ethdev: fix MAC address in telemetry device info
[dpdk.git] / lib / lpm / rte_lpm.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  * Copyright(c) 2020 Arm Limited
4  */
5
6 #include <string.h>
7 #include <stdint.h>
8 #include <errno.h>
9 #include <stdio.h>
10 #include <sys/queue.h>
11
12 #include <rte_log.h>
13 #include <rte_common.h>
14 #include <rte_malloc.h>
15 #include <rte_eal_memconfig.h>
16 #include <rte_string_fns.h>
17 #include <rte_errno.h>
18 #include <rte_tailq.h>
19
20 #include "rte_lpm.h"
21
22 TAILQ_HEAD(rte_lpm_list, rte_tailq_entry);
23
24 static struct rte_tailq_elem rte_lpm_tailq = {
25         .name = "RTE_LPM",
26 };
27 EAL_REGISTER_TAILQ(rte_lpm_tailq)
28
29 #define MAX_DEPTH_TBL24 24
30
31 enum valid_flag {
32         INVALID = 0,
33         VALID
34 };
35
36 /** @internal Rule structure. */
37 struct rte_lpm_rule {
38         uint32_t ip; /**< Rule IP address. */
39         uint32_t next_hop; /**< Rule next hop. */
40 };
41
42 /** @internal Contains metadata about the rules table. */
43 struct rte_lpm_rule_info {
44         uint32_t used_rules; /**< Used rules so far. */
45         uint32_t first_rule; /**< Indexes the first rule of a given depth. */
46 };
47
48 /** @internal LPM structure. */
49 struct __rte_lpm {
50         /* Exposed LPM data. */
51         struct rte_lpm lpm;
52
53         /* LPM metadata. */
54         char name[RTE_LPM_NAMESIZE];        /**< Name of the lpm. */
55         uint32_t max_rules; /**< Max. balanced rules per lpm. */
56         uint32_t number_tbl8s; /**< Number of tbl8s. */
57         /**< Rule info table. */
58         struct rte_lpm_rule_info rule_info[RTE_LPM_MAX_DEPTH];
59         struct rte_lpm_rule *rules_tbl; /**< LPM rules. */
60
61         /* RCU config. */
62         struct rte_rcu_qsbr *v;         /* RCU QSBR variable. */
63         enum rte_lpm_qsbr_mode rcu_mode;/* Blocking, defer queue. */
64         struct rte_rcu_qsbr_dq *dq;     /* RCU QSBR defer queue. */
65 };
66
67 /* Macro to enable/disable run-time checks. */
68 #if defined(RTE_LIBRTE_LPM_DEBUG)
69 #include <rte_debug.h>
70 #define VERIFY_DEPTH(depth) do {                                \
71         if ((depth == 0) || (depth > RTE_LPM_MAX_DEPTH))        \
72                 rte_panic("LPM: Invalid depth (%u) at line %d", \
73                                 (unsigned)(depth), __LINE__);   \
74 } while (0)
75 #else
76 #define VERIFY_DEPTH(depth)
77 #endif
78
79 /*
80  * Converts a given depth value to its corresponding mask value.
81  *
82  * depth  (IN)          : range = 1 - 32
83  * mask   (OUT)         : 32bit mask
84  */
85 static uint32_t __attribute__((pure))
86 depth_to_mask(uint8_t depth)
87 {
88         VERIFY_DEPTH(depth);
89
90         /* To calculate a mask start with a 1 on the left hand side and right
91          * shift while populating the left hand side with 1's
92          */
93         return (int)0x80000000 >> (depth - 1);
94 }
95
96 /*
97  * Converts given depth value to its corresponding range value.
98  */
99 static uint32_t __attribute__((pure))
100 depth_to_range(uint8_t depth)
101 {
102         VERIFY_DEPTH(depth);
103
104         /*
105          * Calculate tbl24 range. (Note: 2^depth = 1 << depth)
106          */
107         if (depth <= MAX_DEPTH_TBL24)
108                 return 1 << (MAX_DEPTH_TBL24 - depth);
109
110         /* Else if depth is greater than 24 */
111         return 1 << (RTE_LPM_MAX_DEPTH - depth);
112 }
113
114 /*
115  * Find an existing lpm table and return a pointer to it.
116  */
117 struct rte_lpm *
118 rte_lpm_find_existing(const char *name)
119 {
120         struct __rte_lpm *i_lpm = NULL;
121         struct rte_tailq_entry *te;
122         struct rte_lpm_list *lpm_list;
123
124         lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
125
126         rte_mcfg_tailq_read_lock();
127         TAILQ_FOREACH(te, lpm_list, next) {
128                 i_lpm = te->data;
129                 if (strncmp(name, i_lpm->name, RTE_LPM_NAMESIZE) == 0)
130                         break;
131         }
132         rte_mcfg_tailq_read_unlock();
133
134         if (te == NULL) {
135                 rte_errno = ENOENT;
136                 return NULL;
137         }
138
139         return &i_lpm->lpm;
140 }
141
142 /*
143  * Allocates memory for LPM object
144  */
145 struct rte_lpm *
146 rte_lpm_create(const char *name, int socket_id,
147                 const struct rte_lpm_config *config)
148 {
149         char mem_name[RTE_LPM_NAMESIZE];
150         struct __rte_lpm *i_lpm;
151         struct rte_lpm *lpm = NULL;
152         struct rte_tailq_entry *te;
153         uint32_t mem_size, rules_size, tbl8s_size;
154         struct rte_lpm_list *lpm_list;
155
156         lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
157
158         RTE_BUILD_BUG_ON(sizeof(struct rte_lpm_tbl_entry) != 4);
159
160         /* Check user arguments. */
161         if ((name == NULL) || (socket_id < -1) || (config->max_rules == 0)
162                         || config->number_tbl8s > RTE_LPM_MAX_TBL8_NUM_GROUPS) {
163                 rte_errno = EINVAL;
164                 return NULL;
165         }
166
167         snprintf(mem_name, sizeof(mem_name), "LPM_%s", name);
168
169         rte_mcfg_tailq_write_lock();
170
171         /* guarantee there's no existing */
172         TAILQ_FOREACH(te, lpm_list, next) {
173                 i_lpm = te->data;
174                 if (strncmp(name, i_lpm->name, RTE_LPM_NAMESIZE) == 0)
175                         break;
176         }
177
178         if (te != NULL) {
179                 rte_errno = EEXIST;
180                 goto exit;
181         }
182
183         /* Determine the amount of memory to allocate. */
184         mem_size = sizeof(*i_lpm);
185         rules_size = sizeof(struct rte_lpm_rule) * config->max_rules;
186         tbl8s_size = sizeof(struct rte_lpm_tbl_entry) *
187                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES * config->number_tbl8s;
188
189         /* allocate tailq entry */
190         te = rte_zmalloc("LPM_TAILQ_ENTRY", sizeof(*te), 0);
191         if (te == NULL) {
192                 RTE_LOG(ERR, LPM, "Failed to allocate tailq entry\n");
193                 rte_errno = ENOMEM;
194                 goto exit;
195         }
196
197         /* Allocate memory to store the LPM data structures. */
198         i_lpm = rte_zmalloc_socket(mem_name, mem_size,
199                         RTE_CACHE_LINE_SIZE, socket_id);
200         if (i_lpm == NULL) {
201                 RTE_LOG(ERR, LPM, "LPM memory allocation failed\n");
202                 rte_free(te);
203                 rte_errno = ENOMEM;
204                 goto exit;
205         }
206
207         i_lpm->rules_tbl = rte_zmalloc_socket(NULL,
208                         (size_t)rules_size, RTE_CACHE_LINE_SIZE, socket_id);
209
210         if (i_lpm->rules_tbl == NULL) {
211                 RTE_LOG(ERR, LPM, "LPM rules_tbl memory allocation failed\n");
212                 rte_free(i_lpm);
213                 i_lpm = NULL;
214                 rte_free(te);
215                 rte_errno = ENOMEM;
216                 goto exit;
217         }
218
219         i_lpm->lpm.tbl8 = rte_zmalloc_socket(NULL,
220                         (size_t)tbl8s_size, RTE_CACHE_LINE_SIZE, socket_id);
221
222         if (i_lpm->lpm.tbl8 == NULL) {
223                 RTE_LOG(ERR, LPM, "LPM tbl8 memory allocation failed\n");
224                 rte_free(i_lpm->rules_tbl);
225                 rte_free(i_lpm);
226                 i_lpm = NULL;
227                 rte_free(te);
228                 rte_errno = ENOMEM;
229                 goto exit;
230         }
231
232         /* Save user arguments. */
233         i_lpm->max_rules = config->max_rules;
234         i_lpm->number_tbl8s = config->number_tbl8s;
235         strlcpy(i_lpm->name, name, sizeof(i_lpm->name));
236
237         te->data = i_lpm;
238         lpm = &i_lpm->lpm;
239
240         TAILQ_INSERT_TAIL(lpm_list, te, next);
241
242 exit:
243         rte_mcfg_tailq_write_unlock();
244
245         return lpm;
246 }
247
248 /*
249  * Deallocates memory for given LPM table.
250  */
251 void
252 rte_lpm_free(struct rte_lpm *lpm)
253 {
254         struct rte_lpm_list *lpm_list;
255         struct rte_tailq_entry *te;
256         struct __rte_lpm *i_lpm;
257
258         /* Check user arguments. */
259         if (lpm == NULL)
260                 return;
261         i_lpm = container_of(lpm, struct __rte_lpm, lpm);
262
263         lpm_list = RTE_TAILQ_CAST(rte_lpm_tailq.head, rte_lpm_list);
264
265         rte_mcfg_tailq_write_lock();
266
267         /* find our tailq entry */
268         TAILQ_FOREACH(te, lpm_list, next) {
269                 if (te->data == (void *)i_lpm)
270                         break;
271         }
272         if (te != NULL)
273                 TAILQ_REMOVE(lpm_list, te, next);
274
275         rte_mcfg_tailq_write_unlock();
276
277         if (i_lpm->dq != NULL)
278                 rte_rcu_qsbr_dq_delete(i_lpm->dq);
279         rte_free(i_lpm->lpm.tbl8);
280         rte_free(i_lpm->rules_tbl);
281         rte_free(i_lpm);
282         rte_free(te);
283 }
284
285 static void
286 __lpm_rcu_qsbr_free_resource(void *p, void *data, unsigned int n)
287 {
288         struct rte_lpm_tbl_entry *tbl8 = ((struct __rte_lpm *)p)->lpm.tbl8;
289         struct rte_lpm_tbl_entry zero_tbl8_entry = {0};
290         uint32_t tbl8_group_index = *(uint32_t *)data;
291
292         RTE_SET_USED(n);
293         /* Set tbl8 group invalid */
294         __atomic_store(&tbl8[tbl8_group_index], &zero_tbl8_entry,
295                 __ATOMIC_RELAXED);
296 }
297
298 /* Associate QSBR variable with an LPM object.
299  */
300 int
301 rte_lpm_rcu_qsbr_add(struct rte_lpm *lpm, struct rte_lpm_rcu_config *cfg)
302 {
303         struct rte_rcu_qsbr_dq_parameters params = {0};
304         char rcu_dq_name[RTE_RCU_QSBR_DQ_NAMESIZE];
305         struct __rte_lpm *i_lpm;
306
307         if (lpm == NULL || cfg == NULL) {
308                 rte_errno = EINVAL;
309                 return 1;
310         }
311
312         i_lpm = container_of(lpm, struct __rte_lpm, lpm);
313         if (i_lpm->v != NULL) {
314                 rte_errno = EEXIST;
315                 return 1;
316         }
317
318         if (cfg->mode == RTE_LPM_QSBR_MODE_SYNC) {
319                 /* No other things to do. */
320         } else if (cfg->mode == RTE_LPM_QSBR_MODE_DQ) {
321                 /* Init QSBR defer queue. */
322                 snprintf(rcu_dq_name, sizeof(rcu_dq_name),
323                                 "LPM_RCU_%s", i_lpm->name);
324                 params.name = rcu_dq_name;
325                 params.size = cfg->dq_size;
326                 if (params.size == 0)
327                         params.size = i_lpm->number_tbl8s;
328                 params.trigger_reclaim_limit = cfg->reclaim_thd;
329                 params.max_reclaim_size = cfg->reclaim_max;
330                 if (params.max_reclaim_size == 0)
331                         params.max_reclaim_size = RTE_LPM_RCU_DQ_RECLAIM_MAX;
332                 params.esize = sizeof(uint32_t);        /* tbl8 group index */
333                 params.free_fn = __lpm_rcu_qsbr_free_resource;
334                 params.p = i_lpm;
335                 params.v = cfg->v;
336                 i_lpm->dq = rte_rcu_qsbr_dq_create(&params);
337                 if (i_lpm->dq == NULL) {
338                         RTE_LOG(ERR, LPM, "LPM defer queue creation failed\n");
339                         return 1;
340                 }
341         } else {
342                 rte_errno = EINVAL;
343                 return 1;
344         }
345         i_lpm->rcu_mode = cfg->mode;
346         i_lpm->v = cfg->v;
347
348         return 0;
349 }
350
351 /*
352  * Adds a rule to the rule table.
353  *
354  * NOTE: The rule table is split into 32 groups. Each group contains rules that
355  * apply to a specific prefix depth (i.e. group 1 contains rules that apply to
356  * prefixes with a depth of 1 etc.). In the following code (depth - 1) is used
357  * to refer to depth 1 because even though the depth range is 1 - 32, depths
358  * are stored in the rule table from 0 - 31.
359  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
360  */
361 static int32_t
362 rule_add(struct __rte_lpm *i_lpm, uint32_t ip_masked, uint8_t depth,
363         uint32_t next_hop)
364 {
365         uint32_t rule_gindex, rule_index, last_rule;
366         int i;
367
368         VERIFY_DEPTH(depth);
369
370         /* Scan through rule group to see if rule already exists. */
371         if (i_lpm->rule_info[depth - 1].used_rules > 0) {
372
373                 /* rule_gindex stands for rule group index. */
374                 rule_gindex = i_lpm->rule_info[depth - 1].first_rule;
375                 /* Initialise rule_index to point to start of rule group. */
376                 rule_index = rule_gindex;
377                 /* Last rule = Last used rule in this rule group. */
378                 last_rule = rule_gindex + i_lpm->rule_info[depth - 1].used_rules;
379
380                 for (; rule_index < last_rule; rule_index++) {
381
382                         /* If rule already exists update next hop and return. */
383                         if (i_lpm->rules_tbl[rule_index].ip == ip_masked) {
384
385                                 if (i_lpm->rules_tbl[rule_index].next_hop
386                                                 == next_hop)
387                                         return -EEXIST;
388                                 i_lpm->rules_tbl[rule_index].next_hop = next_hop;
389
390                                 return rule_index;
391                         }
392                 }
393
394                 if (rule_index == i_lpm->max_rules)
395                         return -ENOSPC;
396         } else {
397                 /* Calculate the position in which the rule will be stored. */
398                 rule_index = 0;
399
400                 for (i = depth - 1; i > 0; i--) {
401                         if (i_lpm->rule_info[i - 1].used_rules > 0) {
402                                 rule_index = i_lpm->rule_info[i - 1].first_rule
403                                                 + i_lpm->rule_info[i - 1].used_rules;
404                                 break;
405                         }
406                 }
407                 if (rule_index == i_lpm->max_rules)
408                         return -ENOSPC;
409
410                 i_lpm->rule_info[depth - 1].first_rule = rule_index;
411         }
412
413         /* Make room for the new rule in the array. */
414         for (i = RTE_LPM_MAX_DEPTH; i > depth; i--) {
415                 if (i_lpm->rule_info[i - 1].first_rule
416                                 + i_lpm->rule_info[i - 1].used_rules == i_lpm->max_rules)
417                         return -ENOSPC;
418
419                 if (i_lpm->rule_info[i - 1].used_rules > 0) {
420                         i_lpm->rules_tbl[i_lpm->rule_info[i - 1].first_rule
421                                 + i_lpm->rule_info[i - 1].used_rules]
422                                         = i_lpm->rules_tbl[i_lpm->rule_info[i - 1].first_rule];
423                         i_lpm->rule_info[i - 1].first_rule++;
424                 }
425         }
426
427         /* Add the new rule. */
428         i_lpm->rules_tbl[rule_index].ip = ip_masked;
429         i_lpm->rules_tbl[rule_index].next_hop = next_hop;
430
431         /* Increment the used rules counter for this rule group. */
432         i_lpm->rule_info[depth - 1].used_rules++;
433
434         return rule_index;
435 }
436
437 /*
438  * Delete a rule from the rule table.
439  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
440  */
441 static void
442 rule_delete(struct __rte_lpm *i_lpm, int32_t rule_index, uint8_t depth)
443 {
444         int i;
445
446         VERIFY_DEPTH(depth);
447
448         i_lpm->rules_tbl[rule_index] =
449                         i_lpm->rules_tbl[i_lpm->rule_info[depth - 1].first_rule
450                         + i_lpm->rule_info[depth - 1].used_rules - 1];
451
452         for (i = depth; i < RTE_LPM_MAX_DEPTH; i++) {
453                 if (i_lpm->rule_info[i].used_rules > 0) {
454                         i_lpm->rules_tbl[i_lpm->rule_info[i].first_rule - 1] =
455                                         i_lpm->rules_tbl[i_lpm->rule_info[i].first_rule
456                                                 + i_lpm->rule_info[i].used_rules - 1];
457                         i_lpm->rule_info[i].first_rule--;
458                 }
459         }
460
461         i_lpm->rule_info[depth - 1].used_rules--;
462 }
463
464 /*
465  * Finds a rule in rule table.
466  * NOTE: Valid range for depth parameter is 1 .. 32 inclusive.
467  */
468 static int32_t
469 rule_find(struct __rte_lpm *i_lpm, uint32_t ip_masked, uint8_t depth)
470 {
471         uint32_t rule_gindex, last_rule, rule_index;
472
473         VERIFY_DEPTH(depth);
474
475         rule_gindex = i_lpm->rule_info[depth - 1].first_rule;
476         last_rule = rule_gindex + i_lpm->rule_info[depth - 1].used_rules;
477
478         /* Scan used rules at given depth to find rule. */
479         for (rule_index = rule_gindex; rule_index < last_rule; rule_index++) {
480                 /* If rule is found return the rule index. */
481                 if (i_lpm->rules_tbl[rule_index].ip == ip_masked)
482                         return rule_index;
483         }
484
485         /* If rule is not found return -EINVAL. */
486         return -EINVAL;
487 }
488
489 /*
490  * Find, clean and allocate a tbl8.
491  */
492 static int32_t
493 _tbl8_alloc(struct __rte_lpm *i_lpm)
494 {
495         uint32_t group_idx; /* tbl8 group index. */
496         struct rte_lpm_tbl_entry *tbl8_entry;
497
498         /* Scan through tbl8 to find a free (i.e. INVALID) tbl8 group. */
499         for (group_idx = 0; group_idx < i_lpm->number_tbl8s; group_idx++) {
500                 tbl8_entry = &i_lpm->lpm.tbl8[group_idx *
501                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES];
502                 /* If a free tbl8 group is found clean it and set as VALID. */
503                 if (!tbl8_entry->valid_group) {
504                         struct rte_lpm_tbl_entry new_tbl8_entry = {
505                                 .next_hop = 0,
506                                 .valid = INVALID,
507                                 .depth = 0,
508                                 .valid_group = VALID,
509                         };
510
511                         memset(&tbl8_entry[0], 0,
512                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES *
513                                         sizeof(tbl8_entry[0]));
514
515                         __atomic_store(tbl8_entry, &new_tbl8_entry,
516                                         __ATOMIC_RELAXED);
517
518                         /* Return group index for allocated tbl8 group. */
519                         return group_idx;
520                 }
521         }
522
523         /* If there are no tbl8 groups free then return error. */
524         return -ENOSPC;
525 }
526
527 static int32_t
528 tbl8_alloc(struct __rte_lpm *i_lpm)
529 {
530         int32_t group_idx; /* tbl8 group index. */
531
532         group_idx = _tbl8_alloc(i_lpm);
533         if (group_idx == -ENOSPC && i_lpm->dq != NULL) {
534                 /* If there are no tbl8 groups try to reclaim one. */
535                 if (rte_rcu_qsbr_dq_reclaim(i_lpm->dq, 1,
536                                 NULL, NULL, NULL) == 0)
537                         group_idx = _tbl8_alloc(i_lpm);
538         }
539
540         return group_idx;
541 }
542
543 static int32_t
544 tbl8_free(struct __rte_lpm *i_lpm, uint32_t tbl8_group_start)
545 {
546         struct rte_lpm_tbl_entry zero_tbl8_entry = {0};
547         int status;
548
549         if (i_lpm->v == NULL) {
550                 /* Set tbl8 group invalid*/
551                 __atomic_store(&i_lpm->lpm.tbl8[tbl8_group_start], &zero_tbl8_entry,
552                                 __ATOMIC_RELAXED);
553         } else if (i_lpm->rcu_mode == RTE_LPM_QSBR_MODE_SYNC) {
554                 /* Wait for quiescent state change. */
555                 rte_rcu_qsbr_synchronize(i_lpm->v,
556                         RTE_QSBR_THRID_INVALID);
557                 /* Set tbl8 group invalid*/
558                 __atomic_store(&i_lpm->lpm.tbl8[tbl8_group_start], &zero_tbl8_entry,
559                                 __ATOMIC_RELAXED);
560         } else if (i_lpm->rcu_mode == RTE_LPM_QSBR_MODE_DQ) {
561                 /* Push into QSBR defer queue. */
562                 status = rte_rcu_qsbr_dq_enqueue(i_lpm->dq,
563                                 (void *)&tbl8_group_start);
564                 if (status == 1) {
565                         RTE_LOG(ERR, LPM, "Failed to push QSBR FIFO\n");
566                         return -rte_errno;
567                 }
568         }
569
570         return 0;
571 }
572
573 static __rte_noinline int32_t
574 add_depth_small(struct __rte_lpm *i_lpm, uint32_t ip, uint8_t depth,
575                 uint32_t next_hop)
576 {
577 #define group_idx next_hop
578         uint32_t tbl24_index, tbl24_range, tbl8_index, tbl8_group_end, i, j;
579
580         /* Calculate the index into Table24. */
581         tbl24_index = ip >> 8;
582         tbl24_range = depth_to_range(depth);
583
584         for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
585                 /*
586                  * For invalid OR valid and non-extended tbl 24 entries set
587                  * entry.
588                  */
589                 if (!i_lpm->lpm.tbl24[i].valid || (i_lpm->lpm.tbl24[i].valid_group == 0 &&
590                                 i_lpm->lpm.tbl24[i].depth <= depth)) {
591
592                         struct rte_lpm_tbl_entry new_tbl24_entry = {
593                                 .next_hop = next_hop,
594                                 .valid = VALID,
595                                 .valid_group = 0,
596                                 .depth = depth,
597                         };
598
599                         /* Setting tbl24 entry in one go to avoid race
600                          * conditions
601                          */
602                         __atomic_store(&i_lpm->lpm.tbl24[i], &new_tbl24_entry,
603                                         __ATOMIC_RELEASE);
604
605                         continue;
606                 }
607
608                 if (i_lpm->lpm.tbl24[i].valid_group == 1) {
609                         /* If tbl24 entry is valid and extended calculate the
610                          *  index into tbl8.
611                          */
612                         tbl8_index = i_lpm->lpm.tbl24[i].group_idx *
613                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
614                         tbl8_group_end = tbl8_index +
615                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
616
617                         for (j = tbl8_index; j < tbl8_group_end; j++) {
618                                 if (!i_lpm->lpm.tbl8[j].valid ||
619                                                 i_lpm->lpm.tbl8[j].depth <= depth) {
620                                         struct rte_lpm_tbl_entry
621                                                 new_tbl8_entry = {
622                                                 .valid = VALID,
623                                                 .valid_group = VALID,
624                                                 .depth = depth,
625                                                 .next_hop = next_hop,
626                                         };
627
628                                         /*
629                                          * Setting tbl8 entry in one go to avoid
630                                          * race conditions
631                                          */
632                                         __atomic_store(&i_lpm->lpm.tbl8[j],
633                                                 &new_tbl8_entry,
634                                                 __ATOMIC_RELAXED);
635
636                                         continue;
637                                 }
638                         }
639                 }
640         }
641 #undef group_idx
642         return 0;
643 }
644
645 static __rte_noinline int32_t
646 add_depth_big(struct __rte_lpm *i_lpm, uint32_t ip_masked, uint8_t depth,
647                 uint32_t next_hop)
648 {
649 #define group_idx next_hop
650         uint32_t tbl24_index;
651         int32_t tbl8_group_index, tbl8_group_start, tbl8_group_end, tbl8_index,
652                 tbl8_range, i;
653
654         tbl24_index = (ip_masked >> 8);
655         tbl8_range = depth_to_range(depth);
656
657         if (!i_lpm->lpm.tbl24[tbl24_index].valid) {
658                 /* Search for a free tbl8 group. */
659                 tbl8_group_index = tbl8_alloc(i_lpm);
660
661                 /* Check tbl8 allocation was successful. */
662                 if (tbl8_group_index < 0) {
663                         return tbl8_group_index;
664                 }
665
666                 /* Find index into tbl8 and range. */
667                 tbl8_index = (tbl8_group_index *
668                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES) +
669                                 (ip_masked & 0xFF);
670
671                 /* Set tbl8 entry. */
672                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
673                         struct rte_lpm_tbl_entry new_tbl8_entry = {
674                                 .valid = VALID,
675                                 .depth = depth,
676                                 .valid_group = i_lpm->lpm.tbl8[i].valid_group,
677                                 .next_hop = next_hop,
678                         };
679                         __atomic_store(&i_lpm->lpm.tbl8[i], &new_tbl8_entry,
680                                         __ATOMIC_RELAXED);
681                 }
682
683                 /*
684                  * Update tbl24 entry to point to new tbl8 entry. Note: The
685                  * ext_flag and tbl8_index need to be updated simultaneously,
686                  * so assign whole structure in one go
687                  */
688
689                 struct rte_lpm_tbl_entry new_tbl24_entry = {
690                         .group_idx = tbl8_group_index,
691                         .valid = VALID,
692                         .valid_group = 1,
693                         .depth = 0,
694                 };
695
696                 /* The tbl24 entry must be written only after the
697                  * tbl8 entries are written.
698                  */
699                 __atomic_store(&i_lpm->lpm.tbl24[tbl24_index], &new_tbl24_entry,
700                                 __ATOMIC_RELEASE);
701
702         } /* If valid entry but not extended calculate the index into Table8. */
703         else if (i_lpm->lpm.tbl24[tbl24_index].valid_group == 0) {
704                 /* Search for free tbl8 group. */
705                 tbl8_group_index = tbl8_alloc(i_lpm);
706
707                 if (tbl8_group_index < 0) {
708                         return tbl8_group_index;
709                 }
710
711                 tbl8_group_start = tbl8_group_index *
712                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
713                 tbl8_group_end = tbl8_group_start +
714                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
715
716                 /* Populate new tbl8 with tbl24 value. */
717                 for (i = tbl8_group_start; i < tbl8_group_end; i++) {
718                         struct rte_lpm_tbl_entry new_tbl8_entry = {
719                                 .valid = VALID,
720                                 .depth = i_lpm->lpm.tbl24[tbl24_index].depth,
721                                 .valid_group = i_lpm->lpm.tbl8[i].valid_group,
722                                 .next_hop = i_lpm->lpm.tbl24[tbl24_index].next_hop,
723                         };
724                         __atomic_store(&i_lpm->lpm.tbl8[i], &new_tbl8_entry,
725                                         __ATOMIC_RELAXED);
726                 }
727
728                 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
729
730                 /* Insert new rule into the tbl8 entry. */
731                 for (i = tbl8_index; i < tbl8_index + tbl8_range; i++) {
732                         struct rte_lpm_tbl_entry new_tbl8_entry = {
733                                 .valid = VALID,
734                                 .depth = depth,
735                                 .valid_group = i_lpm->lpm.tbl8[i].valid_group,
736                                 .next_hop = next_hop,
737                         };
738                         __atomic_store(&i_lpm->lpm.tbl8[i], &new_tbl8_entry,
739                                         __ATOMIC_RELAXED);
740                 }
741
742                 /*
743                  * Update tbl24 entry to point to new tbl8 entry. Note: The
744                  * ext_flag and tbl8_index need to be updated simultaneously,
745                  * so assign whole structure in one go.
746                  */
747
748                 struct rte_lpm_tbl_entry new_tbl24_entry = {
749                                 .group_idx = tbl8_group_index,
750                                 .valid = VALID,
751                                 .valid_group = 1,
752                                 .depth = 0,
753                 };
754
755                 /* The tbl24 entry must be written only after the
756                  * tbl8 entries are written.
757                  */
758                 __atomic_store(&i_lpm->lpm.tbl24[tbl24_index], &new_tbl24_entry,
759                                 __ATOMIC_RELEASE);
760
761         } else { /*
762                 * If it is valid, extended entry calculate the index into tbl8.
763                 */
764                 tbl8_group_index = i_lpm->lpm.tbl24[tbl24_index].group_idx;
765                 tbl8_group_start = tbl8_group_index *
766                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
767                 tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
768
769                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
770
771                         if (!i_lpm->lpm.tbl8[i].valid ||
772                                         i_lpm->lpm.tbl8[i].depth <= depth) {
773                                 struct rte_lpm_tbl_entry new_tbl8_entry = {
774                                         .valid = VALID,
775                                         .depth = depth,
776                                         .next_hop = next_hop,
777                                         .valid_group = i_lpm->lpm.tbl8[i].valid_group,
778                                 };
779
780                                 /*
781                                  * Setting tbl8 entry in one go to avoid race
782                                  * condition
783                                  */
784                                 __atomic_store(&i_lpm->lpm.tbl8[i], &new_tbl8_entry,
785                                                 __ATOMIC_RELAXED);
786
787                                 continue;
788                         }
789                 }
790         }
791 #undef group_idx
792         return 0;
793 }
794
795 /*
796  * Add a route
797  */
798 int
799 rte_lpm_add(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
800                 uint32_t next_hop)
801 {
802         int32_t rule_index, status = 0;
803         struct __rte_lpm *i_lpm;
804         uint32_t ip_masked;
805
806         /* Check user arguments. */
807         if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM_MAX_DEPTH))
808                 return -EINVAL;
809
810         i_lpm = container_of(lpm, struct __rte_lpm, lpm);
811         ip_masked = ip & depth_to_mask(depth);
812
813         /* Add the rule to the rule table. */
814         rule_index = rule_add(i_lpm, ip_masked, depth, next_hop);
815
816         /* Skip table entries update if The rule is the same as
817          * the rule in the rules table.
818          */
819         if (rule_index == -EEXIST)
820                 return 0;
821
822         /* If the is no space available for new rule return error. */
823         if (rule_index < 0) {
824                 return rule_index;
825         }
826
827         if (depth <= MAX_DEPTH_TBL24) {
828                 status = add_depth_small(i_lpm, ip_masked, depth, next_hop);
829         } else { /* If depth > RTE_LPM_MAX_DEPTH_TBL24 */
830                 status = add_depth_big(i_lpm, ip_masked, depth, next_hop);
831
832                 /*
833                  * If add fails due to exhaustion of tbl8 extensions delete
834                  * rule that was added to rule table.
835                  */
836                 if (status < 0) {
837                         rule_delete(i_lpm, rule_index, depth);
838
839                         return status;
840                 }
841         }
842
843         return 0;
844 }
845
846 /*
847  * Look for a rule in the high-level rules table
848  */
849 int
850 rte_lpm_is_rule_present(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
851 uint32_t *next_hop)
852 {
853         struct __rte_lpm *i_lpm;
854         uint32_t ip_masked;
855         int32_t rule_index;
856
857         /* Check user arguments. */
858         if ((lpm == NULL) ||
859                 (next_hop == NULL) ||
860                 (depth < 1) || (depth > RTE_LPM_MAX_DEPTH))
861                 return -EINVAL;
862
863         /* Look for the rule using rule_find. */
864         i_lpm = container_of(lpm, struct __rte_lpm, lpm);
865         ip_masked = ip & depth_to_mask(depth);
866         rule_index = rule_find(i_lpm, ip_masked, depth);
867
868         if (rule_index >= 0) {
869                 *next_hop = i_lpm->rules_tbl[rule_index].next_hop;
870                 return 1;
871         }
872
873         /* If rule is not found return 0. */
874         return 0;
875 }
876
877 static int32_t
878 find_previous_rule(struct __rte_lpm *i_lpm, uint32_t ip, uint8_t depth,
879                 uint8_t *sub_rule_depth)
880 {
881         int32_t rule_index;
882         uint32_t ip_masked;
883         uint8_t prev_depth;
884
885         for (prev_depth = (uint8_t)(depth - 1); prev_depth > 0; prev_depth--) {
886                 ip_masked = ip & depth_to_mask(prev_depth);
887
888                 rule_index = rule_find(i_lpm, ip_masked, prev_depth);
889
890                 if (rule_index >= 0) {
891                         *sub_rule_depth = prev_depth;
892                         return rule_index;
893                 }
894         }
895
896         return -1;
897 }
898
899 static int32_t
900 delete_depth_small(struct __rte_lpm *i_lpm, uint32_t ip_masked,
901         uint8_t depth, int32_t sub_rule_index, uint8_t sub_rule_depth)
902 {
903 #define group_idx next_hop
904         uint32_t tbl24_range, tbl24_index, tbl8_group_index, tbl8_index, i, j;
905
906         /* Calculate the range and index into Table24. */
907         tbl24_range = depth_to_range(depth);
908         tbl24_index = (ip_masked >> 8);
909         struct rte_lpm_tbl_entry zero_tbl24_entry = {0};
910
911         /*
912          * Firstly check the sub_rule_index. A -1 indicates no replacement rule
913          * and a positive number indicates a sub_rule_index.
914          */
915         if (sub_rule_index < 0) {
916                 /*
917                  * If no replacement rule exists then invalidate entries
918                  * associated with this rule.
919                  */
920                 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
921
922                         if (i_lpm->lpm.tbl24[i].valid_group == 0 &&
923                                         i_lpm->lpm.tbl24[i].depth <= depth) {
924                                 __atomic_store(&i_lpm->lpm.tbl24[i],
925                                         &zero_tbl24_entry, __ATOMIC_RELEASE);
926                         } else if (i_lpm->lpm.tbl24[i].valid_group == 1) {
927                                 /*
928                                  * If TBL24 entry is extended, then there has
929                                  * to be a rule with depth >= 25 in the
930                                  * associated TBL8 group.
931                                  */
932
933                                 tbl8_group_index = i_lpm->lpm.tbl24[i].group_idx;
934                                 tbl8_index = tbl8_group_index *
935                                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
936
937                                 for (j = tbl8_index; j < (tbl8_index +
938                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
939
940                                         if (i_lpm->lpm.tbl8[j].depth <= depth)
941                                                 i_lpm->lpm.tbl8[j].valid = INVALID;
942                                 }
943                         }
944                 }
945         } else {
946                 /*
947                  * If a replacement rule exists then modify entries
948                  * associated with this rule.
949                  */
950
951                 struct rte_lpm_tbl_entry new_tbl24_entry = {
952                         .next_hop = i_lpm->rules_tbl[sub_rule_index].next_hop,
953                         .valid = VALID,
954                         .valid_group = 0,
955                         .depth = sub_rule_depth,
956                 };
957
958                 struct rte_lpm_tbl_entry new_tbl8_entry = {
959                         .valid = VALID,
960                         .valid_group = VALID,
961                         .depth = sub_rule_depth,
962                         .next_hop = i_lpm->rules_tbl
963                         [sub_rule_index].next_hop,
964                 };
965
966                 for (i = tbl24_index; i < (tbl24_index + tbl24_range); i++) {
967
968                         if (i_lpm->lpm.tbl24[i].valid_group == 0 &&
969                                         i_lpm->lpm.tbl24[i].depth <= depth) {
970                                 __atomic_store(&i_lpm->lpm.tbl24[i], &new_tbl24_entry,
971                                                 __ATOMIC_RELEASE);
972                         } else  if (i_lpm->lpm.tbl24[i].valid_group == 1) {
973                                 /*
974                                  * If TBL24 entry is extended, then there has
975                                  * to be a rule with depth >= 25 in the
976                                  * associated TBL8 group.
977                                  */
978
979                                 tbl8_group_index = i_lpm->lpm.tbl24[i].group_idx;
980                                 tbl8_index = tbl8_group_index *
981                                                 RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
982
983                                 for (j = tbl8_index; j < (tbl8_index +
984                                         RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
985
986                                         if (i_lpm->lpm.tbl8[j].depth <= depth)
987                                                 __atomic_store(&i_lpm->lpm.tbl8[j],
988                                                         &new_tbl8_entry,
989                                                         __ATOMIC_RELAXED);
990                                 }
991                         }
992                 }
993         }
994 #undef group_idx
995         return 0;
996 }
997
998 /*
999  * Checks if table 8 group can be recycled.
1000  *
1001  * Return of -EEXIST means tbl8 is in use and thus can not be recycled.
1002  * Return of -EINVAL means tbl8 is empty and thus can be recycled
1003  * Return of value > -1 means tbl8 is in use but has all the same values and
1004  * thus can be recycled
1005  */
1006 static int32_t
1007 tbl8_recycle_check(struct rte_lpm_tbl_entry *tbl8,
1008                 uint32_t tbl8_group_start)
1009 {
1010         uint32_t tbl8_group_end, i;
1011         tbl8_group_end = tbl8_group_start + RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
1012
1013         /*
1014          * Check the first entry of the given tbl8. If it is invalid we know
1015          * this tbl8 does not contain any rule with a depth < RTE_LPM_MAX_DEPTH
1016          *  (As they would affect all entries in a tbl8) and thus this table
1017          *  can not be recycled.
1018          */
1019         if (tbl8[tbl8_group_start].valid) {
1020                 /*
1021                  * If first entry is valid check if the depth is less than 24
1022                  * and if so check the rest of the entries to verify that they
1023                  * are all of this depth.
1024                  */
1025                 if (tbl8[tbl8_group_start].depth <= MAX_DEPTH_TBL24) {
1026                         for (i = (tbl8_group_start + 1); i < tbl8_group_end;
1027                                         i++) {
1028
1029                                 if (tbl8[i].depth !=
1030                                                 tbl8[tbl8_group_start].depth) {
1031
1032                                         return -EEXIST;
1033                                 }
1034                         }
1035                         /* If all entries are the same return the tb8 index */
1036                         return tbl8_group_start;
1037                 }
1038
1039                 return -EEXIST;
1040         }
1041         /*
1042          * If the first entry is invalid check if the rest of the entries in
1043          * the tbl8 are invalid.
1044          */
1045         for (i = (tbl8_group_start + 1); i < tbl8_group_end; i++) {
1046                 if (tbl8[i].valid)
1047                         return -EEXIST;
1048         }
1049         /* If no valid entries are found then return -EINVAL. */
1050         return -EINVAL;
1051 }
1052
1053 static int32_t
1054 delete_depth_big(struct __rte_lpm *i_lpm, uint32_t ip_masked,
1055         uint8_t depth, int32_t sub_rule_index, uint8_t sub_rule_depth)
1056 {
1057 #define group_idx next_hop
1058         uint32_t tbl24_index, tbl8_group_index, tbl8_group_start, tbl8_index,
1059                         tbl8_range, i;
1060         int32_t tbl8_recycle_index, status = 0;
1061
1062         /*
1063          * Calculate the index into tbl24 and range. Note: All depths larger
1064          * than MAX_DEPTH_TBL24 are associated with only one tbl24 entry.
1065          */
1066         tbl24_index = ip_masked >> 8;
1067
1068         /* Calculate the index into tbl8 and range. */
1069         tbl8_group_index = i_lpm->lpm.tbl24[tbl24_index].group_idx;
1070         tbl8_group_start = tbl8_group_index * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
1071         tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
1072         tbl8_range = depth_to_range(depth);
1073
1074         if (sub_rule_index < 0) {
1075                 /*
1076                  * Loop through the range of entries on tbl8 for which the
1077                  * rule_to_delete must be removed or modified.
1078                  */
1079                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
1080                         if (i_lpm->lpm.tbl8[i].depth <= depth)
1081                                 i_lpm->lpm.tbl8[i].valid = INVALID;
1082                 }
1083         } else {
1084                 /* Set new tbl8 entry. */
1085                 struct rte_lpm_tbl_entry new_tbl8_entry = {
1086                         .valid = VALID,
1087                         .depth = sub_rule_depth,
1088                         .valid_group = i_lpm->lpm.tbl8[tbl8_group_start].valid_group,
1089                         .next_hop = i_lpm->rules_tbl[sub_rule_index].next_hop,
1090                 };
1091
1092                 /*
1093                  * Loop through the range of entries on tbl8 for which the
1094                  * rule_to_delete must be modified.
1095                  */
1096                 for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
1097                         if (i_lpm->lpm.tbl8[i].depth <= depth)
1098                                 __atomic_store(&i_lpm->lpm.tbl8[i], &new_tbl8_entry,
1099                                                 __ATOMIC_RELAXED);
1100                 }
1101         }
1102
1103         /*
1104          * Check if there are any valid entries in this tbl8 group. If all
1105          * tbl8 entries are invalid we can free the tbl8 and invalidate the
1106          * associated tbl24 entry.
1107          */
1108
1109         tbl8_recycle_index = tbl8_recycle_check(i_lpm->lpm.tbl8, tbl8_group_start);
1110
1111         if (tbl8_recycle_index == -EINVAL) {
1112                 /* Set tbl24 before freeing tbl8 to avoid race condition.
1113                  * Prevent the free of the tbl8 group from hoisting.
1114                  */
1115                 i_lpm->lpm.tbl24[tbl24_index].valid = 0;
1116                 __atomic_thread_fence(__ATOMIC_RELEASE);
1117                 status = tbl8_free(i_lpm, tbl8_group_start);
1118         } else if (tbl8_recycle_index > -1) {
1119                 /* Update tbl24 entry. */
1120                 struct rte_lpm_tbl_entry new_tbl24_entry = {
1121                         .next_hop = i_lpm->lpm.tbl8[tbl8_recycle_index].next_hop,
1122                         .valid = VALID,
1123                         .valid_group = 0,
1124                         .depth = i_lpm->lpm.tbl8[tbl8_recycle_index].depth,
1125                 };
1126
1127                 /* Set tbl24 before freeing tbl8 to avoid race condition.
1128                  * Prevent the free of the tbl8 group from hoisting.
1129                  */
1130                 __atomic_store(&i_lpm->lpm.tbl24[tbl24_index], &new_tbl24_entry,
1131                                 __ATOMIC_RELAXED);
1132                 __atomic_thread_fence(__ATOMIC_RELEASE);
1133                 status = tbl8_free(i_lpm, tbl8_group_start);
1134         }
1135 #undef group_idx
1136         return status;
1137 }
1138
1139 /*
1140  * Deletes a rule
1141  */
1142 int
1143 rte_lpm_delete(struct rte_lpm *lpm, uint32_t ip, uint8_t depth)
1144 {
1145         int32_t rule_to_delete_index, sub_rule_index;
1146         struct __rte_lpm *i_lpm;
1147         uint32_t ip_masked;
1148         uint8_t sub_rule_depth;
1149         /*
1150          * Check input arguments. Note: IP must be a positive integer of 32
1151          * bits in length therefore it need not be checked.
1152          */
1153         if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM_MAX_DEPTH)) {
1154                 return -EINVAL;
1155         }
1156
1157         i_lpm = container_of(lpm, struct __rte_lpm, lpm);
1158         ip_masked = ip & depth_to_mask(depth);
1159
1160         /*
1161          * Find the index of the input rule, that needs to be deleted, in the
1162          * rule table.
1163          */
1164         rule_to_delete_index = rule_find(i_lpm, ip_masked, depth);
1165
1166         /*
1167          * Check if rule_to_delete_index was found. If no rule was found the
1168          * function rule_find returns -EINVAL.
1169          */
1170         if (rule_to_delete_index < 0)
1171                 return -EINVAL;
1172
1173         /* Delete the rule from the rule table. */
1174         rule_delete(i_lpm, rule_to_delete_index, depth);
1175
1176         /*
1177          * Find rule to replace the rule_to_delete. If there is no rule to
1178          * replace the rule_to_delete we return -1 and invalidate the table
1179          * entries associated with this rule.
1180          */
1181         sub_rule_depth = 0;
1182         sub_rule_index = find_previous_rule(i_lpm, ip, depth, &sub_rule_depth);
1183
1184         /*
1185          * If the input depth value is less than 25 use function
1186          * delete_depth_small otherwise use delete_depth_big.
1187          */
1188         if (depth <= MAX_DEPTH_TBL24) {
1189                 return delete_depth_small(i_lpm, ip_masked, depth,
1190                                 sub_rule_index, sub_rule_depth);
1191         } else { /* If depth > MAX_DEPTH_TBL24 */
1192                 return delete_depth_big(i_lpm, ip_masked, depth, sub_rule_index,
1193                                 sub_rule_depth);
1194         }
1195 }
1196
1197 /*
1198  * Delete all rules from the LPM table.
1199  */
1200 void
1201 rte_lpm_delete_all(struct rte_lpm *lpm)
1202 {
1203         struct __rte_lpm *i_lpm;
1204
1205         i_lpm = container_of(lpm, struct __rte_lpm, lpm);
1206         /* Zero rule information. */
1207         memset(i_lpm->rule_info, 0, sizeof(i_lpm->rule_info));
1208
1209         /* Zero tbl24. */
1210         memset(i_lpm->lpm.tbl24, 0, sizeof(i_lpm->lpm.tbl24));
1211
1212         /* Zero tbl8. */
1213         memset(i_lpm->lpm.tbl8, 0, sizeof(i_lpm->lpm.tbl8[0])
1214                         * RTE_LPM_TBL8_GROUP_NUM_ENTRIES * i_lpm->number_tbl8s);
1215
1216         /* Delete all rules form the rules table. */
1217         memset(i_lpm->rules_tbl, 0, sizeof(i_lpm->rules_tbl[0]) * i_lpm->max_rules);
1218 }