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 uint32_t per_core_cache;
214 * Cache entry number per core for performance. Should not be
215 * set with release_mem_en.
217 const char *type; /* Memory allocate type name. */
218 void *(*malloc)(uint32_t flags, size_t size, unsigned int align,
220 /* User defined memory allocator. */
221 void (*free)(void *addr); /* User defined memory release. */
224 struct mlx5_indexed_trunk {
225 uint32_t idx; /* Trunk id. */
226 uint32_t prev; /* Previous free trunk in free list. */
227 uint32_t next; /* Next free trunk in free list. */
228 uint32_t free; /* Free entries available */
229 struct rte_bitmap *bmp;
230 uint8_t data[] __rte_cache_aligned; /* Entry data start. */
233 struct mlx5_indexed_cache {
234 struct mlx5_indexed_trunk **trunks;
235 volatile uint32_t n_trunk_valid; /* Trunks allocated. */
236 uint32_t n_trunk; /* Trunk pointer array size. */
242 struct mlx5_ipool_per_lcore {
243 struct mlx5_indexed_cache *lc;
244 uint32_t len; /**< Current cache count. */
245 uint32_t idx[]; /**< Cache objects. */
248 struct mlx5_indexed_pool {
249 struct mlx5_indexed_pool_config cfg; /* Indexed pool configuration. */
250 rte_spinlock_t rsz_lock; /* Pool lock for multiple thread usage. */
251 rte_spinlock_t lcore_lock;
252 /* Dim of trunk pointer array. */
255 uint32_t n_trunk_valid; /* Trunks allocated. */
256 uint32_t n_trunk; /* Trunk pointer array size. */
257 struct mlx5_indexed_trunk **trunks;
258 uint32_t free_list; /* Index to first free trunk. */
261 struct mlx5_indexed_cache *gc;
263 struct mlx5_ipool_per_lcore *cache[RTE_MAX_LCORE + 1];
265 struct rte_bitmap *ibmp;
267 /* Allocate objects bitmap. Use during flush. */
273 uint32_t trunk_avail;
274 uint32_t trunk_empty;
277 uint32_t grow_tbl[]; /* Save the index offset for the grow trunks. */
281 * Return logarithm of the nearest power of two above input value.
287 * Logarithm of the nearest power of two above input value.
289 static inline unsigned int
290 log2above(unsigned int v)
295 for (l = 0, r = 0; (v >> 1); ++l, v >>= 1)
300 /********************************* indexed pool *************************/
303 * This function allocates non-initialized memory entry from pool.
304 * In NUMA systems, the memory entry allocated resides on the same
305 * NUMA socket as the core that calls this function.
307 * Memory entry is allocated from memory trunk, no alignment.
310 * Pointer to indexed memory entry pool.
311 * No initialization required.
313 * Pointer to memory to save allocated index.
314 * Memory index always positive value.
316 * - Pointer to the allocated memory entry.
317 * - NULL on error. Not enough memory, or invalid arguments.
319 void *mlx5_ipool_malloc(struct mlx5_indexed_pool *pool, uint32_t *idx);
322 * This function allocates zero initialized memory entry from pool.
323 * In NUMA systems, the memory entry allocated resides on the same
324 * NUMA socket as the core that calls this function.
326 * Memory entry is allocated from memory trunk, no alignment.
329 * Pointer to indexed memory pool.
330 * No initialization required.
332 * Pointer to memory to save allocated index.
333 * Memory index always positive value.
335 * - Pointer to the allocated memory entry .
336 * - NULL on error. Not enough memory, or invalid arguments.
338 void *mlx5_ipool_zmalloc(struct mlx5_indexed_pool *pool, uint32_t *idx);
341 * This function frees indexed memory entry to pool.
342 * Caller has to make sure that the index is allocated from same pool.
345 * Pointer to indexed memory pool.
347 * Allocated memory entry index.
349 void mlx5_ipool_free(struct mlx5_indexed_pool *pool, uint32_t idx);
352 * This function returns pointer of indexed memory entry from index.
353 * Caller has to make sure that the index is valid, and allocated
357 * Pointer to indexed memory pool.
359 * Allocated memory index.
361 * - Pointer to indexed memory entry.
363 void *mlx5_ipool_get(struct mlx5_indexed_pool *pool, uint32_t idx);
366 * This function creates indexed memory pool.
367 * Caller has to configure the configuration accordingly.
370 * Pointer to indexed memory pool.
372 * Allocated memory index.
374 struct mlx5_indexed_pool *
375 mlx5_ipool_create(struct mlx5_indexed_pool_config *cfg);
378 * This function releases all resources of pool.
379 * Caller has to make sure that all indexes and memories allocated
380 * from this pool not referenced anymore.
383 * Pointer to indexed memory pool.
385 * - non-zero value on error.
388 int mlx5_ipool_destroy(struct mlx5_indexed_pool *pool);
391 * This function dumps debug info of pool.
394 * Pointer to indexed memory pool.
396 void mlx5_ipool_dump(struct mlx5_indexed_pool *pool);
399 * This function flushes all the cache index back to pool trunk.
402 * Pointer to the index memory pool handler.
406 void mlx5_ipool_flush_cache(struct mlx5_indexed_pool *pool);
409 * This function gets the available entry from pos.
412 * Pointer to the index memory pool handler.
414 * Pointer to the index position start from.
417 * - Pointer to the next available entry.
420 void *mlx5_ipool_get_next(struct mlx5_indexed_pool *pool, uint32_t *pos);
423 * This function allocates new empty Three-level table.
426 * The l3t can set as word, double word, quad word or pointer with index.
429 * - Pointer to the allocated l3t.
430 * - NULL on error. Not enough memory, or invalid arguments.
432 struct mlx5_l3t_tbl *mlx5_l3t_create(enum mlx5_l3t_type type);
435 * This function destroys Three-level table.
438 * Pointer to the l3t.
440 void mlx5_l3t_destroy(struct mlx5_l3t_tbl *tbl);
443 * This function gets the index entry from Three-level table.
446 * Pointer to the l3t.
448 * Index to the entry.
450 * Pointer to the memory which saves the entry data.
451 * When function call returns 0, data contains the entry data get from
453 * When function call returns -1, data is not modified.
456 * 0 if success, -1 on error.
459 int32_t mlx5_l3t_get_entry(struct mlx5_l3t_tbl *tbl, uint32_t idx,
460 union mlx5_l3t_data *data);
463 * This function gets the index entry from Three-level table.
465 * If the index entry is not available, allocate new one by callback
466 * function and fill in the entry.
469 * Pointer to the l3t.
471 * Index to the entry.
473 * Pointer to the memory which saves the entry data.
474 * When function call returns 0, data contains the entry data get from
476 * When function call returns -1, data is not modified.
478 * Callback function to allocate new data.
480 * Context for callback function.
483 * 0 if success, -1 on error.
486 int32_t mlx5_l3t_prepare_entry(struct mlx5_l3t_tbl *tbl, uint32_t idx,
487 union mlx5_l3t_data *data,
488 mlx5_l3t_alloc_callback_fn cb, void *ctx);
491 * This function decreases and clear index entry if reference
492 * counter is 0 from Three-level table.
495 * Pointer to the l3t.
497 * Index to the entry.
500 * The remaining reference count, 0 means entry be cleared, -1 on error.
502 int32_t mlx5_l3t_clear_entry(struct mlx5_l3t_tbl *tbl, uint32_t idx);
505 * This function sets the index entry to Three-level table.
506 * If the entry is already set, the EEXIST errno will be given, and
507 * the set data will be filled to the data.
510 * Pointer to the l3t.
512 * Index to the entry.
513 * @param data[in/out]
514 * Pointer to the memory which contains the entry data save to l3t.
515 * If the entry is already set, the set data will be filled.
518 * 0 if success, -1 on error.
520 int32_t mlx5_l3t_set_entry(struct mlx5_l3t_tbl *tbl, uint32_t idx,
521 union mlx5_l3t_data *data);
524 mlx5_l3t_get_next(struct mlx5_l3t_tbl *tbl, uint32_t *pos)
526 struct mlx5_l3t_level_tbl *g_tbl, *m_tbl;
527 uint32_t i, j, k, g_start, m_start, e_start;
530 struct mlx5_l3t_entry_word *w_e_tbl;
531 struct mlx5_l3t_entry_dword *dw_e_tbl;
532 struct mlx5_l3t_entry_qword *qw_e_tbl;
533 struct mlx5_l3t_entry_ptr *ptr_e_tbl;
540 g_start = (idx >> MLX5_L3T_GT_OFFSET) & MLX5_L3T_GT_MASK;
541 m_start = (idx >> MLX5_L3T_MT_OFFSET) & MLX5_L3T_MT_MASK;
542 e_start = idx & MLX5_L3T_ET_MASK;
543 for (i = g_start; i < MLX5_L3T_GT_SIZE; i++) {
544 m_tbl = g_tbl->tbl[i];
546 /* Jump to new table, reset the sub table start. */
551 for (j = m_start; j < MLX5_L3T_MT_SIZE; j++) {
552 if (!m_tbl->tbl[j]) {
554 * Jump to new table, reset the sub table
560 e_tbl = m_tbl->tbl[j];
562 case MLX5_L3T_TYPE_WORD:
563 w_e_tbl = (struct mlx5_l3t_entry_word *)e_tbl;
564 for (k = e_start; k < MLX5_L3T_ET_SIZE; k++) {
565 if (!w_e_tbl->entry[k].data)
567 *pos = (i << MLX5_L3T_GT_OFFSET) |
568 (j << MLX5_L3T_MT_OFFSET) | k;
569 return (void *)&w_e_tbl->entry[k].data;
572 case MLX5_L3T_TYPE_DWORD:
573 dw_e_tbl = (struct mlx5_l3t_entry_dword *)e_tbl;
574 for (k = e_start; k < MLX5_L3T_ET_SIZE; k++) {
575 if (!dw_e_tbl->entry[k].data)
577 *pos = (i << MLX5_L3T_GT_OFFSET) |
578 (j << MLX5_L3T_MT_OFFSET) | k;
579 return (void *)&dw_e_tbl->entry[k].data;
582 case MLX5_L3T_TYPE_QWORD:
583 qw_e_tbl = (struct mlx5_l3t_entry_qword *)e_tbl;
584 for (k = e_start; k < MLX5_L3T_ET_SIZE; k++) {
585 if (!qw_e_tbl->entry[k].data)
587 *pos = (i << MLX5_L3T_GT_OFFSET) |
588 (j << MLX5_L3T_MT_OFFSET) | k;
589 return (void *)&qw_e_tbl->entry[k].data;
593 ptr_e_tbl = (struct mlx5_l3t_entry_ptr *)e_tbl;
594 for (k = e_start; k < MLX5_L3T_ET_SIZE; k++) {
595 if (!ptr_e_tbl->entry[k].data)
597 *pos = (i << MLX5_L3T_GT_OFFSET) |
598 (j << MLX5_L3T_MT_OFFSET) | k;
599 return ptr_e_tbl->entry[k].data;
609 * Macros for linked list based on indexed memory.
610 * Example data structure:
612 * ILIST_ENTRY(uint16_t) next;
617 #define ILIST_ENTRY(type) \
619 type prev; /* Index of previous element. */ \
620 type next; /* Index of next element. */ \
623 #define ILIST_INSERT(pool, head, idx, elem, field) \
626 MLX5_ASSERT((elem) && (idx)); \
627 (elem)->field.next = *(head); \
628 (elem)->field.prev = 0; \
630 (peer) = mlx5_ipool_get(pool, *(head)); \
632 (peer)->field.prev = (idx); \
637 #define ILIST_REMOVE(pool, head, idx, elem, field) \
642 if ((elem)->field.prev) { \
643 (peer) = mlx5_ipool_get \
644 (pool, (elem)->field.prev); \
646 (peer)->field.next = (elem)->field.next;\
648 if ((elem)->field.next) { \
649 (peer) = mlx5_ipool_get \
650 (pool, (elem)->field.next); \
652 (peer)->field.prev = (elem)->field.prev;\
654 if (*(head) == (idx)) \
655 *(head) = (elem)->field.next; \
658 #define ILIST_FOREACH(pool, head, idx, elem, field) \
659 for ((idx) = (head), (elem) = \
660 (idx) ? mlx5_ipool_get(pool, (idx)) : NULL; (elem); \
661 idx = (elem)->field.next, (elem) = \
662 (idx) ? mlx5_ipool_get(pool, idx) : NULL)
664 /* Single index list. */
665 #define SILIST_ENTRY(type) \
667 type next; /* Index of next element. */ \
670 #define SILIST_INSERT(head, idx, elem, field) \
672 MLX5_ASSERT((elem) && (idx)); \
673 (elem)->field.next = *(head); \
677 #define SILIST_FOREACH(pool, head, idx, elem, field) \
678 for ((idx) = (head), (elem) = \
679 (idx) ? mlx5_ipool_get(pool, (idx)) : NULL; (elem); \
680 idx = (elem)->field.next, (elem) = \
681 (idx) ? mlx5_ipool_get(pool, idx) : NULL)
683 #define MLX5_L3T_FOREACH(tbl, idx, entry) \
684 for (idx = 0, (entry) = mlx5_l3t_get_next((tbl), &idx); \
686 idx++, (entry) = mlx5_l3t_get_next((tbl), &idx))
688 #define MLX5_IPOOL_FOREACH(ipool, idx, entry) \
689 for ((idx) = 0, mlx5_ipool_flush_cache((ipool)), \
690 (entry) = mlx5_ipool_get_next((ipool), &idx); \
691 (entry); idx++, (entry) = mlx5_ipool_get_next((ipool), &idx))
693 #endif /* RTE_PMD_MLX5_UTILS_H_ */