1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
10 #include <sys/queue.h>
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
25 #include <mlx5_malloc.h>
27 #include "mlx5_defs.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
36 struct tunnel_default_miss_ctx {
40 struct rte_flow_action_rss action_rss;
41 struct rte_flow_action_queue miss_queue;
42 struct rte_flow_action_jump miss_jump;
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 struct rte_flow *flow,
50 const struct rte_flow_attr *attr,
51 const struct rte_flow_action *app_actions,
53 const struct mlx5_flow_tunnel *tunnel,
54 struct tunnel_default_miss_ctx *ctx,
55 struct rte_flow_error *error);
56 static struct mlx5_flow_tunnel *
57 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
59 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
61 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
62 const struct mlx5_flow_tunnel *tunnel,
63 uint32_t group, uint32_t *table,
64 struct rte_flow_error *error);
66 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
67 static void mlx5_flow_pop_thread_workspace(void);
70 /** Device flow drivers. */
71 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
73 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
75 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
76 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
77 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
78 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
80 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
81 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
84 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
85 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
90 /** Node object of input graph for mlx5_flow_expand_rss(). */
91 struct mlx5_flow_expand_node {
92 const int *const next;
94 * List of next node indexes. Index 0 is interpreted as a terminator.
96 const enum rte_flow_item_type type;
97 /**< Pattern item type of current node. */
100 * RSS types bit-field associated with this node
101 * (see ETH_RSS_* definitions).
104 /**< optional expand field. Default 0 to expand, 1 not go deeper. */
107 /** Object returned by mlx5_flow_expand_rss(). */
108 struct mlx5_flow_expand_rss {
110 /**< Number of entries @p patterns and @p priorities. */
112 struct rte_flow_item *pattern; /**< Expanded pattern array. */
113 uint32_t priority; /**< Priority offset for each expansion. */
117 static enum rte_flow_item_type
118 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
120 enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
121 uint16_t ether_type = 0;
122 uint16_t ether_type_m;
123 uint8_t ip_next_proto = 0;
124 uint8_t ip_next_proto_m;
126 if (item == NULL || item->spec == NULL)
128 switch (item->type) {
129 case RTE_FLOW_ITEM_TYPE_ETH:
131 ether_type_m = ((const struct rte_flow_item_eth *)
134 ether_type_m = rte_flow_item_eth_mask.type;
135 if (ether_type_m != RTE_BE16(0xFFFF))
137 ether_type = ((const struct rte_flow_item_eth *)
139 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
140 ret = RTE_FLOW_ITEM_TYPE_IPV4;
141 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
142 ret = RTE_FLOW_ITEM_TYPE_IPV6;
143 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
144 ret = RTE_FLOW_ITEM_TYPE_VLAN;
146 ret = RTE_FLOW_ITEM_TYPE_END;
148 case RTE_FLOW_ITEM_TYPE_VLAN:
150 ether_type_m = ((const struct rte_flow_item_vlan *)
151 (item->mask))->inner_type;
153 ether_type_m = rte_flow_item_vlan_mask.inner_type;
154 if (ether_type_m != RTE_BE16(0xFFFF))
156 ether_type = ((const struct rte_flow_item_vlan *)
157 (item->spec))->inner_type;
158 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
159 ret = RTE_FLOW_ITEM_TYPE_IPV4;
160 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
161 ret = RTE_FLOW_ITEM_TYPE_IPV6;
162 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
163 ret = RTE_FLOW_ITEM_TYPE_VLAN;
165 ret = RTE_FLOW_ITEM_TYPE_END;
167 case RTE_FLOW_ITEM_TYPE_IPV4:
169 ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
170 (item->mask))->hdr.next_proto_id;
173 rte_flow_item_ipv4_mask.hdr.next_proto_id;
174 if (ip_next_proto_m != 0xFF)
176 ip_next_proto = ((const struct rte_flow_item_ipv4 *)
177 (item->spec))->hdr.next_proto_id;
178 if (ip_next_proto == IPPROTO_UDP)
179 ret = RTE_FLOW_ITEM_TYPE_UDP;
180 else if (ip_next_proto == IPPROTO_TCP)
181 ret = RTE_FLOW_ITEM_TYPE_TCP;
182 else if (ip_next_proto == IPPROTO_IP)
183 ret = RTE_FLOW_ITEM_TYPE_IPV4;
184 else if (ip_next_proto == IPPROTO_IPV6)
185 ret = RTE_FLOW_ITEM_TYPE_IPV6;
187 ret = RTE_FLOW_ITEM_TYPE_END;
189 case RTE_FLOW_ITEM_TYPE_IPV6:
191 ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
192 (item->mask))->hdr.proto;
195 rte_flow_item_ipv6_mask.hdr.proto;
196 if (ip_next_proto_m != 0xFF)
198 ip_next_proto = ((const struct rte_flow_item_ipv6 *)
199 (item->spec))->hdr.proto;
200 if (ip_next_proto == IPPROTO_UDP)
201 ret = RTE_FLOW_ITEM_TYPE_UDP;
202 else if (ip_next_proto == IPPROTO_TCP)
203 ret = RTE_FLOW_ITEM_TYPE_TCP;
204 else if (ip_next_proto == IPPROTO_IP)
205 ret = RTE_FLOW_ITEM_TYPE_IPV4;
206 else if (ip_next_proto == IPPROTO_IPV6)
207 ret = RTE_FLOW_ITEM_TYPE_IPV6;
209 ret = RTE_FLOW_ITEM_TYPE_END;
212 ret = RTE_FLOW_ITEM_TYPE_VOID;
218 #define MLX5_RSS_EXP_ELT_N 16
221 * Expand RSS flows into several possible flows according to the RSS hash
222 * fields requested and the driver capabilities.
225 * Buffer to store the result expansion.
227 * Buffer size in bytes. If 0, @p buf can be NULL.
231 * RSS types to expand (see ETH_RSS_* definitions).
233 * Input graph to expand @p pattern according to @p types.
234 * @param[in] graph_root_index
235 * Index of root node in @p graph, typically 0.
238 * A positive value representing the size of @p buf in bytes regardless of
239 * @p size on success, a negative errno value otherwise and rte_errno is
240 * set, the following errors are defined:
242 * -E2BIG: graph-depth @p graph is too deep.
245 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
246 const struct rte_flow_item *pattern, uint64_t types,
247 const struct mlx5_flow_expand_node graph[],
248 int graph_root_index)
250 const struct rte_flow_item *item;
251 const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
252 const int *next_node;
253 const int *stack[MLX5_RSS_EXP_ELT_N];
255 struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
258 size_t user_pattern_size = 0;
260 const struct mlx5_flow_expand_node *next = NULL;
261 struct rte_flow_item missed_item;
264 const struct rte_flow_item *last_item = NULL;
266 memset(&missed_item, 0, sizeof(missed_item));
267 lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
268 MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
270 buf->entry[0].priority = 0;
271 buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
273 addr = buf->entry[0].pattern;
275 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
276 if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
278 for (i = 0; node->next && node->next[i]; ++i) {
279 next = &graph[node->next[i]];
280 if (next->type == item->type)
285 user_pattern_size += sizeof(*item);
287 user_pattern_size += sizeof(*item); /* Handle END item. */
288 lsize += user_pattern_size;
289 /* Copy the user pattern in the first entry of the buffer. */
291 rte_memcpy(addr, pattern, user_pattern_size);
292 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
295 /* Start expanding. */
296 memset(flow_items, 0, sizeof(flow_items));
297 user_pattern_size -= sizeof(*item);
299 * Check if the last valid item has spec set, need complete pattern,
300 * and the pattern can be used for expansion.
302 missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
303 if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
304 /* Item type END indicates expansion is not required. */
307 if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
310 for (i = 0; node->next && node->next[i]; ++i) {
311 next = &graph[node->next[i]];
312 if (next->type == missed_item.type) {
313 flow_items[0].type = missed_item.type;
314 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
320 if (next && missed) {
321 elt = 2; /* missed item + item end. */
323 lsize += elt * sizeof(*item) + user_pattern_size;
324 if ((node->rss_types & types) && lsize <= size) {
325 buf->entry[buf->entries].priority = 1;
326 buf->entry[buf->entries].pattern = addr;
328 rte_memcpy(addr, buf->entry[0].pattern,
330 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
331 rte_memcpy(addr, flow_items, elt * sizeof(*item));
332 addr = (void *)(((uintptr_t)addr) +
333 elt * sizeof(*item));
336 memset(flow_items, 0, sizeof(flow_items));
337 next_node = node->next;
338 stack[stack_pos] = next_node;
339 node = next_node ? &graph[*next_node] : NULL;
341 flow_items[stack_pos].type = node->type;
342 if (node->rss_types & types) {
344 * compute the number of items to copy from the
345 * expansion and copy it.
346 * When the stack_pos is 0, there are 1 element in it,
347 * plus the addition END item.
350 flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
351 lsize += elt * sizeof(*item) + user_pattern_size;
353 size_t n = elt * sizeof(*item);
355 buf->entry[buf->entries].priority =
356 stack_pos + 1 + missed;
357 buf->entry[buf->entries].pattern = addr;
359 rte_memcpy(addr, buf->entry[0].pattern,
361 addr = (void *)(((uintptr_t)addr) +
363 rte_memcpy(addr, &missed_item,
364 missed * sizeof(*item));
365 addr = (void *)(((uintptr_t)addr) +
366 missed * sizeof(*item));
367 rte_memcpy(addr, flow_items, n);
368 addr = (void *)(((uintptr_t)addr) + n);
372 if (!node->optional && node->next) {
373 next_node = node->next;
374 if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
378 stack[stack_pos] = next_node;
379 } else if (*(next_node + 1)) {
380 /* Follow up with the next possibility. */
383 /* Move to the next path. */
385 next_node = stack[--stack_pos];
387 stack[stack_pos] = next_node;
389 node = *next_node ? &graph[*next_node] : NULL;
394 enum mlx5_expansion {
396 MLX5_EXPANSION_ROOT_OUTER,
397 MLX5_EXPANSION_ROOT_ETH_VLAN,
398 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
399 MLX5_EXPANSION_OUTER_ETH,
400 MLX5_EXPANSION_OUTER_ETH_VLAN,
401 MLX5_EXPANSION_OUTER_VLAN,
402 MLX5_EXPANSION_OUTER_IPV4,
403 MLX5_EXPANSION_OUTER_IPV4_UDP,
404 MLX5_EXPANSION_OUTER_IPV4_TCP,
405 MLX5_EXPANSION_OUTER_IPV6,
406 MLX5_EXPANSION_OUTER_IPV6_UDP,
407 MLX5_EXPANSION_OUTER_IPV6_TCP,
408 MLX5_EXPANSION_VXLAN,
409 MLX5_EXPANSION_VXLAN_GPE,
411 MLX5_EXPANSION_NVGRE,
412 MLX5_EXPANSION_GRE_KEY,
415 MLX5_EXPANSION_ETH_VLAN,
418 MLX5_EXPANSION_IPV4_UDP,
419 MLX5_EXPANSION_IPV4_TCP,
421 MLX5_EXPANSION_IPV6_UDP,
422 MLX5_EXPANSION_IPV6_TCP,
425 /** Supported expansion of items. */
426 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
427 [MLX5_EXPANSION_ROOT] = {
428 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
430 MLX5_EXPANSION_IPV6),
431 .type = RTE_FLOW_ITEM_TYPE_END,
433 [MLX5_EXPANSION_ROOT_OUTER] = {
434 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
435 MLX5_EXPANSION_OUTER_IPV4,
436 MLX5_EXPANSION_OUTER_IPV6),
437 .type = RTE_FLOW_ITEM_TYPE_END,
439 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
440 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
441 .type = RTE_FLOW_ITEM_TYPE_END,
443 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
444 .next = MLX5_FLOW_EXPAND_RSS_NEXT
445 (MLX5_EXPANSION_OUTER_ETH_VLAN),
446 .type = RTE_FLOW_ITEM_TYPE_END,
448 [MLX5_EXPANSION_OUTER_ETH] = {
449 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
450 MLX5_EXPANSION_OUTER_IPV6,
451 MLX5_EXPANSION_MPLS),
452 .type = RTE_FLOW_ITEM_TYPE_ETH,
455 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
456 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
457 .type = RTE_FLOW_ITEM_TYPE_ETH,
460 [MLX5_EXPANSION_OUTER_VLAN] = {
461 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
462 MLX5_EXPANSION_OUTER_IPV6),
463 .type = RTE_FLOW_ITEM_TYPE_VLAN,
465 [MLX5_EXPANSION_OUTER_IPV4] = {
466 .next = MLX5_FLOW_EXPAND_RSS_NEXT
467 (MLX5_EXPANSION_OUTER_IPV4_UDP,
468 MLX5_EXPANSION_OUTER_IPV4_TCP,
470 MLX5_EXPANSION_NVGRE,
472 MLX5_EXPANSION_IPV6),
473 .type = RTE_FLOW_ITEM_TYPE_IPV4,
474 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
475 ETH_RSS_NONFRAG_IPV4_OTHER,
477 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
478 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
479 MLX5_EXPANSION_VXLAN_GPE),
480 .type = RTE_FLOW_ITEM_TYPE_UDP,
481 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
483 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
484 .type = RTE_FLOW_ITEM_TYPE_TCP,
485 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
487 [MLX5_EXPANSION_OUTER_IPV6] = {
488 .next = MLX5_FLOW_EXPAND_RSS_NEXT
489 (MLX5_EXPANSION_OUTER_IPV6_UDP,
490 MLX5_EXPANSION_OUTER_IPV6_TCP,
494 MLX5_EXPANSION_NVGRE),
495 .type = RTE_FLOW_ITEM_TYPE_IPV6,
496 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
497 ETH_RSS_NONFRAG_IPV6_OTHER,
499 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
500 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
501 MLX5_EXPANSION_VXLAN_GPE),
502 .type = RTE_FLOW_ITEM_TYPE_UDP,
503 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
505 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
506 .type = RTE_FLOW_ITEM_TYPE_TCP,
507 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
509 [MLX5_EXPANSION_VXLAN] = {
510 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
512 MLX5_EXPANSION_IPV6),
513 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
515 [MLX5_EXPANSION_VXLAN_GPE] = {
516 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
518 MLX5_EXPANSION_IPV6),
519 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
521 [MLX5_EXPANSION_GRE] = {
522 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
524 MLX5_EXPANSION_GRE_KEY),
525 .type = RTE_FLOW_ITEM_TYPE_GRE,
527 [MLX5_EXPANSION_GRE_KEY] = {
528 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
529 MLX5_EXPANSION_IPV6),
530 .type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
533 [MLX5_EXPANSION_NVGRE] = {
534 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
535 .type = RTE_FLOW_ITEM_TYPE_NVGRE,
537 [MLX5_EXPANSION_MPLS] = {
538 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
539 MLX5_EXPANSION_IPV6),
540 .type = RTE_FLOW_ITEM_TYPE_MPLS,
542 [MLX5_EXPANSION_ETH] = {
543 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
544 MLX5_EXPANSION_IPV6),
545 .type = RTE_FLOW_ITEM_TYPE_ETH,
547 [MLX5_EXPANSION_ETH_VLAN] = {
548 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
549 .type = RTE_FLOW_ITEM_TYPE_ETH,
551 [MLX5_EXPANSION_VLAN] = {
552 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
553 MLX5_EXPANSION_IPV6),
554 .type = RTE_FLOW_ITEM_TYPE_VLAN,
556 [MLX5_EXPANSION_IPV4] = {
557 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
558 MLX5_EXPANSION_IPV4_TCP),
559 .type = RTE_FLOW_ITEM_TYPE_IPV4,
560 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
561 ETH_RSS_NONFRAG_IPV4_OTHER,
563 [MLX5_EXPANSION_IPV4_UDP] = {
564 .type = RTE_FLOW_ITEM_TYPE_UDP,
565 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
567 [MLX5_EXPANSION_IPV4_TCP] = {
568 .type = RTE_FLOW_ITEM_TYPE_TCP,
569 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
571 [MLX5_EXPANSION_IPV6] = {
572 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
573 MLX5_EXPANSION_IPV6_TCP),
574 .type = RTE_FLOW_ITEM_TYPE_IPV6,
575 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
576 ETH_RSS_NONFRAG_IPV6_OTHER,
578 [MLX5_EXPANSION_IPV6_UDP] = {
579 .type = RTE_FLOW_ITEM_TYPE_UDP,
580 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
582 [MLX5_EXPANSION_IPV6_TCP] = {
583 .type = RTE_FLOW_ITEM_TYPE_TCP,
584 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
588 static struct rte_flow_action_handle *
589 mlx5_action_handle_create(struct rte_eth_dev *dev,
590 const struct rte_flow_indir_action_conf *conf,
591 const struct rte_flow_action *action,
592 struct rte_flow_error *error);
593 static int mlx5_action_handle_destroy
594 (struct rte_eth_dev *dev,
595 struct rte_flow_action_handle *handle,
596 struct rte_flow_error *error);
597 static int mlx5_action_handle_update
598 (struct rte_eth_dev *dev,
599 struct rte_flow_action_handle *handle,
601 struct rte_flow_error *error);
602 static int mlx5_action_handle_query
603 (struct rte_eth_dev *dev,
604 const struct rte_flow_action_handle *handle,
606 struct rte_flow_error *error);
608 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
609 struct rte_flow_tunnel *app_tunnel,
610 struct rte_flow_action **actions,
611 uint32_t *num_of_actions,
612 struct rte_flow_error *error);
614 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
615 struct rte_flow_tunnel *app_tunnel,
616 struct rte_flow_item **items,
617 uint32_t *num_of_items,
618 struct rte_flow_error *error);
620 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
621 struct rte_flow_item *pmd_items,
622 uint32_t num_items, struct rte_flow_error *err);
624 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
625 struct rte_flow_action *pmd_actions,
626 uint32_t num_actions,
627 struct rte_flow_error *err);
629 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
631 struct rte_flow_restore_info *info,
632 struct rte_flow_error *err);
634 static const struct rte_flow_ops mlx5_flow_ops = {
635 .validate = mlx5_flow_validate,
636 .create = mlx5_flow_create,
637 .destroy = mlx5_flow_destroy,
638 .flush = mlx5_flow_flush,
639 .isolate = mlx5_flow_isolate,
640 .query = mlx5_flow_query,
641 .dev_dump = mlx5_flow_dev_dump,
642 .get_aged_flows = mlx5_flow_get_aged_flows,
643 .action_handle_create = mlx5_action_handle_create,
644 .action_handle_destroy = mlx5_action_handle_destroy,
645 .action_handle_update = mlx5_action_handle_update,
646 .action_handle_query = mlx5_action_handle_query,
647 .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
648 .tunnel_match = mlx5_flow_tunnel_match,
649 .tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
650 .tunnel_item_release = mlx5_flow_tunnel_item_release,
651 .get_restore_info = mlx5_flow_tunnel_get_restore_info,
654 /* Tunnel information. */
655 struct mlx5_flow_tunnel_info {
656 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
657 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
660 static struct mlx5_flow_tunnel_info tunnels_info[] = {
662 .tunnel = MLX5_FLOW_LAYER_VXLAN,
663 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
666 .tunnel = MLX5_FLOW_LAYER_GENEVE,
667 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
670 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
671 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
674 .tunnel = MLX5_FLOW_LAYER_GRE,
675 .ptype = RTE_PTYPE_TUNNEL_GRE,
678 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
679 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
682 .tunnel = MLX5_FLOW_LAYER_MPLS,
683 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
686 .tunnel = MLX5_FLOW_LAYER_NVGRE,
687 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
690 .tunnel = MLX5_FLOW_LAYER_IPIP,
691 .ptype = RTE_PTYPE_TUNNEL_IP,
694 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
695 .ptype = RTE_PTYPE_TUNNEL_IP,
698 .tunnel = MLX5_FLOW_LAYER_GTP,
699 .ptype = RTE_PTYPE_TUNNEL_GTPU,
706 * Translate tag ID to register.
709 * Pointer to the Ethernet device structure.
711 * The feature that request the register.
713 * The request register ID.
715 * Error description in case of any.
718 * The request register on success, a negative errno
719 * value otherwise and rte_errno is set.
722 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
723 enum mlx5_feature_name feature,
725 struct rte_flow_error *error)
727 struct mlx5_priv *priv = dev->data->dev_private;
728 struct mlx5_dev_config *config = &priv->config;
729 enum modify_reg start_reg;
730 bool skip_mtr_reg = false;
733 case MLX5_HAIRPIN_RX:
735 case MLX5_HAIRPIN_TX:
737 case MLX5_METADATA_RX:
738 switch (config->dv_xmeta_en) {
739 case MLX5_XMETA_MODE_LEGACY:
741 case MLX5_XMETA_MODE_META16:
743 case MLX5_XMETA_MODE_META32:
747 case MLX5_METADATA_TX:
749 case MLX5_METADATA_FDB:
750 switch (config->dv_xmeta_en) {
751 case MLX5_XMETA_MODE_LEGACY:
753 case MLX5_XMETA_MODE_META16:
755 case MLX5_XMETA_MODE_META32:
760 switch (config->dv_xmeta_en) {
761 case MLX5_XMETA_MODE_LEGACY:
763 case MLX5_XMETA_MODE_META16:
765 case MLX5_XMETA_MODE_META32:
771 * If meter color and meter id share one register, flow match
772 * should use the meter color register for match.
774 if (priv->mtr_reg_share)
775 return priv->mtr_color_reg;
777 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
780 case MLX5_ASO_FLOW_HIT:
781 case MLX5_ASO_CONNTRACK:
782 /* All features use the same REG_C. */
783 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
784 return priv->mtr_color_reg;
787 * Metadata COPY_MARK register using is in meter suffix sub
788 * flow while with meter. It's safe to share the same register.
790 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
793 * If meter is enable, it will engage the register for color
794 * match and flow match. If meter color match is not using the
795 * REG_C_2, need to skip the REG_C_x be used by meter color
797 * If meter is disable, free to use all available registers.
799 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
800 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
801 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
802 if (id > (uint32_t)(REG_C_7 - start_reg))
803 return rte_flow_error_set(error, EINVAL,
804 RTE_FLOW_ERROR_TYPE_ITEM,
805 NULL, "invalid tag id");
806 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
807 return rte_flow_error_set(error, ENOTSUP,
808 RTE_FLOW_ERROR_TYPE_ITEM,
809 NULL, "unsupported tag id");
811 * This case means meter is using the REG_C_x great than 2.
812 * Take care not to conflict with meter color REG_C_x.
813 * If the available index REG_C_y >= REG_C_x, skip the
816 if (skip_mtr_reg && config->flow_mreg_c
817 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
818 if (id >= (uint32_t)(REG_C_7 - start_reg))
819 return rte_flow_error_set(error, EINVAL,
820 RTE_FLOW_ERROR_TYPE_ITEM,
821 NULL, "invalid tag id");
822 if (config->flow_mreg_c
823 [id + 1 + start_reg - REG_C_0] != REG_NON)
824 return config->flow_mreg_c
825 [id + 1 + start_reg - REG_C_0];
826 return rte_flow_error_set(error, ENOTSUP,
827 RTE_FLOW_ERROR_TYPE_ITEM,
828 NULL, "unsupported tag id");
830 return config->flow_mreg_c[id + start_reg - REG_C_0];
833 return rte_flow_error_set(error, EINVAL,
834 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
835 NULL, "invalid feature name");
839 * Check extensive flow metadata register support.
842 * Pointer to rte_eth_dev structure.
845 * True if device supports extensive flow metadata register, otherwise false.
848 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
850 struct mlx5_priv *priv = dev->data->dev_private;
851 struct mlx5_dev_config *config = &priv->config;
854 * Having available reg_c can be regarded inclusively as supporting
855 * extensive flow metadata register, which could mean,
856 * - metadata register copy action by modify header.
857 * - 16 modify header actions is supported.
858 * - reg_c's are preserved across different domain (FDB and NIC) on
859 * packet loopback by flow lookup miss.
861 return config->flow_mreg_c[2] != REG_NON;
865 * Get the lowest priority.
868 * Pointer to the Ethernet device structure.
869 * @param[in] attributes
870 * Pointer to device flow rule attributes.
873 * The value of lowest priority of flow.
876 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
877 const struct rte_flow_attr *attr)
879 struct mlx5_priv *priv = dev->data->dev_private;
881 if (!attr->group && !attr->transfer)
882 return priv->config.flow_prio - 2;
883 return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
887 * Calculate matcher priority of the flow.
890 * Pointer to the Ethernet device structure.
892 * Pointer to device flow rule attributes.
893 * @param[in] subpriority
894 * The priority based on the items.
896 * The matcher priority of the flow.
899 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
900 const struct rte_flow_attr *attr,
901 uint32_t subpriority)
903 uint16_t priority = (uint16_t)attr->priority;
904 struct mlx5_priv *priv = dev->data->dev_private;
906 if (!attr->group && !attr->transfer) {
907 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
908 priority = priv->config.flow_prio - 1;
909 return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
911 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
912 priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
913 return priority * 3 + subpriority;
917 * Verify the @p item specifications (spec, last, mask) are compatible with the
921 * Item specification.
923 * @p item->mask or flow default bit-masks.
924 * @param[in] nic_mask
925 * Bit-masks covering supported fields by the NIC to compare with user mask.
927 * Bit-masks size in bytes.
928 * @param[in] range_accepted
929 * True if range of values is accepted for specific fields, false otherwise.
931 * Pointer to error structure.
934 * 0 on success, a negative errno value otherwise and rte_errno is set.
937 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
939 const uint8_t *nic_mask,
942 struct rte_flow_error *error)
946 MLX5_ASSERT(nic_mask);
947 for (i = 0; i < size; ++i)
948 if ((nic_mask[i] | mask[i]) != nic_mask[i])
949 return rte_flow_error_set(error, ENOTSUP,
950 RTE_FLOW_ERROR_TYPE_ITEM,
952 "mask enables non supported"
954 if (!item->spec && (item->mask || item->last))
955 return rte_flow_error_set(error, EINVAL,
956 RTE_FLOW_ERROR_TYPE_ITEM, item,
957 "mask/last without a spec is not"
959 if (item->spec && item->last && !range_accepted) {
965 for (i = 0; i < size; ++i) {
966 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
967 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
969 ret = memcmp(spec, last, size);
971 return rte_flow_error_set(error, EINVAL,
972 RTE_FLOW_ERROR_TYPE_ITEM,
974 "range is not valid");
980 * Adjust the hash fields according to the @p flow information.
982 * @param[in] dev_flow.
983 * Pointer to the mlx5_flow.
985 * 1 when the hash field is for a tunnel item.
986 * @param[in] layer_types
988 * @param[in] hash_fields
992 * The hash fields that should be used.
995 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
996 int tunnel __rte_unused, uint64_t layer_types,
997 uint64_t hash_fields)
999 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1000 int rss_request_inner = rss_desc->level >= 2;
1002 /* Check RSS hash level for tunnel. */
1003 if (tunnel && rss_request_inner)
1004 hash_fields |= IBV_RX_HASH_INNER;
1005 else if (tunnel || rss_request_inner)
1008 /* Check if requested layer matches RSS hash fields. */
1009 if (!(rss_desc->types & layer_types))
1015 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
1016 * if several tunnel rules are used on this queue, the tunnel ptype will be
1020 * Rx queue to update.
1023 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1026 uint32_t tunnel_ptype = 0;
1028 /* Look up for the ptype to use. */
1029 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1030 if (!rxq_ctrl->flow_tunnels_n[i])
1032 if (!tunnel_ptype) {
1033 tunnel_ptype = tunnels_info[i].ptype;
1039 rxq_ctrl->rxq.tunnel = tunnel_ptype;
1043 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1047 * Pointer to the Ethernet device structure.
1048 * @param[in] dev_handle
1049 * Pointer to device flow handle structure.
1052 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1053 struct mlx5_flow_handle *dev_handle)
1055 struct mlx5_priv *priv = dev->data->dev_private;
1056 const int mark = dev_handle->mark;
1057 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1058 struct mlx5_ind_table_obj *ind_tbl = NULL;
1061 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1062 struct mlx5_hrxq *hrxq;
1064 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1065 dev_handle->rix_hrxq);
1067 ind_tbl = hrxq->ind_table;
1068 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1069 struct mlx5_shared_action_rss *shared_rss;
1071 shared_rss = mlx5_ipool_get
1072 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1073 dev_handle->rix_srss);
1075 ind_tbl = shared_rss->ind_tbl;
1079 for (i = 0; i != ind_tbl->queues_n; ++i) {
1080 int idx = ind_tbl->queues[i];
1081 struct mlx5_rxq_ctrl *rxq_ctrl =
1082 container_of((*priv->rxqs)[idx],
1083 struct mlx5_rxq_ctrl, rxq);
1086 * To support metadata register copy on Tx loopback,
1087 * this must be always enabled (metadata may arive
1088 * from other port - not from local flows only.
1090 if (priv->config.dv_flow_en &&
1091 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1092 mlx5_flow_ext_mreg_supported(dev)) {
1093 rxq_ctrl->rxq.mark = 1;
1094 rxq_ctrl->flow_mark_n = 1;
1096 rxq_ctrl->rxq.mark = 1;
1097 rxq_ctrl->flow_mark_n++;
1102 /* Increase the counter matching the flow. */
1103 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1104 if ((tunnels_info[j].tunnel &
1105 dev_handle->layers) ==
1106 tunnels_info[j].tunnel) {
1107 rxq_ctrl->flow_tunnels_n[j]++;
1111 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1117 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1120 * Pointer to the Ethernet device structure.
1122 * Pointer to flow structure.
1125 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1127 struct mlx5_priv *priv = dev->data->dev_private;
1128 uint32_t handle_idx;
1129 struct mlx5_flow_handle *dev_handle;
1131 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1132 handle_idx, dev_handle, next)
1133 flow_drv_rxq_flags_set(dev, dev_handle);
1137 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1138 * device flow if no other flow uses it with the same kind of request.
1141 * Pointer to Ethernet device.
1142 * @param[in] dev_handle
1143 * Pointer to the device flow handle structure.
1146 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1147 struct mlx5_flow_handle *dev_handle)
1149 struct mlx5_priv *priv = dev->data->dev_private;
1150 const int mark = dev_handle->mark;
1151 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1152 struct mlx5_ind_table_obj *ind_tbl = NULL;
1155 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1156 struct mlx5_hrxq *hrxq;
1158 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1159 dev_handle->rix_hrxq);
1161 ind_tbl = hrxq->ind_table;
1162 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1163 struct mlx5_shared_action_rss *shared_rss;
1165 shared_rss = mlx5_ipool_get
1166 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1167 dev_handle->rix_srss);
1169 ind_tbl = shared_rss->ind_tbl;
1173 MLX5_ASSERT(dev->data->dev_started);
1174 for (i = 0; i != ind_tbl->queues_n; ++i) {
1175 int idx = ind_tbl->queues[i];
1176 struct mlx5_rxq_ctrl *rxq_ctrl =
1177 container_of((*priv->rxqs)[idx],
1178 struct mlx5_rxq_ctrl, rxq);
1180 if (priv->config.dv_flow_en &&
1181 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1182 mlx5_flow_ext_mreg_supported(dev)) {
1183 rxq_ctrl->rxq.mark = 1;
1184 rxq_ctrl->flow_mark_n = 1;
1186 rxq_ctrl->flow_mark_n--;
1187 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1192 /* Decrease the counter matching the flow. */
1193 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1194 if ((tunnels_info[j].tunnel &
1195 dev_handle->layers) ==
1196 tunnels_info[j].tunnel) {
1197 rxq_ctrl->flow_tunnels_n[j]--;
1201 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1207 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1208 * @p flow if no other flow uses it with the same kind of request.
1211 * Pointer to Ethernet device.
1213 * Pointer to the flow.
1216 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1218 struct mlx5_priv *priv = dev->data->dev_private;
1219 uint32_t handle_idx;
1220 struct mlx5_flow_handle *dev_handle;
1222 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1223 handle_idx, dev_handle, next)
1224 flow_drv_rxq_flags_trim(dev, dev_handle);
1228 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1231 * Pointer to Ethernet device.
1234 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1236 struct mlx5_priv *priv = dev->data->dev_private;
1239 for (i = 0; i != priv->rxqs_n; ++i) {
1240 struct mlx5_rxq_ctrl *rxq_ctrl;
1243 if (!(*priv->rxqs)[i])
1245 rxq_ctrl = container_of((*priv->rxqs)[i],
1246 struct mlx5_rxq_ctrl, rxq);
1247 rxq_ctrl->flow_mark_n = 0;
1248 rxq_ctrl->rxq.mark = 0;
1249 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1250 rxq_ctrl->flow_tunnels_n[j] = 0;
1251 rxq_ctrl->rxq.tunnel = 0;
1256 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1259 * Pointer to the Ethernet device structure.
1262 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1264 struct mlx5_priv *priv = dev->data->dev_private;
1265 struct mlx5_rxq_data *data;
1268 for (i = 0; i != priv->rxqs_n; ++i) {
1269 if (!(*priv->rxqs)[i])
1271 data = (*priv->rxqs)[i];
1272 if (!rte_flow_dynf_metadata_avail()) {
1273 data->dynf_meta = 0;
1274 data->flow_meta_mask = 0;
1275 data->flow_meta_offset = -1;
1276 data->flow_meta_port_mask = 0;
1278 data->dynf_meta = 1;
1279 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1280 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1281 data->flow_meta_port_mask = (uint32_t)~0;
1282 if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1283 data->flow_meta_port_mask >>= 16;
1289 * return a pointer to the desired action in the list of actions.
1291 * @param[in] actions
1292 * The list of actions to search the action in.
1294 * The action to find.
1297 * Pointer to the action in the list, if found. NULL otherwise.
1299 const struct rte_flow_action *
1300 mlx5_flow_find_action(const struct rte_flow_action *actions,
1301 enum rte_flow_action_type action)
1303 if (actions == NULL)
1305 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1306 if (actions->type == action)
1312 * Validate the flag action.
1314 * @param[in] action_flags
1315 * Bit-fields that holds the actions detected until now.
1317 * Attributes of flow that includes this action.
1319 * Pointer to error structure.
1322 * 0 on success, a negative errno value otherwise and rte_errno is set.
1325 mlx5_flow_validate_action_flag(uint64_t action_flags,
1326 const struct rte_flow_attr *attr,
1327 struct rte_flow_error *error)
1329 if (action_flags & MLX5_FLOW_ACTION_MARK)
1330 return rte_flow_error_set(error, EINVAL,
1331 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1332 "can't mark and flag in same flow");
1333 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1334 return rte_flow_error_set(error, EINVAL,
1335 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1337 " actions in same flow");
1339 return rte_flow_error_set(error, ENOTSUP,
1340 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1341 "flag action not supported for "
1347 * Validate the mark action.
1350 * Pointer to the queue action.
1351 * @param[in] action_flags
1352 * Bit-fields that holds the actions detected until now.
1354 * Attributes of flow that includes this action.
1356 * Pointer to error structure.
1359 * 0 on success, a negative errno value otherwise and rte_errno is set.
1362 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1363 uint64_t action_flags,
1364 const struct rte_flow_attr *attr,
1365 struct rte_flow_error *error)
1367 const struct rte_flow_action_mark *mark = action->conf;
1370 return rte_flow_error_set(error, EINVAL,
1371 RTE_FLOW_ERROR_TYPE_ACTION,
1373 "configuration cannot be null");
1374 if (mark->id >= MLX5_FLOW_MARK_MAX)
1375 return rte_flow_error_set(error, EINVAL,
1376 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1378 "mark id must in 0 <= id < "
1379 RTE_STR(MLX5_FLOW_MARK_MAX));
1380 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1381 return rte_flow_error_set(error, EINVAL,
1382 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1383 "can't flag and mark in same flow");
1384 if (action_flags & MLX5_FLOW_ACTION_MARK)
1385 return rte_flow_error_set(error, EINVAL,
1386 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1387 "can't have 2 mark actions in same"
1390 return rte_flow_error_set(error, ENOTSUP,
1391 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1392 "mark action not supported for "
1398 * Validate the drop action.
1400 * @param[in] action_flags
1401 * Bit-fields that holds the actions detected until now.
1403 * Attributes of flow that includes this action.
1405 * Pointer to error structure.
1408 * 0 on success, a negative errno value otherwise and rte_errno is set.
1411 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1412 const struct rte_flow_attr *attr,
1413 struct rte_flow_error *error)
1416 return rte_flow_error_set(error, ENOTSUP,
1417 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1418 "drop action not supported for "
1424 * Validate the queue action.
1427 * Pointer to the queue action.
1428 * @param[in] action_flags
1429 * Bit-fields that holds the actions detected until now.
1431 * Pointer to the Ethernet device structure.
1433 * Attributes of flow that includes this action.
1435 * Pointer to error structure.
1438 * 0 on success, a negative errno value otherwise and rte_errno is set.
1441 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1442 uint64_t action_flags,
1443 struct rte_eth_dev *dev,
1444 const struct rte_flow_attr *attr,
1445 struct rte_flow_error *error)
1447 struct mlx5_priv *priv = dev->data->dev_private;
1448 const struct rte_flow_action_queue *queue = action->conf;
1450 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1451 return rte_flow_error_set(error, EINVAL,
1452 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1453 "can't have 2 fate actions in"
1456 return rte_flow_error_set(error, EINVAL,
1457 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1458 NULL, "No Rx queues configured");
1459 if (queue->index >= priv->rxqs_n)
1460 return rte_flow_error_set(error, EINVAL,
1461 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1463 "queue index out of range");
1464 if (!(*priv->rxqs)[queue->index])
1465 return rte_flow_error_set(error, EINVAL,
1466 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1468 "queue is not configured");
1470 return rte_flow_error_set(error, ENOTSUP,
1471 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1472 "queue action not supported for "
1478 * Validate the rss action.
1481 * Pointer to the Ethernet device structure.
1483 * Pointer to the queue action.
1485 * Pointer to error structure.
1488 * 0 on success, a negative errno value otherwise and rte_errno is set.
1491 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1492 const struct rte_flow_action *action,
1493 struct rte_flow_error *error)
1495 struct mlx5_priv *priv = dev->data->dev_private;
1496 const struct rte_flow_action_rss *rss = action->conf;
1497 enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1500 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1501 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1502 return rte_flow_error_set(error, ENOTSUP,
1503 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1505 "RSS hash function not supported");
1506 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1511 return rte_flow_error_set(error, ENOTSUP,
1512 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1514 "tunnel RSS is not supported");
1515 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1516 if (rss->key_len == 0 && rss->key != NULL)
1517 return rte_flow_error_set(error, ENOTSUP,
1518 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1520 "RSS hash key length 0");
1521 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1522 return rte_flow_error_set(error, ENOTSUP,
1523 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1525 "RSS hash key too small");
1526 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1527 return rte_flow_error_set(error, ENOTSUP,
1528 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1530 "RSS hash key too large");
1531 if (rss->queue_num > priv->config.ind_table_max_size)
1532 return rte_flow_error_set(error, ENOTSUP,
1533 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1535 "number of queues too large");
1536 if (rss->types & MLX5_RSS_HF_MASK)
1537 return rte_flow_error_set(error, ENOTSUP,
1538 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1540 "some RSS protocols are not"
1542 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1543 !(rss->types & ETH_RSS_IP))
1544 return rte_flow_error_set(error, EINVAL,
1545 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1546 "L3 partial RSS requested but L3 RSS"
1547 " type not specified");
1548 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1549 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1550 return rte_flow_error_set(error, EINVAL,
1551 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1552 "L4 partial RSS requested but L4 RSS"
1553 " type not specified");
1555 return rte_flow_error_set(error, EINVAL,
1556 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1557 NULL, "No Rx queues configured");
1558 if (!rss->queue_num)
1559 return rte_flow_error_set(error, EINVAL,
1560 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1561 NULL, "No queues configured");
1562 for (i = 0; i != rss->queue_num; ++i) {
1563 struct mlx5_rxq_ctrl *rxq_ctrl;
1565 if (rss->queue[i] >= priv->rxqs_n)
1566 return rte_flow_error_set
1568 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1569 &rss->queue[i], "queue index out of range");
1570 if (!(*priv->rxqs)[rss->queue[i]])
1571 return rte_flow_error_set
1572 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1573 &rss->queue[i], "queue is not configured");
1574 rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1575 struct mlx5_rxq_ctrl, rxq);
1577 rxq_type = rxq_ctrl->type;
1578 if (rxq_type != rxq_ctrl->type)
1579 return rte_flow_error_set
1580 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1582 "combining hairpin and regular RSS queues is not supported");
1588 * Validate the rss action.
1591 * Pointer to the queue action.
1592 * @param[in] action_flags
1593 * Bit-fields that holds the actions detected until now.
1595 * Pointer to the Ethernet device structure.
1597 * Attributes of flow that includes this action.
1598 * @param[in] item_flags
1599 * Items that were detected.
1601 * Pointer to error structure.
1604 * 0 on success, a negative errno value otherwise and rte_errno is set.
1607 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1608 uint64_t action_flags,
1609 struct rte_eth_dev *dev,
1610 const struct rte_flow_attr *attr,
1611 uint64_t item_flags,
1612 struct rte_flow_error *error)
1614 const struct rte_flow_action_rss *rss = action->conf;
1615 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1618 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1619 return rte_flow_error_set(error, EINVAL,
1620 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1621 "can't have 2 fate actions"
1623 ret = mlx5_validate_action_rss(dev, action, error);
1627 return rte_flow_error_set(error, ENOTSUP,
1628 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1629 "rss action not supported for "
1631 if (rss->level > 1 && !tunnel)
1632 return rte_flow_error_set(error, EINVAL,
1633 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1634 "inner RSS is not supported for "
1635 "non-tunnel flows");
1636 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1637 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1638 return rte_flow_error_set(error, EINVAL,
1639 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1640 "RSS on eCPRI is not supported now");
1646 * Validate the default miss action.
1648 * @param[in] action_flags
1649 * Bit-fields that holds the actions detected until now.
1651 * Pointer to error structure.
1654 * 0 on success, a negative errno value otherwise and rte_errno is set.
1657 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1658 const struct rte_flow_attr *attr,
1659 struct rte_flow_error *error)
1661 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1662 return rte_flow_error_set(error, EINVAL,
1663 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1664 "can't have 2 fate actions in"
1667 return rte_flow_error_set(error, ENOTSUP,
1668 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1669 "default miss action not supported "
1672 return rte_flow_error_set(error, ENOTSUP,
1673 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1674 "only group 0 is supported");
1676 return rte_flow_error_set(error, ENOTSUP,
1677 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1678 NULL, "transfer is not supported");
1683 * Validate the count action.
1686 * Pointer to the Ethernet device structure.
1688 * Attributes of flow that includes this action.
1690 * Pointer to error structure.
1693 * 0 on success, a negative errno value otherwise and rte_errno is set.
1696 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1697 const struct rte_flow_attr *attr,
1698 struct rte_flow_error *error)
1701 return rte_flow_error_set(error, ENOTSUP,
1702 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1703 "count action not supported for "
1709 * Validate the ASO CT action.
1712 * Pointer to the Ethernet device structure.
1713 * @param[in] conntrack
1714 * Pointer to the CT action profile.
1716 * Pointer to error structure.
1719 * 0 on success, a negative errno value otherwise and rte_errno is set.
1722 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1723 const struct rte_flow_action_conntrack *conntrack,
1724 struct rte_flow_error *error)
1728 if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1729 return rte_flow_error_set(error, EINVAL,
1730 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1731 "Invalid CT state");
1732 if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1733 return rte_flow_error_set(error, EINVAL,
1734 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1735 "Invalid last TCP packet flag");
1740 * Verify the @p attributes will be correctly understood by the NIC and store
1741 * them in the @p flow if everything is correct.
1744 * Pointer to the Ethernet device structure.
1745 * @param[in] attributes
1746 * Pointer to flow attributes
1748 * Pointer to error structure.
1751 * 0 on success, a negative errno value otherwise and rte_errno is set.
1754 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1755 const struct rte_flow_attr *attributes,
1756 struct rte_flow_error *error)
1758 struct mlx5_priv *priv = dev->data->dev_private;
1759 uint32_t priority_max = priv->config.flow_prio - 1;
1761 if (attributes->group)
1762 return rte_flow_error_set(error, ENOTSUP,
1763 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1764 NULL, "groups is not supported");
1765 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1766 attributes->priority >= priority_max)
1767 return rte_flow_error_set(error, ENOTSUP,
1768 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1769 NULL, "priority out of range");
1770 if (attributes->egress)
1771 return rte_flow_error_set(error, ENOTSUP,
1772 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1773 "egress is not supported");
1774 if (attributes->transfer && !priv->config.dv_esw_en)
1775 return rte_flow_error_set(error, ENOTSUP,
1776 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1777 NULL, "transfer is not supported");
1778 if (!attributes->ingress)
1779 return rte_flow_error_set(error, EINVAL,
1780 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1782 "ingress attribute is mandatory");
1787 * Validate ICMP6 item.
1790 * Item specification.
1791 * @param[in] item_flags
1792 * Bit-fields that holds the items detected until now.
1793 * @param[in] ext_vlan_sup
1794 * Whether extended VLAN features are supported or not.
1796 * Pointer to error structure.
1799 * 0 on success, a negative errno value otherwise and rte_errno is set.
1802 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1803 uint64_t item_flags,
1804 uint8_t target_protocol,
1805 struct rte_flow_error *error)
1807 const struct rte_flow_item_icmp6 *mask = item->mask;
1808 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1809 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1810 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1811 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1812 MLX5_FLOW_LAYER_OUTER_L4;
1815 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1816 return rte_flow_error_set(error, EINVAL,
1817 RTE_FLOW_ERROR_TYPE_ITEM, item,
1818 "protocol filtering not compatible"
1819 " with ICMP6 layer");
1820 if (!(item_flags & l3m))
1821 return rte_flow_error_set(error, EINVAL,
1822 RTE_FLOW_ERROR_TYPE_ITEM, item,
1823 "IPv6 is mandatory to filter on"
1825 if (item_flags & l4m)
1826 return rte_flow_error_set(error, EINVAL,
1827 RTE_FLOW_ERROR_TYPE_ITEM, item,
1828 "multiple L4 layers not supported");
1830 mask = &rte_flow_item_icmp6_mask;
1831 ret = mlx5_flow_item_acceptable
1832 (item, (const uint8_t *)mask,
1833 (const uint8_t *)&rte_flow_item_icmp6_mask,
1834 sizeof(struct rte_flow_item_icmp6),
1835 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1842 * Validate ICMP item.
1845 * Item specification.
1846 * @param[in] item_flags
1847 * Bit-fields that holds the items detected until now.
1849 * Pointer to error structure.
1852 * 0 on success, a negative errno value otherwise and rte_errno is set.
1855 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1856 uint64_t item_flags,
1857 uint8_t target_protocol,
1858 struct rte_flow_error *error)
1860 const struct rte_flow_item_icmp *mask = item->mask;
1861 const struct rte_flow_item_icmp nic_mask = {
1862 .hdr.icmp_type = 0xff,
1863 .hdr.icmp_code = 0xff,
1864 .hdr.icmp_ident = RTE_BE16(0xffff),
1865 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1867 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1868 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1869 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1870 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1871 MLX5_FLOW_LAYER_OUTER_L4;
1874 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1875 return rte_flow_error_set(error, EINVAL,
1876 RTE_FLOW_ERROR_TYPE_ITEM, item,
1877 "protocol filtering not compatible"
1878 " with ICMP layer");
1879 if (!(item_flags & l3m))
1880 return rte_flow_error_set(error, EINVAL,
1881 RTE_FLOW_ERROR_TYPE_ITEM, item,
1882 "IPv4 is mandatory to filter"
1884 if (item_flags & l4m)
1885 return rte_flow_error_set(error, EINVAL,
1886 RTE_FLOW_ERROR_TYPE_ITEM, item,
1887 "multiple L4 layers not supported");
1890 ret = mlx5_flow_item_acceptable
1891 (item, (const uint8_t *)mask,
1892 (const uint8_t *)&nic_mask,
1893 sizeof(struct rte_flow_item_icmp),
1894 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1901 * Validate Ethernet item.
1904 * Item specification.
1905 * @param[in] item_flags
1906 * Bit-fields that holds the items detected until now.
1908 * Pointer to error structure.
1911 * 0 on success, a negative errno value otherwise and rte_errno is set.
1914 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1915 uint64_t item_flags, bool ext_vlan_sup,
1916 struct rte_flow_error *error)
1918 const struct rte_flow_item_eth *mask = item->mask;
1919 const struct rte_flow_item_eth nic_mask = {
1920 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1921 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1922 .type = RTE_BE16(0xffff),
1923 .has_vlan = ext_vlan_sup ? 1 : 0,
1926 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1927 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1928 MLX5_FLOW_LAYER_OUTER_L2;
1930 if (item_flags & ethm)
1931 return rte_flow_error_set(error, ENOTSUP,
1932 RTE_FLOW_ERROR_TYPE_ITEM, item,
1933 "multiple L2 layers not supported");
1934 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1935 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1936 return rte_flow_error_set(error, EINVAL,
1937 RTE_FLOW_ERROR_TYPE_ITEM, item,
1938 "L2 layer should not follow "
1940 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1941 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1942 return rte_flow_error_set(error, EINVAL,
1943 RTE_FLOW_ERROR_TYPE_ITEM, item,
1944 "L2 layer should not follow VLAN");
1946 mask = &rte_flow_item_eth_mask;
1947 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1948 (const uint8_t *)&nic_mask,
1949 sizeof(struct rte_flow_item_eth),
1950 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1955 * Validate VLAN item.
1958 * Item specification.
1959 * @param[in] item_flags
1960 * Bit-fields that holds the items detected until now.
1962 * Ethernet device flow is being created on.
1964 * Pointer to error structure.
1967 * 0 on success, a negative errno value otherwise and rte_errno is set.
1970 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1971 uint64_t item_flags,
1972 struct rte_eth_dev *dev,
1973 struct rte_flow_error *error)
1975 const struct rte_flow_item_vlan *spec = item->spec;
1976 const struct rte_flow_item_vlan *mask = item->mask;
1977 const struct rte_flow_item_vlan nic_mask = {
1978 .tci = RTE_BE16(UINT16_MAX),
1979 .inner_type = RTE_BE16(UINT16_MAX),
1981 uint16_t vlan_tag = 0;
1982 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1984 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1985 MLX5_FLOW_LAYER_INNER_L4) :
1986 (MLX5_FLOW_LAYER_OUTER_L3 |
1987 MLX5_FLOW_LAYER_OUTER_L4);
1988 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1989 MLX5_FLOW_LAYER_OUTER_VLAN;
1991 if (item_flags & vlanm)
1992 return rte_flow_error_set(error, EINVAL,
1993 RTE_FLOW_ERROR_TYPE_ITEM, item,
1994 "multiple VLAN layers not supported");
1995 else if ((item_flags & l34m) != 0)
1996 return rte_flow_error_set(error, EINVAL,
1997 RTE_FLOW_ERROR_TYPE_ITEM, item,
1998 "VLAN cannot follow L3/L4 layer");
2000 mask = &rte_flow_item_vlan_mask;
2001 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2002 (const uint8_t *)&nic_mask,
2003 sizeof(struct rte_flow_item_vlan),
2004 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2007 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2008 struct mlx5_priv *priv = dev->data->dev_private;
2010 if (priv->vmwa_context) {
2012 * Non-NULL context means we have a virtual machine
2013 * and SR-IOV enabled, we have to create VLAN interface
2014 * to make hypervisor to setup E-Switch vport
2015 * context correctly. We avoid creating the multiple
2016 * VLAN interfaces, so we cannot support VLAN tag mask.
2018 return rte_flow_error_set(error, EINVAL,
2019 RTE_FLOW_ERROR_TYPE_ITEM,
2021 "VLAN tag mask is not"
2022 " supported in virtual"
2027 vlan_tag = spec->tci;
2028 vlan_tag &= mask->tci;
2031 * From verbs perspective an empty VLAN is equivalent
2032 * to a packet without VLAN layer.
2035 return rte_flow_error_set(error, EINVAL,
2036 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2038 "VLAN cannot be empty");
2043 * Validate IPV4 item.
2046 * Item specification.
2047 * @param[in] item_flags
2048 * Bit-fields that holds the items detected until now.
2049 * @param[in] last_item
2050 * Previous validated item in the pattern items.
2051 * @param[in] ether_type
2052 * Type in the ethernet layer header (including dot1q).
2053 * @param[in] acc_mask
2054 * Acceptable mask, if NULL default internal default mask
2055 * will be used to check whether item fields are supported.
2056 * @param[in] range_accepted
2057 * True if range of values is accepted for specific fields, false otherwise.
2059 * Pointer to error structure.
2062 * 0 on success, a negative errno value otherwise and rte_errno is set.
2065 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2066 uint64_t item_flags,
2068 uint16_t ether_type,
2069 const struct rte_flow_item_ipv4 *acc_mask,
2070 bool range_accepted,
2071 struct rte_flow_error *error)
2073 const struct rte_flow_item_ipv4 *mask = item->mask;
2074 const struct rte_flow_item_ipv4 *spec = item->spec;
2075 const struct rte_flow_item_ipv4 nic_mask = {
2077 .src_addr = RTE_BE32(0xffffffff),
2078 .dst_addr = RTE_BE32(0xffffffff),
2079 .type_of_service = 0xff,
2080 .next_proto_id = 0xff,
2083 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2084 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2085 MLX5_FLOW_LAYER_OUTER_L3;
2086 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2087 MLX5_FLOW_LAYER_OUTER_L4;
2089 uint8_t next_proto = 0xFF;
2090 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2091 MLX5_FLOW_LAYER_OUTER_VLAN |
2092 MLX5_FLOW_LAYER_INNER_VLAN);
2094 if ((last_item & l2_vlan) && ether_type &&
2095 ether_type != RTE_ETHER_TYPE_IPV4)
2096 return rte_flow_error_set(error, EINVAL,
2097 RTE_FLOW_ERROR_TYPE_ITEM, item,
2098 "IPv4 cannot follow L2/VLAN layer "
2099 "which ether type is not IPv4");
2100 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2102 next_proto = mask->hdr.next_proto_id &
2103 spec->hdr.next_proto_id;
2104 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2105 return rte_flow_error_set(error, EINVAL,
2106 RTE_FLOW_ERROR_TYPE_ITEM,
2111 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2112 return rte_flow_error_set(error, EINVAL,
2113 RTE_FLOW_ERROR_TYPE_ITEM, item,
2114 "wrong tunnel type - IPv6 specified "
2115 "but IPv4 item provided");
2116 if (item_flags & l3m)
2117 return rte_flow_error_set(error, ENOTSUP,
2118 RTE_FLOW_ERROR_TYPE_ITEM, item,
2119 "multiple L3 layers not supported");
2120 else if (item_flags & l4m)
2121 return rte_flow_error_set(error, EINVAL,
2122 RTE_FLOW_ERROR_TYPE_ITEM, item,
2123 "L3 cannot follow an L4 layer.");
2124 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2125 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2126 return rte_flow_error_set(error, EINVAL,
2127 RTE_FLOW_ERROR_TYPE_ITEM, item,
2128 "L3 cannot follow an NVGRE layer.");
2130 mask = &rte_flow_item_ipv4_mask;
2131 else if (mask->hdr.next_proto_id != 0 &&
2132 mask->hdr.next_proto_id != 0xff)
2133 return rte_flow_error_set(error, EINVAL,
2134 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2135 "partial mask is not supported"
2137 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2138 acc_mask ? (const uint8_t *)acc_mask
2139 : (const uint8_t *)&nic_mask,
2140 sizeof(struct rte_flow_item_ipv4),
2141 range_accepted, error);
2148 * Validate IPV6 item.
2151 * Item specification.
2152 * @param[in] item_flags
2153 * Bit-fields that holds the items detected until now.
2154 * @param[in] last_item
2155 * Previous validated item in the pattern items.
2156 * @param[in] ether_type
2157 * Type in the ethernet layer header (including dot1q).
2158 * @param[in] acc_mask
2159 * Acceptable mask, if NULL default internal default mask
2160 * will be used to check whether item fields are supported.
2162 * Pointer to error structure.
2165 * 0 on success, a negative errno value otherwise and rte_errno is set.
2168 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2169 uint64_t item_flags,
2171 uint16_t ether_type,
2172 const struct rte_flow_item_ipv6 *acc_mask,
2173 struct rte_flow_error *error)
2175 const struct rte_flow_item_ipv6 *mask = item->mask;
2176 const struct rte_flow_item_ipv6 *spec = item->spec;
2177 const struct rte_flow_item_ipv6 nic_mask = {
2180 "\xff\xff\xff\xff\xff\xff\xff\xff"
2181 "\xff\xff\xff\xff\xff\xff\xff\xff",
2183 "\xff\xff\xff\xff\xff\xff\xff\xff"
2184 "\xff\xff\xff\xff\xff\xff\xff\xff",
2185 .vtc_flow = RTE_BE32(0xffffffff),
2189 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2190 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2191 MLX5_FLOW_LAYER_OUTER_L3;
2192 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2193 MLX5_FLOW_LAYER_OUTER_L4;
2195 uint8_t next_proto = 0xFF;
2196 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2197 MLX5_FLOW_LAYER_OUTER_VLAN |
2198 MLX5_FLOW_LAYER_INNER_VLAN);
2200 if ((last_item & l2_vlan) && ether_type &&
2201 ether_type != RTE_ETHER_TYPE_IPV6)
2202 return rte_flow_error_set(error, EINVAL,
2203 RTE_FLOW_ERROR_TYPE_ITEM, item,
2204 "IPv6 cannot follow L2/VLAN layer "
2205 "which ether type is not IPv6");
2206 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2207 next_proto = spec->hdr.proto;
2208 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2209 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2210 return rte_flow_error_set(error, EINVAL,
2211 RTE_FLOW_ERROR_TYPE_ITEM,
2216 if (next_proto == IPPROTO_HOPOPTS ||
2217 next_proto == IPPROTO_ROUTING ||
2218 next_proto == IPPROTO_FRAGMENT ||
2219 next_proto == IPPROTO_ESP ||
2220 next_proto == IPPROTO_AH ||
2221 next_proto == IPPROTO_DSTOPTS)
2222 return rte_flow_error_set(error, EINVAL,
2223 RTE_FLOW_ERROR_TYPE_ITEM, item,
2224 "IPv6 proto (next header) should "
2225 "not be set as extension header");
2226 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2227 return rte_flow_error_set(error, EINVAL,
2228 RTE_FLOW_ERROR_TYPE_ITEM, item,
2229 "wrong tunnel type - IPv4 specified "
2230 "but IPv6 item provided");
2231 if (item_flags & l3m)
2232 return rte_flow_error_set(error, ENOTSUP,
2233 RTE_FLOW_ERROR_TYPE_ITEM, item,
2234 "multiple L3 layers not supported");
2235 else if (item_flags & l4m)
2236 return rte_flow_error_set(error, EINVAL,
2237 RTE_FLOW_ERROR_TYPE_ITEM, item,
2238 "L3 cannot follow an L4 layer.");
2239 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2240 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2241 return rte_flow_error_set(error, EINVAL,
2242 RTE_FLOW_ERROR_TYPE_ITEM, item,
2243 "L3 cannot follow an NVGRE layer.");
2245 mask = &rte_flow_item_ipv6_mask;
2246 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2247 acc_mask ? (const uint8_t *)acc_mask
2248 : (const uint8_t *)&nic_mask,
2249 sizeof(struct rte_flow_item_ipv6),
2250 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2257 * Validate UDP item.
2260 * Item specification.
2261 * @param[in] item_flags
2262 * Bit-fields that holds the items detected until now.
2263 * @param[in] target_protocol
2264 * The next protocol in the previous item.
2265 * @param[in] flow_mask
2266 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2268 * Pointer to error structure.
2271 * 0 on success, a negative errno value otherwise and rte_errno is set.
2274 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2275 uint64_t item_flags,
2276 uint8_t target_protocol,
2277 struct rte_flow_error *error)
2279 const struct rte_flow_item_udp *mask = item->mask;
2280 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2281 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2282 MLX5_FLOW_LAYER_OUTER_L3;
2283 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2284 MLX5_FLOW_LAYER_OUTER_L4;
2287 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2288 return rte_flow_error_set(error, EINVAL,
2289 RTE_FLOW_ERROR_TYPE_ITEM, item,
2290 "protocol filtering not compatible"
2292 if (!(item_flags & l3m))
2293 return rte_flow_error_set(error, EINVAL,
2294 RTE_FLOW_ERROR_TYPE_ITEM, item,
2295 "L3 is mandatory to filter on L4");
2296 if (item_flags & l4m)
2297 return rte_flow_error_set(error, EINVAL,
2298 RTE_FLOW_ERROR_TYPE_ITEM, item,
2299 "multiple L4 layers not supported");
2301 mask = &rte_flow_item_udp_mask;
2302 ret = mlx5_flow_item_acceptable
2303 (item, (const uint8_t *)mask,
2304 (const uint8_t *)&rte_flow_item_udp_mask,
2305 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2313 * Validate TCP item.
2316 * Item specification.
2317 * @param[in] item_flags
2318 * Bit-fields that holds the items detected until now.
2319 * @param[in] target_protocol
2320 * The next protocol in the previous item.
2322 * Pointer to error structure.
2325 * 0 on success, a negative errno value otherwise and rte_errno is set.
2328 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2329 uint64_t item_flags,
2330 uint8_t target_protocol,
2331 const struct rte_flow_item_tcp *flow_mask,
2332 struct rte_flow_error *error)
2334 const struct rte_flow_item_tcp *mask = item->mask;
2335 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2336 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2337 MLX5_FLOW_LAYER_OUTER_L3;
2338 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2339 MLX5_FLOW_LAYER_OUTER_L4;
2342 MLX5_ASSERT(flow_mask);
2343 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2344 return rte_flow_error_set(error, EINVAL,
2345 RTE_FLOW_ERROR_TYPE_ITEM, item,
2346 "protocol filtering not compatible"
2348 if (!(item_flags & l3m))
2349 return rte_flow_error_set(error, EINVAL,
2350 RTE_FLOW_ERROR_TYPE_ITEM, item,
2351 "L3 is mandatory to filter on L4");
2352 if (item_flags & l4m)
2353 return rte_flow_error_set(error, EINVAL,
2354 RTE_FLOW_ERROR_TYPE_ITEM, item,
2355 "multiple L4 layers not supported");
2357 mask = &rte_flow_item_tcp_mask;
2358 ret = mlx5_flow_item_acceptable
2359 (item, (const uint8_t *)mask,
2360 (const uint8_t *)flow_mask,
2361 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2369 * Validate VXLAN item.
2372 * Item specification.
2373 * @param[in] item_flags
2374 * Bit-fields that holds the items detected until now.
2375 * @param[in] target_protocol
2376 * The next protocol in the previous item.
2378 * Pointer to error structure.
2381 * 0 on success, a negative errno value otherwise and rte_errno is set.
2384 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2385 uint64_t item_flags,
2386 struct rte_flow_error *error)
2388 const struct rte_flow_item_vxlan *spec = item->spec;
2389 const struct rte_flow_item_vxlan *mask = item->mask;
2394 } id = { .vlan_id = 0, };
2397 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2398 return rte_flow_error_set(error, ENOTSUP,
2399 RTE_FLOW_ERROR_TYPE_ITEM, item,
2400 "multiple tunnel layers not"
2403 * Verify only UDPv4 is present as defined in
2404 * https://tools.ietf.org/html/rfc7348
2406 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2407 return rte_flow_error_set(error, EINVAL,
2408 RTE_FLOW_ERROR_TYPE_ITEM, item,
2409 "no outer UDP layer found");
2411 mask = &rte_flow_item_vxlan_mask;
2412 ret = mlx5_flow_item_acceptable
2413 (item, (const uint8_t *)mask,
2414 (const uint8_t *)&rte_flow_item_vxlan_mask,
2415 sizeof(struct rte_flow_item_vxlan),
2416 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2420 memcpy(&id.vni[1], spec->vni, 3);
2421 memcpy(&id.vni[1], mask->vni, 3);
2423 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2424 return rte_flow_error_set(error, ENOTSUP,
2425 RTE_FLOW_ERROR_TYPE_ITEM, item,
2426 "VXLAN tunnel must be fully defined");
2431 * Validate VXLAN_GPE item.
2434 * Item specification.
2435 * @param[in] item_flags
2436 * Bit-fields that holds the items detected until now.
2438 * Pointer to the private data structure.
2439 * @param[in] target_protocol
2440 * The next protocol in the previous item.
2442 * Pointer to error structure.
2445 * 0 on success, a negative errno value otherwise and rte_errno is set.
2448 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2449 uint64_t item_flags,
2450 struct rte_eth_dev *dev,
2451 struct rte_flow_error *error)
2453 struct mlx5_priv *priv = dev->data->dev_private;
2454 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2455 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2460 } id = { .vlan_id = 0, };
2462 if (!priv->config.l3_vxlan_en)
2463 return rte_flow_error_set(error, ENOTSUP,
2464 RTE_FLOW_ERROR_TYPE_ITEM, item,
2465 "L3 VXLAN is not enabled by device"
2466 " parameter and/or not configured in"
2468 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2469 return rte_flow_error_set(error, ENOTSUP,
2470 RTE_FLOW_ERROR_TYPE_ITEM, item,
2471 "multiple tunnel layers not"
2474 * Verify only UDPv4 is present as defined in
2475 * https://tools.ietf.org/html/rfc7348
2477 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2478 return rte_flow_error_set(error, EINVAL,
2479 RTE_FLOW_ERROR_TYPE_ITEM, item,
2480 "no outer UDP layer found");
2482 mask = &rte_flow_item_vxlan_gpe_mask;
2483 ret = mlx5_flow_item_acceptable
2484 (item, (const uint8_t *)mask,
2485 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2486 sizeof(struct rte_flow_item_vxlan_gpe),
2487 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2492 return rte_flow_error_set(error, ENOTSUP,
2493 RTE_FLOW_ERROR_TYPE_ITEM,
2495 "VxLAN-GPE protocol"
2497 memcpy(&id.vni[1], spec->vni, 3);
2498 memcpy(&id.vni[1], mask->vni, 3);
2500 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2501 return rte_flow_error_set(error, ENOTSUP,
2502 RTE_FLOW_ERROR_TYPE_ITEM, item,
2503 "VXLAN-GPE tunnel must be fully"
2508 * Validate GRE Key item.
2511 * Item specification.
2512 * @param[in] item_flags
2513 * Bit flags to mark detected items.
2514 * @param[in] gre_item
2515 * Pointer to gre_item
2517 * Pointer to error structure.
2520 * 0 on success, a negative errno value otherwise and rte_errno is set.
2523 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2524 uint64_t item_flags,
2525 const struct rte_flow_item *gre_item,
2526 struct rte_flow_error *error)
2528 const rte_be32_t *mask = item->mask;
2530 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2531 const struct rte_flow_item_gre *gre_spec;
2532 const struct rte_flow_item_gre *gre_mask;
2534 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2535 return rte_flow_error_set(error, ENOTSUP,
2536 RTE_FLOW_ERROR_TYPE_ITEM, item,
2537 "Multiple GRE key not support");
2538 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2539 return rte_flow_error_set(error, ENOTSUP,
2540 RTE_FLOW_ERROR_TYPE_ITEM, item,
2541 "No preceding GRE header");
2542 if (item_flags & MLX5_FLOW_LAYER_INNER)
2543 return rte_flow_error_set(error, ENOTSUP,
2544 RTE_FLOW_ERROR_TYPE_ITEM, item,
2545 "GRE key following a wrong item");
2546 gre_mask = gre_item->mask;
2548 gre_mask = &rte_flow_item_gre_mask;
2549 gre_spec = gre_item->spec;
2550 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2551 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2552 return rte_flow_error_set(error, EINVAL,
2553 RTE_FLOW_ERROR_TYPE_ITEM, item,
2554 "Key bit must be on");
2557 mask = &gre_key_default_mask;
2558 ret = mlx5_flow_item_acceptable
2559 (item, (const uint8_t *)mask,
2560 (const uint8_t *)&gre_key_default_mask,
2561 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2566 * Validate GRE item.
2569 * Item specification.
2570 * @param[in] item_flags
2571 * Bit flags to mark detected items.
2572 * @param[in] target_protocol
2573 * The next protocol in the previous item.
2575 * Pointer to error structure.
2578 * 0 on success, a negative errno value otherwise and rte_errno is set.
2581 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2582 uint64_t item_flags,
2583 uint8_t target_protocol,
2584 struct rte_flow_error *error)
2586 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2587 const struct rte_flow_item_gre *mask = item->mask;
2589 const struct rte_flow_item_gre nic_mask = {
2590 .c_rsvd0_ver = RTE_BE16(0xB000),
2591 .protocol = RTE_BE16(UINT16_MAX),
2594 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2595 return rte_flow_error_set(error, EINVAL,
2596 RTE_FLOW_ERROR_TYPE_ITEM, item,
2597 "protocol filtering not compatible"
2598 " with this GRE layer");
2599 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2600 return rte_flow_error_set(error, ENOTSUP,
2601 RTE_FLOW_ERROR_TYPE_ITEM, item,
2602 "multiple tunnel layers not"
2604 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2605 return rte_flow_error_set(error, ENOTSUP,
2606 RTE_FLOW_ERROR_TYPE_ITEM, item,
2607 "L3 Layer is missing");
2609 mask = &rte_flow_item_gre_mask;
2610 ret = mlx5_flow_item_acceptable
2611 (item, (const uint8_t *)mask,
2612 (const uint8_t *)&nic_mask,
2613 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2617 #ifndef HAVE_MLX5DV_DR
2618 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2619 if (spec && (spec->protocol & mask->protocol))
2620 return rte_flow_error_set(error, ENOTSUP,
2621 RTE_FLOW_ERROR_TYPE_ITEM, item,
2622 "without MPLS support the"
2623 " specification cannot be used for"
2631 * Validate Geneve item.
2634 * Item specification.
2635 * @param[in] itemFlags
2636 * Bit-fields that holds the items detected until now.
2638 * Pointer to the private data structure.
2640 * Pointer to error structure.
2643 * 0 on success, a negative errno value otherwise and rte_errno is set.
2647 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2648 uint64_t item_flags,
2649 struct rte_eth_dev *dev,
2650 struct rte_flow_error *error)
2652 struct mlx5_priv *priv = dev->data->dev_private;
2653 const struct rte_flow_item_geneve *spec = item->spec;
2654 const struct rte_flow_item_geneve *mask = item->mask;
2657 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2658 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2659 const struct rte_flow_item_geneve nic_mask = {
2660 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2661 .vni = "\xff\xff\xff",
2662 .protocol = RTE_BE16(UINT16_MAX),
2665 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2666 return rte_flow_error_set(error, ENOTSUP,
2667 RTE_FLOW_ERROR_TYPE_ITEM, item,
2668 "L3 Geneve is not enabled by device"
2669 " parameter and/or not configured in"
2671 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2672 return rte_flow_error_set(error, ENOTSUP,
2673 RTE_FLOW_ERROR_TYPE_ITEM, item,
2674 "multiple tunnel layers not"
2677 * Verify only UDPv4 is present as defined in
2678 * https://tools.ietf.org/html/rfc7348
2680 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2681 return rte_flow_error_set(error, EINVAL,
2682 RTE_FLOW_ERROR_TYPE_ITEM, item,
2683 "no outer UDP layer found");
2685 mask = &rte_flow_item_geneve_mask;
2686 ret = mlx5_flow_item_acceptable
2687 (item, (const uint8_t *)mask,
2688 (const uint8_t *)&nic_mask,
2689 sizeof(struct rte_flow_item_geneve),
2690 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2694 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2695 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2696 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2697 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2698 return rte_flow_error_set(error, ENOTSUP,
2699 RTE_FLOW_ERROR_TYPE_ITEM,
2701 "Geneve protocol unsupported"
2702 " fields are being used");
2703 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2704 return rte_flow_error_set
2706 RTE_FLOW_ERROR_TYPE_ITEM,
2708 "Unsupported Geneve options length");
2710 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2711 return rte_flow_error_set
2713 RTE_FLOW_ERROR_TYPE_ITEM, item,
2714 "Geneve tunnel must be fully defined");
2719 * Validate Geneve TLV option item.
2722 * Item specification.
2723 * @param[in] last_item
2724 * Previous validated item in the pattern items.
2725 * @param[in] geneve_item
2726 * Previous GENEVE item specification.
2728 * Pointer to the rte_eth_dev structure.
2730 * Pointer to error structure.
2733 * 0 on success, a negative errno value otherwise and rte_errno is set.
2736 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2738 const struct rte_flow_item *geneve_item,
2739 struct rte_eth_dev *dev,
2740 struct rte_flow_error *error)
2742 struct mlx5_priv *priv = dev->data->dev_private;
2743 struct mlx5_dev_ctx_shared *sh = priv->sh;
2744 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2745 struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2746 uint8_t data_max_supported =
2747 hca_attr->max_geneve_tlv_option_data_len * 4;
2748 struct mlx5_dev_config *config = &priv->config;
2749 const struct rte_flow_item_geneve *geneve_spec;
2750 const struct rte_flow_item_geneve *geneve_mask;
2751 const struct rte_flow_item_geneve_opt *spec = item->spec;
2752 const struct rte_flow_item_geneve_opt *mask = item->mask;
2754 unsigned int data_len;
2755 uint8_t tlv_option_len;
2756 uint16_t optlen_m, optlen_v;
2757 const struct rte_flow_item_geneve_opt full_mask = {
2758 .option_class = RTE_BE16(0xffff),
2759 .option_type = 0xff,
2764 mask = &rte_flow_item_geneve_opt_mask;
2766 return rte_flow_error_set
2767 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2768 "Geneve TLV opt class/type/length must be specified");
2769 if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2770 return rte_flow_error_set
2771 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2772 "Geneve TLV opt length exceeeds the limit (31)");
2773 /* Check if class type and length masks are full. */
2774 if (full_mask.option_class != mask->option_class ||
2775 full_mask.option_type != mask->option_type ||
2776 full_mask.option_len != (mask->option_len & full_mask.option_len))
2777 return rte_flow_error_set
2778 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2779 "Geneve TLV opt class/type/length masks must be full");
2780 /* Check if length is supported */
2781 if ((uint32_t)spec->option_len >
2782 config->hca_attr.max_geneve_tlv_option_data_len)
2783 return rte_flow_error_set
2784 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2785 "Geneve TLV opt length not supported");
2786 if (config->hca_attr.max_geneve_tlv_options > 1)
2788 "max_geneve_tlv_options supports more than 1 option");
2789 /* Check GENEVE item preceding. */
2790 if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2791 return rte_flow_error_set
2792 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2793 "Geneve opt item must be preceded with Geneve item");
2794 geneve_spec = geneve_item->spec;
2795 geneve_mask = geneve_item->mask ? geneve_item->mask :
2796 &rte_flow_item_geneve_mask;
2797 /* Check if GENEVE TLV option size doesn't exceed option length */
2798 if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2799 geneve_spec->ver_opt_len_o_c_rsvd0)) {
2800 tlv_option_len = spec->option_len & mask->option_len;
2801 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2802 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2803 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2804 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2805 if ((optlen_v & optlen_m) <= tlv_option_len)
2806 return rte_flow_error_set
2807 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2808 "GENEVE TLV option length exceeds optlen");
2810 /* Check if length is 0 or data is 0. */
2811 if (spec->data == NULL || spec->option_len == 0)
2812 return rte_flow_error_set
2813 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2814 "Geneve TLV opt with zero data/length not supported");
2815 /* Check not all data & mask are 0. */
2816 data_len = spec->option_len * 4;
2817 if (mask->data == NULL) {
2818 for (i = 0; i < data_len; i++)
2822 return rte_flow_error_set(error, ENOTSUP,
2823 RTE_FLOW_ERROR_TYPE_ITEM, item,
2824 "Can't match on Geneve option data 0");
2826 for (i = 0; i < data_len; i++)
2827 if (spec->data[i] & mask->data[i])
2830 return rte_flow_error_set(error, ENOTSUP,
2831 RTE_FLOW_ERROR_TYPE_ITEM, item,
2832 "Can't match on Geneve option data and mask 0");
2833 /* Check data mask supported. */
2834 for (i = data_max_supported; i < data_len ; i++)
2836 return rte_flow_error_set(error, ENOTSUP,
2837 RTE_FLOW_ERROR_TYPE_ITEM, item,
2838 "Data mask is of unsupported size");
2840 /* Check GENEVE option is supported in NIC. */
2841 if (!config->hca_attr.geneve_tlv_opt)
2842 return rte_flow_error_set
2843 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2844 "Geneve TLV opt not supported");
2845 /* Check if we already have geneve option with different type/class. */
2846 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2847 geneve_opt_resource = sh->geneve_tlv_option_resource;
2848 if (geneve_opt_resource != NULL)
2849 if (geneve_opt_resource->option_class != spec->option_class ||
2850 geneve_opt_resource->option_type != spec->option_type ||
2851 geneve_opt_resource->length != spec->option_len) {
2852 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2853 return rte_flow_error_set(error, ENOTSUP,
2854 RTE_FLOW_ERROR_TYPE_ITEM, item,
2855 "Only one Geneve TLV option supported");
2857 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2862 * Validate MPLS item.
2865 * Pointer to the rte_eth_dev structure.
2867 * Item specification.
2868 * @param[in] item_flags
2869 * Bit-fields that holds the items detected until now.
2870 * @param[in] prev_layer
2871 * The protocol layer indicated in previous item.
2873 * Pointer to error structure.
2876 * 0 on success, a negative errno value otherwise and rte_errno is set.
2879 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2880 const struct rte_flow_item *item __rte_unused,
2881 uint64_t item_flags __rte_unused,
2882 uint64_t prev_layer __rte_unused,
2883 struct rte_flow_error *error)
2885 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2886 const struct rte_flow_item_mpls *mask = item->mask;
2887 struct mlx5_priv *priv = dev->data->dev_private;
2890 if (!priv->config.mpls_en)
2891 return rte_flow_error_set(error, ENOTSUP,
2892 RTE_FLOW_ERROR_TYPE_ITEM, item,
2893 "MPLS not supported or"
2894 " disabled in firmware"
2896 /* MPLS over IP, UDP, GRE is allowed */
2897 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2898 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2899 MLX5_FLOW_LAYER_GRE |
2900 MLX5_FLOW_LAYER_GRE_KEY)))
2901 return rte_flow_error_set(error, EINVAL,
2902 RTE_FLOW_ERROR_TYPE_ITEM, item,
2903 "protocol filtering not compatible"
2904 " with MPLS layer");
2905 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2906 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2907 !(item_flags & MLX5_FLOW_LAYER_GRE))
2908 return rte_flow_error_set(error, ENOTSUP,
2909 RTE_FLOW_ERROR_TYPE_ITEM, item,
2910 "multiple tunnel layers not"
2913 mask = &rte_flow_item_mpls_mask;
2914 ret = mlx5_flow_item_acceptable
2915 (item, (const uint8_t *)mask,
2916 (const uint8_t *)&rte_flow_item_mpls_mask,
2917 sizeof(struct rte_flow_item_mpls),
2918 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2923 return rte_flow_error_set(error, ENOTSUP,
2924 RTE_FLOW_ERROR_TYPE_ITEM, item,
2925 "MPLS is not supported by Verbs, please"
2931 * Validate NVGRE item.
2934 * Item specification.
2935 * @param[in] item_flags
2936 * Bit flags to mark detected items.
2937 * @param[in] target_protocol
2938 * The next protocol in the previous item.
2940 * Pointer to error structure.
2943 * 0 on success, a negative errno value otherwise and rte_errno is set.
2946 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2947 uint64_t item_flags,
2948 uint8_t target_protocol,
2949 struct rte_flow_error *error)
2951 const struct rte_flow_item_nvgre *mask = item->mask;
2954 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2955 return rte_flow_error_set(error, EINVAL,
2956 RTE_FLOW_ERROR_TYPE_ITEM, item,
2957 "protocol filtering not compatible"
2958 " with this GRE layer");
2959 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2960 return rte_flow_error_set(error, ENOTSUP,
2961 RTE_FLOW_ERROR_TYPE_ITEM, item,
2962 "multiple tunnel layers not"
2964 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2965 return rte_flow_error_set(error, ENOTSUP,
2966 RTE_FLOW_ERROR_TYPE_ITEM, item,
2967 "L3 Layer is missing");
2969 mask = &rte_flow_item_nvgre_mask;
2970 ret = mlx5_flow_item_acceptable
2971 (item, (const uint8_t *)mask,
2972 (const uint8_t *)&rte_flow_item_nvgre_mask,
2973 sizeof(struct rte_flow_item_nvgre),
2974 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2981 * Validate eCPRI item.
2984 * Item specification.
2985 * @param[in] item_flags
2986 * Bit-fields that holds the items detected until now.
2987 * @param[in] last_item
2988 * Previous validated item in the pattern items.
2989 * @param[in] ether_type
2990 * Type in the ethernet layer header (including dot1q).
2991 * @param[in] acc_mask
2992 * Acceptable mask, if NULL default internal default mask
2993 * will be used to check whether item fields are supported.
2995 * Pointer to error structure.
2998 * 0 on success, a negative errno value otherwise and rte_errno is set.
3001 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3002 uint64_t item_flags,
3004 uint16_t ether_type,
3005 const struct rte_flow_item_ecpri *acc_mask,
3006 struct rte_flow_error *error)
3008 const struct rte_flow_item_ecpri *mask = item->mask;
3009 const struct rte_flow_item_ecpri nic_mask = {
3013 RTE_BE32(((const struct rte_ecpri_common_hdr) {
3017 .dummy[0] = 0xFFFFFFFF,
3020 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3021 MLX5_FLOW_LAYER_OUTER_VLAN);
3022 struct rte_flow_item_ecpri mask_lo;
3024 if (!(last_item & outer_l2_vlan) &&
3025 last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3026 return rte_flow_error_set(error, EINVAL,
3027 RTE_FLOW_ERROR_TYPE_ITEM, item,
3028 "eCPRI can only follow L2/VLAN layer or UDP layer");
3029 if ((last_item & outer_l2_vlan) && ether_type &&
3030 ether_type != RTE_ETHER_TYPE_ECPRI)
3031 return rte_flow_error_set(error, EINVAL,
3032 RTE_FLOW_ERROR_TYPE_ITEM, item,
3033 "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3034 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3035 return rte_flow_error_set(error, EINVAL,
3036 RTE_FLOW_ERROR_TYPE_ITEM, item,
3037 "eCPRI with tunnel is not supported right now");
3038 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3039 return rte_flow_error_set(error, ENOTSUP,
3040 RTE_FLOW_ERROR_TYPE_ITEM, item,
3041 "multiple L3 layers not supported");
3042 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3043 return rte_flow_error_set(error, EINVAL,
3044 RTE_FLOW_ERROR_TYPE_ITEM, item,
3045 "eCPRI cannot coexist with a TCP layer");
3046 /* In specification, eCPRI could be over UDP layer. */
3047 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3048 return rte_flow_error_set(error, EINVAL,
3049 RTE_FLOW_ERROR_TYPE_ITEM, item,
3050 "eCPRI over UDP layer is not yet supported right now");
3051 /* Mask for type field in common header could be zero. */
3053 mask = &rte_flow_item_ecpri_mask;
3054 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3055 /* Input mask is in big-endian format. */
3056 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3057 return rte_flow_error_set(error, EINVAL,
3058 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3059 "partial mask is not supported for protocol");
3060 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3061 return rte_flow_error_set(error, EINVAL,
3062 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3063 "message header mask must be after a type mask");
3064 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3065 acc_mask ? (const uint8_t *)acc_mask
3066 : (const uint8_t *)&nic_mask,
3067 sizeof(struct rte_flow_item_ecpri),
3068 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3072 * Release resource related QUEUE/RSS action split.
3075 * Pointer to Ethernet device.
3077 * Flow to release id's from.
3080 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3081 struct rte_flow *flow)
3083 struct mlx5_priv *priv = dev->data->dev_private;
3084 uint32_t handle_idx;
3085 struct mlx5_flow_handle *dev_handle;
3087 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3088 handle_idx, dev_handle, next)
3089 if (dev_handle->split_flow_id &&
3090 !dev_handle->is_meter_flow_id)
3091 mlx5_ipool_free(priv->sh->ipool
3092 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3093 dev_handle->split_flow_id);
3097 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3098 const struct rte_flow_attr *attr __rte_unused,
3099 const struct rte_flow_item items[] __rte_unused,
3100 const struct rte_flow_action actions[] __rte_unused,
3101 bool external __rte_unused,
3102 int hairpin __rte_unused,
3103 struct rte_flow_error *error)
3105 return rte_flow_error_set(error, ENOTSUP,
3106 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3109 static struct mlx5_flow *
3110 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3111 const struct rte_flow_attr *attr __rte_unused,
3112 const struct rte_flow_item items[] __rte_unused,
3113 const struct rte_flow_action actions[] __rte_unused,
3114 struct rte_flow_error *error)
3116 rte_flow_error_set(error, ENOTSUP,
3117 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3122 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3123 struct mlx5_flow *dev_flow __rte_unused,
3124 const struct rte_flow_attr *attr __rte_unused,
3125 const struct rte_flow_item items[] __rte_unused,
3126 const struct rte_flow_action actions[] __rte_unused,
3127 struct rte_flow_error *error)
3129 return rte_flow_error_set(error, ENOTSUP,
3130 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3134 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3135 struct rte_flow *flow __rte_unused,
3136 struct rte_flow_error *error)
3138 return rte_flow_error_set(error, ENOTSUP,
3139 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3143 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3144 struct rte_flow *flow __rte_unused)
3149 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3150 struct rte_flow *flow __rte_unused)
3155 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3156 struct rte_flow *flow __rte_unused,
3157 const struct rte_flow_action *actions __rte_unused,
3158 void *data __rte_unused,
3159 struct rte_flow_error *error)
3161 return rte_flow_error_set(error, ENOTSUP,
3162 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3166 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3167 uint32_t domains __rte_unused,
3168 uint32_t flags __rte_unused)
3173 /* Void driver to protect from null pointer reference. */
3174 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3175 .validate = flow_null_validate,
3176 .prepare = flow_null_prepare,
3177 .translate = flow_null_translate,
3178 .apply = flow_null_apply,
3179 .remove = flow_null_remove,
3180 .destroy = flow_null_destroy,
3181 .query = flow_null_query,
3182 .sync_domain = flow_null_sync_domain,
3186 * Select flow driver type according to flow attributes and device
3190 * Pointer to the dev structure.
3192 * Pointer to the flow attributes.
3195 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3197 static enum mlx5_flow_drv_type
3198 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3200 struct mlx5_priv *priv = dev->data->dev_private;
3201 /* The OS can determine first a specific flow type (DV, VERBS) */
3202 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3204 if (type != MLX5_FLOW_TYPE_MAX)
3206 /* If no OS specific type - continue with DV/VERBS selection */
3207 if (attr->transfer && priv->config.dv_esw_en)
3208 type = MLX5_FLOW_TYPE_DV;
3209 if (!attr->transfer)
3210 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3211 MLX5_FLOW_TYPE_VERBS;
3215 #define flow_get_drv_ops(type) flow_drv_ops[type]
3218 * Flow driver validation API. This abstracts calling driver specific functions.
3219 * The type of flow driver is determined according to flow attributes.
3222 * Pointer to the dev structure.
3224 * Pointer to the flow attributes.
3226 * Pointer to the list of items.
3227 * @param[in] actions
3228 * Pointer to the list of actions.
3229 * @param[in] external
3230 * This flow rule is created by request external to PMD.
3231 * @param[in] hairpin
3232 * Number of hairpin TX actions, 0 means classic flow.
3234 * Pointer to the error structure.
3237 * 0 on success, a negative errno value otherwise and rte_errno is set.
3240 flow_drv_validate(struct rte_eth_dev *dev,
3241 const struct rte_flow_attr *attr,
3242 const struct rte_flow_item items[],
3243 const struct rte_flow_action actions[],
3244 bool external, int hairpin, struct rte_flow_error *error)
3246 const struct mlx5_flow_driver_ops *fops;
3247 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3249 fops = flow_get_drv_ops(type);
3250 return fops->validate(dev, attr, items, actions, external,
3255 * Flow driver preparation API. This abstracts calling driver specific
3256 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3257 * calculates the size of memory required for device flow, allocates the memory,
3258 * initializes the device flow and returns the pointer.
3261 * This function initializes device flow structure such as dv or verbs in
3262 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3263 * rest. For example, adding returning device flow to flow->dev_flow list and
3264 * setting backward reference to the flow should be done out of this function.
3265 * layers field is not filled either.
3268 * Pointer to the dev structure.
3270 * Pointer to the flow attributes.
3272 * Pointer to the list of items.
3273 * @param[in] actions
3274 * Pointer to the list of actions.
3275 * @param[in] flow_idx
3276 * This memory pool index to the flow.
3278 * Pointer to the error structure.
3281 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3283 static inline struct mlx5_flow *
3284 flow_drv_prepare(struct rte_eth_dev *dev,
3285 const struct rte_flow *flow,
3286 const struct rte_flow_attr *attr,
3287 const struct rte_flow_item items[],
3288 const struct rte_flow_action actions[],
3290 struct rte_flow_error *error)
3292 const struct mlx5_flow_driver_ops *fops;
3293 enum mlx5_flow_drv_type type = flow->drv_type;
3294 struct mlx5_flow *mlx5_flow = NULL;
3296 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3297 fops = flow_get_drv_ops(type);
3298 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3300 mlx5_flow->flow_idx = flow_idx;
3305 * Flow driver translation API. This abstracts calling driver specific
3306 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3307 * translates a generic flow into a driver flow. flow_drv_prepare() must
3311 * dev_flow->layers could be filled as a result of parsing during translation
3312 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3313 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3314 * flow->actions could be overwritten even though all the expanded dev_flows
3315 * have the same actions.
3318 * Pointer to the rte dev structure.
3319 * @param[in, out] dev_flow
3320 * Pointer to the mlx5 flow.
3322 * Pointer to the flow attributes.
3324 * Pointer to the list of items.
3325 * @param[in] actions
3326 * Pointer to the list of actions.
3328 * Pointer to the error structure.
3331 * 0 on success, a negative errno value otherwise and rte_errno is set.
3334 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3335 const struct rte_flow_attr *attr,
3336 const struct rte_flow_item items[],
3337 const struct rte_flow_action actions[],
3338 struct rte_flow_error *error)
3340 const struct mlx5_flow_driver_ops *fops;
3341 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3343 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3344 fops = flow_get_drv_ops(type);
3345 return fops->translate(dev, dev_flow, attr, items, actions, error);
3349 * Flow driver apply API. This abstracts calling driver specific functions.
3350 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3351 * translated driver flows on to device. flow_drv_translate() must precede.
3354 * Pointer to Ethernet device structure.
3355 * @param[in, out] flow
3356 * Pointer to flow structure.
3358 * Pointer to error structure.
3361 * 0 on success, a negative errno value otherwise and rte_errno is set.
3364 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3365 struct rte_flow_error *error)
3367 const struct mlx5_flow_driver_ops *fops;
3368 enum mlx5_flow_drv_type type = flow->drv_type;
3370 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3371 fops = flow_get_drv_ops(type);
3372 return fops->apply(dev, flow, error);
3376 * Flow driver destroy API. This abstracts calling driver specific functions.
3377 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3378 * on device and releases resources of the flow.
3381 * Pointer to Ethernet device.
3382 * @param[in, out] flow
3383 * Pointer to flow structure.
3386 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3388 const struct mlx5_flow_driver_ops *fops;
3389 enum mlx5_flow_drv_type type = flow->drv_type;
3391 flow_mreg_split_qrss_release(dev, flow);
3392 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3393 fops = flow_get_drv_ops(type);
3394 fops->destroy(dev, flow);
3398 * Flow driver find RSS policy tbl API. This abstracts calling driver
3399 * specific functions. Parent flow (rte_flow) should have driver
3400 * type (drv_type). It will find the RSS policy table that has the rss_desc.
3403 * Pointer to Ethernet device.
3404 * @param[in, out] flow
3405 * Pointer to flow structure.
3407 * Pointer to meter policy table.
3408 * @param[in] rss_desc
3409 * Pointer to rss_desc
3411 static struct mlx5_flow_meter_sub_policy *
3412 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3413 struct rte_flow *flow,
3414 struct mlx5_flow_meter_policy *policy,
3415 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3417 const struct mlx5_flow_driver_ops *fops;
3418 enum mlx5_flow_drv_type type = flow->drv_type;
3420 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3421 fops = flow_get_drv_ops(type);
3422 return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3426 * Get RSS action from the action list.
3429 * Pointer to Ethernet device.
3430 * @param[in] actions
3431 * Pointer to the list of actions.
3433 * Parent flow structure pointer.
3436 * Pointer to the RSS action if exist, else return NULL.
3438 static const struct rte_flow_action_rss*
3439 flow_get_rss_action(struct rte_eth_dev *dev,
3440 const struct rte_flow_action actions[])
3442 struct mlx5_priv *priv = dev->data->dev_private;
3443 const struct rte_flow_action_rss *rss = NULL;
3445 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3446 switch (actions->type) {
3447 case RTE_FLOW_ACTION_TYPE_RSS:
3448 rss = actions->conf;
3450 case RTE_FLOW_ACTION_TYPE_SAMPLE:
3452 const struct rte_flow_action_sample *sample =
3454 const struct rte_flow_action *act = sample->actions;
3455 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3456 if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3460 case RTE_FLOW_ACTION_TYPE_METER:
3463 struct mlx5_flow_meter_info *fm;
3464 struct mlx5_flow_meter_policy *policy;
3465 const struct rte_flow_action_meter *mtr = actions->conf;
3467 fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3469 policy = mlx5_flow_meter_policy_find(dev,
3470 fm->policy_id, NULL);
3471 if (policy && policy->is_rss)
3473 policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3485 * Get ASO age action by index.
3488 * Pointer to the Ethernet device structure.
3489 * @param[in] age_idx
3490 * Index to the ASO age action.
3493 * The specified ASO age action.
3495 struct mlx5_aso_age_action*
3496 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3498 uint16_t pool_idx = age_idx & UINT16_MAX;
3499 uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3500 struct mlx5_priv *priv = dev->data->dev_private;
3501 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3502 struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3504 return &pool->actions[offset - 1];
3507 /* maps indirect action to translated direct in some actions array */
3508 struct mlx5_translated_action_handle {
3509 struct rte_flow_action_handle *action; /**< Indirect action handle. */
3510 int index; /**< Index in related array of rte_flow_action. */
3514 * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3515 * direct action if translation possible.
3516 * This functionality used to run same execution path for both direct and
3517 * indirect actions on flow create. All necessary preparations for indirect
3518 * action handling should be performed on *handle* actions list returned
3522 * Pointer to Ethernet device.
3523 * @param[in] actions
3524 * List of actions to translate.
3525 * @param[out] handle
3526 * List to store translated indirect action object handles.
3527 * @param[in, out] indir_n
3528 * Size of *handle* array. On return should be updated with number of
3529 * indirect actions retrieved from the *actions* list.
3530 * @param[out] translated_actions
3531 * List of actions where all indirect actions were translated to direct
3532 * if possible. NULL if no translation took place.
3534 * Pointer to the error structure.
3537 * 0 on success, a negative errno value otherwise and rte_errno is set.
3540 flow_action_handles_translate(struct rte_eth_dev *dev,
3541 const struct rte_flow_action actions[],
3542 struct mlx5_translated_action_handle *handle,
3544 struct rte_flow_action **translated_actions,
3545 struct rte_flow_error *error)
3547 struct mlx5_priv *priv = dev->data->dev_private;
3548 struct rte_flow_action *translated = NULL;
3549 size_t actions_size;
3552 struct mlx5_translated_action_handle *handle_end = NULL;
3554 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3555 if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3557 if (copied_n == *indir_n) {
3558 return rte_flow_error_set
3559 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3560 NULL, "too many shared actions");
3562 rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3563 sizeof(actions[n].conf));
3564 handle[copied_n].index = n;
3568 *indir_n = copied_n;
3571 actions_size = sizeof(struct rte_flow_action) * n;
3572 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3577 memcpy(translated, actions, actions_size);
3578 for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3579 struct mlx5_shared_action_rss *shared_rss;
3580 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3581 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3582 uint32_t idx = act_idx &
3583 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3586 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3587 shared_rss = mlx5_ipool_get
3588 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3589 translated[handle->index].type =
3590 RTE_FLOW_ACTION_TYPE_RSS;
3591 translated[handle->index].conf =
3592 &shared_rss->origin;
3594 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3595 translated[handle->index].type =
3596 (enum rte_flow_action_type)
3597 MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3598 translated[handle->index].conf = (void *)(uintptr_t)idx;
3600 case MLX5_INDIRECT_ACTION_TYPE_AGE:
3601 if (priv->sh->flow_hit_aso_en) {
3602 translated[handle->index].type =
3603 (enum rte_flow_action_type)
3604 MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3605 translated[handle->index].conf =
3606 (void *)(uintptr_t)idx;
3610 case MLX5_INDIRECT_ACTION_TYPE_CT:
3611 if (priv->sh->ct_aso_en) {
3612 translated[handle->index].type =
3613 RTE_FLOW_ACTION_TYPE_CONNTRACK;
3614 translated[handle->index].conf =
3615 (void *)(uintptr_t)idx;
3620 mlx5_free(translated);
3621 return rte_flow_error_set
3622 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3623 NULL, "invalid indirect action type");
3626 *translated_actions = translated;
3631 * Get Shared RSS action from the action list.
3634 * Pointer to Ethernet device.
3636 * Pointer to the list of actions.
3637 * @param[in] shared_n
3638 * Actions list length.
3641 * The MLX5 RSS action ID if exists, otherwise return 0.
3644 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3645 struct mlx5_translated_action_handle *handle,
3648 struct mlx5_translated_action_handle *handle_end;
3649 struct mlx5_priv *priv = dev->data->dev_private;
3650 struct mlx5_shared_action_rss *shared_rss;
3653 for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3654 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3655 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3656 uint32_t idx = act_idx &
3657 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3659 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3660 shared_rss = mlx5_ipool_get
3661 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3663 __atomic_add_fetch(&shared_rss->refcnt, 1,
3674 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3676 const struct rte_flow_item *item;
3677 unsigned int has_vlan = 0;
3679 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3680 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3686 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3687 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3688 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3689 MLX5_EXPANSION_ROOT_OUTER;
3693 * Get layer flags from the prefix flow.
3695 * Some flows may be split to several subflows, the prefix subflow gets the
3696 * match items and the suffix sub flow gets the actions.
3697 * Some actions need the user defined match item flags to get the detail for
3699 * This function helps the suffix flow to get the item layer flags from prefix
3702 * @param[in] dev_flow
3703 * Pointer the created preifx subflow.
3706 * The layers get from prefix subflow.
3708 static inline uint64_t
3709 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3711 uint64_t layers = 0;
3714 * Layers bits could be localization, but usually the compiler will
3715 * help to do the optimization work for source code.
3716 * If no decap actions, use the layers directly.
3718 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3719 return dev_flow->handle->layers;
3720 /* Convert L3 layers with decap action. */
3721 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3722 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3723 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3724 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3725 /* Convert L4 layers with decap action. */
3726 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3727 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3728 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3729 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3734 * Get metadata split action information.
3736 * @param[in] actions
3737 * Pointer to the list of actions.
3739 * Pointer to the return pointer.
3740 * @param[out] qrss_type
3741 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3742 * if no QUEUE/RSS is found.
3743 * @param[out] encap_idx
3744 * Pointer to the index of the encap action if exists, otherwise the last
3748 * Total number of actions.
3751 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3752 const struct rte_flow_action **qrss,
3755 const struct rte_flow_action_raw_encap *raw_encap;
3757 int raw_decap_idx = -1;
3760 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3761 switch (actions->type) {
3762 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3763 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3764 *encap_idx = actions_n;
3766 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3767 raw_decap_idx = actions_n;
3769 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3770 raw_encap = actions->conf;
3771 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3772 *encap_idx = raw_decap_idx != -1 ?
3773 raw_decap_idx : actions_n;
3775 case RTE_FLOW_ACTION_TYPE_QUEUE:
3776 case RTE_FLOW_ACTION_TYPE_RSS:
3784 if (*encap_idx == -1)
3785 *encap_idx = actions_n;
3786 /* Count RTE_FLOW_ACTION_TYPE_END. */
3787 return actions_n + 1;
3791 * Check if the action will change packet.
3794 * Pointer to Ethernet device.
3799 * true if action will change packet, false otherwise.
3801 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3802 enum rte_flow_action_type type)
3804 struct mlx5_priv *priv = dev->data->dev_private;
3807 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3808 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3809 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3810 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3811 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3812 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3813 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3814 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3815 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3816 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3817 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3818 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3819 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3820 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3821 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3822 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3823 case RTE_FLOW_ACTION_TYPE_SET_META:
3824 case RTE_FLOW_ACTION_TYPE_SET_TAG:
3825 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3826 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3827 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3828 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3829 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3830 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3831 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3832 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3833 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3834 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3835 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3837 case RTE_FLOW_ACTION_TYPE_FLAG:
3838 case RTE_FLOW_ACTION_TYPE_MARK:
3839 if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3849 * Check meter action from the action list.
3852 * Pointer to Ethernet device.
3853 * @param[in] actions
3854 * Pointer to the list of actions.
3855 * @param[out] has_mtr
3856 * Pointer to the meter exist flag.
3857 * @param[out] has_modify
3858 * Pointer to the flag showing there's packet change action.
3859 * @param[out] meter_id
3860 * Pointer to the meter id.
3863 * Total number of actions.
3866 flow_check_meter_action(struct rte_eth_dev *dev,
3867 const struct rte_flow_action actions[],
3868 bool *has_mtr, bool *has_modify, uint32_t *meter_id)
3870 const struct rte_flow_action_meter *mtr = NULL;
3873 MLX5_ASSERT(has_mtr);
3875 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3876 switch (actions->type) {
3877 case RTE_FLOW_ACTION_TYPE_METER:
3878 mtr = actions->conf;
3879 *meter_id = mtr->mtr_id;
3886 *has_modify |= flow_check_modify_action_type(dev,
3890 /* Count RTE_FLOW_ACTION_TYPE_END. */
3891 return actions_n + 1;
3895 * Check if the flow should be split due to hairpin.
3896 * The reason for the split is that in current HW we can't
3897 * support encap and push-vlan on Rx, so if a flow contains
3898 * these actions we move it to Tx.
3901 * Pointer to Ethernet device.
3903 * Flow rule attributes.
3904 * @param[in] actions
3905 * Associated actions (list terminated by the END action).
3908 * > 0 the number of actions and the flow should be split,
3909 * 0 when no split required.
3912 flow_check_hairpin_split(struct rte_eth_dev *dev,
3913 const struct rte_flow_attr *attr,
3914 const struct rte_flow_action actions[])
3916 int queue_action = 0;
3919 const struct rte_flow_action_queue *queue;
3920 const struct rte_flow_action_rss *rss;
3921 const struct rte_flow_action_raw_encap *raw_encap;
3922 const struct rte_eth_hairpin_conf *conf;
3926 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3927 switch (actions->type) {
3928 case RTE_FLOW_ACTION_TYPE_QUEUE:
3929 queue = actions->conf;
3932 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3933 if (conf == NULL || conf->tx_explicit != 0)
3938 case RTE_FLOW_ACTION_TYPE_RSS:
3939 rss = actions->conf;
3940 if (rss == NULL || rss->queue_num == 0)
3942 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3943 if (conf == NULL || conf->tx_explicit != 0)
3948 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3949 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3950 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3951 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3952 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3956 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3957 raw_encap = actions->conf;
3958 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3967 if (split && queue_action)
3972 /* Declare flow create/destroy prototype in advance. */
3974 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3975 const struct rte_flow_attr *attr,
3976 const struct rte_flow_item items[],
3977 const struct rte_flow_action actions[],
3978 bool external, struct rte_flow_error *error);
3981 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3985 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3986 struct mlx5_hlist_entry *entry,
3987 uint64_t key, void *cb_ctx __rte_unused)
3989 struct mlx5_flow_mreg_copy_resource *mcp_res =
3990 container_of(entry, typeof(*mcp_res), hlist_ent);
3992 return mcp_res->mark_id != key;
3995 struct mlx5_hlist_entry *
3996 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3999 struct rte_eth_dev *dev = list->ctx;
4000 struct mlx5_priv *priv = dev->data->dev_private;
4001 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4002 struct mlx5_flow_mreg_copy_resource *mcp_res;
4003 struct rte_flow_error *error = ctx->error;
4006 uint32_t mark_id = key;
4007 struct rte_flow_attr attr = {
4008 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4011 struct mlx5_rte_flow_item_tag tag_spec = {
4014 struct rte_flow_item items[] = {
4015 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4017 struct rte_flow_action_mark ftag = {
4020 struct mlx5_flow_action_copy_mreg cp_mreg = {
4024 struct rte_flow_action_jump jump = {
4025 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4027 struct rte_flow_action actions[] = {
4028 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4031 /* Fill the register fileds in the flow. */
4032 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4036 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4040 /* Provide the full width of FLAG specific value. */
4041 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4042 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4043 /* Build a new flow. */
4044 if (mark_id != MLX5_DEFAULT_COPY_ID) {
4045 items[0] = (struct rte_flow_item){
4046 .type = (enum rte_flow_item_type)
4047 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4050 items[1] = (struct rte_flow_item){
4051 .type = RTE_FLOW_ITEM_TYPE_END,
4053 actions[0] = (struct rte_flow_action){
4054 .type = (enum rte_flow_action_type)
4055 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4058 actions[1] = (struct rte_flow_action){
4059 .type = (enum rte_flow_action_type)
4060 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4063 actions[2] = (struct rte_flow_action){
4064 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4067 actions[3] = (struct rte_flow_action){
4068 .type = RTE_FLOW_ACTION_TYPE_END,
4071 /* Default rule, wildcard match. */
4072 attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4073 items[0] = (struct rte_flow_item){
4074 .type = RTE_FLOW_ITEM_TYPE_END,
4076 actions[0] = (struct rte_flow_action){
4077 .type = (enum rte_flow_action_type)
4078 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4081 actions[1] = (struct rte_flow_action){
4082 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4085 actions[2] = (struct rte_flow_action){
4086 .type = RTE_FLOW_ACTION_TYPE_END,
4089 /* Build a new entry. */
4090 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4096 mcp_res->mark_id = mark_id;
4098 * The copy Flows are not included in any list. There
4099 * ones are referenced from other Flows and can not
4100 * be applied, removed, deleted in ardbitrary order
4101 * by list traversing.
4103 mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
4104 actions, false, error);
4105 if (!mcp_res->rix_flow) {
4106 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4109 return &mcp_res->hlist_ent;
4113 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4115 * As mark_id is unique, if there's already a registered flow for the mark_id,
4116 * return by increasing the reference counter of the resource. Otherwise, create
4117 * the resource (mcp_res) and flow.
4120 * - If ingress port is ANY and reg_c[1] is mark_id,
4121 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4123 * For default flow (zero mark_id), flow is like,
4124 * - If ingress port is ANY,
4125 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
4128 * Pointer to Ethernet device.
4130 * ID of MARK action, zero means default flow for META.
4132 * Perform verbose error reporting if not NULL.
4135 * Associated resource on success, NULL otherwise and rte_errno is set.
4137 static struct mlx5_flow_mreg_copy_resource *
4138 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4139 struct rte_flow_error *error)
4141 struct mlx5_priv *priv = dev->data->dev_private;
4142 struct mlx5_hlist_entry *entry;
4143 struct mlx5_flow_cb_ctx ctx = {
4148 /* Check if already registered. */
4149 MLX5_ASSERT(priv->mreg_cp_tbl);
4150 entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4153 return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4158 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
4160 struct mlx5_flow_mreg_copy_resource *mcp_res =
4161 container_of(entry, typeof(*mcp_res), hlist_ent);
4162 struct rte_eth_dev *dev = list->ctx;
4163 struct mlx5_priv *priv = dev->data->dev_private;
4165 MLX5_ASSERT(mcp_res->rix_flow);
4166 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
4167 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4171 * Release flow in RX_CP_TBL.
4174 * Pointer to Ethernet device.
4176 * Parent flow for wich copying is provided.
4179 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4180 struct rte_flow *flow)
4182 struct mlx5_flow_mreg_copy_resource *mcp_res;
4183 struct mlx5_priv *priv = dev->data->dev_private;
4185 if (!flow->rix_mreg_copy)
4187 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4188 flow->rix_mreg_copy);
4189 if (!mcp_res || !priv->mreg_cp_tbl)
4191 MLX5_ASSERT(mcp_res->rix_flow);
4192 mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4193 flow->rix_mreg_copy = 0;
4197 * Remove the default copy action from RX_CP_TBL.
4199 * This functions is called in the mlx5_dev_start(). No thread safe
4203 * Pointer to Ethernet device.
4206 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4208 struct mlx5_hlist_entry *entry;
4209 struct mlx5_priv *priv = dev->data->dev_private;
4211 /* Check if default flow is registered. */
4212 if (!priv->mreg_cp_tbl)
4214 entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4215 MLX5_DEFAULT_COPY_ID, NULL);
4218 mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4222 * Add the default copy action in in RX_CP_TBL.
4224 * This functions is called in the mlx5_dev_start(). No thread safe
4228 * Pointer to Ethernet device.
4230 * Perform verbose error reporting if not NULL.
4233 * 0 for success, negative value otherwise and rte_errno is set.
4236 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4237 struct rte_flow_error *error)
4239 struct mlx5_priv *priv = dev->data->dev_private;
4240 struct mlx5_flow_mreg_copy_resource *mcp_res;
4242 /* Check whether extensive metadata feature is engaged. */
4243 if (!priv->config.dv_flow_en ||
4244 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4245 !mlx5_flow_ext_mreg_supported(dev) ||
4246 !priv->sh->dv_regc0_mask)
4249 * Add default mreg copy flow may be called multiple time, but
4250 * only be called once in stop. Avoid register it twice.
4252 if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4254 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4261 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4263 * All the flow having Q/RSS action should be split by
4264 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4265 * performs the following,
4266 * - CQE->flow_tag := reg_c[1] (MARK)
4267 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4268 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4269 * but there should be a flow per each MARK ID set by MARK action.
4271 * For the aforementioned reason, if there's a MARK action in flow's action
4272 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4273 * the MARK ID to CQE's flow_tag like,
4274 * - If reg_c[1] is mark_id,
4275 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4277 * For SET_META action which stores value in reg_c[0], as the destination is
4278 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4279 * MARK ID means the default flow. The default flow looks like,
4280 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4283 * Pointer to Ethernet device.
4285 * Pointer to flow structure.
4286 * @param[in] actions
4287 * Pointer to the list of actions.
4289 * Perform verbose error reporting if not NULL.
4292 * 0 on success, negative value otherwise and rte_errno is set.
4295 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4296 struct rte_flow *flow,
4297 const struct rte_flow_action *actions,
4298 struct rte_flow_error *error)
4300 struct mlx5_priv *priv = dev->data->dev_private;
4301 struct mlx5_dev_config *config = &priv->config;
4302 struct mlx5_flow_mreg_copy_resource *mcp_res;
4303 const struct rte_flow_action_mark *mark;
4305 /* Check whether extensive metadata feature is engaged. */
4306 if (!config->dv_flow_en ||
4307 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4308 !mlx5_flow_ext_mreg_supported(dev) ||
4309 !priv->sh->dv_regc0_mask)
4311 /* Find MARK action. */
4312 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4313 switch (actions->type) {
4314 case RTE_FLOW_ACTION_TYPE_FLAG:
4315 mcp_res = flow_mreg_add_copy_action
4316 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4319 flow->rix_mreg_copy = mcp_res->idx;
4321 case RTE_FLOW_ACTION_TYPE_MARK:
4322 mark = (const struct rte_flow_action_mark *)
4325 flow_mreg_add_copy_action(dev, mark->id, error);
4328 flow->rix_mreg_copy = mcp_res->idx;
4337 #define MLX5_MAX_SPLIT_ACTIONS 24
4338 #define MLX5_MAX_SPLIT_ITEMS 24
4341 * Split the hairpin flow.
4342 * Since HW can't support encap and push-vlan on Rx, we move these
4344 * If the count action is after the encap then we also
4345 * move the count action. in this case the count will also measure
4349 * Pointer to Ethernet device.
4350 * @param[in] actions
4351 * Associated actions (list terminated by the END action).
4352 * @param[out] actions_rx
4354 * @param[out] actions_tx
4356 * @param[out] pattern_tx
4357 * The pattern items for the Tx flow.
4358 * @param[out] flow_id
4359 * The flow ID connected to this flow.
4365 flow_hairpin_split(struct rte_eth_dev *dev,
4366 const struct rte_flow_action actions[],
4367 struct rte_flow_action actions_rx[],
4368 struct rte_flow_action actions_tx[],
4369 struct rte_flow_item pattern_tx[],
4372 const struct rte_flow_action_raw_encap *raw_encap;
4373 const struct rte_flow_action_raw_decap *raw_decap;
4374 struct mlx5_rte_flow_action_set_tag *set_tag;
4375 struct rte_flow_action *tag_action;
4376 struct mlx5_rte_flow_item_tag *tag_item;
4377 struct rte_flow_item *item;
4381 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4382 switch (actions->type) {
4383 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4384 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4385 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4386 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4387 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4388 rte_memcpy(actions_tx, actions,
4389 sizeof(struct rte_flow_action));
4392 case RTE_FLOW_ACTION_TYPE_COUNT:
4394 rte_memcpy(actions_tx, actions,
4395 sizeof(struct rte_flow_action));
4398 rte_memcpy(actions_rx, actions,
4399 sizeof(struct rte_flow_action));
4403 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4404 raw_encap = actions->conf;
4405 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4406 memcpy(actions_tx, actions,
4407 sizeof(struct rte_flow_action));
4411 rte_memcpy(actions_rx, actions,
4412 sizeof(struct rte_flow_action));
4416 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4417 raw_decap = actions->conf;
4418 if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4419 memcpy(actions_tx, actions,
4420 sizeof(struct rte_flow_action));
4423 rte_memcpy(actions_rx, actions,
4424 sizeof(struct rte_flow_action));
4429 rte_memcpy(actions_rx, actions,
4430 sizeof(struct rte_flow_action));
4435 /* Add set meta action and end action for the Rx flow. */
4436 tag_action = actions_rx;
4437 tag_action->type = (enum rte_flow_action_type)
4438 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4440 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4442 set_tag = (void *)actions_rx;
4443 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4444 .id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4447 MLX5_ASSERT(set_tag->id > REG_NON);
4448 tag_action->conf = set_tag;
4449 /* Create Tx item list. */
4450 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4451 addr = (void *)&pattern_tx[2];
4453 item->type = (enum rte_flow_item_type)
4454 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4455 tag_item = (void *)addr;
4456 tag_item->data = flow_id;
4457 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4458 MLX5_ASSERT(set_tag->id > REG_NON);
4459 item->spec = tag_item;
4460 addr += sizeof(struct mlx5_rte_flow_item_tag);
4461 tag_item = (void *)addr;
4462 tag_item->data = UINT32_MAX;
4463 tag_item->id = UINT16_MAX;
4464 item->mask = tag_item;
4467 item->type = RTE_FLOW_ITEM_TYPE_END;
4472 * The last stage of splitting chain, just creates the subflow
4473 * without any modification.
4476 * Pointer to Ethernet device.
4478 * Parent flow structure pointer.
4479 * @param[in, out] sub_flow
4480 * Pointer to return the created subflow, may be NULL.
4482 * Flow rule attributes.
4484 * Pattern specification (list terminated by the END pattern item).
4485 * @param[in] actions
4486 * Associated actions (list terminated by the END action).
4487 * @param[in] flow_split_info
4488 * Pointer to flow split info structure.
4490 * Perform verbose error reporting if not NULL.
4492 * 0 on success, negative value otherwise
4495 flow_create_split_inner(struct rte_eth_dev *dev,
4496 struct rte_flow *flow,
4497 struct mlx5_flow **sub_flow,
4498 const struct rte_flow_attr *attr,
4499 const struct rte_flow_item items[],
4500 const struct rte_flow_action actions[],
4501 struct mlx5_flow_split_info *flow_split_info,
4502 struct rte_flow_error *error)
4504 struct mlx5_flow *dev_flow;
4506 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4507 flow_split_info->flow_idx, error);
4510 dev_flow->flow = flow;
4511 dev_flow->external = flow_split_info->external;
4512 dev_flow->skip_scale = flow_split_info->skip_scale;
4513 /* Subflow object was created, we must include one in the list. */
4514 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4515 dev_flow->handle, next);
4517 * If dev_flow is as one of the suffix flow, some actions in suffix
4518 * flow may need some user defined item layer flags, and pass the
4519 * Metadate rxq mark flag to suffix flow as well.
4521 if (flow_split_info->prefix_layers)
4522 dev_flow->handle->layers = flow_split_info->prefix_layers;
4523 if (flow_split_info->prefix_mark)
4524 dev_flow->handle->mark = 1;
4526 *sub_flow = dev_flow;
4527 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4528 dev_flow->dv.table_id = flow_split_info->table_id;
4530 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4534 * Get the sub policy of a meter.
4537 * Pointer to Ethernet device.
4539 * Parent flow structure pointer.
4540 * @param[in] policy_id;
4543 * Flow rule attributes.
4545 * Pattern specification (list terminated by the END pattern item).
4547 * Perform verbose error reporting if not NULL.
4550 * Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4552 static struct mlx5_flow_meter_sub_policy *
4553 get_meter_sub_policy(struct rte_eth_dev *dev,
4554 struct rte_flow *flow,
4556 const struct rte_flow_attr *attr,
4557 const struct rte_flow_item items[],
4558 struct rte_flow_error *error)
4560 struct mlx5_flow_meter_policy *policy;
4561 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4563 policy = mlx5_flow_meter_policy_find(dev, policy_id, NULL);
4565 rte_flow_error_set(error, EINVAL,
4566 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4567 "Failed to find Meter Policy.");
4570 if (policy->is_rss) {
4571 struct mlx5_flow_workspace *wks =
4572 mlx5_flow_get_thread_workspace();
4573 struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4574 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4579 * This is a tmp dev_flow,
4580 * no need to register any matcher for it in translate.
4582 wks->skip_matcher_reg = 1;
4583 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4584 struct mlx5_flow dev_flow = {0};
4585 struct mlx5_flow_handle dev_handle = { {0} };
4586 const void *rss_act = policy->act_cnt[i].rss->conf;
4587 struct rte_flow_action rss_actions[2] = {
4589 .type = RTE_FLOW_ACTION_TYPE_RSS,
4593 .type = RTE_FLOW_ACTION_TYPE_END,
4598 dev_flow.handle = &dev_handle;
4599 dev_flow.ingress = attr->ingress;
4600 dev_flow.flow = flow;
4601 dev_flow.external = 0;
4602 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4603 dev_flow.dv.transfer = attr->transfer;
4605 /* Translate RSS action to get rss hash fields. */
4606 if (flow_drv_translate(dev, &dev_flow, attr,
4607 items, rss_actions, error))
4609 rss_desc_v[i] = wks->rss_desc;
4610 rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4611 rss_desc_v[i].hash_fields = dev_flow.hash_fields;
4612 rss_desc_v[i].queue_num = rss_desc_v[i].hash_fields ?
4613 rss_desc_v[i].queue_num : 1;
4614 rss_desc[i] = &rss_desc_v[i];
4616 sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4617 flow, policy, rss_desc);
4619 enum mlx5_meter_domain mtr_domain =
4620 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4621 attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4622 MLX5_MTR_DOMAIN_INGRESS;
4623 sub_policy = policy->sub_policys[mtr_domain][0];
4626 rte_flow_error_set(error, EINVAL,
4627 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4628 "Failed to get meter sub-policy.");
4636 * Split the meter flow.
4638 * As meter flow will split to three sub flow, other than meter
4639 * action, the other actions make sense to only meter accepts
4640 * the packet. If it need to be dropped, no other additional
4641 * actions should be take.
4643 * One kind of special action which decapsulates the L3 tunnel
4644 * header will be in the prefix sub flow, as not to take the
4645 * L3 tunnel header into account.
4648 * Pointer to Ethernet device.
4650 * Parent flow structure pointer.
4652 * Pointer to flow meter structure.
4654 * Flow rule attributes.
4656 * Pattern specification (list terminated by the END pattern item).
4657 * @param[out] sfx_items
4658 * Suffix flow match items (list terminated by the END pattern item).
4659 * @param[in] actions
4660 * Associated actions (list terminated by the END action).
4661 * @param[out] actions_sfx
4662 * Suffix flow actions.
4663 * @param[out] actions_pre
4664 * Prefix flow actions.
4665 * @param[out] mtr_flow_id
4666 * Pointer to meter flow id.
4668 * Perform verbose error reporting if not NULL.
4671 * 0 on success, a negative errno value otherwise and rte_errno is set.
4674 flow_meter_split_prep(struct rte_eth_dev *dev,
4675 struct rte_flow *flow,
4676 struct mlx5_flow_meter_info *fm,
4677 const struct rte_flow_attr *attr,
4678 const struct rte_flow_item items[],
4679 struct rte_flow_item sfx_items[],
4680 const struct rte_flow_action actions[],
4681 struct rte_flow_action actions_sfx[],
4682 struct rte_flow_action actions_pre[],
4683 uint32_t *mtr_flow_id,
4684 struct rte_flow_error *error)
4686 struct mlx5_priv *priv = dev->data->dev_private;
4687 struct rte_flow_action *tag_action = NULL;
4688 struct rte_flow_item *tag_item;
4689 struct mlx5_rte_flow_action_set_tag *set_tag;
4690 const struct rte_flow_action_raw_encap *raw_encap;
4691 const struct rte_flow_action_raw_decap *raw_decap;
4692 struct mlx5_rte_flow_item_tag *tag_item_spec;
4693 struct mlx5_rte_flow_item_tag *tag_item_mask;
4694 uint32_t tag_id = 0;
4695 bool copy_vlan = false;
4696 struct rte_flow_action *hw_mtr_action;
4697 struct rte_flow_action *action_pre_head = NULL;
4698 bool mtr_first = priv->sh->meter_aso_en &&
4700 (attr->transfer && priv->representor_id != UINT16_MAX));
4701 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4702 uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4703 MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4704 uint32_t flow_id = 0;
4705 uint32_t flow_id_reversed = 0;
4706 uint8_t flow_id_bits = 0;
4709 /* For ASO meter, meter must be before tag in TX direction. */
4711 action_pre_head = actions_pre++;
4712 /* Leave space for tag action. */
4713 tag_action = actions_pre++;
4715 /* Prepare the actions for prefix and suffix flow. */
4716 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4717 struct rte_flow_action *action_cur = NULL;
4719 switch (actions->type) {
4720 case RTE_FLOW_ACTION_TYPE_METER:
4722 action_cur = action_pre_head;
4724 /* Leave space for tag action. */
4725 tag_action = actions_pre++;
4726 action_cur = actions_pre++;
4729 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4730 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4731 action_cur = actions_pre++;
4733 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4734 raw_encap = actions->conf;
4735 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4736 action_cur = actions_pre++;
4738 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4739 raw_decap = actions->conf;
4740 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4741 action_cur = actions_pre++;
4743 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4744 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4751 action_cur = (fm->def_policy) ?
4752 actions_sfx++ : actions_pre++;
4753 memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4755 /* Add end action to the actions. */
4756 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4757 if (priv->sh->meter_aso_en) {
4759 * For ASO meter, need to add an extra jump action explicitly,
4760 * to jump from meter to policer table.
4762 struct mlx5_flow_meter_sub_policy *sub_policy;
4763 struct mlx5_flow_tbl_data_entry *tbl_data;
4765 if (!fm->def_policy) {
4766 sub_policy = get_meter_sub_policy(dev, flow,
4767 fm->policy_id, attr,
4772 enum mlx5_meter_domain mtr_domain =
4773 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4774 attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4775 MLX5_MTR_DOMAIN_INGRESS;
4778 &priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
4780 tbl_data = container_of(sub_policy->tbl_rsc,
4781 struct mlx5_flow_tbl_data_entry, tbl);
4782 hw_mtr_action = actions_pre++;
4783 hw_mtr_action->type = (enum rte_flow_action_type)
4784 MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
4785 hw_mtr_action->conf = tbl_data->jump.action;
4787 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4790 return rte_flow_error_set(error, ENOMEM,
4791 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4792 "No tag action space.");
4794 tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
4797 /* Only default-policy Meter creates mtr flow id. */
4798 if (fm->def_policy) {
4799 mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4801 return rte_flow_error_set(error, ENOMEM,
4802 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4803 "Failed to allocate meter flow id.");
4804 flow_id = tag_id - 1;
4805 flow_id_bits = (!flow_id) ? 1 :
4806 (MLX5_REG_BITS - __builtin_clz(flow_id));
4807 if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
4809 mlx5_ipool_free(fm->flow_ipool, tag_id);
4810 return rte_flow_error_set(error, EINVAL,
4811 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4812 "Meter flow id exceeds max limit.");
4814 if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
4815 priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
4817 /* Prepare the suffix subflow items. */
4818 tag_item = sfx_items++;
4819 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4820 int item_type = items->type;
4822 switch (item_type) {
4823 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4824 memcpy(sfx_items, items, sizeof(*sfx_items));
4827 case RTE_FLOW_ITEM_TYPE_VLAN:
4829 memcpy(sfx_items, items, sizeof(*sfx_items));
4831 * Convert to internal match item, it is used
4832 * for vlan push and set vid.
4834 sfx_items->type = (enum rte_flow_item_type)
4835 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4843 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4845 /* Build tag actions and items for meter_id/meter flow_id. */
4846 set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
4847 tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4848 tag_item_mask = tag_item_spec + 1;
4849 /* Both flow_id and meter_id share the same register. */
4850 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4851 .id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
4853 .offset = mtr_id_offset,
4854 .length = mtr_reg_bits,
4855 .data = flow->meter,
4858 * The color Reg bits used by flow_id are growing from
4859 * msb to lsb, so must do bit reverse for flow_id val in RegC.
4861 for (shift = 0; shift < flow_id_bits; shift++)
4862 flow_id_reversed = (flow_id_reversed << 1) |
4863 ((flow_id >> shift) & 0x1);
4865 flow_id_reversed << (mtr_reg_bits - flow_id_bits);
4866 tag_item_spec->id = set_tag->id;
4867 tag_item_spec->data = set_tag->data << mtr_id_offset;
4868 tag_item_mask->data = UINT32_MAX << mtr_id_offset;
4869 tag_action->type = (enum rte_flow_action_type)
4870 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4871 tag_action->conf = set_tag;
4872 tag_item->type = (enum rte_flow_item_type)
4873 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4874 tag_item->spec = tag_item_spec;
4875 tag_item->last = NULL;
4876 tag_item->mask = tag_item_mask;
4879 *mtr_flow_id = tag_id;
4884 * Split action list having QUEUE/RSS for metadata register copy.
4886 * Once Q/RSS action is detected in user's action list, the flow action
4887 * should be split in order to copy metadata registers, which will happen in
4889 * - CQE->flow_tag := reg_c[1] (MARK)
4890 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4891 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4892 * This is because the last action of each flow must be a terminal action
4893 * (QUEUE, RSS or DROP).
4895 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4896 * stored and kept in the mlx5_flow structure per each sub_flow.
4898 * The Q/RSS action is replaced with,
4899 * - SET_TAG, setting the allocated flow ID to reg_c[2].
4900 * And the following JUMP action is added at the end,
4901 * - JUMP, to RX_CP_TBL.
4903 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4904 * flow_create_split_metadata() routine. The flow will look like,
4905 * - If flow ID matches (reg_c[2]), perform Q/RSS.
4908 * Pointer to Ethernet device.
4909 * @param[out] split_actions
4910 * Pointer to store split actions to jump to CP_TBL.
4911 * @param[in] actions
4912 * Pointer to the list of original flow actions.
4914 * Pointer to the Q/RSS action.
4915 * @param[in] actions_n
4916 * Number of original actions.
4918 * Perform verbose error reporting if not NULL.
4921 * non-zero unique flow_id on success, otherwise 0 and
4922 * error/rte_error are set.
4925 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4926 struct rte_flow_action *split_actions,
4927 const struct rte_flow_action *actions,
4928 const struct rte_flow_action *qrss,
4929 int actions_n, struct rte_flow_error *error)
4931 struct mlx5_priv *priv = dev->data->dev_private;
4932 struct mlx5_rte_flow_action_set_tag *set_tag;
4933 struct rte_flow_action_jump *jump;
4934 const int qrss_idx = qrss - actions;
4935 uint32_t flow_id = 0;
4939 * Given actions will be split
4940 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4941 * - Add jump to mreg CP_TBL.
4942 * As a result, there will be one more action.
4945 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4946 set_tag = (void *)(split_actions + actions_n);
4948 * If tag action is not set to void(it means we are not the meter
4949 * suffix flow), add the tag action. Since meter suffix flow already
4950 * has the tag added.
4952 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4954 * Allocate the new subflow ID. This one is unique within
4955 * device and not shared with representors. Otherwise,
4956 * we would have to resolve multi-thread access synch
4957 * issue. Each flow on the shared device is appended
4958 * with source vport identifier, so the resulting
4959 * flows will be unique in the shared (by master and
4960 * representors) domain even if they have coinciding
4963 mlx5_ipool_malloc(priv->sh->ipool
4964 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4966 return rte_flow_error_set(error, ENOMEM,
4967 RTE_FLOW_ERROR_TYPE_ACTION,
4968 NULL, "can't allocate id "
4969 "for split Q/RSS subflow");
4970 /* Internal SET_TAG action to set flow ID. */
4971 *set_tag = (struct mlx5_rte_flow_action_set_tag){
4974 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4978 /* Construct new actions array. */
4979 /* Replace QUEUE/RSS action. */
4980 split_actions[qrss_idx] = (struct rte_flow_action){
4981 .type = (enum rte_flow_action_type)
4982 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4986 /* JUMP action to jump to mreg copy table (CP_TBL). */
4987 jump = (void *)(set_tag + 1);
4988 *jump = (struct rte_flow_action_jump){
4989 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4991 split_actions[actions_n - 2] = (struct rte_flow_action){
4992 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4995 split_actions[actions_n - 1] = (struct rte_flow_action){
4996 .type = RTE_FLOW_ACTION_TYPE_END,
5002 * Extend the given action list for Tx metadata copy.
5004 * Copy the given action list to the ext_actions and add flow metadata register
5005 * copy action in order to copy reg_a set by WQE to reg_c[0].
5007 * @param[out] ext_actions
5008 * Pointer to the extended action list.
5009 * @param[in] actions
5010 * Pointer to the list of actions.
5011 * @param[in] actions_n
5012 * Number of actions in the list.
5014 * Perform verbose error reporting if not NULL.
5015 * @param[in] encap_idx
5016 * The encap action inndex.
5019 * 0 on success, negative value otherwise
5022 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5023 struct rte_flow_action *ext_actions,
5024 const struct rte_flow_action *actions,
5025 int actions_n, struct rte_flow_error *error,
5028 struct mlx5_flow_action_copy_mreg *cp_mreg =
5029 (struct mlx5_flow_action_copy_mreg *)
5030 (ext_actions + actions_n + 1);
5033 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5037 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5042 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5043 if (encap_idx == actions_n - 1) {
5044 ext_actions[actions_n - 1] = (struct rte_flow_action){
5045 .type = (enum rte_flow_action_type)
5046 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5049 ext_actions[actions_n] = (struct rte_flow_action){
5050 .type = RTE_FLOW_ACTION_TYPE_END,
5053 ext_actions[encap_idx] = (struct rte_flow_action){
5054 .type = (enum rte_flow_action_type)
5055 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5058 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5059 sizeof(*ext_actions) * (actions_n - encap_idx));
5065 * Check the match action from the action list.
5067 * @param[in] actions
5068 * Pointer to the list of actions.
5070 * Flow rule attributes.
5072 * The action to be check if exist.
5073 * @param[out] match_action_pos
5074 * Pointer to the position of the matched action if exists, otherwise is -1.
5075 * @param[out] qrss_action_pos
5076 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5077 * @param[out] modify_after_mirror
5078 * Pointer to the flag of modify action after FDB mirroring.
5081 * > 0 the total number of actions.
5082 * 0 if not found match action in action list.
5085 flow_check_match_action(const struct rte_flow_action actions[],
5086 const struct rte_flow_attr *attr,
5087 enum rte_flow_action_type action,
5088 int *match_action_pos, int *qrss_action_pos,
5089 int *modify_after_mirror)
5091 const struct rte_flow_action_sample *sample;
5098 *match_action_pos = -1;
5099 *qrss_action_pos = -1;
5100 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5101 if (actions->type == action) {
5103 *match_action_pos = actions_n;
5105 switch (actions->type) {
5106 case RTE_FLOW_ACTION_TYPE_QUEUE:
5107 case RTE_FLOW_ACTION_TYPE_RSS:
5108 *qrss_action_pos = actions_n;
5110 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5111 sample = actions->conf;
5112 ratio = sample->ratio;
5113 sub_type = ((const struct rte_flow_action *)
5114 (sample->actions))->type;
5115 if (ratio == 1 && attr->transfer)
5118 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5119 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5120 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5121 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5122 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5123 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5124 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5125 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5126 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5127 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5128 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5129 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5130 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5131 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5132 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5133 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5134 case RTE_FLOW_ACTION_TYPE_FLAG:
5135 case RTE_FLOW_ACTION_TYPE_MARK:
5136 case RTE_FLOW_ACTION_TYPE_SET_META:
5137 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5138 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5139 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5140 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5141 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5142 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5143 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5144 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5145 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5146 case RTE_FLOW_ACTION_TYPE_METER:
5148 *modify_after_mirror = 1;
5155 if (flag && fdb_mirror && !*modify_after_mirror) {
5156 /* FDB mirroring uses the destination array to implement
5157 * instead of FLOW_SAMPLER object.
5159 if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5162 /* Count RTE_FLOW_ACTION_TYPE_END. */
5163 return flag ? actions_n + 1 : 0;
5166 #define SAMPLE_SUFFIX_ITEM 2
5169 * Split the sample flow.
5171 * As sample flow will split to two sub flow, sample flow with
5172 * sample action, the other actions will move to new suffix flow.
5174 * Also add unique tag id with tag action in the sample flow,
5175 * the same tag id will be as match in the suffix flow.
5178 * Pointer to Ethernet device.
5179 * @param[in] add_tag
5180 * Add extra tag action flag.
5181 * @param[out] sfx_items
5182 * Suffix flow match items (list terminated by the END pattern item).
5183 * @param[in] actions
5184 * Associated actions (list terminated by the END action).
5185 * @param[out] actions_sfx
5186 * Suffix flow actions.
5187 * @param[out] actions_pre
5188 * Prefix flow actions.
5189 * @param[in] actions_n
5190 * The total number of actions.
5191 * @param[in] sample_action_pos
5192 * The sample action position.
5193 * @param[in] qrss_action_pos
5194 * The Queue/RSS action position.
5195 * @param[in] jump_table
5196 * Add extra jump action flag.
5198 * Perform verbose error reporting if not NULL.
5201 * 0 on success, or unique flow_id, a negative errno value
5202 * otherwise and rte_errno is set.
5205 flow_sample_split_prep(struct rte_eth_dev *dev,
5207 struct rte_flow_item sfx_items[],
5208 const struct rte_flow_action actions[],
5209 struct rte_flow_action actions_sfx[],
5210 struct rte_flow_action actions_pre[],
5212 int sample_action_pos,
5213 int qrss_action_pos,
5215 struct rte_flow_error *error)
5217 struct mlx5_priv *priv = dev->data->dev_private;
5218 struct mlx5_rte_flow_action_set_tag *set_tag;
5219 struct mlx5_rte_flow_item_tag *tag_spec;
5220 struct mlx5_rte_flow_item_tag *tag_mask;
5221 struct rte_flow_action_jump *jump_action;
5222 uint32_t tag_id = 0;
5224 int append_index = 0;
5227 if (sample_action_pos < 0)
5228 return rte_flow_error_set(error, EINVAL,
5229 RTE_FLOW_ERROR_TYPE_ACTION,
5230 NULL, "invalid position of sample "
5232 /* Prepare the actions for prefix and suffix flow. */
5233 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5234 index = qrss_action_pos;
5235 /* Put the preceding the Queue/RSS action into prefix flow. */
5237 memcpy(actions_pre, actions,
5238 sizeof(struct rte_flow_action) * index);
5239 /* Put others preceding the sample action into prefix flow. */
5240 if (sample_action_pos > index + 1)
5241 memcpy(actions_pre + index, actions + index + 1,
5242 sizeof(struct rte_flow_action) *
5243 (sample_action_pos - index - 1));
5244 index = sample_action_pos - 1;
5245 /* Put Queue/RSS action into Suffix flow. */
5246 memcpy(actions_sfx, actions + qrss_action_pos,
5247 sizeof(struct rte_flow_action));
5250 index = sample_action_pos;
5252 memcpy(actions_pre, actions,
5253 sizeof(struct rte_flow_action) * index);
5255 /* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5256 * For CX6DX and above, metadata registers Cx preserve their value,
5257 * add an extra tag action for NIC-RX and E-Switch Domain.
5260 /* Prepare the prefix tag action. */
5262 set_tag = (void *)(actions_pre + actions_n + append_index);
5263 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5266 mlx5_ipool_malloc(priv->sh->ipool
5267 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5268 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5272 /* Prepare the suffix subflow items. */
5273 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5274 tag_spec->data = tag_id;
5275 tag_spec->id = set_tag->id;
5276 tag_mask = tag_spec + 1;
5277 tag_mask->data = UINT32_MAX;
5278 sfx_items[0] = (struct rte_flow_item){
5279 .type = (enum rte_flow_item_type)
5280 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5285 sfx_items[1] = (struct rte_flow_item){
5286 .type = (enum rte_flow_item_type)
5287 RTE_FLOW_ITEM_TYPE_END,
5289 /* Prepare the tag action in prefix subflow. */
5290 actions_pre[index++] =
5291 (struct rte_flow_action){
5292 .type = (enum rte_flow_action_type)
5293 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5297 memcpy(actions_pre + index, actions + sample_action_pos,
5298 sizeof(struct rte_flow_action));
5300 /* For the modify action after the sample action in E-Switch mirroring,
5301 * Add the extra jump action in prefix subflow and jump into the next
5302 * table, then do the modify action in the new table.
5305 /* Prepare the prefix jump action. */
5307 jump_action = (void *)(actions_pre + actions_n + append_index);
5308 jump_action->group = jump_table;
5309 actions_pre[index++] =
5310 (struct rte_flow_action){
5311 .type = (enum rte_flow_action_type)
5312 RTE_FLOW_ACTION_TYPE_JUMP,
5313 .conf = jump_action,
5316 actions_pre[index] = (struct rte_flow_action){
5317 .type = (enum rte_flow_action_type)
5318 RTE_FLOW_ACTION_TYPE_END,
5320 /* Put the actions after sample into Suffix flow. */
5321 memcpy(actions_sfx, actions + sample_action_pos + 1,
5322 sizeof(struct rte_flow_action) *
5323 (actions_n - sample_action_pos - 1));
5328 * The splitting for metadata feature.
5330 * - Q/RSS action on NIC Rx should be split in order to pass by
5331 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5332 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5334 * - All the actions on NIC Tx should have a mreg copy action to
5335 * copy reg_a from WQE to reg_c[0].
5338 * Pointer to Ethernet device.
5340 * Parent flow structure pointer.
5342 * Flow rule attributes.
5344 * Pattern specification (list terminated by the END pattern item).
5345 * @param[in] actions
5346 * Associated actions (list terminated by the END action).
5347 * @param[in] flow_split_info
5348 * Pointer to flow split info structure.
5350 * Perform verbose error reporting if not NULL.
5352 * 0 on success, negative value otherwise
5355 flow_create_split_metadata(struct rte_eth_dev *dev,
5356 struct rte_flow *flow,
5357 const struct rte_flow_attr *attr,
5358 const struct rte_flow_item items[],
5359 const struct rte_flow_action actions[],
5360 struct mlx5_flow_split_info *flow_split_info,
5361 struct rte_flow_error *error)
5363 struct mlx5_priv *priv = dev->data->dev_private;
5364 struct mlx5_dev_config *config = &priv->config;
5365 const struct rte_flow_action *qrss = NULL;
5366 struct rte_flow_action *ext_actions = NULL;
5367 struct mlx5_flow *dev_flow = NULL;
5368 uint32_t qrss_id = 0;
5375 /* Check whether extensive metadata feature is engaged. */
5376 if (!config->dv_flow_en ||
5377 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5378 !mlx5_flow_ext_mreg_supported(dev))
5379 return flow_create_split_inner(dev, flow, NULL, attr, items,
5380 actions, flow_split_info, error);
5381 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5384 /* Exclude hairpin flows from splitting. */
5385 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5386 const struct rte_flow_action_queue *queue;
5389 if (mlx5_rxq_get_type(dev, queue->index) ==
5390 MLX5_RXQ_TYPE_HAIRPIN)
5392 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5393 const struct rte_flow_action_rss *rss;
5396 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5397 MLX5_RXQ_TYPE_HAIRPIN)
5402 /* Check if it is in meter suffix table. */
5403 mtr_sfx = attr->group == (attr->transfer ?
5404 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5405 MLX5_FLOW_TABLE_LEVEL_METER);
5407 * Q/RSS action on NIC Rx should be split in order to pass by
5408 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5409 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5411 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5412 sizeof(struct rte_flow_action_set_tag) +
5413 sizeof(struct rte_flow_action_jump);
5414 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5417 return rte_flow_error_set(error, ENOMEM,
5418 RTE_FLOW_ERROR_TYPE_ACTION,
5419 NULL, "no memory to split "
5422 * If we are the suffix flow of meter, tag already exist.
5423 * Set the tag action to void.
5426 ext_actions[qrss - actions].type =
5427 RTE_FLOW_ACTION_TYPE_VOID;
5429 ext_actions[qrss - actions].type =
5430 (enum rte_flow_action_type)
5431 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5433 * Create the new actions list with removed Q/RSS action
5434 * and appended set tag and jump to register copy table
5435 * (RX_CP_TBL). We should preallocate unique tag ID here
5436 * in advance, because it is needed for set tag action.
5438 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5439 qrss, actions_n, error);
5440 if (!mtr_sfx && !qrss_id) {
5444 } else if (attr->egress && !attr->transfer) {
5446 * All the actions on NIC Tx should have a metadata register
5447 * copy action to copy reg_a from WQE to reg_c[meta]
5449 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5450 sizeof(struct mlx5_flow_action_copy_mreg);
5451 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5454 return rte_flow_error_set(error, ENOMEM,
5455 RTE_FLOW_ERROR_TYPE_ACTION,
5456 NULL, "no memory to split "
5458 /* Create the action list appended with copy register. */
5459 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5460 actions_n, error, encap_idx);
5464 /* Add the unmodified original or prefix subflow. */
5465 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5466 items, ext_actions ? ext_actions :
5467 actions, flow_split_info, error);
5470 MLX5_ASSERT(dev_flow);
5472 const struct rte_flow_attr q_attr = {
5473 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5476 /* Internal PMD action to set register. */
5477 struct mlx5_rte_flow_item_tag q_tag_spec = {
5481 struct rte_flow_item q_items[] = {
5483 .type = (enum rte_flow_item_type)
5484 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5485 .spec = &q_tag_spec,
5490 .type = RTE_FLOW_ITEM_TYPE_END,
5493 struct rte_flow_action q_actions[] = {
5499 .type = RTE_FLOW_ACTION_TYPE_END,
5502 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5505 * Configure the tag item only if there is no meter subflow.
5506 * Since tag is already marked in the meter suffix subflow
5507 * we can just use the meter suffix items as is.
5510 /* Not meter subflow. */
5511 MLX5_ASSERT(!mtr_sfx);
5513 * Put unique id in prefix flow due to it is destroyed
5514 * after suffix flow and id will be freed after there
5515 * is no actual flows with this id and identifier
5516 * reallocation becomes possible (for example, for
5517 * other flows in other threads).
5519 dev_flow->handle->split_flow_id = qrss_id;
5520 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5524 q_tag_spec.id = ret;
5527 /* Add suffix subflow to execute Q/RSS. */
5528 flow_split_info->prefix_layers = layers;
5529 flow_split_info->prefix_mark = 0;
5530 ret = flow_create_split_inner(dev, flow, &dev_flow,
5531 &q_attr, mtr_sfx ? items :
5533 flow_split_info, error);
5536 /* qrss ID should be freed if failed. */
5538 MLX5_ASSERT(dev_flow);
5543 * We do not destroy the partially created sub_flows in case of error.
5544 * These ones are included into parent flow list and will be destroyed
5545 * by flow_drv_destroy.
5547 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5549 mlx5_free(ext_actions);
5554 * Create meter internal drop flow with the original pattern.
5557 * Pointer to Ethernet device.
5559 * Parent flow structure pointer.
5561 * Flow rule attributes.
5563 * Pattern specification (list terminated by the END pattern item).
5564 * @param[in] flow_split_info
5565 * Pointer to flow split info structure.
5567 * Pointer to flow meter structure.
5569 * Perform verbose error reporting if not NULL.
5571 * 0 on success, negative value otherwise
5574 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5575 struct rte_flow *flow,
5576 const struct rte_flow_attr *attr,
5577 const struct rte_flow_item items[],
5578 struct mlx5_flow_split_info *flow_split_info,
5579 struct mlx5_flow_meter_info *fm,
5580 struct rte_flow_error *error)
5582 struct mlx5_flow *dev_flow = NULL;
5583 struct rte_flow_attr drop_attr = *attr;
5584 struct rte_flow_action drop_actions[3];
5585 struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5587 MLX5_ASSERT(fm->drop_cnt);
5588 drop_actions[0].type =
5589 (enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5590 drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5591 drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5592 drop_actions[1].conf = NULL;
5593 drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5594 drop_actions[2].conf = NULL;
5595 drop_split_info.external = false;
5596 drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5597 drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5598 drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5599 return flow_create_split_inner(dev, flow, &dev_flow,
5600 &drop_attr, items, drop_actions,
5601 &drop_split_info, error);
5605 * The splitting for meter feature.
5607 * - The meter flow will be split to two flows as prefix and
5608 * suffix flow. The packets make sense only it pass the prefix
5611 * - Reg_C_5 is used for the packet to match betweend prefix and
5615 * Pointer to Ethernet device.
5617 * Parent flow structure pointer.
5619 * Flow rule attributes.
5621 * Pattern specification (list terminated by the END pattern item).
5622 * @param[in] actions
5623 * Associated actions (list terminated by the END action).
5624 * @param[in] flow_split_info
5625 * Pointer to flow split info structure.
5627 * Perform verbose error reporting if not NULL.
5629 * 0 on success, negative value otherwise
5632 flow_create_split_meter(struct rte_eth_dev *dev,
5633 struct rte_flow *flow,
5634 const struct rte_flow_attr *attr,
5635 const struct rte_flow_item items[],
5636 const struct rte_flow_action actions[],
5637 struct mlx5_flow_split_info *flow_split_info,
5638 struct rte_flow_error *error)
5640 struct mlx5_priv *priv = dev->data->dev_private;
5641 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5642 struct rte_flow_action *sfx_actions = NULL;
5643 struct rte_flow_action *pre_actions = NULL;
5644 struct rte_flow_item *sfx_items = NULL;
5645 struct mlx5_flow *dev_flow = NULL;
5646 struct rte_flow_attr sfx_attr = *attr;
5647 struct mlx5_flow_meter_info *fm = NULL;
5648 uint8_t skip_scale_restore;
5649 bool has_mtr = false;
5650 bool has_modify = false;
5651 bool set_mtr_reg = true;
5652 uint32_t meter_id = 0;
5653 uint32_t mtr_idx = 0;
5654 uint32_t mtr_flow_id = 0;
5661 actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5662 &has_modify, &meter_id);
5665 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5667 return rte_flow_error_set(error, EINVAL,
5668 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5669 NULL, "Meter not found.");
5671 fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5673 return rte_flow_error_set(error, EINVAL,
5674 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5675 NULL, "Meter not found.");
5676 ret = mlx5_flow_meter_attach(priv, fm,
5680 flow->meter = mtr_idx;
5685 * If it isn't default-policy Meter, and
5686 * 1. There's no action in flow to change
5687 * packet (modify/encap/decap etc.), OR
5688 * 2. No drop count needed for this meter.
5689 * no need to use regC to save meter id anymore.
5691 if (!fm->def_policy && (!has_modify || !fm->drop_cnt))
5692 set_mtr_reg = false;
5693 /* Prefix actions: meter, decap, encap, tag, jump, end. */
5694 act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5695 sizeof(struct mlx5_rte_flow_action_set_tag);
5696 /* Suffix items: tag, vlan, port id, end. */
5697 #define METER_SUFFIX_ITEM 4
5698 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5699 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5700 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5703 return rte_flow_error_set(error, ENOMEM,
5704 RTE_FLOW_ERROR_TYPE_ACTION,
5705 NULL, "no memory to split "
5707 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5709 /* There's no suffix flow for meter of non-default policy. */
5710 if (!fm->def_policy)
5711 pre_actions = sfx_actions + 1;
5713 pre_actions = sfx_actions + actions_n;
5714 ret = flow_meter_split_prep(dev, flow, fm, &sfx_attr,
5715 items, sfx_items, actions,
5716 sfx_actions, pre_actions,
5717 (set_mtr_reg ? &mtr_flow_id : NULL),
5723 /* Add the prefix subflow. */
5724 flow_split_info->prefix_mark = 0;
5725 skip_scale_restore = flow_split_info->skip_scale;
5726 flow_split_info->skip_scale |=
5727 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5728 ret = flow_create_split_inner(dev, flow, &dev_flow,
5729 attr, items, pre_actions,
5730 flow_split_info, error);
5731 flow_split_info->skip_scale = skip_scale_restore;
5734 mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
5739 dev_flow->handle->split_flow_id = mtr_flow_id;
5740 dev_flow->handle->is_meter_flow_id = 1;
5742 if (!fm->def_policy) {
5743 if (!set_mtr_reg && fm->drop_cnt)
5745 flow_meter_create_drop_flow_with_org_pattern(dev, flow,
5751 /* Setting the sfx group atrr. */
5752 sfx_attr.group = sfx_attr.transfer ?
5753 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5754 MLX5_FLOW_TABLE_LEVEL_METER;
5755 flow_split_info->prefix_layers =
5756 flow_get_prefix_layer_flags(dev_flow);
5757 flow_split_info->prefix_mark = dev_flow->handle->mark;
5758 flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
5760 /* Add the prefix subflow. */
5761 ret = flow_create_split_metadata(dev, flow,
5762 &sfx_attr, sfx_items ?
5764 sfx_actions ? sfx_actions : actions,
5765 flow_split_info, error);
5768 mlx5_free(sfx_actions);
5773 * The splitting for sample feature.
5775 * Once Sample action is detected in the action list, the flow actions should
5776 * be split into prefix sub flow and suffix sub flow.
5778 * The original items remain in the prefix sub flow, all actions preceding the
5779 * sample action and the sample action itself will be copied to the prefix
5780 * sub flow, the actions following the sample action will be copied to the
5781 * suffix sub flow, Queue action always be located in the suffix sub flow.
5783 * In order to make the packet from prefix sub flow matches with suffix sub
5784 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5785 * flow uses tag item with the unique flow id.
5788 * Pointer to Ethernet device.
5790 * Parent flow structure pointer.
5792 * Flow rule attributes.
5794 * Pattern specification (list terminated by the END pattern item).
5795 * @param[in] actions
5796 * Associated actions (list terminated by the END action).
5797 * @param[in] flow_split_info
5798 * Pointer to flow split info structure.
5800 * Perform verbose error reporting if not NULL.
5802 * 0 on success, negative value otherwise
5805 flow_create_split_sample(struct rte_eth_dev *dev,
5806 struct rte_flow *flow,
5807 const struct rte_flow_attr *attr,
5808 const struct rte_flow_item items[],
5809 const struct rte_flow_action actions[],
5810 struct mlx5_flow_split_info *flow_split_info,
5811 struct rte_flow_error *error)
5813 struct mlx5_priv *priv = dev->data->dev_private;
5814 struct rte_flow_action *sfx_actions = NULL;
5815 struct rte_flow_action *pre_actions = NULL;
5816 struct rte_flow_item *sfx_items = NULL;
5817 struct mlx5_flow *dev_flow = NULL;
5818 struct rte_flow_attr sfx_attr = *attr;
5819 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5820 struct mlx5_flow_dv_sample_resource *sample_res;
5821 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5822 struct mlx5_flow_tbl_resource *sfx_tbl;
5826 uint32_t fdb_tx = 0;
5829 int sample_action_pos;
5830 int qrss_action_pos;
5832 int modify_after_mirror = 0;
5833 uint16_t jump_table = 0;
5834 const uint32_t next_ft_step = 1;
5837 if (priv->sampler_en)
5838 actions_n = flow_check_match_action(actions, attr,
5839 RTE_FLOW_ACTION_TYPE_SAMPLE,
5840 &sample_action_pos, &qrss_action_pos,
5841 &modify_after_mirror);
5843 /* The prefix actions must includes sample, tag, end. */
5844 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5845 + sizeof(struct mlx5_rte_flow_action_set_tag);
5846 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5847 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5848 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5849 item_size), 0, SOCKET_ID_ANY);
5851 return rte_flow_error_set(error, ENOMEM,
5852 RTE_FLOW_ERROR_TYPE_ACTION,
5853 NULL, "no memory to split "
5855 /* The representor_id is -1 for uplink. */
5856 fdb_tx = (attr->transfer && priv->representor_id != -1);
5858 * When reg_c_preserve is set, metadata registers Cx preserve
5859 * their value even through packet duplication.
5861 add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5863 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5865 if (modify_after_mirror)
5866 jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5868 pre_actions = sfx_actions + actions_n;
5869 tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5870 actions, sfx_actions,
5871 pre_actions, actions_n,
5873 qrss_action_pos, jump_table,
5875 if (tag_id < 0 || (add_tag && !tag_id)) {
5879 if (modify_after_mirror)
5880 flow_split_info->skip_scale =
5881 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5882 /* Add the prefix subflow. */
5883 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5885 flow_split_info, error);
5890 dev_flow->handle->split_flow_id = tag_id;
5891 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5892 if (!modify_after_mirror) {
5893 /* Set the sfx group attr. */
5894 sample_res = (struct mlx5_flow_dv_sample_resource *)
5895 dev_flow->dv.sample_res;
5896 sfx_tbl = (struct mlx5_flow_tbl_resource *)
5897 sample_res->normal_path_tbl;
5898 sfx_tbl_data = container_of(sfx_tbl,
5899 struct mlx5_flow_tbl_data_entry,
5901 sfx_attr.group = sfx_attr.transfer ?
5902 (sfx_tbl_data->level - 1) : sfx_tbl_data->level;
5904 MLX5_ASSERT(attr->transfer);
5905 sfx_attr.group = jump_table;
5907 flow_split_info->prefix_layers =
5908 flow_get_prefix_layer_flags(dev_flow);
5909 flow_split_info->prefix_mark = dev_flow->handle->mark;
5910 /* Suffix group level already be scaled with factor, set
5911 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5912 * again in translation.
5914 flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5917 /* Add the suffix subflow. */
5918 ret = flow_create_split_meter(dev, flow, &sfx_attr,
5919 sfx_items ? sfx_items : items,
5920 sfx_actions ? sfx_actions : actions,
5921 flow_split_info, error);
5924 mlx5_free(sfx_actions);
5929 * Split the flow to subflow set. The splitters might be linked
5930 * in the chain, like this:
5931 * flow_create_split_outer() calls:
5932 * flow_create_split_meter() calls:
5933 * flow_create_split_metadata(meter_subflow_0) calls:
5934 * flow_create_split_inner(metadata_subflow_0)
5935 * flow_create_split_inner(metadata_subflow_1)
5936 * flow_create_split_inner(metadata_subflow_2)
5937 * flow_create_split_metadata(meter_subflow_1) calls:
5938 * flow_create_split_inner(metadata_subflow_0)
5939 * flow_create_split_inner(metadata_subflow_1)
5940 * flow_create_split_inner(metadata_subflow_2)
5942 * This provide flexible way to add new levels of flow splitting.
5943 * The all of successfully created subflows are included to the
5944 * parent flow dev_flow list.
5947 * Pointer to Ethernet device.
5949 * Parent flow structure pointer.
5951 * Flow rule attributes.
5953 * Pattern specification (list terminated by the END pattern item).
5954 * @param[in] actions
5955 * Associated actions (list terminated by the END action).
5956 * @param[in] flow_split_info
5957 * Pointer to flow split info structure.
5959 * Perform verbose error reporting if not NULL.
5961 * 0 on success, negative value otherwise
5964 flow_create_split_outer(struct rte_eth_dev *dev,
5965 struct rte_flow *flow,
5966 const struct rte_flow_attr *attr,
5967 const struct rte_flow_item items[],
5968 const struct rte_flow_action actions[],
5969 struct mlx5_flow_split_info *flow_split_info,
5970 struct rte_flow_error *error)
5974 ret = flow_create_split_sample(dev, flow, attr, items,
5975 actions, flow_split_info, error);
5976 MLX5_ASSERT(ret <= 0);
5980 static inline struct mlx5_flow_tunnel *
5981 flow_tunnel_from_rule(const struct mlx5_flow *flow)
5983 struct mlx5_flow_tunnel *tunnel;
5985 #pragma GCC diagnostic push
5986 #pragma GCC diagnostic ignored "-Wcast-qual"
5987 tunnel = (typeof(tunnel))flow->tunnel;
5988 #pragma GCC diagnostic pop
5994 * Adjust flow RSS workspace if needed.
5997 * Pointer to thread flow work space.
5999 * Pointer to RSS descriptor.
6000 * @param[in] nrssq_num
6001 * New RSS queue number.
6004 * 0 on success, -1 otherwise and rte_errno is set.
6007 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6008 struct mlx5_flow_rss_desc *rss_desc,
6011 if (likely(nrssq_num <= wks->rssq_num))
6013 rss_desc->queue = realloc(rss_desc->queue,
6014 sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6015 if (!rss_desc->queue) {
6019 wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6024 * Create a flow and add it to @p list.
6027 * Pointer to Ethernet device.
6029 * Pointer to a TAILQ flow list. If this parameter NULL,
6030 * no list insertion occurred, flow is just created,
6031 * this is caller's responsibility to track the
6034 * Flow rule attributes.
6036 * Pattern specification (list terminated by the END pattern item).
6037 * @param[in] actions
6038 * Associated actions (list terminated by the END action).
6039 * @param[in] external
6040 * This flow rule is created by request external to PMD.
6042 * Perform verbose error reporting if not NULL.
6045 * A flow index on success, 0 otherwise and rte_errno is set.
6048 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
6049 const struct rte_flow_attr *attr,
6050 const struct rte_flow_item items[],
6051 const struct rte_flow_action original_actions[],
6052 bool external, struct rte_flow_error *error)
6054 struct mlx5_priv *priv = dev->data->dev_private;
6055 struct rte_flow *flow = NULL;
6056 struct mlx5_flow *dev_flow;
6057 const struct rte_flow_action_rss *rss = NULL;
6058 struct mlx5_translated_action_handle
6059 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6060 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6062 struct mlx5_flow_expand_rss buf;
6063 uint8_t buffer[2048];
6066 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6067 uint8_t buffer[2048];
6070 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6071 uint8_t buffer[2048];
6072 } actions_hairpin_tx;
6074 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6075 uint8_t buffer[2048];
6077 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6078 struct mlx5_flow_rss_desc *rss_desc;
6079 const struct rte_flow_action *p_actions_rx;
6083 struct rte_flow_attr attr_tx = { .priority = 0 };
6084 const struct rte_flow_action *actions;
6085 struct rte_flow_action *translated_actions = NULL;
6086 struct mlx5_flow_tunnel *tunnel;
6087 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6088 struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6089 struct mlx5_flow_split_info flow_split_info = {
6090 .external = !!external,
6100 rss_desc = &wks->rss_desc;
6101 ret = flow_action_handles_translate(dev, original_actions,
6104 &translated_actions, error);
6106 MLX5_ASSERT(translated_actions == NULL);
6109 actions = translated_actions ? translated_actions : original_actions;
6110 p_actions_rx = actions;
6111 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6112 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6113 external, hairpin_flow, error);
6115 goto error_before_hairpin_split;
6116 flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
6119 goto error_before_hairpin_split;
6121 if (hairpin_flow > 0) {
6122 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6124 goto error_before_hairpin_split;
6126 flow_hairpin_split(dev, actions, actions_rx.actions,
6127 actions_hairpin_tx.actions, items_tx.items,
6129 p_actions_rx = actions_rx.actions;
6131 flow_split_info.flow_idx = idx;
6132 flow->drv_type = flow_get_drv_type(dev, attr);
6133 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6134 flow->drv_type < MLX5_FLOW_TYPE_MAX);
6135 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6136 /* RSS Action only works on NIC RX domain */
6137 if (attr->ingress && !attr->transfer)
6138 rss = flow_get_rss_action(dev, p_actions_rx);
6140 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6143 * The following information is required by
6144 * mlx5_flow_hashfields_adjust() in advance.
6146 rss_desc->level = rss->level;
6147 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6148 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6150 flow->dev_handles = 0;
6151 if (rss && rss->types) {
6152 unsigned int graph_root;
6154 graph_root = find_graph_root(items, rss->level);
6155 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6157 mlx5_support_expansion, graph_root);
6158 MLX5_ASSERT(ret > 0 &&
6159 (unsigned int)ret < sizeof(expand_buffer.buffer));
6162 buf->entry[0].pattern = (void *)(uintptr_t)items;
6164 rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6166 for (i = 0; i < buf->entries; ++i) {
6167 /* Initialize flow split data. */
6168 flow_split_info.prefix_layers = 0;
6169 flow_split_info.prefix_mark = 0;
6170 flow_split_info.skip_scale = 0;
6172 * The splitter may create multiple dev_flows,
6173 * depending on configuration. In the simplest
6174 * case it just creates unmodified original flow.
6176 ret = flow_create_split_outer(dev, flow, attr,
6177 buf->entry[i].pattern,
6178 p_actions_rx, &flow_split_info,
6182 if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6183 ret = flow_tunnel_add_default_miss(dev, flow, attr,
6186 wks->flows[0].tunnel,
6190 mlx5_free(default_miss_ctx.queue);
6195 /* Create the tx flow. */
6197 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6198 attr_tx.ingress = 0;
6200 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6201 actions_hairpin_tx.actions,
6205 dev_flow->flow = flow;
6206 dev_flow->external = 0;
6207 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6208 dev_flow->handle, next);
6209 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6211 actions_hairpin_tx.actions, error);
6216 * Update the metadata register copy table. If extensive
6217 * metadata feature is enabled and registers are supported
6218 * we might create the extra rte_flow for each unique
6219 * MARK/FLAG action ID.
6221 * The table is updated for ingress Flows only, because
6222 * the egress Flows belong to the different device and
6223 * copy table should be updated in peer NIC Rx domain.
6225 if (attr->ingress &&
6226 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6227 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6232 * If the flow is external (from application) OR device is started,
6233 * OR mreg discover, then apply immediately.
6235 if (external || dev->data->dev_started ||
6236 (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6237 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6238 ret = flow_drv_apply(dev, flow, error);
6243 rte_spinlock_lock(&priv->flow_list_lock);
6244 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
6246 rte_spinlock_unlock(&priv->flow_list_lock);
6248 flow_rxq_flags_set(dev, flow);
6249 rte_free(translated_actions);
6250 tunnel = flow_tunnel_from_rule(wks->flows);
6253 flow->tunnel_id = tunnel->tunnel_id;
6254 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6255 mlx5_free(default_miss_ctx.queue);
6257 mlx5_flow_pop_thread_workspace();
6261 ret = rte_errno; /* Save rte_errno before cleanup. */
6262 flow_mreg_del_copy_action(dev, flow);
6263 flow_drv_destroy(dev, flow);
6264 if (rss_desc->shared_rss)
6265 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6267 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6268 rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6269 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
6270 rte_errno = ret; /* Restore rte_errno. */
6273 mlx5_flow_pop_thread_workspace();
6274 error_before_hairpin_split:
6275 rte_free(translated_actions);
6280 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6281 * incoming packets to table 1.
6283 * Other flow rules, requested for group n, will be created in
6284 * e-switch table n+1.
6285 * Jump action to e-switch group n will be created to group n+1.
6287 * Used when working in switchdev mode, to utilise advantages of table 1
6291 * Pointer to Ethernet device.
6294 * Pointer to flow on success, NULL otherwise and rte_errno is set.
6297 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6299 const struct rte_flow_attr attr = {
6306 const struct rte_flow_item pattern = {
6307 .type = RTE_FLOW_ITEM_TYPE_END,
6309 struct rte_flow_action_jump jump = {
6312 const struct rte_flow_action actions[] = {
6314 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6318 .type = RTE_FLOW_ACTION_TYPE_END,
6321 struct mlx5_priv *priv = dev->data->dev_private;
6322 struct rte_flow_error error;
6324 return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
6326 actions, false, &error);
6330 * Validate a flow supported by the NIC.
6332 * @see rte_flow_validate()
6336 mlx5_flow_validate(struct rte_eth_dev *dev,
6337 const struct rte_flow_attr *attr,
6338 const struct rte_flow_item items[],
6339 const struct rte_flow_action original_actions[],
6340 struct rte_flow_error *error)
6343 struct mlx5_translated_action_handle
6344 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6345 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6346 const struct rte_flow_action *actions;
6347 struct rte_flow_action *translated_actions = NULL;
6348 int ret = flow_action_handles_translate(dev, original_actions,
6351 &translated_actions, error);
6355 actions = translated_actions ? translated_actions : original_actions;
6356 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6357 ret = flow_drv_validate(dev, attr, items, actions,
6358 true, hairpin_flow, error);
6359 rte_free(translated_actions);
6366 * @see rte_flow_create()
6370 mlx5_flow_create(struct rte_eth_dev *dev,
6371 const struct rte_flow_attr *attr,
6372 const struct rte_flow_item items[],
6373 const struct rte_flow_action actions[],
6374 struct rte_flow_error *error)
6376 struct mlx5_priv *priv = dev->data->dev_private;
6379 * If the device is not started yet, it is not allowed to created a
6380 * flow from application. PMD default flows and traffic control flows
6383 if (unlikely(!dev->data->dev_started)) {
6384 DRV_LOG(DEBUG, "port %u is not started when "
6385 "inserting a flow", dev->data->port_id);
6386 rte_flow_error_set(error, ENODEV,
6387 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6389 "port not started");
6393 return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
6394 attr, items, actions, true, error);
6398 * Destroy a flow in a list.
6401 * Pointer to Ethernet device.
6403 * Pointer to the Indexed flow list. If this parameter NULL,
6404 * there is no flow removal from the list. Be noted that as
6405 * flow is add to the indexed list, memory of the indexed
6406 * list points to maybe changed as flow destroyed.
6407 * @param[in] flow_idx
6408 * Index of flow to destroy.
6411 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
6414 struct mlx5_priv *priv = dev->data->dev_private;
6415 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6416 [MLX5_IPOOL_RTE_FLOW], flow_idx);
6421 * Update RX queue flags only if port is started, otherwise it is
6424 if (dev->data->dev_started)
6425 flow_rxq_flags_trim(dev, flow);
6426 flow_drv_destroy(dev, flow);
6428 rte_spinlock_lock(&priv->flow_list_lock);
6429 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
6430 flow_idx, flow, next);
6431 rte_spinlock_unlock(&priv->flow_list_lock);
6434 struct mlx5_flow_tunnel *tunnel;
6436 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6438 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6439 mlx5_flow_tunnel_free(dev, tunnel);
6441 flow_mreg_del_copy_action(dev, flow);
6442 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6446 * Destroy all flows.
6449 * Pointer to Ethernet device.
6451 * Pointer to the Indexed flow list.
6453 * If flushing is called avtively.
6456 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
6458 uint32_t num_flushed = 0;
6461 flow_list_destroy(dev, list, *list);
6465 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6466 dev->data->port_id, num_flushed);
6471 * Stop all default actions for flows.
6474 * Pointer to Ethernet device.
6477 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6479 flow_mreg_del_default_copy_action(dev);
6480 flow_rxq_flags_clear(dev);
6484 * Start all default actions for flows.
6487 * Pointer to Ethernet device.
6489 * 0 on success, a negative errno value otherwise and rte_errno is set.
6492 mlx5_flow_start_default(struct rte_eth_dev *dev)
6494 struct rte_flow_error error;
6496 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6497 return flow_mreg_add_default_copy_action(dev, &error);
6501 * Release key of thread specific flow workspace data.
6504 flow_release_workspace(void *data)
6506 struct mlx5_flow_workspace *wks = data;
6507 struct mlx5_flow_workspace *next;
6511 free(wks->rss_desc.queue);
6518 * Get thread specific current flow workspace.
6520 * @return pointer to thread specific flow workspace data, NULL on error.
6522 struct mlx5_flow_workspace*
6523 mlx5_flow_get_thread_workspace(void)
6525 struct mlx5_flow_workspace *data;
6527 data = mlx5_flow_os_get_specific_workspace();
6528 MLX5_ASSERT(data && data->inuse);
6529 if (!data || !data->inuse)
6530 DRV_LOG(ERR, "flow workspace not initialized.");
6535 * Allocate and init new flow workspace.
6537 * @return pointer to flow workspace data, NULL on error.
6539 static struct mlx5_flow_workspace*
6540 flow_alloc_thread_workspace(void)
6542 struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6545 DRV_LOG(ERR, "Failed to allocate flow workspace "
6549 data->rss_desc.queue = calloc(1,
6550 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6551 if (!data->rss_desc.queue)
6553 data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6556 if (data->rss_desc.queue)
6557 free(data->rss_desc.queue);
6563 * Get new thread specific flow workspace.
6565 * If current workspace inuse, create new one and set as current.
6567 * @return pointer to thread specific flow workspace data, NULL on error.
6569 static struct mlx5_flow_workspace*
6570 mlx5_flow_push_thread_workspace(void)
6572 struct mlx5_flow_workspace *curr;
6573 struct mlx5_flow_workspace *data;
6575 curr = mlx5_flow_os_get_specific_workspace();
6577 data = flow_alloc_thread_workspace();
6580 } else if (!curr->inuse) {
6582 } else if (curr->next) {
6585 data = flow_alloc_thread_workspace();
6593 /* Set as current workspace */
6594 if (mlx5_flow_os_set_specific_workspace(data))
6595 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6600 * Close current thread specific flow workspace.
6602 * If previous workspace available, set it as current.
6604 * @return pointer to thread specific flow workspace data, NULL on error.
6607 mlx5_flow_pop_thread_workspace(void)
6609 struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6614 DRV_LOG(ERR, "Failed to close unused flow workspace.");
6620 if (mlx5_flow_os_set_specific_workspace(data->prev))
6621 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6625 * Verify the flow list is empty
6628 * Pointer to Ethernet device.
6630 * @return the number of flows not released.
6633 mlx5_flow_verify(struct rte_eth_dev *dev)
6635 struct mlx5_priv *priv = dev->data->dev_private;
6636 struct rte_flow *flow;
6640 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6642 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6643 dev->data->port_id, (void *)flow);
6650 * Enable default hairpin egress flow.
6653 * Pointer to Ethernet device.
6658 * 0 on success, a negative errno value otherwise and rte_errno is set.
6661 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6664 struct mlx5_priv *priv = dev->data->dev_private;
6665 const struct rte_flow_attr attr = {
6669 struct mlx5_rte_flow_item_tx_queue queue_spec = {
6672 struct mlx5_rte_flow_item_tx_queue queue_mask = {
6673 .queue = UINT32_MAX,
6675 struct rte_flow_item items[] = {
6677 .type = (enum rte_flow_item_type)
6678 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6679 .spec = &queue_spec,
6681 .mask = &queue_mask,
6684 .type = RTE_FLOW_ITEM_TYPE_END,
6687 struct rte_flow_action_jump jump = {
6688 .group = MLX5_HAIRPIN_TX_TABLE,
6690 struct rte_flow_action actions[2];
6692 struct rte_flow_error error;
6694 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6695 actions[0].conf = &jump;
6696 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6697 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6698 &attr, items, actions, false, &error);
6701 "Failed to create ctrl flow: rte_errno(%d),"
6702 " type(%d), message(%s)",
6703 rte_errno, error.type,
6704 error.message ? error.message : " (no stated reason)");
6711 * Enable a control flow configured from the control plane.
6714 * Pointer to Ethernet device.
6716 * An Ethernet flow spec to apply.
6718 * An Ethernet flow mask to apply.
6720 * A VLAN flow spec to apply.
6722 * A VLAN flow mask to apply.
6725 * 0 on success, a negative errno value otherwise and rte_errno is set.
6728 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6729 struct rte_flow_item_eth *eth_spec,
6730 struct rte_flow_item_eth *eth_mask,
6731 struct rte_flow_item_vlan *vlan_spec,
6732 struct rte_flow_item_vlan *vlan_mask)
6734 struct mlx5_priv *priv = dev->data->dev_private;
6735 const struct rte_flow_attr attr = {
6737 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6739 struct rte_flow_item items[] = {
6741 .type = RTE_FLOW_ITEM_TYPE_ETH,
6747 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6748 RTE_FLOW_ITEM_TYPE_END,
6754 .type = RTE_FLOW_ITEM_TYPE_END,
6757 uint16_t queue[priv->reta_idx_n];
6758 struct rte_flow_action_rss action_rss = {
6759 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6761 .types = priv->rss_conf.rss_hf,
6762 .key_len = priv->rss_conf.rss_key_len,
6763 .queue_num = priv->reta_idx_n,
6764 .key = priv->rss_conf.rss_key,
6767 struct rte_flow_action actions[] = {
6769 .type = RTE_FLOW_ACTION_TYPE_RSS,
6770 .conf = &action_rss,
6773 .type = RTE_FLOW_ACTION_TYPE_END,
6777 struct rte_flow_error error;
6780 if (!priv->reta_idx_n || !priv->rxqs_n) {
6783 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6784 action_rss.types = 0;
6785 for (i = 0; i != priv->reta_idx_n; ++i)
6786 queue[i] = (*priv->reta_idx)[i];
6787 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6788 &attr, items, actions, false, &error);
6795 * Enable a flow control configured from the control plane.
6798 * Pointer to Ethernet device.
6800 * An Ethernet flow spec to apply.
6802 * An Ethernet flow mask to apply.
6805 * 0 on success, a negative errno value otherwise and rte_errno is set.
6808 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6809 struct rte_flow_item_eth *eth_spec,
6810 struct rte_flow_item_eth *eth_mask)
6812 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6816 * Create default miss flow rule matching lacp traffic
6819 * Pointer to Ethernet device.
6821 * An Ethernet flow spec to apply.
6824 * 0 on success, a negative errno value otherwise and rte_errno is set.
6827 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6829 struct mlx5_priv *priv = dev->data->dev_private;
6831 * The LACP matching is done by only using ether type since using
6832 * a multicast dst mac causes kernel to give low priority to this flow.
6834 static const struct rte_flow_item_eth lacp_spec = {
6835 .type = RTE_BE16(0x8809),
6837 static const struct rte_flow_item_eth lacp_mask = {
6840 const struct rte_flow_attr attr = {
6843 struct rte_flow_item items[] = {
6845 .type = RTE_FLOW_ITEM_TYPE_ETH,
6850 .type = RTE_FLOW_ITEM_TYPE_END,
6853 struct rte_flow_action actions[] = {
6855 .type = (enum rte_flow_action_type)
6856 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6859 .type = RTE_FLOW_ACTION_TYPE_END,
6862 struct rte_flow_error error;
6863 uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6864 &attr, items, actions, false, &error);
6874 * @see rte_flow_destroy()
6878 mlx5_flow_destroy(struct rte_eth_dev *dev,
6879 struct rte_flow *flow,
6880 struct rte_flow_error *error __rte_unused)
6882 struct mlx5_priv *priv = dev->data->dev_private;
6884 flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6889 * Destroy all flows.
6891 * @see rte_flow_flush()
6895 mlx5_flow_flush(struct rte_eth_dev *dev,
6896 struct rte_flow_error *error __rte_unused)
6898 struct mlx5_priv *priv = dev->data->dev_private;
6900 mlx5_flow_list_flush(dev, &priv->flows, false);
6907 * @see rte_flow_isolate()
6911 mlx5_flow_isolate(struct rte_eth_dev *dev,
6913 struct rte_flow_error *error)
6915 struct mlx5_priv *priv = dev->data->dev_private;
6917 if (dev->data->dev_started) {
6918 rte_flow_error_set(error, EBUSY,
6919 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6921 "port must be stopped first");
6924 priv->isolated = !!enable;
6926 dev->dev_ops = &mlx5_dev_ops_isolate;
6928 dev->dev_ops = &mlx5_dev_ops;
6930 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6931 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6939 * @see rte_flow_query()
6943 flow_drv_query(struct rte_eth_dev *dev,
6945 const struct rte_flow_action *actions,
6947 struct rte_flow_error *error)
6949 struct mlx5_priv *priv = dev->data->dev_private;
6950 const struct mlx5_flow_driver_ops *fops;
6951 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6952 [MLX5_IPOOL_RTE_FLOW],
6954 enum mlx5_flow_drv_type ftype;
6957 return rte_flow_error_set(error, ENOENT,
6958 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6960 "invalid flow handle");
6962 ftype = flow->drv_type;
6963 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6964 fops = flow_get_drv_ops(ftype);
6966 return fops->query(dev, flow, actions, data, error);
6972 * @see rte_flow_query()
6976 mlx5_flow_query(struct rte_eth_dev *dev,
6977 struct rte_flow *flow,
6978 const struct rte_flow_action *actions,
6980 struct rte_flow_error *error)
6984 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6992 * Get rte_flow callbacks.
6995 * Pointer to Ethernet device structure.
6997 * Pointer to operation-specific structure.
7002 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7003 const struct rte_flow_ops **ops)
7005 *ops = &mlx5_flow_ops;
7010 * Validate meter policy actions.
7011 * Dispatcher for action type specific validation.
7014 * Pointer to the Ethernet device structure.
7016 * The meter policy action object to validate.
7018 * Attributes of flow to determine steering domain.
7019 * @param[out] is_rss
7021 * @param[out] domain_bitmap
7023 * @param[out] is_def_policy
7024 * Is default policy or not.
7026 * Perform verbose error reporting if not NULL. Initialized in case of
7030 * 0 on success, otherwise negative errno value.
7033 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7034 const struct rte_flow_action *actions[RTE_COLORS],
7035 struct rte_flow_attr *attr,
7037 uint8_t *domain_bitmap,
7038 bool *is_def_policy,
7039 struct rte_mtr_error *error)
7041 const struct mlx5_flow_driver_ops *fops;
7043 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7044 return fops->validate_mtr_acts(dev, actions, attr,
7045 is_rss, domain_bitmap, is_def_policy, error);
7049 * Destroy the meter table set.
7052 * Pointer to Ethernet device.
7053 * @param[in] mtr_policy
7054 * Meter policy struct.
7057 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7058 struct mlx5_flow_meter_policy *mtr_policy)
7060 const struct mlx5_flow_driver_ops *fops;
7062 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7063 fops->destroy_mtr_acts(dev, mtr_policy);
7067 * Create policy action, lock free,
7068 * (mutex should be acquired by caller).
7069 * Dispatcher for action type specific call.
7072 * Pointer to the Ethernet device structure.
7073 * @param[in] mtr_policy
7074 * Meter policy struct.
7076 * Action specification used to create meter actions.
7078 * Perform verbose error reporting if not NULL. Initialized in case of
7082 * 0 on success, otherwise negative errno value.
7085 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7086 struct mlx5_flow_meter_policy *mtr_policy,
7087 const struct rte_flow_action *actions[RTE_COLORS],
7088 struct rte_mtr_error *error)
7090 const struct mlx5_flow_driver_ops *fops;
7092 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7093 return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7097 * Create policy rules, lock free,
7098 * (mutex should be acquired by caller).
7099 * Dispatcher for action type specific call.
7102 * Pointer to the Ethernet device structure.
7103 * @param[in] mtr_policy
7104 * Meter policy struct.
7107 * 0 on success, -1 otherwise.
7110 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7111 struct mlx5_flow_meter_policy *mtr_policy)
7113 const struct mlx5_flow_driver_ops *fops;
7115 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7116 return fops->create_policy_rules(dev, mtr_policy);
7120 * Destroy policy rules, lock free,
7121 * (mutex should be acquired by caller).
7122 * Dispatcher for action type specific call.
7125 * Pointer to the Ethernet device structure.
7126 * @param[in] mtr_policy
7127 * Meter policy struct.
7130 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7131 struct mlx5_flow_meter_policy *mtr_policy)
7133 const struct mlx5_flow_driver_ops *fops;
7135 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7136 fops->destroy_policy_rules(dev, mtr_policy);
7140 * Destroy the default policy table set.
7143 * Pointer to Ethernet device.
7146 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7148 const struct mlx5_flow_driver_ops *fops;
7150 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7151 fops->destroy_def_policy(dev);
7155 * Destroy the default policy table set.
7158 * Pointer to Ethernet device.
7161 * 0 on success, -1 otherwise.
7164 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7166 const struct mlx5_flow_driver_ops *fops;
7168 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7169 return fops->create_def_policy(dev);
7173 * Create the needed meter and suffix tables.
7176 * Pointer to Ethernet device.
7179 * 0 on success, -1 otherwise.
7182 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7183 struct mlx5_flow_meter_info *fm,
7185 uint8_t domain_bitmap)
7187 const struct mlx5_flow_driver_ops *fops;
7189 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7190 return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7194 * Destroy the meter table set.
7197 * Pointer to Ethernet device.
7199 * Pointer to the meter table set.
7202 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7203 struct mlx5_flow_meter_info *fm)
7205 const struct mlx5_flow_driver_ops *fops;
7207 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7208 fops->destroy_mtr_tbls(dev, fm);
7212 * Destroy the global meter drop table.
7215 * Pointer to Ethernet device.
7218 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7220 const struct mlx5_flow_driver_ops *fops;
7222 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7223 fops->destroy_mtr_drop_tbls(dev);
7227 * Allocate the needed aso flow meter id.
7230 * Pointer to Ethernet device.
7233 * Index to aso flow meter on success, NULL otherwise.
7236 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7238 const struct mlx5_flow_driver_ops *fops;
7240 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7241 return fops->create_meter(dev);
7245 * Free the aso flow meter id.
7248 * Pointer to Ethernet device.
7249 * @param[in] mtr_idx
7250 * Index to aso flow meter to be free.
7256 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7258 const struct mlx5_flow_driver_ops *fops;
7260 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7261 fops->free_meter(dev, mtr_idx);
7265 * Allocate a counter.
7268 * Pointer to Ethernet device structure.
7271 * Index to allocated counter on success, 0 otherwise.
7274 mlx5_counter_alloc(struct rte_eth_dev *dev)
7276 const struct mlx5_flow_driver_ops *fops;
7277 struct rte_flow_attr attr = { .transfer = 0 };
7279 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7280 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7281 return fops->counter_alloc(dev);
7284 "port %u counter allocate is not supported.",
7285 dev->data->port_id);
7293 * Pointer to Ethernet device structure.
7295 * Index to counter to be free.
7298 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7300 const struct mlx5_flow_driver_ops *fops;
7301 struct rte_flow_attr attr = { .transfer = 0 };
7303 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7304 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7305 fops->counter_free(dev, cnt);
7309 "port %u counter free is not supported.",
7310 dev->data->port_id);
7314 * Query counter statistics.
7317 * Pointer to Ethernet device structure.
7319 * Index to counter to query.
7321 * Set to clear counter statistics.
7323 * The counter hits packets number to save.
7325 * The counter hits bytes number to save.
7328 * 0 on success, a negative errno value otherwise.
7331 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7332 bool clear, uint64_t *pkts, uint64_t *bytes)
7334 const struct mlx5_flow_driver_ops *fops;
7335 struct rte_flow_attr attr = { .transfer = 0 };
7337 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7338 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7339 return fops->counter_query(dev, cnt, clear, pkts, bytes);
7342 "port %u counter query is not supported.",
7343 dev->data->port_id);
7348 * Allocate a new memory for the counter values wrapped by all the needed
7352 * Pointer to mlx5_dev_ctx_shared object.
7355 * 0 on success, a negative errno value otherwise.
7358 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7360 struct mlx5_devx_mkey_attr mkey_attr;
7361 struct mlx5_counter_stats_mem_mng *mem_mng;
7362 volatile struct flow_counter_stats *raw_data;
7363 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7364 int size = (sizeof(struct flow_counter_stats) *
7365 MLX5_COUNTERS_PER_POOL +
7366 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7367 sizeof(struct mlx5_counter_stats_mem_mng);
7368 size_t pgsize = rte_mem_page_size();
7372 if (pgsize == (size_t)-1) {
7373 DRV_LOG(ERR, "Failed to get mem page size");
7377 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7382 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7383 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7384 mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7385 IBV_ACCESS_LOCAL_WRITE);
7386 if (!mem_mng->umem) {
7391 memset(&mkey_attr, 0, sizeof(mkey_attr));
7392 mkey_attr.addr = (uintptr_t)mem;
7393 mkey_attr.size = size;
7394 mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7395 mkey_attr.pd = sh->pdn;
7396 mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7397 mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7398 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7400 mlx5_os_umem_dereg(mem_mng->umem);
7405 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7406 raw_data = (volatile struct flow_counter_stats *)mem;
7407 for (i = 0; i < raws_n; ++i) {
7408 mem_mng->raws[i].mem_mng = mem_mng;
7409 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7411 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7412 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7413 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7415 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7416 sh->cmng.mem_mng = mem_mng;
7421 * Set the statistic memory to the new counter pool.
7424 * Pointer to mlx5_dev_ctx_shared object.
7426 * Pointer to the pool to set the statistic memory.
7429 * 0 on success, a negative errno value otherwise.
7432 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7433 struct mlx5_flow_counter_pool *pool)
7435 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7436 /* Resize statistic memory once used out. */
7437 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7438 mlx5_flow_create_counter_stat_mem_mng(sh)) {
7439 DRV_LOG(ERR, "Cannot resize counter stat mem.");
7442 rte_spinlock_lock(&pool->sl);
7443 pool->raw = cmng->mem_mng->raws + pool->index %
7444 MLX5_CNT_CONTAINER_RESIZE;
7445 rte_spinlock_unlock(&pool->sl);
7446 pool->raw_hw = NULL;
7450 #define MLX5_POOL_QUERY_FREQ_US 1000000
7453 * Set the periodic procedure for triggering asynchronous batch queries for all
7454 * the counter pools.
7457 * Pointer to mlx5_dev_ctx_shared object.
7460 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7462 uint32_t pools_n, us;
7464 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7465 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7466 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7467 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7468 sh->cmng.query_thread_on = 0;
7469 DRV_LOG(ERR, "Cannot reinitialize query alarm");
7471 sh->cmng.query_thread_on = 1;
7476 * The periodic procedure for triggering asynchronous batch queries for all the
7477 * counter pools. This function is probably called by the host thread.
7480 * The parameter for the alarm process.
7483 mlx5_flow_query_alarm(void *arg)
7485 struct mlx5_dev_ctx_shared *sh = arg;
7487 uint16_t pool_index = sh->cmng.pool_index;
7488 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7489 struct mlx5_flow_counter_pool *pool;
7492 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7494 rte_spinlock_lock(&cmng->pool_update_sl);
7495 pool = cmng->pools[pool_index];
7496 n_valid = cmng->n_valid;
7497 rte_spinlock_unlock(&cmng->pool_update_sl);
7498 /* Set the statistic memory to the new created pool. */
7499 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7502 /* There is a pool query in progress. */
7505 LIST_FIRST(&sh->cmng.free_stat_raws);
7507 /* No free counter statistics raw memory. */
7510 * Identify the counters released between query trigger and query
7511 * handle more efficiently. The counter released in this gap period
7512 * should wait for a new round of query as the new arrived packets
7513 * will not be taken into account.
7516 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7517 MLX5_COUNTERS_PER_POOL,
7519 pool->raw_hw->mem_mng->dm->id,
7523 (uint64_t)(uintptr_t)pool);
7525 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7526 " %d", pool->min_dcs->id);
7527 pool->raw_hw = NULL;
7530 LIST_REMOVE(pool->raw_hw, next);
7531 sh->cmng.pending_queries++;
7533 if (pool_index >= n_valid)
7536 sh->cmng.pool_index = pool_index;
7537 mlx5_set_query_alarm(sh);
7541 * Check and callback event for new aged flow in the counter pool
7544 * Pointer to mlx5_dev_ctx_shared object.
7546 * Pointer to Current counter pool.
7549 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7550 struct mlx5_flow_counter_pool *pool)
7552 struct mlx5_priv *priv;
7553 struct mlx5_flow_counter *cnt;
7554 struct mlx5_age_info *age_info;
7555 struct mlx5_age_param *age_param;
7556 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7557 struct mlx5_counter_stats_raw *prev = pool->raw;
7558 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7559 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7560 uint16_t expected = AGE_CANDIDATE;
7563 pool->time_of_last_age_check = curr_time;
7564 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7565 cnt = MLX5_POOL_GET_CNT(pool, i);
7566 age_param = MLX5_CNT_TO_AGE(cnt);
7567 if (__atomic_load_n(&age_param->state,
7568 __ATOMIC_RELAXED) != AGE_CANDIDATE)
7570 if (cur->data[i].hits != prev->data[i].hits) {
7571 __atomic_store_n(&age_param->sec_since_last_hit, 0,
7575 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7577 __ATOMIC_RELAXED) <= age_param->timeout)
7580 * Hold the lock first, or if between the
7581 * state AGE_TMOUT and tailq operation the
7582 * release happened, the release procedure
7583 * may delete a non-existent tailq node.
7585 priv = rte_eth_devices[age_param->port_id].data->dev_private;
7586 age_info = GET_PORT_AGE_INFO(priv);
7587 rte_spinlock_lock(&age_info->aged_sl);
7588 if (__atomic_compare_exchange_n(&age_param->state, &expected,
7591 __ATOMIC_RELAXED)) {
7592 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7593 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7595 rte_spinlock_unlock(&age_info->aged_sl);
7597 mlx5_age_event_prepare(sh);
7601 * Handler for the HW respond about ready values from an asynchronous batch
7602 * query. This function is probably called by the host thread.
7605 * The pointer to the shared device context.
7606 * @param[in] async_id
7607 * The Devx async ID.
7609 * The status of the completion.
7612 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7613 uint64_t async_id, int status)
7615 struct mlx5_flow_counter_pool *pool =
7616 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7617 struct mlx5_counter_stats_raw *raw_to_free;
7618 uint8_t query_gen = pool->query_gen ^ 1;
7619 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7620 enum mlx5_counter_type cnt_type =
7621 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7622 MLX5_COUNTER_TYPE_ORIGIN;
7624 if (unlikely(status)) {
7625 raw_to_free = pool->raw_hw;
7627 raw_to_free = pool->raw;
7629 mlx5_flow_aging_check(sh, pool);
7630 rte_spinlock_lock(&pool->sl);
7631 pool->raw = pool->raw_hw;
7632 rte_spinlock_unlock(&pool->sl);
7633 /* Be sure the new raw counters data is updated in memory. */
7635 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7636 rte_spinlock_lock(&cmng->csl[cnt_type]);
7637 TAILQ_CONCAT(&cmng->counters[cnt_type],
7638 &pool->counters[query_gen], next);
7639 rte_spinlock_unlock(&cmng->csl[cnt_type]);
7642 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7643 pool->raw_hw = NULL;
7644 sh->cmng.pending_queries--;
7648 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7649 const struct flow_grp_info *grp_info,
7650 struct rte_flow_error *error)
7652 if (grp_info->transfer && grp_info->external &&
7653 grp_info->fdb_def_rule) {
7654 if (group == UINT32_MAX)
7655 return rte_flow_error_set
7657 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7659 "group index not supported");
7664 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7669 * Translate the rte_flow group index to HW table value.
7671 * If tunnel offload is disabled, all group ids converted to flow table
7672 * id using the standard method.
7673 * If tunnel offload is enabled, group id can be converted using the
7674 * standard or tunnel conversion method. Group conversion method
7675 * selection depends on flags in `grp_info` parameter:
7676 * - Internal (grp_info.external == 0) groups conversion uses the
7678 * - Group ids in JUMP action converted with the tunnel conversion.
7679 * - Group id in rule attribute conversion depends on a rule type and
7681 * ** non zero group attributes converted with the tunnel method
7682 * ** zero group attribute in non-tunnel rule is converted using the
7683 * standard method - there's only one root table
7684 * ** zero group attribute in steer tunnel rule is converted with the
7685 * standard method - single root table
7686 * ** zero group attribute in match tunnel rule is a special OvS
7687 * case: that value is used for portability reasons. That group
7688 * id is converted with the tunnel conversion method.
7693 * PMD tunnel offload object
7695 * rte_flow group index value.
7698 * @param[in] grp_info
7699 * flags used for conversion
7701 * Pointer to error structure.
7704 * 0 on success, a negative errno value otherwise and rte_errno is set.
7707 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7708 const struct mlx5_flow_tunnel *tunnel,
7709 uint32_t group, uint32_t *table,
7710 const struct flow_grp_info *grp_info,
7711 struct rte_flow_error *error)
7714 bool standard_translation;
7716 if (!grp_info->skip_scale && grp_info->external &&
7717 group < MLX5_MAX_TABLES_EXTERNAL)
7718 group *= MLX5_FLOW_TABLE_FACTOR;
7719 if (is_tunnel_offload_active(dev)) {
7720 standard_translation = !grp_info->external ||
7721 grp_info->std_tbl_fix;
7723 standard_translation = true;
7726 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7727 dev->data->port_id, group, grp_info->transfer,
7728 grp_info->external, grp_info->fdb_def_rule,
7729 standard_translation ? "STANDARD" : "TUNNEL");
7730 if (standard_translation)
7731 ret = flow_group_to_table(dev->data->port_id, group, table,
7734 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7741 * Discover availability of metadata reg_c's.
7743 * Iteratively use test flows to check availability.
7746 * Pointer to the Ethernet device structure.
7749 * 0 on success, a negative errno value otherwise and rte_errno is set.
7752 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7754 struct mlx5_priv *priv = dev->data->dev_private;
7755 struct mlx5_dev_config *config = &priv->config;
7756 enum modify_reg idx;
7759 /* reg_c[0] and reg_c[1] are reserved. */
7760 config->flow_mreg_c[n++] = REG_C_0;
7761 config->flow_mreg_c[n++] = REG_C_1;
7762 /* Discover availability of other reg_c's. */
7763 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7764 struct rte_flow_attr attr = {
7765 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7766 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7769 struct rte_flow_item items[] = {
7771 .type = RTE_FLOW_ITEM_TYPE_END,
7774 struct rte_flow_action actions[] = {
7776 .type = (enum rte_flow_action_type)
7777 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7778 .conf = &(struct mlx5_flow_action_copy_mreg){
7784 .type = RTE_FLOW_ACTION_TYPE_JUMP,
7785 .conf = &(struct rte_flow_action_jump){
7786 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7790 .type = RTE_FLOW_ACTION_TYPE_END,
7794 struct rte_flow *flow;
7795 struct rte_flow_error error;
7797 if (!config->dv_flow_en)
7799 /* Create internal flow, validation skips copy action. */
7800 flow_idx = flow_list_create(dev, NULL, &attr, items,
7801 actions, false, &error);
7802 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7806 config->flow_mreg_c[n++] = idx;
7807 flow_list_destroy(dev, NULL, flow_idx);
7809 for (; n < MLX5_MREG_C_NUM; ++n)
7810 config->flow_mreg_c[n] = REG_NON;
7815 * Dump flow raw hw data to file
7818 * The pointer to Ethernet device.
7820 * A pointer to a file for output.
7822 * Perform verbose error reporting if not NULL. PMDs initialize this
7823 * structure in case of error only.
7825 * 0 on success, a nagative value otherwise.
7828 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
7830 struct rte_flow_error *error __rte_unused)
7832 struct mlx5_priv *priv = dev->data->dev_private;
7833 struct mlx5_dev_ctx_shared *sh = priv->sh;
7834 uint32_t handle_idx;
7836 struct mlx5_flow_handle *dh;
7837 struct rte_flow *flow;
7839 if (!priv->config.dv_flow_en) {
7840 if (fputs("device dv flow disabled\n", file) <= 0)
7847 return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
7849 sh->tx_domain, file);
7851 flow = mlx5_ipool_get(priv->sh->ipool
7852 [MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
7856 handle_idx = flow->dev_handles;
7857 while (handle_idx) {
7858 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
7863 ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
7868 handle_idx = dh->next.next;
7874 * Get aged-out flows.
7877 * Pointer to the Ethernet device structure.
7878 * @param[in] context
7879 * The address of an array of pointers to the aged-out flows contexts.
7880 * @param[in] nb_countexts
7881 * The length of context array pointers.
7883 * Perform verbose error reporting if not NULL. Initialized in case of
7887 * how many contexts get in success, otherwise negative errno value.
7888 * if nb_contexts is 0, return the amount of all aged contexts.
7889 * if nb_contexts is not 0 , return the amount of aged flows reported
7890 * in the context array.
7893 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7894 uint32_t nb_contexts, struct rte_flow_error *error)
7896 const struct mlx5_flow_driver_ops *fops;
7897 struct rte_flow_attr attr = { .transfer = 0 };
7899 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7900 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7901 return fops->get_aged_flows(dev, contexts, nb_contexts,
7905 "port %u get aged flows is not supported.",
7906 dev->data->port_id);
7910 /* Wrapper for driver action_validate op callback */
7912 flow_drv_action_validate(struct rte_eth_dev *dev,
7913 const struct rte_flow_indir_action_conf *conf,
7914 const struct rte_flow_action *action,
7915 const struct mlx5_flow_driver_ops *fops,
7916 struct rte_flow_error *error)
7918 static const char err_msg[] = "indirect action validation unsupported";
7920 if (!fops->action_validate) {
7921 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7922 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7926 return fops->action_validate(dev, conf, action, error);
7930 * Destroys the shared action by handle.
7933 * Pointer to Ethernet device structure.
7935 * Handle for the indirect action object to be destroyed.
7937 * Perform verbose error reporting if not NULL. PMDs initialize this
7938 * structure in case of error only.
7941 * 0 on success, a negative errno value otherwise and rte_errno is set.
7943 * @note: wrapper for driver action_create op callback.
7946 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
7947 struct rte_flow_action_handle *handle,
7948 struct rte_flow_error *error)
7950 static const char err_msg[] = "indirect action destruction unsupported";
7951 struct rte_flow_attr attr = { .transfer = 0 };
7952 const struct mlx5_flow_driver_ops *fops =
7953 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7955 if (!fops->action_destroy) {
7956 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7957 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7961 return fops->action_destroy(dev, handle, error);
7964 /* Wrapper for driver action_destroy op callback */
7966 flow_drv_action_update(struct rte_eth_dev *dev,
7967 struct rte_flow_action_handle *handle,
7969 const struct mlx5_flow_driver_ops *fops,
7970 struct rte_flow_error *error)
7972 static const char err_msg[] = "indirect action update unsupported";
7974 if (!fops->action_update) {
7975 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7976 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7980 return fops->action_update(dev, handle, update, error);
7983 /* Wrapper for driver action_destroy op callback */
7985 flow_drv_action_query(struct rte_eth_dev *dev,
7986 const struct rte_flow_action_handle *handle,
7988 const struct mlx5_flow_driver_ops *fops,
7989 struct rte_flow_error *error)
7991 static const char err_msg[] = "indirect action query unsupported";
7993 if (!fops->action_query) {
7994 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7995 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7999 return fops->action_query(dev, handle, data, error);
8003 * Create indirect action for reuse in multiple flow rules.
8006 * Pointer to Ethernet device structure.
8008 * Pointer to indirect action object configuration.
8010 * Action configuration for indirect action object creation.
8012 * Perform verbose error reporting if not NULL. PMDs initialize this
8013 * structure in case of error only.
8015 * A valid handle in case of success, NULL otherwise and rte_errno is set.
8017 static struct rte_flow_action_handle *
8018 mlx5_action_handle_create(struct rte_eth_dev *dev,
8019 const struct rte_flow_indir_action_conf *conf,
8020 const struct rte_flow_action *action,
8021 struct rte_flow_error *error)
8023 static const char err_msg[] = "indirect action creation unsupported";
8024 struct rte_flow_attr attr = { .transfer = 0 };
8025 const struct mlx5_flow_driver_ops *fops =
8026 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8028 if (flow_drv_action_validate(dev, conf, action, fops, error))
8030 if (!fops->action_create) {
8031 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8032 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8036 return fops->action_create(dev, conf, action, error);
8040 * Updates inplace the indirect action configuration pointed by *handle*
8041 * with the configuration provided as *update* argument.
8042 * The update of the indirect action configuration effects all flow rules
8043 * reusing the action via handle.
8046 * Pointer to Ethernet device structure.
8048 * Handle for the indirect action to be updated.
8050 * Action specification used to modify the action pointed by handle.
8051 * *update* could be of same type with the action pointed by the *handle*
8052 * handle argument, or some other structures like a wrapper, depending on
8053 * the indirect action type.
8055 * Perform verbose error reporting if not NULL. PMDs initialize this
8056 * structure in case of error only.
8059 * 0 on success, a negative errno value otherwise and rte_errno is set.
8062 mlx5_action_handle_update(struct rte_eth_dev *dev,
8063 struct rte_flow_action_handle *handle,
8065 struct rte_flow_error *error)
8067 struct rte_flow_attr attr = { .transfer = 0 };
8068 const struct mlx5_flow_driver_ops *fops =
8069 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8072 ret = flow_drv_action_validate(dev, NULL,
8073 (const struct rte_flow_action *)update, fops, error);
8076 return flow_drv_action_update(dev, handle, update, fops,
8081 * Query the indirect action by handle.
8083 * This function allows retrieving action-specific data such as counters.
8084 * Data is gathered by special action which may be present/referenced in
8085 * more than one flow rule definition.
8087 * see @RTE_FLOW_ACTION_TYPE_COUNT
8090 * Pointer to Ethernet device structure.
8092 * Handle for the indirect action to query.
8093 * @param[in, out] data
8094 * Pointer to storage for the associated query data type.
8096 * Perform verbose error reporting if not NULL. PMDs initialize this
8097 * structure in case of error only.
8100 * 0 on success, a negative errno value otherwise and rte_errno is set.
8103 mlx5_action_handle_query(struct rte_eth_dev *dev,
8104 const struct rte_flow_action_handle *handle,
8106 struct rte_flow_error *error)
8108 struct rte_flow_attr attr = { .transfer = 0 };
8109 const struct mlx5_flow_driver_ops *fops =
8110 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8112 return flow_drv_action_query(dev, handle, data, fops, error);
8116 * Destroy all indirect actions (shared RSS).
8119 * Pointer to Ethernet device.
8122 * 0 on success, a negative errno value otherwise and rte_errno is set.
8125 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8127 struct rte_flow_error error;
8128 struct mlx5_priv *priv = dev->data->dev_private;
8129 struct mlx5_shared_action_rss *shared_rss;
8133 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8134 priv->rss_shared_actions, idx, shared_rss, next) {
8135 ret |= mlx5_action_handle_destroy(dev,
8136 (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8141 #ifndef HAVE_MLX5DV_DR
8142 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8144 #define MLX5_DOMAIN_SYNC_FLOW \
8145 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8148 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8150 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8151 const struct mlx5_flow_driver_ops *fops;
8153 struct rte_flow_attr attr = { .transfer = 0 };
8155 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8156 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8162 const struct mlx5_flow_tunnel *
8163 mlx5_get_tof(const struct rte_flow_item *item,
8164 const struct rte_flow_action *action,
8165 enum mlx5_tof_rule_type *rule_type)
8167 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8168 if (item->type == (typeof(item->type))
8169 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8170 *rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8171 return flow_items_to_tunnel(item);
8174 for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8175 if (action->type == (typeof(action->type))
8176 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8177 *rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8178 return flow_actions_to_tunnel(action);
8185 * tunnel offload functionalilty is defined for DV environment only
8187 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8189 union tunnel_offload_mark {
8192 uint32_t app_reserve:8;
8193 uint32_t table_id:15;
8194 uint32_t transfer:1;
8195 uint32_t _unused_:8;
8200 mlx5_access_tunnel_offload_db
8201 (struct rte_eth_dev *dev,
8202 bool (*match)(struct rte_eth_dev *,
8203 struct mlx5_flow_tunnel *, const void *),
8204 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8205 void (*miss)(struct rte_eth_dev *, void *),
8206 void *ctx, bool lock_op);
8209 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8210 struct rte_flow *flow,
8211 const struct rte_flow_attr *attr,
8212 const struct rte_flow_action *app_actions,
8214 const struct mlx5_flow_tunnel *tunnel,
8215 struct tunnel_default_miss_ctx *ctx,
8216 struct rte_flow_error *error)
8218 struct mlx5_priv *priv = dev->data->dev_private;
8219 struct mlx5_flow *dev_flow;
8220 struct rte_flow_attr miss_attr = *attr;
8221 const struct rte_flow_item miss_items[2] = {
8223 .type = RTE_FLOW_ITEM_TYPE_ETH,
8229 .type = RTE_FLOW_ITEM_TYPE_END,
8235 union tunnel_offload_mark mark_id;
8236 struct rte_flow_action_mark miss_mark;
8237 struct rte_flow_action miss_actions[3] = {
8238 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8239 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
8241 const struct rte_flow_action_jump *jump_data;
8242 uint32_t i, flow_table = 0; /* prevent compilation warning */
8243 struct flow_grp_info grp_info = {
8245 .transfer = attr->transfer,
8246 .fdb_def_rule = !!priv->fdb_def_rule,
8251 if (!attr->transfer) {
8254 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8255 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8256 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8259 return rte_flow_error_set
8261 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8262 NULL, "invalid default miss RSS");
8263 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8264 ctx->action_rss.level = 0,
8265 ctx->action_rss.types = priv->rss_conf.rss_hf,
8266 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8267 ctx->action_rss.queue_num = priv->reta_idx_n,
8268 ctx->action_rss.key = priv->rss_conf.rss_key,
8269 ctx->action_rss.queue = ctx->queue;
8270 if (!priv->reta_idx_n || !priv->rxqs_n)
8271 return rte_flow_error_set
8273 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8274 NULL, "invalid port configuration");
8275 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8276 ctx->action_rss.types = 0;
8277 for (i = 0; i != priv->reta_idx_n; ++i)
8278 ctx->queue[i] = (*priv->reta_idx)[i];
8280 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8281 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8283 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8284 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8285 jump_data = app_actions->conf;
8286 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8287 miss_attr.group = jump_data->group;
8288 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8289 &flow_table, &grp_info, error);
8291 return rte_flow_error_set(error, EINVAL,
8292 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8293 NULL, "invalid tunnel id");
8294 mark_id.app_reserve = 0;
8295 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8296 mark_id.transfer = !!attr->transfer;
8297 mark_id._unused_ = 0;
8298 miss_mark.id = mark_id.val;
8299 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8300 miss_items, miss_actions, flow_idx, error);
8303 dev_flow->flow = flow;
8304 dev_flow->external = true;
8305 dev_flow->tunnel = tunnel;
8306 dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8307 /* Subflow object was created, we must include one in the list. */
8308 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8309 dev_flow->handle, next);
8311 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8312 dev->data->port_id, tunnel->app_tunnel.type,
8313 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8314 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8315 miss_actions, error);
8317 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8323 static const struct mlx5_flow_tbl_data_entry *
8324 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8326 struct mlx5_priv *priv = dev->data->dev_private;
8327 struct mlx5_dev_ctx_shared *sh = priv->sh;
8328 struct mlx5_hlist_entry *he;
8329 union tunnel_offload_mark mbits = { .val = mark };
8330 union mlx5_flow_tbl_key table_key = {
8332 .level = tunnel_id_to_flow_tbl(mbits.table_id),
8336 .is_fdb = !!mbits.transfer,
8340 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
8342 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8346 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
8347 struct mlx5_hlist_entry *entry)
8349 struct mlx5_dev_ctx_shared *sh = list->ctx;
8350 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8352 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8353 tunnel_flow_tbl_to_id(tte->flow_table));
8358 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
8359 struct mlx5_hlist_entry *entry,
8360 uint64_t key, void *cb_ctx __rte_unused)
8362 union tunnel_tbl_key tbl = {
8365 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8367 return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8370 static struct mlx5_hlist_entry *
8371 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
8372 void *ctx __rte_unused)
8374 struct mlx5_dev_ctx_shared *sh = list->ctx;
8375 struct tunnel_tbl_entry *tte;
8376 union tunnel_tbl_key tbl = {
8380 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8385 mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8387 if (tte->flow_table >= MLX5_MAX_TABLES) {
8388 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8390 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8393 } else if (!tte->flow_table) {
8396 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8397 tte->tunnel_id = tbl.tunnel_id;
8398 tte->group = tbl.group;
8407 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8408 const struct mlx5_flow_tunnel *tunnel,
8409 uint32_t group, uint32_t *table,
8410 struct rte_flow_error *error)
8412 struct mlx5_hlist_entry *he;
8413 struct tunnel_tbl_entry *tte;
8414 union tunnel_tbl_key key = {
8415 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8418 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8419 struct mlx5_hlist *group_hash;
8421 group_hash = tunnel ? tunnel->groups : thub->groups;
8422 he = mlx5_hlist_register(group_hash, key.val, NULL);
8424 return rte_flow_error_set(error, EINVAL,
8425 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8427 "tunnel group index not supported");
8428 tte = container_of(he, typeof(*tte), hash);
8429 *table = tte->flow_table;
8430 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8431 dev->data->port_id, key.tunnel_id, group, *table);
8436 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8437 struct mlx5_flow_tunnel *tunnel)
8439 struct mlx5_priv *priv = dev->data->dev_private;
8440 struct mlx5_indexed_pool *ipool;
8442 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8443 dev->data->port_id, tunnel->tunnel_id);
8444 LIST_REMOVE(tunnel, chain);
8445 mlx5_hlist_destroy(tunnel->groups);
8446 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8447 mlx5_ipool_free(ipool, tunnel->tunnel_id);
8451 mlx5_access_tunnel_offload_db
8452 (struct rte_eth_dev *dev,
8453 bool (*match)(struct rte_eth_dev *,
8454 struct mlx5_flow_tunnel *, const void *),
8455 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8456 void (*miss)(struct rte_eth_dev *, void *),
8457 void *ctx, bool lock_op)
8459 bool verdict = false;
8460 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8461 struct mlx5_flow_tunnel *tunnel;
8463 rte_spinlock_lock(&thub->sl);
8464 LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8465 verdict = match(dev, tunnel, (const void *)ctx);
8470 rte_spinlock_unlock(&thub->sl);
8472 hit(dev, tunnel, ctx);
8473 if (!verdict && miss)
8476 rte_spinlock_unlock(&thub->sl);
8481 struct tunnel_db_find_tunnel_id_ctx {
8483 struct mlx5_flow_tunnel *tunnel;
8487 find_tunnel_id_match(struct rte_eth_dev *dev,
8488 struct mlx5_flow_tunnel *tunnel, const void *x)
8490 const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8493 return tunnel->tunnel_id == ctx->tunnel_id;
8497 find_tunnel_id_hit(struct rte_eth_dev *dev,
8498 struct mlx5_flow_tunnel *tunnel, void *x)
8500 struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8502 ctx->tunnel = tunnel;
8505 static struct mlx5_flow_tunnel *
8506 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8508 struct tunnel_db_find_tunnel_id_ctx ctx = {
8512 mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8513 find_tunnel_id_hit, NULL, &ctx, true);
8518 static struct mlx5_flow_tunnel *
8519 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8520 const struct rte_flow_tunnel *app_tunnel)
8522 struct mlx5_priv *priv = dev->data->dev_private;
8523 struct mlx5_indexed_pool *ipool;
8524 struct mlx5_flow_tunnel *tunnel;
8527 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8528 tunnel = mlx5_ipool_zmalloc(ipool, &id);
8531 if (id >= MLX5_MAX_TUNNELS) {
8532 mlx5_ipool_free(ipool, id);
8533 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8536 tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
8537 mlx5_flow_tunnel_grp2tbl_create_cb,
8538 mlx5_flow_tunnel_grp2tbl_match_cb,
8539 mlx5_flow_tunnel_grp2tbl_remove_cb);
8540 if (!tunnel->groups) {
8541 mlx5_ipool_free(ipool, id);
8544 tunnel->groups->ctx = priv->sh;
8545 /* initiate new PMD tunnel */
8546 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8547 tunnel->tunnel_id = id;
8548 tunnel->action.type = (typeof(tunnel->action.type))
8549 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8550 tunnel->action.conf = tunnel;
8551 tunnel->item.type = (typeof(tunnel->item.type))
8552 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8553 tunnel->item.spec = tunnel;
8554 tunnel->item.last = NULL;
8555 tunnel->item.mask = NULL;
8557 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8558 dev->data->port_id, tunnel->tunnel_id);
8563 struct tunnel_db_get_tunnel_ctx {
8564 const struct rte_flow_tunnel *app_tunnel;
8565 struct mlx5_flow_tunnel *tunnel;
8568 static bool get_tunnel_match(struct rte_eth_dev *dev,
8569 struct mlx5_flow_tunnel *tunnel, const void *x)
8571 const struct tunnel_db_get_tunnel_ctx *ctx = x;
8574 return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
8575 sizeof(*ctx->app_tunnel));
8578 static void get_tunnel_hit(struct rte_eth_dev *dev,
8579 struct mlx5_flow_tunnel *tunnel, void *x)
8581 /* called under tunnel spinlock protection */
8582 struct tunnel_db_get_tunnel_ctx *ctx = x;
8586 ctx->tunnel = tunnel;
8589 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8591 /* called under tunnel spinlock protection */
8592 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8593 struct tunnel_db_get_tunnel_ctx *ctx = x;
8595 rte_spinlock_unlock(&thub->sl);
8596 ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8597 rte_spinlock_lock(&thub->sl);
8599 ctx->tunnel->refctn = 1;
8600 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8606 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8607 const struct rte_flow_tunnel *app_tunnel,
8608 struct mlx5_flow_tunnel **tunnel)
8610 struct tunnel_db_get_tunnel_ctx ctx = {
8611 .app_tunnel = app_tunnel,
8614 mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8615 get_tunnel_miss, &ctx, true);
8616 *tunnel = ctx.tunnel;
8617 return ctx.tunnel ? 0 : -ENOMEM;
8620 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8622 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8626 if (!LIST_EMPTY(&thub->tunnels))
8627 DRV_LOG(WARNING, "port %u tunnels present", port_id);
8628 mlx5_hlist_destroy(thub->groups);
8632 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8635 struct mlx5_flow_tunnel_hub *thub;
8637 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8641 LIST_INIT(&thub->tunnels);
8642 rte_spinlock_init(&thub->sl);
8643 thub->groups = mlx5_hlist_create("flow groups",
8644 rte_align32pow2(MLX5_MAX_TABLES), 0,
8645 0, mlx5_flow_tunnel_grp2tbl_create_cb,
8646 mlx5_flow_tunnel_grp2tbl_match_cb,
8647 mlx5_flow_tunnel_grp2tbl_remove_cb);
8648 if (!thub->groups) {
8652 thub->groups->ctx = sh;
8653 sh->tunnel_hub = thub;
8659 mlx5_hlist_destroy(thub->groups);
8666 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
8667 struct rte_flow_tunnel *tunnel,
8668 const char *err_msg)
8671 if (!is_tunnel_offload_active(dev)) {
8672 err_msg = "tunnel offload was not activated";
8674 } else if (!tunnel) {
8675 err_msg = "no application tunnel";
8679 switch (tunnel->type) {
8681 err_msg = "unsupported tunnel type";
8683 case RTE_FLOW_ITEM_TYPE_VXLAN:
8692 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
8693 struct rte_flow_tunnel *app_tunnel,
8694 struct rte_flow_action **actions,
8695 uint32_t *num_of_actions,
8696 struct rte_flow_error *error)
8699 struct mlx5_flow_tunnel *tunnel;
8700 const char *err_msg = NULL;
8701 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8704 return rte_flow_error_set(error, EINVAL,
8705 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8707 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8709 return rte_flow_error_set(error, ret,
8710 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8711 "failed to initialize pmd tunnel");
8713 *actions = &tunnel->action;
8714 *num_of_actions = 1;
8719 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
8720 struct rte_flow_tunnel *app_tunnel,
8721 struct rte_flow_item **items,
8722 uint32_t *num_of_items,
8723 struct rte_flow_error *error)
8726 struct mlx5_flow_tunnel *tunnel;
8727 const char *err_msg = NULL;
8728 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8731 return rte_flow_error_set(error, EINVAL,
8732 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8734 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8736 return rte_flow_error_set(error, ret,
8737 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8738 "failed to initialize pmd tunnel");
8740 *items = &tunnel->item;
8745 struct tunnel_db_element_release_ctx {
8746 struct rte_flow_item *items;
8747 struct rte_flow_action *actions;
8748 uint32_t num_elements;
8749 struct rte_flow_error *error;
8754 tunnel_element_release_match(struct rte_eth_dev *dev,
8755 struct mlx5_flow_tunnel *tunnel, const void *x)
8757 const struct tunnel_db_element_release_ctx *ctx = x;
8760 if (ctx->num_elements != 1)
8762 else if (ctx->items)
8763 return ctx->items == &tunnel->item;
8764 else if (ctx->actions)
8765 return ctx->actions == &tunnel->action;
8771 tunnel_element_release_hit(struct rte_eth_dev *dev,
8772 struct mlx5_flow_tunnel *tunnel, void *x)
8774 struct tunnel_db_element_release_ctx *ctx = x;
8776 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8777 mlx5_flow_tunnel_free(dev, tunnel);
8781 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8783 struct tunnel_db_element_release_ctx *ctx = x;
8785 ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8786 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8787 "invalid argument");
8791 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8792 struct rte_flow_item *pmd_items,
8793 uint32_t num_items, struct rte_flow_error *err)
8795 struct tunnel_db_element_release_ctx ctx = {
8798 .num_elements = num_items,
8802 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8803 tunnel_element_release_hit,
8804 tunnel_element_release_miss, &ctx, false);
8810 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8811 struct rte_flow_action *pmd_actions,
8812 uint32_t num_actions, struct rte_flow_error *err)
8814 struct tunnel_db_element_release_ctx ctx = {
8816 .actions = pmd_actions,
8817 .num_elements = num_actions,
8821 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8822 tunnel_element_release_hit,
8823 tunnel_element_release_miss, &ctx, false);
8829 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8831 struct rte_flow_restore_info *info,
8832 struct rte_flow_error *err)
8834 uint64_t ol_flags = m->ol_flags;
8835 const struct mlx5_flow_tbl_data_entry *tble;
8836 const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8838 if (!is_tunnel_offload_active(dev)) {
8843 if ((ol_flags & mask) != mask)
8845 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8847 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8848 dev->data->port_id, m->hash.fdir.hi);
8851 MLX5_ASSERT(tble->tunnel);
8852 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8853 info->group_id = tble->group_id;
8854 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8855 RTE_FLOW_RESTORE_INFO_GROUP_ID |
8856 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8861 return rte_flow_error_set(err, EINVAL,
8862 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8863 "failed to get restore info");
8866 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8868 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8869 __rte_unused struct rte_flow_tunnel *app_tunnel,
8870 __rte_unused struct rte_flow_action **actions,
8871 __rte_unused uint32_t *num_of_actions,
8872 __rte_unused struct rte_flow_error *error)
8878 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8879 __rte_unused struct rte_flow_tunnel *app_tunnel,
8880 __rte_unused struct rte_flow_item **items,
8881 __rte_unused uint32_t *num_of_items,
8882 __rte_unused struct rte_flow_error *error)
8888 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8889 __rte_unused struct rte_flow_item *pmd_items,
8890 __rte_unused uint32_t num_items,
8891 __rte_unused struct rte_flow_error *err)
8897 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8898 __rte_unused struct rte_flow_action *pmd_action,
8899 __rte_unused uint32_t num_actions,
8900 __rte_unused struct rte_flow_error *err)
8906 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8907 __rte_unused struct rte_mbuf *m,
8908 __rte_unused struct rte_flow_restore_info *i,
8909 __rte_unused struct rte_flow_error *err)
8915 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8916 __rte_unused struct rte_flow *flow,
8917 __rte_unused const struct rte_flow_attr *attr,
8918 __rte_unused const struct rte_flow_action *actions,
8919 __rte_unused uint32_t flow_idx,
8920 __rte_unused const struct mlx5_flow_tunnel *tunnel,
8921 __rte_unused struct tunnel_default_miss_ctx *ctx,
8922 __rte_unused struct rte_flow_error *error)
8927 static struct mlx5_flow_tunnel *
8928 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8929 __rte_unused uint32_t id)
8935 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8936 __rte_unused struct mlx5_flow_tunnel *tunnel)
8941 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8942 __rte_unused const struct mlx5_flow_tunnel *t,
8943 __rte_unused uint32_t group,
8944 __rte_unused uint32_t *table,
8945 struct rte_flow_error *error)
8947 return rte_flow_error_set(error, ENOTSUP,
8948 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8949 "tunnel offload requires DV support");
8953 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8954 __rte_unused uint16_t port_id)
8957 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */