1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
3 * Copyright(c) 2020 Arm Limited
11 #include <sys/queue.h>
14 #include <rte_branch_prediction.h>
15 #include <rte_common.h>
16 #include <rte_memory.h> /* for definition of RTE_CACHE_LINE_SIZE */
17 #include <rte_malloc.h>
19 #include <rte_eal_memconfig.h>
20 #include <rte_per_lcore.h>
21 #include <rte_string_fns.h>
22 #include <rte_errno.h>
23 #include <rte_rwlock.h>
24 #include <rte_spinlock.h>
25 #include <rte_tailq.h>
29 TAILQ_HEAD(rte_lpm_list, rte_tailq_entry);
31 static struct rte_tailq_elem rte_lpm_tailq = {
34 EAL_REGISTER_TAILQ(rte_lpm_tailq)
36 #define MAX_DEPTH_TBL24 24
43 /** @internal LPM structure. */
49 struct rte_rcu_qsbr *v; /* RCU QSBR variable. */
50 enum rte_lpm_qsbr_mode rcu_mode;/* Blocking, defer queue. */
51 struct rte_rcu_qsbr_dq *dq; /* RCU QSBR defer queue. */
54 /* Macro to enable/disable run-time checks. */
55 #if defined(RTE_LIBRTE_LPM_DEBUG)
56 #include <rte_debug.h>
57 #define VERIFY_DEPTH(depth) do { \
58 if ((depth == 0) || (depth > RTE_LPM_MAX_DEPTH)) \
59 rte_panic("LPM: Invalid depth (%u) at line %d", \
60 (unsigned)(depth), __LINE__); \
63 #define VERIFY_DEPTH(depth)
67 * Converts a given depth value to its corresponding mask value.
69 * depth (IN) : range = 1 - 32
70 * mask (OUT) : 32bit mask
72 static uint32_t __attribute__((pure))
73 depth_to_mask(uint8_t depth)
77 /* To calculate a mask start with a 1 on the left hand side and right
78 * shift while populating the left hand side with 1's
80 return (int)0x80000000 >> (depth - 1);
84 * Converts given depth value to its corresponding range value.
86 static uint32_t __attribute__((pure))
87 depth_to_range(uint8_t depth)
92 * Calculate tbl24 range. (Note: 2^depth = 1 << depth)
94 if (depth <= MAX_DEPTH_TBL24)
95 return 1 << (MAX_DEPTH_TBL24 - depth);
97 /* Else if depth is greater than 24 */
98 return 1 << (RTE_LPM_MAX_DEPTH - depth);
102 * Find an existing lpm table and return a pointer to it.
105 rte_lpm_find_existing(const char *name)
107 struct rte_lpm *l = NULL;
108 struct rte_tailq_entry *te;
109 struct rte_lpm_list *lpm_list;
111 lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
113 rte_mcfg_tailq_read_lock();
114 TAILQ_FOREACH(te, lpm_list, next) {
116 if (strncmp(name, l->name, RTE_LPM_NAMESIZE) == 0)
119 rte_mcfg_tailq_read_unlock();
130 * Allocates memory for LPM object
133 rte_lpm_create(const char *name, int socket_id,
134 const struct rte_lpm_config *config)
136 char mem_name[RTE_LPM_NAMESIZE];
137 struct __rte_lpm *internal_lpm;
138 struct rte_lpm *lpm = NULL;
139 struct rte_tailq_entry *te;
140 uint32_t mem_size, rules_size, tbl8s_size;
141 struct rte_lpm_list *lpm_list;
143 lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
145 RTE_BUILD_BUG_ON(sizeof(struct rte_lpm_tbl_entry) != 4);
147 /* Check user arguments. */
148 if ((name == NULL) || (socket_id < -1) || (config->max_rules == 0)
149 || config->number_tbl8s > RTE_LPM_MAX_TBL8_NUM_GROUPS) {
154 snprintf(mem_name, sizeof(mem_name), "LPM_%s", name);
156 rte_mcfg_tailq_write_lock();
158 /* guarantee there's no existing */
159 TAILQ_FOREACH(te, lpm_list, next) {
161 if (strncmp(name, lpm->name, RTE_LPM_NAMESIZE) == 0)
171 /* Determine the amount of memory to allocate. */
172 mem_size = sizeof(*internal_lpm);
173 rules_size = sizeof(struct rte_lpm_rule) * config->max_rules;
174 tbl8s_size = sizeof(struct rte_lpm_tbl_entry) *
175 RTE_LPM_TBL8_GROUP_NUM_ENTRIES * config->number_tbl8s;
177 /* allocate tailq entry */
178 te = rte_zmalloc("LPM_TAILQ_ENTRY", sizeof(*te), 0);
180 RTE_LOG(ERR, LPM, "Failed to allocate tailq entry\n");
185 /* Allocate memory to store the LPM data structures. */
186 internal_lpm = rte_zmalloc_socket(mem_name, mem_size,
187 RTE_CACHE_LINE_SIZE, socket_id);
188 if (internal_lpm == NULL) {
189 RTE_LOG(ERR, LPM, "LPM memory allocation failed\n");
195 lpm = &internal_lpm->lpm;
196 lpm->rules_tbl = rte_zmalloc_socket(NULL,
197 (size_t)rules_size, RTE_CACHE_LINE_SIZE, socket_id);
199 if (lpm->rules_tbl == NULL) {
200 RTE_LOG(ERR, LPM, "LPM rules_tbl memory allocation failed\n");
201 rte_free(internal_lpm);
209 lpm->tbl8 = rte_zmalloc_socket(NULL,
210 (size_t)tbl8s_size, RTE_CACHE_LINE_SIZE, socket_id);
212 if (lpm->tbl8 == NULL) {
213 RTE_LOG(ERR, LPM, "LPM tbl8 memory allocation failed\n");
214 rte_free(lpm->rules_tbl);
215 rte_free(internal_lpm);
223 /* Save user arguments. */
224 lpm->max_rules = config->max_rules;
225 lpm->number_tbl8s = config->number_tbl8s;
226 strlcpy(lpm->name, name, sizeof(lpm->name));
230 TAILQ_INSERT_TAIL(lpm_list, te, next);
233 rte_mcfg_tailq_write_unlock();
239 * Deallocates memory for given LPM table.
242 rte_lpm_free(struct rte_lpm *lpm)
244 struct __rte_lpm *internal_lpm;
245 struct rte_lpm_list *lpm_list;
246 struct rte_tailq_entry *te;
248 /* Check user arguments. */
252 lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
254 rte_mcfg_tailq_write_lock();
256 /* find our tailq entry */
257 TAILQ_FOREACH(te, lpm_list, next) {
258 if (te->data == (void *) lpm)
262 TAILQ_REMOVE(lpm_list, te, next);
264 rte_mcfg_tailq_write_unlock();
266 internal_lpm = container_of(lpm, struct __rte_lpm, lpm);
267 if (internal_lpm->dq != NULL)
268 rte_rcu_qsbr_dq_delete(internal_lpm->dq);
270 rte_free(lpm->rules_tbl);
276 __lpm_rcu_qsbr_free_resource(void *p, void *data, unsigned int n)
278 struct rte_lpm_tbl_entry *tbl8 = ((struct rte_lpm *)p)->tbl8;
279 struct rte_lpm_tbl_entry zero_tbl8_entry = {0};
280 uint32_t tbl8_group_index = *(uint32_t *)data;
283 /* Set tbl8 group invalid */
284 __atomic_store(&tbl8[tbl8_group_index], &zero_tbl8_entry,
288 /* Associate QSBR variable with an LPM object.
291 rte_lpm_rcu_qsbr_add(struct rte_lpm *lpm, struct rte_lpm_rcu_config *cfg,
292 struct rte_rcu_qsbr_dq **dq)
294 struct rte_rcu_qsbr_dq_parameters params = {0};
295 char rcu_dq_name[RTE_RCU_QSBR_DQ_NAMESIZE];
296 struct __rte_lpm *internal_lpm;
298 if (lpm == NULL || cfg == NULL) {
303 internal_lpm = container_of(lpm, struct __rte_lpm, lpm);
304 if (internal_lpm->v != NULL) {
309 if (cfg->mode == RTE_LPM_QSBR_MODE_SYNC) {
310 /* No other things to do. */
311 } else if (cfg->mode == RTE_LPM_QSBR_MODE_DQ) {
312 /* Init QSBR defer queue. */
313 snprintf(rcu_dq_name, sizeof(rcu_dq_name),
314 "LPM_RCU_%s", lpm->name);
315 params.name = rcu_dq_name;
316 params.size = cfg->dq_size;
317 if (params.size == 0)
318 params.size = lpm->number_tbl8s;
319 params.trigger_reclaim_limit = cfg->reclaim_thd;
320 params.max_reclaim_size = cfg->reclaim_max;
321 if (params.max_reclaim_size == 0)
322 params.max_reclaim_size = RTE_LPM_RCU_DQ_RECLAIM_MAX;
323 params.esize = sizeof(uint32_t); /* tbl8 group index */
324 params.free_fn = __lpm_rcu_qsbr_free_resource;
327 internal_lpm->dq = rte_rcu_qsbr_dq_create(¶ms);
328 if (internal_lpm->dq == NULL) {
329 RTE_LOG(ERR, LPM, "LPM defer queue creation failed\n");
333 *dq = internal_lpm->dq;
338 internal_lpm->rcu_mode = cfg->mode;
339 internal_lpm->v = cfg->v;
345 * Adds a rule to the rule table.
347 * NOTE: The rule table is split into 32 groups. Each group contains rules that
348 * apply to a specific prefix depth (i.e. group 1 contains rules that apply to
349 * prefixes with a depth of 1 etc.). In the following code (depth - 1) is used
350 * to refer to depth 1 because even though the depth range is 1 - 32, depths
351 * are stored in the rule table from 0 - 31.
352 * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
355 rule_add(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
358 uint32_t rule_gindex, rule_index, last_rule;
363 /* Scan through rule group to see if rule already exists. */
364 if (lpm->rule_info[depth - 1].used_rules > 0) {
366 /* rule_gindex stands for rule group index. */
367 rule_gindex = lpm->rule_info[depth - 1].first_rule;
368 /* Initialise rule_index to point to start of rule group. */
369 rule_index = rule_gindex;
370 /* Last rule = Last used rule in this rule group. */
371 last_rule = rule_gindex + lpm->rule_info[depth - 1].used_rules;
373 for (; rule_index < last_rule; rule_index++) {
375 /* If rule already exists update next hop and return. */
376 if (lpm->rules_tbl[rule_index].ip == ip_masked) {
378 if (lpm->rules_tbl[rule_index].next_hop
381 lpm->rules_tbl[rule_index].next_hop = next_hop;
387 if (rule_index == lpm->max_rules)
390 /* Calculate the position in which the rule will be stored. */
393 for (i = depth - 1; i > 0; i--) {
394 if (lpm->rule_info[i - 1].used_rules > 0) {
395 rule_index = lpm->rule_info[i - 1].first_rule
396 + lpm->rule_info[i - 1].used_rules;
400 if (rule_index == lpm->max_rules)
403 lpm->rule_info[depth - 1].first_rule = rule_index;
406 /* Make room for the new rule in the array. */
407 for (i = RTE_LPM_MAX_DEPTH; i > depth; i--) {
408 if (lpm->rule_info[i - 1].first_rule
409 + lpm->rule_info[i - 1].used_rules == lpm->max_rules)
412 if (lpm->rule_info[i - 1].used_rules > 0) {
413 lpm->rules_tbl[lpm->rule_info[i - 1].first_rule
414 + lpm->rule_info[i - 1].used_rules]
415 = lpm->rules_tbl[lpm->rule_info[i - 1].first_rule];
416 lpm->rule_info[i - 1].first_rule++;
420 /* Add the new rule. */
421 lpm->rules_tbl[rule_index].ip = ip_masked;
422 lpm->rules_tbl[rule_index].next_hop = next_hop;
424 /* Increment the used rules counter for this rule group. */
425 lpm->rule_info[depth - 1].used_rules++;
431 * Delete a rule from the rule table.
432 * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
435 rule_delete(struct rte_lpm *lpm, int32_t rule_index, uint8_t depth)
441 lpm->rules_tbl[rule_index] =
442 lpm->rules_tbl[lpm->rule_info[depth - 1].first_rule
443 + lpm->rule_info[depth - 1].used_rules - 1];
445 for (i = depth; i < RTE_LPM_MAX_DEPTH; i++) {
446 if (lpm->rule_info[i].used_rules > 0) {
447 lpm->rules_tbl[lpm->rule_info[i].first_rule - 1] =
448 lpm->rules_tbl[lpm->rule_info[i].first_rule
449 + lpm->rule_info[i].used_rules - 1];
450 lpm->rule_info[i].first_rule--;
454 lpm->rule_info[depth - 1].used_rules--;
458 * Finds a rule in rule table.
459 * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
462 rule_find(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth)
464 uint32_t rule_gindex, last_rule, rule_index;
468 rule_gindex = lpm->rule_info[depth - 1].first_rule;
469 last_rule = rule_gindex + lpm->rule_info[depth - 1].used_rules;
471 /* Scan used rules at given depth to find rule. */
472 for (rule_index = rule_gindex; rule_index < last_rule; rule_index++) {
473 /* If rule is found return the rule index. */
474 if (lpm->rules_tbl[rule_index].ip == ip_masked)
478 /* If rule is not found return -EINVAL. */
483 * Find, clean and allocate a tbl8.
486 _tbl8_alloc(struct rte_lpm *lpm)
488 uint32_t group_idx; /* tbl8 group index. */
489 struct rte_lpm_tbl_entry *tbl8_entry;
491 /* Scan through tbl8 to find a free (i.e. INVALID) tbl8 group. */
492 for (group_idx = 0; group_idx < lpm->number_tbl8s; group_idx++) {
493 tbl8_entry = &lpm->tbl8[group_idx *
494 RTE_LPM_TBL8_GROUP_NUM_ENTRIES];
495 /* If a free tbl8 group is found clean it and set as VALID. */
496 if (!tbl8_entry->valid_group) {
497 struct rte_lpm_tbl_entry new_tbl8_entry = {
501 .valid_group = VALID,
504 memset(&tbl8_entry[0], 0,
505 RTE_LPM_TBL8_GROUP_NUM_ENTRIES *
506 sizeof(tbl8_entry[0]));
508 __atomic_store(tbl8_entry, &new_tbl8_entry,
511 /* Return group index for allocated tbl8 group. */
516 /* If there are no tbl8 groups free then return error. */
521 tbl8_alloc(struct rte_lpm *lpm)
523 int32_t group_idx; /* tbl8 group index. */
524 struct __rte_lpm *internal_lpm;
526 internal_lpm = container_of(lpm, struct __rte_lpm, lpm);
527 group_idx = _tbl8_alloc(lpm);
528 if (group_idx == -ENOSPC && internal_lpm->dq != NULL) {
529 /* If there are no tbl8 groups try to reclaim one. */
530 if (rte_rcu_qsbr_dq_reclaim(internal_lpm->dq, 1,
531 NULL, NULL, NULL) == 0)
532 group_idx = _tbl8_alloc(lpm);
539 tbl8_free(struct rte_lpm *lpm, uint32_t tbl8_group_start)
541 struct rte_lpm_tbl_entry zero_tbl8_entry = {0};
542 struct __rte_lpm *internal_lpm;
544 internal_lpm = container_of(lpm, struct __rte_lpm, lpm);
545 if (internal_lpm->v == NULL) {
546 /* Set tbl8 group invalid*/
547 __atomic_store(&lpm->tbl8[tbl8_group_start], &zero_tbl8_entry,
549 } else if (internal_lpm->rcu_mode == RTE_LPM_QSBR_MODE_SYNC) {
550 /* Wait for quiescent state change. */
551 rte_rcu_qsbr_synchronize(internal_lpm->v,
552 RTE_QSBR_THRID_INVALID);
553 /* Set tbl8 group invalid*/
554 __atomic_store(&lpm->tbl8[tbl8_group_start], &zero_tbl8_entry,
556 } else if (internal_lpm->rcu_mode == RTE_LPM_QSBR_MODE_DQ) {
557 /* Push into QSBR defer queue. */
558 rte_rcu_qsbr_dq_enqueue(internal_lpm->dq,
559 (void *)&tbl8_group_start);
563 static __rte_noinline int32_t
564 add_depth_small(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
567 #define group_idx next_hop
568 uint32_t tbl24_index, tbl24_range, tbl8_index, tbl8_group_end, i, j;
570 /* Calculate the index into Table24. */
571 tbl24_index = ip >> 8;
572 tbl24_range = depth_to_range(depth);
574 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
576 * For invalid OR valid and non-extended tbl 24 entries set
579 if (!lpm->tbl24[i].valid || (lpm->tbl24[i].valid_group == 0 &&
580 lpm->tbl24[i].depth <= depth)) {
582 struct rte_lpm_tbl_entry new_tbl24_entry = {
583 .next_hop = next_hop,
589 /* Setting tbl24 entry in one go to avoid race
592 __atomic_store(&lpm->tbl24[i], &new_tbl24_entry,
598 if (lpm->tbl24[i].valid_group == 1) {
599 /* If tbl24 entry is valid and extended calculate the
602 tbl8_index = lpm->tbl24[i].group_idx *
603 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
604 tbl8_group_end = tbl8_index +
605 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
607 for (j = tbl8_index; j < tbl8_group_end; j++) {
608 if (!lpm->tbl8[j].valid ||
609 lpm->tbl8[j].depth <= depth) {
610 struct rte_lpm_tbl_entry
613 .valid_group = VALID,
615 .next_hop = next_hop,
619 * Setting tbl8 entry in one go to avoid
622 __atomic_store(&lpm->tbl8[j],
635 static __rte_noinline int32_t
636 add_depth_big(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
639 #define group_idx next_hop
640 uint32_t tbl24_index;
641 int32_t tbl8_group_index, tbl8_group_start, tbl8_group_end, tbl8_index,
644 tbl24_index = (ip_masked >> 8);
645 tbl8_range = depth_to_range(depth);
647 if (!lpm->tbl24[tbl24_index].valid) {
648 /* Search for a free tbl8 group. */
649 tbl8_group_index = tbl8_alloc(lpm);
651 /* Check tbl8 allocation was successful. */
652 if (tbl8_group_index < 0) {
653 return tbl8_group_index;
656 /* Find index into tbl8 and range. */
657 tbl8_index = (tbl8_group_index *
658 RTE_LPM_TBL8_GROUP_NUM_ENTRIES) +
661 /* Set tbl8 entry. */
662 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
663 struct rte_lpm_tbl_entry new_tbl8_entry = {
666 .valid_group = lpm->tbl8[i].valid_group,
667 .next_hop = next_hop,
669 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
674 * Update tbl24 entry to point to new tbl8 entry. Note: The
675 * ext_flag and tbl8_index need to be updated simultaneously,
676 * so assign whole structure in one go
679 struct rte_lpm_tbl_entry new_tbl24_entry = {
680 .group_idx = tbl8_group_index,
686 /* The tbl24 entry must be written only after the
687 * tbl8 entries are written.
689 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
692 } /* If valid entry but not extended calculate the index into Table8. */
693 else if (lpm->tbl24[tbl24_index].valid_group == 0) {
694 /* Search for free tbl8 group. */
695 tbl8_group_index = tbl8_alloc(lpm);
697 if (tbl8_group_index < 0) {
698 return tbl8_group_index;
701 tbl8_group_start = tbl8_group_index *
702 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
703 tbl8_group_end = tbl8_group_start +
704 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
706 /* Populate new tbl8 with tbl24 value. */
707 for (i = tbl8_group_start; i < tbl8_group_end; i++) {
708 struct rte_lpm_tbl_entry new_tbl8_entry = {
710 .depth = lpm->tbl24[tbl24_index].depth,
711 .valid_group = lpm->tbl8[i].valid_group,
712 .next_hop = lpm->tbl24[tbl24_index].next_hop,
714 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
718 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
720 /* Insert new rule into the tbl8 entry. */
721 for (i = tbl8_index; i < tbl8_index + tbl8_range; i++) {
722 struct rte_lpm_tbl_entry new_tbl8_entry = {
725 .valid_group = lpm->tbl8[i].valid_group,
726 .next_hop = next_hop,
728 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
733 * Update tbl24 entry to point to new tbl8 entry. Note: The
734 * ext_flag and tbl8_index need to be updated simultaneously,
735 * so assign whole structure in one go.
738 struct rte_lpm_tbl_entry new_tbl24_entry = {
739 .group_idx = tbl8_group_index,
745 /* The tbl24 entry must be written only after the
746 * tbl8 entries are written.
748 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
752 * If it is valid, extended entry calculate the index into tbl8.
754 tbl8_group_index = lpm->tbl24[tbl24_index].group_idx;
755 tbl8_group_start = tbl8_group_index *
756 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
757 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
759 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
761 if (!lpm->tbl8[i].valid ||
762 lpm->tbl8[i].depth <= depth) {
763 struct rte_lpm_tbl_entry new_tbl8_entry = {
766 .next_hop = next_hop,
767 .valid_group = lpm->tbl8[i].valid_group,
771 * Setting tbl8 entry in one go to avoid race
774 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
789 rte_lpm_add(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
792 int32_t rule_index, status = 0;
795 /* Check user arguments. */
796 if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM_MAX_DEPTH))
799 ip_masked = ip & depth_to_mask(depth);
801 /* Add the rule to the rule table. */
802 rule_index = rule_add(lpm, ip_masked, depth, next_hop);
804 /* Skip table entries update if The rule is the same as
805 * the rule in the rules table.
807 if (rule_index == -EEXIST)
810 /* If the is no space available for new rule return error. */
811 if (rule_index < 0) {
815 if (depth <= MAX_DEPTH_TBL24) {
816 status = add_depth_small(lpm, ip_masked, depth, next_hop);
817 } else { /* If depth > RTE_LPM_MAX_DEPTH_TBL24 */
818 status = add_depth_big(lpm, ip_masked, depth, next_hop);
821 * If add fails due to exhaustion of tbl8 extensions delete
822 * rule that was added to rule table.
825 rule_delete(lpm, rule_index, depth);
835 * Look for a rule in the high-level rules table
838 rte_lpm_is_rule_present(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
844 /* Check user arguments. */
846 (next_hop == NULL) ||
847 (depth < 1) || (depth > RTE_LPM_MAX_DEPTH))
850 /* Look for the rule using rule_find. */
851 ip_masked = ip & depth_to_mask(depth);
852 rule_index = rule_find(lpm, ip_masked, depth);
854 if (rule_index >= 0) {
855 *next_hop = lpm->rules_tbl[rule_index].next_hop;
859 /* If rule is not found return 0. */
864 find_previous_rule(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
865 uint8_t *sub_rule_depth)
871 for (prev_depth = (uint8_t)(depth - 1); prev_depth > 0; prev_depth--) {
872 ip_masked = ip & depth_to_mask(prev_depth);
874 rule_index = rule_find(lpm, ip_masked, prev_depth);
876 if (rule_index >= 0) {
877 *sub_rule_depth = prev_depth;
886 delete_depth_small(struct rte_lpm *lpm, uint32_t ip_masked,
887 uint8_t depth, int32_t sub_rule_index, uint8_t sub_rule_depth)
889 #define group_idx next_hop
890 uint32_t tbl24_range, tbl24_index, tbl8_group_index, tbl8_index, i, j;
892 /* Calculate the range and index into Table24. */
893 tbl24_range = depth_to_range(depth);
894 tbl24_index = (ip_masked >> 8);
895 struct rte_lpm_tbl_entry zero_tbl24_entry = {0};
898 * Firstly check the sub_rule_index. A -1 indicates no replacement rule
899 * and a positive number indicates a sub_rule_index.
901 if (sub_rule_index < 0) {
903 * If no replacement rule exists then invalidate entries
904 * associated with this rule.
906 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
908 if (lpm->tbl24[i].valid_group == 0 &&
909 lpm->tbl24[i].depth <= depth) {
910 __atomic_store(&lpm->tbl24[i],
911 &zero_tbl24_entry, __ATOMIC_RELEASE);
912 } else if (lpm->tbl24[i].valid_group == 1) {
914 * If TBL24 entry is extended, then there has
915 * to be a rule with depth >= 25 in the
916 * associated TBL8 group.
919 tbl8_group_index = lpm->tbl24[i].group_idx;
920 tbl8_index = tbl8_group_index *
921 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
923 for (j = tbl8_index; j < (tbl8_index +
924 RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
926 if (lpm->tbl8[j].depth <= depth)
927 lpm->tbl8[j].valid = INVALID;
933 * If a replacement rule exists then modify entries
934 * associated with this rule.
937 struct rte_lpm_tbl_entry new_tbl24_entry = {
938 .next_hop = lpm->rules_tbl[sub_rule_index].next_hop,
941 .depth = sub_rule_depth,
944 struct rte_lpm_tbl_entry new_tbl8_entry = {
946 .valid_group = VALID,
947 .depth = sub_rule_depth,
948 .next_hop = lpm->rules_tbl
949 [sub_rule_index].next_hop,
952 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
954 if (lpm->tbl24[i].valid_group == 0 &&
955 lpm->tbl24[i].depth <= depth) {
956 __atomic_store(&lpm->tbl24[i], &new_tbl24_entry,
958 } else if (lpm->tbl24[i].valid_group == 1) {
960 * If TBL24 entry is extended, then there has
961 * to be a rule with depth >= 25 in the
962 * associated TBL8 group.
965 tbl8_group_index = lpm->tbl24[i].group_idx;
966 tbl8_index = tbl8_group_index *
967 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
969 for (j = tbl8_index; j < (tbl8_index +
970 RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
972 if (lpm->tbl8[j].depth <= depth)
973 __atomic_store(&lpm->tbl8[j],
985 * Checks if table 8 group can be recycled.
987 * Return of -EEXIST means tbl8 is in use and thus can not be recycled.
988 * Return of -EINVAL means tbl8 is empty and thus can be recycled
989 * Return of value > -1 means tbl8 is in use but has all the same values and
990 * thus can be recycled
993 tbl8_recycle_check(struct rte_lpm_tbl_entry *tbl8,
994 uint32_t tbl8_group_start)
996 uint32_t tbl8_group_end, i;
997 tbl8_group_end = tbl8_group_start + RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
1000 * Check the first entry of the given tbl8. If it is invalid we know
1001 * this tbl8 does not contain any rule with a depth < RTE_LPM_MAX_DEPTH
1002 * (As they would affect all entries in a tbl8) and thus this table
1003 * can not be recycled.
1005 if (tbl8[tbl8_group_start].valid) {
1007 * If first entry is valid check if the depth is less than 24
1008 * and if so check the rest of the entries to verify that they
1009 * are all of this depth.
1011 if (tbl8[tbl8_group_start].depth <= MAX_DEPTH_TBL24) {
1012 for (i = (tbl8_group_start + 1); i < tbl8_group_end;
1015 if (tbl8[i].depth !=
1016 tbl8[tbl8_group_start].depth) {
1021 /* If all entries are the same return the tb8 index */
1022 return tbl8_group_start;
1028 * If the first entry is invalid check if the rest of the entries in
1029 * the tbl8 are invalid.
1031 for (i = (tbl8_group_start + 1); i < tbl8_group_end; i++) {
1035 /* If no valid entries are found then return -EINVAL. */
1040 delete_depth_big(struct rte_lpm *lpm, uint32_t ip_masked,
1041 uint8_t depth, int32_t sub_rule_index, uint8_t sub_rule_depth)
1043 #define group_idx next_hop
1044 uint32_t tbl24_index, tbl8_group_index, tbl8_group_start, tbl8_index,
1046 int32_t tbl8_recycle_index;
1049 * Calculate the index into tbl24 and range. Note: All depths larger
1050 * than MAX_DEPTH_TBL24 are associated with only one tbl24 entry.
1052 tbl24_index = ip_masked >> 8;
1054 /* Calculate the index into tbl8 and range. */
1055 tbl8_group_index = lpm->tbl24[tbl24_index].group_idx;
1056 tbl8_group_start = tbl8_group_index * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
1057 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
1058 tbl8_range = depth_to_range(depth);
1060 if (sub_rule_index < 0) {
1062 * Loop through the range of entries on tbl8 for which the
1063 * rule_to_delete must be removed or modified.
1065 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
1066 if (lpm->tbl8[i].depth <= depth)
1067 lpm->tbl8[i].valid = INVALID;
1070 /* Set new tbl8 entry. */
1071 struct rte_lpm_tbl_entry new_tbl8_entry = {
1073 .depth = sub_rule_depth,
1074 .valid_group = lpm->tbl8[tbl8_group_start].valid_group,
1075 .next_hop = lpm->rules_tbl[sub_rule_index].next_hop,
1079 * Loop through the range of entries on tbl8 for which the
1080 * rule_to_delete must be modified.
1082 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
1083 if (lpm->tbl8[i].depth <= depth)
1084 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
1090 * Check if there are any valid entries in this tbl8 group. If all
1091 * tbl8 entries are invalid we can free the tbl8 and invalidate the
1092 * associated tbl24 entry.
1095 tbl8_recycle_index = tbl8_recycle_check(lpm->tbl8, tbl8_group_start);
1097 if (tbl8_recycle_index == -EINVAL) {
1098 /* Set tbl24 before freeing tbl8 to avoid race condition.
1099 * Prevent the free of the tbl8 group from hoisting.
1101 lpm->tbl24[tbl24_index].valid = 0;
1102 __atomic_thread_fence(__ATOMIC_RELEASE);
1103 tbl8_free(lpm, tbl8_group_start);
1104 } else if (tbl8_recycle_index > -1) {
1105 /* Update tbl24 entry. */
1106 struct rte_lpm_tbl_entry new_tbl24_entry = {
1107 .next_hop = lpm->tbl8[tbl8_recycle_index].next_hop,
1110 .depth = lpm->tbl8[tbl8_recycle_index].depth,
1113 /* Set tbl24 before freeing tbl8 to avoid race condition.
1114 * Prevent the free of the tbl8 group from hoisting.
1116 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
1118 __atomic_thread_fence(__ATOMIC_RELEASE);
1119 tbl8_free(lpm, tbl8_group_start);
1129 rte_lpm_delete(struct rte_lpm *lpm, uint32_t ip, uint8_t depth)
1131 int32_t rule_to_delete_index, sub_rule_index;
1133 uint8_t sub_rule_depth;
1135 * Check input arguments. Note: IP must be a positive integer of 32
1136 * bits in length therefore it need not be checked.
1138 if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM_MAX_DEPTH)) {
1142 ip_masked = ip & depth_to_mask(depth);
1145 * Find the index of the input rule, that needs to be deleted, in the
1148 rule_to_delete_index = rule_find(lpm, ip_masked, depth);
1151 * Check if rule_to_delete_index was found. If no rule was found the
1152 * function rule_find returns -EINVAL.
1154 if (rule_to_delete_index < 0)
1157 /* Delete the rule from the rule table. */
1158 rule_delete(lpm, rule_to_delete_index, depth);
1161 * Find rule to replace the rule_to_delete. If there is no rule to
1162 * replace the rule_to_delete we return -1 and invalidate the table
1163 * entries associated with this rule.
1166 sub_rule_index = find_previous_rule(lpm, ip, depth, &sub_rule_depth);
1169 * If the input depth value is less than 25 use function
1170 * delete_depth_small otherwise use delete_depth_big.
1172 if (depth <= MAX_DEPTH_TBL24) {
1173 return delete_depth_small(lpm, ip_masked, depth,
1174 sub_rule_index, sub_rule_depth);
1175 } else { /* If depth > MAX_DEPTH_TBL24 */
1176 return delete_depth_big(lpm, ip_masked, depth, sub_rule_index,
1182 * Delete all rules from the LPM table.
1185 rte_lpm_delete_all(struct rte_lpm *lpm)
1187 /* Zero rule information. */
1188 memset(lpm->rule_info, 0, sizeof(lpm->rule_info));
1191 memset(lpm->tbl24, 0, sizeof(lpm->tbl24));
1194 memset(lpm->tbl8, 0, sizeof(lpm->tbl8[0])
1195 * RTE_LPM_TBL8_GROUP_NUM_ENTRIES * lpm->number_tbl8s);
1197 /* Delete all rules form the rules table. */
1198 memset(lpm->rules_tbl, 0, sizeof(lpm->rules_tbl[0]) * lpm->max_rules);