1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
3 * Copyright(c) 2020 Arm Limited
11 * RTE Longest Prefix Match (LPM)
15 #include <sys/queue.h>
18 #include <rte_branch_prediction.h>
19 #include <rte_byteorder.h>
20 #include <rte_config.h>
21 #include <rte_memory.h>
22 #include <rte_common.h>
24 #include <rte_rcu_qsbr.h>
30 /** Max number of characters in LPM name. */
31 #define RTE_LPM_NAMESIZE 32
33 /** Maximum depth value possible for IPv4 LPM. */
34 #define RTE_LPM_MAX_DEPTH 32
36 /** @internal Total number of tbl24 entries. */
37 #define RTE_LPM_TBL24_NUM_ENTRIES (1 << 24)
39 /** @internal Number of entries in a tbl8 group. */
40 #define RTE_LPM_TBL8_GROUP_NUM_ENTRIES 256
42 /** @internal Max number of tbl8 groups in the tbl8. */
43 #define RTE_LPM_MAX_TBL8_NUM_GROUPS (1 << 24)
45 /** @internal Total number of tbl8 groups in the tbl8. */
46 #define RTE_LPM_TBL8_NUM_GROUPS 256
48 /** @internal Total number of tbl8 entries. */
49 #define RTE_LPM_TBL8_NUM_ENTRIES (RTE_LPM_TBL8_NUM_GROUPS * \
50 RTE_LPM_TBL8_GROUP_NUM_ENTRIES)
52 /** @internal Macro to enable/disable run-time checks. */
53 #if defined(RTE_LIBRTE_LPM_DEBUG)
54 #define RTE_LPM_RETURN_IF_TRUE(cond, retval) do { \
55 if (cond) return (retval); \
58 #define RTE_LPM_RETURN_IF_TRUE(cond, retval)
61 /** @internal bitmask with valid and valid_group fields set */
62 #define RTE_LPM_VALID_EXT_ENTRY_BITMASK 0x03000000
64 /** Bitmask used to indicate successful lookup */
65 #define RTE_LPM_LOOKUP_SUCCESS 0x01000000
67 /** @internal Default RCU defer queue entries to reclaim in one go. */
68 #define RTE_LPM_RCU_DQ_RECLAIM_MAX 16
70 /** RCU reclamation modes */
71 enum rte_lpm_qsbr_mode {
72 /** Create defer queue for reclaim. */
73 RTE_LPM_QSBR_MODE_DQ = 0,
74 /** Use blocking mode reclaim. No defer queue created. */
75 RTE_LPM_QSBR_MODE_SYNC
78 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
79 /** @internal Tbl24 entry structure. */
81 struct rte_lpm_tbl_entry {
83 * Stores Next hop (tbl8 or tbl24 when valid_group is not set) or
84 * a group index pointing to a tbl8 structure (tbl24 only, when
87 uint32_t next_hop :24;
88 /* Using single uint8_t to store 3 values. */
89 uint32_t valid :1; /**< Validation flag. */
92 * - valid_group == 0: entry stores a next hop
93 * - valid_group == 1: entry stores a group_index pointing to a tbl8
95 * - valid_group indicates whether the current tbl8 is in use or not
97 uint32_t valid_group :1;
98 uint32_t depth :6; /**< Rule depth. */
104 struct rte_lpm_tbl_entry {
106 uint32_t valid_group :1;
108 uint32_t next_hop :24;
114 /** LPM configuration structure. */
115 struct rte_lpm_config {
116 uint32_t max_rules; /**< Max number of rules. */
117 uint32_t number_tbl8s; /**< Number of tbl8s to allocate. */
118 int flags; /**< This field is currently unused. */
121 /** @internal Rule structure. */
122 struct rte_lpm_rule {
123 uint32_t ip; /**< Rule IP address. */
124 uint32_t next_hop; /**< Rule next hop. */
127 /** @internal Contains metadata about the rules table. */
128 struct rte_lpm_rule_info {
129 uint32_t used_rules; /**< Used rules so far. */
130 uint32_t first_rule; /**< Indexes the first rule of a given depth. */
133 /** @internal LPM structure. */
136 char name[RTE_LPM_NAMESIZE]; /**< Name of the lpm. */
137 uint32_t max_rules; /**< Max. balanced rules per lpm. */
138 uint32_t number_tbl8s; /**< Number of tbl8s. */
139 struct rte_lpm_rule_info rule_info[RTE_LPM_MAX_DEPTH]; /**< Rule info table. */
142 struct rte_lpm_tbl_entry tbl24[RTE_LPM_TBL24_NUM_ENTRIES]
143 __rte_cache_aligned; /**< LPM tbl24 table. */
144 struct rte_lpm_tbl_entry *tbl8; /**< LPM tbl8 table. */
145 struct rte_lpm_rule *rules_tbl; /**< LPM rules. */
148 /** LPM RCU QSBR configuration structure. */
149 struct rte_lpm_rcu_config {
150 struct rte_rcu_qsbr *v; /* RCU QSBR variable. */
151 /* Mode of RCU QSBR. RTE_LPM_QSBR_MODE_xxx
152 * '0' for default: create defer queue for reclaim.
154 enum rte_lpm_qsbr_mode mode;
155 uint32_t dq_size; /* RCU defer queue size.
156 * default: lpm->number_tbl8s.
158 uint32_t reclaim_thd; /* Threshold to trigger auto reclaim. */
159 uint32_t reclaim_max; /* Max entries to reclaim in one go.
160 * default: RTE_LPM_RCU_DQ_RECLAIM_MAX.
165 * Create an LPM object.
170 * NUMA socket ID for LPM table memory allocation
172 * Structure containing the configuration
174 * Handle to LPM object on success, NULL otherwise with rte_errno set
175 * to an appropriate values. Possible rte_errno values include:
176 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
177 * - E_RTE_SECONDARY - function was called from a secondary process instance
178 * - EINVAL - invalid parameter passed to function
179 * - ENOSPC - the maximum number of memzones has already been allocated
180 * - EEXIST - a memzone with the same name already exists
181 * - ENOMEM - no appropriate memory area found in which to create memzone
184 rte_lpm_create(const char *name, int socket_id,
185 const struct rte_lpm_config *config);
188 * Find an existing LPM object and return a pointer to it.
191 * Name of the lpm object as passed to rte_lpm_create()
193 * Pointer to lpm object or NULL if object not found with rte_errno
194 * set appropriately. Possible rte_errno values include:
195 * - ENOENT - required entry not available to return.
198 rte_lpm_find_existing(const char *name);
201 * Free an LPM object.
209 rte_lpm_free(struct rte_lpm *lpm);
213 * @b EXPERIMENTAL: this API may change without prior notice
215 * Associate RCU QSBR variable with an LPM object.
218 * the lpm object to add RCU QSBR
220 * RCU QSBR configuration
223 * On error - 1 with error code set in rte_errno.
224 * Possible rte_errno codes are:
225 * - EINVAL - invalid pointer
226 * - EEXIST - already added QSBR
227 * - ENOMEM - memory allocation failure
230 int rte_lpm_rcu_qsbr_add(struct rte_lpm *lpm, struct rte_lpm_rcu_config *cfg);
233 * Add a rule to the LPM table.
238 * IP of the rule to be added to the LPM table
240 * Depth of the rule to be added to the LPM table
242 * Next hop of the rule to be added to the LPM table
244 * 0 on success, negative value otherwise
247 rte_lpm_add(struct rte_lpm *lpm, uint32_t ip, uint8_t depth, uint32_t next_hop);
250 * Check if a rule is present in the LPM table,
251 * and provide its next hop if it is.
256 * IP of the rule to be searched
258 * Depth of the rule to searched
260 * Next hop of the rule (valid only if it is found)
262 * 1 if the rule exists, 0 if it does not, a negative value on failure
265 rte_lpm_is_rule_present(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
269 * Delete a rule from the LPM table.
274 * IP of the rule to be deleted from the LPM table
276 * Depth of the rule to be deleted from the LPM table
278 * 0 on success, negative value otherwise
281 rte_lpm_delete(struct rte_lpm *lpm, uint32_t ip, uint8_t depth);
284 * Delete all rules from the LPM table.
290 rte_lpm_delete_all(struct rte_lpm *lpm);
293 * Lookup an IP into the LPM table.
298 * IP to be looked up in the LPM table
300 * Next hop of the most specific rule found for IP (valid on lookup hit only)
302 * -EINVAL for incorrect arguments, -ENOENT on lookup miss, 0 on lookup hit
305 rte_lpm_lookup(struct rte_lpm *lpm, uint32_t ip, uint32_t *next_hop)
307 unsigned tbl24_index = (ip >> 8);
309 const uint32_t *ptbl;
311 /* DEBUG: Check user input arguments. */
312 RTE_LPM_RETURN_IF_TRUE(((lpm == NULL) || (next_hop == NULL)), -EINVAL);
314 /* Copy tbl24 entry */
315 ptbl = (const uint32_t *)(&lpm->tbl24[tbl24_index]);
318 /* Memory ordering is not required in lookup. Because dataflow
319 * dependency exists, compiler or HW won't be able to re-order
322 /* Copy tbl8 entry (only if needed) */
323 if (unlikely((tbl_entry & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
324 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
326 unsigned tbl8_index = (uint8_t)ip +
327 (((uint32_t)tbl_entry & 0x00FFFFFF) *
328 RTE_LPM_TBL8_GROUP_NUM_ENTRIES);
330 ptbl = (const uint32_t *)&lpm->tbl8[tbl8_index];
334 *next_hop = ((uint32_t)tbl_entry & 0x00FFFFFF);
335 return (tbl_entry & RTE_LPM_LOOKUP_SUCCESS) ? 0 : -ENOENT;
339 * Lookup multiple IP addresses in an LPM table. This may be implemented as a
340 * macro, so the address of the function should not be used.
345 * Array of IPs to be looked up in the LPM table
347 * Next hop of the most specific rule found for IP (valid on lookup hit only).
348 * This is an array of two byte values. The most significant byte in each
349 * value says whether the lookup was successful (bitmask
350 * RTE_LPM_LOOKUP_SUCCESS is set). The least significant byte is the
353 * Number of elements in ips (and next_hops) array to lookup. This should be a
354 * compile time constant, and divisible by 8 for best performance.
356 * -EINVAL for incorrect arguments, otherwise 0
358 #define rte_lpm_lookup_bulk(lpm, ips, next_hops, n) \
359 rte_lpm_lookup_bulk_func(lpm, ips, next_hops, n)
362 rte_lpm_lookup_bulk_func(const struct rte_lpm *lpm, const uint32_t *ips,
363 uint32_t *next_hops, const unsigned n)
366 unsigned tbl24_indexes[n];
367 const uint32_t *ptbl;
369 /* DEBUG: Check user input arguments. */
370 RTE_LPM_RETURN_IF_TRUE(((lpm == NULL) || (ips == NULL) ||
371 (next_hops == NULL)), -EINVAL);
373 for (i = 0; i < n; i++) {
374 tbl24_indexes[i] = ips[i] >> 8;
377 for (i = 0; i < n; i++) {
378 /* Simply copy tbl24 entry to output */
379 ptbl = (const uint32_t *)&lpm->tbl24[tbl24_indexes[i]];
380 next_hops[i] = *ptbl;
382 /* Overwrite output with tbl8 entry if needed */
383 if (unlikely((next_hops[i] & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
384 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
386 unsigned tbl8_index = (uint8_t)ips[i] +
387 (((uint32_t)next_hops[i] & 0x00FFFFFF) *
388 RTE_LPM_TBL8_GROUP_NUM_ENTRIES);
390 ptbl = (const uint32_t *)&lpm->tbl8[tbl8_index];
391 next_hops[i] = *ptbl;
397 /* Mask four results. */
398 #define RTE_LPM_MASKX4_RES UINT64_C(0x00ffffff00ffffff)
401 * Lookup four IP addresses in an LPM table.
406 * Four IPs to be looked up in the LPM table
408 * Next hop of the most specific rule found for IP (valid on lookup hit only).
409 * This is an 4 elements array of two byte values.
410 * If the lookup was successful for the given IP, then least significant byte
411 * of the corresponding element is the actual next hop and the most
412 * significant byte is zero.
413 * If the lookup for the given IP failed, then corresponding element would
414 * contain default value, see description of then next parameter.
416 * Default value to populate into corresponding element of hop[] array,
417 * if lookup would fail.
420 rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
423 #if defined(RTE_ARCH_ARM) || defined(RTE_ARCH_ARM64)
424 #include "rte_lpm_neon.h"
425 #elif defined(RTE_ARCH_PPC_64)
426 #include "rte_lpm_altivec.h"
428 #include "rte_lpm_sse.h"
435 #endif /* _RTE_LPM_H_ */