xref: /freebsd/sys/contrib/dpdk_rte_lpm/rte_lpm.c (revision 537d1343)
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))
depth_to_mask(uint8_t depth)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))
depth_to_range(uint8_t depth)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 *
rte_lpm_create(const char * name,int socket_id,const struct rte_lpm_config * config)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
rte_lpm_free(struct rte_lpm * lpm)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
tbl8_alloc(struct rte_lpm_tbl_entry * tbl8,uint32_t number_tbl8s)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
tbl8_free(struct rte_lpm_tbl_entry * tbl8,uint32_t tbl8_group_start)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
add_depth_small(struct rte_lpm * lpm,uint32_t ip,uint8_t depth,uint32_t next_hop)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
add_depth_big(struct rte_lpm * lpm,uint32_t ip_masked,uint8_t depth,uint32_t next_hop)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
rte_lpm_add(struct rte_lpm * lpm,uint32_t ip,uint8_t depth,uint32_t next_hop)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
delete_depth_small(struct rte_lpm * lpm,uint32_t ip_masked,uint8_t depth,uint32_t sub_rule_nhop,uint8_t sub_rule_depth)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
tbl8_recycle_check(struct rte_lpm_tbl_entry * tbl8,uint32_t tbl8_group_start)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
delete_depth_big(struct rte_lpm * lpm,uint32_t ip_masked,uint8_t depth,uint32_t sub_rule_nhop,uint8_t sub_rule_depth)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
rte_lpm_delete(struct rte_lpm * lpm,uint32_t ip,uint8_t depth,uint8_t sub_rule_depth,uint32_t sub_rule_nhop)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
rte_lpm_delete_all(struct rte_lpm * lpm)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 }
1108