4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 #include <rte_common.h>
38 #include <rte_malloc.h>
41 #include "rte_table_hash.h"
44 #define RTE_TABLE_HASH_KEY_SIZE 32
46 #define RTE_BUCKET_ENTRY_VALID 0x1LLU
48 struct rte_bucket_4_32 {
50 uint64_t signature[4 + 1];
52 struct rte_bucket_4_32 *next;
55 /* Cache lines 1 and 2 */
62 struct rte_table_hash {
63 /* Input parameters */
65 uint32_t n_entries_per_bucket;
69 uint32_t signature_offset;
71 rte_table_hash_op_hash f_hash;
74 /* Extendible buckets */
75 uint32_t n_buckets_ext;
80 uint8_t memory[0] __rte_cache_aligned;
84 check_params_create_lru(struct rte_table_hash_key32_lru_params *params) {
86 if (params->n_entries == 0) {
87 RTE_LOG(ERR, TABLE, "%s: n_entries is zero\n", __func__);
91 /* signature offset */
92 if ((params->signature_offset & 0x3) != 0) {
93 RTE_LOG(ERR, TABLE, "%s: invalid signature offset\n", __func__);
98 if ((params->key_offset & 0x7) != 0) {
99 RTE_LOG(ERR, TABLE, "%s: invalid key offset\n", __func__);
104 if (params->f_hash == NULL) {
105 RTE_LOG(ERR, TABLE, "%s: f_hash function pointer is NULL\n",
114 rte_table_hash_create_key32_lru(void *params,
118 struct rte_table_hash_key32_lru_params *p =
119 (struct rte_table_hash_key32_lru_params *) params;
120 struct rte_table_hash *f;
121 uint32_t n_buckets, n_entries_per_bucket, key_size, bucket_size_cl;
122 uint32_t total_size, i;
124 /* Check input parameters */
125 if ((check_params_create_lru(p) != 0) ||
126 ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) ||
127 ((sizeof(struct rte_bucket_4_32) % CACHE_LINE_SIZE) != 0)) {
130 n_entries_per_bucket = 4;
133 /* Memory allocation */
134 n_buckets = rte_align32pow2((p->n_entries + n_entries_per_bucket - 1) /
135 n_entries_per_bucket);
136 bucket_size_cl = (sizeof(struct rte_bucket_4_32) + n_entries_per_bucket
137 * entry_size + CACHE_LINE_SIZE - 1) / CACHE_LINE_SIZE;
138 total_size = sizeof(struct rte_table_hash) + n_buckets *
139 bucket_size_cl * CACHE_LINE_SIZE;
141 f = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id);
144 "%s: Cannot allocate %u bytes for hash table\n",
145 __func__, total_size);
149 "%s: Hash table memory footprint is %u bytes\n", __func__,
152 /* Memory initialization */
153 f->n_buckets = n_buckets;
154 f->n_entries_per_bucket = n_entries_per_bucket;
155 f->key_size = key_size;
156 f->entry_size = entry_size;
157 f->bucket_size = bucket_size_cl * CACHE_LINE_SIZE;
158 f->signature_offset = p->signature_offset;
159 f->key_offset = p->key_offset;
160 f->f_hash = p->f_hash;
163 for (i = 0; i < n_buckets; i++) {
164 struct rte_bucket_4_32 *bucket;
166 bucket = (struct rte_bucket_4_32 *) &f->memory[i *
168 bucket->lru_list = 0x0000000100020003LLU;
175 rte_table_hash_free_key32_lru(void *table)
177 struct rte_table_hash *f = (struct rte_table_hash *) table;
179 /* Check input parameters */
181 RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__);
190 rte_table_hash_entry_add_key32_lru(
197 struct rte_table_hash *f = (struct rte_table_hash *) table;
198 struct rte_bucket_4_32 *bucket;
199 uint64_t signature, pos;
200 uint32_t bucket_index, i;
202 signature = f->f_hash(key, f->key_size, f->seed);
203 bucket_index = signature & (f->n_buckets - 1);
204 bucket = (struct rte_bucket_4_32 *)
205 &f->memory[bucket_index * f->bucket_size];
206 signature |= RTE_BUCKET_ENTRY_VALID;
208 /* Key is present in the bucket */
209 for (i = 0; i < 4; i++) {
210 uint64_t bucket_signature = bucket->signature[i];
211 uint8_t *bucket_key = (uint8_t *) bucket->key[i];
213 if ((bucket_signature == signature) &&
214 (memcmp(key, bucket_key, f->key_size) == 0)) {
215 uint8_t *bucket_data = &bucket->data[i * f->entry_size];
217 memcpy(bucket_data, entry, f->entry_size);
218 lru_update(bucket, i);
220 *entry_ptr = (void *) bucket_data;
225 /* Key is not present in the bucket */
226 for (i = 0; i < 4; i++) {
227 uint64_t bucket_signature = bucket->signature[i];
228 uint8_t *bucket_key = (uint8_t *) bucket->key[i];
230 if (bucket_signature == 0) {
231 uint8_t *bucket_data = &bucket->data[i * f->entry_size];
233 bucket->signature[i] = signature;
234 memcpy(bucket_key, key, f->key_size);
235 memcpy(bucket_data, entry, f->entry_size);
236 lru_update(bucket, i);
238 *entry_ptr = (void *) bucket_data;
244 /* Bucket full: replace LRU entry */
245 pos = lru_pos(bucket);
246 bucket->signature[pos] = signature;
247 memcpy(bucket->key[pos], key, f->key_size);
248 memcpy(&bucket->data[pos * f->entry_size], entry, f->entry_size);
249 lru_update(bucket, pos);
251 *entry_ptr = (void *) &bucket->data[pos * f->entry_size];
257 rte_table_hash_entry_delete_key32_lru(
263 struct rte_table_hash *f = (struct rte_table_hash *) table;
264 struct rte_bucket_4_32 *bucket;
266 uint32_t bucket_index, i;
268 signature = f->f_hash(key, f->key_size, f->seed);
269 bucket_index = signature & (f->n_buckets - 1);
270 bucket = (struct rte_bucket_4_32 *)
271 &f->memory[bucket_index * f->bucket_size];
272 signature |= RTE_BUCKET_ENTRY_VALID;
274 /* Key is present in the bucket */
275 for (i = 0; i < 4; i++) {
276 uint64_t bucket_signature = bucket->signature[i];
277 uint8_t *bucket_key = (uint8_t *) bucket->key[i];
279 if ((bucket_signature == signature) &&
280 (memcmp(key, bucket_key, f->key_size) == 0)) {
281 uint8_t *bucket_data = &bucket->data[i * f->entry_size];
283 bucket->signature[i] = 0;
286 memcpy(entry, bucket_data, f->entry_size);
292 /* Key is not present in the bucket */
298 check_params_create_ext(struct rte_table_hash_key32_ext_params *params) {
300 if (params->n_entries == 0) {
301 RTE_LOG(ERR, TABLE, "%s: n_entries is zero\n", __func__);
306 if (params->n_entries_ext == 0) {
307 RTE_LOG(ERR, TABLE, "%s: n_entries_ext is zero\n", __func__);
311 /* signature offset */
312 if ((params->signature_offset & 0x3) != 0) {
313 RTE_LOG(ERR, TABLE, "%s: invalid signature offset\n", __func__);
318 if ((params->key_offset & 0x7) != 0) {
319 RTE_LOG(ERR, TABLE, "%s: invalid key offset\n", __func__);
324 if (params->f_hash == NULL) {
325 RTE_LOG(ERR, TABLE, "%s: f_hash function pointer is NULL\n",
334 rte_table_hash_create_key32_ext(void *params,
338 struct rte_table_hash_key32_ext_params *p =
339 (struct rte_table_hash_key32_ext_params *) params;
340 struct rte_table_hash *f;
341 uint32_t n_buckets, n_buckets_ext, n_entries_per_bucket;
342 uint32_t key_size, bucket_size_cl, stack_size_cl, total_size, i;
344 /* Check input parameters */
345 if ((check_params_create_ext(p) != 0) ||
346 ((sizeof(struct rte_table_hash) % CACHE_LINE_SIZE) != 0) ||
347 ((sizeof(struct rte_bucket_4_32) % CACHE_LINE_SIZE) != 0))
350 n_entries_per_bucket = 4;
353 /* Memory allocation */
354 n_buckets = rte_align32pow2((p->n_entries + n_entries_per_bucket - 1) /
355 n_entries_per_bucket);
356 n_buckets_ext = (p->n_entries_ext + n_entries_per_bucket - 1) /
357 n_entries_per_bucket;
358 bucket_size_cl = (sizeof(struct rte_bucket_4_32) + n_entries_per_bucket
359 * entry_size + CACHE_LINE_SIZE - 1) / CACHE_LINE_SIZE;
360 stack_size_cl = (n_buckets_ext * sizeof(uint32_t) + CACHE_LINE_SIZE - 1)
362 total_size = sizeof(struct rte_table_hash) +
363 ((n_buckets + n_buckets_ext) * bucket_size_cl + stack_size_cl) *
366 f = rte_zmalloc_socket("TABLE", total_size, CACHE_LINE_SIZE, socket_id);
369 "%s: Cannot allocate %u bytes for hash table\n",
370 __func__, total_size);
374 "%s: Hash table memory footprint is %u bytes\n", __func__,
377 /* Memory initialization */
378 f->n_buckets = n_buckets;
379 f->n_entries_per_bucket = n_entries_per_bucket;
380 f->key_size = key_size;
381 f->entry_size = entry_size;
382 f->bucket_size = bucket_size_cl * CACHE_LINE_SIZE;
383 f->signature_offset = p->signature_offset;
384 f->key_offset = p->key_offset;
385 f->f_hash = p->f_hash;
388 f->n_buckets_ext = n_buckets_ext;
389 f->stack_pos = n_buckets_ext;
390 f->stack = (uint32_t *)
391 &f->memory[(n_buckets + n_buckets_ext) * f->bucket_size];
393 for (i = 0; i < n_buckets_ext; i++)
400 rte_table_hash_free_key32_ext(void *table)
402 struct rte_table_hash *f = (struct rte_table_hash *) table;
404 /* Check input parameters */
406 RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__);
415 rte_table_hash_entry_add_key32_ext(
422 struct rte_table_hash *f = (struct rte_table_hash *) table;
423 struct rte_bucket_4_32 *bucket0, *bucket, *bucket_prev;
425 uint32_t bucket_index, i;
427 signature = f->f_hash(key, f->key_size, f->seed);
428 bucket_index = signature & (f->n_buckets - 1);
429 bucket0 = (struct rte_bucket_4_32 *)
430 &f->memory[bucket_index * f->bucket_size];
431 signature |= RTE_BUCKET_ENTRY_VALID;
433 /* Key is present in the bucket */
434 for (bucket = bucket0; bucket != NULL; bucket = bucket->next) {
435 for (i = 0; i < 4; i++) {
436 uint64_t bucket_signature = bucket->signature[i];
437 uint8_t *bucket_key = (uint8_t *) bucket->key[i];
439 if ((bucket_signature == signature) &&
440 (memcmp(key, bucket_key, f->key_size) == 0)) {
441 uint8_t *bucket_data = &bucket->data[i *
444 memcpy(bucket_data, entry, f->entry_size);
446 *entry_ptr = (void *) bucket_data;
453 /* Key is not present in the bucket */
454 for (bucket_prev = NULL, bucket = bucket0; bucket != NULL;
455 bucket_prev = bucket, bucket = bucket->next)
456 for (i = 0; i < 4; i++) {
457 uint64_t bucket_signature = bucket->signature[i];
458 uint8_t *bucket_key = (uint8_t *) bucket->key[i];
460 if (bucket_signature == 0) {
461 uint8_t *bucket_data = &bucket->data[i *
464 bucket->signature[i] = signature;
465 memcpy(bucket_key, key, f->key_size);
466 memcpy(bucket_data, entry, f->entry_size);
468 *entry_ptr = (void *) bucket_data;
474 /* Bucket full: extend bucket */
475 if (f->stack_pos > 0) {
476 bucket_index = f->stack[--f->stack_pos];
478 bucket = (struct rte_bucket_4_32 *)
479 &f->memory[(f->n_buckets + bucket_index) *
481 bucket_prev->next = bucket;
482 bucket_prev->next_valid = 1;
484 bucket->signature[0] = signature;
485 memcpy(bucket->key[0], key, f->key_size);
486 memcpy(&bucket->data[0], entry, f->entry_size);
488 *entry_ptr = (void *) &bucket->data[0];
496 rte_table_hash_entry_delete_key32_ext(
502 struct rte_table_hash *f = (struct rte_table_hash *) table;
503 struct rte_bucket_4_32 *bucket0, *bucket, *bucket_prev;
505 uint32_t bucket_index, i;
507 signature = f->f_hash(key, f->key_size, f->seed);
508 bucket_index = signature & (f->n_buckets - 1);
509 bucket0 = (struct rte_bucket_4_32 *)
510 &f->memory[bucket_index * f->bucket_size];
511 signature |= RTE_BUCKET_ENTRY_VALID;
513 /* Key is present in the bucket */
514 for (bucket_prev = NULL, bucket = bucket0; bucket != NULL;
515 bucket_prev = bucket, bucket = bucket->next)
516 for (i = 0; i < 4; i++) {
517 uint64_t bucket_signature = bucket->signature[i];
518 uint8_t *bucket_key = (uint8_t *) bucket->key[i];
520 if ((bucket_signature == signature) &&
521 (memcmp(key, bucket_key, f->key_size) == 0)) {
522 uint8_t *bucket_data = &bucket->data[i *
525 bucket->signature[i] = 0;
528 memcpy(entry, bucket_data,
531 if ((bucket->signature[0] == 0) &&
532 (bucket->signature[1] == 0) &&
533 (bucket->signature[2] == 0) &&
534 (bucket->signature[3] == 0) &&
535 (bucket_prev != NULL)) {
536 bucket_prev->next = bucket->next;
537 bucket_prev->next_valid =
541 sizeof(struct rte_bucket_4_32));
542 bucket_index = (bucket -
543 ((struct rte_bucket_4_32 *)
544 f->memory)) - f->n_buckets;
545 f->stack[f->stack_pos++] = bucket_index;
552 /* Key is not present in the bucket */
557 #define lookup_key32_cmp(key_in, bucket, pos) \
559 uint64_t xor[4][4], or[4], signature[4]; \
561 signature[0] = ((~bucket->signature[0]) & 1); \
562 signature[1] = ((~bucket->signature[1]) & 1); \
563 signature[2] = ((~bucket->signature[2]) & 1); \
564 signature[3] = ((~bucket->signature[3]) & 1); \
566 xor[0][0] = key_in[0] ^ bucket->key[0][0]; \
567 xor[0][1] = key_in[1] ^ bucket->key[0][1]; \
568 xor[0][2] = key_in[2] ^ bucket->key[0][2]; \
569 xor[0][3] = key_in[3] ^ bucket->key[0][3]; \
571 xor[1][0] = key_in[0] ^ bucket->key[1][0]; \
572 xor[1][1] = key_in[1] ^ bucket->key[1][1]; \
573 xor[1][2] = key_in[2] ^ bucket->key[1][2]; \
574 xor[1][3] = key_in[3] ^ bucket->key[1][3]; \
576 xor[2][0] = key_in[0] ^ bucket->key[2][0]; \
577 xor[2][1] = key_in[1] ^ bucket->key[2][1]; \
578 xor[2][2] = key_in[2] ^ bucket->key[2][2]; \
579 xor[2][3] = key_in[3] ^ bucket->key[2][3]; \
581 xor[3][0] = key_in[0] ^ bucket->key[3][0]; \
582 xor[3][1] = key_in[1] ^ bucket->key[3][1]; \
583 xor[3][2] = key_in[2] ^ bucket->key[3][2]; \
584 xor[3][3] = key_in[3] ^ bucket->key[3][3]; \
586 or[0] = xor[0][0] | xor[0][1] | xor[0][2] | xor[0][3] | signature[0];\
587 or[1] = xor[1][0] | xor[1][1] | xor[1][2] | xor[1][3] | signature[1];\
588 or[2] = xor[2][0] | xor[2][1] | xor[2][2] | xor[2][3] | signature[2];\
589 or[3] = xor[3][0] | xor[3][1] | xor[3][2] | xor[3][3] | signature[3];\
602 #define lookup1_stage0(pkt0_index, mbuf0, pkts, pkts_mask) \
606 pkt0_index = __builtin_ctzll(pkts_mask); \
607 pkt_mask = 1LLU << pkt0_index; \
608 pkts_mask &= ~pkt_mask; \
610 mbuf0 = pkts[pkt0_index]; \
611 rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf0, 0)); \
614 #define lookup1_stage1(mbuf1, bucket1, f) \
616 uint64_t signature; \
617 uint32_t bucket_index; \
619 signature = RTE_MBUF_METADATA_UINT32(mbuf1, f->signature_offset);\
620 bucket_index = signature & (f->n_buckets - 1); \
621 bucket1 = (struct rte_bucket_4_32 *) \
622 &f->memory[bucket_index * f->bucket_size]; \
623 rte_prefetch0(bucket1); \
624 rte_prefetch0((void *)(((uintptr_t) bucket1) + CACHE_LINE_SIZE));\
625 rte_prefetch0((void *)(((uintptr_t) bucket1) + 2 * CACHE_LINE_SIZE));\
628 #define lookup1_stage2_lru(pkt2_index, mbuf2, bucket2, \
629 pkts_mask_out, entries, f) \
636 key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\
638 lookup_key32_cmp(key, bucket2, pos); \
640 pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\
641 pkts_mask_out |= pkt_mask; \
643 a = (void *) &bucket2->data[pos * f->entry_size]; \
645 entries[pkt2_index] = a; \
646 lru_update(bucket2, pos); \
649 #define lookup1_stage2_ext(pkt2_index, mbuf2, bucket2, pkts_mask_out,\
650 entries, buckets_mask, buckets, keys, f) \
652 struct rte_bucket_4_32 *bucket_next; \
654 uint64_t pkt_mask, bucket_mask; \
658 key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\
660 lookup_key32_cmp(key, bucket2, pos); \
662 pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\
663 pkts_mask_out |= pkt_mask; \
665 a = (void *) &bucket2->data[pos * f->entry_size]; \
667 entries[pkt2_index] = a; \
669 bucket_mask = (~pkt_mask) & (bucket2->next_valid << pkt2_index);\
670 buckets_mask |= bucket_mask; \
671 bucket_next = bucket2->next; \
672 buckets[pkt2_index] = bucket_next; \
673 keys[pkt2_index] = key; \
676 #define lookup_grinder(pkt_index, buckets, keys, pkts_mask_out, \
677 entries, buckets_mask, f) \
679 struct rte_bucket_4_32 *bucket, *bucket_next; \
681 uint64_t pkt_mask, bucket_mask; \
685 bucket = buckets[pkt_index]; \
686 key = keys[pkt_index]; \
688 lookup_key32_cmp(key, bucket, pos); \
690 pkt_mask = (bucket->signature[pos] & 1LLU) << pkt_index;\
691 pkts_mask_out |= pkt_mask; \
693 a = (void *) &bucket->data[pos * f->entry_size]; \
695 entries[pkt_index] = a; \
697 bucket_mask = (~pkt_mask) & (bucket->next_valid << pkt_index);\
698 buckets_mask |= bucket_mask; \
699 bucket_next = bucket->next; \
700 rte_prefetch0(bucket_next); \
701 rte_prefetch0((void *)(((uintptr_t) bucket_next) + CACHE_LINE_SIZE));\
702 rte_prefetch0((void *)(((uintptr_t) bucket_next) + \
703 2 * CACHE_LINE_SIZE)); \
704 buckets[pkt_index] = bucket_next; \
705 keys[pkt_index] = key; \
708 #define lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01,\
711 uint64_t pkt00_mask, pkt01_mask; \
713 pkt00_index = __builtin_ctzll(pkts_mask); \
714 pkt00_mask = 1LLU << pkt00_index; \
715 pkts_mask &= ~pkt00_mask; \
717 mbuf00 = pkts[pkt00_index]; \
718 rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \
720 pkt01_index = __builtin_ctzll(pkts_mask); \
721 pkt01_mask = 1LLU << pkt01_index; \
722 pkts_mask &= ~pkt01_mask; \
724 mbuf01 = pkts[pkt01_index]; \
725 rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \
728 #define lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,\
729 mbuf00, mbuf01, pkts, pkts_mask) \
731 uint64_t pkt00_mask, pkt01_mask; \
733 pkt00_index = __builtin_ctzll(pkts_mask); \
734 pkt00_mask = 1LLU << pkt00_index; \
735 pkts_mask &= ~pkt00_mask; \
737 mbuf00 = pkts[pkt00_index]; \
738 rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, 0)); \
740 pkt01_index = __builtin_ctzll(pkts_mask); \
741 if (pkts_mask == 0) \
742 pkt01_index = pkt00_index; \
744 pkt01_mask = 1LLU << pkt01_index; \
745 pkts_mask &= ~pkt01_mask; \
747 mbuf01 = pkts[pkt01_index]; \
748 rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, 0)); \
751 #define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f) \
753 uint64_t signature10, signature11; \
754 uint32_t bucket10_index, bucket11_index; \
756 signature10 = RTE_MBUF_METADATA_UINT32(mbuf10, f->signature_offset);\
757 bucket10_index = signature10 & (f->n_buckets - 1); \
758 bucket10 = (struct rte_bucket_4_32 *) \
759 &f->memory[bucket10_index * f->bucket_size]; \
760 rte_prefetch0(bucket10); \
761 rte_prefetch0((void *)(((uintptr_t) bucket10) + CACHE_LINE_SIZE));\
762 rte_prefetch0((void *)(((uintptr_t) bucket10) + 2 * CACHE_LINE_SIZE));\
764 signature11 = RTE_MBUF_METADATA_UINT32(mbuf11, f->signature_offset);\
765 bucket11_index = signature11 & (f->n_buckets - 1); \
766 bucket11 = (struct rte_bucket_4_32 *) \
767 &f->memory[bucket11_index * f->bucket_size]; \
768 rte_prefetch0(bucket11); \
769 rte_prefetch0((void *)(((uintptr_t) bucket11) + CACHE_LINE_SIZE));\
770 rte_prefetch0((void *)(((uintptr_t) bucket11) + 2 * CACHE_LINE_SIZE));\
773 #define lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21,\
774 bucket20, bucket21, pkts_mask_out, entries, f) \
777 uint64_t pkt20_mask, pkt21_mask; \
778 uint64_t *key20, *key21; \
779 uint32_t pos20, pos21; \
781 key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\
782 key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\
784 lookup_key32_cmp(key20, bucket20, pos20); \
785 lookup_key32_cmp(key21, bucket21, pos21); \
787 pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\
788 pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\
789 pkts_mask_out |= pkt20_mask | pkt21_mask; \
791 a20 = (void *) &bucket20->data[pos20 * f->entry_size]; \
792 a21 = (void *) &bucket21->data[pos21 * f->entry_size]; \
793 rte_prefetch0(a20); \
794 rte_prefetch0(a21); \
795 entries[pkt20_index] = a20; \
796 entries[pkt21_index] = a21; \
797 lru_update(bucket20, pos20); \
798 lru_update(bucket21, pos21); \
801 #define lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, bucket20, \
802 bucket21, pkts_mask_out, entries, buckets_mask, buckets, keys, f)\
804 struct rte_bucket_4_32 *bucket20_next, *bucket21_next; \
806 uint64_t pkt20_mask, pkt21_mask, bucket20_mask, bucket21_mask;\
807 uint64_t *key20, *key21; \
808 uint32_t pos20, pos21; \
810 key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\
811 key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\
813 lookup_key32_cmp(key20, bucket20, pos20); \
814 lookup_key32_cmp(key21, bucket21, pos21); \
816 pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\
817 pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\
818 pkts_mask_out |= pkt20_mask | pkt21_mask; \
820 a20 = (void *) &bucket20->data[pos20 * f->entry_size]; \
821 a21 = (void *) &bucket21->data[pos21 * f->entry_size]; \
822 rte_prefetch0(a20); \
823 rte_prefetch0(a21); \
824 entries[pkt20_index] = a20; \
825 entries[pkt21_index] = a21; \
827 bucket20_mask = (~pkt20_mask) & (bucket20->next_valid << pkt20_index);\
828 bucket21_mask = (~pkt21_mask) & (bucket21->next_valid << pkt21_index);\
829 buckets_mask |= bucket20_mask | bucket21_mask; \
830 bucket20_next = bucket20->next; \
831 bucket21_next = bucket21->next; \
832 buckets[pkt20_index] = bucket20_next; \
833 buckets[pkt21_index] = bucket21_next; \
834 keys[pkt20_index] = key20; \
835 keys[pkt21_index] = key21; \
839 rte_table_hash_lookup_key32_lru(
841 struct rte_mbuf **pkts,
843 uint64_t *lookup_hit_mask,
846 struct rte_table_hash *f = (struct rte_table_hash *) table;
847 struct rte_bucket_4_32 *bucket10, *bucket11, *bucket20, *bucket21;
848 struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21;
849 uint32_t pkt00_index, pkt01_index, pkt10_index;
850 uint32_t pkt11_index, pkt20_index, pkt21_index;
851 uint64_t pkts_mask_out = 0;
853 /* Cannot run the pipeline with less than 5 packets */
854 if (__builtin_popcountll(pkts_mask) < 5) {
855 for ( ; pkts_mask; ) {
856 struct rte_bucket_4_32 *bucket;
857 struct rte_mbuf *mbuf;
860 lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask);
861 lookup1_stage1(mbuf, bucket, f);
862 lookup1_stage2_lru(pkt_index, mbuf, bucket,
863 pkts_mask_out, entries, f);
866 *lookup_hit_mask = pkts_mask_out;
874 /* Pipeline stage 0 */
875 lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts,
881 pkt10_index = pkt00_index;
882 pkt11_index = pkt01_index;
884 /* Pipeline stage 0 */
885 lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts,
888 /* Pipeline stage 1 */
889 lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
895 for ( ; pkts_mask; ) {
903 pkt20_index = pkt10_index;
904 pkt21_index = pkt11_index;
905 pkt10_index = pkt00_index;
906 pkt11_index = pkt01_index;
908 /* Pipeline stage 0 */
909 lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,
910 mbuf00, mbuf01, pkts, pkts_mask);
912 /* Pipeline stage 1 */
913 lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
915 /* Pipeline stage 2 */
916 lookup2_stage2_lru(pkt20_index, pkt21_index,
917 mbuf20, mbuf21, bucket20, bucket21, pkts_mask_out,
932 pkt20_index = pkt10_index;
933 pkt21_index = pkt11_index;
934 pkt10_index = pkt00_index;
935 pkt11_index = pkt01_index;
937 /* Pipeline stage 1 */
938 lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
940 /* Pipeline stage 2 */
941 lookup2_stage2_lru(pkt20_index, pkt21_index,
942 mbuf20, mbuf21, bucket20, bucket21, pkts_mask_out, entries, f);
949 pkt20_index = pkt10_index;
950 pkt21_index = pkt11_index;
952 /* Pipeline stage 2 */
953 lookup2_stage2_lru(pkt20_index, pkt21_index,
954 mbuf20, mbuf21, bucket20, bucket21, pkts_mask_out, entries, f);
956 *lookup_hit_mask = pkts_mask_out;
958 } /* rte_table_hash_lookup_key32_lru() */
961 rte_table_hash_lookup_key32_ext(
963 struct rte_mbuf **pkts,
965 uint64_t *lookup_hit_mask,
968 struct rte_table_hash *f = (struct rte_table_hash *) table;
969 struct rte_bucket_4_32 *bucket10, *bucket11, *bucket20, *bucket21;
970 struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21;
971 uint32_t pkt00_index, pkt01_index, pkt10_index;
972 uint32_t pkt11_index, pkt20_index, pkt21_index;
973 uint64_t pkts_mask_out = 0, buckets_mask = 0;
974 struct rte_bucket_4_32 *buckets[RTE_PORT_IN_BURST_SIZE_MAX];
975 uint64_t *keys[RTE_PORT_IN_BURST_SIZE_MAX];
977 /* Cannot run the pipeline with less than 5 packets */
978 if (__builtin_popcountll(pkts_mask) < 5) {
979 for ( ; pkts_mask; ) {
980 struct rte_bucket_4_32 *bucket;
981 struct rte_mbuf *mbuf;
984 lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask);
985 lookup1_stage1(mbuf, bucket, f);
986 lookup1_stage2_ext(pkt_index, mbuf, bucket,
987 pkts_mask_out, entries, buckets_mask, buckets,
991 *lookup_hit_mask = pkts_mask_out;
999 /* Pipeline stage 0 */
1000 lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts,
1006 pkt10_index = pkt00_index;
1007 pkt11_index = pkt01_index;
1009 /* Pipeline stage 0 */
1010 lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts,
1013 /* Pipeline stage 1 */
1014 lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
1020 for ( ; pkts_mask; ) {
1022 bucket20 = bucket10;
1023 bucket21 = bucket11;
1028 pkt20_index = pkt10_index;
1029 pkt21_index = pkt11_index;
1030 pkt10_index = pkt00_index;
1031 pkt11_index = pkt01_index;
1033 /* Pipeline stage 0 */
1034 lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,
1035 mbuf00, mbuf01, pkts, pkts_mask);
1037 /* Pipeline stage 1 */
1038 lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
1040 /* Pipeline stage 2 */
1041 lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21,
1042 bucket20, bucket21, pkts_mask_out, entries,
1043 buckets_mask, buckets, keys, f);
1051 bucket20 = bucket10;
1052 bucket21 = bucket11;
1057 pkt20_index = pkt10_index;
1058 pkt21_index = pkt11_index;
1059 pkt10_index = pkt00_index;
1060 pkt11_index = pkt01_index;
1062 /* Pipeline stage 1 */
1063 lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
1065 /* Pipeline stage 2 */
1066 lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21,
1067 bucket20, bucket21, pkts_mask_out, entries,
1068 buckets_mask, buckets, keys, f);
1071 bucket20 = bucket10;
1072 bucket21 = bucket11;
1075 pkt20_index = pkt10_index;
1076 pkt21_index = pkt11_index;
1078 /* Pipeline stage 2 */
1079 lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21,
1080 bucket20, bucket21, pkts_mask_out, entries,
1081 buckets_mask, buckets, keys, f);
1083 /* Grind next buckets */
1084 for ( ; buckets_mask; ) {
1085 uint64_t buckets_mask_next = 0;
1087 for ( ; buckets_mask; ) {
1091 pkt_index = __builtin_ctzll(buckets_mask);
1092 pkt_mask = 1LLU << pkt_index;
1093 buckets_mask &= ~pkt_mask;
1095 lookup_grinder(pkt_index, buckets, keys, pkts_mask_out,
1096 entries, buckets_mask_next, f);
1099 buckets_mask = buckets_mask_next;
1102 *lookup_hit_mask = pkts_mask_out;
1104 } /* rte_table_hash_lookup_key32_ext() */
1106 struct rte_table_ops rte_table_hash_key32_lru_ops = {
1107 .f_create = rte_table_hash_create_key32_lru,
1108 .f_free = rte_table_hash_free_key32_lru,
1109 .f_add = rte_table_hash_entry_add_key32_lru,
1110 .f_delete = rte_table_hash_entry_delete_key32_lru,
1111 .f_lookup = rte_table_hash_lookup_key32_lru,
1114 struct rte_table_ops rte_table_hash_key32_ext_ops = {
1115 .f_create = rte_table_hash_create_key32_ext,
1116 .f_free = rte_table_hash_free_key32_ext,
1117 .f_add = rte_table_hash_entry_add_key32_ext,
1118 .f_delete = rte_table_hash_entry_delete_key32_ext,
1119 .f_lookup = rte_table_hash_lookup_key32_ext,