diff options
Diffstat (limited to 'lib/librte_table')
| -rw-r--r-- | lib/librte_table/Makefile | 85 | ||||
| -rw-r--r-- | lib/librte_table/rte_lru.h | 213 | ||||
| -rw-r--r-- | lib/librte_table/rte_table.h | 301 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_acl.c | 835 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_acl.h | 95 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_array.c | 237 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_array.h | 76 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash.h | 370 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_ext.c | 1159 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_key16.c | 1515 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_key32.c | 1144 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_key8.c | 1471 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_hash_lru.c | 1102 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_lpm.c | 393 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_lpm.h | 124 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_lpm_ipv6.c | 398 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_lpm_ipv6.h | 122 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_stub.c | 121 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_stub.h | 62 | ||||
| -rw-r--r-- | lib/librte_table/rte_table_version.map | 28 |
20 files changed, 9851 insertions, 0 deletions
diff --git a/lib/librte_table/Makefile b/lib/librte_table/Makefile new file mode 100644 index 00000000..7f02af3c --- /dev/null +++ b/lib/librte_table/Makefile @@ -0,0 +1,85 @@ +# 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 $(RTE_SDK)/mk/rte.vars.mk + +# +# library name +# +LIB = librte_table.a + +CFLAGS += -O3 +CFLAGS += $(WERROR_FLAGS) + +EXPORT_MAP := rte_table_version.map + +LIBABIVER := 2 + +# +# all source are stored in SRCS-y +# +SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_lpm.c +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 +SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_array.c +SRCS-$(CONFIG_RTE_LIBRTE_TABLE) += rte_table_stub.c + +# install includes +SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_table.h +SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_table_lpm.h +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-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_lru.h +SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_table_array.h +SYMLINK-$(CONFIG_RTE_LIBRTE_TABLE)-include += rte_table_stub.h + +# this lib depends upon: +DEPDIRS-$(CONFIG_RTE_LIBRTE_TABLE) := lib/librte_eal +DEPDIRS-$(CONFIG_RTE_LIBRTE_TABLE) += lib/librte_mbuf +DEPDIRS-$(CONFIG_RTE_LIBRTE_TABLE) += lib/librte_mempool +DEPDIRS-$(CONFIG_RTE_LIBRTE_TABLE) += lib/librte_port +DEPDIRS-$(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 00000000..e87e062d --- /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 <stdint.h> + +#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 <x86intrin.h> +#else +#include <emmintrin.h> +#include <smmintrin.h> +#include <xmmintrin.h> +#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 <x86intrin.h> +#else +#include <emmintrin.h> +#include <smmintrin.h> +#include <xmmintrin.h> +#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.h b/lib/librte_table/rte_table.h new file mode 100644 index 00000000..d3446a56 --- /dev/null +++ b/lib/librte_table/rte_table.h @@ -0,0 +1,301 @@ +/*- + * 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_H__ +#define __INCLUDE_RTE_TABLE_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Table + * + * This tool is part of the DPDK Packet Framework tool suite and provides + * a standard interface to implement different types of lookup tables for data + * plane processing. + * + * Virtually any search algorithm that can uniquely associate data to a lookup + * key can be fitted under this lookup table abstraction. For the flow table + * use-case, the lookup key is an n-tuple of packet fields that uniquely + * identifies a traffic flow, while data represents actions and action + * meta-data associated with the same traffic flow. + * + ***/ + +#include <stdint.h> +#include <rte_port.h> + +struct rte_mbuf; + +/** Lookup table statistics */ +struct rte_table_stats { + uint64_t n_pkts_in; + uint64_t n_pkts_lookup_miss; +}; + +/** + * Lookup table create + * + * @param params + * Parameters for lookup table creation. The underlying data structure is + * different for each lookup table type. + * @param socket_id + * CPU socket ID (e.g. for memory allocation purpose) + * @param entry_size + * Data size of each lookup table entry (measured in bytes) + * @return + * Handle to lookup table instance + */ +typedef void* (*rte_table_op_create)(void *params, int socket_id, + uint32_t entry_size); + +/** + * Lookup table free + * + * @param table + * Handle to lookup table instance + * @return + * 0 on success, error code otherwise + */ +typedef int (*rte_table_op_free)(void *table); + +/** + * Lookup table entry add + * + * @param table + * Handle to lookup table instance + * @param key + * Lookup key + * @param entry + * Data to be associated with the current key. This parameter has to point to + * a valid memory buffer where the first entry_size bytes (table create + * parameter) are populated with the data. + * @param key_found + * After successful invocation, *key_found is set to a value different than 0 + * if the current key is already present in the table and to 0 if not. This + * pointer has to be set to a valid memory location before the table entry add + * function is called. + * @param entry_ptr + * After successful invocation, *entry_ptr stores the handle to the table + * entry containing the data associated with the current key. This handle can + * be used to perform further read-write accesses to this entry. This handle + * is valid until the key is deleted from the table or the same key is + * re-added to the table, typically to associate it with different data. This + * pointer has to be set to a valid memory location before the function is + * called. + * @return + * 0 on success, error code otherwise + */ +typedef int (*rte_table_op_entry_add)( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr); + +/** + * Lookup table entry delete + * + * @param table + * Handle to lookup table instance + * @param key + * Lookup key + * @param key_found + * After successful invocation, *key_found is set to a value different than 0 + * if the current key was present in the table before the delete operation + * was performed and to 0 if not. This pointer has to be set to a valid + * memory location before the table entry delete function is called. + * @param entry + * After successful invocation, if the key is found in the table (*key found + * is different than 0 after function call is completed) and entry points to + * a valid buffer (entry is set to a value different than NULL before the + * function is called), then the first entry_size bytes (table create + * parameter) in *entry store a copy of table entry that contained the data + * associated with the current key before the key was deleted. + * @return + * 0 on success, error code otherwise + */ +typedef int (*rte_table_op_entry_delete)( + void *table, + void *key, + int *key_found, + void *entry); + +/** + * Lookup table entry add bulk + * + * @param table + * Handle to lookup table instance + * @param key + * Array containing lookup keys + * @param entries + * Array containing data to be associated with each key. Every item in the + * array has to point to a valid memory buffer where the first entry_size + * bytes (table create parameter) are populated with the data. + * @param n_keys + * Number of keys to add + * @param key_found + * After successful invocation, key_found for every item in the array is set + * to a value different than 0 if the current key is already present in the + * table and to 0 if not. This pointer has to be set to a valid memory + * location before the table entry add function is called. + * @param entries_ptr + * After successful invocation, array *entries_ptr stores the handle to the + * table entry containing the data associated with every key. This handle can + * be used to perform further read-write accesses to this entry. This handle + * is valid until the key is deleted from the table or the same key is + * re-added to the table, typically to associate it with different data. This + * pointer has to be set to a valid memory location before the function is + * called. + * @return + * 0 on success, error code otherwise + */ +typedef int (*rte_table_op_entry_add_bulk)( + void *table, + void **keys, + void **entries, + uint32_t n_keys, + int *key_found, + void **entries_ptr); + +/** + * Lookup table entry delete bulk + * + * @param table + * Handle to lookup table instance + * @param key + * Array containing lookup keys + * @param n_keys + * Number of keys to delete + * @param key_found + * After successful invocation, key_found for every item in the array is set + * to a value different than 0if the current key was present in the table + * before the delete operation was performed and to 0 if not. This pointer + * has to be set to a valid memory location before the table entry delete + * function is called. + * @param entries + * If entries pointer is NULL, this pointer is ignored for every entry found. + * Else, after successful invocation, if specific key is found in the table + * (key_found is different than 0 for this item after function call is + * completed) and item of entry array points to a valid buffer (entry is set + * to a value different than NULL before the function is called), then the + * first entry_size bytes (table create parameter) in *entry store a copy of + * table entry that contained the data associated with the current key before + * the key was deleted. + * @return + * 0 on success, error code otherwise + */ +typedef int (*rte_table_op_entry_delete_bulk)( + void *table, + void **keys, + uint32_t n_keys, + int *key_found, + void **entries); + +/** + * Lookup table lookup + * + * @param table + * Handle to lookup table instance + * @param pkts + * Burst of input packets specified as array of up to 64 pointers to struct + * rte_mbuf + * @param pkts_mask + * 64-bit bitmask specifying which packets in the input burst are valid. When + * pkts_mask bit n is set, then element n of pkts array is pointing to a + * valid packet. Otherwise, element n of pkts array does not point to a valid + * packet, therefore it will not be accessed. + * @param lookup_hit_mask + * Once the table lookup operation is completed, this 64-bit bitmask + * specifies which of the valid packets in the input burst resulted in lookup + * hit. For each valid input packet (pkts_mask bit n is set), the following + * are true on lookup hit: lookup_hit_mask bit n is set, element n of entries + * array is valid and it points to the lookup table entry that was hit. For + * each valid input packet (pkts_mask bit n is set), the following are true + * on lookup miss: lookup_hit_mask bit n is not set and element n of entries + * array is not valid. + * @param entries + * Once the table lookup operation is completed, this array provides the + * lookup table entries that were hit, as described above. It is required + * that this array is always pre-allocated by the caller of this function + * with exactly 64 elements. The implementation is allowed to speculatively + * modify the elements of this array, so elements marked as invalid in + * lookup_hit_mask once the table lookup operation is completed might have + * been modified by this function. + * @return + * 0 on success, error code otherwise + */ +typedef int (*rte_table_op_lookup)( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries); + +/** + * Lookup table stats read + * + * @param table + * Handle to lookup table instance + * @param stats + * Handle to table stats struct to copy data + * @param clear + * Flag indicating that stats should be cleared after read + * + * @return + * Error code or 0 on success. + */ +typedef int (*rte_table_op_stats_read)( + void *table, + struct rte_table_stats *stats, + int clear); + +/** Lookup table interface defining the lookup table operation */ +struct rte_table_ops { + rte_table_op_create f_create; /**< Create */ + rte_table_op_free f_free; /**< Free */ + rte_table_op_entry_add f_add; /**< Entry add */ + rte_table_op_entry_delete f_delete; /**< Entry delete */ + rte_table_op_entry_add_bulk f_add_bulk; /**< Add entry bulk */ + rte_table_op_entry_delete_bulk f_delete_bulk; /**< Delete entry bulk */ + rte_table_op_lookup f_lookup; /**< Lookup */ + rte_table_op_stats_read f_stats; /**< Stats */ +}; + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/lib/librte_table/rte_table_acl.c b/lib/librte_table/rte_table_acl.c new file mode 100644 index 00000000..c1eb8488 --- /dev/null +++ b/lib/librte_table/rte_table_acl.c @@ -0,0 +1,835 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_log.h> + +#include "rte_table_acl.h" +#include <rte_ether.h> + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_ACL_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_ACL_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_ACL_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_ACL_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +struct rte_table_acl { + struct rte_table_stats stats; + + /* Low-level ACL table */ + char name[2][RTE_ACL_NAMESIZE]; + struct rte_acl_param acl_params; /* for creating low level acl table */ + struct rte_acl_config cfg; /* Holds the field definitions (metadata) */ + struct rte_acl_ctx *ctx; + uint32_t name_id; + + /* Input parameters */ + uint32_t n_rules; + uint32_t entry_size; + + /* Internal tables */ + uint8_t *action_table; + struct rte_acl_rule **acl_rule_list; /* Array of pointers to rules */ + uint8_t *acl_rule_memory; /* Memory to store the rules */ + + /* Memory to store the action table and stack of free entries */ + uint8_t memory[0] __rte_cache_aligned; +}; + + +static void * +rte_table_acl_create( + void *params, + int socket_id, + uint32_t entry_size) +{ + struct rte_table_acl_params *p = (struct rte_table_acl_params *) params; + struct rte_table_acl *acl; + uint32_t action_table_size, acl_rule_list_size, acl_rule_memory_size; + uint32_t total_size; + + RTE_BUILD_BUG_ON(((sizeof(struct rte_table_acl) % RTE_CACHE_LINE_SIZE) + != 0)); + + /* Check input parameters */ + if (p == NULL) { + RTE_LOG(ERR, TABLE, "%s: Invalid value for params\n", __func__); + return NULL; + } + if (p->name == NULL) { + RTE_LOG(ERR, TABLE, "%s: Invalid value for name\n", __func__); + return NULL; + } + if (p->n_rules == 0) { + RTE_LOG(ERR, TABLE, "%s: Invalid value for n_rules\n", + __func__); + return NULL; + } + if ((p->n_rule_fields == 0) || + (p->n_rule_fields > RTE_ACL_MAX_FIELDS)) { + RTE_LOG(ERR, TABLE, "%s: Invalid value for n_rule_fields\n", + __func__); + return NULL; + } + + entry_size = RTE_ALIGN(entry_size, sizeof(uint64_t)); + + /* Memory allocation */ + action_table_size = RTE_CACHE_LINE_ROUNDUP(p->n_rules * entry_size); + acl_rule_list_size = + RTE_CACHE_LINE_ROUNDUP(p->n_rules * sizeof(struct rte_acl_rule *)); + acl_rule_memory_size = RTE_CACHE_LINE_ROUNDUP(p->n_rules * + RTE_ACL_RULE_SZ(p->n_rule_fields)); + total_size = sizeof(struct rte_table_acl) + action_table_size + + acl_rule_list_size + acl_rule_memory_size; + + acl = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, + socket_id); + if (acl == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for ACL table\n", + __func__, total_size); + return NULL; + } + + acl->action_table = &acl->memory[0]; + acl->acl_rule_list = + (struct rte_acl_rule **) &acl->memory[action_table_size]; + acl->acl_rule_memory = (uint8_t *) + &acl->memory[action_table_size + acl_rule_list_size]; + + /* Initialization of internal fields */ + snprintf(acl->name[0], RTE_ACL_NAMESIZE, "%s_a", p->name); + snprintf(acl->name[1], RTE_ACL_NAMESIZE, "%s_b", p->name); + acl->name_id = 1; + + acl->acl_params.name = acl->name[acl->name_id]; + acl->acl_params.socket_id = socket_id; + acl->acl_params.rule_size = RTE_ACL_RULE_SZ(p->n_rule_fields); + acl->acl_params.max_rule_num = p->n_rules; + + acl->cfg.num_categories = 1; + acl->cfg.num_fields = p->n_rule_fields; + memcpy(&acl->cfg.defs[0], &p->field_format[0], + p->n_rule_fields * sizeof(struct rte_acl_field_def)); + + acl->ctx = NULL; + + acl->n_rules = p->n_rules; + acl->entry_size = entry_size; + + return acl; +} + +static int +rte_table_acl_free(void *table) +{ + struct rte_table_acl *acl = (struct rte_table_acl *) table; + + /* Check input parameters */ + if (table == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + /* Free previously allocated resources */ + if (acl->ctx != NULL) + rte_acl_free(acl->ctx); + + rte_free(acl); + + return 0; +} + +RTE_ACL_RULE_DEF(rte_pipeline_acl_rule, RTE_ACL_MAX_FIELDS); + +static int +rte_table_acl_build(struct rte_table_acl *acl, struct rte_acl_ctx **acl_ctx) +{ + struct rte_acl_ctx *ctx = NULL; + uint32_t n_rules, i; + int status; + + /* Create low level ACL table */ + ctx = rte_acl_create(&acl->acl_params); + if (ctx == NULL) { + RTE_LOG(ERR, TABLE, "%s: Cannot create low level ACL table\n", + __func__); + return -1; + } + + /* Add rules to low level ACL table */ + n_rules = 0; + for (i = 1; i < acl->n_rules; i++) { + if (acl->acl_rule_list[i] != NULL) { + status = rte_acl_add_rules(ctx, acl->acl_rule_list[i], + 1); + if (status != 0) { + RTE_LOG(ERR, TABLE, + "%s: Cannot add rule to low level ACL table\n", + __func__); + rte_acl_free(ctx); + return -1; + } + + n_rules++; + } + } + + if (n_rules == 0) { + rte_acl_free(ctx); + *acl_ctx = NULL; + return 0; + } + + /* Build low level ACl table */ + status = rte_acl_build(ctx, &acl->cfg); + if (status != 0) { + RTE_LOG(ERR, TABLE, + "%s: Cannot build the low level ACL table\n", + __func__); + rte_acl_free(ctx); + return -1; + } + + rte_acl_dump(ctx); + + *acl_ctx = ctx; + return 0; +} + +static int +rte_table_acl_entry_add( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_acl *acl = (struct rte_table_acl *) table; + struct rte_table_acl_rule_add_params *rule = + (struct rte_table_acl_rule_add_params *) key; + struct rte_pipeline_acl_rule acl_rule; + struct rte_acl_rule *rule_location; + struct rte_acl_ctx *ctx; + uint32_t free_pos, free_pos_valid, i; + int status; + + /* Check input parameters */ + if (table == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (key == NULL) { + RTE_LOG(ERR, TABLE, "%s: key parameter is NULL\n", __func__); + return -EINVAL; + } + if (entry == NULL) { + RTE_LOG(ERR, TABLE, "%s: entry parameter is NULL\n", __func__); + return -EINVAL; + } + if (key_found == NULL) { + RTE_LOG(ERR, TABLE, "%s: key_found parameter is NULL\n", + __func__); + return -EINVAL; + } + if (entry_ptr == NULL) { + RTE_LOG(ERR, TABLE, "%s: entry_ptr parameter is NULL\n", + __func__); + return -EINVAL; + } + if (rule->priority > RTE_ACL_MAX_PRIORITY) { + RTE_LOG(ERR, TABLE, "%s: Priority is too high\n", __func__); + return -EINVAL; + } + + /* Setup rule data structure */ + memset(&acl_rule, 0, sizeof(acl_rule)); + acl_rule.data.category_mask = 1; + acl_rule.data.priority = RTE_ACL_MAX_PRIORITY - rule->priority; + acl_rule.data.userdata = 0; /* To be set up later */ + memcpy(&acl_rule.field[0], + &rule->field_value[0], + acl->cfg.num_fields * sizeof(struct rte_acl_field)); + + /* Look to see if the rule exists already in the table */ + free_pos = 0; + free_pos_valid = 0; + for (i = 1; i < acl->n_rules; i++) { + if (acl->acl_rule_list[i] == NULL) { + if (free_pos_valid == 0) { + free_pos = i; + free_pos_valid = 1; + } + + continue; + } + + /* Compare the key fields */ + status = memcmp(&acl->acl_rule_list[i]->field[0], + &rule->field_value[0], + acl->cfg.num_fields * sizeof(struct rte_acl_field)); + + /* Rule found: update data associated with the rule */ + if (status == 0) { + *key_found = 1; + *entry_ptr = &acl->memory[i * acl->entry_size]; + memcpy(*entry_ptr, entry, acl->entry_size); + + return 0; + } + } + + /* Return if max rules */ + if (free_pos_valid == 0) { + RTE_LOG(ERR, TABLE, "%s: Max number of rules reached\n", + __func__); + return -ENOSPC; + } + + /* Add the new rule to the rule set */ + acl_rule.data.userdata = free_pos; + rule_location = (struct rte_acl_rule *) + &acl->acl_rule_memory[free_pos * acl->acl_params.rule_size]; + memcpy(rule_location, &acl_rule, acl->acl_params.rule_size); + acl->acl_rule_list[free_pos] = rule_location; + + /* Build low level ACL table */ + acl->name_id ^= 1; + acl->acl_params.name = acl->name[acl->name_id]; + status = rte_table_acl_build(acl, &ctx); + if (status != 0) { + /* Roll back changes */ + acl->acl_rule_list[free_pos] = NULL; + acl->name_id ^= 1; + + return -EINVAL; + } + + /* Commit changes */ + if (acl->ctx != NULL) + rte_acl_free(acl->ctx); + acl->ctx = ctx; + *key_found = 0; + *entry_ptr = &acl->memory[free_pos * acl->entry_size]; + memcpy(*entry_ptr, entry, acl->entry_size); + + return 0; +} + +static int +rte_table_acl_entry_delete( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_acl *acl = (struct rte_table_acl *) table; + struct rte_table_acl_rule_delete_params *rule = + (struct rte_table_acl_rule_delete_params *) key; + struct rte_acl_rule *deleted_rule = NULL; + struct rte_acl_ctx *ctx; + uint32_t pos, pos_valid, i; + int status; + + /* Check input parameters */ + if (table == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (key == NULL) { + RTE_LOG(ERR, TABLE, "%s: key parameter is NULL\n", __func__); + return -EINVAL; + } + if (key_found == NULL) { + RTE_LOG(ERR, TABLE, "%s: key_found parameter is NULL\n", + __func__); + return -EINVAL; + } + + /* Look for the rule in the table */ + pos = 0; + pos_valid = 0; + for (i = 1; i < acl->n_rules; i++) { + if (acl->acl_rule_list[i] != NULL) { + /* Compare the key fields */ + status = memcmp(&acl->acl_rule_list[i]->field[0], + &rule->field_value[0], acl->cfg.num_fields * + sizeof(struct rte_acl_field)); + + /* Rule found: remove from table */ + if (status == 0) { + pos = i; + pos_valid = 1; + + deleted_rule = acl->acl_rule_list[i]; + acl->acl_rule_list[i] = NULL; + } + } + } + + /* Return if rule not found */ + if (pos_valid == 0) { + *key_found = 0; + return 0; + } + + /* Build low level ACL table */ + acl->name_id ^= 1; + acl->acl_params.name = acl->name[acl->name_id]; + status = rte_table_acl_build(acl, &ctx); + if (status != 0) { + /* Roll back changes */ + acl->acl_rule_list[pos] = deleted_rule; + acl->name_id ^= 1; + + return -EINVAL; + } + + /* Commit changes */ + if (acl->ctx != NULL) + rte_acl_free(acl->ctx); + + acl->ctx = ctx; + *key_found = 1; + if (entry != NULL) + memcpy(entry, &acl->memory[pos * acl->entry_size], + acl->entry_size); + + return 0; +} + +static int +rte_table_acl_entry_add_bulk( + void *table, + void **keys, + void **entries, + uint32_t n_keys, + int *key_found, + void **entries_ptr) +{ + struct rte_table_acl *acl = (struct rte_table_acl *) table; + struct rte_acl_ctx *ctx; + uint32_t rule_pos[n_keys]; + uint32_t i; + int err = 0, build = 0; + int status; + + /* Check input parameters */ + if (table == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (keys == NULL) { + RTE_LOG(ERR, TABLE, "%s: keys parameter is NULL\n", __func__); + return -EINVAL; + } + if (entries == NULL) { + RTE_LOG(ERR, TABLE, "%s: entries parameter is NULL\n", __func__); + return -EINVAL; + } + if (n_keys == 0) { + RTE_LOG(ERR, TABLE, "%s: 0 rules to add\n", __func__); + return -EINVAL; + } + if (key_found == NULL) { + RTE_LOG(ERR, TABLE, "%s: key_found parameter is NULL\n", + __func__); + return -EINVAL; + } + if (entries_ptr == NULL) { + RTE_LOG(ERR, TABLE, "%s: entries_ptr parameter is NULL\n", + __func__); + return -EINVAL; + } + + /* Check input parameters in arrays */ + for (i = 0; i < n_keys; i++) { + struct rte_table_acl_rule_add_params *rule; + + if (keys[i] == NULL) { + RTE_LOG(ERR, TABLE, "%s: keys[%" PRIu32 "] parameter is NULL\n", + __func__, i); + return -EINVAL; + } + + if (entries[i] == NULL) { + RTE_LOG(ERR, TABLE, "%s: entries[%" PRIu32 "] parameter is NULL\n", + __func__, i); + return -EINVAL; + } + + if (entries_ptr[i] == NULL) { + RTE_LOG(ERR, TABLE, "%s: entries_ptr[%" PRIu32 "] parameter is NULL\n", + __func__, i); + return -EINVAL; + } + + rule = (struct rte_table_acl_rule_add_params *) keys[i]; + if (rule->priority > RTE_ACL_MAX_PRIORITY) { + RTE_LOG(ERR, TABLE, "%s: Priority is too high\n", __func__); + return -EINVAL; + } + } + + memset(rule_pos, 0, n_keys * sizeof(uint32_t)); + memset(key_found, 0, n_keys * sizeof(int)); + for (i = 0; i < n_keys; i++) { + struct rte_table_acl_rule_add_params *rule = + (struct rte_table_acl_rule_add_params *) keys[i]; + struct rte_pipeline_acl_rule acl_rule; + struct rte_acl_rule *rule_location; + uint32_t free_pos, free_pos_valid, j; + + /* Setup rule data structure */ + memset(&acl_rule, 0, sizeof(acl_rule)); + acl_rule.data.category_mask = 1; + acl_rule.data.priority = RTE_ACL_MAX_PRIORITY - rule->priority; + acl_rule.data.userdata = 0; /* To be set up later */ + memcpy(&acl_rule.field[0], + &rule->field_value[0], + acl->cfg.num_fields * sizeof(struct rte_acl_field)); + + /* Look to see if the rule exists already in the table */ + free_pos = 0; + free_pos_valid = 0; + for (j = 1; j < acl->n_rules; j++) { + if (acl->acl_rule_list[j] == NULL) { + if (free_pos_valid == 0) { + free_pos = j; + free_pos_valid = 1; + } + + continue; + } + + /* Compare the key fields */ + status = memcmp(&acl->acl_rule_list[j]->field[0], + &rule->field_value[0], + acl->cfg.num_fields * sizeof(struct rte_acl_field)); + + /* Rule found: update data associated with the rule */ + if (status == 0) { + key_found[i] = 1; + entries_ptr[i] = &acl->memory[j * acl->entry_size]; + memcpy(entries_ptr[i], entries[i], acl->entry_size); + + break; + } + } + + /* Key already in the table */ + if (key_found[i] != 0) + continue; + + /* Maximum number of rules reached */ + if (free_pos_valid == 0) { + err = 1; + break; + } + + /* Add the new rule to the rule set */ + acl_rule.data.userdata = free_pos; + rule_location = (struct rte_acl_rule *) + &acl->acl_rule_memory[free_pos * acl->acl_params.rule_size]; + memcpy(rule_location, &acl_rule, acl->acl_params.rule_size); + acl->acl_rule_list[free_pos] = rule_location; + rule_pos[i] = free_pos; + build = 1; + } + + if (err != 0) { + for (i = 0; i < n_keys; i++) { + if (rule_pos[i] == 0) + continue; + + acl->acl_rule_list[rule_pos[i]] = NULL; + } + + return -ENOSPC; + } + + if (build == 0) + return 0; + + /* Build low level ACL table */ + acl->name_id ^= 1; + acl->acl_params.name = acl->name[acl->name_id]; + status = rte_table_acl_build(acl, &ctx); + if (status != 0) { + /* Roll back changes */ + for (i = 0; i < n_keys; i++) { + if (rule_pos[i] == 0) + continue; + + acl->acl_rule_list[rule_pos[i]] = NULL; + } + acl->name_id ^= 1; + + return -EINVAL; + } + + /* Commit changes */ + if (acl->ctx != NULL) + rte_acl_free(acl->ctx); + acl->ctx = ctx; + + for (i = 0; i < n_keys; i++) { + if (rule_pos[i] == 0) + continue; + + key_found[i] = 0; + entries_ptr[i] = &acl->memory[rule_pos[i] * acl->entry_size]; + memcpy(entries_ptr[i], entries[i], acl->entry_size); + } + + return 0; +} + +static int +rte_table_acl_entry_delete_bulk( + void *table, + void **keys, + uint32_t n_keys, + int *key_found, + void **entries) +{ + struct rte_table_acl *acl = (struct rte_table_acl *) table; + struct rte_acl_rule *deleted_rules[n_keys]; + uint32_t rule_pos[n_keys]; + struct rte_acl_ctx *ctx; + uint32_t i; + int status; + int build = 0; + + /* Check input parameters */ + if (table == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (keys == NULL) { + RTE_LOG(ERR, TABLE, "%s: key parameter is NULL\n", __func__); + return -EINVAL; + } + if (n_keys == 0) { + RTE_LOG(ERR, TABLE, "%s: 0 rules to delete\n", __func__); + return -EINVAL; + } + if (key_found == NULL) { + RTE_LOG(ERR, TABLE, "%s: key_found parameter is NULL\n", + __func__); + return -EINVAL; + } + + for (i = 0; i < n_keys; i++) { + if (keys[i] == NULL) { + RTE_LOG(ERR, TABLE, "%s: keys[%" PRIu32 "] parameter is NULL\n", + __func__, i); + return -EINVAL; + } + } + + memset(deleted_rules, 0, n_keys * sizeof(struct rte_acl_rule *)); + memset(rule_pos, 0, n_keys * sizeof(uint32_t)); + for (i = 0; i < n_keys; i++) { + struct rte_table_acl_rule_delete_params *rule = + (struct rte_table_acl_rule_delete_params *) keys[i]; + uint32_t pos_valid, j; + + /* Look for the rule in the table */ + pos_valid = 0; + for (j = 1; j < acl->n_rules; j++) { + if (acl->acl_rule_list[j] == NULL) + continue; + + /* Compare the key fields */ + status = memcmp(&acl->acl_rule_list[j]->field[0], + &rule->field_value[0], + acl->cfg.num_fields * sizeof(struct rte_acl_field)); + + /* Rule found: remove from table */ + if (status == 0) { + pos_valid = 1; + + deleted_rules[i] = acl->acl_rule_list[j]; + acl->acl_rule_list[j] = NULL; + rule_pos[i] = j; + + build = 1; + } + } + + if (pos_valid == 0) { + key_found[i] = 0; + continue; + } + } + + /* Return if no changes to acl table */ + if (build == 0) { + return 0; + } + + /* Build low level ACL table */ + acl->name_id ^= 1; + acl->acl_params.name = acl->name[acl->name_id]; + status = rte_table_acl_build(acl, &ctx); + if (status != 0) { + /* Roll back changes */ + for (i = 0; i < n_keys; i++) { + if (rule_pos[i] == 0) + continue; + + acl->acl_rule_list[rule_pos[i]] = deleted_rules[i]; + } + + acl->name_id ^= 1; + + return -EINVAL; + } + + /* Commit changes */ + if (acl->ctx != NULL) + rte_acl_free(acl->ctx); + + acl->ctx = ctx; + for (i = 0; i < n_keys; i++) { + if (rule_pos[i] == 0) + continue; + + key_found[i] = 1; + if (entries != NULL && entries[i] != NULL) + memcpy(entries[i], &acl->memory[rule_pos[i] * acl->entry_size], + acl->entry_size); + } + + return 0; +} + +static int +rte_table_acl_lookup( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_acl *acl = (struct rte_table_acl *) table; + const uint8_t *pkts_data[RTE_PORT_IN_BURST_SIZE_MAX]; + uint32_t results[RTE_PORT_IN_BURST_SIZE_MAX]; + uint64_t pkts_out_mask; + uint32_t n_pkts, i, j; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_ACL_STATS_PKTS_IN_ADD(acl, n_pkts_in); + + /* Input conversion */ + for (i = 0, j = 0; i < (uint32_t)(RTE_PORT_IN_BURST_SIZE_MAX - + __builtin_clzll(pkts_mask)); i++) { + uint64_t pkt_mask = 1LLU << i; + + if (pkt_mask & pkts_mask) { + pkts_data[j] = rte_pktmbuf_mtod(pkts[i], uint8_t *); + j++; + } + } + n_pkts = j; + + /* Low-level ACL table lookup */ + if (acl->ctx != NULL) + rte_acl_classify(acl->ctx, pkts_data, results, n_pkts, 1); + else + n_pkts = 0; + + /* Output conversion */ + pkts_out_mask = 0; + for (i = 0; i < n_pkts; i++) { + uint32_t action_table_pos = results[i]; + uint32_t pkt_pos = __builtin_ctzll(pkts_mask); + uint64_t pkt_mask = 1LLU << pkt_pos; + + pkts_mask &= ~pkt_mask; + + if (action_table_pos != RTE_ACL_INVALID_USERDATA) { + pkts_out_mask |= pkt_mask; + entries[pkt_pos] = (void *) + &acl->memory[action_table_pos * + acl->entry_size]; + rte_prefetch0(entries[pkt_pos]); + } + } + + *lookup_hit_mask = pkts_out_mask; + RTE_TABLE_ACL_STATS_PKTS_LOOKUP_MISS(acl, n_pkts_in - __builtin_popcountll(pkts_out_mask)); + + return 0; +} + +static int +rte_table_acl_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_acl *acl = (struct rte_table_acl *) table; + + if (stats != NULL) + memcpy(stats, &acl->stats, sizeof(acl->stats)); + + if (clear) + memset(&acl->stats, 0, sizeof(acl->stats)); + + return 0; +} + +struct rte_table_ops rte_table_acl_ops = { + .f_create = rte_table_acl_create, + .f_free = rte_table_acl_free, + .f_add = rte_table_acl_entry_add, + .f_delete = rte_table_acl_entry_delete, + .f_add_bulk = rte_table_acl_entry_add_bulk, + .f_delete_bulk = rte_table_acl_entry_delete_bulk, + .f_lookup = rte_table_acl_lookup, + .f_stats = rte_table_acl_stats_read, +}; diff --git a/lib/librte_table/rte_table_acl.h b/lib/librte_table/rte_table_acl.h new file mode 100644 index 00000000..a9cc0328 --- /dev/null +++ b/lib/librte_table/rte_table_acl.h @@ -0,0 +1,95 @@ +/*- + * 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_ACL_H__ +#define __INCLUDE_RTE_TABLE_ACL_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Table ACL + * + * This table uses the Access Control List (ACL) algorithm to uniquely + * associate data to lookup keys. + * + * Use-cases: Firewall rule database, etc. + * + ***/ + +#include <stdint.h> + +#include "rte_acl.h" + +#include "rte_table.h" + +/** ACL table parameters */ +struct rte_table_acl_params { + /** Name */ + const char *name; + + /** Maximum number of ACL rules in the table */ + uint32_t n_rules; + + /** Number of fields in the ACL rule specification */ + uint32_t n_rule_fields; + + /** Format specification of the fields of the ACL rule */ + struct rte_acl_field_def field_format[RTE_ACL_MAX_FIELDS]; +}; + +/** ACL rule specification for entry add operation */ +struct rte_table_acl_rule_add_params { + /** ACL rule priority, with 0 as the highest priority */ + int32_t priority; + + /** Values for the fields of the ACL rule to be added to the table */ + struct rte_acl_field field_value[RTE_ACL_MAX_FIELDS]; +}; + +/** ACL rule specification for entry delete operation */ +struct rte_table_acl_rule_delete_params { + /** Values for the fields of the ACL rule to be deleted from table */ + struct rte_acl_field field_value[RTE_ACL_MAX_FIELDS]; +}; + +/** ACL table operations */ +extern struct rte_table_ops rte_table_acl_ops; + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/lib/librte_table/rte_table_array.c b/lib/librte_table/rte_table_array.c new file mode 100644 index 00000000..3bb68d11 --- /dev/null +++ b/lib/librte_table/rte_table_array.c @@ -0,0 +1,237 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_log.h> + +#include "rte_table_array.h" + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_ARRAY_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_ARRAY_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_ARRAY_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_ARRAY_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +struct rte_table_array { + struct rte_table_stats stats; + + /* Input parameters */ + uint32_t entry_size; + uint32_t n_entries; + uint32_t offset; + + /* Internal fields */ + uint32_t entry_pos_mask; + + /* Internal table */ + uint8_t array[0] __rte_cache_aligned; +} __rte_cache_aligned; + +static void * +rte_table_array_create(void *params, int socket_id, uint32_t entry_size) +{ + struct rte_table_array_params *p = + (struct rte_table_array_params *) params; + struct rte_table_array *t; + uint32_t total_cl_size, total_size; + + /* Check input parameters */ + if ((p == NULL) || + (p->n_entries == 0) || + (!rte_is_power_of_2(p->n_entries))) + return NULL; + + /* Memory allocation */ + total_cl_size = (sizeof(struct rte_table_array) + + RTE_CACHE_LINE_SIZE) / RTE_CACHE_LINE_SIZE; + total_cl_size += (p->n_entries * entry_size + + RTE_CACHE_LINE_SIZE) / RTE_CACHE_LINE_SIZE; + total_size = total_cl_size * RTE_CACHE_LINE_SIZE; + t = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + if (t == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for array table\n", + __func__, total_size); + return NULL; + } + + /* Memory initialization */ + t->entry_size = entry_size; + t->n_entries = p->n_entries; + t->offset = p->offset; + t->entry_pos_mask = t->n_entries - 1; + + return t; +} + +static int +rte_table_array_free(void *table) +{ + struct rte_table_array *t = (struct rte_table_array *) table; + + /* Check input parameters */ + if (t == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + /* Free previously allocated resources */ + rte_free(t); + + return 0; +} + +static int +rte_table_array_entry_add( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_array *t = (struct rte_table_array *) table; + struct rte_table_array_key *k = (struct rte_table_array_key *) key; + uint8_t *table_entry; + + /* Check input parameters */ + if (table == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (key == NULL) { + RTE_LOG(ERR, TABLE, "%s: key parameter is NULL\n", __func__); + return -EINVAL; + } + if (entry == NULL) { + RTE_LOG(ERR, TABLE, "%s: entry parameter is NULL\n", __func__); + return -EINVAL; + } + if (key_found == NULL) { + RTE_LOG(ERR, TABLE, "%s: key_found parameter is NULL\n", + __func__); + return -EINVAL; + } + if (entry_ptr == NULL) { + RTE_LOG(ERR, TABLE, "%s: entry_ptr parameter is NULL\n", + __func__); + return -EINVAL; + } + + table_entry = &t->array[k->pos * t->entry_size]; + memcpy(table_entry, entry, t->entry_size); + *key_found = 1; + *entry_ptr = (void *) table_entry; + + return 0; +} + +static int +rte_table_array_lookup( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_array *t = (struct rte_table_array *) table; + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_ARRAY_STATS_PKTS_IN_ADD(t, n_pkts_in); + *lookup_hit_mask = pkts_mask; + + if ((pkts_mask & (pkts_mask + 1)) == 0) { + uint64_t n_pkts = __builtin_popcountll(pkts_mask); + uint32_t i; + + for (i = 0; i < n_pkts; i++) { + struct rte_mbuf *pkt = pkts[i]; + uint32_t entry_pos = RTE_MBUF_METADATA_UINT32(pkt, + t->offset) & t->entry_pos_mask; + + entries[i] = (void *) &t->array[entry_pos * + t->entry_size]; + } + } else { + for ( ; pkts_mask; ) { + uint32_t pkt_index = __builtin_ctzll(pkts_mask); + uint64_t pkt_mask = 1LLU << pkt_index; + struct rte_mbuf *pkt = pkts[pkt_index]; + uint32_t entry_pos = RTE_MBUF_METADATA_UINT32(pkt, + t->offset) & t->entry_pos_mask; + + entries[pkt_index] = (void *) &t->array[entry_pos * + t->entry_size]; + pkts_mask &= ~pkt_mask; + } + } + + return 0; +} + +static int +rte_table_array_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_array *array = (struct rte_table_array *) table; + + if (stats != NULL) + memcpy(stats, &array->stats, sizeof(array->stats)); + + if (clear) + memset(&array->stats, 0, sizeof(array->stats)); + + return 0; +} + +struct rte_table_ops rte_table_array_ops = { + .f_create = rte_table_array_create, + .f_free = rte_table_array_free, + .f_add = rte_table_array_entry_add, + .f_delete = NULL, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_array_lookup, + .f_stats = rte_table_array_stats_read, +}; diff --git a/lib/librte_table/rte_table_array.h b/lib/librte_table/rte_table_array.h new file mode 100644 index 00000000..9521119e --- /dev/null +++ b/lib/librte_table/rte_table_array.h @@ -0,0 +1,76 @@ +/*- + * 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_ARRAY_H__ +#define __INCLUDE_RTE_TABLE_ARRAY_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Table Array + * + * Simple array indexing. Lookup key is the array entry index. + * + ***/ + +#include <stdint.h> + +#include "rte_table.h" + +/** Array table parameters */ +struct rte_table_array_params { + /** Number of array entries. Has to be a power of two. */ + uint32_t n_entries; + + /** Byte offset within input packet meta-data where lookup key (i.e. the + array entry index) is located. */ + uint32_t offset; +}; + +/** Array table key format */ +struct rte_table_array_key { + /** Array entry index */ + uint32_t pos; +}; + +/** Array table operations */ +extern struct rte_table_ops rte_table_array_ops; + +#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 00000000..9d17516a --- /dev/null +++ b/lib/librte_table/rte_table_hash.h @@ -0,0 +1,370 @@ +/*- + * 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 <stdint.h> + +#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; + + /** Bit-mask to be AND-ed to the key on lookup */ + uint8_t *key_mask; +}; + +/** LRU hash table operations for pre-computed key signature */ +extern struct rte_table_ops rte_table_hash_key8_lru_ops; + +/** LRU hash table operations for key signature computed on lookup ("do-sig") */ +extern struct rte_table_ops rte_table_hash_key8_lru_dosig_ops; + +/** Extendible bucket hash table parameters */ +struct rte_table_hash_key8_ext_params { + /** Maximum number of entries (and keys) in the table */ + uint32_t n_entries; + + /** Number of entries (and keys) for hash table bucket extensions. Each + bucket is extended in increments of 4 keys. */ + uint32_t n_entries_ext; + + /** Hash function */ + rte_table_hash_op_hash f_hash; + + /** Seed for the hash function */ + uint64_t seed; + + /** Byte offset within packet meta-data where the 4-byte key signature + is located. Valid for pre-computed key signature tables, ignored for + do-sig tables. */ + uint32_t signature_offset; + + /** Byte offset within packet meta-data where the key is located */ + uint32_t key_offset; + + /** Bit-mask to be AND-ed to the key on lookup */ + uint8_t *key_mask; +}; + +/** Extendible bucket hash table operations for pre-computed key signature */ +extern struct rte_table_ops rte_table_hash_key8_ext_ops; + +/** Extendible bucket hash table operations for key signature computed on + lookup ("do-sig") */ +extern struct rte_table_ops rte_table_hash_key8_ext_dosig_ops; + +/** + * 16-byte key hash tables + * + */ +/** LRU hash table parameters */ +struct rte_table_hash_key16_lru_params { + /** Maximum number of entries (and keys) in the table */ + uint32_t n_entries; + + /** Hash function */ + rte_table_hash_op_hash f_hash; + + /** Seed for the hash function */ + uint64_t seed; + + /** Byte offset within packet meta-data where the 4-byte key signature + is located. Valid for pre-computed key signature tables, ignored for + do-sig tables. */ + uint32_t signature_offset; + + /** Byte offset within packet meta-data where the key is located */ + uint32_t key_offset; + + /** Bit-mask to be AND-ed to the key on lookup */ + uint8_t *key_mask; +}; + +/** LRU hash table operations for pre-computed key signature */ +extern struct rte_table_ops rte_table_hash_key16_lru_ops; + +/** LRU hash table operations for key signature computed on lookup + ("do-sig") */ +extern struct rte_table_ops rte_table_hash_key16_lru_dosig_ops; + +/** Extendible bucket hash table parameters */ +struct rte_table_hash_key16_ext_params { + /** Maximum number of entries (and keys) in the table */ + uint32_t n_entries; + + /** Number of entries (and keys) for hash table bucket extensions. Each + bucket is extended in increments of 4 keys. */ + uint32_t n_entries_ext; + + /** Hash function */ + rte_table_hash_op_hash f_hash; + + /** Seed for the hash function */ + uint64_t seed; + + /** Byte offset within packet meta-data where the 4-byte key signature + is located. Valid for pre-computed key signature tables, ignored for + do-sig tables. */ + uint32_t signature_offset; + + /** Byte offset within packet meta-data where the key is located */ + uint32_t key_offset; + + /** Bit-mask to be AND-ed to the key on lookup */ + uint8_t *key_mask; +}; + +/** Extendible bucket operations for pre-computed key signature */ +extern struct rte_table_ops rte_table_hash_key16_ext_ops; + +/** Extendible bucket hash table operations for key signature computed on + lookup ("do-sig") */ +extern struct rte_table_ops rte_table_hash_key16_ext_dosig_ops; + +/** + * 32-byte key hash tables + * + */ +/** LRU hash table parameters */ +struct rte_table_hash_key32_lru_params { + /** Maximum number of entries (and keys) in the table */ + uint32_t n_entries; + + /** Hash function */ + rte_table_hash_op_hash f_hash; + + /** Seed for the hash function */ + uint64_t seed; + + /** Byte offset within packet meta-data where the 4-byte key signature + is located. Valid for pre-computed key signature tables, ignored for + do-sig tables. */ + uint32_t signature_offset; + + /** Byte offset within packet meta-data where the key is located */ + uint32_t key_offset; +}; + +/** LRU hash table operations for pre-computed key signature */ +extern struct rte_table_ops rte_table_hash_key32_lru_ops; + +/** Extendible bucket hash table parameters */ +struct rte_table_hash_key32_ext_params { + /** Maximum number of entries (and keys) in the table */ + uint32_t n_entries; + + /** Number of entries (and keys) for hash table bucket extensions. Each + bucket is extended in increments of 4 keys. */ + uint32_t n_entries_ext; + + /** Hash function */ + rte_table_hash_op_hash f_hash; + + /** Seed for the hash function */ + uint64_t seed; + + /** Byte offset within packet meta-data where the 4-byte key signature + is located. Valid for pre-computed key signature tables, ignored for + do-sig tables. */ + uint32_t signature_offset; + + /** Byte offset within packet meta-data where the key is located */ + uint32_t key_offset; +}; + +/** Extendible bucket hash table operations */ +extern struct rte_table_ops rte_table_hash_key32_ext_ops; + +#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 00000000..e283a3d1 --- /dev/null +++ b/lib/librte_table/rte_table_hash_ext.c @@ -0,0 +1,1159 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_log.h> + +#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) + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_HASH_EXT_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_HASH_EXT_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_HASH_EXT_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_HASH_EXT_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +struct grinder { + struct bucket *bkt; + uint64_t sig; + uint64_t match; + uint32_t key_index; +}; + +struct rte_table_hash { + struct rte_table_stats stats; + + /* 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; + } + + 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; + 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) % RTE_CACHE_LINE_SIZE) != 0) || + (sizeof(struct bucket) != (RTE_CACHE_LINE_SIZE / 2))) + return NULL; + + /* Memory allocation */ + table_meta_sz = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_table_hash)); + bucket_sz = RTE_CACHE_LINE_ROUNDUP(p->n_buckets * sizeof(struct bucket)); + bucket_ext_sz = + RTE_CACHE_LINE_ROUNDUP(p->n_buckets_ext * sizeof(struct bucket)); + key_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * p->key_size); + key_stack_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * sizeof(uint32_t)); + bkt_ext_stack_sz = + RTE_CACHE_LINE_ROUNDUP(p->n_buckets_ext * sizeof(uint32_t)); + data_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * entry_size); + total_size = table_meta_sz + bucket_sz + bucket_ext_sz + key_sz + + key_stack_sz + bkt_ext_stack_sz + data_sz; + + t = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + if (t == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + 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(entry_size); + + /* Tables */ + bucket_offset = 0; + 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)) { + /* Unchain bucket */ + BUCKET_NEXT_COPY(bkt_prev, bkt); + + /* Clear bucket */ + memset(bkt, 0, sizeof(struct bucket)); + + /* 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; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_EXT_STATS_PKTS_IN_ADD(t, n_pkts_in); + + 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; + RTE_TABLE_HASH_EXT_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(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; \ + uint32_t key_offset = t->key_offset; \ + \ + 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, key_offset));\ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));\ +} + +#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; \ + uint32_t key_offset = t->key_offset; \ + \ + 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, key_offset));\ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));\ +} + +#define lookup2_stage1(t, g, pkts, pkt10_index, pkt11_index) \ +{ \ + struct grinder *g10, *g11; \ + uint64_t sig10, sig11, bkt10_index, bkt11_index; \ + struct rte_mbuf *mbuf10, *mbuf11; \ + struct bucket *bkt10, *bkt11, *buckets = t->buckets; \ + uint64_t bucket_mask = t->bucket_mask; \ + uint32_t signature_offset = t->signature_offset; \ + \ + mbuf10 = pkts[pkt10_index]; \ + sig10 = (uint64_t) RTE_MBUF_METADATA_UINT32(mbuf10, signature_offset);\ + bkt10_index = sig10 & bucket_mask; \ + bkt10 = &buckets[bkt10_index]; \ + \ + mbuf11 = pkts[pkt11_index]; \ + sig11 = (uint64_t) RTE_MBUF_METADATA_UINT32(mbuf11, signature_offset);\ + bkt11_index = sig11 & bucket_mask; \ + bkt11 = &buckets[bkt11_index]; \ + \ + rte_prefetch0(bkt10); \ + rte_prefetch0(bkt11); \ + \ + g10 = &g[pkt10_index]; \ + g10->sig = sig10; \ + g10->bkt = bkt10; \ + \ + g11 = &g[pkt11_index]; \ + g11->sig = sig11; \ + g11->bkt = bkt11; \ +} + +#define lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index) \ +{ \ + 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; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_EXT_STATS_PKTS_IN_ADD(t, n_pkts_in); + + /* Cannot run the pipeline with less than 7 packets */ + if (__builtin_popcountll(pkts_mask) < 7) + 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; + RTE_TABLE_HASH_EXT_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return status; +} + +static int rte_table_hash_ext_lookup_dosig( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + struct grinder *g = t->grinders; + uint64_t pkt00_index, pkt01_index, pkt10_index, pkt11_index; + uint64_t pkt20_index, pkt21_index, pkt30_index, pkt31_index; + uint64_t pkts_mask_out = 0, pkts_mask_match_many = 0; + int status = 0; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_EXT_STATS_PKTS_IN_ADD(t, n_pkts_in); + + /* Cannot run the pipeline with less than 7 packets */ + if (__builtin_popcountll(pkts_mask) < 7) + 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; + RTE_TABLE_HASH_EXT_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return status; +} + +static int +rte_table_hash_ext_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + + if (stats != NULL) + memcpy(stats, &t->stats, sizeof(t->stats)); + + if (clear) + memset(&t->stats, 0, sizeof(t->stats)); + + return 0; +} + +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_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_ext_lookup, + .f_stats = rte_table_hash_ext_stats_read, +}; + +struct rte_table_ops rte_table_hash_ext_dosig_ops = { + .f_create = rte_table_hash_ext_create, + .f_free = rte_table_hash_ext_free, + .f_add = rte_table_hash_ext_entry_add, + .f_delete = rte_table_hash_ext_entry_delete, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_ext_lookup_dosig, + .f_stats = rte_table_hash_ext_stats_read, +}; diff --git a/lib/librte_table/rte_table_hash_key16.c b/lib/librte_table/rte_table_hash_key16.c new file mode 100644 index 00000000..b7e000fd --- /dev/null +++ b/lib/librte_table/rte_table_hash_key16.c @@ -0,0 +1,1515 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_log.h> + +#include "rte_table_hash.h" +#include "rte_lru.h" + +#define RTE_TABLE_HASH_KEY_SIZE 16 + +#define RTE_BUCKET_ENTRY_VALID 0x1LLU + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +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 { + struct rte_table_stats stats; + + /* Input parameters */ + uint32_t n_buckets; + uint32_t n_entries_per_bucket; + uint32_t key_size; + uint32_t entry_size; + uint32_t bucket_size; + uint32_t signature_offset; + uint32_t key_offset; + uint64_t key_mask[2]; + rte_table_hash_op_hash f_hash; + 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; + } + + /* 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) % RTE_CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_16) % RTE_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 + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + n_buckets * + bucket_size_cl * RTE_CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + 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 * RTE_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; + + if (p->key_mask != NULL) { + f->key_mask[0] = ((uint64_t *)p->key_mask)[0]; + f->key_mask[1] = ((uint64_t *)p->key_mask)[1]; + } else { + f->key_mask[0] = 0xFFFFFFFFFFFFFFFFLLU; + f->key_mask[1] = 0xFFFFFFFFFFFFFFFFLLU; + } + + 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; + } + + /* 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) % RTE_CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_16) % RTE_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 + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; + stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + RTE_CACHE_LINE_SIZE - 1) + / RTE_CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + + ((n_buckets + n_buckets_ext) * bucket_size_cl + stack_size_cl) * + RTE_CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + 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 * RTE_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; + + if (p->key_mask != NULL) { + f->key_mask[0] = (((uint64_t *)p->key_mask)[0]); + f->key_mask[1] = (((uint64_t *)p->key_mask)[1]); + } else { + f->key_mask[0] = 0xFFFFFFFFFFFFFFFFLLU; + f->key_mask[1] = 0xFFFFFFFFFFFFFFFFLLU; + } + + 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 = (((uint8_t *)bucket - + (uint8_t *)f->memory)/f->bucket_size) - 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, f) \ +{ \ + uint64_t pkt_mask; \ + uint32_t key_offset = f->key_offset;\ + \ + 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, key_offset));\ +} + +#define lookup1_stage1(mbuf1, bucket1, f) \ +{ \ + uint64_t signature; \ + uint32_t bucket_index; \ + \ + signature = RTE_MBUF_METADATA_UINT32(mbuf1, f->signature_offset);\ + bucket_index = signature & (f->n_buckets - 1); \ + bucket1 = (struct rte_bucket_4_16 *) \ + &f->memory[bucket_index * f->bucket_size]; \ + rte_prefetch0(bucket1); \ + rte_prefetch0((void *)(((uintptr_t) bucket1) + RTE_CACHE_LINE_SIZE));\ +} + +#define lookup1_stage1_dosig(mbuf1, bucket1, f) \ +{ \ + uint64_t *key; \ + uint64_t signature = 0; \ + uint32_t bucket_index; \ + uint64_t hash_key_buffer[2]; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf1, f->key_offset);\ + \ + hash_key_buffer[0] = key[0] & f->key_mask[0]; \ + hash_key_buffer[1] = key[1] & f->key_mask[1]; \ + signature = f->f_hash(hash_key_buffer, \ + RTE_TABLE_HASH_KEY_SIZE, f->seed); \ + \ + bucket_index = signature & (f->n_buckets - 1); \ + bucket1 = (struct rte_bucket_4_16 *) \ + &f->memory[bucket_index * f->bucket_size]; \ + rte_prefetch0(bucket1); \ + rte_prefetch0((void *)(((uintptr_t) bucket1) + RTE_CACHE_LINE_SIZE));\ +} + +#define lookup1_stage2_lru(pkt2_index, mbuf2, bucket2, \ + pkts_mask_out, entries, f) \ +{ \ + void *a; \ + uint64_t pkt_mask; \ + uint64_t *key; \ + uint64_t hash_key_buffer[2]; \ + uint32_t pos; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + hash_key_buffer[0] = key[0] & f->key_mask[0]; \ + hash_key_buffer[1] = key[1] & f->key_mask[1]; \ + \ + lookup_key16_cmp(hash_key_buffer, bucket2, pos); \ + \ + 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; \ + uint64_t hash_key_buffer[2]; \ + uint32_t pos; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + hash_key_buffer[0] = key[0] & f->key_mask[0]; \ + hash_key_buffer[1] = key[1] & f->key_mask[1]; \ + \ + lookup_key16_cmp(hash_key_buffer, bucket2, pos); \ + \ + 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; \ + uint64_t hash_key_buffer[2]; \ + uint32_t pos; \ + \ + bucket = buckets[pkt_index]; \ + key = keys[pkt_index]; \ + hash_key_buffer[0] = key[0] & f->key_mask[0]; \ + hash_key_buffer[1] = key[1] & f->key_mask[1]; \ + \ + lookup_key16_cmp(hash_key_buffer, bucket, pos); \ + \ + 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) + RTE_CACHE_LINE_SIZE));\ + buckets[pkt_index] = bucket_next; \ + keys[pkt_index] = key; \ +} + +#define lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01,\ + pkts, pkts_mask, f) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + uint32_t key_offset = f->key_offset; \ + \ + 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, key_offset));\ + \ + 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, key_offset));\ +} + +#define lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,\ + mbuf00, mbuf01, pkts, pkts_mask, f) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + uint32_t key_offset = f->key_offset; \ + \ + 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, key_offset)); \ + \ + 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, key_offset)); \ +} + +#define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f) \ +{ \ + uint64_t signature10, signature11; \ + uint32_t bucket10_index, bucket11_index; \ + \ + signature10 = RTE_MBUF_METADATA_UINT32(mbuf10, f->signature_offset);\ + bucket10_index = signature10 & (f->n_buckets - 1); \ + bucket10 = (struct rte_bucket_4_16 *) \ + &f->memory[bucket10_index * f->bucket_size]; \ + rte_prefetch0(bucket10); \ + rte_prefetch0((void *)(((uintptr_t) bucket10) + RTE_CACHE_LINE_SIZE));\ + \ + signature11 = RTE_MBUF_METADATA_UINT32(mbuf11, f->signature_offset);\ + bucket11_index = signature11 & (f->n_buckets - 1); \ + bucket11 = (struct rte_bucket_4_16 *) \ + &f->memory[bucket11_index * f->bucket_size]; \ + rte_prefetch0(bucket11); \ + rte_prefetch0((void *)(((uintptr_t) bucket11) + RTE_CACHE_LINE_SIZE));\ +} + +#define lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f) \ +{ \ + uint64_t *key10, *key11; \ + uint64_t hash_offset_buffer[2]; \ + uint64_t signature10, signature11; \ + uint32_t bucket10_index, bucket11_index; \ + \ + key10 = RTE_MBUF_METADATA_UINT64_PTR(mbuf10, f->key_offset);\ + hash_offset_buffer[0] = key10[0] & f->key_mask[0]; \ + hash_offset_buffer[1] = key10[1] & f->key_mask[1]; \ + signature10 = f->f_hash(hash_offset_buffer, \ + RTE_TABLE_HASH_KEY_SIZE, f->seed);\ + bucket10_index = signature10 & (f->n_buckets - 1); \ + bucket10 = (struct rte_bucket_4_16 *) \ + &f->memory[bucket10_index * f->bucket_size]; \ + rte_prefetch0(bucket10); \ + rte_prefetch0((void *)(((uintptr_t) bucket10) + RTE_CACHE_LINE_SIZE));\ + \ + key11 = RTE_MBUF_METADATA_UINT64_PTR(mbuf11, f->key_offset);\ + hash_offset_buffer[0] = key11[0] & f->key_mask[0]; \ + hash_offset_buffer[1] = key11[1] & f->key_mask[1]; \ + signature11 = f->f_hash(hash_offset_buffer, \ + RTE_TABLE_HASH_KEY_SIZE, f->seed);\ + bucket11_index = signature11 & (f->n_buckets - 1); \ + bucket11 = (struct rte_bucket_4_16 *) \ + &f->memory[bucket11_index * f->bucket_size]; \ + rte_prefetch0(bucket11); \ + rte_prefetch0((void *)(((uintptr_t) bucket11) + RTE_CACHE_LINE_SIZE));\ +} + +#define lookup2_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; \ + uint64_t hash_key_buffer20[2]; \ + uint64_t hash_key_buffer21[2]; \ + uint32_t pos20, pos21; \ + \ + key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ + key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ + hash_key_buffer20[0] = key20[0] & f->key_mask[0]; \ + hash_key_buffer20[1] = key20[1] & f->key_mask[1]; \ + hash_key_buffer21[0] = key21[0] & f->key_mask[0]; \ + hash_key_buffer21[1] = key21[1] & f->key_mask[1]; \ + \ + lookup_key16_cmp(hash_key_buffer20, bucket20, pos20); \ + lookup_key16_cmp(hash_key_buffer21, bucket21, pos21); \ + \ + 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; \ + uint64_t hash_key_buffer20[2]; \ + uint64_t hash_key_buffer21[2]; \ + uint32_t pos20, pos21; \ + \ + key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ + key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ + hash_key_buffer20[0] = key20[0] & f->key_mask[0]; \ + hash_key_buffer20[1] = key20[1] & f->key_mask[1]; \ + hash_key_buffer21[0] = key21[0] & f->key_mask[0]; \ + hash_key_buffer21[1] = key21[1] & f->key_mask[1]; \ + \ + lookup_key16_cmp(hash_key_buffer20, bucket20, pos20); \ + lookup_key16_cmp(hash_key_buffer21, bucket21, pos21); \ + \ + 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; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_16 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, + n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(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; + RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - + __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key16_lru() */ + +static int +rte_table_hash_lookup_key16_lru_dosig( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_16 *bucket10, *bucket11, *bucket20, *bucket21; + struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; + uint32_t pkt00_index, pkt01_index, pkt10_index; + uint32_t pkt11_index, pkt20_index, pkt21_index; + uint64_t pkts_mask_out = 0; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + + RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_16 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1_dosig(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - + __builtin_popcountll(pkts_mask_out)); + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* Pipeline stage 2 */ + lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, f); + } + + /* + * Pipeline flush + * + */ + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* Pipeline stage 2 */ + lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, f); + + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + + /* Pipeline stage 2 */ + lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, f); + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - + __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key16_lru_dosig() */ + +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]; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_16 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, + buckets, keys, f); + } + + goto grind_next_buckets; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(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: + /* Grind next buckets */ + for ( ; buckets_mask; ) { + uint64_t buckets_mask_next = 0; + + for ( ; buckets_mask; ) { + uint64_t pkt_mask; + uint32_t pkt_index; + + pkt_index = __builtin_ctzll(buckets_mask); + pkt_mask = 1LLU << pkt_index; + buckets_mask &= ~pkt_mask; + + lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, + entries, buckets_mask_next, f); + } + + buckets_mask = buckets_mask_next; + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - + __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key16_ext() */ + +static int +rte_table_hash_lookup_key16_ext_dosig( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_16 *bucket10, *bucket11, *bucket20, *bucket21; + struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; + uint32_t pkt00_index, pkt01_index, pkt10_index; + uint32_t pkt11_index, pkt20_index, pkt21_index; + uint64_t pkts_mask_out = 0, buckets_mask = 0; + struct rte_bucket_4_16 *buckets[RTE_PORT_IN_BURST_SIZE_MAX]; + uint64_t *keys[RTE_PORT_IN_BURST_SIZE_MAX]; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + + RTE_TABLE_HASH_KEY16_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_16 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1_dosig(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, + buckets, keys, f); + } + + goto grind_next_buckets; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* Pipeline stage 2 */ + lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, + buckets_mask, buckets, keys, f); + } + + /* + * Pipeline flush + * + */ + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* Pipeline stage 2 */ + lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, + buckets_mask, buckets, keys, f); + + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + + /* Pipeline stage 2 */ + lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, + buckets_mask, buckets, keys, f); + +grind_next_buckets: + /* Grind next buckets */ + for ( ; buckets_mask; ) { + uint64_t buckets_mask_next = 0; + + for ( ; buckets_mask; ) { + uint64_t pkt_mask; + uint32_t pkt_index; + + pkt_index = __builtin_ctzll(buckets_mask); + pkt_mask = 1LLU << pkt_index; + buckets_mask &= ~pkt_mask; + + lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, + entries, buckets_mask_next, f); + } + + buckets_mask = buckets_mask_next; + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY16_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - + __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key16_ext_dosig() */ + +static int +rte_table_hash_key16_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + + if (stats != NULL) + memcpy(stats, &t->stats, sizeof(t->stats)); + + if (clear) + memset(&t->stats, 0, sizeof(t->stats)); + + return 0; +} + +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_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lookup_key16_lru, + .f_stats = rte_table_hash_key16_stats_read, +}; + +struct rte_table_ops rte_table_hash_key16_lru_dosig_ops = { + .f_create = rte_table_hash_create_key16_lru, + .f_free = rte_table_hash_free_key16_lru, + .f_add = rte_table_hash_entry_add_key16_lru, + .f_delete = rte_table_hash_entry_delete_key16_lru, + .f_lookup = rte_table_hash_lookup_key16_lru_dosig, + .f_stats = rte_table_hash_key16_stats_read, +}; + +struct rte_table_ops rte_table_hash_key16_ext_ops = { + .f_create = rte_table_hash_create_key16_ext, + .f_free = rte_table_hash_free_key16_ext, + .f_add = rte_table_hash_entry_add_key16_ext, + .f_delete = rte_table_hash_entry_delete_key16_ext, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lookup_key16_ext, + .f_stats = rte_table_hash_key16_stats_read, +}; + +struct rte_table_ops rte_table_hash_key16_ext_dosig_ops = { + .f_create = rte_table_hash_create_key16_ext, + .f_free = rte_table_hash_free_key16_ext, + .f_add = rte_table_hash_entry_add_key16_ext, + .f_delete = rte_table_hash_entry_delete_key16_ext, + .f_lookup = rte_table_hash_lookup_key16_ext_dosig, + .f_stats = rte_table_hash_key16_stats_read, +}; diff --git a/lib/librte_table/rte_table_hash_key32.c b/lib/librte_table/rte_table_hash_key32.c new file mode 100644 index 00000000..a7aba492 --- /dev/null +++ b/lib/librte_table/rte_table_hash_key32.c @@ -0,0 +1,1144 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_log.h> + +#include "rte_table_hash.h" +#include "rte_lru.h" + +#define RTE_TABLE_HASH_KEY_SIZE 32 + +#define RTE_BUCKET_ENTRY_VALID 0x1LLU + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_HASH_KEY32_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_HASH_KEY32_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +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 { + struct rte_table_stats stats; + + /* 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; + } + + /* 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) % RTE_CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_32) % RTE_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 + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + n_buckets * + bucket_size_cl * RTE_CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + 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 * RTE_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; + } + + /* 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) % RTE_CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_32) % RTE_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 + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; + stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + RTE_CACHE_LINE_SIZE - 1) + / RTE_CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + + ((n_buckets + n_buckets_ext) * bucket_size_cl + stack_size_cl) * + RTE_CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + 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 * RTE_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 = (((uint8_t *)bucket - + (uint8_t *)f->memory)/f->bucket_size) - 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, f) \ +{ \ + uint64_t pkt_mask; \ + uint32_t key_offset = f->key_offset; \ + \ + 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, key_offset));\ +} + +#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) + RTE_CACHE_LINE_SIZE));\ + rte_prefetch0((void *)(((uintptr_t) bucket1) + 2 * RTE_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) + RTE_CACHE_LINE_SIZE));\ + rte_prefetch0((void *)(((uintptr_t) bucket_next) + \ + 2 * RTE_CACHE_LINE_SIZE)); \ + buckets[pkt_index] = bucket_next; \ + keys[pkt_index] = key; \ +} + +#define lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01,\ + pkts, pkts_mask, f) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + uint32_t key_offset = f->key_offset; \ + \ + 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, key_offset));\ + \ + 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, key_offset));\ +} + +#define lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,\ + mbuf00, mbuf01, pkts, pkts_mask, f) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + uint32_t key_offset = f->key_offset; \ + \ + 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, key_offset)); \ + \ + 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, key_offset)); \ +} + +#define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f) \ +{ \ + uint64_t signature10, signature11; \ + uint32_t bucket10_index, bucket11_index; \ + \ + signature10 = RTE_MBUF_METADATA_UINT32(mbuf10, f->signature_offset);\ + bucket10_index = signature10 & (f->n_buckets - 1); \ + bucket10 = (struct rte_bucket_4_32 *) \ + &f->memory[bucket10_index * f->bucket_size]; \ + rte_prefetch0(bucket10); \ + rte_prefetch0((void *)(((uintptr_t) bucket10) + RTE_CACHE_LINE_SIZE));\ + rte_prefetch0((void *)(((uintptr_t) bucket10) + 2 * RTE_CACHE_LINE_SIZE));\ + \ + signature11 = RTE_MBUF_METADATA_UINT32(mbuf11, f->signature_offset);\ + bucket11_index = signature11 & (f->n_buckets - 1); \ + bucket11 = (struct rte_bucket_4_32 *) \ + &f->memory[bucket11_index * f->bucket_size]; \ + rte_prefetch0(bucket11); \ + rte_prefetch0((void *)(((uintptr_t) bucket11) + RTE_CACHE_LINE_SIZE));\ + rte_prefetch0((void *)(((uintptr_t) bucket11) + 2 * RTE_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; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY32_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_32 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(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; + RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(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]; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY32_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_32 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, buckets, + keys, f); + } + + goto grind_next_buckets; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(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: + /* Grind next buckets */ + for ( ; buckets_mask; ) { + uint64_t buckets_mask_next = 0; + + for ( ; buckets_mask; ) { + uint64_t pkt_mask; + uint32_t pkt_index; + + pkt_index = __builtin_ctzll(buckets_mask); + pkt_mask = 1LLU << pkt_index; + buckets_mask &= ~pkt_mask; + + lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, + entries, buckets_mask_next, f); + } + + buckets_mask = buckets_mask_next; + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key32_ext() */ + +static int +rte_table_hash_key32_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + + if (stats != NULL) + memcpy(stats, &t->stats, sizeof(t->stats)); + + if (clear) + memset(&t->stats, 0, sizeof(t->stats)); + + return 0; +} + +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_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lookup_key32_lru, + .f_stats = rte_table_hash_key32_stats_read, +}; + +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_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lookup_key32_ext, + .f_stats = rte_table_hash_key32_stats_read, +}; 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 00000000..e2e2bdc4 --- /dev/null +++ b/lib/librte_table/rte_table_hash_key8.c @@ -0,0 +1,1471 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_log.h> + +#include "rte_table_hash.h" +#include "rte_lru.h" + +#define RTE_TABLE_HASH_KEY_SIZE 8 + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_HASH_KEY8_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_HASH_KEY8_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +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 { + struct rte_table_stats stats; + + /* Input parameters */ + uint32_t n_buckets; + uint32_t n_entries_per_bucket; + uint32_t key_size; + uint32_t entry_size; + uint32_t bucket_size; + uint32_t signature_offset; + uint32_t key_offset; + uint64_t key_mask; + rte_table_hash_op_hash f_hash; + 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; + } + + /* 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) % RTE_CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_8) % RTE_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 + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + n_buckets * + bucket_size_cl * RTE_CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + 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 * RTE_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; + + if (p->key_mask != NULL) + f->key_mask = ((uint64_t *)p->key_mask)[0]; + else + f->key_mask = 0xFFFFFFFFFFFFFFFFLLU; + + 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; + } + + /* 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) % RTE_CACHE_LINE_SIZE) != 0) || + ((sizeof(struct rte_bucket_4_8) % RTE_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 + RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE; + stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + RTE_CACHE_LINE_SIZE - 1) + / RTE_CACHE_LINE_SIZE; + total_size = sizeof(struct rte_table_hash) + ((n_buckets + + n_buckets_ext) * bucket_size_cl + stack_size_cl) * + RTE_CACHE_LINE_SIZE; + + f = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + 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 * RTE_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]; + + if (p->key_mask != NULL) + f->key_mask = ((uint64_t *)p->key_mask)[0]; + else + f->key_mask = 0xFFFFFFFFFFFFFFFFLLU; + + 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 = (((uint8_t *)bucket - + (uint8_t *)f->memory)/f->bucket_size) - 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, f) \ +{ \ + uint64_t pkt_mask; \ + uint32_t key_offset = f->key_offset;\ + \ + 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, key_offset)); \ +} + +#define lookup1_stage1(mbuf1, bucket1, f) \ +{ \ + uint64_t signature; \ + uint32_t bucket_index; \ + \ + signature = RTE_MBUF_METADATA_UINT32(mbuf1, f->signature_offset);\ + bucket_index = signature & (f->n_buckets - 1); \ + bucket1 = (struct rte_bucket_4_8 *) \ + &f->memory[bucket_index * f->bucket_size]; \ + rte_prefetch0(bucket1); \ +} + +#define lookup1_stage1_dosig(mbuf1, bucket1, f) \ +{ \ + uint64_t *key; \ + uint64_t signature; \ + uint32_t bucket_index; \ + uint64_t hash_key_buffer; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf1, f->key_offset);\ + hash_key_buffer = *key & f->key_mask; \ + signature = f->f_hash(&hash_key_buffer, \ + RTE_TABLE_HASH_KEY_SIZE, f->seed); \ + 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; \ + uint64_t hash_key_buffer; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + hash_key_buffer = key[0] & f->key_mask; \ + \ + lookup_key8_cmp((&hash_key_buffer), bucket2, pos); \ + \ + 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; \ + uint64_t hash_key_buffer; \ + \ + key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\ + hash_key_buffer = *key & f->key_mask; \ + \ + lookup_key8_cmp((&hash_key_buffer), bucket2, pos); \ + \ + 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; \ + uint64_t hash_key_buffer; \ + \ + bucket = buckets[pkt_index]; \ + key = keys[pkt_index]; \ + hash_key_buffer = (*key) & f->key_mask; \ + \ + lookup_key8_cmp((&hash_key_buffer), bucket, pos); \ + \ + 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, f) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + uint32_t key_offset = f->key_offset; \ + \ + 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, key_offset));\ + \ + 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, key_offset));\ +} + +#define lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,\ + mbuf00, mbuf01, pkts, pkts_mask, f) \ +{ \ + uint64_t pkt00_mask, pkt01_mask; \ + uint32_t key_offset = f->key_offset; \ + \ + 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, key_offset));\ + \ + 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, key_offset));\ +} + +#define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f) \ +{ \ + uint64_t signature10, signature11; \ + uint32_t bucket10_index, bucket11_index; \ + \ + signature10 = RTE_MBUF_METADATA_UINT32(mbuf10, f->signature_offset);\ + bucket10_index = signature10 & (f->n_buckets - 1); \ + bucket10 = (struct rte_bucket_4_8 *) \ + &f->memory[bucket10_index * f->bucket_size]; \ + rte_prefetch0(bucket10); \ + \ + signature11 = RTE_MBUF_METADATA_UINT32(mbuf11, f->signature_offset);\ + bucket11_index = signature11 & (f->n_buckets - 1); \ + bucket11 = (struct rte_bucket_4_8 *) \ + &f->memory[bucket11_index * f->bucket_size]; \ + rte_prefetch0(bucket11); \ +} + +#define lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f)\ +{ \ + uint64_t *key10, *key11; \ + uint64_t hash_offset_buffer10; \ + uint64_t hash_offset_buffer11; \ + uint64_t signature10, signature11; \ + uint32_t bucket10_index, bucket11_index; \ + rte_table_hash_op_hash f_hash = f->f_hash; \ + 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);\ + hash_offset_buffer10 = *key10 & f->key_mask; \ + hash_offset_buffer11 = *key11 & f->key_mask; \ + \ + signature10 = f_hash(&hash_offset_buffer10, \ + RTE_TABLE_HASH_KEY_SIZE, seed); \ + bucket10_index = signature10 & (f->n_buckets - 1); \ + bucket10 = (struct rte_bucket_4_8 *) \ + &f->memory[bucket10_index * f->bucket_size]; \ + rte_prefetch0(bucket10); \ + \ + signature11 = f_hash(&hash_offset_buffer11, \ + RTE_TABLE_HASH_KEY_SIZE, seed); \ + 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; \ + uint64_t hash_offset_buffer20; \ + uint64_t hash_offset_buffer21; \ + uint32_t pos20, pos21; \ + \ + key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ + key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ + hash_offset_buffer20 = *key20 & f->key_mask; \ + hash_offset_buffer21 = *key21 & f->key_mask; \ + \ + lookup_key8_cmp((&hash_offset_buffer20), bucket20, pos20);\ + lookup_key8_cmp((&hash_offset_buffer21), bucket21, pos21);\ + \ + 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; \ + uint64_t hash_offset_buffer20; \ + uint64_t hash_offset_buffer21; \ + uint32_t pos20, pos21; \ + \ + key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\ + key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\ + hash_offset_buffer20 = *key20 & f->key_mask; \ + hash_offset_buffer21 = *key21 & f->key_mask; \ + \ + lookup_key8_cmp((&hash_offset_buffer20), bucket20, pos20);\ + lookup_key8_cmp((&hash_offset_buffer21), bucket21, pos21);\ + \ + 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; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY8_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_8 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(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; + RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key8_lru() */ + +static int +rte_table_hash_lookup_key8_lru_dosig( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_8 *bucket10, *bucket11, *bucket20, *bucket21; + struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; + uint32_t pkt00_index, pkt01_index, pkt10_index; + uint32_t pkt11_index, pkt20_index, pkt21_index; + uint64_t pkts_mask_out = 0; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY8_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_8 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1_dosig(mbuf, bucket, f); + lookup1_stage2_lru(pkt_index, mbuf, bucket, + pkts_mask_out, entries, f); + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* Pipeline stage 2 */ + lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, f); + } + + /* + * Pipeline flush + * + */ + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* Pipeline stage 2 */ + lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, f); + + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + + /* Pipeline stage 2 */ + lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, f); + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key8_lru_dosig() */ + +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]; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY8_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_8 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, buckets, + keys, f); + } + + goto grind_next_buckets; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1(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: + /* Grind next buckets */ + for ( ; buckets_mask; ) { + uint64_t buckets_mask_next = 0; + + for ( ; buckets_mask; ) { + uint64_t pkt_mask; + uint32_t pkt_index; + + pkt_index = __builtin_ctzll(buckets_mask); + pkt_mask = 1LLU << pkt_index; + buckets_mask &= ~pkt_mask; + + lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, + entries, buckets_mask_next, f); + } + + buckets_mask = buckets_mask_next; + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key8_ext() */ + +static int +rte_table_hash_lookup_key8_ext_dosig( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_hash *f = (struct rte_table_hash *) table; + struct rte_bucket_4_8 *bucket10, *bucket11, *bucket20, *bucket21; + struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21; + uint32_t pkt00_index, pkt01_index, pkt10_index; + uint32_t pkt11_index, pkt20_index, pkt21_index; + uint64_t pkts_mask_out = 0, buckets_mask = 0; + struct rte_bucket_4_8 *buckets[RTE_PORT_IN_BURST_SIZE_MAX]; + uint64_t *keys[RTE_PORT_IN_BURST_SIZE_MAX]; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_KEY8_STATS_PKTS_IN_ADD(f, n_pkts_in); + + /* Cannot run the pipeline with less than 5 packets */ + if (__builtin_popcountll(pkts_mask) < 5) { + for ( ; pkts_mask; ) { + struct rte_bucket_4_8 *bucket; + struct rte_mbuf *mbuf; + uint32_t pkt_index; + + lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f); + lookup1_stage1_dosig(mbuf, bucket, f); + lookup1_stage2_ext(pkt_index, mbuf, bucket, + pkts_mask_out, entries, buckets_mask, + buckets, keys, f); + } + + goto grind_next_buckets; + } + + /* + * Pipeline fill + * + */ + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline feed */ + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts, + pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(pkt00_index, pkt01_index, + mbuf00, mbuf01, pkts, pkts_mask, f); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* Pipeline stage 2 */ + lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, + buckets_mask, buckets, keys, f); + } + + /* + * Pipeline flush + * + */ + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + mbuf10 = mbuf00; + mbuf11 = mbuf01; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(mbuf10, mbuf11, bucket10, bucket11, f); + + /* Pipeline stage 2 */ + lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, + buckets_mask, buckets, keys, f); + + /* Pipeline feed */ + bucket20 = bucket10; + bucket21 = bucket11; + mbuf20 = mbuf10; + mbuf21 = mbuf11; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + + /* Pipeline stage 2 */ + lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, + bucket20, bucket21, pkts_mask_out, entries, + buckets_mask, buckets, keys, f); + +grind_next_buckets: + /* Grind next buckets */ + for ( ; buckets_mask; ) { + uint64_t buckets_mask_next = 0; + + for ( ; buckets_mask; ) { + uint64_t pkt_mask; + uint32_t pkt_index; + + pkt_index = __builtin_ctzll(buckets_mask); + pkt_mask = 1LLU << pkt_index; + buckets_mask &= ~pkt_mask; + + lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, + entries, buckets_mask_next, f); + } + + buckets_mask = buckets_mask_next; + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_KEY8_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return 0; +} /* rte_table_hash_lookup_key8_dosig_ext() */ + +static int +rte_table_hash_key8_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + + if (stats != NULL) + memcpy(stats, &t->stats, sizeof(t->stats)); + + if (clear) + memset(&t->stats, 0, sizeof(t->stats)); + + return 0; +} + +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_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lookup_key8_lru, + .f_stats = rte_table_hash_key8_stats_read, +}; + +struct rte_table_ops rte_table_hash_key8_lru_dosig_ops = { + .f_create = rte_table_hash_create_key8_lru, + .f_free = rte_table_hash_free_key8_lru, + .f_add = rte_table_hash_entry_add_key8_lru, + .f_delete = rte_table_hash_entry_delete_key8_lru, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lookup_key8_lru_dosig, + .f_stats = rte_table_hash_key8_stats_read, +}; + +struct rte_table_ops rte_table_hash_key8_ext_ops = { + .f_create = rte_table_hash_create_key8_ext, + .f_free = rte_table_hash_free_key8_ext, + .f_add = rte_table_hash_entry_add_key8_ext, + .f_delete = rte_table_hash_entry_delete_key8_ext, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lookup_key8_ext, + .f_stats = rte_table_hash_key8_stats_read, +}; + +struct rte_table_ops rte_table_hash_key8_ext_dosig_ops = { + .f_create = rte_table_hash_create_key8_ext, + .f_free = rte_table_hash_free_key8_ext, + .f_add = rte_table_hash_entry_add_key8_ext, + .f_delete = rte_table_hash_entry_delete_key8_ext, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lookup_key8_ext_dosig, + .f_stats = rte_table_hash_key8_stats_read, +}; diff --git a/lib/librte_table/rte_table_hash_lru.c b/lib/librte_table/rte_table_hash_lru.c new file mode 100644 index 00000000..407c62ab --- /dev/null +++ b/lib/librte_table/rte_table_hash_lru.c @@ -0,0 +1,1102 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_log.h> + +#include "rte_table_hash.h" +#include "rte_lru.h" + +#define KEYS_PER_BUCKET 4 + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_HASH_LRU_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_HASH_LRU_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_HASH_LRU_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_HASH_LRU_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +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 { + struct rte_table_stats stats; + + /* 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; + } + + 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; + 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) % RTE_CACHE_LINE_SIZE) != 0) || + (sizeof(struct bucket) != (RTE_CACHE_LINE_SIZE / 2))) { + return NULL; + } + + /* Memory allocation */ + table_meta_sz = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_table_hash)); + bucket_sz = RTE_CACHE_LINE_ROUNDUP(p->n_buckets * sizeof(struct bucket)); + key_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * p->key_size); + key_stack_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * sizeof(uint32_t)); + data_sz = RTE_CACHE_LINE_ROUNDUP(p->n_keys * entry_size); + total_size = table_meta_sz + bucket_sz + key_sz + key_stack_sz + + data_sz; + + t = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, socket_id); + if (t == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for hash table\n", + __func__, total_size); + 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(entry_size); + + /* Tables */ + bucket_offset = 0; + 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; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_LRU_STATS_PKTS_IN_ADD(t, n_pkts_in); + + 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; + RTE_TABLE_HASH_LRU_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(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; \ + uint32_t key_offset = t->key_offset; \ + \ + 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, key_offset));\ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));\ +} + +#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; \ + uint32_t key_offset = t->key_offset; \ + \ + 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, key_offset));\ + rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));\ +} + +#define lookup2_stage1(t, g, pkts, pkt10_index, pkt11_index) \ +{ \ + struct grinder *g10, *g11; \ + uint64_t sig10, sig11, bkt10_index, bkt11_index; \ + struct rte_mbuf *mbuf10, *mbuf11; \ + struct bucket *bkt10, *bkt11, *buckets = t->buckets; \ + uint64_t bucket_mask = t->bucket_mask; \ + uint32_t signature_offset = t->signature_offset; \ + \ + mbuf10 = pkts[pkt10_index]; \ + sig10 = (uint64_t) RTE_MBUF_METADATA_UINT32(mbuf10, signature_offset);\ + bkt10_index = sig10 & bucket_mask; \ + bkt10 = &buckets[bkt10_index]; \ + \ + mbuf11 = pkts[pkt11_index]; \ + sig11 = (uint64_t) RTE_MBUF_METADATA_UINT32(mbuf11, signature_offset);\ + bkt11_index = sig11 & bucket_mask; \ + bkt11 = &buckets[bkt11_index]; \ + \ + rte_prefetch0(bkt10); \ + rte_prefetch0(bkt11); \ + \ + g10 = &g[pkt10_index]; \ + g10->sig = sig10; \ + g10->bkt = bkt10; \ + \ + g11 = &g[pkt11_index]; \ + g11->sig = sig11; \ + g11->bkt = bkt11; \ +} + +#define lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index)\ +{ \ + 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; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_LRU_STATS_PKTS_IN_ADD(t, n_pkts_in); + + /* Cannot run the pipeline with less than 7 packets */ + if (__builtin_popcountll(pkts_mask) < 7) + return rte_table_hash_lru_lookup_unoptimized(table, pkts, + pkts_mask, lookup_hit_mask, entries, 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; + RTE_TABLE_HASH_LRU_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return status; +} + +static int rte_table_hash_lru_lookup_dosig( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + struct grinder *g = t->grinders; + uint64_t pkt00_index, pkt01_index, pkt10_index, pkt11_index; + uint64_t pkt20_index, pkt21_index, pkt30_index, pkt31_index; + uint64_t pkts_mask_out = 0, pkts_mask_match_many = 0; + int status = 0; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_HASH_LRU_STATS_PKTS_IN_ADD(t, n_pkts_in); + + /* Cannot run the pipeline with less than 7 packets */ + if (__builtin_popcountll(pkts_mask) < 7) + return rte_table_hash_lru_lookup_unoptimized(table, pkts, + pkts_mask, lookup_hit_mask, entries, 1); + + /* Pipeline stage 0 */ + lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); + + /* Pipeline feed */ + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); + + /* Pipeline feed */ + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0(t, g, pkts, pkts_mask, pkt00_index, pkt01_index); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); + + /* Pipeline stage 2 */ + lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); + + /* + * Pipeline run + * + */ + for ( ; pkts_mask; ) { + /* Pipeline feed */ + pkt30_index = pkt20_index; + pkt31_index = pkt21_index; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 0 */ + lookup2_stage0_with_odd_support(t, g, pkts, pkts_mask, + pkt00_index, pkt01_index); + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); + + /* Pipeline stage 2 */ + lookup2_stage2(t, g, pkt20_index, pkt21_index, + pkts_mask_match_many); + + /* Pipeline stage 3 */ + lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, + pkts_mask_out, entries); + } + + /* Pipeline feed */ + pkt30_index = pkt20_index; + pkt31_index = pkt21_index; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + pkt10_index = pkt00_index; + pkt11_index = pkt01_index; + + /* Pipeline stage 1 */ + lookup2_stage1_dosig(t, g, pkts, pkt10_index, pkt11_index); + + /* Pipeline stage 2 */ + lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); + + /* Pipeline stage 3 */ + lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, + entries); + + /* Pipeline feed */ + pkt30_index = pkt20_index; + pkt31_index = pkt21_index; + pkt20_index = pkt10_index; + pkt21_index = pkt11_index; + + /* Pipeline stage 2 */ + lookup2_stage2(t, g, pkt20_index, pkt21_index, pkts_mask_match_many); + + /* Pipeline stage 3 */ + lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, + entries); + + /* Pipeline feed */ + pkt30_index = pkt20_index; + pkt31_index = pkt21_index; + + /* Pipeline stage 3 */ + lookup2_stage3(t, g, pkts, pkt30_index, pkt31_index, pkts_mask_out, + entries); + + /* Slow path */ + pkts_mask_match_many &= ~pkts_mask_out; + if (pkts_mask_match_many) { + uint64_t pkts_mask_out_slow = 0; + + status = rte_table_hash_lru_lookup_unoptimized(table, pkts, + pkts_mask_match_many, &pkts_mask_out_slow, entries, 1); + pkts_mask_out |= pkts_mask_out_slow; + } + + *lookup_hit_mask = pkts_mask_out; + RTE_TABLE_HASH_LRU_STATS_PKTS_LOOKUP_MISS(t, n_pkts_in - __builtin_popcountll(pkts_mask_out)); + return status; +} + +static int +rte_table_hash_lru_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_hash *t = (struct rte_table_hash *) table; + + if (stats != NULL) + memcpy(stats, &t->stats, sizeof(t->stats)); + + if (clear) + memset(&t->stats, 0, sizeof(t->stats)); + + return 0; +} + +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_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lru_lookup, + .f_stats = rte_table_hash_lru_stats_read, +}; + +struct rte_table_ops rte_table_hash_lru_dosig_ops = { + .f_create = rte_table_hash_lru_create, + .f_free = rte_table_hash_lru_free, + .f_add = rte_table_hash_lru_entry_add, + .f_delete = rte_table_hash_lru_entry_delete, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_hash_lru_lookup_dosig, + .f_stats = rte_table_hash_lru_stats_read, +}; diff --git a/lib/librte_table/rte_table_lpm.c b/lib/librte_table/rte_table_lpm.c new file mode 100644 index 00000000..cdeb0f5a --- /dev/null +++ b/lib/librte_table/rte_table_lpm.c @@ -0,0 +1,393 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_byteorder.h> +#include <rte_log.h> +#include <rte_lpm.h> + +#include "rte_table_lpm.h" + +#define RTE_TABLE_LPM_MAX_NEXT_HOPS 256 + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_LPM_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_LPM_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_LPM_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_LPM_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +struct rte_table_lpm { + struct rte_table_stats stats; + + /* Input parameters */ + uint32_t entry_size; + uint32_t entry_unique_size; + uint32_t n_rules; + uint32_t offset; + + /* Handle to low-level LPM table */ + struct rte_lpm *lpm; + + /* Next Hop Table (NHT) */ + uint32_t nht_users[RTE_TABLE_LPM_MAX_NEXT_HOPS]; + uint32_t nht[0] __rte_cache_aligned; +}; + +static void * +rte_table_lpm_create(void *params, int socket_id, uint32_t entry_size) +{ + struct rte_table_lpm_params *p = (struct rte_table_lpm_params *) params; + struct rte_table_lpm *lpm; + struct rte_lpm_config lpm_config; + + uint32_t total_size, nht_size; + + /* Check input parameters */ + if (p == NULL) { + RTE_LOG(ERR, TABLE, "%s: NULL input parameters\n", __func__); + return NULL; + } + if (p->n_rules == 0) { + RTE_LOG(ERR, TABLE, "%s: Invalid n_rules\n", __func__); + return NULL; + } + if (p->number_tbl8s == 0) { + RTE_LOG(ERR, TABLE, "%s: Invalid number_tbl8s\n", __func__); + return NULL; + } + if (p->entry_unique_size == 0) { + RTE_LOG(ERR, TABLE, "%s: Invalid entry_unique_size\n", + __func__); + return NULL; + } + if (p->entry_unique_size > entry_size) { + RTE_LOG(ERR, TABLE, "%s: Invalid entry_unique_size\n", + __func__); + return NULL; + } + if (p->name == NULL) { + RTE_LOG(ERR, TABLE, "%s: Table name is NULL\n", + __func__); + return NULL; + } + entry_size = RTE_ALIGN(entry_size, sizeof(uint64_t)); + + /* Memory allocation */ + nht_size = RTE_TABLE_LPM_MAX_NEXT_HOPS * entry_size; + total_size = sizeof(struct rte_table_lpm) + nht_size; + lpm = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, + socket_id); + if (lpm == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for LPM table\n", + __func__, total_size); + return NULL; + } + + /* LPM low-level table creation */ + lpm_config.max_rules = p->n_rules; + lpm_config.number_tbl8s = p->number_tbl8s; + lpm_config.flags = p->flags; + lpm->lpm = rte_lpm_create(p->name, socket_id, &lpm_config); + + if (lpm->lpm == NULL) { + rte_free(lpm); + RTE_LOG(ERR, TABLE, "Unable to create low-level LPM table\n"); + return NULL; + } + + /* Memory initialization */ + lpm->entry_size = entry_size; + lpm->entry_unique_size = p->entry_unique_size; + lpm->n_rules = p->n_rules; + lpm->offset = p->offset; + + return lpm; +} + +static int +rte_table_lpm_free(void *table) +{ + struct rte_table_lpm *lpm = (struct rte_table_lpm *) table; + + /* Check input parameters */ + if (lpm == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + /* Free previously allocated resources */ + rte_lpm_free(lpm->lpm); + rte_free(lpm); + + return 0; +} + +static int +nht_find_free(struct rte_table_lpm *lpm, uint32_t *pos) +{ + uint32_t i; + + for (i = 0; i < RTE_TABLE_LPM_MAX_NEXT_HOPS; i++) { + if (lpm->nht_users[i] == 0) { + *pos = i; + return 1; + } + } + + return 0; +} + +static int +nht_find_existing(struct rte_table_lpm *lpm, void *entry, uint32_t *pos) +{ + uint32_t i; + + for (i = 0; i < RTE_TABLE_LPM_MAX_NEXT_HOPS; i++) { + uint32_t *nht_entry = &lpm->nht[i * lpm->entry_size]; + + if ((lpm->nht_users[i] > 0) && (memcmp(nht_entry, entry, + lpm->entry_unique_size) == 0)) { + *pos = i; + return 1; + } + } + + return 0; +} + +static int +rte_table_lpm_entry_add( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_lpm *lpm = (struct rte_table_lpm *) table; + struct rte_table_lpm_key *ip_prefix = (struct rte_table_lpm_key *) key; + uint32_t nht_pos, nht_pos0_valid; + int status; + uint32_t nht_pos0 = 0; + + /* Check input parameters */ + if (lpm == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (ip_prefix == NULL) { + RTE_LOG(ERR, TABLE, "%s: ip_prefix parameter is NULL\n", + __func__); + return -EINVAL; + } + if (entry == NULL) { + RTE_LOG(ERR, TABLE, "%s: entry parameter is NULL\n", __func__); + return -EINVAL; + } + + if ((ip_prefix->depth == 0) || (ip_prefix->depth > 32)) { + RTE_LOG(ERR, TABLE, "%s: invalid depth (%d)\n", + __func__, ip_prefix->depth); + return -EINVAL; + } + + /* Check if rule is already present in the table */ + status = rte_lpm_is_rule_present(lpm->lpm, ip_prefix->ip, + ip_prefix->depth, &nht_pos0); + nht_pos0_valid = status > 0; + + /* Find existing or free NHT entry */ + if (nht_find_existing(lpm, entry, &nht_pos) == 0) { + uint32_t *nht_entry; + + if (nht_find_free(lpm, &nht_pos) == 0) { + RTE_LOG(ERR, TABLE, "%s: NHT full\n", __func__); + return -1; + } + + nht_entry = &lpm->nht[nht_pos * lpm->entry_size]; + memcpy(nht_entry, entry, lpm->entry_size); + } + + /* Add rule to low level LPM table */ + if (rte_lpm_add(lpm->lpm, ip_prefix->ip, ip_prefix->depth, nht_pos) < 0) { + RTE_LOG(ERR, TABLE, "%s: LPM rule add failed\n", __func__); + return -1; + } + + /* Commit NHT changes */ + lpm->nht_users[nht_pos]++; + lpm->nht_users[nht_pos0] -= nht_pos0_valid; + + *key_found = nht_pos0_valid; + *entry_ptr = (void *) &lpm->nht[nht_pos * lpm->entry_size]; + return 0; +} + +static int +rte_table_lpm_entry_delete( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_lpm *lpm = (struct rte_table_lpm *) table; + struct rte_table_lpm_key *ip_prefix = (struct rte_table_lpm_key *) key; + uint32_t nht_pos; + int status; + + /* Check input parameters */ + if (lpm == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (ip_prefix == NULL) { + RTE_LOG(ERR, TABLE, "%s: ip_prefix parameter is NULL\n", + __func__); + return -EINVAL; + } + if ((ip_prefix->depth == 0) || (ip_prefix->depth > 32)) { + RTE_LOG(ERR, TABLE, "%s: invalid depth (%d)\n", __func__, + ip_prefix->depth); + return -EINVAL; + } + + /* Return if rule is not present in the table */ + status = rte_lpm_is_rule_present(lpm->lpm, ip_prefix->ip, + ip_prefix->depth, &nht_pos); + if (status < 0) { + RTE_LOG(ERR, TABLE, "%s: LPM algorithmic error\n", __func__); + return -1; + } + if (status == 0) { + *key_found = 0; + return 0; + } + + /* Delete rule from the low-level LPM table */ + status = rte_lpm_delete(lpm->lpm, ip_prefix->ip, ip_prefix->depth); + if (status) { + RTE_LOG(ERR, TABLE, "%s: LPM rule delete failed\n", __func__); + return -1; + } + + /* Commit NHT changes */ + lpm->nht_users[nht_pos]--; + + *key_found = 1; + if (entry) + memcpy(entry, &lpm->nht[nht_pos * lpm->entry_size], + lpm->entry_size); + + return 0; +} + +static int +rte_table_lpm_lookup( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_lpm *lpm = (struct rte_table_lpm *) table; + uint64_t pkts_out_mask = 0; + uint32_t i; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_LPM_STATS_PKTS_IN_ADD(lpm, n_pkts_in); + + pkts_out_mask = 0; + for (i = 0; i < (uint32_t)(RTE_PORT_IN_BURST_SIZE_MAX - + __builtin_clzll(pkts_mask)); i++) { + uint64_t pkt_mask = 1LLU << i; + + if (pkt_mask & pkts_mask) { + struct rte_mbuf *pkt = pkts[i]; + uint32_t ip = rte_bswap32( + RTE_MBUF_METADATA_UINT32(pkt, lpm->offset)); + int status; + uint32_t nht_pos; + + status = rte_lpm_lookup(lpm->lpm, ip, &nht_pos); + if (status == 0) { + pkts_out_mask |= pkt_mask; + entries[i] = (void *) &lpm->nht[nht_pos * + lpm->entry_size]; + } + } + } + + *lookup_hit_mask = pkts_out_mask; + RTE_TABLE_LPM_STATS_PKTS_LOOKUP_MISS(lpm, n_pkts_in - __builtin_popcountll(pkts_out_mask)); + return 0; +} + +static int +rte_table_lpm_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_lpm *t = (struct rte_table_lpm *) table; + + if (stats != NULL) + memcpy(stats, &t->stats, sizeof(t->stats)); + + if (clear) + memset(&t->stats, 0, sizeof(t->stats)); + + return 0; +} + +struct rte_table_ops rte_table_lpm_ops = { + .f_create = rte_table_lpm_create, + .f_free = rte_table_lpm_free, + .f_add = rte_table_lpm_entry_add, + .f_delete = rte_table_lpm_entry_delete, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_lpm_lookup, + .f_stats = rte_table_lpm_stats_read, +}; diff --git a/lib/librte_table/rte_table_lpm.h b/lib/librte_table/rte_table_lpm.h new file mode 100644 index 00000000..f3033234 --- /dev/null +++ b/lib/librte_table/rte_table_lpm.h @@ -0,0 +1,124 @@ +/*- + * 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_LPM_H__ +#define __INCLUDE_RTE_TABLE_LPM_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Table LPM for IPv4 + * + * This table uses the Longest Prefix Match (LPM) algorithm to uniquely + * associate data to lookup keys. + * + * Use-case: IP routing table. Routes that are added to the table associate a + * next hop to an IP prefix. The IP prefix is specified as IP address and depth + * and cover for a multitude of lookup keys (i.e. destination IP addresses) + * that all share the same data (i.e. next hop). The next hop information + * typically contains the output interface ID, the IP address of the next hop + * station (which is part of the same IP network the output interface is + * connected to) and other flags and counters. + * + * The LPM primitive only allows associating an 8-bit number (next hop ID) to + * an IP prefix, while a routing table can potentially contain thousands of + * routes or even more. This means that the same next hop ID (and next hop + * information) has to be shared by multiple routes, which makes sense, as + * multiple remote networks could be reached through the same next hop. + * Therefore, when a route is added or updated, the LPM table has to check + * whether the same next hop is already in use before using a new next hop ID + * for this route. + * + * The comparison between different next hops is done for the first + * “entry_unique_size” bytes of the next hop information (configurable + * parameter), which have to uniquely identify the next hop, therefore the user + * has to carefully manage the format of the LPM table entry (i.e. the next + * hop information) so that any next hop data that changes value during + * run-time (e.g. counters) is placed outside of this area. + * + ***/ + +#include <stdint.h> + +#include "rte_table.h" + +/** LPM table parameters */ +struct rte_table_lpm_params { + /** Table name */ + const char *name; + + /** Maximum number of LPM rules (i.e. IP routes) */ + uint32_t n_rules; + + /**< Number of tbl8s to allocate. */ + uint32_t number_tbl8s; + + /**< This field is currently unused. */ + int flags; + + /** Number of bytes at the start of the table entry that uniquely + identify the entry. Cannot be bigger than table entry size. */ + uint32_t entry_unique_size; + + /** Byte offset within input packet meta-data where lookup key (i.e. + the destination IP address) is located. */ + uint32_t offset; +}; + +/** LPM table rule (i.e. route), specified as IP prefix. While the key used by +the lookup operation is the destination IP address (read from the input packet +meta-data), the entry add and entry delete operations work with LPM rules, with +each rule covering for a multitude of lookup keys (destination IP addresses) +that share the same data (next hop). */ +struct rte_table_lpm_key { + /** IP address */ + uint32_t ip; + + /** IP address depth. The most significant "depth" bits of the IP + address specify the network part of the IP address, while the rest of + the bits specify the host part of the address and are ignored for the + purpose of route specification. */ + uint8_t depth; +}; + +/** LPM table operations */ +extern struct rte_table_ops rte_table_lpm_ops; + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/lib/librte_table/rte_table_lpm_ipv6.c b/lib/librte_table/rte_table_lpm_ipv6.c new file mode 100644 index 00000000..836f4cf6 --- /dev/null +++ b/lib/librte_table/rte_table_lpm_ipv6.c @@ -0,0 +1,398 @@ +/*- + * 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 <string.h> +#include <stdio.h> + +#include <rte_common.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_byteorder.h> +#include <rte_log.h> +#include <rte_lpm6.h> + +#include "rte_table_lpm_ipv6.h" + +#define RTE_TABLE_LPM_MAX_NEXT_HOPS 256 + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_LPM_IPV6_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_LPM_IPV6_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_LPM_IPV6_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_LPM_IPV6_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +struct rte_table_lpm_ipv6 { + struct rte_table_stats stats; + + /* Input parameters */ + uint32_t entry_size; + uint32_t entry_unique_size; + uint32_t n_rules; + uint32_t offset; + + /* Handle to low-level LPM table */ + struct rte_lpm6 *lpm; + + /* Next Hop Table (NHT) */ + uint32_t nht_users[RTE_TABLE_LPM_MAX_NEXT_HOPS]; + uint8_t nht[0] __rte_cache_aligned; +}; + +static void * +rte_table_lpm_ipv6_create(void *params, int socket_id, uint32_t entry_size) +{ + struct rte_table_lpm_ipv6_params *p = + (struct rte_table_lpm_ipv6_params *) params; + struct rte_table_lpm_ipv6 *lpm; + struct rte_lpm6_config lpm6_config; + uint32_t total_size, nht_size; + + /* Check input parameters */ + if (p == NULL) { + RTE_LOG(ERR, TABLE, "%s: NULL input parameters\n", __func__); + return NULL; + } + if (p->n_rules == 0) { + RTE_LOG(ERR, TABLE, "%s: Invalid n_rules\n", __func__); + return NULL; + } + if (p->number_tbl8s == 0) { + RTE_LOG(ERR, TABLE, "%s: Invalid n_rules\n", __func__); + return NULL; + } + if (p->entry_unique_size == 0) { + RTE_LOG(ERR, TABLE, "%s: Invalid entry_unique_size\n", + __func__); + return NULL; + } + if (p->entry_unique_size > entry_size) { + RTE_LOG(ERR, TABLE, "%s: Invalid entry_unique_size\n", + __func__); + return NULL; + } + if (p->name == NULL) { + RTE_LOG(ERR, TABLE, "%s: Table name is NULL\n", + __func__); + return NULL; + } + entry_size = RTE_ALIGN(entry_size, sizeof(uint64_t)); + + /* Memory allocation */ + nht_size = RTE_TABLE_LPM_MAX_NEXT_HOPS * entry_size; + total_size = sizeof(struct rte_table_lpm_ipv6) + nht_size; + lpm = rte_zmalloc_socket("TABLE", total_size, RTE_CACHE_LINE_SIZE, + socket_id); + if (lpm == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for LPM IPv6 table\n", + __func__, total_size); + return NULL; + } + + /* LPM low-level table creation */ + lpm6_config.max_rules = p->n_rules; + lpm6_config.number_tbl8s = p->number_tbl8s; + lpm6_config.flags = 0; + lpm->lpm = rte_lpm6_create(p->name, socket_id, &lpm6_config); + if (lpm->lpm == NULL) { + rte_free(lpm); + RTE_LOG(ERR, TABLE, + "Unable to create low-level LPM IPv6 table\n"); + return NULL; + } + + /* Memory initialization */ + lpm->entry_size = entry_size; + lpm->entry_unique_size = p->entry_unique_size; + lpm->n_rules = p->n_rules; + lpm->offset = p->offset; + + return lpm; +} + +static int +rte_table_lpm_ipv6_free(void *table) +{ + struct rte_table_lpm_ipv6 *lpm = (struct rte_table_lpm_ipv6 *) table; + + /* Check input parameters */ + if (lpm == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + + /* Free previously allocated resources */ + rte_lpm6_free(lpm->lpm); + rte_free(lpm); + + return 0; +} + +static int +nht_find_free(struct rte_table_lpm_ipv6 *lpm, uint32_t *pos) +{ + uint32_t i; + + for (i = 0; i < RTE_TABLE_LPM_MAX_NEXT_HOPS; i++) { + if (lpm->nht_users[i] == 0) { + *pos = i; + return 1; + } + } + + return 0; +} + +static int +nht_find_existing(struct rte_table_lpm_ipv6 *lpm, void *entry, uint32_t *pos) +{ + uint32_t i; + + for (i = 0; i < RTE_TABLE_LPM_MAX_NEXT_HOPS; i++) { + uint8_t *nht_entry = &lpm->nht[i * lpm->entry_size]; + + if ((lpm->nht_users[i] > 0) && (memcmp(nht_entry, entry, + lpm->entry_unique_size) == 0)) { + *pos = i; + return 1; + } + } + + return 0; +} + +static int +rte_table_lpm_ipv6_entry_add( + void *table, + void *key, + void *entry, + int *key_found, + void **entry_ptr) +{ + struct rte_table_lpm_ipv6 *lpm = (struct rte_table_lpm_ipv6 *) table; + struct rte_table_lpm_ipv6_key *ip_prefix = + (struct rte_table_lpm_ipv6_key *) key; + uint32_t nht_pos, nht_pos0_valid; + int status; + uint8_t nht_pos0; + + /* Check input parameters */ + if (lpm == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (ip_prefix == NULL) { + RTE_LOG(ERR, TABLE, "%s: ip_prefix parameter is NULL\n", + __func__); + return -EINVAL; + } + if (entry == NULL) { + RTE_LOG(ERR, TABLE, "%s: entry parameter is NULL\n", __func__); + return -EINVAL; + } + + if ((ip_prefix->depth == 0) || (ip_prefix->depth > 128)) { + RTE_LOG(ERR, TABLE, "%s: invalid depth (%d)\n", __func__, + ip_prefix->depth); + return -EINVAL; + } + + /* Check if rule is already present in the table */ + status = rte_lpm6_is_rule_present(lpm->lpm, ip_prefix->ip, + ip_prefix->depth, &nht_pos0); + nht_pos0_valid = status > 0; + + /* Find existing or free NHT entry */ + if (nht_find_existing(lpm, entry, &nht_pos) == 0) { + uint8_t *nht_entry; + + if (nht_find_free(lpm, &nht_pos) == 0) { + RTE_LOG(ERR, TABLE, "%s: NHT full\n", __func__); + return -1; + } + + nht_entry = &lpm->nht[nht_pos * lpm->entry_size]; + memcpy(nht_entry, entry, lpm->entry_size); + } + + /* Add rule to low level LPM table */ + if (rte_lpm6_add(lpm->lpm, ip_prefix->ip, ip_prefix->depth, + (uint8_t) nht_pos) < 0) { + RTE_LOG(ERR, TABLE, "%s: LPM IPv6 rule add failed\n", __func__); + return -1; + } + + /* Commit NHT changes */ + lpm->nht_users[nht_pos]++; + lpm->nht_users[nht_pos0] -= nht_pos0_valid; + + *key_found = nht_pos0_valid; + *entry_ptr = (void *) &lpm->nht[nht_pos * lpm->entry_size]; + return 0; +} + +static int +rte_table_lpm_ipv6_entry_delete( + void *table, + void *key, + int *key_found, + void *entry) +{ + struct rte_table_lpm_ipv6 *lpm = (struct rte_table_lpm_ipv6 *) table; + struct rte_table_lpm_ipv6_key *ip_prefix = + (struct rte_table_lpm_ipv6_key *) key; + uint8_t nht_pos; + int status; + + /* Check input parameters */ + if (lpm == NULL) { + RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__); + return -EINVAL; + } + if (ip_prefix == NULL) { + RTE_LOG(ERR, TABLE, "%s: ip_prefix parameter is NULL\n", + __func__); + return -EINVAL; + } + if ((ip_prefix->depth == 0) || (ip_prefix->depth > 128)) { + RTE_LOG(ERR, TABLE, "%s: invalid depth (%d)\n", __func__, + ip_prefix->depth); + return -EINVAL; + } + + /* Return if rule is not present in the table */ + status = rte_lpm6_is_rule_present(lpm->lpm, ip_prefix->ip, + ip_prefix->depth, &nht_pos); + if (status < 0) { + RTE_LOG(ERR, TABLE, "%s: LPM IPv6 algorithmic error\n", + __func__); + return -1; + } + if (status == 0) { + *key_found = 0; + return 0; + } + + /* Delete rule from the low-level LPM table */ + status = rte_lpm6_delete(lpm->lpm, ip_prefix->ip, ip_prefix->depth); + if (status) { + RTE_LOG(ERR, TABLE, "%s: LPM IPv6 rule delete failed\n", + __func__); + return -1; + } + + /* Commit NHT changes */ + lpm->nht_users[nht_pos]--; + + *key_found = 1; + if (entry) + memcpy(entry, &lpm->nht[nht_pos * lpm->entry_size], + lpm->entry_size); + + return 0; +} + +static int +rte_table_lpm_ipv6_lookup( + void *table, + struct rte_mbuf **pkts, + uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + void **entries) +{ + struct rte_table_lpm_ipv6 *lpm = (struct rte_table_lpm_ipv6 *) table; + uint64_t pkts_out_mask = 0; + uint32_t i; + + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + RTE_TABLE_LPM_IPV6_STATS_PKTS_IN_ADD(lpm, n_pkts_in); + + pkts_out_mask = 0; + for (i = 0; i < (uint32_t)(RTE_PORT_IN_BURST_SIZE_MAX - + __builtin_clzll(pkts_mask)); i++) { + uint64_t pkt_mask = 1LLU << i; + + if (pkt_mask & pkts_mask) { + struct rte_mbuf *pkt = pkts[i]; + uint8_t *ip = RTE_MBUF_METADATA_UINT8_PTR(pkt, + lpm->offset); + int status; + uint8_t nht_pos; + + status = rte_lpm6_lookup(lpm->lpm, ip, &nht_pos); + if (status == 0) { + pkts_out_mask |= pkt_mask; + entries[i] = (void *) &lpm->nht[nht_pos * + lpm->entry_size]; + } + } + } + + *lookup_hit_mask = pkts_out_mask; + RTE_TABLE_LPM_IPV6_STATS_PKTS_LOOKUP_MISS(lpm, n_pkts_in - __builtin_popcountll(pkts_out_mask)); + return 0; +} + +static int +rte_table_lpm_ipv6_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_lpm_ipv6 *t = (struct rte_table_lpm_ipv6 *) table; + + if (stats != NULL) + memcpy(stats, &t->stats, sizeof(t->stats)); + + if (clear) + memset(&t->stats, 0, sizeof(t->stats)); + + return 0; +} + +struct rte_table_ops rte_table_lpm_ipv6_ops = { + .f_create = rte_table_lpm_ipv6_create, + .f_free = rte_table_lpm_ipv6_free, + .f_add = rte_table_lpm_ipv6_entry_add, + .f_delete = rte_table_lpm_ipv6_entry_delete, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_lpm_ipv6_lookup, + .f_stats = rte_table_lpm_ipv6_stats_read, +}; diff --git a/lib/librte_table/rte_table_lpm_ipv6.h b/lib/librte_table/rte_table_lpm_ipv6.h new file mode 100644 index 00000000..43aea399 --- /dev/null +++ b/lib/librte_table/rte_table_lpm_ipv6.h @@ -0,0 +1,122 @@ +/*- + * 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_LPM_IPV6_H__ +#define __INCLUDE_RTE_TABLE_LPM_IPV6_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Table LPM for IPv6 + * + * This table uses the Longest Prefix Match (LPM) algorithm to uniquely + * associate data to lookup keys. + * + * Use-case: IP routing table. Routes that are added to the table associate a + * next hop to an IP prefix. The IP prefix is specified as IP address and depth + * and cover for a multitude of lookup keys (i.e. destination IP addresses) + * that all share the same data (i.e. next hop). The next hop information + * typically contains the output interface ID, the IP address of the next hop + * station (which is part of the same IP network the output interface is + * connected to) and other flags and counters. + * + * The LPM primitive only allows associating an 8-bit number (next hop ID) to + * an IP prefix, while a routing table can potentially contain thousands of + * routes or even more. This means that the same next hop ID (and next hop + * information) has to be shared by multiple routes, which makes sense, as + * multiple remote networks could be reached through the same next hop. + * Therefore, when a route is added or updated, the LPM table has to check + * whether the same next hop is already in use before using a new next hop ID + * for this route. + * + * The comparison between different next hops is done for the first + * “entry_unique_size” bytes of the next hop information (configurable + * parameter), which have to uniquely identify the next hop, therefore the user + * has to carefully manage the format of the LPM table entry (i.e. the next + * hop information) so that any next hop data that changes value during + * run-time (e.g. counters) is placed outside of this area. + * + ***/ + +#include <stdint.h> + +#include "rte_table.h" + +#define RTE_LPM_IPV6_ADDR_SIZE 16 + +/** LPM table parameters */ +struct rte_table_lpm_ipv6_params { + /** Table name */ + const char *name; + + /** Maximum number of LPM rules (i.e. IP routes) */ + uint32_t n_rules; + + uint32_t number_tbl8s; + + /** Number of bytes at the start of the table entry that uniquely + identify the entry. Cannot be bigger than table entry size. */ + uint32_t entry_unique_size; + + /** Byte offset within input packet meta-data where lookup key (i.e. + the destination IP address) is located. */ + uint32_t offset; +}; + +/** LPM table rule (i.e. route), specified as IP prefix. While the key used by +the lookup operation is the destination IP address (read from the input packet +meta-data), the entry add and entry delete operations work with LPM rules, with +each rule covering for a multitude of lookup keys (destination IP addresses) +that share the same data (next hop). */ +struct rte_table_lpm_ipv6_key { + /** IP address */ + uint8_t ip[RTE_LPM_IPV6_ADDR_SIZE]; + + /** IP address depth. The most significant "depth" bits of the IP + address specify the network part of the IP address, while the rest of + the bits specify the host part of the address and are ignored for the + purpose of route specification. */ + uint8_t depth; +}; + +/** LPM table operations */ +extern struct rte_table_ops rte_table_lpm_ipv6_ops; + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/lib/librte_table/rte_table_stub.c b/lib/librte_table/rte_table_stub.c new file mode 100644 index 00000000..691d681a --- /dev/null +++ b/lib/librte_table/rte_table_stub.c @@ -0,0 +1,121 @@ +/*- + * 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 <string.h> + +#include <rte_mbuf.h> +#include <rte_malloc.h> + +#include "rte_table_stub.h" + +#ifdef RTE_TABLE_STATS_COLLECT + +#define RTE_TABLE_LPM_STATS_PKTS_IN_ADD(table, val) \ + table->stats.n_pkts_in += val +#define RTE_TABLE_LPM_STATS_PKTS_LOOKUP_MISS(table, val) \ + table->stats.n_pkts_lookup_miss += val + +#else + +#define RTE_TABLE_LPM_STATS_PKTS_IN_ADD(table, val) +#define RTE_TABLE_LPM_STATS_PKTS_LOOKUP_MISS(table, val) + +#endif + +struct rte_table_stub { + struct rte_table_stats stats; +}; + +static void * +rte_table_stub_create(__rte_unused void *params, + __rte_unused int socket_id, + __rte_unused uint32_t entry_size) +{ + struct rte_table_stub *stub; + uint32_t size; + + size = sizeof(struct rte_table_stub); + stub = rte_zmalloc_socket("TABLE", size, RTE_CACHE_LINE_SIZE, + socket_id); + if (stub == NULL) { + RTE_LOG(ERR, TABLE, + "%s: Cannot allocate %u bytes for stub table\n", + __func__, size); + return NULL; + } + + return stub; +} + +static int +rte_table_stub_lookup( + __rte_unused void *table, + __rte_unused struct rte_mbuf **pkts, + __rte_unused uint64_t pkts_mask, + uint64_t *lookup_hit_mask, + __rte_unused void **entries) +{ + __rte_unused struct rte_table_stub *stub = (struct rte_table_stub *) table; + __rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask); + + RTE_TABLE_LPM_STATS_PKTS_IN_ADD(stub, n_pkts_in); + *lookup_hit_mask = 0; + RTE_TABLE_LPM_STATS_PKTS_LOOKUP_MISS(stub, n_pkts_in); + + return 0; +} + +static int +rte_table_stub_stats_read(void *table, struct rte_table_stats *stats, int clear) +{ + struct rte_table_stub *t = (struct rte_table_stub *) table; + + if (stats != NULL) + memcpy(stats, &t->stats, sizeof(t->stats)); + + if (clear) + memset(&t->stats, 0, sizeof(t->stats)); + + return 0; +} + +struct rte_table_ops rte_table_stub_ops = { + .f_create = rte_table_stub_create, + .f_free = NULL, + .f_add = NULL, + .f_delete = NULL, + .f_add_bulk = NULL, + .f_delete_bulk = NULL, + .f_lookup = rte_table_stub_lookup, + .f_stats = rte_table_stub_stats_read, +}; diff --git a/lib/librte_table/rte_table_stub.h b/lib/librte_table/rte_table_stub.h new file mode 100644 index 00000000..e75340b0 --- /dev/null +++ b/lib/librte_table/rte_table_stub.h @@ -0,0 +1,62 @@ +/*- + * 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_STUB_H__ +#define __INCLUDE_RTE_TABLE_STUB_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Table Stub + * + * The stub table lookup operation produces lookup miss for all input packets. + * + ***/ + +#include <stdint.h> + +#include "rte_table.h" + +/** Stub table parameters: NONE */ + +/** Stub table operations */ +extern struct rte_table_ops rte_table_stub_ops; + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/lib/librte_table/rte_table_version.map b/lib/librte_table/rte_table_version.map new file mode 100644 index 00000000..21386984 --- /dev/null +++ b/lib/librte_table/rte_table_version.map @@ -0,0 +1,28 @@ +DPDK_2.0 { + global: + + rte_table_acl_ops; + rte_table_array_ops; + rte_table_hash_ext_ops; + rte_table_hash_key8_ext_dosig_ops; + rte_table_hash_key8_ext_ops; + rte_table_hash_key8_lru_dosig_ops; + rte_table_hash_key8_lru_ops; + rte_table_hash_key16_ext_ops; + rte_table_hash_key16_lru_ops; + rte_table_hash_key32_ext_ops; + rte_table_hash_key32_lru_ops; + rte_table_hash_lru_ops; + rte_table_lpm_ipv6_ops; + rte_table_lpm_ops; + rte_table_stub_ops; + + local: *; +}; + +DPDK_2.2 { + global: + + rte_table_hash_key16_ext_dosig_ops; + +} DPDK_2.0; |