sys/contrib/dpdk_rte_lpm/rte_lpm.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation
   3  */
   4
   5 #include <sys/param.h>
   6 #include <sys/ctype.h>
   7 #include <sys/systm.h>
   8 #include <sys/lock.h>
   9 #include <sys/rwlock.h>
  10 #include <sys/malloc.h>
  11 #include <sys/mbuf.h>
  12 #include <sys/socket.h>
  13 #include <sys/kernel.h>
  14
  15 int errno = 0, rte_errno = 0;
  16
  17 #if 0
  18 #include <string.h>
  19 #include <stdint.h>
  20 #include <errno.h>
  21 #include <stdarg.h>
  22 #include <stdio.h>
  23 #include <sys/queue.h>
  24
  25 #include <rte_log.h>
  26 #include <rte_branch_prediction.h>
  27 #include <rte_common.h>
  28 #include <rte_memory.h>        /* for definition of RTE_CACHE_LINE_SIZE */
  29 #include <rte_malloc.h>
  30 #include <rte_eal.h>
  31 #include <rte_eal_memconfig.h>
  32 #include <rte_per_lcore.h>
  33 #include <rte_string_fns.h>
  34 #include <rte_errno.h>
  35 #include <rte_rwlock.h>
  36 #include <rte_spinlock.h>
  37 #include <rte_tailq.h>
  38 #endif
  39
  40 #include "rte_shim.h"
  41 #include "rte_lpm.h"
  42
  43 #if 0
  44 TAILQ_HEAD(rte_lpm_list, rte_tailq_entry);
  45
  46 static struct rte_tailq_elem rte_lpm_tailq = {
  47         .name = "RTE_LPM",
  48 };
  49 EAL_REGISTER_TAILQ(rte_lpm_tailq)
  50 #endif
  51
  52 #define MAX_DEPTH_TBL24 24
  53
  54 enum valid_flag {
  55         INVALID = 0,
  56         VALID
  57 };
  58
  59 /* Macro to enable/disable run-time checks. */
  60 #if defined(RTE_LIBRTE_LPM_DEBUG)
  61 #include <rte_debug.h>
  62 #define VERIFY_DEPTH(depth) do {                                \
  63         if ((depth == 0) || (depth > RTE_LPM_MAX_DEPTH))        \
  64                 rte_panic("LPM: Invalid depth (%u) at line %d", \
  65                                 (unsigned)(depth), __LINE__);   \
  66 } while (0)
  67 #else
  68 #define VERIFY_DEPTH(depth)
  69 #endif
  70
  71 /*
  72  * Converts a given depth value to its corresponding mask value.
  73  *
  74  * depth  (IN)          : range = 1 - 32
  75  * mask   (OUT)         : 32bit mask
  76  */
  77 static uint32_t __attribute__((pure))
  78 depth_to_mask(uint8_t depth)
  79 {
  80         VERIFY_DEPTH(depth);
  81
  82         /* To calculate a mask start with a 1 on the left hand side and right
  83          * shift while populating the left hand side with 1's
  84          */
  85         return (int)0x80000000 >> (depth - 1);
  86 }
  87
  88 /*
  89  * Converts given depth value to its corresponding range value.
  90  */
  91 static uint32_t __attribute__((pure))
  92 depth_to_range(uint8_t depth)
  93 {
  94         VERIFY_DEPTH(depth);
  95
  96         /*
  97          * Calculate tbl24 range. (Note: 2^depth = 1 << depth)
  98          */
  99         if (depth <= MAX_DEPTH_TBL24)
 100                 return 1 << (MAX_DEPTH_TBL24 - depth);
 101
 102         /* Else if depth is greater than 24 */
 103         return 1 << (RTE_LPM_MAX_DEPTH - depth);
 104 }
 105
 106 #if 0
 107 /*
 108  * Find an existing lpm table and return a pointer to it.
 109  */
 110 struct rte_lpm *
 111 rte_lpm_find_existing(const char *name)
 112 {
 113         struct rte_lpm *l = NULL;
 114         struct rte_tailq_entry *te;
 115         struct rte_lpm_list *lpm_list;
 116
 117         lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
 118
 119         rte_mcfg_tailq_read_lock();
 120         TAILQ_FOREACH(te, lpm_list, next) {
 121                 l = te->data;
 122                 if (strncmp(name, l->name, RTE_LPM_NAMESIZE) == 0)
 123                         break;
 124         }
 125         rte_mcfg_tailq_read_unlock();
 126
 127         if (te == NULL) {
 128                 rte_errno = ENOENT;
 129                 return NULL;
 130         }
 131
 132         return l;
 133 }
 134 #endif
 135
 136 /*
 137  * Allocates memory for LPM object
 138  */
 139 struct rte_lpm *
 140 rte_lpm_create(const char *name, int socket_id,
 141                 const struct rte_lpm_config *config)
 142 {
 143         char mem_name[RTE_LPM_NAMESIZE];
 144         struct rte_lpm *lpm = NULL;
 145         //struct rte_tailq_entry *te;
 146         uint32_t mem_size, rules_size, tbl8s_size;
 147         //struct rte_lpm_list *lpm_list;
 148
 149         //lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
 150
 151         RTE_BUILD_BUG_ON(sizeof(struct rte_lpm_tbl_entry) != 4);
 152
 153         /* Check user arguments. */
 154         if ((name == NULL) || (socket_id < -1)
 155                         || config->number_tbl8s > RTE_LPM_MAX_TBL8_NUM_GROUPS) {
 156                 rte_errno = EINVAL;
 157                 return NULL;
 158         }
 159
 160         snprintf(mem_name, sizeof(mem_name), "LPM_%s", name);
 161
 162         /* Determine the amount of memory to allocate. */
 163         mem_size = sizeof(*lpm);
 164         rules_size = sizeof(struct rte_lpm_rule) * config->max_rules;
 165         tbl8s_size = (sizeof(struct rte_lpm_tbl_entry) *
 166                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES * config->number_tbl8s);
 167
 168 #if 0
 169         rte_mcfg_tailq_write_lock();
 170
 171         /* guarantee there's no existing */
 172         TAILQ_FOREACH(te, lpm_list, next) {
 173                 lpm = te->data;
 174                 if (strncmp(name, lpm->name, RTE_LPM_NAMESIZE) == 0)
 175                         break;
 176         }
 177
 178         if (te != NULL) {
 179                 lpm = NULL;
 180                 rte_errno = EEXIST;
 181                 goto exit;
 182         }
 183
 184         /* allocate tailq entry */
 185         te = rte_zmalloc("LPM_TAILQ_ENTRY", sizeof(*te), 0);
 186         if (te == NULL) {
 187                 RTE_LOG(ERR, LPM, "Failed to allocate tailq entry\n");
 188                 rte_errno = ENOMEM;
 189                 goto exit;
 190         }
 191 #endif
 192
 193         /* Allocate memory to store the LPM data structures. */
 194         lpm = rte_zmalloc_socket(mem_name, mem_size,
 195                         RTE_CACHE_LINE_SIZE, socket_id);
 196         if (lpm == NULL) {
 197                 RTE_LOG(ERR, LPM, "LPM memory allocation failed\n");
 198                 //rte_free(te);
 199                 rte_errno = ENOMEM;
 200                 goto exit;
 201         }
 202
 203         lpm->rules_tbl = rte_zmalloc_socket(NULL,
 204                         (size_t)rules_size, RTE_CACHE_LINE_SIZE, socket_id);
 205
 206         if (lpm->rules_tbl == NULL) {
 207                 RTE_LOG(ERR, LPM, "LPM rules_tbl memory allocation failed\n");
 208                 rte_free(lpm);
 209                 lpm = NULL;
 210                 //rte_free(te);
 211                 rte_errno = ENOMEM;
 212                 goto exit;
 213         }
 214
 215         lpm->tbl8 = rte_zmalloc_socket(NULL,
 216                         (size_t)tbl8s_size, RTE_CACHE_LINE_SIZE, socket_id);
 217
 218         if (lpm->tbl8 == NULL) {
 219                 RTE_LOG(ERR, LPM, "LPM tbl8 memory allocation failed\n");
 220                 rte_free(lpm->rules_tbl);
 221                 rte_free(lpm);
 222                 lpm = NULL;
 223                 //rte_free(te);
 224                 rte_errno = ENOMEM;
 225                 goto exit;
 226         }
 227
 228         /* Save user arguments. */
 229         lpm->max_rules = config->max_rules;
 230         lpm->number_tbl8s = config->number_tbl8s;
 231         strlcpy(lpm->name, name, sizeof(lpm->name));
 232
 233         //te->data = lpm;
 234
 235         //TAILQ_INSERT_TAIL(lpm_list, te, next);
 236
 237 exit:
 238         rte_mcfg_tailq_write_unlock();
 239
 240         return lpm;
 241 }
 242
 243 /*
 244  * Deallocates memory for given LPM table.
 245  */
 246 void
 247 rte_lpm_free(struct rte_lpm *lpm)
 248 {
 249 #if 0
 250         struct rte_lpm_list *lpm_list;
 251         struct rte_tailq_entry *te;
 252
 253         /* Check user arguments. */
 254         if (lpm == NULL)
 255                 return;
 256
 257         lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
 258
 259         rte_mcfg_tailq_write_lock();
 260
 261         /* find our tailq entry */
 262         TAILQ_FOREACH(te, lpm_list, next) {
 263                 if (te->data == (void *) lpm)
 264                         break;
 265         }
 266         if (te != NULL)
 267                 TAILQ_REMOVE(lpm_list, te, next);
 268
 269         rte_mcfg_tailq_write_unlock();
 270 #endif
 271
 272         rte_free(lpm->tbl8);
 273         rte_free(lpm->rules_tbl);
 274         rte_free(lpm);
 275         //rte_free(te);
 276 }
 277
 278 #if 0
 279 /*
 280  * Adds a rule to the rule table.
 281  *
 282  * NOTE: The rule table is split into 32 groups. Each group contains rules that
 283  * apply to a specific prefix depth (i.e. group 1 contains rules that apply to
 284  * prefixes with a depth of 1 etc.). In the following code (depth - 1) is used
 285  * to refer to depth 1 because even though the depth range is 1 - 32, depths
 286  * are stored in the rule table from 0 - 31.
 287  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
 288  */
 289 static int32_t
 290 rule_add(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
 291         uint32_t next_hop)
 292 {
 293         uint32_t rule_gindex, rule_index, last_rule;
 294         int i;
 295
 296         VERIFY_DEPTH(depth);
 297
 298         /* Scan through rule group to see if rule already exists. */
 299         if (lpm->rule_info[depth - 1].used_rules > 0) {
 300
 301                 /* rule_gindex stands for rule group index. */
 302                 rule_gindex = lpm->rule_info[depth - 1].first_rule;
 303                 /* Initialise rule_index to point to start of rule group. */
 304                 rule_index = rule_gindex;
 305                 /* Last rule = Last used rule in this rule group. */
 306                 last_rule = rule_gindex + lpm->rule_info[depth - 1].used_rules;
 307
 308                 for (; rule_index < last_rule; rule_index++) {
 309
 310                         /* If rule already exists update next hop and return. */
 311                         if (lpm->rules_tbl[rule_index].ip == ip_masked) {
 312
 313                                 if (lpm->rules_tbl[rule_index].next_hop
 314                                                 == next_hop)
 315                                         return -EEXIST;
 316                                 lpm->rules_tbl[rule_index].next_hop = next_hop;
 317
 318                                 return rule_index;
 319                         }
 320                 }
 321
 322                 if (rule_index == lpm->max_rules)
 323                         return -ENOSPC;
 324         } else {
 325                 /* Calculate the position in which the rule will be stored. */
 326                 rule_index = 0;
 327
 328                 for (i = depth - 1; i > 0; i--) {
 329                         if (lpm->rule_info[i - 1].used_rules > 0) {
 330                                 rule_index = lpm->rule_info[i - 1].first_rule
 331                                                 + lpm->rule_info[i - 1].used_rules;
 332                                 break;
 333                         }
 334                 }
 335                 if (rule_index == lpm->max_rules)
 336                         return -ENOSPC;
 337
 338                 lpm->rule_info[depth - 1].first_rule = rule_index;
 339         }
 340
 341         /* Make room for the new rule in the array. */
 342         for (i = RTE_LPM_MAX_DEPTH; i > depth; i--) {
 343                 if (lpm->rule_info[i - 1].first_rule
 344                                 + lpm->rule_info[i - 1].used_rules == lpm->max_rules)
 345                         return -ENOSPC;
 346
 347                 if (lpm->rule_info[i - 1].used_rules > 0) {
 348                         lpm->rules_tbl[lpm->rule_info[i - 1].first_rule
 349                                 + lpm->rule_info[i - 1].used_rules]
 350                                         = lpm->rules_tbl[lpm->rule_info[i - 1].first_rule];
 351                         lpm->rule_info[i - 1].first_rule++;
 352                 }
 353         }
 354
 355         /* Add the new rule. */
 356         lpm->rules_tbl[rule_index].ip = ip_masked;
 357         lpm->rules_tbl[rule_index].next_hop = next_hop;
 358
 359         /* Increment the used rules counter for this rule group. */
 360         lpm->rule_info[depth - 1].used_rules++;
 361
 362         return rule_index;
 363 }
 364
 365 /*
 366  * Delete a rule from the rule table.
 367  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
 368  */
 369 static void
 370 rule_delete(struct rte_lpm *lpm, int32_t rule_index, uint8_t depth)
 371 {
 372         int i;
 373
 374         VERIFY_DEPTH(depth);
 375
 376         lpm->rules_tbl[rule_index] =
 377                         lpm->rules_tbl[lpm->rule_info[depth - 1].first_rule
 378                         + lpm->rule_info[depth - 1].used_rules - 1];
 379
 380         for (i = depth; i < RTE_LPM_MAX_DEPTH; i++) {
 381                 if (lpm->rule_info[i].used_rules > 0) {
 382                         lpm->rules_tbl[lpm->rule_info[i].first_rule - 1] =
 383                                         lpm->rules_tbl[lpm->rule_info[i].first_rule
 384                                                 + lpm->rule_info[i].used_rules - 1];
 385                         lpm->rule_info[i].first_rule--;
 386                 }
 387         }
 388
 389         lpm->rule_info[depth - 1].used_rules--;
 390 }
 391
 392 /*
 393  * Finds a rule in rule table.
 394  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
 395  */
 396 static int32_t
 397 rule_find(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth)
 398 {
 399         uint32_t rule_gindex, last_rule, rule_index;
 400
 401         VERIFY_DEPTH(depth);
 402
 403         rule_gindex = lpm->rule_info[depth - 1].first_rule;
 404         last_rule = rule_gindex + lpm->rule_info[depth - 1].used_rules;
 405
 406         /* Scan used rules at given depth to find rule. */
 407         for (rule_index = rule_gindex; rule_index < last_rule; rule_index++) {
 408                 /* If rule is found return the rule index. */
 409                 if (lpm->rules_tbl[rule_index].ip == ip_masked)
 410                         return rule_index;
 411         }
 412
 413         /* If rule is not found return -EINVAL. */
 414         return -EINVAL;
 415 }
 416 #endif
 417
 418 /*
 419  * Find, clean and allocate a tbl8.
 420  */
 421 static int32_t
 422 tbl8_alloc(struct rte_lpm_tbl_entry *tbl8, uint32_t number_tbl8s)
 423 {
 424         uint32_t group_idx; /* tbl8 group index. */
 425         struct rte_lpm_tbl_entry *tbl8_entry;
 426
 427         /* Scan through tbl8 to find a free (i.e. INVALID) tbl8 group. */
 428         for (group_idx = 0; group_idx < number_tbl8s; group_idx++) {
 429                 tbl8_entry = &tbl8[group_idx * RTE_LPM_TBL8_GROUP_NUM_ENTRIES];
 430                 /* If a free tbl8 group is found clean it and set as VALID. */
 431                 if (!tbl8_entry->valid_group) {
 432                         struct rte_lpm_tbl_entry new_tbl8_entry = {
 433                                 .next_hop = 0,
 434                                 .valid = INVALID,
 435                                 .depth = 0,
 436                                 .valid_group = VALID,
 437                         };
 438
 439                         memset(&tbl8_entry[0], 0,
 440                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES *
 441                                         sizeof(tbl8_entry[0]));
 442
 443                         __atomic_store(tbl8_entry, &new_tbl8_entry,
 444                                         __ATOMIC_RELAXED);
 445
 446                         /* Return group index for allocated tbl8 group. */
 447                         return group_idx;
 448                 }
 449         }
 450
 451         /* If there are no tbl8 groups free then return error. */
 452         return -ENOSPC;
 453 }
 454
 455 static void
 456 tbl8_free(struct rte_lpm_tbl_entry *tbl8, uint32_t tbl8_group_start)
 457 {
 458         /* Set tbl8 group invalid*/
 459         struct rte_lpm_tbl_entry zero_tbl8_entry = {0};
 460
 461         __atomic_store(&tbl8[tbl8_group_start], &zero_tbl8_entry,
 462                         __ATOMIC_RELAXED);
 463 }
 464
 465 static __rte_noinline int32_t
 466 add_depth_small(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 467                 uint32_t next_hop)
 468 {
 469 #define group_idx next_hop
 470         uint32_t tbl24_index, tbl24_range, tbl8_index, tbl8_group_end, i, j;
 471
 472         /* Calculate the index into Table24. */
 473         tbl24_index = ip >> 8;
 474         tbl24_range = depth_to_range(depth);
 475
 476         for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
 477                 /*
 478                  * For invalid OR valid and non-extended tbl 24 entries set
 479                  * entry.
 480                  */
 481                 if (!lpm->tbl24[i].valid || (lpm->tbl24[i].valid_group == 0 &&
 482                                 lpm->tbl24[i].depth <= depth)) {
 483
 484                         struct rte_lpm_tbl_entry new_tbl24_entry = {
 485                                 .next_hop = next_hop,
 486                                 .valid = VALID,
 487                                 .valid_group = 0,
 488                                 .depth = depth,
 489                         };
 490
 491                         /* Setting tbl24 entry in one go to avoid race
 492                          * conditions
 493                          */
 494                         __atomic_store(&lpm->tbl24[i], &new_tbl24_entry,
 495                                         __ATOMIC_RELEASE);
 496
 497                         continue;
 498                 }
 499
 500                 if (lpm->tbl24[i].valid_group == 1) {
 501                         /* If tbl24 entry is valid and extended calculate the
 502                          *  index into tbl8.
 503                          */
 504                         tbl8_index = lpm->tbl24[i].group_idx *
 505                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 506                         tbl8_group_end = tbl8_index +
 507                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 508
 509                         for (j = tbl8_index; j < tbl8_group_end; j++) {
 510                                 if (!lpm->tbl8[j].valid ||
 511                                                 lpm->tbl8[j].depth <= depth) {
 512                                         struct rte_lpm_tbl_entry
 513                                                 new_tbl8_entry = {
 514                                                 .valid = VALID,
 515                                                 .valid_group = VALID,
 516                                                 .depth = depth,
 517                                                 .next_hop = next_hop,
 518                                         };
 519
 520                                         /*
 521                                          * Setting tbl8 entry in one go to avoid
 522                                          * race conditions
 523                                          */
 524                                         __atomic_store(&lpm->tbl8[j],
 525                                                 &new_tbl8_entry,
 526                                                 __ATOMIC_RELAXED);
 527
 528                                         continue;
 529                                 }
 530                         }
 531                 }
 532         }
 533 #undef group_idx
 534         return 0;
 535 }
 536
 537 static __rte_noinline int32_t
 538 add_depth_big(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
 539                 uint32_t next_hop)
 540 {
 541 #define group_idx next_hop
 542         uint32_t tbl24_index;
 543         int32_t tbl8_group_index, tbl8_group_start, tbl8_group_end, tbl8_index,
 544                 tbl8_range, i;
 545
 546         tbl24_index = (ip_masked >> 8);
 547         tbl8_range = depth_to_range(depth);
 548
 549         if (!lpm->tbl24[tbl24_index].valid) {
 550                 /* Search for a free tbl8 group. */
 551                 tbl8_group_index = tbl8_alloc(lpm->tbl8, lpm->number_tbl8s);
 552
 553                 /* Check tbl8 allocation was successful. */
 554                 if (tbl8_group_index < 0) {
 555                         return tbl8_group_index;
 556                 }
 557
 558                 /* Find index into tbl8 and range. */
 559                 tbl8_index = (tbl8_group_index *
 560                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES) +
 561                                 (ip_masked & 0xFF);
 562
 563                 /* Set tbl8 entry. */
 564                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 565                         struct rte_lpm_tbl_entry new_tbl8_entry = {
 566                                 .valid = VALID,
 567                                 .depth = depth,
 568                                 .valid_group = lpm->tbl8[i].valid_group,
 569                                 .next_hop = next_hop,
 570                         };
 571                         __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 572                                         __ATOMIC_RELAXED);
 573                 }
 574
 575                 /*
 576                  * Update tbl24 entry to point to new tbl8 entry. Note: The
 577                  * ext_flag and tbl8_index need to be updated simultaneously,
 578                  * so assign whole structure in one go
 579                  */
 580
 581                 struct rte_lpm_tbl_entry new_tbl24_entry = {
 582                         .group_idx = tbl8_group_index,
 583                         .valid = VALID,
 584                         .valid_group = 1,
 585                         .depth = 0,
 586                 };
 587
 588                 /* The tbl24 entry must be written only after the
 589                  * tbl8 entries are written.
 590                  */
 591                 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
 592                                 __ATOMIC_RELEASE);
 593
 594         } /* If valid entry but not extended calculate the index into Table8. */
 595         else if (lpm->tbl24[tbl24_index].valid_group == 0) {
 596                 /* Search for free tbl8 group. */
 597                 tbl8_group_index = tbl8_alloc(lpm->tbl8, lpm->number_tbl8s);
 598
 599                 if (tbl8_group_index < 0) {
 600                         return tbl8_group_index;
 601                 }
 602
 603                 tbl8_group_start = tbl8_group_index *
 604                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 605                 tbl8_group_end = tbl8_group_start +
 606                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 607
 608                 /* Populate new tbl8 with tbl24 value. */
 609                 for (i = tbl8_group_start; i < tbl8_group_end; i++) {
 610                         struct rte_lpm_tbl_entry new_tbl8_entry = {
 611                                 .valid = VALID,
 612                                 .depth = lpm->tbl24[tbl24_index].depth,
 613                                 .valid_group = lpm->tbl8[i].valid_group,
 614                                 .next_hop = lpm->tbl24[tbl24_index].next_hop,
 615                         };
 616                         __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 617                                         __ATOMIC_RELAXED);
 618                 }
 619
 620                 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
 621
 622                 /* Insert new rule into the tbl8 entry. */
 623                 for (i = tbl8_index; i < tbl8_index + tbl8_range; i++) {
 624                         struct rte_lpm_tbl_entry new_tbl8_entry = {
 625                                 .valid = VALID,
 626                                 .depth = depth,
 627                                 .valid_group = lpm->tbl8[i].valid_group,
 628                                 .next_hop = next_hop,
 629                         };
 630                         __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 631                                         __ATOMIC_RELAXED);
 632                 }
 633
 634                 /*
 635                  * Update tbl24 entry to point to new tbl8 entry. Note: The
 636                  * ext_flag and tbl8_index need to be updated simultaneously,
 637                  * so assign whole structure in one go.
 638                  */
 639
 640                 struct rte_lpm_tbl_entry new_tbl24_entry = {
 641                                 .group_idx = tbl8_group_index,
 642                                 .valid = VALID,
 643                                 .valid_group = 1,
 644                                 .depth = 0,
 645                 };
 646
 647                 /* The tbl24 entry must be written only after the
 648                  * tbl8 entries are written.
 649                  */
 650                 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
 651                                 __ATOMIC_RELEASE);
 652
 653         } else { /*
 654                 * If it is valid, extended entry calculate the index into tbl8.
 655                 */
 656                 tbl8_group_index = lpm->tbl24[tbl24_index].group_idx;
 657                 tbl8_group_start = tbl8_group_index *
 658                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 659                 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
 660
 661                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 662
 663                         if (!lpm->tbl8[i].valid ||
 664                                         lpm->tbl8[i].depth <= depth) {
 665                                 struct rte_lpm_tbl_entry new_tbl8_entry = {
 666                                         .valid = VALID,
 667                                         .depth = depth,
 668                                         .next_hop = next_hop,
 669                                         .valid_group = lpm->tbl8[i].valid_group,
 670                                 };
 671
 672                                 /*
 673                                  * Setting tbl8 entry in one go to avoid race
 674                                  * condition
 675                                  */
 676                                 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 677                                                 __ATOMIC_RELAXED);
 678
 679                                 continue;
 680                         }
 681                 }
 682         }
 683 #undef group_idx
 684         return 0;
 685 }
 686
 687 /*
 688  * Add a route
 689  */
 690 int
 691 rte_lpm_add(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 692                 uint32_t next_hop)
 693 {
 694         int32_t status = 0;
 695         uint32_t ip_masked;
 696
 697         /* Check user arguments. */
 698         if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM_MAX_DEPTH))
 699                 return -EINVAL;
 700
 701         ip_masked = ip & depth_to_mask(depth);
 702
 703 #if 0
 704         /* Add the rule to the rule table. */
 705         rule_index = rule_add(lpm, ip_masked, depth, next_hop);
 706
 707         /* Skip table entries update if The rule is the same as
 708          * the rule in the rules table.
 709          */
 710         if (rule_index == -EEXIST)
 711                 return 0;
 712
 713         /* If the is no space available for new rule return error. */
 714         if (rule_index < 0) {
 715                 return rule_index;
 716         }
 717 #endif
 718
 719         if (depth <= MAX_DEPTH_TBL24) {
 720                 status = add_depth_small(lpm, ip_masked, depth, next_hop);
 721         } else { /* If depth > RTE_LPM_MAX_DEPTH_TBL24 */
 722                 status = add_depth_big(lpm, ip_masked, depth, next_hop);
 723
 724                 /*
 725                  * If add fails due to exhaustion of tbl8 extensions delete
 726                  * rule that was added to rule table.
 727                  */
 728                 if (status < 0) {
 729                         //rule_delete(lpm, rule_index, depth);
 730
 731                         return status;
 732                 }
 733         }
 734
 735         return 0;
 736 }
 737
 738 #if 0
 739 /*
 740  * Look for a rule in the high-level rules table
 741  */
 742 int
 743 rte_lpm_is_rule_present(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 744 uint32_t *next_hop)
 745 {
 746         uint32_t ip_masked;
 747         int32_t rule_index;
 748
 749         /* Check user arguments. */
 750         if ((lpm == NULL) ||
 751                 (next_hop == NULL) ||
 752                 (depth < 1) || (depth > RTE_LPM_MAX_DEPTH))
 753                 return -EINVAL;
 754
 755         /* Look for the rule using rule_find. */
 756         ip_masked = ip & depth_to_mask(depth);
 757         rule_index = rule_find(lpm, ip_masked, depth);
 758
 759         if (rule_index >= 0) {
 760                 *next_hop = lpm->rules_tbl[rule_index].next_hop;
 761                 return 1;
 762         }
 763
 764         /* If rule is not found return 0. */
 765         return 0;
 766 }
 767
 768 static int32_t
 769 find_previous_rule(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 770                 uint8_t *sub_rule_depth)
 771 {
 772         int32_t rule_index;
 773         uint32_t ip_masked;
 774         uint8_t prev_depth;
 775
 776         for (prev_depth = (uint8_t)(depth - 1); prev_depth > 0; prev_depth--) {
 777                 ip_masked = ip & depth_to_mask(prev_depth);
 778
 779                 rule_index = rule_find(lpm, ip_masked, prev_depth);
 780
 781                 if (rule_index >= 0) {
 782                         *sub_rule_depth = prev_depth;
 783                         return rule_index;
 784                 }
 785         }
 786
 787         return -1;
 788 }
 789 #endif
 790
 791 static int32_t
 792 delete_depth_small(struct rte_lpm *lpm, uint32_t ip_masked,
 793         uint8_t depth, uint32_t sub_rule_nhop, uint8_t sub_rule_depth)
 794 {
 795 #define group_idx next_hop
 796         uint32_t tbl24_range, tbl24_index, tbl8_group_index, tbl8_index, i, j;
 797
 798         /* Calculate the range and index into Table24. */
 799         tbl24_range = depth_to_range(depth);
 800         tbl24_index = (ip_masked >> 8);
 801         struct rte_lpm_tbl_entry zero_tbl24_entry = {0};
 802
 803         /*
 804          * Firstly check the sub_rule_index. A -1 indicates no replacement rule
 805          * and a positive number indicates a sub_rule_index.
 806          */
 807         if (sub_rule_nhop == 0) {
 808                 /*
 809                  * If no replacement rule exists then invalidate entries
 810                  * associated with this rule.
 811                  */
 812                 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
 813
 814                         if (lpm->tbl24[i].valid_group == 0 &&
 815                                         lpm->tbl24[i].depth <= depth) {
 816                                 __atomic_store(&lpm->tbl24[i],
 817                                         &zero_tbl24_entry, __ATOMIC_RELEASE);
 818                         } else if (lpm->tbl24[i].valid_group == 1) {
 819                                 /*
 820                                  * If TBL24 entry is extended, then there has
 821                                  * to be a rule with depth >= 25 in the
 822                                  * associated TBL8 group.
 823                                  */
 824
 825                                 tbl8_group_index = lpm->tbl24[i].group_idx;
 826                                 tbl8_index = tbl8_group_index *
 827                                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 828
 829                                 for (j = tbl8_index; j < (tbl8_index +
 830                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
 831
 832                                         if (lpm->tbl8[j].depth <= depth)
 833                                                 lpm->tbl8[j].valid = INVALID;
 834                                 }
 835                         }
 836                 }
 837         } else {
 838                 /*
 839                  * If a replacement rule exists then modify entries
 840                  * associated with this rule.
 841                  */
 842
 843                 struct rte_lpm_tbl_entry new_tbl24_entry = {
 844                         .next_hop = sub_rule_nhop,
 845                         .valid = VALID,
 846                         .valid_group = 0,
 847                         .depth = sub_rule_depth,
 848                 };
 849
 850                 struct rte_lpm_tbl_entry new_tbl8_entry = {
 851                         .valid = VALID,
 852                         .valid_group = VALID,
 853                         .depth = sub_rule_depth,
 854                         .next_hop = sub_rule_nhop,
 855                 };
 856
 857                 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
 858
 859                         if (lpm->tbl24[i].valid_group == 0 &&
 860                                         lpm->tbl24[i].depth <= depth) {
 861                                 __atomic_store(&lpm->tbl24[i], &new_tbl24_entry,
 862                                                 __ATOMIC_RELEASE);
 863                         } else  if (lpm->tbl24[i].valid_group == 1) {
 864                                 /*
 865                                  * If TBL24 entry is extended, then there has
 866                                  * to be a rule with depth >= 25 in the
 867                                  * associated TBL8 group.
 868                                  */
 869
 870                                 tbl8_group_index = lpm->tbl24[i].group_idx;
 871                                 tbl8_index = tbl8_group_index *
 872                                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 873
 874                                 for (j = tbl8_index; j < (tbl8_index +
 875                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
 876
 877                                         if (lpm->tbl8[j].depth <= depth)
 878                                                 __atomic_store(&lpm->tbl8[j],
 879                                                         &new_tbl8_entry,
 880                                                         __ATOMIC_RELAXED);
 881                                 }
 882                         }
 883                 }
 884         }
 885 #undef group_idx
 886         return 0;
 887 }
 888
 889 /*
 890  * Checks if table 8 group can be recycled.
 891  *
 892  * Return of -EEXIST means tbl8 is in use and thus can not be recycled.
 893  * Return of -EINVAL means tbl8 is empty and thus can be recycled
 894  * Return of value > -1 means tbl8 is in use but has all the same values and
 895  * thus can be recycled
 896  */
 897 static int32_t
 898 tbl8_recycle_check(struct rte_lpm_tbl_entry *tbl8,
 899                 uint32_t tbl8_group_start)
 900 {
 901         uint32_t tbl8_group_end, i;
 902         tbl8_group_end = tbl8_group_start + RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 903
 904         /*
 905          * Check the first entry of the given tbl8. If it is invalid we know
 906          * this tbl8 does not contain any rule with a depth < RTE_LPM_MAX_DEPTH
 907          *  (As they would affect all entries in a tbl8) and thus this table
 908          *  can not be recycled.
 909          */
 910         if (tbl8[tbl8_group_start].valid) {
 911                 /*
 912                  * If first entry is valid check if the depth is less than 24
 913                  * and if so check the rest of the entries to verify that they
 914                  * are all of this depth.
 915                  */
 916                 if (tbl8[tbl8_group_start].depth <= MAX_DEPTH_TBL24) {
 917                         for (i = (tbl8_group_start + 1); i < tbl8_group_end;
 918                                         i++) {
 919
 920                                 if (tbl8[i].depth !=
 921                                                 tbl8[tbl8_group_start].depth) {
 922
 923                                         return -EEXIST;
 924                                 }
 925                         }
 926                         /* If all entries are the same return the tb8 index */
 927                         return tbl8_group_start;
 928                 }
 929
 930                 return -EEXIST;
 931         }
 932         /*
 933          * If the first entry is invalid check if the rest of the entries in
 934          * the tbl8 are invalid.
 935          */
 936         for (i = (tbl8_group_start + 1); i < tbl8_group_end; i++) {
 937                 if (tbl8[i].valid)
 938                         return -EEXIST;
 939         }
 940         /* If no valid entries are found then return -EINVAL. */
 941         return -EINVAL;
 942 }
 943
 944 static int32_t
 945 delete_depth_big(struct rte_lpm *lpm, uint32_t ip_masked,
 946         uint8_t depth, uint32_t sub_rule_nhop, uint8_t sub_rule_depth)
 947 {
 948 #define group_idx next_hop
 949         uint32_t tbl24_index, tbl8_group_index, tbl8_group_start, tbl8_index,
 950                         tbl8_range, i;
 951         int32_t tbl8_recycle_index;
 952
 953         /*
 954          * Calculate the index into tbl24 and range. Note: All depths larger
 955          * than MAX_DEPTH_TBL24 are associated with only one tbl24 entry.
 956          */
 957         tbl24_index = ip_masked >> 8;
 958
 959         /* Calculate the index into tbl8 and range. */
 960         tbl8_group_index = lpm->tbl24[tbl24_index].group_idx;
 961         tbl8_group_start = tbl8_group_index * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
 962         tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
 963         tbl8_range = depth_to_range(depth);
 964
 965         if (sub_rule_nhop == 0) {
 966                 /*
 967                  * Loop through the range of entries on tbl8 for which the
 968                  * rule_to_delete must be removed or modified.
 969                  */
 970                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 971                         if (lpm->tbl8[i].depth <= depth)
 972                                 lpm->tbl8[i].valid = INVALID;
 973                 }
 974         } else {
 975                 /* Set new tbl8 entry. */
 976                 struct rte_lpm_tbl_entry new_tbl8_entry = {
 977                         .valid = VALID,
 978                         .depth = sub_rule_depth,
 979                         .valid_group = lpm->tbl8[tbl8_group_start].valid_group,
 980                         .next_hop = sub_rule_nhop,
 981                 };
 982
 983                 /*
 984                  * Loop through the range of entries on tbl8 for which the
 985                  * rule_to_delete must be modified.
 986                  */
 987                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 988                         if (lpm->tbl8[i].depth <= depth)
 989                                 __atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
 990                                                 __ATOMIC_RELAXED);
 991                 }
 992         }
 993
 994         /*
 995          * Check if there are any valid entries in this tbl8 group. If all
 996          * tbl8 entries are invalid we can free the tbl8 and invalidate the
 997          * associated tbl24 entry.
 998          */
 999
1000         tbl8_recycle_index = tbl8_recycle_check(lpm->tbl8, tbl8_group_start);
1001
1002         if (tbl8_recycle_index == -EINVAL) {
1003                 /* Set tbl24 before freeing tbl8 to avoid race condition.
1004                  * Prevent the free of the tbl8 group from hoisting.
1005                  */
1006                 lpm->tbl24[tbl24_index].valid = 0;
1007                 __atomic_thread_fence(__ATOMIC_RELEASE);
1008                 tbl8_free(lpm->tbl8, tbl8_group_start);
1009         } else if (tbl8_recycle_index > -1) {
1010                 /* Update tbl24 entry. */
1011                 struct rte_lpm_tbl_entry new_tbl24_entry = {
1012                         .next_hop = lpm->tbl8[tbl8_recycle_index].next_hop,
1013                         .valid = VALID,
1014                         .valid_group = 0,
1015                         .depth = lpm->tbl8[tbl8_recycle_index].depth,
1016                 };
1017
1018                 /* Set tbl24 before freeing tbl8 to avoid race condition.
1019                  * Prevent the free of the tbl8 group from hoisting.
1020                  */
1021                 __atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
1022                                 __ATOMIC_RELAXED);
1023                 __atomic_thread_fence(__ATOMIC_RELEASE);
1024                 tbl8_free(lpm->tbl8, tbl8_group_start);
1025         }
1026 #undef group_idx
1027         return 0;
1028 }
1029
1030 /*
1031  * Deletes a rule
1032  */
1033 int
1034 rte_lpm_delete(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
1035         uint8_t sub_rule_depth, uint32_t sub_rule_nhop)
1036 {
1037         //int32_t rule_to_delete_index;
1038         uint32_t ip_masked;
1039         //uint8_t sub_rule_depth;
1040         /*
1041          * Check input arguments. Note: IP must be a positive integer of 32
1042          * bits in length therefore it need not be checked.
1043          */
1044         if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM_MAX_DEPTH)) {
1045                 return -EINVAL;
1046         }
1047
1048         ip_masked = ip & depth_to_mask(depth);
1049
1050 #if 0
1051         /*
1052          * Find the index of the input rule, that needs to be deleted, in the
1053          * rule table.
1054          */
1055         rule_to_delete_index = rule_find(lpm, ip_masked, depth);
1056
1057         /*
1058          * Check if rule_to_delete_index was found. If no rule was found the
1059          * function rule_find returns -EINVAL.
1060          */
1061         if (rule_to_delete_index < 0)
1062                 return -EINVAL;
1063
1064         /* Delete the rule from the rule table. */
1065         rule_delete(lpm, rule_to_delete_index, depth);
1066 #endif
1067
1068         /*
1069          * Find rule to replace the rule_to_delete. If there is no rule to
1070          * replace the rule_to_delete we return -1 and invalidate the table
1071          * entries associated with this rule.
1072          */
1073         //sub_rule_depth = *psub_rule_depth;
1074         //sub_rule_index = find_previous_rule(lpm, ip, depth, &sub_rule_depth);
1075
1076         /*
1077          * If the input depth value is less than 25 use function
1078          * delete_depth_small otherwise use delete_depth_big.
1079          */
1080         if (depth <= MAX_DEPTH_TBL24) {
1081                 return delete_depth_small(lpm, ip_masked, depth,
1082                                 sub_rule_nhop, sub_rule_depth);
1083         } else { /* If depth > MAX_DEPTH_TBL24 */
1084                 return delete_depth_big(lpm, ip_masked, depth, sub_rule_nhop,
1085                                 sub_rule_depth);
1086         }
1087 }
1088
1089 /*
1090  * Delete all rules from the LPM table.
1091  */
1092 void
1093 rte_lpm_delete_all(struct rte_lpm *lpm)
1094 {
1095         /* Zero rule information. */
1096         memset(lpm->rule_info, 0, sizeof(lpm->rule_info));
1097
1098         /* Zero tbl24. */
1099         memset(lpm->tbl24, 0, sizeof(lpm->tbl24));
1100
1101         /* Zero tbl8. */
1102         memset(lpm->tbl8, 0, sizeof(lpm->tbl8[0])
1103                         * RTE_LPM_TBL8_GROUP_NUM_ENTRIES * lpm->number_tbl8s);
1104
1105         /* Delete all rules form the rules table. */
1106         memset(lpm->rules_tbl, 0, sizeof(lpm->rules_tbl[0]) * lpm->max_rules);
1107 }