \ - g11 = &g[pkt11_index]; \ - g11->sig = sig11; \ - g11->bkt = bkt11; \ -} - -#define lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index)\ +#define lookup2_stage1(t, g, pkts, pkt10_index, pkt11_index)\ { \ struct grinder *g10, *g11; \ uint64_t sig10, sig11, bkt10_index, bkt11_index; \ @@ -662,13 +692,13 @@ static int rte_table_hash_lru_lookup_unoptimized( \ mbuf10 = pkts[pkt10_index]; \ key10 = RTE_MBUF_METADATA_UINT8_PTR(mbuf10, key_offset);\ - sig10 = (uint64_t) f_hash(key10, key_size, seed); \ + sig10 = (uint64_t) f_hash(key10, t->key_mask, key_size, seed);\ bkt10_index = sig10 & bucket_mask; \ bkt10 = &buckets[bkt10_index]; \ \ mbuf11 = pkts[pkt11_index]; \ key11 = RTE_MBUF_METADATA_UINT8_PTR(mbuf11, key_offset);\ - sig11 = (uint64_t) f_hash(key11, key_size, seed); \ + sig11 = (uint64_t) f_hash(key11, t->key_mask, key_size, seed);\ bkt11_index = sig11 & bucket_mask; \ bkt11 = &buckets[bkt11_index]; \ \ @@ -819,7 +849,7 @@ static int rte_table_hash_lru_lookup( /* Cannot run the pipeline with less than 7 packets */ if (__builtin_popcountll(pkts_mask) < 7) return rte_table_hash_lru_lookup_unoptimized(table, pkts, - pkts_mask, lookup_hit_mask, entries, 0); + pkts_mask, lookup_hit_mask, entries); /* Pipeline stage 0 */ lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); @@ -923,140 +953,7 @@ static int rte_table_hash_lru_lookup( uint64_t pkts_mask_out_slow = 0; status = rte_table_hash_lru_lookup_unoptimized(table, pkts, - pkts_mask_match_many, &pkts_mask_out_slow, entries, 0); - pkts_mask_out |= pkts_mask_out_slow; - } - - *lookup_hit_mask = pkts_mask_out; - RTE_TABLE_HASH_LRU_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(pkts_mask_out)); - return status; -} - -static int rte_table_hash_lru_lookup_dosig( - void *table, - struct rte_mbuf **pkts, - uint64_t pkts_mask, - uint64_t *lookup_hit_mask, - void **entries) -{ - struct rte_table_hash *t = (struct rte_table_hash *) table; - struct grinder *g = t->grinders; - uint64_t pkt00_index, pkt01_index, pkt10_index, pkt11_index; - uint64_t pkt20_index, pkt21_index, pkt30_index, pkt31_index; - uint64_t pkts_mask_out = 0, pkts_mask_match_many = 0; - int status = 0; - - __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); - RTE_TABLE_HASH_LRU_STATS_PKTS_IN_ADD(t, n_pkts_in); - - /* Cannot run the pipeline with less than 7 packets */ - if (__builtin_popcountll(pkts_mask) < 7) - return rte_table_hash_lru_lookup_unoptimized(table, pkts, - pkts_mask, lookup_hit_mask, entries, 1); - - /* Pipeline stage 0 */ - lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); - - /* Pipeline feed */ - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); - - /* Pipeline feed */ - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); - - /* Pipeline stage 2 */ - lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); - - /* - * Pipeline run - * - */ - for ( ; pkts_mask; ) { - /* Pipeline feed */ - pkt30_index = pkt20_index; - pkt31_index = pkt21_index; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 0 */ - lookup2_stage0_with_odd_support(t, g, pkts, pkts_mask, - pkt00_index, pkt01_index); - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); - - /* Pipeline stage 2 */ - lookup2_stage2(t, g, pkt20_index, pkt21_index, - pkts_mask_match_many); - - /* Pipeline stage 3 */ - lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, - pkts_mask_out, entries); - } - - /* Pipeline feed */ - pkt30_index = pkt20_index; - pkt31_index = pkt21_index; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - pkt10_index = pkt00_index; - pkt11_index = pkt01_index; - - /* Pipeline stage 1 */ - lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); - - /* Pipeline stage 2 */ - lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); - - /* Pipeline stage 3 */ - lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, - entries); - - /* Pipeline feed */ - pkt30_index = pkt20_index; - pkt31_index = pkt21_index; - pkt20_index = pkt10_index; - pkt21_index = pkt11_index; - - /* Pipeline stage 2 */ - lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); - - /* Pipeline stage 3 */ - lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, - entries); - - /* Pipeline feed */ - pkt30_index = pkt20_index; - pkt31_index = pkt21_index; - - /* Pipeline stage 3 */ - lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, - entries); - - /* Slow path */ - pkts_mask_match_many &= ~pkts_mask_out; - if (pkts_mask_match_many) { - uint64_t pkts_mask_out_slow = 0; - - status = rte_table_hash_lru_lookup_unoptimized(table, pkts, - pkts_mask_match_many, &pkts_mask_out_slow, entries, 1); + pkts_mask_match_many, &pkts_mask_out_slow, entries); pkts_mask_out |= pkts_mask_out_slow; } @@ -1089,14 +986,3 @@ struct rte_table_ops rte_table_hash_lru_ops = { .f_lookup = rte_table_hash_lru_lookup, .f_stats = rte_table_hash_lru_stats_read, }; - -struct rte_table_ops rte_table_hash_lru_dosig_ops = { - .f_create = rte_table_hash_lru_create, - .f_free = rte_table_hash_lru_free, - .f_add = rte_table_hash_lru_entry_add, - .f_delete = rte_table_hash_lru_entry_delete, - .f_add_bulk = NULL, - .f_delete_bulk = NULL, - .f_lookup = rte_table_hash_lru_lookup_dosig, - .f_stats = rte_table_hash_lru_stats_read, -}; diff --git a/lib/librte_table/rte_table_version.map b/lib/librte_table/rte_table_version.map index e1eaa275..6237252b 100644 --- a/lib/librte_table/rte_table_version.map +++ b/lib/librte_table/rte_table_version.map @@ -1,19 +1,16 @@ -DPDK_2.0 { +DPDK_17.11 { global: rte_table_acl_ops; rte_table_array_ops; - rte_table_hash_ext_dosig_ops; + rte_table_hash_cuckoo_ops; rte_table_hash_ext_ops; - rte_table_hash_key8_ext_dosig_ops; - rte_table_hash_key8_ext_ops; - rte_table_hash_key8_lru_dosig_ops; - rte_table_hash_key8_lru_ops; rte_table_hash_key16_ext_ops; rte_table_hash_key16_lru_ops; rte_table_hash_key32_ext_ops; rte_table_hash_key32_lru_ops; - rte_table_hash_lru_dosig_ops; + rte_table_hash_key8_ext_ops; + rte_table_hash_key8_lru_ops; rte_table_hash_lru_ops; rte_table_lpm_ipv6_ops; rte_table_lpm_ops; @@ -21,18 +18,3 @@ DPDK_2.0 { local: *; }; - -DPDK_2.2 { - global: - - rte_table_hash_key16_ext_dosig_ops; - rte_table_hash_key16_lru_dosig_ops; - -}; - -DPDK_16.07 { - global: - - rte_table_hash_cuckoo_dosig_ops; - -} DPDK_2.0; |