lib/librte_lpm/rte_lpm.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation
   3  */
   4
   5 #include <string.h>
   6 #include <stdint.h>
   7 #include <errno.h>
   8 #include <stdarg.h>
   9 #include <stdio.h>
  10 #include <sys/queue.h>
  11
  12 #include <rte_log.h>
  13 #include <rte_branch_prediction.h>
  14 #include <rte_common.h>
  15 #include <rte_memory.h>        /* for definition of RTE_CACHE_LINE_SIZE */
  16 #include <rte_malloc.h>
  17 #include <rte_eal.h>
  18 #include <rte_eal_memconfig.h>
  19 #include <rte_per_lcore.h>
  20 #include <rte_string_fns.h>
  21 #include <rte_errno.h>
  22 #include <rte_rwlock.h>
  23 #include <rte_spinlock.h>
  24 #include <rte_tailq.h>
  25
  26 #include "rte_lpm.h"
  27
  28 TAILQ_HEAD(rte_lpm_list, rte_tailq_entry);
  29
  30 static struct rte_tailq_elem rte_lpm_tailq = {
  31         .name = "RTE_LPM",
  32 };
  33 EAL_REGISTER_TAILQ(rte_lpm_tailq)
  34
  35 #define MAX_DEPTH_TBL24 24
  36
  37 enum valid_flag {
  38         INVALID = 0,
  39         VALID
  40 };
  41
  42 /* Macro to enable/disable run-time checks. */
  43 #if defined(RTE_LIBRTE_LPM_DEBUG)
  44 #include <rte_debug.h>
  45 #define VERIFY_DEPTH(depth) do {                                \
  46         if ((depth == 0) || (depth > RTE_LPM_MAX_DEPTH))        \
  47                 rte_panic("LPM: Invalid depth (%u) at line %d", \
  48                                 (unsigned)(depth), __LINE__);   \
  49 } while (0)
  50 #else
  51 #define VERIFY_DEPTH(depth)
  52 #endif
  53
  54 /*
  55  * Converts a given depth value to its corresponding mask value.
  56  *
  57  * depth  (IN)          : range = 1 - 32
  58  * mask   (OUT)         : 32bit mask
  59  */
  60 static uint32_t __attribute__((pure))
  61 depth_to_mask(uint8_t depth)
  62 {
  63         VERIFY_DEPTH(depth);
  64
  65         /* To calculate a mask start with a 1 on the left hand side and right
  66          * shift while populating the left hand side with 1's
  67          */
  68         return (int)0x80000000 >> (depth - 1);
  69 }
  70
  71 /*
  72  * Converts given depth value to its corresponding range value.
  73  */
  74 static uint32_t __attribute__((pure))
  75 depth_to_range(uint8_t depth)
  76 {
  77         VERIFY_DEPTH(depth);
  78
  79         /*
  80          * Calculate tbl24 range. (Note: 2^depth = 1 << depth)
  81          */
  82         if (depth <= MAX_DEPTH_TBL24)
  83                 return 1 << (MAX_DEPTH_TBL24 - depth);
  84
  85         /* Else if depth is greater than 24 */
  86         return 1 << (RTE_LPM_MAX_DEPTH - depth);
  87 }
  88
  89 /*
  90  * Find an existing lpm table and return a pointer to it.
  91  */
  92 struct rte_lpm *
  93 rte_lpm_find_existing(const char *name)
  94 {
  95         struct rte_lpm *l = NULL;
  96         struct rte_tailq_entry *te;
  97         struct rte_lpm_list *lpm_list;
  98
  99         lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
 100
 101         rte_mcfg_tailq_read_lock();
 102         TAILQ_FOREACH(te, lpm_list, next) {
 103                 l = te->data;
 104                 if (strncmp(name, l->name, RTE_LPM_NAMESIZE) == 0)
 105                         break;
 106         }
 107         rte_mcfg_tailq_read_unlock();
 108
 109         if (te == NULL) {
 110                 rte_errno = ENOENT;
 111                 return NULL;
 112         }
 113
 114         return l;
 115 }
 116
 117 /*
 118  * Allocates memory for LPM object
 119  */
 120 struct rte_lpm *
 121 rte_lpm_create(const char *name, int socket_id,
 122                 const struct rte_lpm_config *config)
 123 {
 124         char mem_name[RTE_LPM_NAMESIZE];
 125         struct rte_lpm *lpm = NULL;
 126         struct rte_tailq_entry *te;
 127         uint32_t mem_size, rules_size, tbl8s_size;
 128         struct rte_lpm_list *lpm_list;
 129
 130         lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
 131
 132         RTE_BUILD_BUG_ON(sizeof(struct rte_lpm_tbl_entry) != 4);
 133
 134         /* Check user arguments. */
 135         if ((name == NULL) || (socket_id < -1) || (config->max_rules == 0)
 136                         || config->number_tbl8s > RTE_LPM_MAX_TBL8_NUM_GROUPS) {
 137                 rte_errno = EINVAL;
 138                 return NULL;
 139         }
 140
 141         snprintf(mem_name, sizeof(mem_name), "LPM_%s", name);
 142
 143         /* Determine the amount of memory to allocate. */
 144         mem_size = sizeof(*lpm);
 145         rules_size = sizeof(struct rte_lpm_rule) * config->max_rules;
 146         tbl8s_size = (sizeof(struct rte_lpm_tbl_entry) *
 147                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES * config->number_tbl8s);
 148
 149         rte_mcfg_tailq_write_lock();
 150
 151         /* guarantee there's no existing */
 152         TAILQ_FOREACH(te, lpm_list, next) {
 153                 lpm = te->data;
 154                 if (strncmp(name, lpm->name, RTE_LPM_NAMESIZE) == 0)
 155                         break;
 156         }
 157
 158         if (te != NULL) {
 159                 lpm = NULL;
 160                 rte_errno = EEXIST;
 161                 goto exit;
 162         }
 163
 164         /* allocate tailq entry */
 165         te = rte_zmalloc("LPM_TAILQ_ENTRY", sizeof(*te), 0);
 166         if (te == NULL) {
 167                 RTE_LOG(ERR, LPM, "Failed to allocate tailq entry\n");
 168                 rte_errno = ENOMEM;
 169                 goto exit;
 170         }
 171
 172         /* Allocate memory to store the LPM data structures. */
 173         lpm = rte_zmalloc_socket(mem_name, mem_size,
 174                         RTE_CACHE_LINE_SIZE, socket_id);
 175         if (lpm == NULL) {
 176                 RTE_LOG(ERR, LPM, "LPM memory allocation failed\n");
 177                 rte_free(te);
 178                 rte_errno = ENOMEM;
 179                 goto exit;
 180         }
 181
 182         lpm->rules_tbl = rte_zmalloc_socket(NULL,
 183                         (size_t)rules_size, RTE_CACHE_LINE_SIZE, socket_id);
 184
 185         if (lpm->rules_tbl == NULL) {
 186                 RTE_LOG(ERR, LPM, "LPM rules_tbl memory allocation failed\n");
 187                 rte_free(lpm);
 188                 lpm = NULL;
 189                 rte_free(te);
 190                 rte_errno = ENOMEM;
 191                 goto exit;
 192         }
 193
 194         lpm->tbl8 = rte_zmalloc_socket(NULL,
 195                         (size_t)tbl8s_size, RTE_CACHE_LINE_SIZE, socket_id);
 196
 197         if (lpm->tbl8 == NULL) {
 198                 RTE_LOG(ERR, LPM, "LPM tbl8 memory allocation failed\n");
 199                 rte_free(lpm->rules_tbl);
 200                 rte_free(lpm);
 201                 lpm = NULL;
 202                 rte_free(te);
 203                 rte_errno = ENOMEM;
 204                 goto exit;
 205         }
 206
 207         /* Save user arguments. */
 208         lpm->max_rules = config->max_rules;
 209         lpm->number_tbl8s = config->number_tbl8s;
 210         strlcpy(lpm->name, name, sizeof(lpm->name));
 211
 212         te->data = lpm;
 213
 214         TAILQ_INSERT_TAIL(lpm_list, te, next);
 215
 216 exit:
 217         rte_mcfg_tailq_write_unlock();
 218
 219         return lpm;
 220 }
 221
 222 /*
 223  * Deallocates memory for given LPM table.
 224  */
 225 void
 226 rte_lpm_free(struct rte_lpm *lpm)
 227 {
 228         struct rte_lpm_list *lpm_list;
 229         struct rte_tailq_entry *te;
 230
 231         /* Check user arguments. */
 232         if (lpm == NULL)
 233                 return;
 234
 235         lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
 236
 237         rte_mcfg_tailq_write_lock();
 238
 239         /* find our tailq entry */
 240         TAILQ_FOREACH(te, lpm_list, next) {
 241                 if (te->data == (void *) lpm)
 242                         break;
 243         }
 244         if (te != NULL)
 245                 TAILQ_REMOVE(lpm_list, te, next);
 246
 247         rte_mcfg_tailq_write_unlock();
 248
 249         rte_free(lpm->tbl8);
 250         rte_free(lpm->rules_tbl);
 251         rte_free(lpm);
 252         rte_free(te);
 253 }
 254
 255 /*
 256  * Adds a rule to the rule table.
 257  *
 258  * NOTE: The rule table is split into 32 groups. Each group contains rules that
 259  * apply to a specific prefix depth (i.e. group 1 contains rules that apply to
 260  * prefixes with a depth of 1 etc.). In the following code (depth - 1) is used
 261  * to refer to depth 1 because even though the depth range is 1 - 32, depths
 262  * are stored in the rule table from 0 - 31.
 263  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
 264  */
 265 static int32_t
 266 rule_add(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
 267         uint32_t next_hop)
 268 {
 269         uint32_t rule_gindex, rule_index, last_rule;
 270         int i;
 271
 272         VERIFY_DEPTH(depth);
 273
 274         /* Scan through rule group to see if rule already exists. */
 275         if (lpm->rule_info[depth - 1].used_rules > 0) {
 276
 277                 /* rule_gindex stands for rule group index. */
 278                 rule_gindex = lpm->rule_info[depth - 1].first_rule;
 279                 /* Initialise rule_index to point to start of rule group. */
 280                 rule_index = rule_gindex;
 281                 /* Last rule = Last used rule in this rule group. */
 282                 last_rule = rule_gindex + lpm->rule_info[depth - 1].used_rules;
 283
 284                 for (; rule_index < last_rule; rule_index++) {
 285
 286                         /* If rule already exists update next hop and return. */
 287                         if (lpm->rules_tbl[rule_index].ip == ip_masked) {
 288
 289                                 if (lpm->rules_tbl[rule_index].next_hop
 290                                                 == next_hop)
 291                                         return -EEXIST;
 292                                 lpm->rules_tbl[rule_index].next_hop = next_hop;
 293
 294                                 return rule_index;
 295                         }
 296                 }
 297
 298                 if (rule_index == lpm->max_rules)
 299                         return -ENOSPC;
 300         } else {
 301                 /* Calculate the position in which the rule will be stored. */
 302                 rule_index = 0;
 303
 304                 for (i = depth - 1; i > 0; i--) {
 305                         if (lpm->rule_info[i - 1].used_rules > 0) {
 306                                 rule_index = lpm->rule_info[i - 1].first_rule
 307                                                 + lpm->rule_info[i - 1].used_rules;
 308                                 break;
 309                         }
 310                 }
 311                 if (rule_index == lpm->max_rules)
 312                         return -ENOSPC;
 313
 314                 lpm->rule_info[depth - 1].first_rule = rule_index;
 315         }
 316
 317         /* Make room for the new rule in the array. */
 318         for (i = RTE_LPM_MAX_DEPTH; i > depth; i--) {
 319                 if (lpm->rule_info[i - 1].first_rule
 320                                 + lpm->rule_info[i - 1].used_rules == lpm->max_rules)
 321                         return -ENOSPC;
 322
 323                 if (lpm->rule_info[i - 1].used_rules > 0) {
 324                         lpm->rules_tbl[lpm->rule_info[i - 1].first_rule
 325                                 + lpm->rule_info[i - 1].used_rules]
 326                                         = lpm->rules_tbl[lpm->rule_info[i - 1].first_rule];
 327                         lpm->rule_info[i - 1].first_rule++;
 328                 }
 329         }
 330
 331         /* Add the new rule. */
 332         lpm->rules_tbl[rule_index].ip = ip_masked;
 333         lpm->rules_tbl[rule_index].next_hop = next_hop;
 334
 335         /* Increment the used rules counter for this rule group. */
 336         lpm->rule_info[depth - 1].used_rules++;
 337
 338         return rule_index;
 339 }
 340
 341 /*
 342  * Delete a rule from the rule table.
 343  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
 344  */
 345 static void
 346 rule_delete(struct rte_lpm *lpm, int32_t rule_index, uint8_t depth)
 347 {
 348         int i;
 349
 350         VERIFY_DEPTH(depth);
 351
 352         lpm->rules_tbl[rule_index] =
 353                         lpm->rules_tbl[lpm->rule_info[depth - 1].first_rule
 354                         + lpm->rule_info[depth - 1].used_rules - 1];
 355
 356         for (i = depth; i < RTE_LPM_MAX_DEPTH; i++) {
 357                 if (lpm->rule_info[i].used_rules > 0) {
 358                         lpm->rules_tbl[lpm->rule_info[i].first_rule - 1] =
 359                                         lpm->rules_tbl[lpm->rule_info[i].first_rule
 360                                                 + lpm->rule_info[i].used_rules - 1];
 361                         lpm->rule_info[i].first_rule--;
 362                 }
 363         }
 364
 365         lpm->rule_info[depth - 1].used_rules--;
 366 }
 367
 368 /*
 369  * Finds a rule in rule table.
 370  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
 371  */
 372 static int32_t
 373 rule_find(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth)
 374 {
 375         uint32_t rule_gindex, last_rule, rule_index;
 376
 377         VERIFY_DEPTH(depth);
 378
 379         rule_gindex = lpm->rule_info[depth - 1].first_rule;
 380         last_rule = rule_gindex + lpm->rule_info[depth - 1].used_rules;
 381
 382         /* Scan used rules at given depth to find rule. */
 383         for (rule_index = rule_gindex; rule_index < last_rule; rule_index++) {
 384                 /* If rule is found return the rule index. */
 385                 if (lpm->rules_tbl[rule_index].ip == ip_masked)
 386                         return rule_index;
 387         }
 388
 389         /* If rule is not found return -EINVAL. */
 390         return -EINVAL;
 391 }
 392
 393 /*
 394  * Find, clean and allocate a tbl8.
 395  */
 396 static int32_t
 397 tbl8_alloc(struct rte_lpm_tbl_entry *tbl8, uint32_t number_tbl8s)
 398 {
 399         uint32_t group_idx; /* tbl8 group index. */
 400         struct rte_lpm_tbl_entry *tbl8_entry;
 401
 402         /* Scan through tbl8 to find a free (i.e. INVALID) tbl8 group. */
 403         for (group_idx = 0; group_idx < number_tbl8s; group_idx++) {
 404                 tbl8_entry = &tbl8[group_idx * RTE_LPM_TBL8_GROUP_NUM_ENTRIES];
 405                 /* If a free tbl8 group is found clean it and set as VALID. */
 406                 if (!tbl8_entry->valid_group) {
 407                         struct rte_lpm_tbl_entry new_tbl8_entry = {
 408                                 .next_hop = 0,
 409                                 .valid = INVALID,
 410                                 .depth = 0,
 411                                 .valid_group = VALID,
 412                         };
 413
 414                         memset(&tbl8_entry[0], 0,
 415                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES *
 416                                         sizeof(tbl8_entry[0]));
 417
 418                         __atomic_store(tbl8_entry, &new_tbl8_entry,
 419                                         __ATOMIC_RELAXED);
 420
 421                         /* Return group index for allocated tbl8 group. */
 422                         return group_idx;
 423                 }
 424         }
 425
 426         /* If there are no tbl8 groups free then return error. */
 427         return -ENOSPC;
 428 }
 429
 430 static void
 431 tbl8_free(struct rte_lpm_tbl_entry *tbl8, uint32_t tbl8_group_start)
 432 {
 433         /* Set tbl8 group invalid*/
 434         struct rte_lpm_tbl_entry zero_tbl8_entry = {0};
 435
 436         __atomic_store(&tbl8[tbl8_group_start], &zero_tbl8_entry,
 437                         __ATOMIC_RELAXED);
 438 }
 439
 440 static __rte_noinline int32_t
 441 add_depth_small(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 442                 uint32_t next_hop)
 443 {
 444 #define group_idx next_hop
 445         uint32_t tbl24_index, tbl24_range, tbl8_index, tbl8_group_end, i, j;
 446
 447         /* Calculate the index into Table24. */
 448         tbl24_index = ip >> 8;
 449         tbl24_range = depth_to_range(depth);
 450
 451         for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
 452                 /*
 453                  * For invalid OR valid and non-extended tbl 24 entries set
 454                  * entry.
 455                  */
 456                 if (!lpm->tbl24[i].valid || (lpm->tbl24[i].valid_group == 0 &&
 457                                 lpm->tbl24[i].depth <= depth)) {
 458
 459                         struct rte_lpm_tbl_entry new_tbl24_entry = {
 460                                 .next_hop = next_hop,
 461                                 .valid = VALID,
 462                                 .valid_group = 0,
 463                                 .depth = depth,
 464                         };
 465
 466                         /* Setting tbl24 entry in one go to avoid race
 467                          * conditions
 468                          */
 469                         __atomic_store(&lpm->tbl24[i], &new_tbl24_entry,
 470                                         __ATOMIC_RELEASE);
 471
 472                         continue;
 473                 }
 474
 475                 if (lpm->tbl24[i].valid_group == 1) {
 476                         /* If tbl24 entry is valid and extended calculate the
 477                          *  index into tbl8.
 478                          */
 479                         tbl8_index = lpm->tbl24[i].group_idx *
 480                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 481                         tbl8_group_end = tbl8_index +
 482                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 483
 484                         for (j = tbl8_index; j < tbl8_group_end; j++) {
 485                                 if (!lpm->tbl8[j].valid ||
 486                                                 lpm->tbl8[j].depth <= depth) {
 487                                         struct rte_lpm_tbl_entry
 488                                                 new_tbl8_entry = {
 489                                                 .valid = VALID,
 490                                                 .valid_group = VALID,
 491                                                 .depth = depth,
 492                                                 .next_hop = next_hop,
 493                                         };
 494
 495                                         /*
 496                                          * Setting tbl8 entry in one go to avoid
 497                                          * race conditions
 498                                          */
 499                                         __atomic_store(&lpm->tbl8[j],
 500                                                 &new_tbl8_entry,
 501                                                 __ATOMIC_RELAXED);
 502
 503                                         continue;
 504                                 }
 505                         }
 506                 }
 507         }
 508 #undef group_idx
 509         return 0;
 510 }
 511
 512 static __rte_noinline int32_t
 513 add_depth_big(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
 514                 uint32_t next_hop)
 515 {
 516 #define group_idx next_hop
 517         uint32_t tbl24_index;
 518         int32_t tbl8_group_index, tbl8_group_start, tbl8_group_end, tbl8_index,
 519                 tbl8_range, i;
 520
 521         tbl24_index = (ip_masked >> 8);
 522         tbl8_range = depth_to_range(depth);
 523
 524         if (!lpm->tbl24[tbl24_index].valid) {
 525                 /* Search for a free tbl8 group. */
 526                 tbl8_group_index = tbl8_alloc(lpm->tbl8, lpm->number_tbl8s);
 527
 528                 /* Check tbl8 allocation was successful. */
 529                 if (tbl8_group_index < 0) {
 530                         return tbl8_group_index;
 531                 }
 532
 533                 /* Find index into tbl8 and range. */
 534                 tbl8_index = (tbl8_group_index *
 535                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES) +
 536                                 (ip_masked & 0xFF);
 537
 538                 /* Set tbl8 entry. */
 539                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 540                         struct rte_lpm_tbl_entry new_tbl8_entry = {
 541                                 .valid = VALID,
 542                                 .depth = depth,
 543                                 .valid_group = lpm->tbl8[i].valid_group,
 544                                 .next_hop = next_hop,
 545                         };
 546                         __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 547                                         __ATOMIC_RELAXED);
 548                 }
 549
 550                 /*
 551                  * Update tbl24 entry to point to new tbl8 entry. Note: The
 552                  * ext_flag and tbl8_index need to be updated simultaneously,
 553                  * so assign whole structure in one go
 554                  */
 555
 556                 struct rte_lpm_tbl_entry new_tbl24_entry = {
 557                         .group_idx = tbl8_group_index,
 558                         .valid = VALID,
 559                         .valid_group = 1,
 560                         .depth = 0,
 561                 };
 562
 563                 /* The tbl24 entry must be written only after the
 564                  * tbl8 entries are written.
 565                  */
 566                 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
 567                                 __ATOMIC_RELEASE);
 568
 569         } /* If valid entry but not extended calculate the index into Table8. */
 570         else if (lpm->tbl24[tbl24_index].valid_group == 0) {
 571                 /* Search for free tbl8 group. */
 572                 tbl8_group_index = tbl8_alloc(lpm->tbl8, lpm->number_tbl8s);
 573
 574                 if (tbl8_group_index < 0) {
 575                         return tbl8_group_index;
 576                 }
 577
 578                 tbl8_group_start = tbl8_group_index *
 579                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 580                 tbl8_group_end = tbl8_group_start +
 581                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 582
 583                 /* Populate new tbl8 with tbl24 value. */
 584                 for (i = tbl8_group_start; i < tbl8_group_end; i++) {
 585                         struct rte_lpm_tbl_entry new_tbl8_entry = {
 586                                 .valid = VALID,
 587                                 .depth = lpm->tbl24[tbl24_index].depth,
 588                                 .valid_group = lpm->tbl8[i].valid_group,
 589                                 .next_hop = lpm->tbl24[tbl24_index].next_hop,
 590                         };
 591                         __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 592                                         __ATOMIC_RELAXED);
 593                 }
 594
 595                 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
 596
 597                 /* Insert new rule into the tbl8 entry. */
 598                 for (i = tbl8_index; i < tbl8_index + tbl8_range; i++) {
 599                         struct rte_lpm_tbl_entry new_tbl8_entry = {
 600                                 .valid = VALID,
 601                                 .depth = depth,
 602                                 .valid_group = lpm->tbl8[i].valid_group,
 603                                 .next_hop = next_hop,
 604                         };
 605                         __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 606                                         __ATOMIC_RELAXED);
 607                 }
 608
 609                 /*
 610                  * Update tbl24 entry to point to new tbl8 entry. Note: The
 611                  * ext_flag and tbl8_index need to be updated simultaneously,
 612                  * so assign whole structure in one go.
 613                  */
 614
 615                 struct rte_lpm_tbl_entry new_tbl24_entry = {
 616                                 .group_idx = tbl8_group_index,
 617                                 .valid = VALID,
 618                                 .valid_group = 1,
 619                                 .depth = 0,
 620                 };
 621
 622                 /* The tbl24 entry must be written only after the
 623                  * tbl8 entries are written.
 624                  */
 625                 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
 626                                 __ATOMIC_RELEASE);
 627
 628         } else { /*
 629                 * If it is valid, extended entry calculate the index into tbl8.
 630                 */
 631                 tbl8_group_index = lpm->tbl24[tbl24_index].group_idx;
 632                 tbl8_group_start = tbl8_group_index *
 633                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 634                 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
 635
 636                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 637
 638                         if (!lpm->tbl8[i].valid ||
 639                                         lpm->tbl8[i].depth <= depth) {
 640                                 struct rte_lpm_tbl_entry new_tbl8_entry = {
 641                                         .valid = VALID,
 642                                         .depth = depth,
 643                                         .next_hop = next_hop,
 644                                         .valid_group = lpm->tbl8[i].valid_group,
 645                                 };
 646
 647                                 /*
 648                                  * Setting tbl8 entry in one go to avoid race
 649                                  * condition
 650                                  */
 651                                 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 652                                                 __ATOMIC_RELAXED);
 653
 654                                 continue;
 655                         }
 656                 }
 657         }
 658 #undef group_idx
 659         return 0;
 660 }
 661
 662 /*
 663  * Add a route
 664  */
 665 int
 666 rte_lpm_add(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 667                 uint32_t next_hop)
 668 {
 669         int32_t rule_index, status = 0;
 670         uint32_t ip_masked;
 671
 672         /* Check user arguments. */
 673         if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM_MAX_DEPTH))
 674                 return -EINVAL;
 675
 676         ip_masked = ip & depth_to_mask(depth);
 677
 678         /* Add the rule to the rule table. */
 679         rule_index = rule_add(lpm, ip_masked, depth, next_hop);
 680
 681         /* Skip table entries update if The rule is the same as
 682          * the rule in the rules table.
 683          */
 684         if (rule_index == -EEXIST)
 685                 return 0;
 686
 687         /* If the is no space available for new rule return error. */
 688         if (rule_index < 0) {
 689                 return rule_index;
 690         }
 691
 692         if (depth <= MAX_DEPTH_TBL24) {
 693                 status = add_depth_small(lpm, ip_masked, depth, next_hop);
 694         } else { /* If depth > RTE_LPM_MAX_DEPTH_TBL24 */
 695                 status = add_depth_big(lpm, ip_masked, depth, next_hop);
 696
 697                 /*
 698                  * If add fails due to exhaustion of tbl8 extensions delete
 699                  * rule that was added to rule table.
 700                  */
 701                 if (status < 0) {
 702                         rule_delete(lpm, rule_index, depth);
 703
 704                         return status;
 705                 }
 706         }
 707
 708         return 0;
 709 }
 710
 711 /*
 712  * Look for a rule in the high-level rules table
 713  */
 714 int
 715 rte_lpm_is_rule_present(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 716 uint32_t *next_hop)
 717 {
 718         uint32_t ip_masked;
 719         int32_t rule_index;
 720
 721         /* Check user arguments. */
 722         if ((lpm == NULL) ||
 723                 (next_hop == NULL) ||
 724                 (depth < 1) || (depth > RTE_LPM_MAX_DEPTH))
 725                 return -EINVAL;
 726
 727         /* Look for the rule using rule_find. */
 728         ip_masked = ip & depth_to_mask(depth);
 729         rule_index = rule_find(lpm, ip_masked, depth);
 730
 731         if (rule_index >= 0) {
 732                 *next_hop = lpm->rules_tbl[rule_index].next_hop;
 733                 return 1;
 734         }
 735
 736         /* If rule is not found return 0. */
 737         return 0;
 738 }
 739
 740 static int32_t
 741 find_previous_rule(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 742                 uint8_t *sub_rule_depth)
 743 {
 744         int32_t rule_index;
 745         uint32_t ip_masked;
 746         uint8_t prev_depth;
 747
 748         for (prev_depth = (uint8_t)(depth - 1); prev_depth > 0; prev_depth--) {
 749                 ip_masked = ip & depth_to_mask(prev_depth);
 750
 751                 rule_index = rule_find(lpm, ip_masked, prev_depth);
 752
 753                 if (rule_index >= 0) {
 754                         *sub_rule_depth = prev_depth;
 755                         return rule_index;
 756                 }
 757         }
 758
 759         return -1;
 760 }
 761
 762 static int32_t
 763 delete_depth_small(struct rte_lpm *lpm, uint32_t ip_masked,
 764         uint8_t depth, int32_t sub_rule_index, uint8_t sub_rule_depth)
 765 {
 766 #define group_idx next_hop
 767         uint32_t tbl24_range, tbl24_index, tbl8_group_index, tbl8_index, i, j;
 768
 769         /* Calculate the range and index into Table24. */
 770         tbl24_range = depth_to_range(depth);
 771         tbl24_index = (ip_masked >> 8);
 772         struct rte_lpm_tbl_entry zero_tbl24_entry = {0};
 773
 774         /*
 775          * Firstly check the sub_rule_index. A -1 indicates no replacement rule
 776          * and a positive number indicates a sub_rule_index.
 777          */
 778         if (sub_rule_index < 0) {
 779                 /*
 780                  * If no replacement rule exists then invalidate entries
 781                  * associated with this rule.
 782                  */
 783                 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
 784
 785                         if (lpm->tbl24[i].valid_group == 0 &&
 786                                         lpm->tbl24[i].depth <= depth) {
 787                                 __atomic_store(&lpm->tbl24[i],
 788                                         &zero_tbl24_entry, __ATOMIC_RELEASE);
 789                         } else if (lpm->tbl24[i].valid_group == 1) {
 790                                 /*
 791                                  * If TBL24 entry is extended, then there has
 792                                  * to be a rule with depth >= 25 in the
 793                                  * associated TBL8 group.
 794                                  */
 795
 796                                 tbl8_group_index = lpm->tbl24[i].group_idx;
 797                                 tbl8_index = tbl8_group_index *
 798                                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 799
 800                                 for (j = tbl8_index; j < (tbl8_index +
 801                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
 802
 803                                         if (lpm->tbl8[j].depth <= depth)
 804                                                 lpm->tbl8[j].valid = INVALID;
 805                                 }
 806                         }
 807                 }
 808         } else {
 809                 /*
 810                  * If a replacement rule exists then modify entries
 811                  * associated with this rule.
 812                  */
 813
 814                 struct rte_lpm_tbl_entry new_tbl24_entry = {
 815                         .next_hop = lpm->rules_tbl[sub_rule_index].next_hop,
 816                         .valid = VALID,
 817                         .valid_group = 0,
 818                         .depth = sub_rule_depth,
 819                 };
 820
 821                 struct rte_lpm_tbl_entry new_tbl8_entry = {
 822                         .valid = VALID,
 823                         .valid_group = VALID,
 824                         .depth = sub_rule_depth,
 825                         .next_hop = lpm->rules_tbl
 826                         [sub_rule_index].next_hop,
 827                 };
 828
 829                 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
 830
 831                         if (lpm->tbl24[i].valid_group == 0 &&
 832                                         lpm->tbl24[i].depth <= depth) {
 833                                 __atomic_store(&lpm->tbl24[i], &new_tbl24_entry,
 834                                                 __ATOMIC_RELEASE);
 835                         } else  if (lpm->tbl24[i].valid_group == 1) {
 836                                 /*
 837                                  * If TBL24 entry is extended, then there has
 838                                  * to be a rule with depth >= 25 in the
 839                                  * associated TBL8 group.
 840                                  */
 841
 842                                 tbl8_group_index = lpm->tbl24[i].group_idx;
 843                                 tbl8_index = tbl8_group_index *
 844                                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 845
 846                                 for (j = tbl8_index; j < (tbl8_index +
 847                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
 848
 849                                         if (lpm->tbl8[j].depth <= depth)
 850                                                 __atomic_store(&lpm->tbl8[j],
 851                                                         &new_tbl8_entry,
 852                                                         __ATOMIC_RELAXED);
 853                                 }
 854                         }
 855                 }
 856         }
 857 #undef group_idx
 858         return 0;
 859 }
 860
 861 /*
 862  * Checks if table 8 group can be recycled.
 863  *
 864  * Return of -EEXIST means tbl8 is in use and thus can not be recycled.
 865  * Return of -EINVAL means tbl8 is empty and thus can be recycled
 866  * Return of value > -1 means tbl8 is in use but has all the same values and
 867  * thus can be recycled
 868  */
 869 static int32_t
 870 tbl8_recycle_check(struct rte_lpm_tbl_entry *tbl8,
 871                 uint32_t tbl8_group_start)
 872 {
 873         uint32_t tbl8_group_end, i;
 874         tbl8_group_end = tbl8_group_start + RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 875
 876         /*
 877          * Check the first entry of the given tbl8. If it is invalid we know
 878          * this tbl8 does not contain any rule with a depth < RTE_LPM_MAX_DEPTH
 879          *  (As they would affect all entries in a tbl8) and thus this table
 880          *  can not be recycled.
 881          */
 882         if (tbl8[tbl8_group_start].valid) {
 883                 /*
 884                  * If first entry is valid check if the depth is less than 24
 885                  * and if so check the rest of the entries to verify that they
 886                  * are all of this depth.
 887                  */
 888                 if (tbl8[tbl8_group_start].depth <= MAX_DEPTH_TBL24) {
 889                         for (i = (tbl8_group_start + 1); i < tbl8_group_end;
 890                                         i++) {
 891
 892                                 if (tbl8[i].depth !=
 893                                                 tbl8[tbl8_group_start].depth) {
 894
 895                                         return -EEXIST;
 896                                 }
 897                         }
 898                         /* If all entries are the same return the tb8 index */
 899                         return tbl8_group_start;
 900                 }
 901
 902                 return -EEXIST;
 903         }
 904         /*
 905          * If the first entry is invalid check if the rest of the entries in
 906          * the tbl8 are invalid.
 907          */
 908         for (i = (tbl8_group_start + 1); i < tbl8_group_end; i++) {
 909                 if (tbl8[i].valid)
 910                         return -EEXIST;
 911         }
 912         /* If no valid entries are found then return -EINVAL. */
 913         return -EINVAL;
 914 }
 915
 916 static int32_t
 917 delete_depth_big(struct rte_lpm *lpm, uint32_t ip_masked,
 918         uint8_t depth, int32_t sub_rule_index, uint8_t sub_rule_depth)
 919 {
 920 #define group_idx next_hop
 921         uint32_t tbl24_index, tbl8_group_index, tbl8_group_start, tbl8_index,
 922                         tbl8_range, i;
 923         int32_t tbl8_recycle_index;
 924
 925         /*
 926          * Calculate the index into tbl24 and range. Note: All depths larger
 927          * than MAX_DEPTH_TBL24 are associated with only one tbl24 entry.
 928          */
 929         tbl24_index = ip_masked >> 8;
 930
 931         /* Calculate the index into tbl8 and range. */
 932         tbl8_group_index = lpm->tbl24[tbl24_index].group_idx;
 933         tbl8_group_start = tbl8_group_index * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 934         tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
 935         tbl8_range = depth_to_range(depth);
 936
 937         if (sub_rule_index < 0) {
 938                 /*
 939                  * Loop through the range of entries on tbl8 for which the
 940                  * rule_to_delete must be removed or modified.
 941                  */
 942                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 943                         if (lpm->tbl8[i].depth <= depth)
 944                                 lpm->tbl8[i].valid = INVALID;
 945                 }
 946         } else {
 947                 /* Set new tbl8 entry. */
 948                 struct rte_lpm_tbl_entry new_tbl8_entry = {
 949                         .valid = VALID,
 950                         .depth = sub_rule_depth,
 951                         .valid_group = lpm->tbl8[tbl8_group_start].valid_group,
 952                         .next_hop = lpm->rules_tbl[sub_rule_index].next_hop,
 953                 };
 954
 955                 /*
 956                  * Loop through the range of entries on tbl8 for which the
 957                  * rule_to_delete must be modified.
 958                  */
 959                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 960                         if (lpm->tbl8[i].depth <= depth)
 961                                 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 962                                                 __ATOMIC_RELAXED);
 963                 }
 964         }
 965
 966         /*
 967          * Check if there are any valid entries in this tbl8 group. If all
 968          * tbl8 entries are invalid we can free the tbl8 and invalidate the
 969          * associated tbl24 entry.
 970          */
 971
 972         tbl8_recycle_index = tbl8_recycle_check(lpm->tbl8, tbl8_group_start);
 973
 974         if (tbl8_recycle_index == -EINVAL) {
 975                 /* Set tbl24 before freeing tbl8 to avoid race condition.
 976                  * Prevent the free of the tbl8 group from hoisting.
 977                  */
 978                 lpm->tbl24[tbl24_index].valid = 0;
 979                 __atomic_thread_fence(__ATOMIC_RELEASE);
 980                 tbl8_free(lpm->tbl8, tbl8_group_start);
 981         } else if (tbl8_recycle_index > -1) {
 982                 /* Update tbl24 entry. */
 983                 struct rte_lpm_tbl_entry new_tbl24_entry = {
 984                         .next_hop = lpm->tbl8[tbl8_recycle_index].next_hop,
 985                         .valid = VALID,
 986                         .valid_group = 0,
 987                         .depth = lpm->tbl8[tbl8_recycle_index].depth,
 988                 };
 989
 990                 /* Set tbl24 before freeing tbl8 to avoid race condition.
 991                  * Prevent the free of the tbl8 group from hoisting.
 992                  */
 993                 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
 994                                 __ATOMIC_RELAXED);
 995                 __atomic_thread_fence(__ATOMIC_RELEASE);
 996                 tbl8_free(lpm->tbl8, tbl8_group_start);
 997         }
 998 #undef group_idx
 999         return 0;
1000 }
1001
1002 /*
1003  * Deletes a rule
1004  */
1005 int
1006 rte_lpm_delete(struct rte_lpm *lpm, uint32_t ip, uint8_t depth)
1007 {
1008         int32_t rule_to_delete_index, sub_rule_index;
1009         uint32_t ip_masked;
1010         uint8_t sub_rule_depth;
1011         /*
1012          * Check input arguments. Note: IP must be a positive integer of 32
1013          * bits in length therefore it need not be checked.
1014          */
1015         if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM_MAX_DEPTH)) {
1016                 return -EINVAL;
1017         }
1018
1019         ip_masked = ip & depth_to_mask(depth);
1020
1021         /*
1022          * Find the index of the input rule, that needs to be deleted, in the
1023          * rule table.
1024          */
1025         rule_to_delete_index = rule_find(lpm, ip_masked, depth);
1026
1027         /*
1028          * Check if rule_to_delete_index was found. If no rule was found the
1029          * function rule_find returns -EINVAL.
1030          */
1031         if (rule_to_delete_index < 0)
1032                 return -EINVAL;
1033
1034         /* Delete the rule from the rule table. */
1035         rule_delete(lpm, rule_to_delete_index, depth);
1036
1037         /*
1038          * Find rule to replace the rule_to_delete. If there is no rule to
1039          * replace the rule_to_delete we return -1 and invalidate the table
1040          * entries associated with this rule.
1041          */
1042         sub_rule_depth = 0;
1043         sub_rule_index = find_previous_rule(lpm, ip, depth, &sub_rule_depth);
1044
1045         /*
1046          * If the input depth value is less than 25 use function
1047          * delete_depth_small otherwise use delete_depth_big.
1048          */
1049         if (depth <= MAX_DEPTH_TBL24) {
1050                 return delete_depth_small(lpm, ip_masked, depth,
1051                                 sub_rule_index, sub_rule_depth);
1052         } else { /* If depth > MAX_DEPTH_TBL24 */
1053                 return delete_depth_big(lpm, ip_masked, depth, sub_rule_index,
1054                                 sub_rule_depth);
1055         }
1056 }
1057
1058 /*
1059  * Delete all rules from the LPM table.
1060  */
1061 void
1062 rte_lpm_delete_all(struct rte_lpm *lpm)
1063 {
1064         /* Zero rule information. */
1065         memset(lpm->rule_info, 0, sizeof(lpm->rule_info));
1066
1067         /* Zero tbl24. */
1068         memset(lpm->tbl24, 0, sizeof(lpm->tbl24));
1069
1070         /* Zero tbl8. */
1071         memset(lpm->tbl8, 0, sizeof(lpm->tbl8[0])
1072                         * RTE_LPM_TBL8_GROUP_NUM_ENTRIES * lpm->number_tbl8s);
1073
1074         /* Delete all rules form the rules table. */
1075         memset(lpm->rules_tbl, 0, sizeof(lpm->rules_tbl[0]) * lpm->max_rules);
1076 }