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