1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2015 6WIND S.A.
3 * Copyright 2015 Mellanox Technologies, Ltd
6 #ifndef RTE_PMD_MLX5_UTILS_H_
7 #define RTE_PMD_MLX5_UTILS_H_
15 #include <rte_spinlock.h>
16 #include <rte_rwlock.h>
17 #include <rte_memory.h>
18 #include <rte_bitmap.h>
20 #include <mlx5_common.h>
21 #include <mlx5_common_utils.h>
23 #include "mlx5_defs.h"
25 /* Convert a bit number to the corresponding 64-bit mask */
26 #define MLX5_BITSHIFT(v) (UINT64_C(1) << (v))
28 /* Save and restore errno around argument evaluation. */
29 #define ERRNO_SAFE(x) ((errno = (int []){ errno, ((x), 0) }[0]))
31 extern int mlx5_logtype;
33 #define MLX5_NET_LOG_PREFIX "mlx5_net"
35 /* Generic printf()-like logging macro with automatic line feed. */
36 #define DRV_LOG(level, ...) \
37 PMD_DRV_LOG_(level, mlx5_logtype, MLX5_NET_LOG_PREFIX, \
38 __VA_ARGS__ PMD_DRV_LOG_STRIP PMD_DRV_LOG_OPAREN, \
41 /* Convenience macros for accessing mbuf fields. */
42 #define NEXT(m) ((m)->next)
43 #define DATA_LEN(m) ((m)->data_len)
44 #define PKT_LEN(m) ((m)->pkt_len)
45 #define DATA_OFF(m) ((m)->data_off)
46 #define SET_DATA_OFF(m, o) ((m)->data_off = (o))
47 #define NB_SEGS(m) ((m)->nb_segs)
48 #define PORT(m) ((m)->port)
50 /* Transpose flags. Useful to convert IBV to DPDK flags. */
51 #define TRANSPOSE(val, from, to) \
53 (((val) & (from)) / ((from) / (to))) : \
54 (((val) & (from)) * ((to) / (from))))
57 * For the case which data is linked with sequence increased index, the
58 * array table will be more efficiect than hash table once need to serarch
59 * one data entry in large numbers of entries. Since the traditional hash
60 * tables has fixed table size, when huge numbers of data saved to the hash
61 * table, it also comes lots of hash conflict.
63 * But simple array table also has fixed size, allocates all the needed
64 * memory at once will waste lots of memory. For the case don't know the
65 * exactly number of entries will be impossible to allocate the array.
67 * Then the multiple level table helps to balance the two disadvantages.
68 * Allocate a global high level table with sub table entries at first,
69 * the global table contains the sub table entries, and the sub table will
70 * be allocated only once the corresponding index entry need to be saved.
71 * e.g. for up to 32-bits index, three level table with 10-10-12 splitting,
72 * with sequence increased index, the memory grows with every 4K entries.
74 * The currently implementation introduces 10-10-12 32-bits splitting
75 * Three-Level table to help the cases which have millions of enties to
76 * save. The index entries can be addressed directly by the index, no
77 * search will be needed.q
80 /* L3 table global table define. */
81 #define MLX5_L3T_GT_OFFSET 22
82 #define MLX5_L3T_GT_SIZE (1 << 10)
83 #define MLX5_L3T_GT_MASK (MLX5_L3T_GT_SIZE - 1)
85 /* L3 table middle table define. */
86 #define MLX5_L3T_MT_OFFSET 12
87 #define MLX5_L3T_MT_SIZE (1 << 10)
88 #define MLX5_L3T_MT_MASK (MLX5_L3T_MT_SIZE - 1)
90 /* L3 table entry table define. */
91 #define MLX5_L3T_ET_OFFSET 0
92 #define MLX5_L3T_ET_SIZE (1 << 12)
93 #define MLX5_L3T_ET_MASK (MLX5_L3T_ET_SIZE - 1)
97 MLX5_L3T_TYPE_WORD = 0,
104 struct mlx5_indexed_pool;
106 /* Generic data struct. */
107 union mlx5_l3t_data {
114 /* L3 level table data structure. */
115 struct mlx5_l3t_level_tbl {
116 uint64_t ref_cnt; /* Table ref_cnt. */
117 void *tbl[]; /* Table array. */
120 /* L3 word entry table data structure. */
121 struct mlx5_l3t_entry_word {
122 uint32_t idx; /* Table index. */
123 uint64_t ref_cnt; /* Table ref_cnt. */
127 } entry[MLX5_L3T_ET_SIZE]; /* Entry array */
130 /* L3 double word entry table data structure. */
131 struct mlx5_l3t_entry_dword {
132 uint32_t idx; /* Table index. */
133 uint64_t ref_cnt; /* Table ref_cnt. */
137 } entry[MLX5_L3T_ET_SIZE]; /* Entry array */
140 /* L3 quad word entry table data structure. */
141 struct mlx5_l3t_entry_qword {
142 uint32_t idx; /* Table index. */
143 uint64_t ref_cnt; /* Table ref_cnt. */
147 } entry[MLX5_L3T_ET_SIZE]; /* Entry array */
150 /* L3 pointer entry table data structure. */
151 struct mlx5_l3t_entry_ptr {
152 uint32_t idx; /* Table index. */
153 uint64_t ref_cnt; /* Table ref_cnt. */
157 } entry[MLX5_L3T_ET_SIZE]; /* Entry array */
160 /* L3 table data structure. */
161 struct mlx5_l3t_tbl {
162 enum mlx5_l3t_type type; /* Table type. */
163 struct mlx5_indexed_pool *eip;
164 /* Table index pool handles. */
165 struct mlx5_l3t_level_tbl *tbl; /* Global table index. */
166 rte_spinlock_t sl; /* The table lock. */
169 /** Type of function that is used to handle the data before freeing. */
170 typedef int32_t (*mlx5_l3t_alloc_callback_fn)(void *ctx,
171 union mlx5_l3t_data *data);
174 * The indexed memory entry index is made up of trunk index and offset of
175 * the entry in the trunk. Since the entry index is 32 bits, in case user
176 * prefers to have small trunks, user can change the macro below to a big
177 * number which helps the pool contains more trunks with lots of entries
180 #define TRUNK_IDX_BITS 16
181 #define TRUNK_MAX_IDX ((1 << TRUNK_IDX_BITS) - 1)
182 #define TRUNK_INVALID TRUNK_MAX_IDX
183 #define MLX5_IPOOL_DEFAULT_TRUNK_SIZE (1 << (28 - TRUNK_IDX_BITS))
184 #ifdef RTE_LIBRTE_MLX5_DEBUG
188 struct mlx5_indexed_pool_config {
189 uint32_t size; /* Pool entry size. */
190 uint32_t trunk_size:22;
192 * Trunk entry number. Must be power of 2. It can be increased
193 * if trunk_grow enable. The trunk entry number increases with
194 * left shift grow_shift. Trunks with index are after grow_trunk
195 * will keep the entry number same with the last grow trunk.
197 uint32_t grow_trunk:4;
199 * Trunks with entry number increase in the pool. Set it to 0
200 * to make the pool works as trunk entry fixed pool. It works
201 * only if grow_shift is not 0.
203 uint32_t grow_shift:4;
205 * Trunk entry number increase shift value, stop after grow_trunk.
206 * It works only if grow_trunk is not 0.
208 uint32_t need_lock:1;
209 /* Lock is needed for multiple thread usage. */
210 uint32_t release_mem_en:1; /* Rlease trunk when it is free. */
211 uint32_t max_idx; /* The maximum index can be allocated. */
212 const char *type; /* Memory allocate type name. */
213 void *(*malloc)(uint32_t flags, size_t size, unsigned int align,
215 /* User defined memory allocator. */
216 void (*free)(void *addr); /* User defined memory release. */
219 struct mlx5_indexed_trunk {
220 uint32_t idx; /* Trunk id. */
221 uint32_t prev; /* Previous free trunk in free list. */
222 uint32_t next; /* Next free trunk in free list. */
223 uint32_t free; /* Free entries available */
224 struct rte_bitmap *bmp;
225 uint8_t data[] __rte_cache_aligned; /* Entry data start. */
228 struct mlx5_indexed_pool {
229 struct mlx5_indexed_pool_config cfg; /* Indexed pool configuration. */
230 rte_spinlock_t lock; /* Pool lock for multiple thread usage. */
231 uint32_t n_trunk_valid; /* Trunks allocated. */
232 uint32_t n_trunk; /* Trunk pointer array size. */
233 /* Dim of trunk pointer array. */
234 struct mlx5_indexed_trunk **trunks;
235 uint32_t free_list; /* Index to first free trunk. */
239 uint32_t trunk_avail;
240 uint32_t trunk_empty;
243 uint32_t grow_tbl[]; /* Save the index offset for the grow trunks. */
247 * Return logarithm of the nearest power of two above input value.
253 * Logarithm of the nearest power of two above input value.
255 static inline unsigned int
256 log2above(unsigned int v)
261 for (l = 0, r = 0; (v >> 1); ++l, v >>= 1)
266 /************************ cache list *****************************/
268 /** Maximum size of string for naming. */
269 #define MLX5_NAME_SIZE 32
271 struct mlx5_cache_list;
274 * Structure of the entry in the cache list, user should define its own struct
275 * that contains this in order to store the data.
277 struct mlx5_cache_entry {
278 LIST_ENTRY(mlx5_cache_entry) next; /* Entry pointers in the list. */
279 uint32_t ref_cnt; /* Reference count. */
283 * Type of callback function for entry removal.
288 * The entry in the list.
290 typedef void (*mlx5_cache_remove_cb)(struct mlx5_cache_list *list,
291 struct mlx5_cache_entry *entry);
294 * Type of function for user defined matching.
299 * The entry in the list.
301 * The pointer to new entry context.
304 * 0 if matching, non-zero number otherwise.
306 typedef int (*mlx5_cache_match_cb)(struct mlx5_cache_list *list,
307 struct mlx5_cache_entry *entry, void *ctx);
310 * Type of function for user defined cache list entry creation.
315 * The new allocated entry, NULL if list entry size unspecified,
316 * New entry has to be allocated in callback and return.
318 * The pointer to new entry context.
321 * Pointer of entry on success, NULL otherwise.
323 typedef struct mlx5_cache_entry *(*mlx5_cache_create_cb)
324 (struct mlx5_cache_list *list,
325 struct mlx5_cache_entry *entry,
329 * Linked cache list structure.
331 * Entry in cache list could be reused if entry already exists,
332 * reference count will increase and the existing entry returns.
334 * When destroy an entry from list, decrease reference count and only
335 * destroy when no further reference.
337 * Linked list cache is designed for limited number of entries cache,
338 * read mostly, less modification.
340 * For huge amount of entries cache, please consider hash list cache.
343 struct mlx5_cache_list {
344 char name[MLX5_NAME_SIZE]; /**< Name of the cache list. */
345 uint32_t entry_sz; /**< Entry size, 0: use create callback. */
346 rte_rwlock_t lock; /* read/write lock. */
347 uint32_t gen_cnt; /* List modification will update generation count. */
348 uint32_t count; /* number of entries in list. */
349 void *ctx; /* user objects target to callback. */
350 mlx5_cache_create_cb cb_create; /**< entry create callback. */
351 mlx5_cache_match_cb cb_match; /**< entry match callback. */
352 mlx5_cache_remove_cb cb_remove; /**< entry remove callback. */
353 LIST_HEAD(mlx5_cache_head, mlx5_cache_entry) head;
357 * Initialize a cache list.
360 * Pointer to the hast list table.
362 * Name of the cache list.
364 * Entry size to allocate, 0 to allocate by creation callback.
366 * Pointer to the list context data.
368 * Callback function for entry create.
370 * Callback function for entry match.
372 * Callback function for entry remove.
374 * 0 on success, otherwise failure.
376 int mlx5_cache_list_init(struct mlx5_cache_list *list,
377 const char *name, uint32_t entry_size, void *ctx,
378 mlx5_cache_create_cb cb_create,
379 mlx5_cache_match_cb cb_match,
380 mlx5_cache_remove_cb cb_remove);
383 * Search an entry matching the key.
385 * Result returned might be destroyed by other thread, must use
386 * this function only in main thread.
389 * Pointer to the cache list.
391 * Common context parameter used by entry callback function.
394 * Pointer of the cache entry if found, NULL otherwise.
396 struct mlx5_cache_entry *mlx5_cache_lookup(struct mlx5_cache_list *list,
400 * Reuse or create an entry to the cache list.
403 * Pointer to the hast list table.
405 * Common context parameter used by callback function.
408 * registered entry on success, NULL otherwise
410 struct mlx5_cache_entry *mlx5_cache_register(struct mlx5_cache_list *list,
414 * Remove an entry from the cache list.
416 * User should guarantee the validity of the entry.
419 * Pointer to the hast list.
421 * Entry to be removed from the cache list table.
423 * 0 on entry removed, 1 on entry still referenced.
425 int mlx5_cache_unregister(struct mlx5_cache_list *list,
426 struct mlx5_cache_entry *entry);
429 * Destroy the cache list.
432 * Pointer to the cache list.
434 void mlx5_cache_list_destroy(struct mlx5_cache_list *list);
437 * Get entry number from the cache list.
440 * Pointer to the hast list.
442 * Cache list entry number.
445 mlx5_cache_list_get_entry_num(struct mlx5_cache_list *list);
447 /********************************* indexed pool *************************/
450 * This function allocates non-initialized memory entry from pool.
451 * In NUMA systems, the memory entry allocated resides on the same
452 * NUMA socket as the core that calls this function.
454 * Memory entry is allocated from memory trunk, no alignment.
457 * Pointer to indexed memory entry pool.
458 * No initialization required.
460 * Pointer to memory to save allocated index.
461 * Memory index always positive value.
463 * - Pointer to the allocated memory entry.
464 * - NULL on error. Not enough memory, or invalid arguments.
466 void *mlx5_ipool_malloc(struct mlx5_indexed_pool *pool, uint32_t *idx);
469 * This function allocates zero initialized memory entry from pool.
470 * In NUMA systems, the memory entry allocated resides on the same
471 * NUMA socket as the core that calls this function.
473 * Memory entry is allocated from memory trunk, no alignment.
476 * Pointer to indexed memory pool.
477 * No initialization required.
479 * Pointer to memory to save allocated index.
480 * Memory index always positive value.
482 * - Pointer to the allocated memory entry .
483 * - NULL on error. Not enough memory, or invalid arguments.
485 void *mlx5_ipool_zmalloc(struct mlx5_indexed_pool *pool, uint32_t *idx);
488 * This function frees indexed memory entry to pool.
489 * Caller has to make sure that the index is allocated from same pool.
492 * Pointer to indexed memory pool.
494 * Allocated memory entry index.
496 void mlx5_ipool_free(struct mlx5_indexed_pool *pool, uint32_t idx);
499 * This function returns pointer of indexed memory entry from index.
500 * Caller has to make sure that the index is valid, and allocated
504 * Pointer to indexed memory pool.
506 * Allocated memory index.
508 * - Pointer to indexed memory entry.
510 void *mlx5_ipool_get(struct mlx5_indexed_pool *pool, uint32_t idx);
513 * This function creates indexed memory pool.
514 * Caller has to configure the configuration accordingly.
517 * Pointer to indexed memory pool.
519 * Allocated memory index.
521 struct mlx5_indexed_pool *
522 mlx5_ipool_create(struct mlx5_indexed_pool_config *cfg);
525 * This function releases all resources of pool.
526 * Caller has to make sure that all indexes and memories allocated
527 * from this pool not referenced anymore.
530 * Pointer to indexed memory pool.
532 * - non-zero value on error.
535 int mlx5_ipool_destroy(struct mlx5_indexed_pool *pool);
538 * This function dumps debug info of pool.
541 * Pointer to indexed memory pool.
543 void mlx5_ipool_dump(struct mlx5_indexed_pool *pool);
546 * This function allocates new empty Three-level table.
549 * The l3t can set as word, double word, quad word or pointer with index.
552 * - Pointer to the allocated l3t.
553 * - NULL on error. Not enough memory, or invalid arguments.
555 struct mlx5_l3t_tbl *mlx5_l3t_create(enum mlx5_l3t_type type);
558 * This function destroys Three-level table.
561 * Pointer to the l3t.
563 void mlx5_l3t_destroy(struct mlx5_l3t_tbl *tbl);
566 * This function gets the index entry from Three-level table.
569 * Pointer to the l3t.
571 * Index to the entry.
573 * Pointer to the memory which saves the entry data.
574 * When function call returns 0, data contains the entry data get from
576 * When function call returns -1, data is not modified.
579 * 0 if success, -1 on error.
582 int32_t mlx5_l3t_get_entry(struct mlx5_l3t_tbl *tbl, uint32_t idx,
583 union mlx5_l3t_data *data);
586 * This function gets the index entry from Three-level table.
588 * If the index entry is not available, allocate new one by callback
589 * function and fill in the entry.
592 * Pointer to the l3t.
594 * Index to the entry.
596 * Pointer to the memory which saves the entry data.
597 * When function call returns 0, data contains the entry data get from
599 * When function call returns -1, data is not modified.
601 * Callback function to allocate new data.
603 * Context for callback function.
606 * 0 if success, -1 on error.
609 int32_t mlx5_l3t_prepare_entry(struct mlx5_l3t_tbl *tbl, uint32_t idx,
610 union mlx5_l3t_data *data,
611 mlx5_l3t_alloc_callback_fn cb, void *ctx);
614 * This function decreases and clear index entry if reference
615 * counter is 0 from Three-level table.
618 * Pointer to the l3t.
620 * Index to the entry.
623 * The remaining reference count, 0 means entry be cleared, -1 on error.
625 int32_t mlx5_l3t_clear_entry(struct mlx5_l3t_tbl *tbl, uint32_t idx);
628 * This function sets the index entry to Three-level table.
629 * If the entry is already set, the EEXIST errno will be given, and
630 * the set data will be filled to the data.
633 * Pointer to the l3t.
635 * Index to the entry.
636 * @param data[in/out]
637 * Pointer to the memory which contains the entry data save to l3t.
638 * If the entry is already set, the set data will be filled.
641 * 0 if success, -1 on error.
643 int32_t mlx5_l3t_set_entry(struct mlx5_l3t_tbl *tbl, uint32_t idx,
644 union mlx5_l3t_data *data);
647 mlx5_l3t_get_next(struct mlx5_l3t_tbl *tbl, uint32_t *pos)
649 struct mlx5_l3t_level_tbl *g_tbl, *m_tbl;
650 uint32_t i, j, k, g_start, m_start, e_start;
653 struct mlx5_l3t_entry_word *w_e_tbl;
654 struct mlx5_l3t_entry_dword *dw_e_tbl;
655 struct mlx5_l3t_entry_qword *qw_e_tbl;
656 struct mlx5_l3t_entry_ptr *ptr_e_tbl;
663 g_start = (idx >> MLX5_L3T_GT_OFFSET) & MLX5_L3T_GT_MASK;
664 m_start = (idx >> MLX5_L3T_MT_OFFSET) & MLX5_L3T_MT_MASK;
665 e_start = idx & MLX5_L3T_ET_MASK;
666 for (i = g_start; i < MLX5_L3T_GT_SIZE; i++) {
667 m_tbl = g_tbl->tbl[i];
669 /* Jump to new table, reset the sub table start. */
674 for (j = m_start; j < MLX5_L3T_MT_SIZE; j++) {
675 if (!m_tbl->tbl[j]) {
677 * Jump to new table, reset the sub table
683 e_tbl = m_tbl->tbl[j];
685 case MLX5_L3T_TYPE_WORD:
686 w_e_tbl = (struct mlx5_l3t_entry_word *)e_tbl;
687 for (k = e_start; k < MLX5_L3T_ET_SIZE; k++) {
688 if (!w_e_tbl->entry[k].data)
690 *pos = (i << MLX5_L3T_GT_OFFSET) |
691 (j << MLX5_L3T_MT_OFFSET) | k;
692 return (void *)&w_e_tbl->entry[k].data;
695 case MLX5_L3T_TYPE_DWORD:
696 dw_e_tbl = (struct mlx5_l3t_entry_dword *)e_tbl;
697 for (k = e_start; k < MLX5_L3T_ET_SIZE; k++) {
698 if (!dw_e_tbl->entry[k].data)
700 *pos = (i << MLX5_L3T_GT_OFFSET) |
701 (j << MLX5_L3T_MT_OFFSET) | k;
702 return (void *)&dw_e_tbl->entry[k].data;
705 case MLX5_L3T_TYPE_QWORD:
706 qw_e_tbl = (struct mlx5_l3t_entry_qword *)e_tbl;
707 for (k = e_start; k < MLX5_L3T_ET_SIZE; k++) {
708 if (!qw_e_tbl->entry[k].data)
710 *pos = (i << MLX5_L3T_GT_OFFSET) |
711 (j << MLX5_L3T_MT_OFFSET) | k;
712 return (void *)&qw_e_tbl->entry[k].data;
716 ptr_e_tbl = (struct mlx5_l3t_entry_ptr *)e_tbl;
717 for (k = e_start; k < MLX5_L3T_ET_SIZE; k++) {
718 if (!ptr_e_tbl->entry[k].data)
720 *pos = (i << MLX5_L3T_GT_OFFSET) |
721 (j << MLX5_L3T_MT_OFFSET) | k;
722 return ptr_e_tbl->entry[k].data;
732 * Macros for linked list based on indexed memory.
733 * Example data structure:
735 * ILIST_ENTRY(uint16_t) next;
740 #define ILIST_ENTRY(type) \
742 type prev; /* Index of previous element. */ \
743 type next; /* Index of next element. */ \
746 #define ILIST_INSERT(pool, head, idx, elem, field) \
749 MLX5_ASSERT((elem) && (idx)); \
750 (elem)->field.next = *(head); \
751 (elem)->field.prev = 0; \
753 (peer) = mlx5_ipool_get(pool, *(head)); \
755 (peer)->field.prev = (idx); \
760 #define ILIST_REMOVE(pool, head, idx, elem, field) \
765 if ((elem)->field.prev) { \
766 (peer) = mlx5_ipool_get \
767 (pool, (elem)->field.prev); \
769 (peer)->field.next = (elem)->field.next;\
771 if ((elem)->field.next) { \
772 (peer) = mlx5_ipool_get \
773 (pool, (elem)->field.next); \
775 (peer)->field.prev = (elem)->field.prev;\
777 if (*(head) == (idx)) \
778 *(head) = (elem)->field.next; \
781 #define ILIST_FOREACH(pool, head, idx, elem, field) \
782 for ((idx) = (head), (elem) = \
783 (idx) ? mlx5_ipool_get(pool, (idx)) : NULL; (elem); \
784 idx = (elem)->field.next, (elem) = \
785 (idx) ? mlx5_ipool_get(pool, idx) : NULL)
787 /* Single index list. */
788 #define SILIST_ENTRY(type) \
790 type next; /* Index of next element. */ \
793 #define SILIST_INSERT(head, idx, elem, field) \
795 MLX5_ASSERT((elem) && (idx)); \
796 (elem)->field.next = *(head); \
800 #define SILIST_FOREACH(pool, head, idx, elem, field) \
801 for ((idx) = (head), (elem) = \
802 (idx) ? mlx5_ipool_get(pool, (idx)) : NULL; (elem); \
803 idx = (elem)->field.next, (elem) = \
804 (idx) ? mlx5_ipool_get(pool, idx) : NULL)
806 #define MLX5_L3T_FOREACH(tbl, idx, entry) \
807 for (idx = 0, (entry) = mlx5_l3t_get_next((tbl), &idx); \
809 idx++, (entry) = mlx5_l3t_get_next((tbl), &idx))
811 #endif /* RTE_PMD_MLX5_UTILS_H_ */