From: Cristian Dumitrescu Date: Wed, 4 Jun 2014 18:08:31 +0000 (+0100) Subject: table: hash X-Git-Tag: spdx-start~10690 X-Git-Url: http://git.droids-corp.org/?a=commitdiff_plain;h=8aa327214c;p=dpdk.git table: hash Various types of hash tables presented under the Packet Framework toolbox. Hash table types: 1. Extendible bucket (ext): when bucket is full, bucket is extended with more keys 2. Least Recently Used (LRU): when bucket is full, the LRU entry is discarded 3. Pre-computed key signature: RX core extracts the key n-tuple from the packet, computes the key signature and saves the key and key signature within the packet meta-data; flow classification core performs the actual lookup (the bucket search stage) after reading the key and key signature from packet meta-data 4. Signature computed on-the-fly (do-sig version): the same CPU core extracts the key n-tuple from pkt, computes key signature and performs the table lookup 5. Configurable key size or optimized for single key size (8-byte, 16-byte and 32-byte key sizes) Signed-off-by: Cristian Dumitrescu Acked-by: Pablo de Lara Guarch Acked by: Ivan Boule --- diff --git a/doc/doxy-api-index.md b/doc/doxy-api-index.md index f1309057f1..ae5dadd73f 100644 --- a/doc/doxy-api-index.md +++ b/doc/doxy-api-index.md @@ -112,6 +112,7 @@ There are many libraries, so their headers may be grouped by topics: [lpm IPv4] (@ref rte_table_lpm.h), [lpm IPv6] (@ref rte_table_lpm_ipv6.h), [ACL] (@ref rte_table_acl.h), + [hash] (@ref rte_table_hash.h), - **basic**: [approx fraction] (@ref rte_approx.h), diff --git a/lib/librte_table/Makefile b/lib/librte_table/Makefile index 9eecb11e88..26246fb60a 100644 --- a/lib/librte_table/Makefile +++ b/lib/librte_table/Makefile @@ -47,6 +47,11 @@ SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_lpm_ipv6.c ifeq ($(CONFIG_RTE_LIBRTE_ACL),y) SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_acl.c endif +SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_hash_key8.c +SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_hash_key16.c +SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_hash_key32.c +SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_hash_ext.c +SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_hash_lru.c # install includes SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_table.h @@ -55,6 +60,8 @@ SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_table_lpm_ipv6.h ifeq ($(CONFIG_RTE_LIBRTE_ACL),y) SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_table_acl.h endif +SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_table_hash.h +SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_lru.h # this lib depends upon: DEPDIRS-$(CONFIG_RTE_LIBRTE_TABLE) := lib/librte_eal @@ -66,5 +73,6 @@ DEPDIRS-$(CONFIG_RTE_LIBRTE_TABLE) += lib/librte_lpm ifeq ($(CONFIG_RTE_LIBRTE_ACL),y) DEPDIRS-$(CONFIG_RTE_LIBRTE_TABLE) += lib/librte_acl endif +DEPDIRS-$(CONFIG_RTE_LIBRTE_TABLE) += lib/librte_hash include $(RTE_SDK)/mk/rte.lib.mk diff --git a/lib/librte_table/rte_lru.h b/lib/librte_table/rte_lru.h new file mode 100644 index 0000000000..e87e062d00 --- /dev/null +++ b/lib/librte_table/rte_lru.h @@ -0,0 +1,213 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 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: + * + * * 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. + */ + +#ifndef __INCLUDE_RTE_LRU_H__ +#define __INCLUDE_RTE_LRU_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +#include + +#ifdef __INTEL_COMPILER +#define GCC_VERSION (0) +#else +#define GCC_VERSION (__GNUC__ * 10000+__GNUC_MINOR__*100 + __GNUC_PATCHLEVEL__) +#endif + +#ifndef RTE_TABLE_HASH_LRU_STRATEGY +#ifdef __SSE4_2__ +#define RTE_TABLE_HASH_LRU_STRATEGY 2 +#else /* if no SSE, use simple scalar version */ +#define RTE_TABLE_HASH_LRU_STRATEGY 1 +#endif +#endif + +#ifndef RTE_ARCH_X86_64 +#undef RTE_TABLE_HASH_LRU_STRATEGY +#define RTE_TABLE_HASH_LRU_STRATEGY 1 +#endif + +#if (RTE_TABLE_HASH_LRU_STRATEGY < 0) || (RTE_TABLE_HASH_LRU_STRATEGY > 3) +#error Invalid value for RTE_TABLE_HASH_LRU_STRATEGY +#endif + +#if RTE_TABLE_HASH_LRU_STRATEGY == 0 + +#define lru_init(bucket) \ +do \ + bucket = bucket; \ +while (0) + +#define lru_pos(bucket) (bucket->lru_list & 0xFFFFLLU) + +#define lru_update(bucket, mru_val) \ +do { \ + bucket = bucket; \ + mru_val = mru_val; \ +} while (0) + +#elif RTE_TABLE_HASH_LRU_STRATEGY == 1 + +#define lru_init(bucket) \ +do \ + bucket->lru_list = 0x0000000100020003LLU; \ +while (0) + +#define lru_pos(bucket) (bucket->lru_list & 0xFFFFLLU) + +#define lru_update(bucket, mru_val) \ +do { \ + uint64_t x, pos, x0, x1, x2, mask; \ + \ + x = bucket->lru_list; \ + \ + pos = 4; \ + if ((x >> 48) == ((uint64_t) mru_val)) \ + pos = 3; \ + \ + if (((x >> 32) & 0xFFFFLLU) == ((uint64_t) mru_val)) \ + pos = 2; \ + \ + if (((x >> 16) & 0xFFFFLLU) == ((uint64_t) mru_val)) \ + pos = 1; \ + \ + if ((x & 0xFFFFLLU) == ((uint64_t) mru_val)) \ + pos = 0; \ + \ + \ + pos <<= 4; \ + mask = (~0LLU) << pos; \ + x0 = x & (~mask); \ + x1 = (x >> 16) & mask; \ + x2 = (x << (48 - pos)) & (0xFFFFLLU << 48); \ + x = x0 | x1 | x2; \ + \ + if (pos != 64) \ + bucket->lru_list = x; \ +} while (0) + +#elif RTE_TABLE_HASH_LRU_STRATEGY == 2 + +#if GCC_VERSION > 40306 +#include +#else +#include +#include +#include +#endif + +#define lru_init(bucket) \ +do \ + bucket->lru_list = 0x0000000100020003LLU; \ +while (0) + +#define lru_pos(bucket) (bucket->lru_list & 0xFFFFLLU) + +#define lru_update(bucket, mru_val) \ +do { \ + /* set up the masks for all possible shuffles, depends on pos */\ + static uint64_t masks[10] = { \ + /* Shuffle order; Make Zero (see _mm_shuffle_epi8 manual) */\ + 0x0100070605040302, 0x8080808080808080, \ + 0x0302070605040100, 0x8080808080808080, \ + 0x0504070603020100, 0x8080808080808080, \ + 0x0706050403020100, 0x8080808080808080, \ + 0x0706050403020100, 0x8080808080808080}; \ + /* load up one register with repeats of mru-val */ \ + uint64_t mru2 = mru_val; \ + uint64_t mru3 = mru2 | (mru2 << 16); \ + uint64_t lru = bucket->lru_list; \ + /* XOR to cause the word we're looking for to go to zero */ \ + uint64_t mru = lru ^ ((mru3 << 32) | mru3); \ + __m128i c = _mm_cvtsi64_si128(mru); \ + __m128i b = _mm_cvtsi64_si128(lru); \ + /* Find the minimum value (first zero word, if it's in there) */\ + __m128i d = _mm_minpos_epu16(c); \ + /* Second word is the index to found word (first word is the value) */\ + unsigned pos = _mm_extract_epi16(d, 1); \ + /* move the recently used location to top of list */ \ + __m128i k = _mm_shuffle_epi8(b, *((__m128i *) &masks[2 * pos]));\ + /* Finally, update the original list with the reordered data */ \ + bucket->lru_list = _mm_extract_epi64(k, 0); \ + /* Phwew! */ \ +} while (0) + +#elif RTE_TABLE_HASH_LRU_STRATEGY == 3 + +#if GCC_VERSION > 40306 +#include +#else +#include +#include +#include +#endif + +#define lru_init(bucket) \ +do \ + bucket->lru_list = ~0LLU; \ +while (0) + + +static inline int +f_lru_pos(uint64_t lru_list) +{ + __m128i lst = _mm_set_epi64x((uint64_t)-1, lru_list); + __m128i min = _mm_minpos_epu16(lst); + return _mm_extract_epi16(min, 1); +} +#define lru_pos(bucket) f_lru_pos(bucket->lru_list) + +#define lru_update(bucket, mru_val) \ +do { \ + const uint64_t orvals[] = {0xFFFFLLU, 0xFFFFLLU << 16, \ + 0xFFFFLLU << 32, 0xFFFFLLU << 48, 0LLU}; \ + const uint64_t decs[] = {0x1000100010001LLU, 0}; \ + __m128i lru = _mm_cvtsi64_si128(bucket->lru_list); \ + __m128i vdec = _mm_cvtsi64_si128(decs[mru_val>>2]); \ + lru = _mm_subs_epu16(lru, vdec); \ + bucket->lru_list = _mm_extract_epi64(lru, 0) | orvals[mru_val]; \ +} while (0) + +#else + +#error "Incorrect value for RTE_TABLE_HASH_LRU_STRATEGY" + +#endif + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/lib/librte_table/rte_table_hash.h b/lib/librte_table/rte_table_hash.h new file mode 100644 index 0000000000..9181942a7b --- /dev/null +++ b/lib/librte_table/rte_table_hash.h @@ -0,0 +1,350 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 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: + * + * * 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. + */ + +#ifndef __INCLUDE_RTE_TABLE_HASH_H__ +#define __INCLUDE_RTE_TABLE_HASH_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Table Hash + * + * 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 + * deleted and the new key is added in its place. The number of keys in + * each bucket never grows bigger than 4. The logic to pick the key to + * be dropped from the bucket is LRU. The hash table lookup operation + * maintains the order in which the keys in the same bucket are hit, so + * every time a key is hit, it becomes the new Most Recently Used (MRU) + * key, i.e. the most unlikely candidate for drop. When a key is added + * 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. + * 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 + * configurable and always a multiple of 4). On key add operation, the + * allocation of a group of 4 keys only happens successfully within the + * limit of free keys, otherwise the key add operation fails. On key + * delete operation, a group of 4 keys is freed back to the pool of free + * keys when the key to be deleted is the only key that was used within + * its group of 4 keys at that time. On key lookup operation, if the + * current bucket is in extended state and a match is not found in the + * 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: + * a. Configurable key size + * b. Single key size (8-byte, 16-byte or 32-byte key size) + * + ***/ +#include + +#include "rte_table.h" + +/** Hash function */ +typedef uint64_t (*rte_table_hash_op_hash)( + void *key, + uint32_t key_size, + uint64_t seed); + +/** + * Hash tables with configurable key size + * + */ +/** Extendible bucket hash table parameters */ +struct rte_table_hash_ext_params { + /** 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; + + /** Maximum number of keys */ + uint32_t n_keys; + + /** Number of hash table buckets. Each bucket stores up to 4 keys. */ + uint32_t n_buckets; + + /** 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; +}; + +/** 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; +}; + +/** 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; +}; + +/** Extendible bucket hash table operations for pre-computed key signature */ +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; +}; + +/** LRU hash table operations for pre-computed key signature */ +extern struct rte_table_ops rte_table_hash_key16_lru_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; +}; + +/** Extendible bucket operations for pre-computed key signature */ +extern struct rte_table_ops rte_table_hash_key16_ext_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; + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/lib/librte_table/rte_table_hash_ext.c b/lib/librte_table/rte_table_hash_ext.c new file mode 100644 index 0000000000..6e26d9844b --- /dev/null +++ b/lib/librte_table/rte_table_hash_ext.c @@ -0,0 +1,1122 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 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: + * + * * 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. + */ + +#include +#include + +#include +#include +#include +#include + +#include "rte_table_hash.h" + +#define KEYS_PER_BUCKET 4 + +struct bucket { + union { + uintptr_t next; + uint64_t lru_list; + }; + uint16_t sig[KEYS_PER_BUCKET]; + uint32_t key_pos[KEYS_PER_BUCKET]; +}; + +#define BUCKET_NEXT(bucket) \ + ((void *) ((bucket)->next & (~1LU))) + +#define BUCKET_NEXT_VALID(bucket) \ + ((bucket)->next & 1LU) + +#define BUCKET_NEXT_SET(bucket, bucket_next) \ +do \ + (bucket)->next = (((uintptr_t) ((void *) (bucket_next))) | 1LU);\ +while (0) + +#define BUCKET_NEXT_SET_NULL(bucket) \ +do \ + (bucket)->next = 0; \ +while (0) + +#define BUCKET_NEXT_COPY(bucket, bucket2) \ +do \ + (bucket)->next = (bucket2)->next; \ +while (0) + +struct grinder { + struct bucket *bkt; + uint64_t sig; + uint64_t match; + uint32_t key_index; +}; + +struct rte_table_hash { + /* Input parameters */ + uint32_t key_size; + uint32_t entry_size; + uint32_t n_keys; + uint32_t n_buckets; + uint32_t n_buckets_ext; + rte_table_hash_op_hash f_hash; + uint64_t seed; + uint32_t signature_offset; + uint32_t key_offset; + + /* Internal */ + uint64_t bucket_mask; + uint32_t key_size_shl; + uint32_t data_size_shl; + uint32_t key_stack_tos; + uint32_t bkt_ext_stack_tos; + + /* Grinder */ + struct grinder grinders[RTE_PORT_IN_BURST_SIZE_MAX]; + + /* Tables */ + struct bucket *buckets; + struct bucket *buckets_ext; + uint8_t *key_mem; + uint8_t *data_mem; + uint32_t *key_stack; + uint32_t *bkt_ext_stack; + + /* Table memory */ + uint8_t memory[0] __rte_cache_aligned; +}; + +static int +check_params_create(struct rte_table_hash_ext_params *params) +{ + uint32_t n_buckets_min; + + /* key_size */ + if ((params->key_size == 0) || + (!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))) { + 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_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 invalid value\n", __func__); + return -EINVAL; + } + + /* signature offset */ + if ((params->signature_offset & 0x3) != 0) { + RTE_LOG(ERR, TABLE, "%s: signature_offset invalid value\n", + __func__); + return -EINVAL; + } + + /* key offset */ + if ((params->key_offset & 0x7) != 0) { + RTE_LOG(ERR, TABLE, "%s: key_offset invalid value\n", __func__); + return -EINVAL; + } + + return 0; +} + +static void * +rte_table_hash_ext_create(void *params, int socket_id, uint32_t entry_size) +{ + struct rte_table_hash_ext_params *p = + (struct rte_table_hash_ext_params *) params; + struct rte_table_hash *t; + uint32_t total_size, table_meta_sz, table_meta_offset; + 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; + + /* Check input parameters */ + if ((check_params_create(p) != 0) || + (!rte_is_power_of_2(entry_size)) || + ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) || + (sizeof(struct bucket) != (CACHE_LINE_SIZE / 2))) + return NULL; + + /* Memory allocation */ + table_meta_sz = CACHE_LINE_ROUNDUP(sizeof(struct rte_table_hash)); + bucket_sz = CACHE_LINE_ROUNDUP(p->n_buckets * sizeof(struct bucket)); + bucket_ext_sz = + CACHE_LINE_ROUNDUP(p->n_buckets_ext * sizeof(struct bucket)); + key_sz = CACHE_LINE_ROUNDUP(p->n_keys * p->key_size); + key_stack_sz = CACHE_LINE_ROUNDUP(p->n_keys * sizeof(uint32_t)); + bkt_ext_stack_sz = + CACHE_LINE_ROUNDUP(p->n_buckets_ext * sizeof(uint32_t)); + data_sz = 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; + + t = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id); + if (t == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + return NULL; + } + RTE_LOG(INFO, TABLE, "%s (%u-byte key): Hash table memory footprint is " + "%u bytes\n", __func__, p->key_size, 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->f_hash = p->f_hash; + t->seed = p->seed; + t->signature_offset = p->signature_offset; + t->key_offset = p->key_offset; + + /* Internal */ + t->bucket_mask = t->n_buckets - 1; + t->key_size_shl = __builtin_ctzl(p->key_size); + t->data_size_shl = __builtin_ctzl(p->key_size); + + /* Tables */ + table_meta_offset = 0; + bucket_offset = table_meta_offset + table_meta_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->buckets = (struct bucket *) &t->memory[bucket_offset]; + t->buckets_ext = (struct bucket *) &t->memory[bucket_ext_offset]; + t->key_mem = &t->memory[key_offset]; + t->key_stack = (uint32_t *) &t->memory[key_stack_offset]; + t->bkt_ext_stack = (uint32_t *) &t->memory[bkt_ext_stack_offset]; + t->data_mem = &t->memory[data_offset]; + + /* Key stack */ + for (i = 0; i < t->n_keys; i++) + t->key_stack[i] = t->n_keys - 1 - i; + t->key_stack_tos = t->n_keys; + + /* Bucket ext stack */ + for (i = 0; i < t->n_buckets_ext; i++) + t->bkt_ext_stack[i] = t->n_buckets_ext - 1 - i; + t->bkt_ext_stack_tos = t->n_buckets_ext; + + return t; +} + +static int +rte_table_hash_ext_free(void *table) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + + /* Check input parameters */ + if (t == NULL) + return -EINVAL; + + rte_free(t); + return 0; +} + +static int +rte_table_hash_ext_entry_add(void *table, void *key, void *entry, + int *key_found, void **entry_ptr) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + struct bucket *bkt0, *bkt, *bkt_prev; + uint64_t sig; + uint32_t bkt_index, i; + + sig = t->f_hash(key, t->key_size, t->seed); + bkt_index = sig & t->bucket_mask; + bkt0 = &t->buckets[bkt_index]; + sig = (sig >> 16) | 1LLU; + + /* Key is present in the bucket */ + for (bkt = bkt0; bkt != NULL; bkt = BUCKET_NEXT(bkt)) + for (i = 0; i < KEYS_PER_BUCKET; i++) { + uint64_t bkt_sig = (uint64_t) bkt->sig[i]; + uint32_t bkt_key_index = bkt->key_pos[i]; + 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)) { + uint8_t *data = &t->data_mem[bkt_key_index << + t->data_size_shl]; + + memcpy(data, entry, t->entry_size); + *key_found = 1; + *entry_ptr = (void *) data; + return 0; + } + } + + /* Key is not present in the bucket */ + for (bkt_prev = NULL, bkt = bkt0; bkt != NULL; bkt_prev = bkt, + bkt = BUCKET_NEXT(bkt)) + for (i = 0; i < KEYS_PER_BUCKET; i++) { + uint64_t bkt_sig = (uint64_t) bkt->sig[i]; + + if (bkt_sig == 0) { + uint32_t bkt_key_index; + uint8_t *bkt_key, *data; + + /* Allocate new key */ + if (t->key_stack_tos == 0) /* No free keys */ + return -ENOSPC; + + bkt_key_index = t->key_stack[ + --t->key_stack_tos]; + + /* Install new key */ + bkt_key = &t->key_mem[bkt_key_index << + t->key_size_shl]; + data = &t->data_mem[bkt_key_index << + t->data_size_shl]; + + bkt->sig[i] = (uint16_t) sig; + bkt->key_pos[i] = bkt_key_index; + memcpy(bkt_key, key, t->key_size); + memcpy(data, entry, t->entry_size); + + *key_found = 0; + *entry_ptr = (void *) data; + return 0; + } + } + + /* Bucket full: extend bucket */ + if ((t->bkt_ext_stack_tos > 0) && (t->key_stack_tos > 0)) { + uint32_t bkt_key_index; + uint8_t *bkt_key, *data; + + /* Allocate new bucket ext */ + bkt_index = t->bkt_ext_stack[--t->bkt_ext_stack_tos]; + bkt = &t->buckets_ext[bkt_index]; + + /* Chain the new bucket ext */ + BUCKET_NEXT_SET(bkt_prev, bkt); + BUCKET_NEXT_SET_NULL(bkt); + + /* Allocate new key */ + bkt_key_index = t->key_stack[--t->key_stack_tos]; + bkt_key = &t->key_mem[bkt_key_index << t->key_size_shl]; + + data = &t->data_mem[bkt_key_index << t->data_size_shl]; + + /* 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); + memcpy(data, entry, t->entry_size); + + *key_found = 0; + *entry_ptr = (void *) data; + return 0; + } + + return -ENOSPC; +} + +static int +rte_table_hash_ext_entry_delete(void *table, void *key, int *key_found, +void *entry) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + struct bucket *bkt0, *bkt, *bkt_prev; + uint64_t sig; + uint32_t bkt_index, i; + + sig = t->f_hash(key, t->key_size, t->seed); + bkt_index = sig & t->bucket_mask; + bkt0 = &t->buckets[bkt_index]; + sig = (sig >> 16) | 1LLU; + + /* Key is present in the bucket */ + for (bkt_prev = NULL, bkt = bkt0; bkt != NULL; bkt_prev = bkt, + bkt = BUCKET_NEXT(bkt)) + for (i = 0; i < KEYS_PER_BUCKET; i++) { + uint64_t bkt_sig = (uint64_t) bkt->sig[i]; + uint32_t bkt_key_index = bkt->key_pos[i]; + 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)) { + uint8_t *data = &t->data_mem[bkt_key_index << + t->data_size_shl]; + + /* Uninstall key from bucket */ + bkt->sig[i] = 0; + *key_found = 1; + if (entry) + memcpy(entry, data, t->entry_size); + + /* Free key */ + t->key_stack[t->key_stack_tos++] = + bkt_key_index; + + /*Check if bucket is unused */ + if ((bkt_prev != NULL) && + (bkt->sig[0] == 0) && (bkt->sig[1] == 0) && + (bkt->sig[2] == 0) && (bkt->sig[3] == 0)) { + /* Clear bucket */ + memset(bkt, 0, sizeof(struct bucket)); + + /* Unchain bucket */ + BUCKET_NEXT_COPY(bkt_prev, bkt); + + /* Free bucket back to buckets ext */ + bkt_index = bkt - t->buckets_ext; + t->bkt_ext_stack[t->bkt_ext_stack_tos++] + = bkt_index; + } + + return 0; + } + } + + /* Key is not present in the bucket */ + *key_found = 0; + return 0; +} + +static int rte_table_hash_ext_lookup_unoptimized( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries, + int dosig) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + uint64_t pkts_mask_out = 0; + + for ( ; pkts_mask; ) { + struct bucket *bkt0, *bkt; + struct rte_mbuf *pkt; + uint8_t *key; + uint64_t pkt_mask, sig; + uint32_t pkt_index, bkt_index, i; + + pkt_index = __builtin_ctzll(pkts_mask); + pkt_mask = 1LLU << pkt_index; + pkts_mask &= ~pkt_mask; + + 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); + + bkt_index = sig & t->bucket_mask; + bkt0 = &t->buckets[bkt_index]; + sig = (sig >> 16) | 1LLU; + + /* Key is present in the bucket */ + for (bkt = bkt0; bkt != NULL; bkt = BUCKET_NEXT(bkt)) + for (i = 0; i < KEYS_PER_BUCKET; i++) { + uint64_t bkt_sig = (uint64_t) bkt->sig[i]; + uint32_t bkt_key_index = bkt->key_pos[i]; + 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)) { + uint8_t *data = &t->data_mem[ + bkt_key_index << t->data_size_shl]; + + pkts_mask_out |= pkt_mask; + entries[pkt_index] = (void *) data; + break; + } + } + } + + *lookup_hit_mask = pkts_mask_out; + return 0; +} + +/*** + * + * mask = match bitmask + * match = at least one match + * match_many = more than one match + * match_pos = position of first match + * + *---------------------------------------- + * mask match match_many match_pos + *---------------------------------------- + * 0000 0 0 00 + * 0001 1 0 00 + * 0010 1 0 01 + * 0011 1 1 00 + *---------------------------------------- + * 0100 1 0 10 + * 0101 1 1 00 + * 0110 1 1 01 + * 0111 1 1 00 + *---------------------------------------- + * 1000 1 0 11 + * 1001 1 1 00 + * 1010 1 1 01 + * 1011 1 1 00 + *---------------------------------------- + * 1100 1 1 10 + * 1101 1 1 00 + * 1110 1 1 01 + * 1111 1 1 00 + *---------------------------------------- + * + * match = 1111_1111_1111_1110 + * match_many = 1111_1110_1110_1000 + * match_pos = 0001_0010_0001_0011__0001_0010_0001_0000 + * + * match = 0xFFFELLU + * match_many = 0xFEE8LLU + * match_pos = 0x12131210LLU + * + ***/ + +#define LUT_MATCH 0xFFFELLU +#define LUT_MATCH_MANY 0xFEE8LLU +#define LUT_MATCH_POS 0x12131210LLU + +#define lookup_cmp_sig(mbuf_sig, bucket, match, match_many, match_pos) \ +{ \ + uint64_t bucket_sig[4], mask[4], mask_all; \ + \ + bucket_sig[0] = bucket->sig[0]; \ + bucket_sig[1] = bucket->sig[1]; \ + bucket_sig[2] = bucket->sig[2]; \ + bucket_sig[3] = bucket->sig[3]; \ + \ + bucket_sig[0] ^= mbuf_sig; \ + bucket_sig[1] ^= mbuf_sig; \ + bucket_sig[2] ^= mbuf_sig; \ + bucket_sig[3] ^= mbuf_sig; \ + \ + mask[0] = 0; \ + mask[1] = 0; \ + mask[2] = 0; \ + mask[3] = 0; \ + \ + if (bucket_sig[0] == 0) \ + mask[0] = 1; \ + if (bucket_sig[1] == 0) \ + mask[1] = 2; \ + if (bucket_sig[2] == 0) \ + mask[2] = 4; \ + if (bucket_sig[3] == 0) \ + mask[3] = 8; \ + \ + mask_all = (mask[0] | mask[1]) | (mask[2] | mask[3]); \ + \ + match = (LUT_MATCH >> mask_all) & 1; \ + match_many = (LUT_MATCH_MANY >> mask_all) & 1; \ + match_pos = (LUT_MATCH_POS >> (mask_all << 1)) & 3; \ +} + +#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; \ + } \ +} + +#define lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + struct rte_mbuf *mbuf00, *mbuf01; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + mbuf00 = pkts[pkt00_index]; \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + mbuf01 = pkts[pkt01_index]; \ + \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#define lookup2_stage0_with_odd_support(t, g, pkts, pkts_mask, pkt00_index, \ + pkt01_index) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + struct rte_mbuf *mbuf00, *mbuf01; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + mbuf00 = pkts[pkt00_index]; \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + if (pkts_mask == 0) \ + pkt01_index = pkt00_index; \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + mbuf01 = pkts[pkt01_index]; \ + \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#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) \ +{ \ + struct grinder *g10, *g11; \ + uint64_t sig10, sig11, bkt10_index, bkt11_index; \ + struct rte_mbuf *mbuf10, *mbuf11; \ + struct bucket *bkt10, *bkt11, *buckets = t->buckets; \ + uint8_t *key10, *key11; \ + uint64_t bucket_mask = t->bucket_mask; \ + rte_table_hash_op_hash f_hash = t->f_hash; \ + uint64_t seed = t->seed; \ + uint32_t key_size = t->key_size; \ + uint32_t key_offset = t->key_offset; \ + \ + mbuf10 = pkts[pkt10_index]; \ + key10 = RTE_MBUF_METADATA_UINT8_PTR(mbuf10, key_offset); \ + sig10 = (uint64_t) f_hash(key10, 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); \ + 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_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many)\ +{ \ + struct grinder *g20, *g21; \ + uint64_t sig20, sig21; \ + struct bucket *bkt20, *bkt21; \ + uint8_t *key20, *key21, *key_mem = t->key_mem; \ + uint64_t match20, match21, match_many20, match_many21; \ + uint64_t match_pos20, match_pos21; \ + uint32_t key20_index, key21_index, key_size_shl = t->key_size_shl;\ + \ + g20 = &g[pkt20_index]; \ + sig20 = g20->sig; \ + bkt20 = g20->bkt; \ + sig20 = (sig20 >> 16) | 1LLU; \ + lookup_cmp_sig(sig20, bkt20, match20, match_many20, match_pos20);\ + match20 <<= pkt20_index; \ + match_many20 |= BUCKET_NEXT_VALID(bkt20); \ + match_many20 <<= pkt20_index; \ + key20_index = bkt20->key_pos[match_pos20]; \ + key20 = &key_mem[key20_index << key_size_shl]; \ + \ + g21 = &g[pkt21_index]; \ + sig21 = g21->sig; \ + bkt21 = g21->bkt; \ + sig21 = (sig21 >> 16) | 1LLU; \ + lookup_cmp_sig(sig21, bkt21, match21, match_many21, match_pos21);\ + match21 <<= pkt21_index; \ + match_many21 |= BUCKET_NEXT_VALID(bkt21); \ + match_many21 <<= pkt21_index; \ + key21_index = bkt21->key_pos[match_pos21]; \ + key21 = &key_mem[key21_index << key_size_shl]; \ + \ + rte_prefetch0(key20); \ + rte_prefetch0(key21); \ + \ + pkts_mask_match_many |= match_many20 | match_many21; \ + \ + g20->match = match20; \ + g20->key_index = key20_index; \ + \ + g21->match = match21; \ + g21->key_index = key21_index; \ +} + +#define lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, \ + entries) \ +{ \ + struct grinder *g30, *g31; \ + struct rte_mbuf *mbuf30, *mbuf31; \ + uint8_t *key30, *key31, *key_mem = t->key_mem; \ + uint8_t *data30, *data31, *data_mem = t->data_mem; \ + uint64_t match30, match31, match_key30, match_key31, match_keys;\ + uint32_t key30_index, key31_index; \ + uint32_t key_size_shl = t->key_size_shl; \ + uint32_t data_size_shl = t->data_size_shl; \ + \ + mbuf30 = pkts[pkt30_index]; \ + g30 = &g[pkt30_index]; \ + match30 = g30->match; \ + key30_index = g30->key_index; \ + key30 = &key_mem[key30_index << key_size_shl]; \ + lookup_cmp_key(mbuf30, key30, match_key30, t); \ + match_key30 <<= pkt30_index; \ + match_key30 &= match30; \ + data30 = &data_mem[key30_index << data_size_shl]; \ + entries[pkt30_index] = data30; \ + \ + mbuf31 = pkts[pkt31_index]; \ + g31 = &g[pkt31_index]; \ + match31 = g31->match; \ + key31_index = g31->key_index; \ + key31 = &key_mem[key31_index << key_size_shl]; \ + lookup_cmp_key(mbuf31, key31, match_key31, t); \ + match_key31 <<= pkt31_index; \ + match_key31 &= match31; \ + data31 = &data_mem[key31_index << data_size_shl]; \ + entries[pkt31_index] = data31; \ + \ + rte_prefetch0(data30); \ + rte_prefetch0(data31); \ + \ + match_keys = match_key30 | match_key31; \ + pkts_mask_out |= match_keys; \ +} + +/*** +* The lookup function implements a 4-stage pipeline, with each stage processing +* two different packets. The purpose of pipelined implementation is to hide the +* latency of prefetching the data structures and loosen the data dependency +* between instructions. +* +* p00 _______ p10 _______ p20 _______ p30 _______ +*----->| |----->| |----->| |----->| |-----> +* | 0 | | 1 | | 2 | | 3 | +*----->|_______|----->|_______|----->|_______|----->|_______|-----> +* p01 p11 p21 p31 +* +* The naming convention is: +* pXY = packet Y of stage X, X = 0 .. 3, Y = 0 .. 1 +* +***/ +static int rte_table_hash_ext_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 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; + + /* Cannot run the pipeline with less than 7 packets */ + if (__builtin_popcountll(pkts_mask) < 7) + return rte_table_hash_ext_lookup_unoptimized(table, pkts, + pkts_mask, lookup_hit_mask, entries, 0); + + /* 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(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(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(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(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, 0); + pkts_mask_out |= pkts_mask_out_slow; + } + + *lookup_hit_mask = 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; + + /* Cannot run the pipeline with less than 7 packets */ + if (__builtin_popcountll(pkts_mask) < 7) + return rte_table_hash_ext_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_ext_lookup_unoptimized(table, pkts, + pkts_mask_match_many, &pkts_mask_out_slow, entries, 1); + pkts_mask_out |= pkts_mask_out_slow; + } + + *lookup_hit_mask = pkts_mask_out; + return status; +} + +struct rte_table_ops rte_table_hash_ext_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_lookup = rte_table_hash_ext_lookup, +}; + +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_lookup = rte_table_hash_ext_lookup_dosig, +}; diff --git a/lib/librte_table/rte_table_hash_key16.c b/lib/librte_table/rte_table_hash_key16.c new file mode 100644 index 0000000000..f5ec87d4de --- /dev/null +++ b/lib/librte_table/rte_table_hash_key16.c @@ -0,0 +1,1100 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 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: + * + * * 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. + */ +#include +#include + +#include +#include +#include +#include + +#include "rte_table_hash.h" +#include "rte_lru.h" + +#define RTE_TABLE_HASH_KEY_SIZE 16 + +#define RTE_BUCKET_ENTRY_VALID 0x1LLU + +struct rte_bucket_4_16 { + /* Cache line 0 */ + uint64_t signature[4 + 1]; + uint64_t lru_list; + struct rte_bucket_4_16 *next; + uint64_t next_valid; + + /* Cache line 1 */ + uint64_t key[4][2]; + + /* Cache line 2 */ + uint8_t data[0]; +}; + +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; + rte_table_hash_op_hash f_hash; + uint64_t seed; + + /* Extendible buckets */ + uint32_t n_buckets_ext; + uint32_t stack_pos; + uint32_t *stack; + + /* Lookup table */ + uint8_t memory[0] __rte_cache_aligned; +}; + +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__); + return -EINVAL; + } + + /* signature offset */ + if ((params->signature_offset & 0x3) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid signature_offset\n", __func__); + return -EINVAL; + } + + /* key offset */ + if ((params->key_offset & 0x7) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid key_offset\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_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 *f; + uint32_t n_buckets, n_entries_per_bucket, + key_size, bucket_size_cl, total_size, i; + + /* Check input parameters */ + if ((check_params_create_lru(p) != 0) || + ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_16) % CACHE_LINE_SIZE) != 0)) + return NULL; + n_entries_per_bucket = 4; + key_size = 16; + + /* 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 + CACHE_LINE_SIZE - 1) / CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + n_buckets * + bucket_size_cl * CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id); + if (f == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + return NULL; + } + RTE_LOG(INFO, TABLE, + "%s: Hash table memory footprint is %u bytes\n", + __func__, total_size); + + /* Memory initialization */ + f->n_buckets = n_buckets; + f->n_entries_per_bucket = n_entries_per_bucket; + f->key_size = key_size; + f->entry_size = entry_size; + f->bucket_size = bucket_size_cl * CACHE_LINE_SIZE; + f->signature_offset = p->signature_offset; + f->key_offset = p->key_offset; + f->f_hash = p->f_hash; + f->seed = p->seed; + + for (i = 0; i < n_buckets; i++) { + struct rte_bucket_4_16 *bucket; + + bucket = (struct rte_bucket_4_16 *) &f->memory[i * + f->bucket_size]; + lru_init(bucket); + } + + return f; +} + +static int +rte_table_hash_free_key16_lru(void *table) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + + /* Check input parameters */ + if (f == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + rte_free(f); + return 0; +} + +static int +rte_table_hash_entry_add_key16_lru( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_16 *bucket; + uint64_t signature, pos; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + 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]; + + if ((bucket_signature == signature) && + (memcmp(key, bucket_key, f->key_size) == 0)) { + uint8_t *bucket_data = &bucket->data[i * f->entry_size]; + + memcpy(bucket_data, entry, f->entry_size); + lru_update(bucket, i); + *key_found = 1; + *entry_ptr = (void *) bucket_data; + return 0; + } + } + + /* 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]; + + 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); + memcpy(bucket_data, entry, f->entry_size); + lru_update(bucket, i); + *key_found = 0; + *entry_ptr = (void *) bucket_data; + + return 0; + } + } + + /* Bucket full: replace LRU entry */ + pos = lru_pos(bucket); + bucket->signature[pos] = signature; + memcpy(bucket->key[pos], key, f->key_size); + memcpy(&bucket->data[pos * f->entry_size], entry, f->entry_size); + lru_update(bucket, pos); + *key_found = 0; + *entry_ptr = (void *) &bucket->data[pos * f->entry_size]; + + return 0; +} + +static int +rte_table_hash_entry_delete_key16_lru( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_16 *bucket; + uint64_t signature; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + 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]; + + if ((bucket_signature == signature) && + (memcmp(key, bucket_key, f->key_size) == 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); + return 0; + } + } + + /* Key is not present in the bucket */ + *key_found = 0; + 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; + } + + /* signature offset */ + if ((params->signature_offset & 0x3) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid signature offset\n", __func__); + return -EINVAL; + } + + /* key offset */ + if ((params->key_offset & 0x7) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid key offset\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 *f; + uint32_t n_buckets, n_buckets_ext, n_entries_per_bucket, key_size, + bucket_size_cl, stack_size_cl, total_size, i; + + /* Check input parameters */ + if ((check_params_create_ext(p) != 0) || + ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_16) % CACHE_LINE_SIZE) != 0)) + return NULL; + + n_entries_per_bucket = 4; + key_size = 16; + + /* 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 + CACHE_LINE_SIZE - 1) / CACHE_LINE_SIZE; + stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + CACHE_LINE_SIZE - 1) + / CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + + ((n_buckets + n_buckets_ext) * bucket_size_cl + stack_size_cl) * + CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id); + if (f == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + return NULL; + } + RTE_LOG(INFO, TABLE, + "%s: Hash table memory footprint is %u bytes\n", + __func__, total_size); + + /* Memory initialization */ + f->n_buckets = n_buckets; + f->n_entries_per_bucket = n_entries_per_bucket; + f->key_size = key_size; + f->entry_size = entry_size; + f->bucket_size = bucket_size_cl * CACHE_LINE_SIZE; + f->signature_offset = p->signature_offset; + f->key_offset = p->key_offset; + f->f_hash = p->f_hash; + f->seed = p->seed; + + 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; + + return f; +} + +static int +rte_table_hash_free_key16_ext(void *table) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + + /* Check input parameters */ + if (f == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + rte_free(f); + return 0; +} + +static int +rte_table_hash_entry_add_key16_ext( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_16 *bucket0, *bucket, *bucket_prev; + uint64_t signature; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + 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]; + + if ((bucket_signature == signature) && + (memcmp(key, bucket_key, f->key_size) == 0)) { + uint8_t *bucket_data = &bucket->data[i * + f->entry_size]; + + memcpy(bucket_data, entry, f->entry_size); + *key_found = 1; + *entry_ptr = (void *) bucket_data; + return 0; + } + } + + /* Key is not present in the bucket */ + for (bucket_prev = NULL, bucket = bucket0; bucket != NULL; + 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]; + + 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); + memcpy(bucket_data, entry, f->entry_size); + *key_found = 0; + *entry_ptr = (void *) bucket_data; + + return 0; + } + } + + /* Bucket full: extend bucket */ + if (f->stack_pos > 0) { + bucket_index = f->stack[--f->stack_pos]; + + bucket = (struct rte_bucket_4_16 *) &f->memory[(f->n_buckets + + bucket_index) * f->bucket_size]; + bucket_prev->next = bucket; + bucket_prev->next_valid = 1; + + bucket->signature[0] = signature; + memcpy(bucket->key[0], key, f->key_size); + memcpy(&bucket->data[0], entry, f->entry_size); + *key_found = 0; + *entry_ptr = (void *) &bucket->data[0]; + return 0; + } + + return -ENOSPC; +} + +static int +rte_table_hash_entry_delete_key16_ext( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_16 *bucket0, *bucket, *bucket_prev; + uint64_t signature; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + signature |= RTE_BUCKET_ENTRY_VALID; + + /* Key is present in the bucket */ + for (bucket_prev = NULL, bucket = bucket0; bucket != NULL; + 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]; + + if ((bucket_signature == signature) && + (memcmp(key, bucket_key, f->key_size) == 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); + + if ((bucket->signature[0] == 0) && + (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; + + memset(bucket, 0, + sizeof(struct rte_bucket_4_16)); + bucket_index = (bucket - + ((struct rte_bucket_4_16 *) + f->memory)) - f->n_buckets; + f->stack[f->stack_pos++] = bucket_index; + } + + return 0; + } + } + + /* Key is not present in the bucket */ + *key_found = 0; + return 0; +} + +#define lookup_key16_cmp(key_in, bucket, pos) \ +{ \ + uint64_t xor[4][2], or[4], signature[4]; \ + \ + 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[1][0] = key_in[0] ^ bucket->key[1][0]; \ + xor[1][1] = key_in[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[3][0] = key_in[0] ^ bucket->key[3][0]; \ + xor[3][1] = key_in[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]; \ + or[2] = xor[2][0] | xor[2][1] | signature[2]; \ + or[3] = xor[3][0] | xor[3][1] | signature[3]; \ + \ + pos = 4; \ + if (or[0] == 0) \ + pos = 0; \ + if (or[1] == 0) \ + pos = 1; \ + if (or[2] == 0) \ + pos = 2; \ + if (or[3] == 0) \ + pos = 3; \ +} + +#define lookup1_stage0(pkt0_index, mbuf0, pkts, pkts_mask) \ +{ \ + uint64_t pkt_mask; \ + \ + pkt0_index = __builtin_ctzll(pkts_mask); \ + pkt_mask = 1LLU << pkt0_index; \ + pkts_mask &= ~pkt_mask; \ + \ + mbuf0 = pkts[pkt0_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf0, 0)); \ +} + +#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) + CACHE_LINE_SIZE));\ +} + +#define lookup1_stage2_lru(pkt2_index, mbuf2, bucket2, \ + pkts_mask_out, entries, f) \ +{ \ + void *a; \ + uint64_t pkt_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + \ + lookup_key16_cmp(key, bucket2, pos); \ + \ + pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket2->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt2_index] = a; \ + lru_update(bucket2, pos); \ +} + +#define lookup1_stage2_ext(pkt2_index, mbuf2, bucket2, pkts_mask_out, entries, \ + buckets_mask, buckets, keys, f) \ +{ \ + struct rte_bucket_4_16 *bucket_next; \ + void *a; \ + uint64_t pkt_mask, bucket_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + \ + lookup_key16_cmp(key, bucket2, pos); \ + \ + pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket2->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt2_index] = a; \ + \ + bucket_mask = (~pkt_mask) & (bucket2->next_valid << pkt2_index);\ + buckets_mask |= bucket_mask; \ + bucket_next = bucket2->next; \ + buckets[pkt2_index] = bucket_next; \ + keys[pkt2_index] = key; \ +} + +#define lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, entries,\ + buckets_mask, f) \ +{ \ + struct rte_bucket_4_16 *bucket, *bucket_next; \ + void *a; \ + uint64_t pkt_mask, bucket_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + bucket = buckets[pkt_index]; \ + key = keys[pkt_index]; \ + \ + lookup_key16_cmp(key, bucket, pos); \ + \ + pkt_mask = (bucket->signature[pos] & 1LLU) << pkt_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt_index] = a; \ + \ + bucket_mask = (~pkt_mask) & (bucket->next_valid << pkt_index);\ + buckets_mask |= bucket_mask; \ + bucket_next = bucket->next; \ + rte_prefetch0(bucket_next); \ + rte_prefetch0((void *)(((uintptr_t) bucket_next) + CACHE_LINE_SIZE));\ + buckets[pkt_index] = bucket_next; \ + keys[pkt_index] = key; \ +} + +#define lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01,\ + pkts, pkts_mask) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + \ + mbuf00 = pkts[pkt00_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + \ + mbuf01 = pkts[pkt01_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#define lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,\ + mbuf00, mbuf01, pkts, pkts_mask) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + \ + mbuf00 = pkts[pkt00_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + if (pkts_mask == 0) \ + pkt01_index = pkt00_index; \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + \ + mbuf01 = pkts[pkt01_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#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) + 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) + CACHE_LINE_SIZE));\ +} + +#define lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21,\ + bucket20, bucket21, pkts_mask_out, entries, f) \ +{ \ + void *a20, *a21; \ + uint64_t pkt20_mask, pkt21_mask; \ + uint64_t *key20, *key21; \ + uint32_t pos20, pos21; \ + \ + key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ + key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ + \ + lookup_key16_cmp(key20, bucket20, pos20); \ + lookup_key16_cmp(key21, bucket21, pos21); \ + \ + pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\ + pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\ + pkts_mask_out |= pkt20_mask | pkt21_mask; \ + \ + a20 = (void *) &bucket20->data[pos20 * f->entry_size]; \ + a21 = (void *) &bucket21->data[pos21 * f->entry_size]; \ + rte_prefetch0(a20); \ + rte_prefetch0(a21); \ + entries[pkt20_index] = a20; \ + entries[pkt21_index] = a21; \ + lru_update(bucket20, pos20); \ + lru_update(bucket21, pos21); \ +} + +#define lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, bucket20, \ + bucket21, pkts_mask_out, entries, buckets_mask, buckets, keys, f) \ +{ \ + struct rte_bucket_4_16 *bucket20_next, *bucket21_next; \ + void *a20, *a21; \ + uint64_t pkt20_mask, pkt21_mask, bucket20_mask, bucket21_mask;\ + uint64_t *key20, *key21; \ + uint32_t pos20, pos21; \ + \ + key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ + key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ + \ + lookup_key16_cmp(key20, bucket20, pos20); \ + lookup_key16_cmp(key21, bucket21, pos21); \ + \ + pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\ + pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\ + pkts_mask_out |= pkt20_mask | pkt21_mask; \ + \ + a20 = (void *) &bucket20->data[pos20 * f->entry_size]; \ + a21 = (void *) &bucket21->data[pos21 * f->entry_size]; \ + rte_prefetch0(a20); \ + rte_prefetch0(a21); \ + entries[pkt20_index] = a20; \ + entries[pkt21_index] = a21; \ + \ + bucket20_mask = (~pkt20_mask) & (bucket20->next_valid << pkt20_index);\ + bucket21_mask = (~pkt21_mask) & (bucket21->next_valid << pkt21_index);\ + buckets_mask |= bucket20_mask | bucket21_mask; \ + bucket20_next = bucket20->next; \ + bucket21_next = bucket21->next; \ + buckets[pkt20_index] = bucket20_next; \ + buckets[pkt21_index] = bucket21_next; \ + keys[pkt20_index] = key20; \ + keys[pkt21_index] = key21; \ +} + +static int +rte_table_hash_lookup_key16_lru( + 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; + + /* 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); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask); + + /* 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); + + /* Pipeline stage 1 */ + lookup2_stage1(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); + + /* Pipeline stage 1 */ + lookup2_stage1(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(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; + return 0; +} /* rte_table_hash_lookup_key16_lru() */ + +static int +rte_table_hash_lookup_key16_ext( + 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]; + + /* 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); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, + buckets, keys, f); + } + + *lookup_hit_mask = pkts_mask_out; + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask); + + /* 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); + + /* Pipeline stage 1 */ + lookup2_stage1(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); + + /* Pipeline stage 1 */ + lookup2_stage1(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(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 */ + 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; + return 0; +} /* rte_table_hash_lookup_key16_ext() */ + +struct rte_table_ops rte_table_hash_key16_lru_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, +}; + +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, + .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, +}; diff --git a/lib/librte_table/rte_table_hash_key32.c b/lib/librte_table/rte_table_hash_key32.c new file mode 100644 index 0000000000..e8f481270e --- /dev/null +++ b/lib/librte_table/rte_table_hash_key32.c @@ -0,0 +1,1120 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 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: + * + * * 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. + */ +#include +#include + +#include +#include +#include +#include + +#include "rte_table_hash.h" +#include "rte_lru.h" + +#define RTE_TABLE_HASH_KEY_SIZE 32 + +#define RTE_BUCKET_ENTRY_VALID 0x1LLU + +struct rte_bucket_4_32 { + /* Cache line 0 */ + uint64_t signature[4 + 1]; + uint64_t lru_list; + struct rte_bucket_4_32 *next; + uint64_t next_valid; + + /* Cache lines 1 and 2 */ + uint64_t key[4][4]; + + /* Cache line 3 */ + uint8_t data[0]; +}; + +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; + rte_table_hash_op_hash f_hash; + uint64_t seed; + + /* Extendible buckets */ + uint32_t n_buckets_ext; + uint32_t stack_pos; + uint32_t *stack; + + /* Lookup table */ + uint8_t memory[0] __rte_cache_aligned; +}; + +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__); + return -EINVAL; + } + + /* signature offset */ + if ((params->signature_offset & 0x3) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid signature offset\n", __func__); + return -EINVAL; + } + + /* key offset */ + if ((params->key_offset & 0x7) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid key offset\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_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 *f; + uint32_t n_buckets, n_entries_per_bucket, key_size, bucket_size_cl; + uint32_t total_size, i; + + /* Check input parameters */ + if ((check_params_create_lru(p) != 0) || + ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_32) % CACHE_LINE_SIZE) != 0)) { + return NULL; + } + n_entries_per_bucket = 4; + key_size = 32; + + /* 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 + CACHE_LINE_SIZE - 1) / CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + n_buckets * + bucket_size_cl * CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id); + if (f == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + return NULL; + } + RTE_LOG(INFO, TABLE, + "%s: Hash table memory footprint is %u bytes\n", __func__, + total_size); + + /* Memory initialization */ + f->n_buckets = n_buckets; + f->n_entries_per_bucket = n_entries_per_bucket; + f->key_size = key_size; + f->entry_size = entry_size; + f->bucket_size = bucket_size_cl * CACHE_LINE_SIZE; + f->signature_offset = p->signature_offset; + f->key_offset = p->key_offset; + f->f_hash = p->f_hash; + f->seed = p->seed; + + for (i = 0; i < n_buckets; i++) { + struct rte_bucket_4_32 *bucket; + + bucket = (struct rte_bucket_4_32 *) &f->memory[i * + f->bucket_size]; + bucket->lru_list = 0x0000000100020003LLU; + } + + return f; +} + +static int +rte_table_hash_free_key32_lru(void *table) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + + /* Check input parameters */ + if (f == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + rte_free(f); + return 0; +} + +static int +rte_table_hash_entry_add_key32_lru( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_32 *bucket; + uint64_t signature, pos; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + 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]; + + if ((bucket_signature == signature) && + (memcmp(key, bucket_key, f->key_size) == 0)) { + uint8_t *bucket_data = &bucket->data[i * f->entry_size]; + + memcpy(bucket_data, entry, f->entry_size); + lru_update(bucket, i); + *key_found = 1; + *entry_ptr = (void *) bucket_data; + return 0; + } + } + + /* 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]; + + 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); + memcpy(bucket_data, entry, f->entry_size); + lru_update(bucket, i); + *key_found = 0; + *entry_ptr = (void *) bucket_data; + + return 0; + } + } + + /* Bucket full: replace LRU entry */ + pos = lru_pos(bucket); + bucket->signature[pos] = signature; + memcpy(bucket->key[pos], key, f->key_size); + memcpy(&bucket->data[pos * f->entry_size], entry, f->entry_size); + lru_update(bucket, pos); + *key_found = 0; + *entry_ptr = (void *) &bucket->data[pos * f->entry_size]; + + return 0; +} + +static int +rte_table_hash_entry_delete_key32_lru( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_32 *bucket; + uint64_t signature; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + 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]; + + if ((bucket_signature == signature) && + (memcmp(key, bucket_key, f->key_size) == 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); + + return 0; + } + } + + /* Key is not present in the bucket */ + *key_found = 0; + 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; + } + + /* signature offset */ + if ((params->signature_offset & 0x3) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid signature offset\n", __func__); + return -EINVAL; + } + + /* key offset */ + if ((params->key_offset & 0x7) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid key offset\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 = + (struct rte_table_hash_key32_ext_params *) 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; + + /* Check input parameters */ + if ((check_params_create_ext(p) != 0) || + ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_32) % CACHE_LINE_SIZE) != 0)) + return NULL; + + n_entries_per_bucket = 4; + key_size = 32; + + /* 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 + CACHE_LINE_SIZE - 1) / CACHE_LINE_SIZE; + stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + CACHE_LINE_SIZE - 1) + / CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + + ((n_buckets + n_buckets_ext) * bucket_size_cl + stack_size_cl) * + CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id); + if (f == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + return NULL; + } + RTE_LOG(INFO, TABLE, + "%s: Hash table memory footprint is %u bytes\n", __func__, + total_size); + + /* Memory initialization */ + f->n_buckets = n_buckets; + f->n_entries_per_bucket = n_entries_per_bucket; + f->key_size = key_size; + f->entry_size = entry_size; + f->bucket_size = bucket_size_cl * CACHE_LINE_SIZE; + f->signature_offset = p->signature_offset; + f->key_offset = p->key_offset; + f->f_hash = p->f_hash; + f->seed = p->seed; + + 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; + + return f; +} + +static int +rte_table_hash_free_key32_ext(void *table) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + + /* Check input parameters */ + if (f == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + rte_free(f); + return 0; +} + +static int +rte_table_hash_entry_add_key32_ext( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_32 *bucket0, *bucket, *bucket_prev; + uint64_t signature; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + 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]; + + if ((bucket_signature == signature) && + (memcmp(key, bucket_key, f->key_size) == 0)) { + uint8_t *bucket_data = &bucket->data[i * + f->entry_size]; + + memcpy(bucket_data, entry, f->entry_size); + *key_found = 1; + *entry_ptr = (void *) bucket_data; + + return 0; + } + } + } + + /* Key is not present in the bucket */ + for (bucket_prev = NULL, bucket = bucket0; bucket != NULL; + 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]; + + 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); + memcpy(bucket_data, entry, f->entry_size); + *key_found = 0; + *entry_ptr = (void *) bucket_data; + + return 0; + } + } + + /* Bucket full: extend bucket */ + if (f->stack_pos > 0) { + bucket_index = f->stack[--f->stack_pos]; + + bucket = (struct rte_bucket_4_32 *) + &f->memory[(f->n_buckets + bucket_index) * + f->bucket_size]; + bucket_prev->next = bucket; + bucket_prev->next_valid = 1; + + bucket->signature[0] = signature; + memcpy(bucket->key[0], key, f->key_size); + memcpy(&bucket->data[0], entry, f->entry_size); + *key_found = 0; + *entry_ptr = (void *) &bucket->data[0]; + return 0; + } + + return -ENOSPC; +} + +static int +rte_table_hash_entry_delete_key32_ext( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_32 *bucket0, *bucket, *bucket_prev; + uint64_t signature; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + signature |= RTE_BUCKET_ENTRY_VALID; + + /* Key is present in the bucket */ + for (bucket_prev = NULL, bucket = bucket0; bucket != NULL; + 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]; + + if ((bucket_signature == signature) && + (memcmp(key, bucket_key, f->key_size) == 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); + + if ((bucket->signature[0] == 0) && + (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; + + memset(bucket, 0, + sizeof(struct rte_bucket_4_32)); + bucket_index = (bucket - + ((struct rte_bucket_4_32 *) + f->memory)) - f->n_buckets; + f->stack[f->stack_pos++] = bucket_index; + } + + return 0; + } + } + + /* Key is not present in the bucket */ + *key_found = 0; + return 0; +} + +#define lookup_key32_cmp(key_in, bucket, pos) \ +{ \ + uint64_t xor[4][4], or[4], signature[4]; \ + \ + 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[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[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[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]; \ + \ + 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];\ + or[2] = xor[2][0] | xor[2][1] | xor[2][2] | xor[2][3] | signature[2];\ + or[3] = xor[3][0] | xor[3][1] | xor[3][2] | xor[3][3] | signature[3];\ + \ + pos = 4; \ + if (or[0] == 0) \ + pos = 0; \ + if (or[1] == 0) \ + pos = 1; \ + if (or[2] == 0) \ + pos = 2; \ + if (or[3] == 0) \ + pos = 3; \ +} + +#define lookup1_stage0(pkt0_index, mbuf0, pkts, pkts_mask) \ +{ \ + uint64_t pkt_mask; \ + \ + pkt0_index = __builtin_ctzll(pkts_mask); \ + pkt_mask = 1LLU << pkt0_index; \ + pkts_mask &= ~pkt_mask; \ + \ + mbuf0 = pkts[pkt0_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf0, 0)); \ +} + +#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_32 *) \ + &f->memory[bucket_index * f->bucket_size]; \ + rte_prefetch0(bucket1); \ + rte_prefetch0((void *)(((uintptr_t) bucket1) + CACHE_LINE_SIZE));\ + rte_prefetch0((void *)(((uintptr_t) bucket1) + 2 * CACHE_LINE_SIZE));\ +} + +#define lookup1_stage2_lru(pkt2_index, mbuf2, bucket2, \ + pkts_mask_out, entries, f) \ +{ \ + void *a; \ + uint64_t pkt_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + \ + lookup_key32_cmp(key, bucket2, pos); \ + \ + pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket2->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt2_index] = a; \ + lru_update(bucket2, pos); \ +} + +#define lookup1_stage2_ext(pkt2_index, mbuf2, bucket2, pkts_mask_out,\ + entries, buckets_mask, buckets, keys, f) \ +{ \ + struct rte_bucket_4_32 *bucket_next; \ + void *a; \ + uint64_t pkt_mask, bucket_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + \ + lookup_key32_cmp(key, bucket2, pos); \ + \ + pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket2->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt2_index] = a; \ + \ + bucket_mask = (~pkt_mask) & (bucket2->next_valid << pkt2_index);\ + buckets_mask |= bucket_mask; \ + bucket_next = bucket2->next; \ + buckets[pkt2_index] = bucket_next; \ + keys[pkt2_index] = key; \ +} + +#define lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, \ + entries, buckets_mask, f) \ +{ \ + struct rte_bucket_4_32 *bucket, *bucket_next; \ + void *a; \ + uint64_t pkt_mask, bucket_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + bucket = buckets[pkt_index]; \ + key = keys[pkt_index]; \ + \ + lookup_key32_cmp(key, bucket, pos); \ + \ + pkt_mask = (bucket->signature[pos] & 1LLU) << pkt_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt_index] = a; \ + \ + bucket_mask = (~pkt_mask) & (bucket->next_valid << pkt_index);\ + buckets_mask |= bucket_mask; \ + bucket_next = bucket->next; \ + rte_prefetch0(bucket_next); \ + rte_prefetch0((void *)(((uintptr_t) bucket_next) + CACHE_LINE_SIZE));\ + rte_prefetch0((void *)(((uintptr_t) bucket_next) + \ + 2 * CACHE_LINE_SIZE)); \ + buckets[pkt_index] = bucket_next; \ + keys[pkt_index] = key; \ +} + +#define lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01,\ + pkts, pkts_mask) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + \ + mbuf00 = pkts[pkt00_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + \ + mbuf01 = pkts[pkt01_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#define lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,\ + mbuf00, mbuf01, pkts, pkts_mask) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + \ + mbuf00 = pkts[pkt00_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + if (pkts_mask == 0) \ + pkt01_index = pkt00_index; \ + \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + \ + mbuf01 = pkts[pkt01_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#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_32 *) \ + &f->memory[bucket10_index * f->bucket_size]; \ + rte_prefetch0(bucket10); \ + rte_prefetch0((void *)(((uintptr_t) bucket10) + CACHE_LINE_SIZE));\ + rte_prefetch0((void *)(((uintptr_t) bucket10) + 2 * CACHE_LINE_SIZE));\ + \ + signature11 = RTE_MBUF_METADATA_UINT32(mbuf11, f->signature_offset);\ + 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((void *)(((uintptr_t) bucket11) + CACHE_LINE_SIZE));\ + rte_prefetch0((void *)(((uintptr_t) bucket11) + 2 * CACHE_LINE_SIZE));\ +} + +#define lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21,\ + bucket20, bucket21, pkts_mask_out, entries, f) \ +{ \ + void *a20, *a21; \ + uint64_t pkt20_mask, pkt21_mask; \ + uint64_t *key20, *key21; \ + uint32_t pos20, pos21; \ + \ + 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); \ + \ + pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\ + pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\ + pkts_mask_out |= pkt20_mask | pkt21_mask; \ + \ + a20 = (void *) &bucket20->data[pos20 * f->entry_size]; \ + a21 = (void *) &bucket21->data[pos21 * f->entry_size]; \ + rte_prefetch0(a20); \ + rte_prefetch0(a21); \ + entries[pkt20_index] = a20; \ + entries[pkt21_index] = a21; \ + lru_update(bucket20, pos20); \ + lru_update(bucket21, pos21); \ +} + +#define lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, bucket20, \ + bucket21, pkts_mask_out, entries, buckets_mask, buckets, keys, f)\ +{ \ + struct rte_bucket_4_32 *bucket20_next, *bucket21_next; \ + void *a20, *a21; \ + uint64_t pkt20_mask, pkt21_mask, bucket20_mask, bucket21_mask;\ + uint64_t *key20, *key21; \ + uint32_t pos20, pos21; \ + \ + 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); \ + \ + pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\ + pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\ + pkts_mask_out |= pkt20_mask | pkt21_mask; \ + \ + a20 = (void *) &bucket20->data[pos20 * f->entry_size]; \ + a21 = (void *) &bucket21->data[pos21 * f->entry_size]; \ + rte_prefetch0(a20); \ + rte_prefetch0(a21); \ + entries[pkt20_index] = a20; \ + entries[pkt21_index] = a21; \ + \ + bucket20_mask = (~pkt20_mask) & (bucket20->next_valid << pkt20_index);\ + bucket21_mask = (~pkt21_mask) & (bucket21->next_valid << pkt21_index);\ + buckets_mask |= bucket20_mask | bucket21_mask; \ + bucket20_next = bucket20->next; \ + bucket21_next = bucket21->next; \ + buckets[pkt20_index] = bucket20_next; \ + buckets[pkt21_index] = bucket21_next; \ + keys[pkt20_index] = key20; \ + keys[pkt21_index] = key21; \ +} + +static int +rte_table_hash_lookup_key32_lru( + 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_32 *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; + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_32 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask); + + /* 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); + + /* Pipeline stage 1 */ + lookup2_stage1(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); + + /* Pipeline stage 1 */ + lookup2_stage1(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(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; + return 0; +} /* rte_table_hash_lookup_key32_lru() */ + +static int +rte_table_hash_lookup_key32_ext( + 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_32 *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_32 *buckets[RTE_PORT_IN_BURST_SIZE_MAX]; + uint64_t *keys[RTE_PORT_IN_BURST_SIZE_MAX]; + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_32 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, buckets, + keys, f); + } + + *lookup_hit_mask = pkts_mask_out; + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask); + + /* 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); + + /* Pipeline stage 1 */ + lookup2_stage1(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); + + /* Pipeline stage 1 */ + lookup2_stage1(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(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 */ + 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; + return 0; +} /* rte_table_hash_lookup_key32_ext() */ + +struct rte_table_ops rte_table_hash_key32_lru_ops = { + .f_create = rte_table_hash_create_key32_lru, + .f_free = rte_table_hash_free_key32_lru, + .f_add = rte_table_hash_entry_add_key32_lru, + .f_delete = rte_table_hash_entry_delete_key32_lru, + .f_lookup = rte_table_hash_lookup_key32_lru, +}; + +struct rte_table_ops rte_table_hash_key32_ext_ops = { + .f_create = rte_table_hash_create_key32_ext, + .f_free = rte_table_hash_free_key32_ext, + .f_add = rte_table_hash_entry_add_key32_ext, + .f_delete = rte_table_hash_entry_delete_key32_ext, + .f_lookup = rte_table_hash_lookup_key32_ext, +}; diff --git a/lib/librte_table/rte_table_hash_key8.c b/lib/librte_table/rte_table_hash_key8.c new file mode 100644 index 0000000000..d60c96eaa3 --- /dev/null +++ b/lib/librte_table/rte_table_hash_key8.c @@ -0,0 +1,1398 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 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: + * + * * 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. + */ +#include +#include + +#include +#include +#include +#include + +#include "rte_table_hash.h" +#include "rte_lru.h" + +#define RTE_TABLE_HASH_KEY_SIZE 8 + +struct rte_bucket_4_8 { + /* Cache line 0 */ + uint64_t signature; + uint64_t lru_list; + struct rte_bucket_4_8 *next; + uint64_t next_valid; + + uint64_t key[4]; + + /* Cache line 1 */ + uint8_t data[0]; +}; + +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; + rte_table_hash_op_hash f_hash; + uint64_t seed; + + /* Extendible buckets */ + uint32_t n_buckets_ext; + uint32_t stack_pos; + uint32_t *stack; + + /* Lookup table */ + uint8_t memory[0] __rte_cache_aligned; +}; + +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__); + return -EINVAL; + } + + /* signature offset */ + if ((params->signature_offset & 0x3) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid signature_offset\n", __func__); + return -EINVAL; + } + + /* key offset */ + if ((params->key_offset & 0x7) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid key_offset\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_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 *f; + uint32_t n_buckets, n_entries_per_bucket, key_size, bucket_size_cl; + uint32_t total_size, i; + + /* Check input parameters */ + if ((check_params_create_lru(p) != 0) || + ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_8) % CACHE_LINE_SIZE) != 0)) { + return NULL; + } + n_entries_per_bucket = 4; + key_size = 8; + + /* 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 + CACHE_LINE_SIZE - 1) / CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + n_buckets * + bucket_size_cl * CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id); + if (f == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + return NULL; + } + RTE_LOG(INFO, TABLE, + "%s: Hash table memory footprint is %u bytes\n", + __func__, total_size); + + /* Memory initialization */ + f->n_buckets = n_buckets; + f->n_entries_per_bucket = n_entries_per_bucket; + f->key_size = key_size; + f->entry_size = entry_size; + f->bucket_size = bucket_size_cl * CACHE_LINE_SIZE; + f->signature_offset = p->signature_offset; + f->key_offset = p->key_offset; + f->f_hash = p->f_hash; + f->seed = p->seed; + + for (i = 0; i < n_buckets; i++) { + struct rte_bucket_4_8 *bucket; + + bucket = (struct rte_bucket_4_8 *) &f->memory[i * + f->bucket_size]; + bucket->lru_list = 0x0000000100020003LLU; + } + + return f; +} + +static int +rte_table_hash_free_key8_lru(void *table) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + + /* Check input parameters */ + if (f == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + rte_free(f); + return 0; +} + +static int +rte_table_hash_entry_add_key8_lru( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_8 *bucket; + uint64_t signature, mask, pos; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + + /* 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]; + + if ((bucket_signature & mask) && + (*((uint64_t *) key) == bucket_key)) { + uint8_t *bucket_data = &bucket->data[i * f->entry_size]; + + memcpy(bucket_data, entry, f->entry_size); + lru_update(bucket, i); + *key_found = 1; + *entry_ptr = (void *) bucket_data; + return 0; + } + } + + /* Key is not present in the bucket */ + for (i = 0, mask = 1LLU; i < 4; i++, mask <<= 1) { + uint64_t bucket_signature = bucket->signature; + + if ((bucket_signature & mask) == 0) { + uint8_t *bucket_data = &bucket->data[i * f->entry_size]; + + bucket->signature |= mask; + bucket->key[i] = *((uint64_t *) key); + memcpy(bucket_data, entry, f->entry_size); + lru_update(bucket, i); + *key_found = 0; + *entry_ptr = (void *) bucket_data; + + return 0; + } + } + + /* Bucket full: replace LRU entry */ + pos = lru_pos(bucket); + bucket->key[pos] = *((uint64_t *) key); + memcpy(&bucket->data[pos * f->entry_size], entry, f->entry_size); + lru_update(bucket, pos); + *key_found = 0; + *entry_ptr = (void *) &bucket->data[pos * f->entry_size]; + + return 0; +} + +static int +rte_table_hash_entry_delete_key8_lru( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_8 *bucket; + uint64_t signature, mask; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + + /* 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]; + + if ((bucket_signature & mask) && + (*((uint64_t *) key) == bucket_key)) { + uint8_t *bucket_data = &bucket->data[i * f->entry_size]; + + bucket->signature &= ~mask; + *key_found = 1; + if (entry) + memcpy(entry, bucket_data, f->entry_size); + + return 0; + } + } + + /* Key is not present in the bucket */ + *key_found = 0; + 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; + } + + /* signature offset */ + if ((params->signature_offset & 0x3) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid signature_offset\n", __func__); + return -EINVAL; + } + + /* key offset */ + if ((params->key_offset & 0x7) != 0) { + RTE_LOG(ERR, TABLE, "%s: invalid key_offset\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 *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; + + /* Check input parameters */ + if ((check_params_create_ext(p) != 0) || + ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_8) % CACHE_LINE_SIZE) != 0)) + return NULL; + + n_entries_per_bucket = 4; + key_size = 8; + + /* 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 + CACHE_LINE_SIZE - 1) / CACHE_LINE_SIZE; + stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + CACHE_LINE_SIZE - 1) + / CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + ((n_buckets + + n_buckets_ext) * bucket_size_cl + stack_size_cl) * + CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id); + if (f == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + return NULL; + } + RTE_LOG(INFO, TABLE, + "%s: Hash table memory footprint is %u bytes\n", + __func__, total_size); + + /* Memory initialization */ + f->n_buckets = n_buckets; + f->n_entries_per_bucket = n_entries_per_bucket; + f->key_size = key_size; + f->entry_size = entry_size; + f->bucket_size = bucket_size_cl * CACHE_LINE_SIZE; + f->signature_offset = p->signature_offset; + f->key_offset = p->key_offset; + f->f_hash = p->f_hash; + f->seed = p->seed; + + 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; + + return f; +} + +static int +rte_table_hash_free_key8_ext(void *table) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + + /* Check input parameters */ + if (f == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + rte_free(f); + return 0; +} + +static int +rte_table_hash_entry_add_key8_ext( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_8 *bucket0, *bucket, *bucket_prev; + uint64_t signature; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + + /* Key is present in the bucket */ + for (bucket = bucket0; bucket != NULL; bucket = bucket->next) { + uint64_t mask; + + for (i = 0, mask = 1LLU; i < 4; i++, mask <<= 1) { + uint64_t bucket_signature = bucket->signature; + uint64_t bucket_key = bucket->key[i]; + + if ((bucket_signature & mask) && + (*((uint64_t *) key) == bucket_key)) { + uint8_t *bucket_data = &bucket->data[i * + f->entry_size]; + + memcpy(bucket_data, entry, f->entry_size); + *key_found = 1; + *entry_ptr = (void *) bucket_data; + return 0; + } + } + } + + /* Key is not present in the bucket */ + for (bucket_prev = NULL, bucket = bucket0; + bucket != NULL; bucket_prev = bucket, bucket = bucket->next) { + uint64_t mask; + + for (i = 0, mask = 1LLU; i < 4; i++, mask <<= 1) { + uint64_t bucket_signature = bucket->signature; + + if ((bucket_signature & mask) == 0) { + uint8_t *bucket_data = &bucket->data[i * + f->entry_size]; + + bucket->signature |= mask; + bucket->key[i] = *((uint64_t *) key); + memcpy(bucket_data, entry, f->entry_size); + *key_found = 0; + *entry_ptr = (void *) bucket_data; + + return 0; + } + } + } + + /* Bucket full: extend bucket */ + if (f->stack_pos > 0) { + bucket_index = f->stack[--f->stack_pos]; + + bucket = (struct rte_bucket_4_8 *) &f->memory[(f->n_buckets + + bucket_index) * f->bucket_size]; + bucket_prev->next = bucket; + bucket_prev->next_valid = 1; + + bucket->signature = 1; + bucket->key[0] = *((uint64_t *) key); + memcpy(&bucket->data[0], entry, f->entry_size); + *key_found = 0; + *entry_ptr = (void *) &bucket->data[0]; + return 0; + } + + return -ENOSPC; +} + +static int +rte_table_hash_entry_delete_key8_ext( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_8 *bucket0, *bucket, *bucket_prev; + uint64_t signature; + uint32_t bucket_index, i; + + signature = f->f_hash(key, 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]; + + /* Key is present in the bucket */ + for (bucket_prev = NULL, bucket = bucket0; bucket != NULL; + bucket_prev = bucket, bucket = bucket->next) { + uint64_t mask; + + for (i = 0, mask = 1LLU; i < 4; i++, mask <<= 1) { + uint64_t bucket_signature = bucket->signature; + uint64_t bucket_key = bucket->key[i]; + + if ((bucket_signature & mask) && + (*((uint64_t *) key) == bucket_key)) { + uint8_t *bucket_data = &bucket->data[i * + f->entry_size]; + + bucket->signature &= ~mask; + *key_found = 1; + if (entry) + memcpy(entry, bucket_data, + f->entry_size); + + if ((bucket->signature == 0) && + (bucket_prev != NULL)) { + bucket_prev->next = bucket->next; + bucket_prev->next_valid = + bucket->next_valid; + + memset(bucket, 0, + sizeof(struct rte_bucket_4_8)); + bucket_index = (bucket - + ((struct rte_bucket_4_8 *) + f->memory)) - f->n_buckets; + f->stack[f->stack_pos++] = bucket_index; + } + + return 0; + } + } + } + + /* Key is not present in the bucket */ + *key_found = 0; + return 0; +} + +#define lookup_key8_cmp(key_in, bucket, pos) \ +{ \ + uint64_t xor[4], signature; \ + \ + 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);\ + \ + pos = 4; \ + if (xor[0] == 0) \ + pos = 0; \ + if (xor[1] == 0) \ + pos = 1; \ + if (xor[2] == 0) \ + pos = 2; \ + if (xor[3] == 0) \ + pos = 3; \ +} + +#define lookup1_stage0(pkt0_index, mbuf0, pkts, pkts_mask) \ +{ \ + uint64_t pkt_mask; \ + \ + pkt0_index = __builtin_ctzll(pkts_mask); \ + pkt_mask = 1LLU << pkt0_index; \ + pkts_mask &= ~pkt_mask; \ + \ + mbuf0 = pkts[pkt0_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf0, 0)); \ +} + +#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; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf1, f->key_offset);\ + signature = f->f_hash(key, RTE_TABLE_HASH_KEY_SIZE, f->seed);\ + 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_stage2_lru(pkt2_index, mbuf2, bucket2, \ + pkts_mask_out, entries, f) \ +{ \ + void *a; \ + uint64_t pkt_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + \ + lookup_key8_cmp(key, bucket2, pos); \ + \ + pkt_mask = ((bucket2->signature >> pos) & 1LLU) << pkt2_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket2->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt2_index] = a; \ + lru_update(bucket2, pos); \ +} + +#define lookup1_stage2_ext(pkt2_index, mbuf2, bucket2, pkts_mask_out,\ + entries, buckets_mask, buckets, keys, f) \ +{ \ + struct rte_bucket_4_8 *bucket_next; \ + void *a; \ + uint64_t pkt_mask, bucket_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + \ + lookup_key8_cmp(key, bucket2, pos); \ + \ + pkt_mask = ((bucket2->signature >> pos) & 1LLU) << pkt2_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket2->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt2_index] = a; \ + \ + bucket_mask = (~pkt_mask) & (bucket2->next_valid << pkt2_index);\ + buckets_mask |= bucket_mask; \ + bucket_next = bucket2->next; \ + buckets[pkt2_index] = bucket_next; \ + keys[pkt2_index] = key; \ +} + +#define lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, entries,\ + buckets_mask, f) \ +{ \ + struct rte_bucket_4_8 *bucket, *bucket_next; \ + void *a; \ + uint64_t pkt_mask, bucket_mask; \ + uint64_t *key; \ + uint32_t pos; \ + \ + bucket = buckets[pkt_index]; \ + key = keys[pkt_index]; \ + \ + lookup_key8_cmp(key, bucket, pos); \ + \ + pkt_mask = ((bucket->signature >> pos) & 1LLU) << pkt_index;\ + pkts_mask_out |= pkt_mask; \ + \ + a = (void *) &bucket->data[pos * f->entry_size]; \ + rte_prefetch0(a); \ + entries[pkt_index] = a; \ + \ + bucket_mask = (~pkt_mask) & (bucket->next_valid << pkt_index);\ + buckets_mask |= bucket_mask; \ + bucket_next = bucket->next; \ + rte_prefetch0(bucket_next); \ + buckets[pkt_index] = bucket_next; \ + keys[pkt_index] = key; \ +} + +#define lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01,\ + pkts, pkts_mask) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + \ + mbuf00 = pkts[pkt00_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + \ + mbuf01 = pkts[pkt01_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#define lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,\ + mbuf00, mbuf01, pkts, pkts_mask) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + \ + mbuf00 = pkts[pkt00_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + if (pkts_mask == 0) \ + pkt01_index = pkt00_index; \ + \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + \ + mbuf01 = pkts[pkt01_index]; \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#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)\ +{ \ + uint64_t *key10, *key11; \ + uint64_t signature10, signature11; \ + uint32_t bucket10_index, bucket11_index; \ + rte_table_hash_op_hash f_hash = f->f_hash; \ + uint64_t seed = f->seed; \ + uint32_t key_offset = f->key_offset; \ + \ + key10 = RTE_MBUF_METADATA_UINT64_PTR(mbuf10, key_offset);\ + key11 = RTE_MBUF_METADATA_UINT64_PTR(mbuf11, key_offset);\ + \ + signature10 = f_hash(key10, RTE_TABLE_HASH_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(key11, RTE_TABLE_HASH_KEY_SIZE, seed);\ + 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_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21,\ + bucket20, bucket21, pkts_mask_out, entries, f) \ +{ \ + void *a20, *a21; \ + uint64_t pkt20_mask, pkt21_mask; \ + uint64_t *key20, *key21; \ + uint32_t pos20, pos21; \ + \ + key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ + key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ + \ + lookup_key8_cmp(key20, bucket20, pos20); \ + lookup_key8_cmp(key21, bucket21, pos21); \ + \ + pkt20_mask = ((bucket20->signature >> pos20) & 1LLU) << pkt20_index;\ + pkt21_mask = ((bucket21->signature >> pos21) & 1LLU) << pkt21_index;\ + pkts_mask_out |= pkt20_mask | pkt21_mask; \ + \ + a20 = (void *) &bucket20->data[pos20 * f->entry_size]; \ + a21 = (void *) &bucket21->data[pos21 * f->entry_size]; \ + rte_prefetch0(a20); \ + rte_prefetch0(a21); \ + entries[pkt20_index] = a20; \ + entries[pkt21_index] = a21; \ + lru_update(bucket20, pos20); \ + lru_update(bucket21, pos21); \ +} + +#define lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, bucket20, \ + bucket21, pkts_mask_out, entries, buckets_mask, buckets, keys, f)\ +{ \ + struct rte_bucket_4_8 *bucket20_next, *bucket21_next; \ + void *a20, *a21; \ + uint64_t pkt20_mask, pkt21_mask, bucket20_mask, bucket21_mask;\ + uint64_t *key20, *key21; \ + uint32_t pos20, pos21; \ + \ + key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ + key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ + \ + lookup_key8_cmp(key20, bucket20, pos20); \ + lookup_key8_cmp(key21, bucket21, pos21); \ + \ + pkt20_mask = ((bucket20->signature >> pos20) & 1LLU) << pkt20_index;\ + pkt21_mask = ((bucket21->signature >> pos21) & 1LLU) << pkt21_index;\ + pkts_mask_out |= pkt20_mask | pkt21_mask; \ + \ + a20 = (void *) &bucket20->data[pos20 * f->entry_size]; \ + a21 = (void *) &bucket21->data[pos21 * f->entry_size]; \ + rte_prefetch0(a20); \ + rte_prefetch0(a21); \ + entries[pkt20_index] = a20; \ + entries[pkt21_index] = a21; \ + \ + bucket20_mask = (~pkt20_mask) & (bucket20->next_valid << pkt20_index);\ + bucket21_mask = (~pkt21_mask) & (bucket21->next_valid << pkt21_index);\ + buckets_mask |= bucket20_mask | bucket21_mask; \ + bucket20_next = bucket20->next; \ + bucket21_next = bucket21->next; \ + buckets[pkt20_index] = bucket20_next; \ + buckets[pkt21_index] = bucket21_next; \ + keys[pkt20_index] = key20; \ + keys[pkt21_index] = key21; \ +} + +static int +rte_table_hash_lookup_key8_lru( + 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, + pkt11_index, pkt20_index, pkt21_index; + uint64_t pkts_mask_out = 0; + + /* 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); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask); + + /* 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); + + /* Pipeline stage 1 */ + lookup2_stage1(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); + + /* Pipeline stage 1 */ + lookup2_stage1(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(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; + 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; + + /* 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); + lookup1_stage1_dosig(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask); + + /* 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); + + /* 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); + + /* 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; + return 0; +} /* rte_table_hash_lookup_key8_lru_dosig() */ + +static int +rte_table_hash_lookup_key8_ext( + 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]; + + /* 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); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, buckets, + keys, f); + } + + *lookup_hit_mask = pkts_mask_out; + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask); + + /* 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); + + /* Pipeline stage 1 */ + lookup2_stage1(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); + + /* Pipeline stage 1 */ + lookup2_stage1(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(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 */ + 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; + 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]; + + /* 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); + lookup1_stage1_dosig(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, + buckets, keys, f); + } + + *lookup_hit_mask = pkts_mask_out; + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask); + + /* 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); + + /* 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); + + /* 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 */ + 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; + return 0; +} /* rte_table_hash_lookup_key8_dosig_ext() */ + +struct rte_table_ops rte_table_hash_key8_lru_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_lookup = rte_table_hash_lookup_key8_lru, +}; + +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_lookup = rte_table_hash_lookup_key8_lru_dosig, +}; + +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, + .f_add = rte_table_hash_entry_add_key8_ext, + .f_delete = rte_table_hash_entry_delete_key8_ext, + .f_lookup = rte_table_hash_lookup_key8_ext, +}; + +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_lookup = rte_table_hash_lookup_key8_ext_dosig, +}; diff --git a/lib/librte_table/rte_table_hash_lru.c b/lib/librte_table/rte_table_hash_lru.c new file mode 100644 index 0000000000..d1a49847f9 --- /dev/null +++ b/lib/librte_table/rte_table_hash_lru.c @@ -0,0 +1,1065 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 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: + * + * * 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. + */ + +#include +#include + +#include +#include +#include +#include + +#include "rte_table_hash.h" +#include "rte_lru.h" + +#define KEYS_PER_BUCKET 4 + +struct bucket { + union { + struct bucket *next; + uint64_t lru_list; + }; + uint16_t sig[KEYS_PER_BUCKET]; + uint32_t key_pos[KEYS_PER_BUCKET]; +}; + +struct grinder { + struct bucket *bkt; + uint64_t sig; + uint64_t match; + uint64_t match_pos; + uint32_t key_index; +}; + +struct rte_table_hash { + /* Input parameters */ + uint32_t key_size; + uint32_t entry_size; + uint32_t n_keys; + uint32_t n_buckets; + rte_table_hash_op_hash f_hash; + uint64_t seed; + uint32_t signature_offset; + uint32_t key_offset; + + /* Internal */ + uint64_t bucket_mask; + uint32_t key_size_shl; + uint32_t data_size_shl; + uint32_t key_stack_tos; + + /* Grinder */ + struct grinder grinders[RTE_PORT_IN_BURST_SIZE_MAX]; + + /* Tables */ + struct bucket *buckets; + uint8_t *key_mem; + uint8_t *data_mem; + uint32_t *key_stack; + + /* Table memory */ + uint8_t memory[0] __rte_cache_aligned; +}; + +static int +check_params_create(struct rte_table_hash_lru_params *params) +{ + uint32_t n_buckets_min; + + /* key_size */ + if ((params->key_size == 0) || + (!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))) { + 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_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 invalid value\n", __func__); + return -EINVAL; + } + + /* signature offset */ + if ((params->signature_offset & 0x3) != 0) { + RTE_LOG(ERR, TABLE, "%s: signature_offset invalid value\n", + __func__); + return -EINVAL; + } + + /* key offset */ + if ((params->key_offset & 0x7) != 0) { + RTE_LOG(ERR, TABLE, "%s: key_offset invalid value\n", __func__); + return -EINVAL; + } + + return 0; +} + +static void * +rte_table_hash_lru_create(void *params, int socket_id, uint32_t entry_size) +{ + struct rte_table_hash_lru_params *p = + (struct rte_table_hash_lru_params *) params; + struct rte_table_hash *t; + uint32_t total_size, table_meta_sz, table_meta_offset; + 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; + + /* Check input parameters */ + if ((check_params_create(p) != 0) || + (!rte_is_power_of_2(entry_size)) || + ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) || + (sizeof(struct bucket) != (CACHE_LINE_SIZE / 2))) { + return NULL; + } + + /* Memory allocation */ + table_meta_sz = CACHE_LINE_ROUNDUP(sizeof(struct rte_table_hash)); + bucket_sz = CACHE_LINE_ROUNDUP(p->n_buckets * sizeof(struct bucket)); + key_sz = CACHE_LINE_ROUNDUP(p->n_keys * p->key_size); + key_stack_sz = CACHE_LINE_ROUNDUP(p->n_keys * sizeof(uint32_t)); + data_sz = CACHE_LINE_ROUNDUP(p->n_keys * entry_size); + total_size = table_meta_sz + bucket_sz + key_sz + key_stack_sz + + data_sz; + + t = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id); + if (t == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + return NULL; + } + RTE_LOG(INFO, TABLE, "%s (%u-byte key): Hash table memory footprint is " + "%u bytes\n", __func__, p->key_size, 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->f_hash = p->f_hash; + t->seed = p->seed; + t->signature_offset = p->signature_offset; + t->key_offset = p->key_offset; + + /* Internal */ + t->bucket_mask = t->n_buckets - 1; + t->key_size_shl = __builtin_ctzl(p->key_size); + t->data_size_shl = __builtin_ctzl(p->key_size); + + /* Tables */ + table_meta_offset = 0; + bucket_offset = table_meta_offset + table_meta_sz; + key_offset = bucket_offset + bucket_sz; + key_stack_offset = key_offset + key_sz; + data_offset = key_stack_offset + key_stack_sz; + + 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 stack */ + for (i = 0; i < t->n_keys; i++) + t->key_stack[i] = t->n_keys - 1 - i; + t->key_stack_tos = t->n_keys; + + /* LRU */ + for (i = 0; i < t->n_buckets; i++) { + struct bucket *bkt = &t->buckets[i]; + + lru_init(bkt); + } + + return t; +} + +static int +rte_table_hash_lru_free(void *table) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + + /* Check input parameters */ + if (t == NULL) + return -EINVAL; + + rte_free(t); + return 0; +} + +static int +rte_table_hash_lru_entry_add(void *table, void *key, void *entry, + int *key_found, void **entry_ptr) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + struct bucket *bkt; + uint64_t sig; + uint32_t bkt_index, i; + + sig = t->f_hash(key, t->key_size, t->seed); + bkt_index = sig & t->bucket_mask; + bkt = &t->buckets[bkt_index]; + sig = (sig >> 16) | 1LLU; + + /* Key is present in the bucket */ + for (i = 0; i < KEYS_PER_BUCKET; i++) { + uint64_t bkt_sig = (uint64_t) bkt->sig[i]; + uint32_t bkt_key_index = bkt->key_pos[i]; + 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)) { + uint8_t *data = &t->data_mem[bkt_key_index << + t->data_size_shl]; + + memcpy(data, entry, t->entry_size); + lru_update(bkt, i); + *key_found = 1; + *entry_ptr = (void *) data; + return 0; + } + } + + /* Key is not present in the bucket */ + for (i = 0; i < KEYS_PER_BUCKET; i++) { + uint64_t bkt_sig = (uint64_t) bkt->sig[i]; + + if (bkt_sig == 0) { + uint32_t bkt_key_index; + uint8_t *bkt_key, *data; + + /* Allocate new key */ + if (t->key_stack_tos == 0) { + /* No keys available */ + return -ENOSPC; + } + bkt_key_index = t->key_stack[--t->key_stack_tos]; + + /* Install new key */ + bkt_key = &t->key_mem[bkt_key_index << t->key_size_shl]; + data = &t->data_mem[bkt_key_index << t->data_size_shl]; + + bkt->sig[i] = (uint16_t) sig; + bkt->key_pos[i] = bkt_key_index; + memcpy(bkt_key, key, t->key_size); + memcpy(data, entry, t->entry_size); + lru_update(bkt, i); + + *key_found = 0; + *entry_ptr = (void *) data; + return 0; + } + } + + /* Bucket full */ + { + uint64_t pos = lru_pos(bkt); + uint32_t bkt_key_index = bkt->key_pos[pos]; + uint8_t *bkt_key = &t->key_mem[bkt_key_index << + t->key_size_shl]; + 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); + memcpy(data, entry, t->entry_size); + lru_update(bkt, pos); + + *key_found = 0; + *entry_ptr = (void *) data; + return 0; + } +} + +static int +rte_table_hash_lru_entry_delete(void *table, void *key, int *key_found, + void *entry) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + struct bucket *bkt; + uint64_t sig; + uint32_t bkt_index, i; + + sig = t->f_hash(key, t->key_size, t->seed); + bkt_index = sig & t->bucket_mask; + bkt = &t->buckets[bkt_index]; + sig = (sig >> 16) | 1LLU; + + /* Key is present in the bucket */ + for (i = 0; i < KEYS_PER_BUCKET; i++) { + uint64_t bkt_sig = (uint64_t) bkt->sig[i]; + uint32_t bkt_key_index = bkt->key_pos[i]; + 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)) { + 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); + return 0; + } + } + + /* Key is not present in the bucket */ + *key_found = 0; + return 0; +} + +static int rte_table_hash_lru_lookup_unoptimized( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries, + int dosig) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + uint64_t pkts_mask_out = 0; + + for ( ; pkts_mask; ) { + struct bucket *bkt; + struct rte_mbuf *pkt; + uint8_t *key; + uint64_t pkt_mask, sig; + uint32_t pkt_index, bkt_index, i; + + pkt_index = __builtin_ctzll(pkts_mask); + pkt_mask = 1LLU << pkt_index; + pkts_mask &= ~pkt_mask; + + 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); + + bkt_index = sig & t->bucket_mask; + bkt = &t->buckets[bkt_index]; + sig = (sig >> 16) | 1LLU; + + /* Key is present in the bucket */ + for (i = 0; i < KEYS_PER_BUCKET; i++) { + uint64_t bkt_sig = (uint64_t) bkt->sig[i]; + uint32_t bkt_key_index = bkt->key_pos[i]; + 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)) { + uint8_t *data = &t->data_mem[bkt_key_index << + t->data_size_shl]; + + lru_update(bkt, i); + pkts_mask_out |= pkt_mask; + entries[pkt_index] = (void *) data; + break; + } + } + } + + *lookup_hit_mask = pkts_mask_out; + return 0; +} + +/*** +* +* mask = match bitmask +* match = at least one match +* match_many = more than one match +* match_pos = position of first match +* +* ---------------------------------------- +* mask match match_many match_pos +* ---------------------------------------- +* 0000 0 0 00 +* 0001 1 0 00 +* 0010 1 0 01 +* 0011 1 1 00 +* ---------------------------------------- +* 0100 1 0 10 +* 0101 1 1 00 +* 0110 1 1 01 +* 0111 1 1 00 +* ---------------------------------------- +* 1000 1 0 11 +* 1001 1 1 00 +* 1010 1 1 01 +* 1011 1 1 00 +* ---------------------------------------- +* 1100 1 1 10 +* 1101 1 1 00 +* 1110 1 1 01 +* 1111 1 1 00 +* ---------------------------------------- +* +* match = 1111_1111_1111_1110 +* match_many = 1111_1110_1110_1000 +* match_pos = 0001_0010_0001_0011__0001_0010_0001_0000 +* +* match = 0xFFFELLU +* match_many = 0xFEE8LLU +* match_pos = 0x12131210LLU +* +***/ + +#define LUT_MATCH 0xFFFELLU +#define LUT_MATCH_MANY 0xFEE8LLU +#define LUT_MATCH_POS 0x12131210LLU + +#define lookup_cmp_sig(mbuf_sig, bucket, match, match_many, match_pos)\ +{ \ + uint64_t bucket_sig[4], mask[4], mask_all; \ + \ + bucket_sig[0] = bucket->sig[0]; \ + bucket_sig[1] = bucket->sig[1]; \ + bucket_sig[2] = bucket->sig[2]; \ + bucket_sig[3] = bucket->sig[3]; \ + \ + bucket_sig[0] ^= mbuf_sig; \ + bucket_sig[1] ^= mbuf_sig; \ + bucket_sig[2] ^= mbuf_sig; \ + bucket_sig[3] ^= mbuf_sig; \ + \ + mask[0] = 0; \ + mask[1] = 0; \ + mask[2] = 0; \ + mask[3] = 0; \ + \ + if (bucket_sig[0] == 0) \ + mask[0] = 1; \ + if (bucket_sig[1] == 0) \ + mask[1] = 2; \ + if (bucket_sig[2] == 0) \ + mask[2] = 4; \ + if (bucket_sig[3] == 0) \ + mask[3] = 8; \ + \ + mask_all = (mask[0] | mask[1]) | (mask[2] | mask[3]); \ + \ + match = (LUT_MATCH >> mask_all) & 1; \ + match_many = (LUT_MATCH_MANY >> mask_all) & 1; \ + match_pos = (LUT_MATCH_POS >> (mask_all << 1)) & 3; \ +} + +#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; \ + } \ +} + +#define lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index)\ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + struct rte_mbuf *mbuf00, *mbuf01; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + mbuf00 = pkts[pkt00_index]; \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + mbuf01 = pkts[pkt01_index]; \ + \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#define lookup2_stage0_with_odd_support(t, g, pkts, pkts_mask, pkt00_index, \ + pkt01_index) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + struct rte_mbuf *mbuf00, *mbuf01; \ + \ + pkt00_index = __builtin_ctzll(pkts_mask); \ + pkt00_mask = 1LLU << pkt00_index; \ + pkts_mask &= ~pkt00_mask; \ + mbuf00 = pkts[pkt00_index]; \ + \ + pkt01_index = __builtin_ctzll(pkts_mask); \ + if (pkts_mask == 0) \ + pkt01_index = pkt00_index; \ + \ + pkt01_mask = 1LLU << pkt01_index; \ + pkts_mask &= ~pkt01_mask; \ + mbuf01 = pkts[pkt01_index]; \ + \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \ +} + +#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)\ +{ \ + struct grinder *g10, *g11; \ + uint64_t sig10, sig11, bkt10_index, bkt11_index; \ + struct rte_mbuf *mbuf10, *mbuf11; \ + struct bucket *bkt10, *bkt11, *buckets = t->buckets; \ + uint8_t *key10, *key11; \ + uint64_t bucket_mask = t->bucket_mask; \ + rte_table_hash_op_hash f_hash = t->f_hash; \ + uint64_t seed = t->seed; \ + uint32_t key_size = t->key_size; \ + uint32_t key_offset = t->key_offset; \ + \ + mbuf10 = pkts[pkt10_index]; \ + key10 = RTE_MBUF_METADATA_UINT8_PTR(mbuf10, key_offset);\ + sig10 = (uint64_t) f_hash(key10, 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); \ + 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_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many)\ +{ \ + struct grinder *g20, *g21; \ + uint64_t sig20, sig21; \ + struct bucket *bkt20, *bkt21; \ + uint8_t *key20, *key21, *key_mem = t->key_mem; \ + uint64_t match20, match21, match_many20, match_many21; \ + uint64_t match_pos20, match_pos21; \ + uint32_t key20_index, key21_index, key_size_shl = t->key_size_shl;\ + \ + g20 = &g[pkt20_index]; \ + sig20 = g20->sig; \ + bkt20 = g20->bkt; \ + sig20 = (sig20 >> 16) | 1LLU; \ + lookup_cmp_sig(sig20, bkt20, match20, match_many20, match_pos20);\ + match20 <<= pkt20_index; \ + match_many20 <<= pkt20_index; \ + key20_index = bkt20->key_pos[match_pos20]; \ + key20 = &key_mem[key20_index << key_size_shl]; \ + \ + g21 = &g[pkt21_index]; \ + sig21 = g21->sig; \ + bkt21 = g21->bkt; \ + sig21 = (sig21 >> 16) | 1LLU; \ + lookup_cmp_sig(sig21, bkt21, match21, match_many21, match_pos21);\ + match21 <<= pkt21_index; \ + match_many21 <<= pkt21_index; \ + key21_index = bkt21->key_pos[match_pos21]; \ + key21 = &key_mem[key21_index << key_size_shl]; \ + \ + rte_prefetch0(key20); \ + rte_prefetch0(key21); \ + \ + pkts_mask_match_many |= match_many20 | match_many21; \ + \ + g20->match = match20; \ + g20->match_pos = match_pos20; \ + g20->key_index = key20_index; \ + \ + g21->match = match21; \ + g21->match_pos = match_pos21; \ + g21->key_index = key21_index; \ +} + +#define lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, \ + entries) \ +{ \ + struct grinder *g30, *g31; \ + struct rte_mbuf *mbuf30, *mbuf31; \ + struct bucket *bkt30, *bkt31; \ + uint8_t *key30, *key31, *key_mem = t->key_mem; \ + uint8_t *data30, *data31, *data_mem = t->data_mem; \ + uint64_t match30, match31, match_pos30, match_pos31; \ + uint64_t match_key30, match_key31, match_keys; \ + uint32_t key30_index, key31_index; \ + uint32_t key_size_shl = t->key_size_shl; \ + uint32_t data_size_shl = t->data_size_shl; \ + \ + mbuf30 = pkts[pkt30_index]; \ + g30 = &g[pkt30_index]; \ + bkt30 = g30->bkt; \ + match30 = g30->match; \ + match_pos30 = g30->match_pos; \ + key30_index = g30->key_index; \ + key30 = &key_mem[key30_index << key_size_shl]; \ + lookup_cmp_key(mbuf30, key30, match_key30, t); \ + match_key30 <<= pkt30_index; \ + match_key30 &= match30; \ + data30 = &data_mem[key30_index << data_size_shl]; \ + entries[pkt30_index] = data30; \ + \ + mbuf31 = pkts[pkt31_index]; \ + g31 = &g[pkt31_index]; \ + bkt31 = g31->bkt; \ + match31 = g31->match; \ + match_pos31 = g31->match_pos; \ + key31_index = g31->key_index; \ + key31 = &key_mem[key31_index << key_size_shl]; \ + lookup_cmp_key(mbuf31, key31, match_key31, t); \ + match_key31 <<= pkt31_index; \ + match_key31 &= match31; \ + data31 = &data_mem[key31_index << data_size_shl]; \ + entries[pkt31_index] = data31; \ + \ + rte_prefetch0(data30); \ + rte_prefetch0(data31); \ + \ + match_keys = match_key30 | match_key31; \ + pkts_mask_out |= match_keys; \ + \ + if (match_key30 == 0) \ + match_pos30 = 4; \ + lru_update(bkt30, match_pos30); \ + \ + if (match_key31 == 0) \ + match_pos31 = 4; \ + lru_update(bkt31, match_pos31); \ +} + +/*** +* The lookup function implements a 4-stage pipeline, with each stage processing +* two different packets. The purpose of pipelined implementation is to hide the +* latency of prefetching the data structures and loosen the data dependency +* between instructions. +* +* p00 _______ p10 _______ p20 _______ p30 _______ +* ----->| |----->| |----->| |----->| |-----> +* | 0 | | 1 | | 2 | | 3 | +* ----->|_______|----->|_______|----->|_______|----->|_______|-----> +* p01 p11 p21 p31 +* +* The naming convention is: +* pXY = packet Y of stage X, X = 0 .. 3, Y = 0 .. 1 +* +***/ +static int rte_table_hash_lru_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 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; + + /* 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); + + /* 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(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(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(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(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, 0); + pkts_mask_out |= pkts_mask_out_slow; + } + + *lookup_hit_mask = 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; + + /* 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_out |= pkts_mask_out_slow; + } + + *lookup_hit_mask = pkts_mask_out; + return status; +} + +struct rte_table_ops rte_table_hash_lru_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_lookup = rte_table_hash_lru_lookup, +}; + +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_lookup = rte_table_hash_lru_lookup_dosig, +};