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 RTE_ETH_RSS_* definitions).
105 * Bit-fields that define how the node is used in the expansion.
106 * (see MLX5_EXPANSION_NODE_* definitions).
110 /* Optional expand field. The expansion alg will not go deeper. */
111 #define MLX5_EXPANSION_NODE_OPTIONAL (UINT64_C(1) << 0)
113 /* The node is not added implicitly as expansion to the flow pattern.
114 * If the node type does not match the flow pattern item type, the
115 * expansion alg will go deeper to its next items.
116 * In the current implementation, the list of next nodes indexes can
117 * have up to one node with this flag set and it has to be the last
118 * node index (before the list terminator).
120 #define MLX5_EXPANSION_NODE_EXPLICIT (UINT64_C(1) << 1)
122 /** Object returned by mlx5_flow_expand_rss(). */
123 struct mlx5_flow_expand_rss {
125 /**< Number of entries @p patterns and @p priorities. */
127 struct rte_flow_item *pattern; /**< Expanded pattern array. */
128 uint32_t priority; /**< Priority offset for each expansion. */
133 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
135 static const struct mlx5_flow_expand_node *
136 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
137 unsigned int item_idx,
138 const struct mlx5_flow_expand_node graph[],
139 const struct mlx5_flow_expand_node *node);
142 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
144 switch (item->type) {
145 case RTE_FLOW_ITEM_TYPE_ETH:
146 case RTE_FLOW_ITEM_TYPE_VLAN:
147 case RTE_FLOW_ITEM_TYPE_IPV4:
148 case RTE_FLOW_ITEM_TYPE_IPV6:
149 case RTE_FLOW_ITEM_TYPE_UDP:
150 case RTE_FLOW_ITEM_TYPE_TCP:
151 case RTE_FLOW_ITEM_TYPE_VXLAN:
152 case RTE_FLOW_ITEM_TYPE_NVGRE:
153 case RTE_FLOW_ITEM_TYPE_GRE:
154 case RTE_FLOW_ITEM_TYPE_GENEVE:
155 case RTE_FLOW_ITEM_TYPE_MPLS:
156 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
157 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
158 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
159 case RTE_FLOW_ITEM_TYPE_GTP:
167 static enum rte_flow_item_type
168 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
170 enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
171 uint16_t ether_type = 0;
172 uint16_t ether_type_m;
173 uint8_t ip_next_proto = 0;
174 uint8_t ip_next_proto_m;
176 if (item == NULL || item->spec == NULL)
178 switch (item->type) {
179 case RTE_FLOW_ITEM_TYPE_ETH:
181 ether_type_m = ((const struct rte_flow_item_eth *)
184 ether_type_m = rte_flow_item_eth_mask.type;
185 if (ether_type_m != RTE_BE16(0xFFFF))
187 ether_type = ((const struct rte_flow_item_eth *)
189 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
190 ret = RTE_FLOW_ITEM_TYPE_IPV4;
191 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
192 ret = RTE_FLOW_ITEM_TYPE_IPV6;
193 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
194 ret = RTE_FLOW_ITEM_TYPE_VLAN;
196 ret = RTE_FLOW_ITEM_TYPE_END;
198 case RTE_FLOW_ITEM_TYPE_VLAN:
200 ether_type_m = ((const struct rte_flow_item_vlan *)
201 (item->mask))->inner_type;
203 ether_type_m = rte_flow_item_vlan_mask.inner_type;
204 if (ether_type_m != RTE_BE16(0xFFFF))
206 ether_type = ((const struct rte_flow_item_vlan *)
207 (item->spec))->inner_type;
208 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
209 ret = RTE_FLOW_ITEM_TYPE_IPV4;
210 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
211 ret = RTE_FLOW_ITEM_TYPE_IPV6;
212 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
213 ret = RTE_FLOW_ITEM_TYPE_VLAN;
215 ret = RTE_FLOW_ITEM_TYPE_END;
217 case RTE_FLOW_ITEM_TYPE_IPV4:
219 ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
220 (item->mask))->hdr.next_proto_id;
223 rte_flow_item_ipv4_mask.hdr.next_proto_id;
224 if (ip_next_proto_m != 0xFF)
226 ip_next_proto = ((const struct rte_flow_item_ipv4 *)
227 (item->spec))->hdr.next_proto_id;
228 if (ip_next_proto == IPPROTO_UDP)
229 ret = RTE_FLOW_ITEM_TYPE_UDP;
230 else if (ip_next_proto == IPPROTO_TCP)
231 ret = RTE_FLOW_ITEM_TYPE_TCP;
232 else if (ip_next_proto == IPPROTO_IP)
233 ret = RTE_FLOW_ITEM_TYPE_IPV4;
234 else if (ip_next_proto == IPPROTO_IPV6)
235 ret = RTE_FLOW_ITEM_TYPE_IPV6;
237 ret = RTE_FLOW_ITEM_TYPE_END;
239 case RTE_FLOW_ITEM_TYPE_IPV6:
241 ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
242 (item->mask))->hdr.proto;
245 rte_flow_item_ipv6_mask.hdr.proto;
246 if (ip_next_proto_m != 0xFF)
248 ip_next_proto = ((const struct rte_flow_item_ipv6 *)
249 (item->spec))->hdr.proto;
250 if (ip_next_proto == IPPROTO_UDP)
251 ret = RTE_FLOW_ITEM_TYPE_UDP;
252 else if (ip_next_proto == IPPROTO_TCP)
253 ret = RTE_FLOW_ITEM_TYPE_TCP;
254 else if (ip_next_proto == IPPROTO_IP)
255 ret = RTE_FLOW_ITEM_TYPE_IPV4;
256 else if (ip_next_proto == IPPROTO_IPV6)
257 ret = RTE_FLOW_ITEM_TYPE_IPV6;
259 ret = RTE_FLOW_ITEM_TYPE_END;
262 ret = RTE_FLOW_ITEM_TYPE_VOID;
269 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
270 const int *next_node)
272 const struct mlx5_flow_expand_node *node = NULL;
273 const int *next = next_node;
275 while (next && *next) {
277 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
278 * flag set, because they were not found in the flow pattern.
280 node = &graph[*next];
281 if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
288 #define MLX5_RSS_EXP_ELT_N 16
291 * Expand RSS flows into several possible flows according to the RSS hash
292 * fields requested and the driver capabilities.
295 * Buffer to store the result expansion.
297 * Buffer size in bytes. If 0, @p buf can be NULL.
301 * RSS types to expand (see RTE_ETH_RSS_* definitions).
303 * Input graph to expand @p pattern according to @p types.
304 * @param[in] graph_root_index
305 * Index of root node in @p graph, typically 0.
308 * A positive value representing the size of @p buf in bytes regardless of
309 * @p size on success, a negative errno value otherwise and rte_errno is
310 * set, the following errors are defined:
312 * -E2BIG: graph-depth @p graph is too deep.
313 * -EINVAL: @p size has not enough space for expanded pattern.
316 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
317 const struct rte_flow_item *pattern, uint64_t types,
318 const struct mlx5_flow_expand_node graph[],
319 int graph_root_index)
321 const struct rte_flow_item *item;
322 const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
323 const int *next_node;
324 const int *stack[MLX5_RSS_EXP_ELT_N];
326 struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
327 unsigned int i, item_idx, last_expand_item_idx = 0;
329 size_t user_pattern_size = 0;
331 const struct mlx5_flow_expand_node *next = NULL;
332 struct rte_flow_item missed_item;
335 const struct rte_flow_item *last_expand_item = NULL;
337 memset(&missed_item, 0, sizeof(missed_item));
338 lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
339 MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
342 buf->entry[0].priority = 0;
343 buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
345 addr = buf->entry[0].pattern;
346 for (item = pattern, item_idx = 0;
347 item->type != RTE_FLOW_ITEM_TYPE_END;
348 item++, item_idx++) {
349 if (!mlx5_flow_is_rss_expandable_item(item)) {
350 user_pattern_size += sizeof(*item);
353 last_expand_item = item;
354 last_expand_item_idx = item_idx;
356 while (node->next && node->next[i]) {
357 next = &graph[node->next[i]];
358 if (next->type == item->type)
360 if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
369 user_pattern_size += sizeof(*item);
371 user_pattern_size += sizeof(*item); /* Handle END item. */
372 lsize += user_pattern_size;
375 /* Copy the user pattern in the first entry of the buffer. */
376 rte_memcpy(addr, pattern, user_pattern_size);
377 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
379 /* Start expanding. */
380 memset(flow_items, 0, sizeof(flow_items));
381 user_pattern_size -= sizeof(*item);
383 * Check if the last valid item has spec set, need complete pattern,
384 * and the pattern can be used for expansion.
386 missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
387 if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
388 /* Item type END indicates expansion is not required. */
391 if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
395 while (node->next && node->next[i]) {
396 next = &graph[node->next[i]];
397 if (next->type == missed_item.type) {
398 flow_items[0].type = missed_item.type;
399 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
402 if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
411 if (next && missed) {
412 elt = 2; /* missed item + item end. */
414 lsize += elt * sizeof(*item) + user_pattern_size;
417 if (node->rss_types & types) {
418 buf->entry[buf->entries].priority = 1;
419 buf->entry[buf->entries].pattern = addr;
421 rte_memcpy(addr, buf->entry[0].pattern,
423 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
424 rte_memcpy(addr, flow_items, elt * sizeof(*item));
425 addr = (void *)(((uintptr_t)addr) +
426 elt * sizeof(*item));
428 } else if (last_expand_item != NULL) {
429 node = mlx5_flow_expand_rss_adjust_node(pattern,
430 last_expand_item_idx, graph, node);
432 memset(flow_items, 0, sizeof(flow_items));
433 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
435 stack[stack_pos] = next_node;
436 node = next_node ? &graph[*next_node] : NULL;
438 flow_items[stack_pos].type = node->type;
439 if (node->rss_types & types) {
442 * compute the number of items to copy from the
443 * expansion and copy it.
444 * When the stack_pos is 0, there are 1 element in it,
445 * plus the addition END item.
448 flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
449 lsize += elt * sizeof(*item) + user_pattern_size;
452 n = elt * sizeof(*item);
453 buf->entry[buf->entries].priority =
454 stack_pos + 1 + missed;
455 buf->entry[buf->entries].pattern = addr;
457 rte_memcpy(addr, buf->entry[0].pattern,
459 addr = (void *)(((uintptr_t)addr) +
461 rte_memcpy(addr, &missed_item,
462 missed * sizeof(*item));
463 addr = (void *)(((uintptr_t)addr) +
464 missed * sizeof(*item));
465 rte_memcpy(addr, flow_items, n);
466 addr = (void *)(((uintptr_t)addr) + n);
469 if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
471 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
473 if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
477 stack[stack_pos] = next_node;
478 } else if (*(next_node + 1)) {
479 /* Follow up with the next possibility. */
480 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
482 } else if (!stack_pos) {
484 * Completing the traverse over the different paths.
485 * The next_node is advanced to the terminator.
489 /* Move to the next path. */
491 next_node = stack[--stack_pos];
496 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
498 stack[stack_pos] = next_node;
500 node = next_node && *next_node ? &graph[*next_node] : NULL;
505 enum mlx5_expansion {
507 MLX5_EXPANSION_ROOT_OUTER,
508 MLX5_EXPANSION_OUTER_ETH,
509 MLX5_EXPANSION_OUTER_VLAN,
510 MLX5_EXPANSION_OUTER_IPV4,
511 MLX5_EXPANSION_OUTER_IPV4_UDP,
512 MLX5_EXPANSION_OUTER_IPV4_TCP,
513 MLX5_EXPANSION_OUTER_IPV6,
514 MLX5_EXPANSION_OUTER_IPV6_UDP,
515 MLX5_EXPANSION_OUTER_IPV6_TCP,
516 MLX5_EXPANSION_VXLAN,
517 MLX5_EXPANSION_STD_VXLAN,
518 MLX5_EXPANSION_L3_VXLAN,
519 MLX5_EXPANSION_VXLAN_GPE,
521 MLX5_EXPANSION_NVGRE,
522 MLX5_EXPANSION_GRE_KEY,
527 MLX5_EXPANSION_IPV4_UDP,
528 MLX5_EXPANSION_IPV4_TCP,
530 MLX5_EXPANSION_IPV6_UDP,
531 MLX5_EXPANSION_IPV6_TCP,
532 MLX5_EXPANSION_IPV6_FRAG_EXT,
536 /** Supported expansion of items. */
537 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
538 [MLX5_EXPANSION_ROOT] = {
539 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
541 MLX5_EXPANSION_IPV6),
542 .type = RTE_FLOW_ITEM_TYPE_END,
544 [MLX5_EXPANSION_ROOT_OUTER] = {
545 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
546 MLX5_EXPANSION_OUTER_IPV4,
547 MLX5_EXPANSION_OUTER_IPV6),
548 .type = RTE_FLOW_ITEM_TYPE_END,
550 [MLX5_EXPANSION_OUTER_ETH] = {
551 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
552 .type = RTE_FLOW_ITEM_TYPE_ETH,
555 [MLX5_EXPANSION_OUTER_VLAN] = {
556 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
557 MLX5_EXPANSION_OUTER_IPV6),
558 .type = RTE_FLOW_ITEM_TYPE_VLAN,
559 .node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
561 [MLX5_EXPANSION_OUTER_IPV4] = {
562 .next = MLX5_FLOW_EXPAND_RSS_NEXT
563 (MLX5_EXPANSION_OUTER_IPV4_UDP,
564 MLX5_EXPANSION_OUTER_IPV4_TCP,
566 MLX5_EXPANSION_NVGRE,
568 MLX5_EXPANSION_IPV6),
569 .type = RTE_FLOW_ITEM_TYPE_IPV4,
570 .rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
571 RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
573 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
574 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
575 MLX5_EXPANSION_VXLAN_GPE,
578 .type = RTE_FLOW_ITEM_TYPE_UDP,
579 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
581 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
582 .type = RTE_FLOW_ITEM_TYPE_TCP,
583 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
585 [MLX5_EXPANSION_OUTER_IPV6] = {
586 .next = MLX5_FLOW_EXPAND_RSS_NEXT
587 (MLX5_EXPANSION_OUTER_IPV6_UDP,
588 MLX5_EXPANSION_OUTER_IPV6_TCP,
592 MLX5_EXPANSION_NVGRE),
593 .type = RTE_FLOW_ITEM_TYPE_IPV6,
594 .rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
595 RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
597 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
598 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
599 MLX5_EXPANSION_VXLAN_GPE,
602 .type = RTE_FLOW_ITEM_TYPE_UDP,
603 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
605 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
606 .type = RTE_FLOW_ITEM_TYPE_TCP,
607 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
609 [MLX5_EXPANSION_VXLAN] = {
610 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
612 MLX5_EXPANSION_IPV6),
613 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
615 [MLX5_EXPANSION_STD_VXLAN] = {
616 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
617 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
619 [MLX5_EXPANSION_L3_VXLAN] = {
620 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
621 MLX5_EXPANSION_IPV6),
622 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
624 [MLX5_EXPANSION_VXLAN_GPE] = {
625 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
627 MLX5_EXPANSION_IPV6),
628 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
630 [MLX5_EXPANSION_GRE] = {
631 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
633 MLX5_EXPANSION_GRE_KEY,
634 MLX5_EXPANSION_MPLS),
635 .type = RTE_FLOW_ITEM_TYPE_GRE,
637 [MLX5_EXPANSION_GRE_KEY] = {
638 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
640 MLX5_EXPANSION_MPLS),
641 .type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
642 .node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
644 [MLX5_EXPANSION_NVGRE] = {
645 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
646 .type = RTE_FLOW_ITEM_TYPE_NVGRE,
648 [MLX5_EXPANSION_MPLS] = {
649 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
652 .type = RTE_FLOW_ITEM_TYPE_MPLS,
653 .node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
655 [MLX5_EXPANSION_ETH] = {
656 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
657 .type = RTE_FLOW_ITEM_TYPE_ETH,
659 [MLX5_EXPANSION_VLAN] = {
660 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
661 MLX5_EXPANSION_IPV6),
662 .type = RTE_FLOW_ITEM_TYPE_VLAN,
663 .node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
665 [MLX5_EXPANSION_IPV4] = {
666 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
667 MLX5_EXPANSION_IPV4_TCP),
668 .type = RTE_FLOW_ITEM_TYPE_IPV4,
669 .rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
670 RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
672 [MLX5_EXPANSION_IPV4_UDP] = {
673 .type = RTE_FLOW_ITEM_TYPE_UDP,
674 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
676 [MLX5_EXPANSION_IPV4_TCP] = {
677 .type = RTE_FLOW_ITEM_TYPE_TCP,
678 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
680 [MLX5_EXPANSION_IPV6] = {
681 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
682 MLX5_EXPANSION_IPV6_TCP,
683 MLX5_EXPANSION_IPV6_FRAG_EXT),
684 .type = RTE_FLOW_ITEM_TYPE_IPV6,
685 .rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
686 RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
688 [MLX5_EXPANSION_IPV6_UDP] = {
689 .type = RTE_FLOW_ITEM_TYPE_UDP,
690 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
692 [MLX5_EXPANSION_IPV6_TCP] = {
693 .type = RTE_FLOW_ITEM_TYPE_TCP,
694 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
696 [MLX5_EXPANSION_IPV6_FRAG_EXT] = {
697 .type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
699 [MLX5_EXPANSION_GTP] = {
700 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
701 MLX5_EXPANSION_IPV6),
702 .type = RTE_FLOW_ITEM_TYPE_GTP,
706 static struct rte_flow_action_handle *
707 mlx5_action_handle_create(struct rte_eth_dev *dev,
708 const struct rte_flow_indir_action_conf *conf,
709 const struct rte_flow_action *action,
710 struct rte_flow_error *error);
711 static int mlx5_action_handle_destroy
712 (struct rte_eth_dev *dev,
713 struct rte_flow_action_handle *handle,
714 struct rte_flow_error *error);
715 static int mlx5_action_handle_update
716 (struct rte_eth_dev *dev,
717 struct rte_flow_action_handle *handle,
719 struct rte_flow_error *error);
720 static int mlx5_action_handle_query
721 (struct rte_eth_dev *dev,
722 const struct rte_flow_action_handle *handle,
724 struct rte_flow_error *error);
726 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
727 struct rte_flow_tunnel *app_tunnel,
728 struct rte_flow_action **actions,
729 uint32_t *num_of_actions,
730 struct rte_flow_error *error);
732 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
733 struct rte_flow_tunnel *app_tunnel,
734 struct rte_flow_item **items,
735 uint32_t *num_of_items,
736 struct rte_flow_error *error);
738 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
739 struct rte_flow_item *pmd_items,
740 uint32_t num_items, struct rte_flow_error *err);
742 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
743 struct rte_flow_action *pmd_actions,
744 uint32_t num_actions,
745 struct rte_flow_error *err);
747 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
749 struct rte_flow_restore_info *info,
750 struct rte_flow_error *err);
752 static const struct rte_flow_ops mlx5_flow_ops = {
753 .validate = mlx5_flow_validate,
754 .create = mlx5_flow_create,
755 .destroy = mlx5_flow_destroy,
756 .flush = mlx5_flow_flush,
757 .isolate = mlx5_flow_isolate,
758 .query = mlx5_flow_query,
759 .dev_dump = mlx5_flow_dev_dump,
760 .get_aged_flows = mlx5_flow_get_aged_flows,
761 .action_handle_create = mlx5_action_handle_create,
762 .action_handle_destroy = mlx5_action_handle_destroy,
763 .action_handle_update = mlx5_action_handle_update,
764 .action_handle_query = mlx5_action_handle_query,
765 .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
766 .tunnel_match = mlx5_flow_tunnel_match,
767 .tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
768 .tunnel_item_release = mlx5_flow_tunnel_item_release,
769 .get_restore_info = mlx5_flow_tunnel_get_restore_info,
772 /* Tunnel information. */
773 struct mlx5_flow_tunnel_info {
774 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
775 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
778 static struct mlx5_flow_tunnel_info tunnels_info[] = {
780 .tunnel = MLX5_FLOW_LAYER_VXLAN,
781 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
784 .tunnel = MLX5_FLOW_LAYER_GENEVE,
785 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
788 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
789 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
792 .tunnel = MLX5_FLOW_LAYER_GRE,
793 .ptype = RTE_PTYPE_TUNNEL_GRE,
796 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
797 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
800 .tunnel = MLX5_FLOW_LAYER_MPLS,
801 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
804 .tunnel = MLX5_FLOW_LAYER_NVGRE,
805 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
808 .tunnel = MLX5_FLOW_LAYER_IPIP,
809 .ptype = RTE_PTYPE_TUNNEL_IP,
812 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
813 .ptype = RTE_PTYPE_TUNNEL_IP,
816 .tunnel = MLX5_FLOW_LAYER_GTP,
817 .ptype = RTE_PTYPE_TUNNEL_GTPU,
824 * Translate tag ID to register.
827 * Pointer to the Ethernet device structure.
829 * The feature that request the register.
831 * The request register ID.
833 * Error description in case of any.
836 * The request register on success, a negative errno
837 * value otherwise and rte_errno is set.
840 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
841 enum mlx5_feature_name feature,
843 struct rte_flow_error *error)
845 struct mlx5_priv *priv = dev->data->dev_private;
846 struct mlx5_dev_config *config = &priv->config;
847 enum modify_reg start_reg;
848 bool skip_mtr_reg = false;
851 case MLX5_HAIRPIN_RX:
853 case MLX5_HAIRPIN_TX:
855 case MLX5_METADATA_RX:
856 switch (config->dv_xmeta_en) {
857 case MLX5_XMETA_MODE_LEGACY:
859 case MLX5_XMETA_MODE_META16:
861 case MLX5_XMETA_MODE_META32:
865 case MLX5_METADATA_TX:
867 case MLX5_METADATA_FDB:
868 switch (config->dv_xmeta_en) {
869 case MLX5_XMETA_MODE_LEGACY:
871 case MLX5_XMETA_MODE_META16:
873 case MLX5_XMETA_MODE_META32:
878 switch (config->dv_xmeta_en) {
879 case MLX5_XMETA_MODE_LEGACY:
881 case MLX5_XMETA_MODE_META16:
883 case MLX5_XMETA_MODE_META32:
889 * If meter color and meter id share one register, flow match
890 * should use the meter color register for match.
892 if (priv->mtr_reg_share)
893 return priv->mtr_color_reg;
895 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
898 case MLX5_ASO_FLOW_HIT:
899 case MLX5_ASO_CONNTRACK:
900 /* All features use the same REG_C. */
901 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
902 return priv->mtr_color_reg;
905 * Metadata COPY_MARK register using is in meter suffix sub
906 * flow while with meter. It's safe to share the same register.
908 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
911 * If meter is enable, it will engage the register for color
912 * match and flow match. If meter color match is not using the
913 * REG_C_2, need to skip the REG_C_x be used by meter color
915 * If meter is disable, free to use all available registers.
917 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
918 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
919 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
920 if (id > (uint32_t)(REG_C_7 - start_reg))
921 return rte_flow_error_set(error, EINVAL,
922 RTE_FLOW_ERROR_TYPE_ITEM,
923 NULL, "invalid tag id");
924 if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
925 return rte_flow_error_set(error, ENOTSUP,
926 RTE_FLOW_ERROR_TYPE_ITEM,
927 NULL, "unsupported tag id");
929 * This case means meter is using the REG_C_x great than 2.
930 * Take care not to conflict with meter color REG_C_x.
931 * If the available index REG_C_y >= REG_C_x, skip the
934 if (skip_mtr_reg && priv->sh->flow_mreg_c
935 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
936 if (id >= (uint32_t)(REG_C_7 - start_reg))
937 return rte_flow_error_set(error, EINVAL,
938 RTE_FLOW_ERROR_TYPE_ITEM,
939 NULL, "invalid tag id");
940 if (priv->sh->flow_mreg_c
941 [id + 1 + start_reg - REG_C_0] != REG_NON)
942 return priv->sh->flow_mreg_c
943 [id + 1 + start_reg - REG_C_0];
944 return rte_flow_error_set(error, ENOTSUP,
945 RTE_FLOW_ERROR_TYPE_ITEM,
946 NULL, "unsupported tag id");
948 return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
951 return rte_flow_error_set(error, EINVAL,
952 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
953 NULL, "invalid feature name");
957 * Check extensive flow metadata register support.
960 * Pointer to rte_eth_dev structure.
963 * True if device supports extensive flow metadata register, otherwise false.
966 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
968 struct mlx5_priv *priv = dev->data->dev_private;
971 * Having available reg_c can be regarded inclusively as supporting
972 * extensive flow metadata register, which could mean,
973 * - metadata register copy action by modify header.
974 * - 16 modify header actions is supported.
975 * - reg_c's are preserved across different domain (FDB and NIC) on
976 * packet loopback by flow lookup miss.
978 return priv->sh->flow_mreg_c[2] != REG_NON;
982 * Get the lowest priority.
985 * Pointer to the Ethernet device structure.
986 * @param[in] attributes
987 * Pointer to device flow rule attributes.
990 * The value of lowest priority of flow.
993 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
994 const struct rte_flow_attr *attr)
996 struct mlx5_priv *priv = dev->data->dev_private;
998 if (!attr->group && !attr->transfer)
999 return priv->sh->flow_max_priority - 2;
1000 return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1004 * Calculate matcher priority of the flow.
1007 * Pointer to the Ethernet device structure.
1009 * Pointer to device flow rule attributes.
1010 * @param[in] subpriority
1011 * The priority based on the items.
1012 * @param[in] external
1013 * Flow is user flow.
1015 * The matcher priority of the flow.
1018 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1019 const struct rte_flow_attr *attr,
1020 uint32_t subpriority, bool external)
1022 uint16_t priority = (uint16_t)attr->priority;
1023 struct mlx5_priv *priv = dev->data->dev_private;
1025 if (!attr->group && !attr->transfer) {
1026 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1027 priority = priv->sh->flow_max_priority - 1;
1028 return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1029 } else if (!external && attr->transfer && attr->group == 0 &&
1030 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1031 return (priv->sh->flow_max_priority - 1) * 3;
1033 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1034 priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1035 return priority * 3 + subpriority;
1039 * Verify the @p item specifications (spec, last, mask) are compatible with the
1043 * Item specification.
1045 * @p item->mask or flow default bit-masks.
1046 * @param[in] nic_mask
1047 * Bit-masks covering supported fields by the NIC to compare with user mask.
1049 * Bit-masks size in bytes.
1050 * @param[in] range_accepted
1051 * True if range of values is accepted for specific fields, false otherwise.
1053 * Pointer to error structure.
1056 * 0 on success, a negative errno value otherwise and rte_errno is set.
1059 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1060 const uint8_t *mask,
1061 const uint8_t *nic_mask,
1063 bool range_accepted,
1064 struct rte_flow_error *error)
1068 MLX5_ASSERT(nic_mask);
1069 for (i = 0; i < size; ++i)
1070 if ((nic_mask[i] | mask[i]) != nic_mask[i])
1071 return rte_flow_error_set(error, ENOTSUP,
1072 RTE_FLOW_ERROR_TYPE_ITEM,
1074 "mask enables non supported"
1076 if (!item->spec && (item->mask || item->last))
1077 return rte_flow_error_set(error, EINVAL,
1078 RTE_FLOW_ERROR_TYPE_ITEM, item,
1079 "mask/last without a spec is not"
1081 if (item->spec && item->last && !range_accepted) {
1087 for (i = 0; i < size; ++i) {
1088 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1089 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1091 ret = memcmp(spec, last, size);
1093 return rte_flow_error_set(error, EINVAL,
1094 RTE_FLOW_ERROR_TYPE_ITEM,
1096 "range is not valid");
1102 * Adjust the hash fields according to the @p flow information.
1104 * @param[in] dev_flow.
1105 * Pointer to the mlx5_flow.
1107 * 1 when the hash field is for a tunnel item.
1108 * @param[in] layer_types
1109 * RTE_ETH_RSS_* types.
1110 * @param[in] hash_fields
1114 * The hash fields that should be used.
1117 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1118 int tunnel __rte_unused, uint64_t layer_types,
1119 uint64_t hash_fields)
1121 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1122 int rss_request_inner = rss_desc->level >= 2;
1124 /* Check RSS hash level for tunnel. */
1125 if (tunnel && rss_request_inner)
1126 hash_fields |= IBV_RX_HASH_INNER;
1127 else if (tunnel || rss_request_inner)
1130 /* Check if requested layer matches RSS hash fields. */
1131 if (!(rss_desc->types & layer_types))
1137 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
1138 * if several tunnel rules are used on this queue, the tunnel ptype will be
1142 * Rx queue to update.
1145 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1148 uint32_t tunnel_ptype = 0;
1150 /* Look up for the ptype to use. */
1151 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1152 if (!rxq_ctrl->flow_tunnels_n[i])
1154 if (!tunnel_ptype) {
1155 tunnel_ptype = tunnels_info[i].ptype;
1161 rxq_ctrl->rxq.tunnel = tunnel_ptype;
1165 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1169 * Pointer to the Ethernet device structure.
1170 * @param[in] dev_handle
1171 * Pointer to device flow handle structure.
1174 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1175 struct mlx5_flow_handle *dev_handle)
1177 struct mlx5_priv *priv = dev->data->dev_private;
1178 const int mark = dev_handle->mark;
1179 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1180 struct mlx5_ind_table_obj *ind_tbl = NULL;
1183 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1184 struct mlx5_hrxq *hrxq;
1186 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1187 dev_handle->rix_hrxq);
1189 ind_tbl = hrxq->ind_table;
1190 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1191 struct mlx5_shared_action_rss *shared_rss;
1193 shared_rss = mlx5_ipool_get
1194 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1195 dev_handle->rix_srss);
1197 ind_tbl = shared_rss->ind_tbl;
1201 for (i = 0; i != ind_tbl->queues_n; ++i) {
1202 int idx = ind_tbl->queues[i];
1203 struct mlx5_rxq_ctrl *rxq_ctrl =
1204 container_of((*priv->rxqs)[idx],
1205 struct mlx5_rxq_ctrl, rxq);
1208 * To support metadata register copy on Tx loopback,
1209 * this must be always enabled (metadata may arive
1210 * from other port - not from local flows only.
1212 if (priv->config.dv_flow_en &&
1213 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1214 mlx5_flow_ext_mreg_supported(dev)) {
1215 rxq_ctrl->rxq.mark = 1;
1216 rxq_ctrl->flow_mark_n = 1;
1218 rxq_ctrl->rxq.mark = 1;
1219 rxq_ctrl->flow_mark_n++;
1224 /* Increase the counter matching the flow. */
1225 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1226 if ((tunnels_info[j].tunnel &
1227 dev_handle->layers) ==
1228 tunnels_info[j].tunnel) {
1229 rxq_ctrl->flow_tunnels_n[j]++;
1233 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1239 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1242 * Pointer to the Ethernet device structure.
1244 * Pointer to flow structure.
1247 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1249 struct mlx5_priv *priv = dev->data->dev_private;
1250 uint32_t handle_idx;
1251 struct mlx5_flow_handle *dev_handle;
1253 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1254 handle_idx, dev_handle, next)
1255 flow_drv_rxq_flags_set(dev, dev_handle);
1259 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1260 * device flow if no other flow uses it with the same kind of request.
1263 * Pointer to Ethernet device.
1264 * @param[in] dev_handle
1265 * Pointer to the device flow handle structure.
1268 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1269 struct mlx5_flow_handle *dev_handle)
1271 struct mlx5_priv *priv = dev->data->dev_private;
1272 const int mark = dev_handle->mark;
1273 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1274 struct mlx5_ind_table_obj *ind_tbl = NULL;
1277 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1278 struct mlx5_hrxq *hrxq;
1280 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1281 dev_handle->rix_hrxq);
1283 ind_tbl = hrxq->ind_table;
1284 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1285 struct mlx5_shared_action_rss *shared_rss;
1287 shared_rss = mlx5_ipool_get
1288 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1289 dev_handle->rix_srss);
1291 ind_tbl = shared_rss->ind_tbl;
1295 MLX5_ASSERT(dev->data->dev_started);
1296 for (i = 0; i != ind_tbl->queues_n; ++i) {
1297 int idx = ind_tbl->queues[i];
1298 struct mlx5_rxq_ctrl *rxq_ctrl =
1299 container_of((*priv->rxqs)[idx],
1300 struct mlx5_rxq_ctrl, rxq);
1302 if (priv->config.dv_flow_en &&
1303 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1304 mlx5_flow_ext_mreg_supported(dev)) {
1305 rxq_ctrl->rxq.mark = 1;
1306 rxq_ctrl->flow_mark_n = 1;
1308 rxq_ctrl->flow_mark_n--;
1309 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1314 /* Decrease the counter matching the flow. */
1315 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1316 if ((tunnels_info[j].tunnel &
1317 dev_handle->layers) ==
1318 tunnels_info[j].tunnel) {
1319 rxq_ctrl->flow_tunnels_n[j]--;
1323 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1329 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1330 * @p flow if no other flow uses it with the same kind of request.
1333 * Pointer to Ethernet device.
1335 * Pointer to the flow.
1338 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1340 struct mlx5_priv *priv = dev->data->dev_private;
1341 uint32_t handle_idx;
1342 struct mlx5_flow_handle *dev_handle;
1344 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1345 handle_idx, dev_handle, next)
1346 flow_drv_rxq_flags_trim(dev, dev_handle);
1350 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1353 * Pointer to Ethernet device.
1356 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1358 struct mlx5_priv *priv = dev->data->dev_private;
1361 for (i = 0; i != priv->rxqs_n; ++i) {
1362 struct mlx5_rxq_ctrl *rxq_ctrl;
1365 if (!(*priv->rxqs)[i])
1367 rxq_ctrl = container_of((*priv->rxqs)[i],
1368 struct mlx5_rxq_ctrl, rxq);
1369 rxq_ctrl->flow_mark_n = 0;
1370 rxq_ctrl->rxq.mark = 0;
1371 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1372 rxq_ctrl->flow_tunnels_n[j] = 0;
1373 rxq_ctrl->rxq.tunnel = 0;
1378 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1381 * Pointer to the Ethernet device structure.
1384 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1386 struct mlx5_priv *priv = dev->data->dev_private;
1387 struct mlx5_rxq_data *data;
1390 for (i = 0; i != priv->rxqs_n; ++i) {
1391 if (!(*priv->rxqs)[i])
1393 data = (*priv->rxqs)[i];
1394 if (!rte_flow_dynf_metadata_avail()) {
1395 data->dynf_meta = 0;
1396 data->flow_meta_mask = 0;
1397 data->flow_meta_offset = -1;
1398 data->flow_meta_port_mask = 0;
1400 data->dynf_meta = 1;
1401 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1402 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1403 data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1409 * return a pointer to the desired action in the list of actions.
1411 * @param[in] actions
1412 * The list of actions to search the action in.
1414 * The action to find.
1417 * Pointer to the action in the list, if found. NULL otherwise.
1419 const struct rte_flow_action *
1420 mlx5_flow_find_action(const struct rte_flow_action *actions,
1421 enum rte_flow_action_type action)
1423 if (actions == NULL)
1425 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1426 if (actions->type == action)
1432 * Validate the flag action.
1434 * @param[in] action_flags
1435 * Bit-fields that holds the actions detected until now.
1437 * Attributes of flow that includes this action.
1439 * Pointer to error structure.
1442 * 0 on success, a negative errno value otherwise and rte_errno is set.
1445 mlx5_flow_validate_action_flag(uint64_t action_flags,
1446 const struct rte_flow_attr *attr,
1447 struct rte_flow_error *error)
1449 if (action_flags & MLX5_FLOW_ACTION_MARK)
1450 return rte_flow_error_set(error, EINVAL,
1451 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1452 "can't mark and flag in same flow");
1453 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1454 return rte_flow_error_set(error, EINVAL,
1455 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1457 " actions in same flow");
1459 return rte_flow_error_set(error, ENOTSUP,
1460 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1461 "flag action not supported for "
1467 * Validate the mark action.
1470 * Pointer to the queue action.
1471 * @param[in] action_flags
1472 * Bit-fields that holds the actions detected until now.
1474 * Attributes of flow that includes this action.
1476 * Pointer to error structure.
1479 * 0 on success, a negative errno value otherwise and rte_errno is set.
1482 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1483 uint64_t action_flags,
1484 const struct rte_flow_attr *attr,
1485 struct rte_flow_error *error)
1487 const struct rte_flow_action_mark *mark = action->conf;
1490 return rte_flow_error_set(error, EINVAL,
1491 RTE_FLOW_ERROR_TYPE_ACTION,
1493 "configuration cannot be null");
1494 if (mark->id >= MLX5_FLOW_MARK_MAX)
1495 return rte_flow_error_set(error, EINVAL,
1496 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1498 "mark id must in 0 <= id < "
1499 RTE_STR(MLX5_FLOW_MARK_MAX));
1500 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1501 return rte_flow_error_set(error, EINVAL,
1502 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1503 "can't flag and mark in same flow");
1504 if (action_flags & MLX5_FLOW_ACTION_MARK)
1505 return rte_flow_error_set(error, EINVAL,
1506 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1507 "can't have 2 mark actions in same"
1510 return rte_flow_error_set(error, ENOTSUP,
1511 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1512 "mark action not supported for "
1518 * Validate the drop action.
1520 * @param[in] action_flags
1521 * Bit-fields that holds the actions detected until now.
1523 * Attributes of flow that includes this action.
1525 * Pointer to error structure.
1528 * 0 on success, a negative errno value otherwise and rte_errno is set.
1531 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1532 const struct rte_flow_attr *attr,
1533 struct rte_flow_error *error)
1536 return rte_flow_error_set(error, ENOTSUP,
1537 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1538 "drop action not supported for "
1544 * Validate the queue action.
1547 * Pointer to the queue action.
1548 * @param[in] action_flags
1549 * Bit-fields that holds the actions detected until now.
1551 * Pointer to the Ethernet device structure.
1553 * Attributes of flow that includes this action.
1555 * Pointer to error structure.
1558 * 0 on success, a negative errno value otherwise and rte_errno is set.
1561 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1562 uint64_t action_flags,
1563 struct rte_eth_dev *dev,
1564 const struct rte_flow_attr *attr,
1565 struct rte_flow_error *error)
1567 struct mlx5_priv *priv = dev->data->dev_private;
1568 const struct rte_flow_action_queue *queue = action->conf;
1570 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1571 return rte_flow_error_set(error, EINVAL,
1572 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1573 "can't have 2 fate actions in"
1576 return rte_flow_error_set(error, EINVAL,
1577 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1578 NULL, "No Rx queues configured");
1579 if (queue->index >= priv->rxqs_n)
1580 return rte_flow_error_set(error, EINVAL,
1581 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1583 "queue index out of range");
1584 if (!(*priv->rxqs)[queue->index])
1585 return rte_flow_error_set(error, EINVAL,
1586 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1588 "queue is not configured");
1590 return rte_flow_error_set(error, ENOTSUP,
1591 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1592 "queue action not supported for "
1598 * Validate the rss action.
1601 * Pointer to the Ethernet device structure.
1603 * Pointer to the queue action.
1605 * Pointer to error structure.
1608 * 0 on success, a negative errno value otherwise and rte_errno is set.
1611 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1612 const struct rte_flow_action *action,
1613 struct rte_flow_error *error)
1615 struct mlx5_priv *priv = dev->data->dev_private;
1616 const struct rte_flow_action_rss *rss = action->conf;
1617 enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1620 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1621 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1622 return rte_flow_error_set(error, ENOTSUP,
1623 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1625 "RSS hash function not supported");
1626 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1631 return rte_flow_error_set(error, ENOTSUP,
1632 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1634 "tunnel RSS is not supported");
1635 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1636 if (rss->key_len == 0 && rss->key != NULL)
1637 return rte_flow_error_set(error, ENOTSUP,
1638 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1640 "RSS hash key length 0");
1641 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1642 return rte_flow_error_set(error, ENOTSUP,
1643 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1645 "RSS hash key too small");
1646 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1647 return rte_flow_error_set(error, ENOTSUP,
1648 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1650 "RSS hash key too large");
1651 if (rss->queue_num > priv->config.ind_table_max_size)
1652 return rte_flow_error_set(error, ENOTSUP,
1653 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1655 "number of queues too large");
1656 if (rss->types & MLX5_RSS_HF_MASK)
1657 return rte_flow_error_set(error, ENOTSUP,
1658 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1660 "some RSS protocols are not"
1662 if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1663 !(rss->types & RTE_ETH_RSS_IP))
1664 return rte_flow_error_set(error, EINVAL,
1665 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1666 "L3 partial RSS requested but L3 RSS"
1667 " type not specified");
1668 if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1669 !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1670 return rte_flow_error_set(error, EINVAL,
1671 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1672 "L4 partial RSS requested but L4 RSS"
1673 " type not specified");
1675 return rte_flow_error_set(error, EINVAL,
1676 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1677 NULL, "No Rx queues configured");
1678 if (!rss->queue_num)
1679 return rte_flow_error_set(error, EINVAL,
1680 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1681 NULL, "No queues configured");
1682 for (i = 0; i != rss->queue_num; ++i) {
1683 struct mlx5_rxq_ctrl *rxq_ctrl;
1685 if (rss->queue[i] >= priv->rxqs_n)
1686 return rte_flow_error_set
1688 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1689 &rss->queue[i], "queue index out of range");
1690 if (!(*priv->rxqs)[rss->queue[i]])
1691 return rte_flow_error_set
1692 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1693 &rss->queue[i], "queue is not configured");
1694 rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1695 struct mlx5_rxq_ctrl, rxq);
1697 rxq_type = rxq_ctrl->type;
1698 if (rxq_type != rxq_ctrl->type)
1699 return rte_flow_error_set
1700 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1702 "combining hairpin and regular RSS queues is not supported");
1708 * Validate the rss action.
1711 * Pointer to the queue action.
1712 * @param[in] action_flags
1713 * Bit-fields that holds the actions detected until now.
1715 * Pointer to the Ethernet device structure.
1717 * Attributes of flow that includes this action.
1718 * @param[in] item_flags
1719 * Items that were detected.
1721 * Pointer to error structure.
1724 * 0 on success, a negative errno value otherwise and rte_errno is set.
1727 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1728 uint64_t action_flags,
1729 struct rte_eth_dev *dev,
1730 const struct rte_flow_attr *attr,
1731 uint64_t item_flags,
1732 struct rte_flow_error *error)
1734 const struct rte_flow_action_rss *rss = action->conf;
1735 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1738 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1739 return rte_flow_error_set(error, EINVAL,
1740 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1741 "can't have 2 fate actions"
1743 ret = mlx5_validate_action_rss(dev, action, error);
1747 return rte_flow_error_set(error, ENOTSUP,
1748 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1749 "rss action not supported for "
1751 if (rss->level > 1 && !tunnel)
1752 return rte_flow_error_set(error, EINVAL,
1753 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1754 "inner RSS is not supported for "
1755 "non-tunnel flows");
1756 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1757 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1758 return rte_flow_error_set(error, EINVAL,
1759 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1760 "RSS on eCPRI is not supported now");
1762 if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1764 (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1766 return rte_flow_error_set(error, EINVAL,
1767 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1768 "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1773 * Validate the default miss action.
1775 * @param[in] action_flags
1776 * Bit-fields that holds the actions detected until now.
1778 * Pointer to error structure.
1781 * 0 on success, a negative errno value otherwise and rte_errno is set.
1784 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1785 const struct rte_flow_attr *attr,
1786 struct rte_flow_error *error)
1788 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1789 return rte_flow_error_set(error, EINVAL,
1790 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1791 "can't have 2 fate actions in"
1794 return rte_flow_error_set(error, ENOTSUP,
1795 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1796 "default miss action not supported "
1799 return rte_flow_error_set(error, ENOTSUP,
1800 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1801 "only group 0 is supported");
1803 return rte_flow_error_set(error, ENOTSUP,
1804 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1805 NULL, "transfer is not supported");
1810 * Validate the count action.
1813 * Pointer to the Ethernet device structure.
1815 * Attributes of flow that includes this action.
1817 * Pointer to error structure.
1820 * 0 on success, a negative errno value otherwise and rte_errno is set.
1823 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1824 const struct rte_flow_attr *attr,
1825 struct rte_flow_error *error)
1828 return rte_flow_error_set(error, ENOTSUP,
1829 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1830 "count action not supported for "
1836 * Validate the ASO CT action.
1839 * Pointer to the Ethernet device structure.
1840 * @param[in] conntrack
1841 * Pointer to the CT action profile.
1843 * Pointer to error structure.
1846 * 0 on success, a negative errno value otherwise and rte_errno is set.
1849 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1850 const struct rte_flow_action_conntrack *conntrack,
1851 struct rte_flow_error *error)
1855 if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1856 return rte_flow_error_set(error, EINVAL,
1857 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1858 "Invalid CT state");
1859 if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1860 return rte_flow_error_set(error, EINVAL,
1861 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1862 "Invalid last TCP packet flag");
1867 * Verify the @p attributes will be correctly understood by the NIC and store
1868 * them in the @p flow if everything is correct.
1871 * Pointer to the Ethernet device structure.
1872 * @param[in] attributes
1873 * Pointer to flow attributes
1875 * Pointer to error structure.
1878 * 0 on success, a negative errno value otherwise and rte_errno is set.
1881 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1882 const struct rte_flow_attr *attributes,
1883 struct rte_flow_error *error)
1885 struct mlx5_priv *priv = dev->data->dev_private;
1886 uint32_t priority_max = priv->sh->flow_max_priority - 1;
1888 if (attributes->group)
1889 return rte_flow_error_set(error, ENOTSUP,
1890 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1891 NULL, "groups is not supported");
1892 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1893 attributes->priority >= priority_max)
1894 return rte_flow_error_set(error, ENOTSUP,
1895 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1896 NULL, "priority out of range");
1897 if (attributes->egress)
1898 return rte_flow_error_set(error, ENOTSUP,
1899 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1900 "egress is not supported");
1901 if (attributes->transfer && !priv->config.dv_esw_en)
1902 return rte_flow_error_set(error, ENOTSUP,
1903 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1904 NULL, "transfer is not supported");
1905 if (!attributes->ingress)
1906 return rte_flow_error_set(error, EINVAL,
1907 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1909 "ingress attribute is mandatory");
1914 * Validate ICMP6 item.
1917 * Item specification.
1918 * @param[in] item_flags
1919 * Bit-fields that holds the items detected until now.
1920 * @param[in] ext_vlan_sup
1921 * Whether extended VLAN features are supported or not.
1923 * Pointer to error structure.
1926 * 0 on success, a negative errno value otherwise and rte_errno is set.
1929 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1930 uint64_t item_flags,
1931 uint8_t target_protocol,
1932 struct rte_flow_error *error)
1934 const struct rte_flow_item_icmp6 *mask = item->mask;
1935 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1936 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1937 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1938 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1939 MLX5_FLOW_LAYER_OUTER_L4;
1942 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1943 return rte_flow_error_set(error, EINVAL,
1944 RTE_FLOW_ERROR_TYPE_ITEM, item,
1945 "protocol filtering not compatible"
1946 " with ICMP6 layer");
1947 if (!(item_flags & l3m))
1948 return rte_flow_error_set(error, EINVAL,
1949 RTE_FLOW_ERROR_TYPE_ITEM, item,
1950 "IPv6 is mandatory to filter on"
1952 if (item_flags & l4m)
1953 return rte_flow_error_set(error, EINVAL,
1954 RTE_FLOW_ERROR_TYPE_ITEM, item,
1955 "multiple L4 layers not supported");
1957 mask = &rte_flow_item_icmp6_mask;
1958 ret = mlx5_flow_item_acceptable
1959 (item, (const uint8_t *)mask,
1960 (const uint8_t *)&rte_flow_item_icmp6_mask,
1961 sizeof(struct rte_flow_item_icmp6),
1962 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1969 * Validate ICMP item.
1972 * Item specification.
1973 * @param[in] item_flags
1974 * Bit-fields that holds the items detected until now.
1976 * Pointer to error structure.
1979 * 0 on success, a negative errno value otherwise and rte_errno is set.
1982 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1983 uint64_t item_flags,
1984 uint8_t target_protocol,
1985 struct rte_flow_error *error)
1987 const struct rte_flow_item_icmp *mask = item->mask;
1988 const struct rte_flow_item_icmp nic_mask = {
1989 .hdr.icmp_type = 0xff,
1990 .hdr.icmp_code = 0xff,
1991 .hdr.icmp_ident = RTE_BE16(0xffff),
1992 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1994 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1995 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1996 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1997 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1998 MLX5_FLOW_LAYER_OUTER_L4;
2001 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2002 return rte_flow_error_set(error, EINVAL,
2003 RTE_FLOW_ERROR_TYPE_ITEM, item,
2004 "protocol filtering not compatible"
2005 " with ICMP layer");
2006 if (!(item_flags & l3m))
2007 return rte_flow_error_set(error, EINVAL,
2008 RTE_FLOW_ERROR_TYPE_ITEM, item,
2009 "IPv4 is mandatory to filter"
2011 if (item_flags & l4m)
2012 return rte_flow_error_set(error, EINVAL,
2013 RTE_FLOW_ERROR_TYPE_ITEM, item,
2014 "multiple L4 layers not supported");
2017 ret = mlx5_flow_item_acceptable
2018 (item, (const uint8_t *)mask,
2019 (const uint8_t *)&nic_mask,
2020 sizeof(struct rte_flow_item_icmp),
2021 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2028 * Validate Ethernet item.
2031 * Item specification.
2032 * @param[in] item_flags
2033 * Bit-fields that holds the items detected until now.
2035 * Pointer to error structure.
2038 * 0 on success, a negative errno value otherwise and rte_errno is set.
2041 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2042 uint64_t item_flags, bool ext_vlan_sup,
2043 struct rte_flow_error *error)
2045 const struct rte_flow_item_eth *mask = item->mask;
2046 const struct rte_flow_item_eth nic_mask = {
2047 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2048 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2049 .type = RTE_BE16(0xffff),
2050 .has_vlan = ext_vlan_sup ? 1 : 0,
2053 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2054 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
2055 MLX5_FLOW_LAYER_OUTER_L2;
2057 if (item_flags & ethm)
2058 return rte_flow_error_set(error, ENOTSUP,
2059 RTE_FLOW_ERROR_TYPE_ITEM, item,
2060 "multiple L2 layers not supported");
2061 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2062 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2063 return rte_flow_error_set(error, EINVAL,
2064 RTE_FLOW_ERROR_TYPE_ITEM, item,
2065 "L2 layer should not follow "
2067 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2068 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2069 return rte_flow_error_set(error, EINVAL,
2070 RTE_FLOW_ERROR_TYPE_ITEM, item,
2071 "L2 layer should not follow VLAN");
2072 if (item_flags & MLX5_FLOW_LAYER_GTP)
2073 return rte_flow_error_set(error, EINVAL,
2074 RTE_FLOW_ERROR_TYPE_ITEM, item,
2075 "L2 layer should not follow GTP");
2077 mask = &rte_flow_item_eth_mask;
2078 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2079 (const uint8_t *)&nic_mask,
2080 sizeof(struct rte_flow_item_eth),
2081 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2086 * Validate VLAN item.
2089 * Item specification.
2090 * @param[in] item_flags
2091 * Bit-fields that holds the items detected until now.
2093 * Ethernet device flow is being created on.
2095 * Pointer to error structure.
2098 * 0 on success, a negative errno value otherwise and rte_errno is set.
2101 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2102 uint64_t item_flags,
2103 struct rte_eth_dev *dev,
2104 struct rte_flow_error *error)
2106 const struct rte_flow_item_vlan *spec = item->spec;
2107 const struct rte_flow_item_vlan *mask = item->mask;
2108 const struct rte_flow_item_vlan nic_mask = {
2109 .tci = RTE_BE16(UINT16_MAX),
2110 .inner_type = RTE_BE16(UINT16_MAX),
2112 uint16_t vlan_tag = 0;
2113 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2115 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2116 MLX5_FLOW_LAYER_INNER_L4) :
2117 (MLX5_FLOW_LAYER_OUTER_L3 |
2118 MLX5_FLOW_LAYER_OUTER_L4);
2119 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2120 MLX5_FLOW_LAYER_OUTER_VLAN;
2122 if (item_flags & vlanm)
2123 return rte_flow_error_set(error, EINVAL,
2124 RTE_FLOW_ERROR_TYPE_ITEM, item,
2125 "multiple VLAN layers not supported");
2126 else if ((item_flags & l34m) != 0)
2127 return rte_flow_error_set(error, EINVAL,
2128 RTE_FLOW_ERROR_TYPE_ITEM, item,
2129 "VLAN cannot follow L3/L4 layer");
2131 mask = &rte_flow_item_vlan_mask;
2132 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2133 (const uint8_t *)&nic_mask,
2134 sizeof(struct rte_flow_item_vlan),
2135 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2138 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2139 struct mlx5_priv *priv = dev->data->dev_private;
2141 if (priv->vmwa_context) {
2143 * Non-NULL context means we have a virtual machine
2144 * and SR-IOV enabled, we have to create VLAN interface
2145 * to make hypervisor to setup E-Switch vport
2146 * context correctly. We avoid creating the multiple
2147 * VLAN interfaces, so we cannot support VLAN tag mask.
2149 return rte_flow_error_set(error, EINVAL,
2150 RTE_FLOW_ERROR_TYPE_ITEM,
2152 "VLAN tag mask is not"
2153 " supported in virtual"
2158 vlan_tag = spec->tci;
2159 vlan_tag &= mask->tci;
2162 * From verbs perspective an empty VLAN is equivalent
2163 * to a packet without VLAN layer.
2166 return rte_flow_error_set(error, EINVAL,
2167 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2169 "VLAN cannot be empty");
2174 * Validate IPV4 item.
2177 * Item specification.
2178 * @param[in] item_flags
2179 * Bit-fields that holds the items detected until now.
2180 * @param[in] last_item
2181 * Previous validated item in the pattern items.
2182 * @param[in] ether_type
2183 * Type in the ethernet layer header (including dot1q).
2184 * @param[in] acc_mask
2185 * Acceptable mask, if NULL default internal default mask
2186 * will be used to check whether item fields are supported.
2187 * @param[in] range_accepted
2188 * True if range of values is accepted for specific fields, false otherwise.
2190 * Pointer to error structure.
2193 * 0 on success, a negative errno value otherwise and rte_errno is set.
2196 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2197 uint64_t item_flags,
2199 uint16_t ether_type,
2200 const struct rte_flow_item_ipv4 *acc_mask,
2201 bool range_accepted,
2202 struct rte_flow_error *error)
2204 const struct rte_flow_item_ipv4 *mask = item->mask;
2205 const struct rte_flow_item_ipv4 *spec = item->spec;
2206 const struct rte_flow_item_ipv4 nic_mask = {
2208 .src_addr = RTE_BE32(0xffffffff),
2209 .dst_addr = RTE_BE32(0xffffffff),
2210 .type_of_service = 0xff,
2211 .next_proto_id = 0xff,
2214 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2215 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2216 MLX5_FLOW_LAYER_OUTER_L3;
2217 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2218 MLX5_FLOW_LAYER_OUTER_L4;
2220 uint8_t next_proto = 0xFF;
2221 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2222 MLX5_FLOW_LAYER_OUTER_VLAN |
2223 MLX5_FLOW_LAYER_INNER_VLAN);
2225 if ((last_item & l2_vlan) && ether_type &&
2226 ether_type != RTE_ETHER_TYPE_IPV4)
2227 return rte_flow_error_set(error, EINVAL,
2228 RTE_FLOW_ERROR_TYPE_ITEM, item,
2229 "IPv4 cannot follow L2/VLAN layer "
2230 "which ether type is not IPv4");
2231 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2233 next_proto = mask->hdr.next_proto_id &
2234 spec->hdr.next_proto_id;
2235 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2236 return rte_flow_error_set(error, EINVAL,
2237 RTE_FLOW_ERROR_TYPE_ITEM,
2242 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2243 return rte_flow_error_set(error, EINVAL,
2244 RTE_FLOW_ERROR_TYPE_ITEM, item,
2245 "wrong tunnel type - IPv6 specified "
2246 "but IPv4 item provided");
2247 if (item_flags & l3m)
2248 return rte_flow_error_set(error, ENOTSUP,
2249 RTE_FLOW_ERROR_TYPE_ITEM, item,
2250 "multiple L3 layers not supported");
2251 else if (item_flags & l4m)
2252 return rte_flow_error_set(error, EINVAL,
2253 RTE_FLOW_ERROR_TYPE_ITEM, item,
2254 "L3 cannot follow an L4 layer.");
2255 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2256 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2257 return rte_flow_error_set(error, EINVAL,
2258 RTE_FLOW_ERROR_TYPE_ITEM, item,
2259 "L3 cannot follow an NVGRE layer.");
2261 mask = &rte_flow_item_ipv4_mask;
2262 else if (mask->hdr.next_proto_id != 0 &&
2263 mask->hdr.next_proto_id != 0xff)
2264 return rte_flow_error_set(error, EINVAL,
2265 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2266 "partial mask is not supported"
2268 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2269 acc_mask ? (const uint8_t *)acc_mask
2270 : (const uint8_t *)&nic_mask,
2271 sizeof(struct rte_flow_item_ipv4),
2272 range_accepted, error);
2279 * Validate IPV6 item.
2282 * Item specification.
2283 * @param[in] item_flags
2284 * Bit-fields that holds the items detected until now.
2285 * @param[in] last_item
2286 * Previous validated item in the pattern items.
2287 * @param[in] ether_type
2288 * Type in the ethernet layer header (including dot1q).
2289 * @param[in] acc_mask
2290 * Acceptable mask, if NULL default internal default mask
2291 * will be used to check whether item fields are supported.
2293 * Pointer to error structure.
2296 * 0 on success, a negative errno value otherwise and rte_errno is set.
2299 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2300 uint64_t item_flags,
2302 uint16_t ether_type,
2303 const struct rte_flow_item_ipv6 *acc_mask,
2304 struct rte_flow_error *error)
2306 const struct rte_flow_item_ipv6 *mask = item->mask;
2307 const struct rte_flow_item_ipv6 *spec = item->spec;
2308 const struct rte_flow_item_ipv6 nic_mask = {
2311 "\xff\xff\xff\xff\xff\xff\xff\xff"
2312 "\xff\xff\xff\xff\xff\xff\xff\xff",
2314 "\xff\xff\xff\xff\xff\xff\xff\xff"
2315 "\xff\xff\xff\xff\xff\xff\xff\xff",
2316 .vtc_flow = RTE_BE32(0xffffffff),
2320 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2321 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2322 MLX5_FLOW_LAYER_OUTER_L3;
2323 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2324 MLX5_FLOW_LAYER_OUTER_L4;
2326 uint8_t next_proto = 0xFF;
2327 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2328 MLX5_FLOW_LAYER_OUTER_VLAN |
2329 MLX5_FLOW_LAYER_INNER_VLAN);
2331 if ((last_item & l2_vlan) && ether_type &&
2332 ether_type != RTE_ETHER_TYPE_IPV6)
2333 return rte_flow_error_set(error, EINVAL,
2334 RTE_FLOW_ERROR_TYPE_ITEM, item,
2335 "IPv6 cannot follow L2/VLAN layer "
2336 "which ether type is not IPv6");
2337 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2338 next_proto = spec->hdr.proto;
2339 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2340 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2341 return rte_flow_error_set(error, EINVAL,
2342 RTE_FLOW_ERROR_TYPE_ITEM,
2347 if (next_proto == IPPROTO_HOPOPTS ||
2348 next_proto == IPPROTO_ROUTING ||
2349 next_proto == IPPROTO_FRAGMENT ||
2350 next_proto == IPPROTO_ESP ||
2351 next_proto == IPPROTO_AH ||
2352 next_proto == IPPROTO_DSTOPTS)
2353 return rte_flow_error_set(error, EINVAL,
2354 RTE_FLOW_ERROR_TYPE_ITEM, item,
2355 "IPv6 proto (next header) should "
2356 "not be set as extension header");
2357 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2358 return rte_flow_error_set(error, EINVAL,
2359 RTE_FLOW_ERROR_TYPE_ITEM, item,
2360 "wrong tunnel type - IPv4 specified "
2361 "but IPv6 item provided");
2362 if (item_flags & l3m)
2363 return rte_flow_error_set(error, ENOTSUP,
2364 RTE_FLOW_ERROR_TYPE_ITEM, item,
2365 "multiple L3 layers not supported");
2366 else if (item_flags & l4m)
2367 return rte_flow_error_set(error, EINVAL,
2368 RTE_FLOW_ERROR_TYPE_ITEM, item,
2369 "L3 cannot follow an L4 layer.");
2370 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2371 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2372 return rte_flow_error_set(error, EINVAL,
2373 RTE_FLOW_ERROR_TYPE_ITEM, item,
2374 "L3 cannot follow an NVGRE layer.");
2376 mask = &rte_flow_item_ipv6_mask;
2377 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2378 acc_mask ? (const uint8_t *)acc_mask
2379 : (const uint8_t *)&nic_mask,
2380 sizeof(struct rte_flow_item_ipv6),
2381 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2388 * Validate UDP item.
2391 * Item specification.
2392 * @param[in] item_flags
2393 * Bit-fields that holds the items detected until now.
2394 * @param[in] target_protocol
2395 * The next protocol in the previous item.
2396 * @param[in] flow_mask
2397 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2399 * Pointer to error structure.
2402 * 0 on success, a negative errno value otherwise and rte_errno is set.
2405 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2406 uint64_t item_flags,
2407 uint8_t target_protocol,
2408 struct rte_flow_error *error)
2410 const struct rte_flow_item_udp *mask = item->mask;
2411 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2412 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2413 MLX5_FLOW_LAYER_OUTER_L3;
2414 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2415 MLX5_FLOW_LAYER_OUTER_L4;
2418 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2419 return rte_flow_error_set(error, EINVAL,
2420 RTE_FLOW_ERROR_TYPE_ITEM, item,
2421 "protocol filtering not compatible"
2423 if (!(item_flags & l3m))
2424 return rte_flow_error_set(error, EINVAL,
2425 RTE_FLOW_ERROR_TYPE_ITEM, item,
2426 "L3 is mandatory to filter on L4");
2427 if (item_flags & l4m)
2428 return rte_flow_error_set(error, EINVAL,
2429 RTE_FLOW_ERROR_TYPE_ITEM, item,
2430 "multiple L4 layers not supported");
2432 mask = &rte_flow_item_udp_mask;
2433 ret = mlx5_flow_item_acceptable
2434 (item, (const uint8_t *)mask,
2435 (const uint8_t *)&rte_flow_item_udp_mask,
2436 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2444 * Validate TCP item.
2447 * Item specification.
2448 * @param[in] item_flags
2449 * Bit-fields that holds the items detected until now.
2450 * @param[in] target_protocol
2451 * The next protocol in the previous item.
2453 * Pointer to error structure.
2456 * 0 on success, a negative errno value otherwise and rte_errno is set.
2459 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2460 uint64_t item_flags,
2461 uint8_t target_protocol,
2462 const struct rte_flow_item_tcp *flow_mask,
2463 struct rte_flow_error *error)
2465 const struct rte_flow_item_tcp *mask = item->mask;
2466 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2467 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2468 MLX5_FLOW_LAYER_OUTER_L3;
2469 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2470 MLX5_FLOW_LAYER_OUTER_L4;
2473 MLX5_ASSERT(flow_mask);
2474 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2475 return rte_flow_error_set(error, EINVAL,
2476 RTE_FLOW_ERROR_TYPE_ITEM, item,
2477 "protocol filtering not compatible"
2479 if (!(item_flags & l3m))
2480 return rte_flow_error_set(error, EINVAL,
2481 RTE_FLOW_ERROR_TYPE_ITEM, item,
2482 "L3 is mandatory to filter on L4");
2483 if (item_flags & l4m)
2484 return rte_flow_error_set(error, EINVAL,
2485 RTE_FLOW_ERROR_TYPE_ITEM, item,
2486 "multiple L4 layers not supported");
2488 mask = &rte_flow_item_tcp_mask;
2489 ret = mlx5_flow_item_acceptable
2490 (item, (const uint8_t *)mask,
2491 (const uint8_t *)flow_mask,
2492 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2500 * Validate VXLAN item.
2503 * Pointer to the Ethernet device structure.
2504 * @param[in] udp_dport
2505 * UDP destination port
2507 * Item specification.
2508 * @param[in] item_flags
2509 * Bit-fields that holds the items detected until now.
2511 * Flow rule attributes.
2513 * Pointer to error structure.
2516 * 0 on success, a negative errno value otherwise and rte_errno is set.
2519 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2521 const struct rte_flow_item *item,
2522 uint64_t item_flags,
2523 const struct rte_flow_attr *attr,
2524 struct rte_flow_error *error)
2526 const struct rte_flow_item_vxlan *spec = item->spec;
2527 const struct rte_flow_item_vxlan *mask = item->mask;
2529 struct mlx5_priv *priv = dev->data->dev_private;
2533 } id = { .vlan_id = 0, };
2534 const struct rte_flow_item_vxlan nic_mask = {
2535 .vni = "\xff\xff\xff",
2538 const struct rte_flow_item_vxlan *valid_mask;
2540 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2541 return rte_flow_error_set(error, ENOTSUP,
2542 RTE_FLOW_ERROR_TYPE_ITEM, item,
2543 "multiple tunnel layers not"
2545 valid_mask = &rte_flow_item_vxlan_mask;
2547 * Verify only UDPv4 is present as defined in
2548 * https://tools.ietf.org/html/rfc7348
2550 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2551 return rte_flow_error_set(error, EINVAL,
2552 RTE_FLOW_ERROR_TYPE_ITEM, item,
2553 "no outer UDP layer found");
2555 mask = &rte_flow_item_vxlan_mask;
2557 if (priv->sh->steering_format_version !=
2558 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2559 !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2560 /* FDB domain & NIC domain non-zero group */
2561 if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2562 valid_mask = &nic_mask;
2563 /* Group zero in NIC domain */
2564 if (!attr->group && !attr->transfer &&
2565 priv->sh->tunnel_header_0_1)
2566 valid_mask = &nic_mask;
2568 ret = mlx5_flow_item_acceptable
2569 (item, (const uint8_t *)mask,
2570 (const uint8_t *)valid_mask,
2571 sizeof(struct rte_flow_item_vxlan),
2572 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2576 memcpy(&id.vni[1], spec->vni, 3);
2577 memcpy(&id.vni[1], mask->vni, 3);
2579 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2580 return rte_flow_error_set(error, ENOTSUP,
2581 RTE_FLOW_ERROR_TYPE_ITEM, item,
2582 "VXLAN tunnel must be fully defined");
2587 * Validate VXLAN_GPE item.
2590 * Item specification.
2591 * @param[in] item_flags
2592 * Bit-fields that holds the items detected until now.
2594 * Pointer to the private data structure.
2595 * @param[in] target_protocol
2596 * The next protocol in the previous item.
2598 * Pointer to error structure.
2601 * 0 on success, a negative errno value otherwise and rte_errno is set.
2604 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2605 uint64_t item_flags,
2606 struct rte_eth_dev *dev,
2607 struct rte_flow_error *error)
2609 struct mlx5_priv *priv = dev->data->dev_private;
2610 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2611 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2616 } id = { .vlan_id = 0, };
2618 if (!priv->config.l3_vxlan_en)
2619 return rte_flow_error_set(error, ENOTSUP,
2620 RTE_FLOW_ERROR_TYPE_ITEM, item,
2621 "L3 VXLAN is not enabled by device"
2622 " parameter and/or not configured in"
2624 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2625 return rte_flow_error_set(error, ENOTSUP,
2626 RTE_FLOW_ERROR_TYPE_ITEM, item,
2627 "multiple tunnel layers not"
2630 * Verify only UDPv4 is present as defined in
2631 * https://tools.ietf.org/html/rfc7348
2633 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2634 return rte_flow_error_set(error, EINVAL,
2635 RTE_FLOW_ERROR_TYPE_ITEM, item,
2636 "no outer UDP layer found");
2638 mask = &rte_flow_item_vxlan_gpe_mask;
2639 ret = mlx5_flow_item_acceptable
2640 (item, (const uint8_t *)mask,
2641 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2642 sizeof(struct rte_flow_item_vxlan_gpe),
2643 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2648 return rte_flow_error_set(error, ENOTSUP,
2649 RTE_FLOW_ERROR_TYPE_ITEM,
2651 "VxLAN-GPE protocol"
2653 memcpy(&id.vni[1], spec->vni, 3);
2654 memcpy(&id.vni[1], mask->vni, 3);
2656 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2657 return rte_flow_error_set(error, ENOTSUP,
2658 RTE_FLOW_ERROR_TYPE_ITEM, item,
2659 "VXLAN-GPE tunnel must be fully"
2664 * Validate GRE Key item.
2667 * Item specification.
2668 * @param[in] item_flags
2669 * Bit flags to mark detected items.
2670 * @param[in] gre_item
2671 * Pointer to gre_item
2673 * Pointer to error structure.
2676 * 0 on success, a negative errno value otherwise and rte_errno is set.
2679 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2680 uint64_t item_flags,
2681 const struct rte_flow_item *gre_item,
2682 struct rte_flow_error *error)
2684 const rte_be32_t *mask = item->mask;
2686 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2687 const struct rte_flow_item_gre *gre_spec;
2688 const struct rte_flow_item_gre *gre_mask;
2690 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2691 return rte_flow_error_set(error, ENOTSUP,
2692 RTE_FLOW_ERROR_TYPE_ITEM, item,
2693 "Multiple GRE key not support");
2694 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2695 return rte_flow_error_set(error, ENOTSUP,
2696 RTE_FLOW_ERROR_TYPE_ITEM, item,
2697 "No preceding GRE header");
2698 if (item_flags & MLX5_FLOW_LAYER_INNER)
2699 return rte_flow_error_set(error, ENOTSUP,
2700 RTE_FLOW_ERROR_TYPE_ITEM, item,
2701 "GRE key following a wrong item");
2702 gre_mask = gre_item->mask;
2704 gre_mask = &rte_flow_item_gre_mask;
2705 gre_spec = gre_item->spec;
2706 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2707 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2708 return rte_flow_error_set(error, EINVAL,
2709 RTE_FLOW_ERROR_TYPE_ITEM, item,
2710 "Key bit must be on");
2713 mask = &gre_key_default_mask;
2714 ret = mlx5_flow_item_acceptable
2715 (item, (const uint8_t *)mask,
2716 (const uint8_t *)&gre_key_default_mask,
2717 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2722 * Validate GRE item.
2725 * Item specification.
2726 * @param[in] item_flags
2727 * Bit flags to mark detected items.
2728 * @param[in] target_protocol
2729 * The next protocol in the previous item.
2731 * Pointer to error structure.
2734 * 0 on success, a negative errno value otherwise and rte_errno is set.
2737 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2738 uint64_t item_flags,
2739 uint8_t target_protocol,
2740 struct rte_flow_error *error)
2742 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2743 const struct rte_flow_item_gre *mask = item->mask;
2745 const struct rte_flow_item_gre nic_mask = {
2746 .c_rsvd0_ver = RTE_BE16(0xB000),
2747 .protocol = RTE_BE16(UINT16_MAX),
2750 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2751 return rte_flow_error_set(error, EINVAL,
2752 RTE_FLOW_ERROR_TYPE_ITEM, item,
2753 "protocol filtering not compatible"
2754 " with this GRE layer");
2755 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2756 return rte_flow_error_set(error, ENOTSUP,
2757 RTE_FLOW_ERROR_TYPE_ITEM, item,
2758 "multiple tunnel layers not"
2760 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2761 return rte_flow_error_set(error, ENOTSUP,
2762 RTE_FLOW_ERROR_TYPE_ITEM, item,
2763 "L3 Layer is missing");
2765 mask = &rte_flow_item_gre_mask;
2766 ret = mlx5_flow_item_acceptable
2767 (item, (const uint8_t *)mask,
2768 (const uint8_t *)&nic_mask,
2769 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2773 #ifndef HAVE_MLX5DV_DR
2774 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2775 if (spec && (spec->protocol & mask->protocol))
2776 return rte_flow_error_set(error, ENOTSUP,
2777 RTE_FLOW_ERROR_TYPE_ITEM, item,
2778 "without MPLS support the"
2779 " specification cannot be used for"
2787 * Validate Geneve item.
2790 * Item specification.
2791 * @param[in] itemFlags
2792 * Bit-fields that holds the items detected until now.
2794 * Pointer to the private data structure.
2796 * Pointer to error structure.
2799 * 0 on success, a negative errno value otherwise and rte_errno is set.
2803 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2804 uint64_t item_flags,
2805 struct rte_eth_dev *dev,
2806 struct rte_flow_error *error)
2808 struct mlx5_priv *priv = dev->data->dev_private;
2809 const struct rte_flow_item_geneve *spec = item->spec;
2810 const struct rte_flow_item_geneve *mask = item->mask;
2813 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2814 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2815 const struct rte_flow_item_geneve nic_mask = {
2816 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2817 .vni = "\xff\xff\xff",
2818 .protocol = RTE_BE16(UINT16_MAX),
2821 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2822 return rte_flow_error_set(error, ENOTSUP,
2823 RTE_FLOW_ERROR_TYPE_ITEM, item,
2824 "L3 Geneve is not enabled by device"
2825 " parameter and/or not configured in"
2827 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2828 return rte_flow_error_set(error, ENOTSUP,
2829 RTE_FLOW_ERROR_TYPE_ITEM, item,
2830 "multiple tunnel layers not"
2833 * Verify only UDPv4 is present as defined in
2834 * https://tools.ietf.org/html/rfc7348
2836 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2837 return rte_flow_error_set(error, EINVAL,
2838 RTE_FLOW_ERROR_TYPE_ITEM, item,
2839 "no outer UDP layer found");
2841 mask = &rte_flow_item_geneve_mask;
2842 ret = mlx5_flow_item_acceptable
2843 (item, (const uint8_t *)mask,
2844 (const uint8_t *)&nic_mask,
2845 sizeof(struct rte_flow_item_geneve),
2846 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2850 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2851 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2852 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2853 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2854 return rte_flow_error_set(error, ENOTSUP,
2855 RTE_FLOW_ERROR_TYPE_ITEM,
2857 "Geneve protocol unsupported"
2858 " fields are being used");
2859 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2860 return rte_flow_error_set
2862 RTE_FLOW_ERROR_TYPE_ITEM,
2864 "Unsupported Geneve options length");
2866 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2867 return rte_flow_error_set
2869 RTE_FLOW_ERROR_TYPE_ITEM, item,
2870 "Geneve tunnel must be fully defined");
2875 * Validate Geneve TLV option item.
2878 * Item specification.
2879 * @param[in] last_item
2880 * Previous validated item in the pattern items.
2881 * @param[in] geneve_item
2882 * Previous GENEVE item specification.
2884 * Pointer to the rte_eth_dev structure.
2886 * Pointer to error structure.
2889 * 0 on success, a negative errno value otherwise and rte_errno is set.
2892 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2894 const struct rte_flow_item *geneve_item,
2895 struct rte_eth_dev *dev,
2896 struct rte_flow_error *error)
2898 struct mlx5_priv *priv = dev->data->dev_private;
2899 struct mlx5_dev_ctx_shared *sh = priv->sh;
2900 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2901 struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2902 uint8_t data_max_supported =
2903 hca_attr->max_geneve_tlv_option_data_len * 4;
2904 struct mlx5_dev_config *config = &priv->config;
2905 const struct rte_flow_item_geneve *geneve_spec;
2906 const struct rte_flow_item_geneve *geneve_mask;
2907 const struct rte_flow_item_geneve_opt *spec = item->spec;
2908 const struct rte_flow_item_geneve_opt *mask = item->mask;
2910 unsigned int data_len;
2911 uint8_t tlv_option_len;
2912 uint16_t optlen_m, optlen_v;
2913 const struct rte_flow_item_geneve_opt full_mask = {
2914 .option_class = RTE_BE16(0xffff),
2915 .option_type = 0xff,
2920 mask = &rte_flow_item_geneve_opt_mask;
2922 return rte_flow_error_set
2923 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2924 "Geneve TLV opt class/type/length must be specified");
2925 if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2926 return rte_flow_error_set
2927 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2928 "Geneve TLV opt length exceeeds the limit (31)");
2929 /* Check if class type and length masks are full. */
2930 if (full_mask.option_class != mask->option_class ||
2931 full_mask.option_type != mask->option_type ||
2932 full_mask.option_len != (mask->option_len & full_mask.option_len))
2933 return rte_flow_error_set
2934 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2935 "Geneve TLV opt class/type/length masks must be full");
2936 /* Check if length is supported */
2937 if ((uint32_t)spec->option_len >
2938 config->hca_attr.max_geneve_tlv_option_data_len)
2939 return rte_flow_error_set
2940 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2941 "Geneve TLV opt length not supported");
2942 if (config->hca_attr.max_geneve_tlv_options > 1)
2944 "max_geneve_tlv_options supports more than 1 option");
2945 /* Check GENEVE item preceding. */
2946 if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2947 return rte_flow_error_set
2948 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2949 "Geneve opt item must be preceded with Geneve item");
2950 geneve_spec = geneve_item->spec;
2951 geneve_mask = geneve_item->mask ? geneve_item->mask :
2952 &rte_flow_item_geneve_mask;
2953 /* Check if GENEVE TLV option size doesn't exceed option length */
2954 if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2955 geneve_spec->ver_opt_len_o_c_rsvd0)) {
2956 tlv_option_len = spec->option_len & mask->option_len;
2957 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2958 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2959 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2960 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2961 if ((optlen_v & optlen_m) <= tlv_option_len)
2962 return rte_flow_error_set
2963 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2964 "GENEVE TLV option length exceeds optlen");
2966 /* Check if length is 0 or data is 0. */
2967 if (spec->data == NULL || spec->option_len == 0)
2968 return rte_flow_error_set
2969 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2970 "Geneve TLV opt with zero data/length not supported");
2971 /* Check not all data & mask are 0. */
2972 data_len = spec->option_len * 4;
2973 if (mask->data == NULL) {
2974 for (i = 0; i < data_len; i++)
2978 return rte_flow_error_set(error, ENOTSUP,
2979 RTE_FLOW_ERROR_TYPE_ITEM, item,
2980 "Can't match on Geneve option data 0");
2982 for (i = 0; i < data_len; i++)
2983 if (spec->data[i] & mask->data[i])
2986 return rte_flow_error_set(error, ENOTSUP,
2987 RTE_FLOW_ERROR_TYPE_ITEM, item,
2988 "Can't match on Geneve option data and mask 0");
2989 /* Check data mask supported. */
2990 for (i = data_max_supported; i < data_len ; i++)
2992 return rte_flow_error_set(error, ENOTSUP,
2993 RTE_FLOW_ERROR_TYPE_ITEM, item,
2994 "Data mask is of unsupported size");
2996 /* Check GENEVE option is supported in NIC. */
2997 if (!config->hca_attr.geneve_tlv_opt)
2998 return rte_flow_error_set
2999 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3000 "Geneve TLV opt not supported");
3001 /* Check if we already have geneve option with different type/class. */
3002 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3003 geneve_opt_resource = sh->geneve_tlv_option_resource;
3004 if (geneve_opt_resource != NULL)
3005 if (geneve_opt_resource->option_class != spec->option_class ||
3006 geneve_opt_resource->option_type != spec->option_type ||
3007 geneve_opt_resource->length != spec->option_len) {
3008 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3009 return rte_flow_error_set(error, ENOTSUP,
3010 RTE_FLOW_ERROR_TYPE_ITEM, item,
3011 "Only one Geneve TLV option supported");
3013 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3018 * Validate MPLS item.
3021 * Pointer to the rte_eth_dev structure.
3023 * Item specification.
3024 * @param[in] item_flags
3025 * Bit-fields that holds the items detected until now.
3026 * @param[in] prev_layer
3027 * The protocol layer indicated in previous item.
3029 * Pointer to error structure.
3032 * 0 on success, a negative errno value otherwise and rte_errno is set.
3035 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3036 const struct rte_flow_item *item __rte_unused,
3037 uint64_t item_flags __rte_unused,
3038 uint64_t prev_layer __rte_unused,
3039 struct rte_flow_error *error)
3041 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3042 const struct rte_flow_item_mpls *mask = item->mask;
3043 struct mlx5_priv *priv = dev->data->dev_private;
3046 if (!priv->config.mpls_en)
3047 return rte_flow_error_set(error, ENOTSUP,
3048 RTE_FLOW_ERROR_TYPE_ITEM, item,
3049 "MPLS not supported or"
3050 " disabled in firmware"
3052 /* MPLS over UDP, GRE is allowed */
3053 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3054 MLX5_FLOW_LAYER_GRE |
3055 MLX5_FLOW_LAYER_GRE_KEY)))
3056 return rte_flow_error_set(error, EINVAL,
3057 RTE_FLOW_ERROR_TYPE_ITEM, item,
3058 "protocol filtering not compatible"
3059 " with MPLS layer");
3060 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3061 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3062 !(item_flags & MLX5_FLOW_LAYER_GRE))
3063 return rte_flow_error_set(error, ENOTSUP,
3064 RTE_FLOW_ERROR_TYPE_ITEM, item,
3065 "multiple tunnel layers not"
3068 mask = &rte_flow_item_mpls_mask;
3069 ret = mlx5_flow_item_acceptable
3070 (item, (const uint8_t *)mask,
3071 (const uint8_t *)&rte_flow_item_mpls_mask,
3072 sizeof(struct rte_flow_item_mpls),
3073 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3078 return rte_flow_error_set(error, ENOTSUP,
3079 RTE_FLOW_ERROR_TYPE_ITEM, item,
3080 "MPLS is not supported by Verbs, please"
3086 * Validate NVGRE item.
3089 * Item specification.
3090 * @param[in] item_flags
3091 * Bit flags to mark detected items.
3092 * @param[in] target_protocol
3093 * The next protocol in the previous item.
3095 * Pointer to error structure.
3098 * 0 on success, a negative errno value otherwise and rte_errno is set.
3101 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3102 uint64_t item_flags,
3103 uint8_t target_protocol,
3104 struct rte_flow_error *error)
3106 const struct rte_flow_item_nvgre *mask = item->mask;
3109 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3110 return rte_flow_error_set(error, EINVAL,
3111 RTE_FLOW_ERROR_TYPE_ITEM, item,
3112 "protocol filtering not compatible"
3113 " with this GRE layer");
3114 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3115 return rte_flow_error_set(error, ENOTSUP,
3116 RTE_FLOW_ERROR_TYPE_ITEM, item,
3117 "multiple tunnel layers not"
3119 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3120 return rte_flow_error_set(error, ENOTSUP,
3121 RTE_FLOW_ERROR_TYPE_ITEM, item,
3122 "L3 Layer is missing");
3124 mask = &rte_flow_item_nvgre_mask;
3125 ret = mlx5_flow_item_acceptable
3126 (item, (const uint8_t *)mask,
3127 (const uint8_t *)&rte_flow_item_nvgre_mask,
3128 sizeof(struct rte_flow_item_nvgre),
3129 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3136 * Validate eCPRI item.
3139 * Item specification.
3140 * @param[in] item_flags
3141 * Bit-fields that holds the items detected until now.
3142 * @param[in] last_item
3143 * Previous validated item in the pattern items.
3144 * @param[in] ether_type
3145 * Type in the ethernet layer header (including dot1q).
3146 * @param[in] acc_mask
3147 * Acceptable mask, if NULL default internal default mask
3148 * will be used to check whether item fields are supported.
3150 * Pointer to error structure.
3153 * 0 on success, a negative errno value otherwise and rte_errno is set.
3156 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3157 uint64_t item_flags,
3159 uint16_t ether_type,
3160 const struct rte_flow_item_ecpri *acc_mask,
3161 struct rte_flow_error *error)
3163 const struct rte_flow_item_ecpri *mask = item->mask;
3164 const struct rte_flow_item_ecpri nic_mask = {
3168 RTE_BE32(((const struct rte_ecpri_common_hdr) {
3172 .dummy[0] = 0xFFFFFFFF,
3175 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3176 MLX5_FLOW_LAYER_OUTER_VLAN);
3177 struct rte_flow_item_ecpri mask_lo;
3179 if (!(last_item & outer_l2_vlan) &&
3180 last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3181 return rte_flow_error_set(error, EINVAL,
3182 RTE_FLOW_ERROR_TYPE_ITEM, item,
3183 "eCPRI can only follow L2/VLAN layer or UDP layer");
3184 if ((last_item & outer_l2_vlan) && ether_type &&
3185 ether_type != RTE_ETHER_TYPE_ECPRI)
3186 return rte_flow_error_set(error, EINVAL,
3187 RTE_FLOW_ERROR_TYPE_ITEM, item,
3188 "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3189 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3190 return rte_flow_error_set(error, EINVAL,
3191 RTE_FLOW_ERROR_TYPE_ITEM, item,
3192 "eCPRI with tunnel is not supported right now");
3193 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3194 return rte_flow_error_set(error, ENOTSUP,
3195 RTE_FLOW_ERROR_TYPE_ITEM, item,
3196 "multiple L3 layers not supported");
3197 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3198 return rte_flow_error_set(error, EINVAL,
3199 RTE_FLOW_ERROR_TYPE_ITEM, item,
3200 "eCPRI cannot coexist with a TCP layer");
3201 /* In specification, eCPRI could be over UDP layer. */
3202 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3203 return rte_flow_error_set(error, EINVAL,
3204 RTE_FLOW_ERROR_TYPE_ITEM, item,
3205 "eCPRI over UDP layer is not yet supported right now");
3206 /* Mask for type field in common header could be zero. */
3208 mask = &rte_flow_item_ecpri_mask;
3209 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3210 /* Input mask is in big-endian format. */
3211 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3212 return rte_flow_error_set(error, EINVAL,
3213 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3214 "partial mask is not supported for protocol");
3215 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3216 return rte_flow_error_set(error, EINVAL,
3217 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3218 "message header mask must be after a type mask");
3219 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3220 acc_mask ? (const uint8_t *)acc_mask
3221 : (const uint8_t *)&nic_mask,
3222 sizeof(struct rte_flow_item_ecpri),
3223 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3227 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3228 const struct rte_flow_attr *attr __rte_unused,
3229 const struct rte_flow_item items[] __rte_unused,
3230 const struct rte_flow_action actions[] __rte_unused,
3231 bool external __rte_unused,
3232 int hairpin __rte_unused,
3233 struct rte_flow_error *error)
3235 return rte_flow_error_set(error, ENOTSUP,
3236 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3239 static struct mlx5_flow *
3240 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3241 const struct rte_flow_attr *attr __rte_unused,
3242 const struct rte_flow_item items[] __rte_unused,
3243 const struct rte_flow_action actions[] __rte_unused,
3244 struct rte_flow_error *error)
3246 rte_flow_error_set(error, ENOTSUP,
3247 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3252 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3253 struct mlx5_flow *dev_flow __rte_unused,
3254 const struct rte_flow_attr *attr __rte_unused,
3255 const struct rte_flow_item items[] __rte_unused,
3256 const struct rte_flow_action actions[] __rte_unused,
3257 struct rte_flow_error *error)
3259 return rte_flow_error_set(error, ENOTSUP,
3260 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3264 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3265 struct rte_flow *flow __rte_unused,
3266 struct rte_flow_error *error)
3268 return rte_flow_error_set(error, ENOTSUP,
3269 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3273 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3274 struct rte_flow *flow __rte_unused)
3279 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3280 struct rte_flow *flow __rte_unused)
3285 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3286 struct rte_flow *flow __rte_unused,
3287 const struct rte_flow_action *actions __rte_unused,
3288 void *data __rte_unused,
3289 struct rte_flow_error *error)
3291 return rte_flow_error_set(error, ENOTSUP,
3292 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3296 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3297 uint32_t domains __rte_unused,
3298 uint32_t flags __rte_unused)
3303 /* Void driver to protect from null pointer reference. */
3304 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3305 .validate = flow_null_validate,
3306 .prepare = flow_null_prepare,
3307 .translate = flow_null_translate,
3308 .apply = flow_null_apply,
3309 .remove = flow_null_remove,
3310 .destroy = flow_null_destroy,
3311 .query = flow_null_query,
3312 .sync_domain = flow_null_sync_domain,
3316 * Select flow driver type according to flow attributes and device
3320 * Pointer to the dev structure.
3322 * Pointer to the flow attributes.
3325 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3327 static enum mlx5_flow_drv_type
3328 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3330 struct mlx5_priv *priv = dev->data->dev_private;
3331 /* The OS can determine first a specific flow type (DV, VERBS) */
3332 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3334 if (type != MLX5_FLOW_TYPE_MAX)
3336 /* If no OS specific type - continue with DV/VERBS selection */
3337 if (attr->transfer && priv->config.dv_esw_en)
3338 type = MLX5_FLOW_TYPE_DV;
3339 if (!attr->transfer)
3340 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3341 MLX5_FLOW_TYPE_VERBS;
3345 #define flow_get_drv_ops(type) flow_drv_ops[type]
3348 * Flow driver validation API. This abstracts calling driver specific functions.
3349 * The type of flow driver is determined according to flow attributes.
3352 * Pointer to the dev structure.
3354 * Pointer to the flow attributes.
3356 * Pointer to the list of items.
3357 * @param[in] actions
3358 * Pointer to the list of actions.
3359 * @param[in] external
3360 * This flow rule is created by request external to PMD.
3361 * @param[in] hairpin
3362 * Number of hairpin TX actions, 0 means classic flow.
3364 * Pointer to the error structure.
3367 * 0 on success, a negative errno value otherwise and rte_errno is set.
3370 flow_drv_validate(struct rte_eth_dev *dev,
3371 const struct rte_flow_attr *attr,
3372 const struct rte_flow_item items[],
3373 const struct rte_flow_action actions[],
3374 bool external, int hairpin, struct rte_flow_error *error)
3376 const struct mlx5_flow_driver_ops *fops;
3377 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3379 fops = flow_get_drv_ops(type);
3380 return fops->validate(dev, attr, items, actions, external,
3385 * Flow driver preparation API. This abstracts calling driver specific
3386 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3387 * calculates the size of memory required for device flow, allocates the memory,
3388 * initializes the device flow and returns the pointer.
3391 * This function initializes device flow structure such as dv or verbs in
3392 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3393 * rest. For example, adding returning device flow to flow->dev_flow list and
3394 * setting backward reference to the flow should be done out of this function.
3395 * layers field is not filled either.
3398 * Pointer to the dev structure.
3400 * Pointer to the flow attributes.
3402 * Pointer to the list of items.
3403 * @param[in] actions
3404 * Pointer to the list of actions.
3405 * @param[in] flow_idx
3406 * This memory pool index to the flow.
3408 * Pointer to the error structure.
3411 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3413 static inline struct mlx5_flow *
3414 flow_drv_prepare(struct rte_eth_dev *dev,
3415 const struct rte_flow *flow,
3416 const struct rte_flow_attr *attr,
3417 const struct rte_flow_item items[],
3418 const struct rte_flow_action actions[],
3420 struct rte_flow_error *error)
3422 const struct mlx5_flow_driver_ops *fops;
3423 enum mlx5_flow_drv_type type = flow->drv_type;
3424 struct mlx5_flow *mlx5_flow = NULL;
3426 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3427 fops = flow_get_drv_ops(type);
3428 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3430 mlx5_flow->flow_idx = flow_idx;
3435 * Flow driver translation API. This abstracts calling driver specific
3436 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3437 * translates a generic flow into a driver flow. flow_drv_prepare() must
3441 * dev_flow->layers could be filled as a result of parsing during translation
3442 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3443 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3444 * flow->actions could be overwritten even though all the expanded dev_flows
3445 * have the same actions.
3448 * Pointer to the rte dev structure.
3449 * @param[in, out] dev_flow
3450 * Pointer to the mlx5 flow.
3452 * Pointer to the flow attributes.
3454 * Pointer to the list of items.
3455 * @param[in] actions
3456 * Pointer to the list of actions.
3458 * Pointer to the error structure.
3461 * 0 on success, a negative errno value otherwise and rte_errno is set.
3464 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3465 const struct rte_flow_attr *attr,
3466 const struct rte_flow_item items[],
3467 const struct rte_flow_action actions[],
3468 struct rte_flow_error *error)
3470 const struct mlx5_flow_driver_ops *fops;
3471 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3473 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3474 fops = flow_get_drv_ops(type);
3475 return fops->translate(dev, dev_flow, attr, items, actions, error);
3479 * Flow driver apply API. This abstracts calling driver specific functions.
3480 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3481 * translated driver flows on to device. flow_drv_translate() must precede.
3484 * Pointer to Ethernet device structure.
3485 * @param[in, out] flow
3486 * Pointer to flow structure.
3488 * Pointer to error structure.
3491 * 0 on success, a negative errno value otherwise and rte_errno is set.
3494 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3495 struct rte_flow_error *error)
3497 const struct mlx5_flow_driver_ops *fops;
3498 enum mlx5_flow_drv_type type = flow->drv_type;
3500 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3501 fops = flow_get_drv_ops(type);
3502 return fops->apply(dev, flow, error);
3506 * Flow driver destroy API. This abstracts calling driver specific functions.
3507 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3508 * on device and releases resources of the flow.
3511 * Pointer to Ethernet device.
3512 * @param[in, out] flow
3513 * Pointer to flow structure.
3516 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3518 const struct mlx5_flow_driver_ops *fops;
3519 enum mlx5_flow_drv_type type = flow->drv_type;
3521 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3522 fops = flow_get_drv_ops(type);
3523 fops->destroy(dev, flow);
3527 * Flow driver find RSS policy tbl API. This abstracts calling driver
3528 * specific functions. Parent flow (rte_flow) should have driver
3529 * type (drv_type). It will find the RSS policy table that has the rss_desc.
3532 * Pointer to Ethernet device.
3533 * @param[in, out] flow
3534 * Pointer to flow structure.
3536 * Pointer to meter policy table.
3537 * @param[in] rss_desc
3538 * Pointer to rss_desc
3540 static struct mlx5_flow_meter_sub_policy *
3541 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3542 struct rte_flow *flow,
3543 struct mlx5_flow_meter_policy *policy,
3544 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3546 const struct mlx5_flow_driver_ops *fops;
3547 enum mlx5_flow_drv_type type = flow->drv_type;
3549 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3550 fops = flow_get_drv_ops(type);
3551 return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3555 * Flow driver color tag rule API. This abstracts calling driver
3556 * specific functions. Parent flow (rte_flow) should have driver
3557 * type (drv_type). It will create the color tag rules in hierarchy meter.
3560 * Pointer to Ethernet device.
3561 * @param[in, out] flow
3562 * Pointer to flow structure.
3564 * Pointer to flow meter structure.
3565 * @param[in] src_port
3566 * The src port this extra rule should use.
3568 * The src port id match item.
3570 * Pointer to error structure.
3573 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3574 struct rte_flow *flow,
3575 struct mlx5_flow_meter_info *fm,
3577 const struct rte_flow_item *item,
3578 struct rte_flow_error *error)
3580 const struct mlx5_flow_driver_ops *fops;
3581 enum mlx5_flow_drv_type type = flow->drv_type;
3583 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3584 fops = flow_get_drv_ops(type);
3585 return fops->meter_hierarchy_rule_create(dev, fm,
3586 src_port, item, error);
3590 * Get RSS action from the action list.
3593 * Pointer to Ethernet device.
3594 * @param[in] actions
3595 * Pointer to the list of actions.
3597 * Parent flow structure pointer.
3600 * Pointer to the RSS action if exist, else return NULL.
3602 static const struct rte_flow_action_rss*
3603 flow_get_rss_action(struct rte_eth_dev *dev,
3604 const struct rte_flow_action actions[])
3606 struct mlx5_priv *priv = dev->data->dev_private;
3607 const struct rte_flow_action_rss *rss = NULL;
3608 struct mlx5_meter_policy_action_container *acg;
3609 struct mlx5_meter_policy_action_container *acy;
3611 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3612 switch (actions->type) {
3613 case RTE_FLOW_ACTION_TYPE_RSS:
3614 rss = actions->conf;
3616 case RTE_FLOW_ACTION_TYPE_SAMPLE:
3618 const struct rte_flow_action_sample *sample =
3620 const struct rte_flow_action *act = sample->actions;
3621 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3622 if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3626 case RTE_FLOW_ACTION_TYPE_METER:
3629 struct mlx5_flow_meter_info *fm;
3630 struct mlx5_flow_meter_policy *policy;
3631 const struct rte_flow_action_meter *mtr = actions->conf;
3633 fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3634 if (fm && !fm->def_policy) {
3635 policy = mlx5_flow_meter_policy_find(dev,
3636 fm->policy_id, NULL);
3637 MLX5_ASSERT(policy);
3638 if (policy->is_hierarchy) {
3640 mlx5_flow_meter_hierarchy_get_final_policy(dev,
3645 if (policy->is_rss) {
3647 &policy->act_cnt[RTE_COLOR_GREEN];
3649 &policy->act_cnt[RTE_COLOR_YELLOW];
3650 if (acg->fate_action ==
3651 MLX5_FLOW_FATE_SHARED_RSS)
3652 rss = acg->rss->conf;
3653 else if (acy->fate_action ==
3654 MLX5_FLOW_FATE_SHARED_RSS)
3655 rss = acy->rss->conf;
3668 * Get ASO age action by index.
3671 * Pointer to the Ethernet device structure.
3672 * @param[in] age_idx
3673 * Index to the ASO age action.
3676 * The specified ASO age action.
3678 struct mlx5_aso_age_action*
3679 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3681 uint16_t pool_idx = age_idx & UINT16_MAX;
3682 uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3683 struct mlx5_priv *priv = dev->data->dev_private;
3684 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3685 struct mlx5_aso_age_pool *pool;
3687 rte_rwlock_read_lock(&mng->resize_rwl);
3688 pool = mng->pools[pool_idx];
3689 rte_rwlock_read_unlock(&mng->resize_rwl);
3690 return &pool->actions[offset - 1];
3693 /* maps indirect action to translated direct in some actions array */
3694 struct mlx5_translated_action_handle {
3695 struct rte_flow_action_handle *action; /**< Indirect action handle. */
3696 int index; /**< Index in related array of rte_flow_action. */
3700 * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3701 * direct action if translation possible.
3702 * This functionality used to run same execution path for both direct and
3703 * indirect actions on flow create. All necessary preparations for indirect
3704 * action handling should be performed on *handle* actions list returned
3708 * Pointer to Ethernet device.
3709 * @param[in] actions
3710 * List of actions to translate.
3711 * @param[out] handle
3712 * List to store translated indirect action object handles.
3713 * @param[in, out] indir_n
3714 * Size of *handle* array. On return should be updated with number of
3715 * indirect actions retrieved from the *actions* list.
3716 * @param[out] translated_actions
3717 * List of actions where all indirect actions were translated to direct
3718 * if possible. NULL if no translation took place.
3720 * Pointer to the error structure.
3723 * 0 on success, a negative errno value otherwise and rte_errno is set.
3726 flow_action_handles_translate(struct rte_eth_dev *dev,
3727 const struct rte_flow_action actions[],
3728 struct mlx5_translated_action_handle *handle,
3730 struct rte_flow_action **translated_actions,
3731 struct rte_flow_error *error)
3733 struct mlx5_priv *priv = dev->data->dev_private;
3734 struct rte_flow_action *translated = NULL;
3735 size_t actions_size;
3738 struct mlx5_translated_action_handle *handle_end = NULL;
3740 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3741 if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3743 if (copied_n == *indir_n) {
3744 return rte_flow_error_set
3745 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3746 NULL, "too many shared actions");
3748 rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3749 sizeof(actions[n].conf));
3750 handle[copied_n].index = n;
3754 *indir_n = copied_n;
3757 actions_size = sizeof(struct rte_flow_action) * n;
3758 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3763 memcpy(translated, actions, actions_size);
3764 for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3765 struct mlx5_shared_action_rss *shared_rss;
3766 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3767 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3768 uint32_t idx = act_idx &
3769 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3772 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3773 shared_rss = mlx5_ipool_get
3774 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3775 translated[handle->index].type =
3776 RTE_FLOW_ACTION_TYPE_RSS;
3777 translated[handle->index].conf =
3778 &shared_rss->origin;
3780 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3781 translated[handle->index].type =
3782 (enum rte_flow_action_type)
3783 MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3784 translated[handle->index].conf = (void *)(uintptr_t)idx;
3786 case MLX5_INDIRECT_ACTION_TYPE_AGE:
3787 if (priv->sh->flow_hit_aso_en) {
3788 translated[handle->index].type =
3789 (enum rte_flow_action_type)
3790 MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3791 translated[handle->index].conf =
3792 (void *)(uintptr_t)idx;
3796 case MLX5_INDIRECT_ACTION_TYPE_CT:
3797 if (priv->sh->ct_aso_en) {
3798 translated[handle->index].type =
3799 RTE_FLOW_ACTION_TYPE_CONNTRACK;
3800 translated[handle->index].conf =
3801 (void *)(uintptr_t)idx;
3806 mlx5_free(translated);
3807 return rte_flow_error_set
3808 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3809 NULL, "invalid indirect action type");
3812 *translated_actions = translated;
3817 * Get Shared RSS action from the action list.
3820 * Pointer to Ethernet device.
3822 * Pointer to the list of actions.
3823 * @param[in] shared_n
3824 * Actions list length.
3827 * The MLX5 RSS action ID if exists, otherwise return 0.
3830 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3831 struct mlx5_translated_action_handle *handle,
3834 struct mlx5_translated_action_handle *handle_end;
3835 struct mlx5_priv *priv = dev->data->dev_private;
3836 struct mlx5_shared_action_rss *shared_rss;
3839 for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3840 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3841 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3842 uint32_t idx = act_idx &
3843 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3845 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3846 shared_rss = mlx5_ipool_get
3847 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3849 __atomic_add_fetch(&shared_rss->refcnt, 1,
3860 find_graph_root(uint32_t rss_level)
3862 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3863 MLX5_EXPANSION_ROOT_OUTER;
3867 * Get layer flags from the prefix flow.
3869 * Some flows may be split to several subflows, the prefix subflow gets the
3870 * match items and the suffix sub flow gets the actions.
3871 * Some actions need the user defined match item flags to get the detail for
3873 * This function helps the suffix flow to get the item layer flags from prefix
3876 * @param[in] dev_flow
3877 * Pointer the created preifx subflow.
3880 * The layers get from prefix subflow.
3882 static inline uint64_t
3883 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3885 uint64_t layers = 0;
3888 * Layers bits could be localization, but usually the compiler will
3889 * help to do the optimization work for source code.
3890 * If no decap actions, use the layers directly.
3892 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3893 return dev_flow->handle->layers;
3894 /* Convert L3 layers with decap action. */
3895 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3896 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3897 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3898 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3899 /* Convert L4 layers with decap action. */
3900 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3901 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3902 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3903 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3908 * Get metadata split action information.
3910 * @param[in] actions
3911 * Pointer to the list of actions.
3913 * Pointer to the return pointer.
3914 * @param[out] qrss_type
3915 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3916 * if no QUEUE/RSS is found.
3917 * @param[out] encap_idx
3918 * Pointer to the index of the encap action if exists, otherwise the last
3922 * Total number of actions.
3925 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3926 const struct rte_flow_action **qrss,
3929 const struct rte_flow_action_raw_encap *raw_encap;
3931 int raw_decap_idx = -1;
3934 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3935 switch (actions->type) {
3936 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3937 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3938 *encap_idx = actions_n;
3940 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3941 raw_decap_idx = actions_n;
3943 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3944 raw_encap = actions->conf;
3945 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3946 *encap_idx = raw_decap_idx != -1 ?
3947 raw_decap_idx : actions_n;
3949 case RTE_FLOW_ACTION_TYPE_QUEUE:
3950 case RTE_FLOW_ACTION_TYPE_RSS:
3958 if (*encap_idx == -1)
3959 *encap_idx = actions_n;
3960 /* Count RTE_FLOW_ACTION_TYPE_END. */
3961 return actions_n + 1;
3965 * Check if the action will change packet.
3968 * Pointer to Ethernet device.
3973 * true if action will change packet, false otherwise.
3975 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3976 enum rte_flow_action_type type)
3978 struct mlx5_priv *priv = dev->data->dev_private;
3981 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3982 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3983 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3984 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3985 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3986 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3987 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3988 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3989 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3990 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3991 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3992 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3993 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3994 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3995 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3996 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3997 case RTE_FLOW_ACTION_TYPE_SET_META:
3998 case RTE_FLOW_ACTION_TYPE_SET_TAG:
3999 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4000 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4001 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4002 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4003 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4004 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4005 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4006 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4007 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4008 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4009 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4011 case RTE_FLOW_ACTION_TYPE_FLAG:
4012 case RTE_FLOW_ACTION_TYPE_MARK:
4013 if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4023 * Check meter action from the action list.
4026 * Pointer to Ethernet device.
4027 * @param[in] actions
4028 * Pointer to the list of actions.
4029 * @param[out] has_mtr
4030 * Pointer to the meter exist flag.
4031 * @param[out] has_modify
4032 * Pointer to the flag showing there's packet change action.
4033 * @param[out] meter_id
4034 * Pointer to the meter id.
4037 * Total number of actions.
4040 flow_check_meter_action(struct rte_eth_dev *dev,
4041 const struct rte_flow_action actions[],
4042 bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4044 const struct rte_flow_action_meter *mtr = NULL;
4047 MLX5_ASSERT(has_mtr);
4049 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4050 switch (actions->type) {
4051 case RTE_FLOW_ACTION_TYPE_METER:
4052 mtr = actions->conf;
4053 *meter_id = mtr->mtr_id;
4060 *has_modify |= flow_check_modify_action_type(dev,
4064 /* Count RTE_FLOW_ACTION_TYPE_END. */
4065 return actions_n + 1;
4069 * Check if the flow should be split due to hairpin.
4070 * The reason for the split is that in current HW we can't
4071 * support encap and push-vlan on Rx, so if a flow contains
4072 * these actions we move it to Tx.
4075 * Pointer to Ethernet device.
4077 * Flow rule attributes.
4078 * @param[in] actions
4079 * Associated actions (list terminated by the END action).
4082 * > 0 the number of actions and the flow should be split,
4083 * 0 when no split required.
4086 flow_check_hairpin_split(struct rte_eth_dev *dev,
4087 const struct rte_flow_attr *attr,
4088 const struct rte_flow_action actions[])
4090 int queue_action = 0;
4093 const struct rte_flow_action_queue *queue;
4094 const struct rte_flow_action_rss *rss;
4095 const struct rte_flow_action_raw_encap *raw_encap;
4096 const struct rte_eth_hairpin_conf *conf;
4100 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4101 switch (actions->type) {
4102 case RTE_FLOW_ACTION_TYPE_QUEUE:
4103 queue = actions->conf;
4106 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4107 if (conf == NULL || conf->tx_explicit != 0)
4112 case RTE_FLOW_ACTION_TYPE_RSS:
4113 rss = actions->conf;
4114 if (rss == NULL || rss->queue_num == 0)
4116 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4117 if (conf == NULL || conf->tx_explicit != 0)
4122 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4123 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4124 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4125 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4126 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4130 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4131 raw_encap = actions->conf;
4132 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4141 if (split && queue_action)
4146 /* Declare flow create/destroy prototype in advance. */
4148 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4149 const struct rte_flow_attr *attr,
4150 const struct rte_flow_item items[],
4151 const struct rte_flow_action actions[],
4152 bool external, struct rte_flow_error *error);
4155 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4159 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4160 struct mlx5_list_entry *entry, void *cb_ctx)
4162 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4163 struct mlx5_flow_mreg_copy_resource *mcp_res =
4164 container_of(entry, typeof(*mcp_res), hlist_ent);
4166 return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4169 struct mlx5_list_entry *
4170 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4172 struct rte_eth_dev *dev = tool_ctx;
4173 struct mlx5_priv *priv = dev->data->dev_private;
4174 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4175 struct mlx5_flow_mreg_copy_resource *mcp_res;
4176 struct rte_flow_error *error = ctx->error;
4179 uint32_t mark_id = *(uint32_t *)(ctx->data);
4180 struct rte_flow_attr attr = {
4181 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4184 struct mlx5_rte_flow_item_tag tag_spec = {
4187 struct rte_flow_item items[] = {
4188 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4190 struct rte_flow_action_mark ftag = {
4193 struct mlx5_flow_action_copy_mreg cp_mreg = {
4197 struct rte_flow_action_jump jump = {
4198 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4200 struct rte_flow_action actions[] = {
4201 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4204 /* Fill the register fileds in the flow. */
4205 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4209 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4213 /* Provide the full width of FLAG specific value. */
4214 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4215 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4216 /* Build a new flow. */
4217 if (mark_id != MLX5_DEFAULT_COPY_ID) {
4218 items[0] = (struct rte_flow_item){
4219 .type = (enum rte_flow_item_type)
4220 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4223 items[1] = (struct rte_flow_item){
4224 .type = RTE_FLOW_ITEM_TYPE_END,
4226 actions[0] = (struct rte_flow_action){
4227 .type = (enum rte_flow_action_type)
4228 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4231 actions[1] = (struct rte_flow_action){
4232 .type = (enum rte_flow_action_type)
4233 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4236 actions[2] = (struct rte_flow_action){
4237 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4240 actions[3] = (struct rte_flow_action){
4241 .type = RTE_FLOW_ACTION_TYPE_END,
4244 /* Default rule, wildcard match. */
4245 attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4246 items[0] = (struct rte_flow_item){
4247 .type = RTE_FLOW_ITEM_TYPE_END,
4249 actions[0] = (struct rte_flow_action){
4250 .type = (enum rte_flow_action_type)
4251 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4254 actions[1] = (struct rte_flow_action){
4255 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4258 actions[2] = (struct rte_flow_action){
4259 .type = RTE_FLOW_ACTION_TYPE_END,
4262 /* Build a new entry. */
4263 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4269 mcp_res->mark_id = mark_id;
4271 * The copy Flows are not included in any list. There
4272 * ones are referenced from other Flows and can not
4273 * be applied, removed, deleted in ardbitrary order
4274 * by list traversing.
4276 mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4277 &attr, items, actions, false, error);
4278 if (!mcp_res->rix_flow) {
4279 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4282 return &mcp_res->hlist_ent;
4285 struct mlx5_list_entry *
4286 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4287 void *cb_ctx __rte_unused)
4289 struct rte_eth_dev *dev = tool_ctx;
4290 struct mlx5_priv *priv = dev->data->dev_private;
4291 struct mlx5_flow_mreg_copy_resource *mcp_res;
4294 mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4299 memcpy(mcp_res, oentry, sizeof(*mcp_res));
4301 return &mcp_res->hlist_ent;
4305 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4307 struct mlx5_flow_mreg_copy_resource *mcp_res =
4308 container_of(entry, typeof(*mcp_res), hlist_ent);
4309 struct rte_eth_dev *dev = tool_ctx;
4310 struct mlx5_priv *priv = dev->data->dev_private;
4312 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4316 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4318 * As mark_id is unique, if there's already a registered flow for the mark_id,
4319 * return by increasing the reference counter of the resource. Otherwise, create
4320 * the resource (mcp_res) and flow.
4323 * - If ingress port is ANY and reg_c[1] is mark_id,
4324 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4326 * For default flow (zero mark_id), flow is like,
4327 * - If ingress port is ANY,
4328 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
4331 * Pointer to Ethernet device.
4333 * ID of MARK action, zero means default flow for META.
4335 * Perform verbose error reporting if not NULL.
4338 * Associated resource on success, NULL otherwise and rte_errno is set.
4340 static struct mlx5_flow_mreg_copy_resource *
4341 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4342 struct rte_flow_error *error)
4344 struct mlx5_priv *priv = dev->data->dev_private;
4345 struct mlx5_list_entry *entry;
4346 struct mlx5_flow_cb_ctx ctx = {
4352 /* Check if already registered. */
4353 MLX5_ASSERT(priv->mreg_cp_tbl);
4354 entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4357 return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4362 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4364 struct mlx5_flow_mreg_copy_resource *mcp_res =
4365 container_of(entry, typeof(*mcp_res), hlist_ent);
4366 struct rte_eth_dev *dev = tool_ctx;
4367 struct mlx5_priv *priv = dev->data->dev_private;
4369 MLX5_ASSERT(mcp_res->rix_flow);
4370 flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4371 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4375 * Release flow in RX_CP_TBL.
4378 * Pointer to Ethernet device.
4380 * Parent flow for wich copying is provided.
4383 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4384 struct rte_flow *flow)
4386 struct mlx5_flow_mreg_copy_resource *mcp_res;
4387 struct mlx5_priv *priv = dev->data->dev_private;
4389 if (!flow->rix_mreg_copy)
4391 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4392 flow->rix_mreg_copy);
4393 if (!mcp_res || !priv->mreg_cp_tbl)
4395 MLX5_ASSERT(mcp_res->rix_flow);
4396 mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4397 flow->rix_mreg_copy = 0;
4401 * Remove the default copy action from RX_CP_TBL.
4403 * This functions is called in the mlx5_dev_start(). No thread safe
4407 * Pointer to Ethernet device.
4410 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4412 struct mlx5_list_entry *entry;
4413 struct mlx5_priv *priv = dev->data->dev_private;
4414 struct mlx5_flow_cb_ctx ctx;
4417 /* Check if default flow is registered. */
4418 if (!priv->mreg_cp_tbl)
4420 mark_id = MLX5_DEFAULT_COPY_ID;
4421 ctx.data = &mark_id;
4422 entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4425 mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4429 * Add the default copy action in in RX_CP_TBL.
4431 * This functions is called in the mlx5_dev_start(). No thread safe
4435 * Pointer to Ethernet device.
4437 * Perform verbose error reporting if not NULL.
4440 * 0 for success, negative value otherwise and rte_errno is set.
4443 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4444 struct rte_flow_error *error)
4446 struct mlx5_priv *priv = dev->data->dev_private;
4447 struct mlx5_flow_mreg_copy_resource *mcp_res;
4448 struct mlx5_flow_cb_ctx ctx;
4451 /* Check whether extensive metadata feature is engaged. */
4452 if (!priv->config.dv_flow_en ||
4453 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4454 !mlx5_flow_ext_mreg_supported(dev) ||
4455 !priv->sh->dv_regc0_mask)
4458 * Add default mreg copy flow may be called multiple time, but
4459 * only be called once in stop. Avoid register it twice.
4461 mark_id = MLX5_DEFAULT_COPY_ID;
4462 ctx.data = &mark_id;
4463 if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4465 mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4472 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4474 * All the flow having Q/RSS action should be split by
4475 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4476 * performs the following,
4477 * - CQE->flow_tag := reg_c[1] (MARK)
4478 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4479 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4480 * but there should be a flow per each MARK ID set by MARK action.
4482 * For the aforementioned reason, if there's a MARK action in flow's action
4483 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4484 * the MARK ID to CQE's flow_tag like,
4485 * - If reg_c[1] is mark_id,
4486 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4488 * For SET_META action which stores value in reg_c[0], as the destination is
4489 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4490 * MARK ID means the default flow. The default flow looks like,
4491 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4494 * Pointer to Ethernet device.
4496 * Pointer to flow structure.
4497 * @param[in] actions
4498 * Pointer to the list of actions.
4500 * Perform verbose error reporting if not NULL.
4503 * 0 on success, negative value otherwise and rte_errno is set.
4506 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4507 struct rte_flow *flow,
4508 const struct rte_flow_action *actions,
4509 struct rte_flow_error *error)
4511 struct mlx5_priv *priv = dev->data->dev_private;
4512 struct mlx5_dev_config *config = &priv->config;
4513 struct mlx5_flow_mreg_copy_resource *mcp_res;
4514 const struct rte_flow_action_mark *mark;
4516 /* Check whether extensive metadata feature is engaged. */
4517 if (!config->dv_flow_en ||
4518 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4519 !mlx5_flow_ext_mreg_supported(dev) ||
4520 !priv->sh->dv_regc0_mask)
4522 /* Find MARK action. */
4523 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4524 switch (actions->type) {
4525 case RTE_FLOW_ACTION_TYPE_FLAG:
4526 mcp_res = flow_mreg_add_copy_action
4527 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4530 flow->rix_mreg_copy = mcp_res->idx;
4532 case RTE_FLOW_ACTION_TYPE_MARK:
4533 mark = (const struct rte_flow_action_mark *)
4536 flow_mreg_add_copy_action(dev, mark->id, error);
4539 flow->rix_mreg_copy = mcp_res->idx;
4548 #define MLX5_MAX_SPLIT_ACTIONS 24
4549 #define MLX5_MAX_SPLIT_ITEMS 24
4552 * Split the hairpin flow.
4553 * Since HW can't support encap and push-vlan on Rx, we move these
4555 * If the count action is after the encap then we also
4556 * move the count action. in this case the count will also measure
4560 * Pointer to Ethernet device.
4561 * @param[in] actions
4562 * Associated actions (list terminated by the END action).
4563 * @param[out] actions_rx
4565 * @param[out] actions_tx
4567 * @param[out] pattern_tx
4568 * The pattern items for the Tx flow.
4569 * @param[out] flow_id
4570 * The flow ID connected to this flow.
4576 flow_hairpin_split(struct rte_eth_dev *dev,
4577 const struct rte_flow_action actions[],
4578 struct rte_flow_action actions_rx[],
4579 struct rte_flow_action actions_tx[],
4580 struct rte_flow_item pattern_tx[],
4583 const struct rte_flow_action_raw_encap *raw_encap;
4584 const struct rte_flow_action_raw_decap *raw_decap;
4585 struct mlx5_rte_flow_action_set_tag *set_tag;
4586 struct rte_flow_action *tag_action;
4587 struct mlx5_rte_flow_item_tag *tag_item;
4588 struct rte_flow_item *item;
4592 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4593 switch (actions->type) {
4594 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4595 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4596 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4597 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4598 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4599 rte_memcpy(actions_tx, actions,
4600 sizeof(struct rte_flow_action));
4603 case RTE_FLOW_ACTION_TYPE_COUNT:
4605 rte_memcpy(actions_tx, actions,
4606 sizeof(struct rte_flow_action));
4609 rte_memcpy(actions_rx, actions,
4610 sizeof(struct rte_flow_action));
4614 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4615 raw_encap = actions->conf;
4616 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4617 memcpy(actions_tx, actions,
4618 sizeof(struct rte_flow_action));
4622 rte_memcpy(actions_rx, actions,
4623 sizeof(struct rte_flow_action));
4627 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4628 raw_decap = actions->conf;
4629 if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4630 memcpy(actions_tx, actions,
4631 sizeof(struct rte_flow_action));
4634 rte_memcpy(actions_rx, actions,
4635 sizeof(struct rte_flow_action));
4640 rte_memcpy(actions_rx, actions,
4641 sizeof(struct rte_flow_action));
4646 /* Add set meta action and end action for the Rx flow. */
4647 tag_action = actions_rx;
4648 tag_action->type = (enum rte_flow_action_type)
4649 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4651 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4653 set_tag = (void *)actions_rx;
4654 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4655 .id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4658 MLX5_ASSERT(set_tag->id > REG_NON);
4659 tag_action->conf = set_tag;
4660 /* Create Tx item list. */
4661 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4662 addr = (void *)&pattern_tx[2];
4664 item->type = (enum rte_flow_item_type)
4665 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4666 tag_item = (void *)addr;
4667 tag_item->data = flow_id;
4668 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4669 MLX5_ASSERT(set_tag->id > REG_NON);
4670 item->spec = tag_item;
4671 addr += sizeof(struct mlx5_rte_flow_item_tag);
4672 tag_item = (void *)addr;
4673 tag_item->data = UINT32_MAX;
4674 tag_item->id = UINT16_MAX;
4675 item->mask = tag_item;
4678 item->type = RTE_FLOW_ITEM_TYPE_END;
4683 * The last stage of splitting chain, just creates the subflow
4684 * without any modification.
4687 * Pointer to Ethernet device.
4689 * Parent flow structure pointer.
4690 * @param[in, out] sub_flow
4691 * Pointer to return the created subflow, may be NULL.
4693 * Flow rule attributes.
4695 * Pattern specification (list terminated by the END pattern item).
4696 * @param[in] actions
4697 * Associated actions (list terminated by the END action).
4698 * @param[in] flow_split_info
4699 * Pointer to flow split info structure.
4701 * Perform verbose error reporting if not NULL.
4703 * 0 on success, negative value otherwise
4706 flow_create_split_inner(struct rte_eth_dev *dev,
4707 struct rte_flow *flow,
4708 struct mlx5_flow **sub_flow,
4709 const struct rte_flow_attr *attr,
4710 const struct rte_flow_item items[],
4711 const struct rte_flow_action actions[],
4712 struct mlx5_flow_split_info *flow_split_info,
4713 struct rte_flow_error *error)
4715 struct mlx5_flow *dev_flow;
4717 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4718 flow_split_info->flow_idx, error);
4721 dev_flow->flow = flow;
4722 dev_flow->external = flow_split_info->external;
4723 dev_flow->skip_scale = flow_split_info->skip_scale;
4724 /* Subflow object was created, we must include one in the list. */
4725 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4726 dev_flow->handle, next);
4728 * If dev_flow is as one of the suffix flow, some actions in suffix
4729 * flow may need some user defined item layer flags, and pass the
4730 * Metadate rxq mark flag to suffix flow as well.
4732 if (flow_split_info->prefix_layers)
4733 dev_flow->handle->layers = flow_split_info->prefix_layers;
4734 if (flow_split_info->prefix_mark)
4735 dev_flow->handle->mark = 1;
4737 *sub_flow = dev_flow;
4738 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4739 dev_flow->dv.table_id = flow_split_info->table_id;
4741 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4745 * Get the sub policy of a meter.
4748 * Pointer to Ethernet device.
4750 * Parent flow structure pointer.
4752 * Pointer to thread flow work space.
4754 * Flow rule attributes.
4756 * Pattern specification (list terminated by the END pattern item).
4758 * Perform verbose error reporting if not NULL.
4761 * Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4763 static struct mlx5_flow_meter_sub_policy *
4764 get_meter_sub_policy(struct rte_eth_dev *dev,
4765 struct rte_flow *flow,
4766 struct mlx5_flow_workspace *wks,
4767 const struct rte_flow_attr *attr,
4768 const struct rte_flow_item items[],
4769 struct rte_flow_error *error)
4771 struct mlx5_flow_meter_policy *policy;
4772 struct mlx5_flow_meter_policy *final_policy;
4773 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4775 policy = wks->policy;
4776 final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4777 if (final_policy->is_rss || final_policy->is_queue) {
4778 struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4779 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4783 * This is a tmp dev_flow,
4784 * no need to register any matcher for it in translate.
4786 wks->skip_matcher_reg = 1;
4787 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4788 struct mlx5_flow dev_flow = {0};
4789 struct mlx5_flow_handle dev_handle = { {0} };
4790 uint8_t fate = final_policy->act_cnt[i].fate_action;
4792 if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4793 const struct rte_flow_action_rss *rss_act =
4794 final_policy->act_cnt[i].rss->conf;
4795 struct rte_flow_action rss_actions[2] = {
4797 .type = RTE_FLOW_ACTION_TYPE_RSS,
4801 .type = RTE_FLOW_ACTION_TYPE_END,
4806 dev_flow.handle = &dev_handle;
4807 dev_flow.ingress = attr->ingress;
4808 dev_flow.flow = flow;
4809 dev_flow.external = 0;
4810 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4811 dev_flow.dv.transfer = attr->transfer;
4814 * Translate RSS action to get rss hash fields.
4816 if (flow_drv_translate(dev, &dev_flow, attr,
4817 items, rss_actions, error))
4819 rss_desc_v[i] = wks->rss_desc;
4820 rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4821 rss_desc_v[i].hash_fields =
4822 dev_flow.hash_fields;
4823 rss_desc_v[i].queue_num =
4824 rss_desc_v[i].hash_fields ?
4825 rss_desc_v[i].queue_num : 1;
4826 rss_desc_v[i].tunnel =
4827 !!(dev_flow.handle->layers &
4828 MLX5_FLOW_LAYER_TUNNEL);
4829 /* Use the RSS queues in the containers. */
4830 rss_desc_v[i].queue =
4831 (uint16_t *)(uintptr_t)rss_act->queue;
4832 rss_desc[i] = &rss_desc_v[i];
4833 } else if (fate == MLX5_FLOW_FATE_QUEUE) {
4834 /* This is queue action. */
4835 rss_desc_v[i] = wks->rss_desc;
4836 rss_desc_v[i].key_len = 0;
4837 rss_desc_v[i].hash_fields = 0;
4838 rss_desc_v[i].queue =
4839 &final_policy->act_cnt[i].queue;
4840 rss_desc_v[i].queue_num = 1;
4841 rss_desc[i] = &rss_desc_v[i];
4846 sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4847 flow, policy, rss_desc);
4849 enum mlx5_meter_domain mtr_domain =
4850 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4851 (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4852 MLX5_MTR_DOMAIN_INGRESS);
4853 sub_policy = policy->sub_policys[mtr_domain][0];
4856 rte_flow_error_set(error, EINVAL,
4857 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4858 "Failed to get meter sub-policy.");
4864 * Split the meter flow.
4866 * As meter flow will split to three sub flow, other than meter
4867 * action, the other actions make sense to only meter accepts
4868 * the packet. If it need to be dropped, no other additional
4869 * actions should be take.
4871 * One kind of special action which decapsulates the L3 tunnel
4872 * header will be in the prefix sub flow, as not to take the
4873 * L3 tunnel header into account.
4876 * Pointer to Ethernet device.
4878 * Parent flow structure pointer.
4880 * Pointer to thread flow work space.
4882 * Flow rule attributes.
4884 * Pattern specification (list terminated by the END pattern item).
4885 * @param[out] sfx_items
4886 * Suffix flow match items (list terminated by the END pattern item).
4887 * @param[in] actions
4888 * Associated actions (list terminated by the END action).
4889 * @param[out] actions_sfx
4890 * Suffix flow actions.
4891 * @param[out] actions_pre
4892 * Prefix flow actions.
4893 * @param[out] mtr_flow_id
4894 * Pointer to meter flow id.
4896 * Perform verbose error reporting if not NULL.
4899 * 0 on success, a negative errno value otherwise and rte_errno is set.
4902 flow_meter_split_prep(struct rte_eth_dev *dev,
4903 struct rte_flow *flow,
4904 struct mlx5_flow_workspace *wks,
4905 const struct rte_flow_attr *attr,
4906 const struct rte_flow_item items[],
4907 struct rte_flow_item sfx_items[],
4908 const struct rte_flow_action actions[],
4909 struct rte_flow_action actions_sfx[],
4910 struct rte_flow_action actions_pre[],
4911 uint32_t *mtr_flow_id,
4912 struct rte_flow_error *error)
4914 struct mlx5_priv *priv = dev->data->dev_private;
4915 struct mlx5_flow_meter_info *fm = wks->fm;
4916 struct rte_flow_action *tag_action = NULL;
4917 struct rte_flow_item *tag_item;
4918 struct mlx5_rte_flow_action_set_tag *set_tag;
4919 const struct rte_flow_action_raw_encap *raw_encap;
4920 const struct rte_flow_action_raw_decap *raw_decap;
4921 struct mlx5_rte_flow_item_tag *tag_item_spec;
4922 struct mlx5_rte_flow_item_tag *tag_item_mask;
4923 uint32_t tag_id = 0;
4924 struct rte_flow_item *vlan_item_dst = NULL;
4925 const struct rte_flow_item *vlan_item_src = NULL;
4926 struct rte_flow_action *hw_mtr_action;
4927 struct rte_flow_action *action_pre_head = NULL;
4928 int32_t flow_src_port = priv->representor_id;
4930 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4931 uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4932 MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4933 uint32_t flow_id = 0;
4934 uint32_t flow_id_reversed = 0;
4935 uint8_t flow_id_bits = 0;
4938 /* Prepare the suffix subflow items. */
4939 tag_item = sfx_items++;
4940 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4941 struct mlx5_priv *port_priv;
4942 const struct rte_flow_item_port_id *pid_v;
4943 int item_type = items->type;
4945 switch (item_type) {
4946 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4947 pid_v = items->spec;
4949 port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4951 return rte_flow_error_set(error,
4953 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
4955 "Failed to get port info.");
4956 flow_src_port = port_priv->representor_id;
4957 if (!fm->def_policy && wks->policy->is_hierarchy &&
4958 flow_src_port != priv->representor_id) {
4959 if (flow_drv_mtr_hierarchy_rule_create(dev,
4966 memcpy(sfx_items, items, sizeof(*sfx_items));
4969 case RTE_FLOW_ITEM_TYPE_VLAN:
4970 /* Determine if copy vlan item below. */
4971 vlan_item_src = items;
4972 vlan_item_dst = sfx_items++;
4973 vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
4979 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4981 mtr_first = priv->sh->meter_aso_en &&
4982 (attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
4983 /* For ASO meter, meter must be before tag in TX direction. */
4985 action_pre_head = actions_pre++;
4986 /* Leave space for tag action. */
4987 tag_action = actions_pre++;
4989 /* Prepare the actions for prefix and suffix flow. */
4990 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4991 struct rte_flow_action *action_cur = NULL;
4993 switch (actions->type) {
4994 case RTE_FLOW_ACTION_TYPE_METER:
4996 action_cur = action_pre_head;
4998 /* Leave space for tag action. */
4999 tag_action = actions_pre++;
5000 action_cur = actions_pre++;
5003 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5004 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5005 action_cur = actions_pre++;
5007 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5008 raw_encap = actions->conf;
5009 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5010 action_cur = actions_pre++;
5012 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5013 raw_decap = actions->conf;
5014 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5015 action_cur = actions_pre++;
5017 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5018 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5019 if (vlan_item_dst && vlan_item_src) {
5020 memcpy(vlan_item_dst, vlan_item_src,
5021 sizeof(*vlan_item_dst));
5023 * Convert to internal match item, it is used
5024 * for vlan push and set vid.
5026 vlan_item_dst->type = (enum rte_flow_item_type)
5027 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5034 action_cur = (fm->def_policy) ?
5035 actions_sfx++ : actions_pre++;
5036 memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5038 /* Add end action to the actions. */
5039 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5040 if (priv->sh->meter_aso_en) {
5042 * For ASO meter, need to add an extra jump action explicitly,
5043 * to jump from meter to policer table.
5045 struct mlx5_flow_meter_sub_policy *sub_policy;
5046 struct mlx5_flow_tbl_data_entry *tbl_data;
5048 if (!fm->def_policy) {
5049 sub_policy = get_meter_sub_policy(dev, flow, wks,
5050 attr, items, error);
5054 enum mlx5_meter_domain mtr_domain =
5055 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5056 (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5057 MLX5_MTR_DOMAIN_INGRESS);
5060 &priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5062 tbl_data = container_of(sub_policy->tbl_rsc,
5063 struct mlx5_flow_tbl_data_entry, tbl);
5064 hw_mtr_action = actions_pre++;
5065 hw_mtr_action->type = (enum rte_flow_action_type)
5066 MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5067 hw_mtr_action->conf = tbl_data->jump.action;
5069 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5072 return rte_flow_error_set(error, ENOMEM,
5073 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5074 NULL, "No tag action space.");
5076 tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5079 /* Only default-policy Meter creates mtr flow id. */
5080 if (fm->def_policy) {
5081 mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5083 return rte_flow_error_set(error, ENOMEM,
5084 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5085 "Failed to allocate meter flow id.");
5086 flow_id = tag_id - 1;
5087 flow_id_bits = (!flow_id) ? 1 :
5088 (MLX5_REG_BITS - __builtin_clz(flow_id));
5089 if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5091 mlx5_ipool_free(fm->flow_ipool, tag_id);
5092 return rte_flow_error_set(error, EINVAL,
5093 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5094 "Meter flow id exceeds max limit.");
5096 if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5097 priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5099 /* Build tag actions and items for meter_id/meter flow_id. */
5100 set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5101 tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5102 tag_item_mask = tag_item_spec + 1;
5103 /* Both flow_id and meter_id share the same register. */
5104 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5105 .id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5107 .offset = mtr_id_offset,
5108 .length = mtr_reg_bits,
5109 .data = flow->meter,
5112 * The color Reg bits used by flow_id are growing from
5113 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5115 for (shift = 0; shift < flow_id_bits; shift++)
5116 flow_id_reversed = (flow_id_reversed << 1) |
5117 ((flow_id >> shift) & 0x1);
5119 flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5120 tag_item_spec->id = set_tag->id;
5121 tag_item_spec->data = set_tag->data << mtr_id_offset;
5122 tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5123 tag_action->type = (enum rte_flow_action_type)
5124 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5125 tag_action->conf = set_tag;
5126 tag_item->type = (enum rte_flow_item_type)
5127 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5128 tag_item->spec = tag_item_spec;
5129 tag_item->last = NULL;
5130 tag_item->mask = tag_item_mask;
5133 *mtr_flow_id = tag_id;
5138 * Split action list having QUEUE/RSS for metadata register copy.
5140 * Once Q/RSS action is detected in user's action list, the flow action
5141 * should be split in order to copy metadata registers, which will happen in
5143 * - CQE->flow_tag := reg_c[1] (MARK)
5144 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5145 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5146 * This is because the last action of each flow must be a terminal action
5147 * (QUEUE, RSS or DROP).
5149 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5150 * stored and kept in the mlx5_flow structure per each sub_flow.
5152 * The Q/RSS action is replaced with,
5153 * - SET_TAG, setting the allocated flow ID to reg_c[2].
5154 * And the following JUMP action is added at the end,
5155 * - JUMP, to RX_CP_TBL.
5157 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5158 * flow_create_split_metadata() routine. The flow will look like,
5159 * - If flow ID matches (reg_c[2]), perform Q/RSS.
5162 * Pointer to Ethernet device.
5163 * @param[out] split_actions
5164 * Pointer to store split actions to jump to CP_TBL.
5165 * @param[in] actions
5166 * Pointer to the list of original flow actions.
5168 * Pointer to the Q/RSS action.
5169 * @param[in] actions_n
5170 * Number of original actions.
5172 * Perform verbose error reporting if not NULL.
5175 * non-zero unique flow_id on success, otherwise 0 and
5176 * error/rte_error are set.
5179 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5180 struct rte_flow_action *split_actions,
5181 const struct rte_flow_action *actions,
5182 const struct rte_flow_action *qrss,
5183 int actions_n, struct rte_flow_error *error)
5185 struct mlx5_priv *priv = dev->data->dev_private;
5186 struct mlx5_rte_flow_action_set_tag *set_tag;
5187 struct rte_flow_action_jump *jump;
5188 const int qrss_idx = qrss - actions;
5189 uint32_t flow_id = 0;
5193 * Given actions will be split
5194 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5195 * - Add jump to mreg CP_TBL.
5196 * As a result, there will be one more action.
5199 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5200 set_tag = (void *)(split_actions + actions_n);
5202 * If tag action is not set to void(it means we are not the meter
5203 * suffix flow), add the tag action. Since meter suffix flow already
5204 * has the tag added.
5206 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5208 * Allocate the new subflow ID. This one is unique within
5209 * device and not shared with representors. Otherwise,
5210 * we would have to resolve multi-thread access synch
5211 * issue. Each flow on the shared device is appended
5212 * with source vport identifier, so the resulting
5213 * flows will be unique in the shared (by master and
5214 * representors) domain even if they have coinciding
5217 mlx5_ipool_malloc(priv->sh->ipool
5218 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5220 return rte_flow_error_set(error, ENOMEM,
5221 RTE_FLOW_ERROR_TYPE_ACTION,
5222 NULL, "can't allocate id "
5223 "for split Q/RSS subflow");
5224 /* Internal SET_TAG action to set flow ID. */
5225 *set_tag = (struct mlx5_rte_flow_action_set_tag){
5228 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5232 /* Construct new actions array. */
5233 /* Replace QUEUE/RSS action. */
5234 split_actions[qrss_idx] = (struct rte_flow_action){
5235 .type = (enum rte_flow_action_type)
5236 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5240 /* JUMP action to jump to mreg copy table (CP_TBL). */
5241 jump = (void *)(set_tag + 1);
5242 *jump = (struct rte_flow_action_jump){
5243 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5245 split_actions[actions_n - 2] = (struct rte_flow_action){
5246 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5249 split_actions[actions_n - 1] = (struct rte_flow_action){
5250 .type = RTE_FLOW_ACTION_TYPE_END,
5256 * Extend the given action list for Tx metadata copy.
5258 * Copy the given action list to the ext_actions and add flow metadata register
5259 * copy action in order to copy reg_a set by WQE to reg_c[0].
5261 * @param[out] ext_actions
5262 * Pointer to the extended action list.
5263 * @param[in] actions
5264 * Pointer to the list of actions.
5265 * @param[in] actions_n
5266 * Number of actions in the list.
5268 * Perform verbose error reporting if not NULL.
5269 * @param[in] encap_idx
5270 * The encap action inndex.
5273 * 0 on success, negative value otherwise
5276 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5277 struct rte_flow_action *ext_actions,
5278 const struct rte_flow_action *actions,
5279 int actions_n, struct rte_flow_error *error,
5282 struct mlx5_flow_action_copy_mreg *cp_mreg =
5283 (struct mlx5_flow_action_copy_mreg *)
5284 (ext_actions + actions_n + 1);
5287 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5291 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5296 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5297 if (encap_idx == actions_n - 1) {
5298 ext_actions[actions_n - 1] = (struct rte_flow_action){
5299 .type = (enum rte_flow_action_type)
5300 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5303 ext_actions[actions_n] = (struct rte_flow_action){
5304 .type = RTE_FLOW_ACTION_TYPE_END,
5307 ext_actions[encap_idx] = (struct rte_flow_action){
5308 .type = (enum rte_flow_action_type)
5309 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5312 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5313 sizeof(*ext_actions) * (actions_n - encap_idx));
5319 * Check the match action from the action list.
5321 * @param[in] actions
5322 * Pointer to the list of actions.
5324 * Flow rule attributes.
5326 * The action to be check if exist.
5327 * @param[out] match_action_pos
5328 * Pointer to the position of the matched action if exists, otherwise is -1.
5329 * @param[out] qrss_action_pos
5330 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5331 * @param[out] modify_after_mirror
5332 * Pointer to the flag of modify action after FDB mirroring.
5335 * > 0 the total number of actions.
5336 * 0 if not found match action in action list.
5339 flow_check_match_action(const struct rte_flow_action actions[],
5340 const struct rte_flow_attr *attr,
5341 enum rte_flow_action_type action,
5342 int *match_action_pos, int *qrss_action_pos,
5343 int *modify_after_mirror)
5345 const struct rte_flow_action_sample *sample;
5346 const struct rte_flow_action_raw_decap *decap;
5353 *match_action_pos = -1;
5354 *qrss_action_pos = -1;
5355 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5356 if (actions->type == action) {
5358 *match_action_pos = actions_n;
5360 switch (actions->type) {
5361 case RTE_FLOW_ACTION_TYPE_QUEUE:
5362 case RTE_FLOW_ACTION_TYPE_RSS:
5363 *qrss_action_pos = actions_n;
5365 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5366 sample = actions->conf;
5367 ratio = sample->ratio;
5368 sub_type = ((const struct rte_flow_action *)
5369 (sample->actions))->type;
5370 if (ratio == 1 && attr->transfer)
5373 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5374 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5375 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5376 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5377 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5378 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5379 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5380 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5381 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5382 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5383 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5384 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5385 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5386 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5387 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5388 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5389 case RTE_FLOW_ACTION_TYPE_FLAG:
5390 case RTE_FLOW_ACTION_TYPE_MARK:
5391 case RTE_FLOW_ACTION_TYPE_SET_META:
5392 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5393 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5394 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5395 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5396 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5397 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5398 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5399 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5400 case RTE_FLOW_ACTION_TYPE_METER:
5402 *modify_after_mirror = 1;
5404 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5405 decap = actions->conf;
5406 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5409 if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5410 const struct rte_flow_action_raw_encap *encap =
5413 MLX5_ENCAPSULATION_DECISION_SIZE &&
5415 MLX5_ENCAPSULATION_DECISION_SIZE)
5420 *modify_after_mirror = 1;
5427 if (flag && fdb_mirror && !*modify_after_mirror) {
5428 /* FDB mirroring uses the destination array to implement
5429 * instead of FLOW_SAMPLER object.
5431 if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5434 /* Count RTE_FLOW_ACTION_TYPE_END. */
5435 return flag ? actions_n + 1 : 0;
5438 #define SAMPLE_SUFFIX_ITEM 2
5441 * Split the sample flow.
5443 * As sample flow will split to two sub flow, sample flow with
5444 * sample action, the other actions will move to new suffix flow.
5446 * Also add unique tag id with tag action in the sample flow,
5447 * the same tag id will be as match in the suffix flow.
5450 * Pointer to Ethernet device.
5451 * @param[in] add_tag
5452 * Add extra tag action flag.
5453 * @param[out] sfx_items
5454 * Suffix flow match items (list terminated by the END pattern item).
5455 * @param[in] actions
5456 * Associated actions (list terminated by the END action).
5457 * @param[out] actions_sfx
5458 * Suffix flow actions.
5459 * @param[out] actions_pre
5460 * Prefix flow actions.
5461 * @param[in] actions_n
5462 * The total number of actions.
5463 * @param[in] sample_action_pos
5464 * The sample action position.
5465 * @param[in] qrss_action_pos
5466 * The Queue/RSS action position.
5467 * @param[in] jump_table
5468 * Add extra jump action flag.
5470 * Perform verbose error reporting if not NULL.
5473 * 0 on success, or unique flow_id, a negative errno value
5474 * otherwise and rte_errno is set.
5477 flow_sample_split_prep(struct rte_eth_dev *dev,
5479 struct rte_flow_item sfx_items[],
5480 const struct rte_flow_action actions[],
5481 struct rte_flow_action actions_sfx[],
5482 struct rte_flow_action actions_pre[],
5484 int sample_action_pos,
5485 int qrss_action_pos,
5487 struct rte_flow_error *error)
5489 struct mlx5_priv *priv = dev->data->dev_private;
5490 struct mlx5_rte_flow_action_set_tag *set_tag;
5491 struct mlx5_rte_flow_item_tag *tag_spec;
5492 struct mlx5_rte_flow_item_tag *tag_mask;
5493 struct rte_flow_action_jump *jump_action;
5494 uint32_t tag_id = 0;
5496 int append_index = 0;
5499 if (sample_action_pos < 0)
5500 return rte_flow_error_set(error, EINVAL,
5501 RTE_FLOW_ERROR_TYPE_ACTION,
5502 NULL, "invalid position of sample "
5504 /* Prepare the actions for prefix and suffix flow. */
5505 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5506 index = qrss_action_pos;
5507 /* Put the preceding the Queue/RSS action into prefix flow. */
5509 memcpy(actions_pre, actions,
5510 sizeof(struct rte_flow_action) * index);
5511 /* Put others preceding the sample action into prefix flow. */
5512 if (sample_action_pos > index + 1)
5513 memcpy(actions_pre + index, actions + index + 1,
5514 sizeof(struct rte_flow_action) *
5515 (sample_action_pos - index - 1));
5516 index = sample_action_pos - 1;
5517 /* Put Queue/RSS action into Suffix flow. */
5518 memcpy(actions_sfx, actions + qrss_action_pos,
5519 sizeof(struct rte_flow_action));
5522 index = sample_action_pos;
5524 memcpy(actions_pre, actions,
5525 sizeof(struct rte_flow_action) * index);
5527 /* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5528 * For CX6DX and above, metadata registers Cx preserve their value,
5529 * add an extra tag action for NIC-RX and E-Switch Domain.
5532 /* Prepare the prefix tag action. */
5534 set_tag = (void *)(actions_pre + actions_n + append_index);
5535 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5538 mlx5_ipool_malloc(priv->sh->ipool
5539 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5540 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5544 /* Prepare the suffix subflow items. */
5545 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5546 tag_spec->data = tag_id;
5547 tag_spec->id = set_tag->id;
5548 tag_mask = tag_spec + 1;
5549 tag_mask->data = UINT32_MAX;
5550 sfx_items[0] = (struct rte_flow_item){
5551 .type = (enum rte_flow_item_type)
5552 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5557 sfx_items[1] = (struct rte_flow_item){
5558 .type = (enum rte_flow_item_type)
5559 RTE_FLOW_ITEM_TYPE_END,
5561 /* Prepare the tag action in prefix subflow. */
5562 actions_pre[index++] =
5563 (struct rte_flow_action){
5564 .type = (enum rte_flow_action_type)
5565 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5569 memcpy(actions_pre + index, actions + sample_action_pos,
5570 sizeof(struct rte_flow_action));
5572 /* For the modify action after the sample action in E-Switch mirroring,
5573 * Add the extra jump action in prefix subflow and jump into the next
5574 * table, then do the modify action in the new table.
5577 /* Prepare the prefix jump action. */
5579 jump_action = (void *)(actions_pre + actions_n + append_index);
5580 jump_action->group = jump_table;
5581 actions_pre[index++] =
5582 (struct rte_flow_action){
5583 .type = (enum rte_flow_action_type)
5584 RTE_FLOW_ACTION_TYPE_JUMP,
5585 .conf = jump_action,
5588 actions_pre[index] = (struct rte_flow_action){
5589 .type = (enum rte_flow_action_type)
5590 RTE_FLOW_ACTION_TYPE_END,
5592 /* Put the actions after sample into Suffix flow. */
5593 memcpy(actions_sfx, actions + sample_action_pos + 1,
5594 sizeof(struct rte_flow_action) *
5595 (actions_n - sample_action_pos - 1));
5600 * The splitting for metadata feature.
5602 * - Q/RSS action on NIC Rx should be split in order to pass by
5603 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5604 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5606 * - All the actions on NIC Tx should have a mreg copy action to
5607 * copy reg_a from WQE to reg_c[0].
5610 * Pointer to Ethernet device.
5612 * Parent flow structure pointer.
5614 * Flow rule attributes.
5616 * Pattern specification (list terminated by the END pattern item).
5617 * @param[in] actions
5618 * Associated actions (list terminated by the END action).
5619 * @param[in] flow_split_info
5620 * Pointer to flow split info structure.
5622 * Perform verbose error reporting if not NULL.
5624 * 0 on success, negative value otherwise
5627 flow_create_split_metadata(struct rte_eth_dev *dev,
5628 struct rte_flow *flow,
5629 const struct rte_flow_attr *attr,
5630 const struct rte_flow_item items[],
5631 const struct rte_flow_action actions[],
5632 struct mlx5_flow_split_info *flow_split_info,
5633 struct rte_flow_error *error)
5635 struct mlx5_priv *priv = dev->data->dev_private;
5636 struct mlx5_dev_config *config = &priv->config;
5637 const struct rte_flow_action *qrss = NULL;
5638 struct rte_flow_action *ext_actions = NULL;
5639 struct mlx5_flow *dev_flow = NULL;
5640 uint32_t qrss_id = 0;
5647 /* Check whether extensive metadata feature is engaged. */
5648 if (!config->dv_flow_en ||
5649 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5650 !mlx5_flow_ext_mreg_supported(dev))
5651 return flow_create_split_inner(dev, flow, NULL, attr, items,
5652 actions, flow_split_info, error);
5653 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5656 /* Exclude hairpin flows from splitting. */
5657 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5658 const struct rte_flow_action_queue *queue;
5661 if (mlx5_rxq_get_type(dev, queue->index) ==
5662 MLX5_RXQ_TYPE_HAIRPIN)
5664 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5665 const struct rte_flow_action_rss *rss;
5668 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5669 MLX5_RXQ_TYPE_HAIRPIN)
5674 /* Check if it is in meter suffix table. */
5675 mtr_sfx = attr->group == (attr->transfer ?
5676 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5677 MLX5_FLOW_TABLE_LEVEL_METER);
5679 * Q/RSS action on NIC Rx should be split in order to pass by
5680 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5681 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5683 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5684 sizeof(struct rte_flow_action_set_tag) +
5685 sizeof(struct rte_flow_action_jump);
5686 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5689 return rte_flow_error_set(error, ENOMEM,
5690 RTE_FLOW_ERROR_TYPE_ACTION,
5691 NULL, "no memory to split "
5694 * If we are the suffix flow of meter, tag already exist.
5695 * Set the tag action to void.
5698 ext_actions[qrss - actions].type =
5699 RTE_FLOW_ACTION_TYPE_VOID;
5701 ext_actions[qrss - actions].type =
5702 (enum rte_flow_action_type)
5703 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5705 * Create the new actions list with removed Q/RSS action
5706 * and appended set tag and jump to register copy table
5707 * (RX_CP_TBL). We should preallocate unique tag ID here
5708 * in advance, because it is needed for set tag action.
5710 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5711 qrss, actions_n, error);
5712 if (!mtr_sfx && !qrss_id) {
5716 } else if (attr->egress && !attr->transfer) {
5718 * All the actions on NIC Tx should have a metadata register
5719 * copy action to copy reg_a from WQE to reg_c[meta]
5721 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5722 sizeof(struct mlx5_flow_action_copy_mreg);
5723 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5726 return rte_flow_error_set(error, ENOMEM,
5727 RTE_FLOW_ERROR_TYPE_ACTION,
5728 NULL, "no memory to split "
5730 /* Create the action list appended with copy register. */
5731 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5732 actions_n, error, encap_idx);
5736 /* Add the unmodified original or prefix subflow. */
5737 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5738 items, ext_actions ? ext_actions :
5739 actions, flow_split_info, error);
5742 MLX5_ASSERT(dev_flow);
5744 const struct rte_flow_attr q_attr = {
5745 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5748 /* Internal PMD action to set register. */
5749 struct mlx5_rte_flow_item_tag q_tag_spec = {
5753 struct rte_flow_item q_items[] = {
5755 .type = (enum rte_flow_item_type)
5756 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5757 .spec = &q_tag_spec,
5762 .type = RTE_FLOW_ITEM_TYPE_END,
5765 struct rte_flow_action q_actions[] = {
5771 .type = RTE_FLOW_ACTION_TYPE_END,
5774 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5777 * Configure the tag item only if there is no meter subflow.
5778 * Since tag is already marked in the meter suffix subflow
5779 * we can just use the meter suffix items as is.
5782 /* Not meter subflow. */
5783 MLX5_ASSERT(!mtr_sfx);
5785 * Put unique id in prefix flow due to it is destroyed
5786 * after suffix flow and id will be freed after there
5787 * is no actual flows with this id and identifier
5788 * reallocation becomes possible (for example, for
5789 * other flows in other threads).
5791 dev_flow->handle->split_flow_id = qrss_id;
5792 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5796 q_tag_spec.id = ret;
5799 /* Add suffix subflow to execute Q/RSS. */
5800 flow_split_info->prefix_layers = layers;
5801 flow_split_info->prefix_mark = 0;
5802 ret = flow_create_split_inner(dev, flow, &dev_flow,
5803 &q_attr, mtr_sfx ? items :
5805 flow_split_info, error);
5808 /* qrss ID should be freed if failed. */
5810 MLX5_ASSERT(dev_flow);
5815 * We do not destroy the partially created sub_flows in case of error.
5816 * These ones are included into parent flow list and will be destroyed
5817 * by flow_drv_destroy.
5819 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5821 mlx5_free(ext_actions);
5826 * Create meter internal drop flow with the original pattern.
5829 * Pointer to Ethernet device.
5831 * Parent flow structure pointer.
5833 * Flow rule attributes.
5835 * Pattern specification (list terminated by the END pattern item).
5836 * @param[in] flow_split_info
5837 * Pointer to flow split info structure.
5839 * Pointer to flow meter structure.
5841 * Perform verbose error reporting if not NULL.
5843 * 0 on success, negative value otherwise
5846 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5847 struct rte_flow *flow,
5848 const struct rte_flow_attr *attr,
5849 const struct rte_flow_item items[],
5850 struct mlx5_flow_split_info *flow_split_info,
5851 struct mlx5_flow_meter_info *fm,
5852 struct rte_flow_error *error)
5854 struct mlx5_flow *dev_flow = NULL;
5855 struct rte_flow_attr drop_attr = *attr;
5856 struct rte_flow_action drop_actions[3];
5857 struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5859 MLX5_ASSERT(fm->drop_cnt);
5860 drop_actions[0].type =
5861 (enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5862 drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5863 drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5864 drop_actions[1].conf = NULL;
5865 drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5866 drop_actions[2].conf = NULL;
5867 drop_split_info.external = false;
5868 drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5869 drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5870 drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5871 return flow_create_split_inner(dev, flow, &dev_flow,
5872 &drop_attr, items, drop_actions,
5873 &drop_split_info, error);
5877 * The splitting for meter feature.
5879 * - The meter flow will be split to two flows as prefix and
5880 * suffix flow. The packets make sense only it pass the prefix
5883 * - Reg_C_5 is used for the packet to match betweend prefix and
5887 * Pointer to Ethernet device.
5889 * Parent flow structure pointer.
5891 * Flow rule attributes.
5893 * Pattern specification (list terminated by the END pattern item).
5894 * @param[in] actions
5895 * Associated actions (list terminated by the END action).
5896 * @param[in] flow_split_info
5897 * Pointer to flow split info structure.
5899 * Perform verbose error reporting if not NULL.
5901 * 0 on success, negative value otherwise
5904 flow_create_split_meter(struct rte_eth_dev *dev,
5905 struct rte_flow *flow,
5906 const struct rte_flow_attr *attr,
5907 const struct rte_flow_item items[],
5908 const struct rte_flow_action actions[],
5909 struct mlx5_flow_split_info *flow_split_info,
5910 struct rte_flow_error *error)
5912 struct mlx5_priv *priv = dev->data->dev_private;
5913 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5914 struct rte_flow_action *sfx_actions = NULL;
5915 struct rte_flow_action *pre_actions = NULL;
5916 struct rte_flow_item *sfx_items = NULL;
5917 struct mlx5_flow *dev_flow = NULL;
5918 struct rte_flow_attr sfx_attr = *attr;
5919 struct mlx5_flow_meter_info *fm = NULL;
5920 uint8_t skip_scale_restore;
5921 bool has_mtr = false;
5922 bool has_modify = false;
5923 bool set_mtr_reg = true;
5924 bool is_mtr_hierarchy = false;
5925 uint32_t meter_id = 0;
5926 uint32_t mtr_idx = 0;
5927 uint32_t mtr_flow_id = 0;
5934 actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5935 &has_modify, &meter_id);
5938 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5940 return rte_flow_error_set(error, EINVAL,
5941 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5942 NULL, "Meter not found.");
5944 fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5946 return rte_flow_error_set(error, EINVAL,
5947 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5948 NULL, "Meter not found.");
5949 ret = mlx5_flow_meter_attach(priv, fm,
5953 flow->meter = mtr_idx;
5957 if (!fm->def_policy) {
5958 wks->policy = mlx5_flow_meter_policy_find(dev,
5961 MLX5_ASSERT(wks->policy);
5962 if (wks->policy->is_hierarchy) {
5964 mlx5_flow_meter_hierarchy_get_final_policy(dev,
5966 if (!wks->final_policy)
5967 return rte_flow_error_set(error,
5969 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5970 "Failed to find terminal policy of hierarchy.");
5971 is_mtr_hierarchy = true;
5975 * If it isn't default-policy Meter, and
5976 * 1. There's no action in flow to change
5977 * packet (modify/encap/decap etc.), OR
5978 * 2. No drop count needed for this meter.
5979 * 3. It's not meter hierarchy.
5980 * Then no need to use regC to save meter id anymore.
5982 if (!fm->def_policy && !is_mtr_hierarchy &&
5983 (!has_modify || !fm->drop_cnt))
5984 set_mtr_reg = false;
5985 /* Prefix actions: meter, decap, encap, tag, jump, end. */
5986 act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5987 sizeof(struct mlx5_rte_flow_action_set_tag);
5988 /* Suffix items: tag, vlan, port id, end. */
5989 #define METER_SUFFIX_ITEM 4
5990 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5991 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5992 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5995 return rte_flow_error_set(error, ENOMEM,
5996 RTE_FLOW_ERROR_TYPE_ACTION,
5997 NULL, "no memory to split "
5999 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6001 /* There's no suffix flow for meter of non-default policy. */
6002 if (!fm->def_policy)
6003 pre_actions = sfx_actions + 1;
6005 pre_actions = sfx_actions + actions_n;
6006 ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6007 items, sfx_items, actions,
6008 sfx_actions, pre_actions,
6009 (set_mtr_reg ? &mtr_flow_id : NULL),
6015 /* Add the prefix subflow. */
6016 flow_split_info->prefix_mark = 0;
6017 skip_scale_restore = flow_split_info->skip_scale;
6018 flow_split_info->skip_scale |=
6019 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6020 ret = flow_create_split_inner(dev, flow, &dev_flow,
6021 attr, items, pre_actions,
6022 flow_split_info, error);
6023 flow_split_info->skip_scale = skip_scale_restore;
6026 mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6031 dev_flow->handle->split_flow_id = mtr_flow_id;
6032 dev_flow->handle->is_meter_flow_id = 1;
6034 if (!fm->def_policy) {
6035 if (!set_mtr_reg && fm->drop_cnt)
6037 flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6043 /* Setting the sfx group atrr. */
6044 sfx_attr.group = sfx_attr.transfer ?
6045 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6046 MLX5_FLOW_TABLE_LEVEL_METER;
6047 flow_split_info->prefix_layers =
6048 flow_get_prefix_layer_flags(dev_flow);
6049 flow_split_info->prefix_mark = dev_flow->handle->mark;
6050 flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6052 /* Add the prefix subflow. */
6053 ret = flow_create_split_metadata(dev, flow,
6054 &sfx_attr, sfx_items ?
6056 sfx_actions ? sfx_actions : actions,
6057 flow_split_info, error);
6060 mlx5_free(sfx_actions);
6065 * The splitting for sample feature.
6067 * Once Sample action is detected in the action list, the flow actions should
6068 * be split into prefix sub flow and suffix sub flow.
6070 * The original items remain in the prefix sub flow, all actions preceding the
6071 * sample action and the sample action itself will be copied to the prefix
6072 * sub flow, the actions following the sample action will be copied to the
6073 * suffix sub flow, Queue action always be located in the suffix sub flow.
6075 * In order to make the packet from prefix sub flow matches with suffix sub
6076 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6077 * flow uses tag item with the unique flow id.
6080 * Pointer to Ethernet device.
6082 * Parent flow structure pointer.
6084 * Flow rule attributes.
6086 * Pattern specification (list terminated by the END pattern item).
6087 * @param[in] actions
6088 * Associated actions (list terminated by the END action).
6089 * @param[in] flow_split_info
6090 * Pointer to flow split info structure.
6092 * Perform verbose error reporting if not NULL.
6094 * 0 on success, negative value otherwise
6097 flow_create_split_sample(struct rte_eth_dev *dev,
6098 struct rte_flow *flow,
6099 const struct rte_flow_attr *attr,
6100 const struct rte_flow_item items[],
6101 const struct rte_flow_action actions[],
6102 struct mlx5_flow_split_info *flow_split_info,
6103 struct rte_flow_error *error)
6105 struct mlx5_priv *priv = dev->data->dev_private;
6106 struct rte_flow_action *sfx_actions = NULL;
6107 struct rte_flow_action *pre_actions = NULL;
6108 struct rte_flow_item *sfx_items = NULL;
6109 struct mlx5_flow *dev_flow = NULL;
6110 struct rte_flow_attr sfx_attr = *attr;
6111 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6112 struct mlx5_flow_dv_sample_resource *sample_res;
6113 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6114 struct mlx5_flow_tbl_resource *sfx_tbl;
6118 uint32_t fdb_tx = 0;
6121 int sample_action_pos;
6122 int qrss_action_pos;
6124 int modify_after_mirror = 0;
6125 uint16_t jump_table = 0;
6126 const uint32_t next_ft_step = 1;
6129 if (priv->sampler_en)
6130 actions_n = flow_check_match_action(actions, attr,
6131 RTE_FLOW_ACTION_TYPE_SAMPLE,
6132 &sample_action_pos, &qrss_action_pos,
6133 &modify_after_mirror);
6135 /* The prefix actions must includes sample, tag, end. */
6136 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6137 + sizeof(struct mlx5_rte_flow_action_set_tag);
6138 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6139 sizeof(struct mlx5_rte_flow_item_tag) * 2;
6140 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6141 item_size), 0, SOCKET_ID_ANY);
6143 return rte_flow_error_set(error, ENOMEM,
6144 RTE_FLOW_ERROR_TYPE_ACTION,
6145 NULL, "no memory to split "
6147 /* The representor_id is UINT16_MAX for uplink. */
6148 fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6150 * When reg_c_preserve is set, metadata registers Cx preserve
6151 * their value even through packet duplication.
6153 add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6155 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6157 if (modify_after_mirror)
6158 jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6160 pre_actions = sfx_actions + actions_n;
6161 tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6162 actions, sfx_actions,
6163 pre_actions, actions_n,
6165 qrss_action_pos, jump_table,
6167 if (tag_id < 0 || (add_tag && !tag_id)) {
6171 if (modify_after_mirror)
6172 flow_split_info->skip_scale =
6173 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6174 /* Add the prefix subflow. */
6175 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6177 flow_split_info, error);
6182 dev_flow->handle->split_flow_id = tag_id;
6183 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6184 if (!modify_after_mirror) {
6185 /* Set the sfx group attr. */
6186 sample_res = (struct mlx5_flow_dv_sample_resource *)
6187 dev_flow->dv.sample_res;
6188 sfx_tbl = (struct mlx5_flow_tbl_resource *)
6189 sample_res->normal_path_tbl;
6190 sfx_tbl_data = container_of(sfx_tbl,
6191 struct mlx5_flow_tbl_data_entry,
6193 sfx_attr.group = sfx_attr.transfer ?
6194 (sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6196 MLX5_ASSERT(attr->transfer);
6197 sfx_attr.group = jump_table;
6199 flow_split_info->prefix_layers =
6200 flow_get_prefix_layer_flags(dev_flow);
6201 flow_split_info->prefix_mark = dev_flow->handle->mark;
6202 /* Suffix group level already be scaled with factor, set
6203 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6204 * again in translation.
6206 flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6209 /* Add the suffix subflow. */
6210 ret = flow_create_split_meter(dev, flow, &sfx_attr,
6211 sfx_items ? sfx_items : items,
6212 sfx_actions ? sfx_actions : actions,
6213 flow_split_info, error);
6216 mlx5_free(sfx_actions);
6221 * Split the flow to subflow set. The splitters might be linked
6222 * in the chain, like this:
6223 * flow_create_split_outer() calls:
6224 * flow_create_split_meter() calls:
6225 * flow_create_split_metadata(meter_subflow_0) calls:
6226 * flow_create_split_inner(metadata_subflow_0)
6227 * flow_create_split_inner(metadata_subflow_1)
6228 * flow_create_split_inner(metadata_subflow_2)
6229 * flow_create_split_metadata(meter_subflow_1) calls:
6230 * flow_create_split_inner(metadata_subflow_0)
6231 * flow_create_split_inner(metadata_subflow_1)
6232 * flow_create_split_inner(metadata_subflow_2)
6234 * This provide flexible way to add new levels of flow splitting.
6235 * The all of successfully created subflows are included to the
6236 * parent flow dev_flow list.
6239 * Pointer to Ethernet device.
6241 * Parent flow structure pointer.
6243 * Flow rule attributes.
6245 * Pattern specification (list terminated by the END pattern item).
6246 * @param[in] actions
6247 * Associated actions (list terminated by the END action).
6248 * @param[in] flow_split_info
6249 * Pointer to flow split info structure.
6251 * Perform verbose error reporting if not NULL.
6253 * 0 on success, negative value otherwise
6256 flow_create_split_outer(struct rte_eth_dev *dev,
6257 struct rte_flow *flow,
6258 const struct rte_flow_attr *attr,
6259 const struct rte_flow_item items[],
6260 const struct rte_flow_action actions[],
6261 struct mlx5_flow_split_info *flow_split_info,
6262 struct rte_flow_error *error)
6266 ret = flow_create_split_sample(dev, flow, attr, items,
6267 actions, flow_split_info, error);
6268 MLX5_ASSERT(ret <= 0);
6272 static inline struct mlx5_flow_tunnel *
6273 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6275 struct mlx5_flow_tunnel *tunnel;
6277 #pragma GCC diagnostic push
6278 #pragma GCC diagnostic ignored "-Wcast-qual"
6279 tunnel = (typeof(tunnel))flow->tunnel;
6280 #pragma GCC diagnostic pop
6286 * Adjust flow RSS workspace if needed.
6289 * Pointer to thread flow work space.
6291 * Pointer to RSS descriptor.
6292 * @param[in] nrssq_num
6293 * New RSS queue number.
6296 * 0 on success, -1 otherwise and rte_errno is set.
6299 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6300 struct mlx5_flow_rss_desc *rss_desc,
6303 if (likely(nrssq_num <= wks->rssq_num))
6305 rss_desc->queue = realloc(rss_desc->queue,
6306 sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6307 if (!rss_desc->queue) {
6311 wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6316 * Create a flow and add it to @p list.
6319 * Pointer to Ethernet device.
6321 * Pointer to a TAILQ flow list. If this parameter NULL,
6322 * no list insertion occurred, flow is just created,
6323 * this is caller's responsibility to track the
6326 * Flow rule attributes.
6328 * Pattern specification (list terminated by the END pattern item).
6329 * @param[in] actions
6330 * Associated actions (list terminated by the END action).
6331 * @param[in] external
6332 * This flow rule is created by request external to PMD.
6334 * Perform verbose error reporting if not NULL.
6337 * A flow index on success, 0 otherwise and rte_errno is set.
6340 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6341 const struct rte_flow_attr *attr,
6342 const struct rte_flow_item items[],
6343 const struct rte_flow_action original_actions[],
6344 bool external, struct rte_flow_error *error)
6346 struct mlx5_priv *priv = dev->data->dev_private;
6347 struct rte_flow *flow = NULL;
6348 struct mlx5_flow *dev_flow;
6349 const struct rte_flow_action_rss *rss = NULL;
6350 struct mlx5_translated_action_handle
6351 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6352 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6354 struct mlx5_flow_expand_rss buf;
6355 uint8_t buffer[4096];
6358 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6359 uint8_t buffer[2048];
6362 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6363 uint8_t buffer[2048];
6364 } actions_hairpin_tx;
6366 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6367 uint8_t buffer[2048];
6369 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6370 struct mlx5_flow_rss_desc *rss_desc;
6371 const struct rte_flow_action *p_actions_rx;
6375 struct rte_flow_attr attr_tx = { .priority = 0 };
6376 const struct rte_flow_action *actions;
6377 struct rte_flow_action *translated_actions = NULL;
6378 struct mlx5_flow_tunnel *tunnel;
6379 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6380 struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6381 struct mlx5_flow_split_info flow_split_info = {
6382 .external = !!external,
6392 rss_desc = &wks->rss_desc;
6393 ret = flow_action_handles_translate(dev, original_actions,
6396 &translated_actions, error);
6398 MLX5_ASSERT(translated_actions == NULL);
6401 actions = translated_actions ? translated_actions : original_actions;
6402 p_actions_rx = actions;
6403 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6404 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6405 external, hairpin_flow, error);
6407 goto error_before_hairpin_split;
6408 flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6411 goto error_before_hairpin_split;
6413 if (hairpin_flow > 0) {
6414 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6416 goto error_before_hairpin_split;
6418 flow_hairpin_split(dev, actions, actions_rx.actions,
6419 actions_hairpin_tx.actions, items_tx.items,
6421 p_actions_rx = actions_rx.actions;
6423 flow_split_info.flow_idx = idx;
6424 flow->drv_type = flow_get_drv_type(dev, attr);
6425 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6426 flow->drv_type < MLX5_FLOW_TYPE_MAX);
6427 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6428 /* RSS Action only works on NIC RX domain */
6429 if (attr->ingress && !attr->transfer)
6430 rss = flow_get_rss_action(dev, p_actions_rx);
6432 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6435 * The following information is required by
6436 * mlx5_flow_hashfields_adjust() in advance.
6438 rss_desc->level = rss->level;
6439 /* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6440 rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6442 flow->dev_handles = 0;
6443 if (rss && rss->types) {
6444 unsigned int graph_root;
6446 graph_root = find_graph_root(rss->level);
6447 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6449 mlx5_support_expansion, graph_root);
6450 MLX5_ASSERT(ret > 0 &&
6451 (unsigned int)ret < sizeof(expand_buffer.buffer));
6452 if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6453 for (i = 0; i < buf->entries; ++i)
6454 mlx5_dbg__print_pattern(buf->entry[i].pattern);
6458 buf->entry[0].pattern = (void *)(uintptr_t)items;
6460 rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6462 for (i = 0; i < buf->entries; ++i) {
6463 /* Initialize flow split data. */
6464 flow_split_info.prefix_layers = 0;
6465 flow_split_info.prefix_mark = 0;
6466 flow_split_info.skip_scale = 0;
6468 * The splitter may create multiple dev_flows,
6469 * depending on configuration. In the simplest
6470 * case it just creates unmodified original flow.
6472 ret = flow_create_split_outer(dev, flow, attr,
6473 buf->entry[i].pattern,
6474 p_actions_rx, &flow_split_info,
6478 if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6479 ret = flow_tunnel_add_default_miss(dev, flow, attr,
6482 wks->flows[0].tunnel,
6486 mlx5_free(default_miss_ctx.queue);
6491 /* Create the tx flow. */
6493 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6494 attr_tx.ingress = 0;
6496 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6497 actions_hairpin_tx.actions,
6501 dev_flow->flow = flow;
6502 dev_flow->external = 0;
6503 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6504 dev_flow->handle, next);
6505 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6507 actions_hairpin_tx.actions, error);
6512 * Update the metadata register copy table. If extensive
6513 * metadata feature is enabled and registers are supported
6514 * we might create the extra rte_flow for each unique
6515 * MARK/FLAG action ID.
6517 * The table is updated for ingress Flows only, because
6518 * the egress Flows belong to the different device and
6519 * copy table should be updated in peer NIC Rx domain.
6521 if (attr->ingress &&
6522 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6523 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6528 * If the flow is external (from application) OR device is started,
6529 * OR mreg discover, then apply immediately.
6531 if (external || dev->data->dev_started ||
6532 (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6533 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6534 ret = flow_drv_apply(dev, flow, error);
6539 flow_rxq_flags_set(dev, flow);
6540 rte_free(translated_actions);
6541 tunnel = flow_tunnel_from_rule(wks->flows);
6544 flow->tunnel_id = tunnel->tunnel_id;
6545 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6546 mlx5_free(default_miss_ctx.queue);
6548 mlx5_flow_pop_thread_workspace();
6552 ret = rte_errno; /* Save rte_errno before cleanup. */
6553 flow_mreg_del_copy_action(dev, flow);
6554 flow_drv_destroy(dev, flow);
6555 if (rss_desc->shared_rss)
6556 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6558 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6559 rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6560 mlx5_ipool_free(priv->flows[type], idx);
6561 rte_errno = ret; /* Restore rte_errno. */
6564 mlx5_flow_pop_thread_workspace();
6565 error_before_hairpin_split:
6566 rte_free(translated_actions);
6571 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6572 * incoming packets to table 1.
6574 * Other flow rules, requested for group n, will be created in
6575 * e-switch table n+1.
6576 * Jump action to e-switch group n will be created to group n+1.
6578 * Used when working in switchdev mode, to utilise advantages of table 1
6582 * Pointer to Ethernet device.
6585 * Pointer to flow on success, NULL otherwise and rte_errno is set.
6588 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6590 const struct rte_flow_attr attr = {
6597 const struct rte_flow_item pattern = {
6598 .type = RTE_FLOW_ITEM_TYPE_END,
6600 struct rte_flow_action_jump jump = {
6603 const struct rte_flow_action actions[] = {
6605 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6609 .type = RTE_FLOW_ACTION_TYPE_END,
6612 struct rte_flow_error error;
6614 return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6616 actions, false, &error);
6620 * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6621 * and sq number, directs all packets to peer vport.
6624 * Pointer to Ethernet device.
6629 * Flow ID on success, 0 otherwise and rte_errno is set.
6632 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6634 struct rte_flow_attr attr = {
6636 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6641 struct rte_flow_item_port_id port_spec = {
6642 .id = MLX5_PORT_ESW_MGR,
6644 struct mlx5_rte_flow_item_tx_queue txq_spec = {
6647 struct rte_flow_item pattern[] = {
6649 .type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6653 .type = (enum rte_flow_item_type)
6654 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6658 .type = RTE_FLOW_ITEM_TYPE_END,
6661 struct rte_flow_action_jump jump = {
6664 struct rte_flow_action_port_id port = {
6665 .id = dev->data->port_id,
6667 struct rte_flow_action actions[] = {
6669 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6673 .type = RTE_FLOW_ACTION_TYPE_END,
6676 struct rte_flow_error error;
6679 * Creates group 0, highest priority jump flow.
6680 * Matches txq to bypass kernel packets.
6682 if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6683 false, &error) == 0)
6685 /* Create group 1, lowest priority redirect flow for txq. */
6687 actions[0].conf = &port;
6688 actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6689 return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6690 actions, false, &error);
6694 * Validate a flow supported by the NIC.
6696 * @see rte_flow_validate()
6700 mlx5_flow_validate(struct rte_eth_dev *dev,
6701 const struct rte_flow_attr *attr,
6702 const struct rte_flow_item items[],
6703 const struct rte_flow_action original_actions[],
6704 struct rte_flow_error *error)
6707 struct mlx5_translated_action_handle
6708 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6709 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6710 const struct rte_flow_action *actions;
6711 struct rte_flow_action *translated_actions = NULL;
6712 int ret = flow_action_handles_translate(dev, original_actions,
6715 &translated_actions, error);
6719 actions = translated_actions ? translated_actions : original_actions;
6720 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6721 ret = flow_drv_validate(dev, attr, items, actions,
6722 true, hairpin_flow, error);
6723 rte_free(translated_actions);
6730 * @see rte_flow_create()
6734 mlx5_flow_create(struct rte_eth_dev *dev,
6735 const struct rte_flow_attr *attr,
6736 const struct rte_flow_item items[],
6737 const struct rte_flow_action actions[],
6738 struct rte_flow_error *error)
6741 * If the device is not started yet, it is not allowed to created a
6742 * flow from application. PMD default flows and traffic control flows
6745 if (unlikely(!dev->data->dev_started)) {
6746 DRV_LOG(DEBUG, "port %u is not started when "
6747 "inserting a flow", dev->data->port_id);
6748 rte_flow_error_set(error, ENODEV,
6749 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6751 "port not started");
6755 return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6756 attr, items, actions,
6761 * Destroy a flow in a list.
6764 * Pointer to Ethernet device.
6765 * @param[in] flow_idx
6766 * Index of flow to destroy.
6769 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6772 struct mlx5_priv *priv = dev->data->dev_private;
6773 struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6777 MLX5_ASSERT(flow->type == type);
6779 * Update RX queue flags only if port is started, otherwise it is
6782 if (dev->data->dev_started)
6783 flow_rxq_flags_trim(dev, flow);
6784 flow_drv_destroy(dev, flow);
6786 struct mlx5_flow_tunnel *tunnel;
6788 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6790 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6791 mlx5_flow_tunnel_free(dev, tunnel);
6793 flow_mreg_del_copy_action(dev, flow);
6794 mlx5_ipool_free(priv->flows[type], flow_idx);
6798 * Destroy all flows.
6801 * Pointer to Ethernet device.
6803 * Flow type to be flushed.
6805 * If flushing is called avtively.
6808 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6811 struct mlx5_priv *priv = dev->data->dev_private;
6812 uint32_t num_flushed = 0, fidx = 1;
6813 struct rte_flow *flow;
6815 MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6816 flow_list_destroy(dev, type, fidx);
6820 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6821 dev->data->port_id, num_flushed);
6826 * Stop all default actions for flows.
6829 * Pointer to Ethernet device.
6832 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6834 flow_mreg_del_default_copy_action(dev);
6835 flow_rxq_flags_clear(dev);
6839 * Start all default actions for flows.
6842 * Pointer to Ethernet device.
6844 * 0 on success, a negative errno value otherwise and rte_errno is set.
6847 mlx5_flow_start_default(struct rte_eth_dev *dev)
6849 struct rte_flow_error error;
6851 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6852 return flow_mreg_add_default_copy_action(dev, &error);
6856 * Release key of thread specific flow workspace data.
6859 flow_release_workspace(void *data)
6861 struct mlx5_flow_workspace *wks = data;
6862 struct mlx5_flow_workspace *next;
6866 free(wks->rss_desc.queue);
6873 * Get thread specific current flow workspace.
6875 * @return pointer to thread specific flow workspace data, NULL on error.
6877 struct mlx5_flow_workspace*
6878 mlx5_flow_get_thread_workspace(void)
6880 struct mlx5_flow_workspace *data;
6882 data = mlx5_flow_os_get_specific_workspace();
6883 MLX5_ASSERT(data && data->inuse);
6884 if (!data || !data->inuse)
6885 DRV_LOG(ERR, "flow workspace not initialized.");
6890 * Allocate and init new flow workspace.
6892 * @return pointer to flow workspace data, NULL on error.
6894 static struct mlx5_flow_workspace*
6895 flow_alloc_thread_workspace(void)
6897 struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6900 DRV_LOG(ERR, "Failed to allocate flow workspace "
6904 data->rss_desc.queue = calloc(1,
6905 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6906 if (!data->rss_desc.queue)
6908 data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6911 if (data->rss_desc.queue)
6912 free(data->rss_desc.queue);
6918 * Get new thread specific flow workspace.
6920 * If current workspace inuse, create new one and set as current.
6922 * @return pointer to thread specific flow workspace data, NULL on error.
6924 static struct mlx5_flow_workspace*
6925 mlx5_flow_push_thread_workspace(void)
6927 struct mlx5_flow_workspace *curr;
6928 struct mlx5_flow_workspace *data;
6930 curr = mlx5_flow_os_get_specific_workspace();
6932 data = flow_alloc_thread_workspace();
6935 } else if (!curr->inuse) {
6937 } else if (curr->next) {
6940 data = flow_alloc_thread_workspace();
6948 /* Set as current workspace */
6949 if (mlx5_flow_os_set_specific_workspace(data))
6950 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6955 * Close current thread specific flow workspace.
6957 * If previous workspace available, set it as current.
6959 * @return pointer to thread specific flow workspace data, NULL on error.
6962 mlx5_flow_pop_thread_workspace(void)
6964 struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6969 DRV_LOG(ERR, "Failed to close unused flow workspace.");
6975 if (mlx5_flow_os_set_specific_workspace(data->prev))
6976 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6980 * Verify the flow list is empty
6983 * Pointer to Ethernet device.
6985 * @return the number of flows not released.
6988 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
6990 struct mlx5_priv *priv = dev->data->dev_private;
6991 struct rte_flow *flow;
6995 for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
6996 MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
6997 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6998 dev->data->port_id, (void *)flow);
7006 * Enable default hairpin egress flow.
7009 * Pointer to Ethernet device.
7014 * 0 on success, a negative errno value otherwise and rte_errno is set.
7017 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7020 const struct rte_flow_attr attr = {
7024 struct mlx5_rte_flow_item_tx_queue queue_spec = {
7027 struct mlx5_rte_flow_item_tx_queue queue_mask = {
7028 .queue = UINT32_MAX,
7030 struct rte_flow_item items[] = {
7032 .type = (enum rte_flow_item_type)
7033 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7034 .spec = &queue_spec,
7036 .mask = &queue_mask,
7039 .type = RTE_FLOW_ITEM_TYPE_END,
7042 struct rte_flow_action_jump jump = {
7043 .group = MLX5_HAIRPIN_TX_TABLE,
7045 struct rte_flow_action actions[2];
7047 struct rte_flow_error error;
7049 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7050 actions[0].conf = &jump;
7051 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7052 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7053 &attr, items, actions, false, &error);
7056 "Failed to create ctrl flow: rte_errno(%d),"
7057 " type(%d), message(%s)",
7058 rte_errno, error.type,
7059 error.message ? error.message : " (no stated reason)");
7066 * Enable a control flow configured from the control plane.
7069 * Pointer to Ethernet device.
7071 * An Ethernet flow spec to apply.
7073 * An Ethernet flow mask to apply.
7075 * A VLAN flow spec to apply.
7077 * A VLAN flow mask to apply.
7080 * 0 on success, a negative errno value otherwise and rte_errno is set.
7083 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7084 struct rte_flow_item_eth *eth_spec,
7085 struct rte_flow_item_eth *eth_mask,
7086 struct rte_flow_item_vlan *vlan_spec,
7087 struct rte_flow_item_vlan *vlan_mask)
7089 struct mlx5_priv *priv = dev->data->dev_private;
7090 const struct rte_flow_attr attr = {
7092 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7094 struct rte_flow_item items[] = {
7096 .type = RTE_FLOW_ITEM_TYPE_ETH,
7102 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7103 RTE_FLOW_ITEM_TYPE_END,
7109 .type = RTE_FLOW_ITEM_TYPE_END,
7112 uint16_t queue[priv->reta_idx_n];
7113 struct rte_flow_action_rss action_rss = {
7114 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7116 .types = priv->rss_conf.rss_hf,
7117 .key_len = priv->rss_conf.rss_key_len,
7118 .queue_num = priv->reta_idx_n,
7119 .key = priv->rss_conf.rss_key,
7122 struct rte_flow_action actions[] = {
7124 .type = RTE_FLOW_ACTION_TYPE_RSS,
7125 .conf = &action_rss,
7128 .type = RTE_FLOW_ACTION_TYPE_END,
7132 struct rte_flow_error error;
7135 if (!priv->reta_idx_n || !priv->rxqs_n) {
7138 if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7139 action_rss.types = 0;
7140 for (i = 0; i != priv->reta_idx_n; ++i)
7141 queue[i] = (*priv->reta_idx)[i];
7142 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7143 &attr, items, actions, false, &error);
7150 * Enable a flow control configured from the control plane.
7153 * Pointer to Ethernet device.
7155 * An Ethernet flow spec to apply.
7157 * An Ethernet flow mask to apply.
7160 * 0 on success, a negative errno value otherwise and rte_errno is set.
7163 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7164 struct rte_flow_item_eth *eth_spec,
7165 struct rte_flow_item_eth *eth_mask)
7167 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7171 * Create default miss flow rule matching lacp traffic
7174 * Pointer to Ethernet device.
7176 * An Ethernet flow spec to apply.
7179 * 0 on success, a negative errno value otherwise and rte_errno is set.
7182 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7185 * The LACP matching is done by only using ether type since using
7186 * a multicast dst mac causes kernel to give low priority to this flow.
7188 static const struct rte_flow_item_eth lacp_spec = {
7189 .type = RTE_BE16(0x8809),
7191 static const struct rte_flow_item_eth lacp_mask = {
7194 const struct rte_flow_attr attr = {
7197 struct rte_flow_item items[] = {
7199 .type = RTE_FLOW_ITEM_TYPE_ETH,
7204 .type = RTE_FLOW_ITEM_TYPE_END,
7207 struct rte_flow_action actions[] = {
7209 .type = (enum rte_flow_action_type)
7210 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7213 .type = RTE_FLOW_ACTION_TYPE_END,
7216 struct rte_flow_error error;
7217 uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7218 &attr, items, actions,
7229 * @see rte_flow_destroy()
7233 mlx5_flow_destroy(struct rte_eth_dev *dev,
7234 struct rte_flow *flow,
7235 struct rte_flow_error *error __rte_unused)
7237 flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7238 (uintptr_t)(void *)flow);
7243 * Destroy all flows.
7245 * @see rte_flow_flush()
7249 mlx5_flow_flush(struct rte_eth_dev *dev,
7250 struct rte_flow_error *error __rte_unused)
7252 mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7259 * @see rte_flow_isolate()
7263 mlx5_flow_isolate(struct rte_eth_dev *dev,
7265 struct rte_flow_error *error)
7267 struct mlx5_priv *priv = dev->data->dev_private;
7269 if (dev->data->dev_started) {
7270 rte_flow_error_set(error, EBUSY,
7271 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7273 "port must be stopped first");
7276 priv->isolated = !!enable;
7278 dev->dev_ops = &mlx5_dev_ops_isolate;
7280 dev->dev_ops = &mlx5_dev_ops;
7282 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7283 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7291 * @see rte_flow_query()
7295 flow_drv_query(struct rte_eth_dev *dev,
7297 const struct rte_flow_action *actions,
7299 struct rte_flow_error *error)
7301 struct mlx5_priv *priv = dev->data->dev_private;
7302 const struct mlx5_flow_driver_ops *fops;
7303 struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7305 enum mlx5_flow_drv_type ftype;
7308 return rte_flow_error_set(error, ENOENT,
7309 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7311 "invalid flow handle");
7313 ftype = flow->drv_type;
7314 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7315 fops = flow_get_drv_ops(ftype);
7317 return fops->query(dev, flow, actions, data, error);
7323 * @see rte_flow_query()
7327 mlx5_flow_query(struct rte_eth_dev *dev,
7328 struct rte_flow *flow,
7329 const struct rte_flow_action *actions,
7331 struct rte_flow_error *error)
7335 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7343 * Get rte_flow callbacks.
7346 * Pointer to Ethernet device structure.
7348 * Pointer to operation-specific structure.
7353 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7354 const struct rte_flow_ops **ops)
7356 *ops = &mlx5_flow_ops;
7361 * Validate meter policy actions.
7362 * Dispatcher for action type specific validation.
7365 * Pointer to the Ethernet device structure.
7367 * The meter policy action object to validate.
7369 * Attributes of flow to determine steering domain.
7370 * @param[out] is_rss
7372 * @param[out] domain_bitmap
7374 * @param[out] is_def_policy
7375 * Is default policy or not.
7377 * Perform verbose error reporting if not NULL. Initialized in case of
7381 * 0 on success, otherwise negative errno value.
7384 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7385 const struct rte_flow_action *actions[RTE_COLORS],
7386 struct rte_flow_attr *attr,
7388 uint8_t *domain_bitmap,
7389 uint8_t *policy_mode,
7390 struct rte_mtr_error *error)
7392 const struct mlx5_flow_driver_ops *fops;
7394 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7395 return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7396 domain_bitmap, policy_mode, error);
7400 * Destroy the meter table set.
7403 * Pointer to Ethernet device.
7404 * @param[in] mtr_policy
7405 * Meter policy struct.
7408 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7409 struct mlx5_flow_meter_policy *mtr_policy)
7411 const struct mlx5_flow_driver_ops *fops;
7413 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7414 fops->destroy_mtr_acts(dev, mtr_policy);
7418 * Create policy action, lock free,
7419 * (mutex should be acquired by caller).
7420 * Dispatcher for action type specific call.
7423 * Pointer to the Ethernet device structure.
7424 * @param[in] mtr_policy
7425 * Meter policy struct.
7427 * Action specification used to create meter actions.
7429 * Perform verbose error reporting if not NULL. Initialized in case of
7433 * 0 on success, otherwise negative errno value.
7436 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7437 struct mlx5_flow_meter_policy *mtr_policy,
7438 const struct rte_flow_action *actions[RTE_COLORS],
7439 struct rte_mtr_error *error)
7441 const struct mlx5_flow_driver_ops *fops;
7443 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7444 return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7448 * Create policy rules, lock free,
7449 * (mutex should be acquired by caller).
7450 * Dispatcher for action type specific call.
7453 * Pointer to the Ethernet device structure.
7454 * @param[in] mtr_policy
7455 * Meter policy struct.
7458 * 0 on success, -1 otherwise.
7461 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7462 struct mlx5_flow_meter_policy *mtr_policy)
7464 const struct mlx5_flow_driver_ops *fops;
7466 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7467 return fops->create_policy_rules(dev, mtr_policy);
7471 * Destroy policy rules, lock free,
7472 * (mutex should be acquired by caller).
7473 * Dispatcher for action type specific call.
7476 * Pointer to the Ethernet device structure.
7477 * @param[in] mtr_policy
7478 * Meter policy struct.
7481 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7482 struct mlx5_flow_meter_policy *mtr_policy)
7484 const struct mlx5_flow_driver_ops *fops;
7486 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7487 fops->destroy_policy_rules(dev, mtr_policy);
7491 * Destroy the default policy table set.
7494 * Pointer to Ethernet device.
7497 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7499 const struct mlx5_flow_driver_ops *fops;
7501 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7502 fops->destroy_def_policy(dev);
7506 * Destroy the default policy table set.
7509 * Pointer to Ethernet device.
7512 * 0 on success, -1 otherwise.
7515 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7517 const struct mlx5_flow_driver_ops *fops;
7519 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7520 return fops->create_def_policy(dev);
7524 * Create the needed meter and suffix tables.
7527 * Pointer to Ethernet device.
7530 * 0 on success, -1 otherwise.
7533 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7534 struct mlx5_flow_meter_info *fm,
7536 uint8_t domain_bitmap)
7538 const struct mlx5_flow_driver_ops *fops;
7540 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7541 return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7545 * Destroy the meter table set.
7548 * Pointer to Ethernet device.
7550 * Pointer to the meter table set.
7553 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7554 struct mlx5_flow_meter_info *fm)
7556 const struct mlx5_flow_driver_ops *fops;
7558 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7559 fops->destroy_mtr_tbls(dev, fm);
7563 * Destroy the global meter drop table.
7566 * Pointer to Ethernet device.
7569 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7571 const struct mlx5_flow_driver_ops *fops;
7573 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7574 fops->destroy_mtr_drop_tbls(dev);
7578 * Destroy the sub policy table with RX queue.
7581 * Pointer to Ethernet device.
7582 * @param[in] mtr_policy
7583 * Pointer to meter policy table.
7586 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7587 struct mlx5_flow_meter_policy *mtr_policy)
7589 const struct mlx5_flow_driver_ops *fops;
7591 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7592 fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7596 * Allocate the needed aso flow meter id.
7599 * Pointer to Ethernet device.
7602 * Index to aso flow meter on success, NULL otherwise.
7605 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7607 const struct mlx5_flow_driver_ops *fops;
7609 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7610 return fops->create_meter(dev);
7614 * Free the aso flow meter id.
7617 * Pointer to Ethernet device.
7618 * @param[in] mtr_idx
7619 * Index to aso flow meter to be free.
7625 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7627 const struct mlx5_flow_driver_ops *fops;
7629 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7630 fops->free_meter(dev, mtr_idx);
7634 * Allocate a counter.
7637 * Pointer to Ethernet device structure.
7640 * Index to allocated counter on success, 0 otherwise.
7643 mlx5_counter_alloc(struct rte_eth_dev *dev)
7645 const struct mlx5_flow_driver_ops *fops;
7646 struct rte_flow_attr attr = { .transfer = 0 };
7648 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7649 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7650 return fops->counter_alloc(dev);
7653 "port %u counter allocate is not supported.",
7654 dev->data->port_id);
7662 * Pointer to Ethernet device structure.
7664 * Index to counter to be free.
7667 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7669 const struct mlx5_flow_driver_ops *fops;
7670 struct rte_flow_attr attr = { .transfer = 0 };
7672 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7673 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7674 fops->counter_free(dev, cnt);
7678 "port %u counter free is not supported.",
7679 dev->data->port_id);
7683 * Query counter statistics.
7686 * Pointer to Ethernet device structure.
7688 * Index to counter to query.
7690 * Set to clear counter statistics.
7692 * The counter hits packets number to save.
7694 * The counter hits bytes number to save.
7697 * 0 on success, a negative errno value otherwise.
7700 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7701 bool clear, uint64_t *pkts, uint64_t *bytes)
7703 const struct mlx5_flow_driver_ops *fops;
7704 struct rte_flow_attr attr = { .transfer = 0 };
7706 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7707 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7708 return fops->counter_query(dev, cnt, clear, pkts, bytes);
7711 "port %u counter query is not supported.",
7712 dev->data->port_id);
7717 * Allocate a new memory for the counter values wrapped by all the needed
7721 * Pointer to mlx5_dev_ctx_shared object.
7724 * 0 on success, a negative errno value otherwise.
7727 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7729 struct mlx5_devx_mkey_attr mkey_attr;
7730 struct mlx5_counter_stats_mem_mng *mem_mng;
7731 volatile struct flow_counter_stats *raw_data;
7732 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7733 int size = (sizeof(struct flow_counter_stats) *
7734 MLX5_COUNTERS_PER_POOL +
7735 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7736 sizeof(struct mlx5_counter_stats_mem_mng);
7737 size_t pgsize = rte_mem_page_size();
7741 if (pgsize == (size_t)-1) {
7742 DRV_LOG(ERR, "Failed to get mem page size");
7746 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7751 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7752 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7753 mem_mng->umem = mlx5_os_umem_reg(sh->cdev->ctx, mem, size,
7754 IBV_ACCESS_LOCAL_WRITE);
7755 if (!mem_mng->umem) {
7760 memset(&mkey_attr, 0, sizeof(mkey_attr));
7761 mkey_attr.addr = (uintptr_t)mem;
7762 mkey_attr.size = size;
7763 mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7764 mkey_attr.pd = sh->cdev->pdn;
7765 mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7766 mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7767 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->cdev->ctx, &mkey_attr);
7769 mlx5_os_umem_dereg(mem_mng->umem);
7774 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7775 raw_data = (volatile struct flow_counter_stats *)mem;
7776 for (i = 0; i < raws_n; ++i) {
7777 mem_mng->raws[i].mem_mng = mem_mng;
7778 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7780 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7781 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7782 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7784 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7785 sh->cmng.mem_mng = mem_mng;
7790 * Set the statistic memory to the new counter pool.
7793 * Pointer to mlx5_dev_ctx_shared object.
7795 * Pointer to the pool to set the statistic memory.
7798 * 0 on success, a negative errno value otherwise.
7801 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7802 struct mlx5_flow_counter_pool *pool)
7804 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7805 /* Resize statistic memory once used out. */
7806 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7807 mlx5_flow_create_counter_stat_mem_mng(sh)) {
7808 DRV_LOG(ERR, "Cannot resize counter stat mem.");
7811 rte_spinlock_lock(&pool->sl);
7812 pool->raw = cmng->mem_mng->raws + pool->index %
7813 MLX5_CNT_CONTAINER_RESIZE;
7814 rte_spinlock_unlock(&pool->sl);
7815 pool->raw_hw = NULL;
7819 #define MLX5_POOL_QUERY_FREQ_US 1000000
7822 * Set the periodic procedure for triggering asynchronous batch queries for all
7823 * the counter pools.
7826 * Pointer to mlx5_dev_ctx_shared object.
7829 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7831 uint32_t pools_n, us;
7833 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7834 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7835 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7836 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7837 sh->cmng.query_thread_on = 0;
7838 DRV_LOG(ERR, "Cannot reinitialize query alarm");
7840 sh->cmng.query_thread_on = 1;
7845 * The periodic procedure for triggering asynchronous batch queries for all the
7846 * counter pools. This function is probably called by the host thread.
7849 * The parameter for the alarm process.
7852 mlx5_flow_query_alarm(void *arg)
7854 struct mlx5_dev_ctx_shared *sh = arg;
7856 uint16_t pool_index = sh->cmng.pool_index;
7857 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7858 struct mlx5_flow_counter_pool *pool;
7861 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7863 rte_spinlock_lock(&cmng->pool_update_sl);
7864 pool = cmng->pools[pool_index];
7865 n_valid = cmng->n_valid;
7866 rte_spinlock_unlock(&cmng->pool_update_sl);
7867 /* Set the statistic memory to the new created pool. */
7868 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7871 /* There is a pool query in progress. */
7874 LIST_FIRST(&sh->cmng.free_stat_raws);
7876 /* No free counter statistics raw memory. */
7879 * Identify the counters released between query trigger and query
7880 * handle more efficiently. The counter released in this gap period
7881 * should wait for a new round of query as the new arrived packets
7882 * will not be taken into account.
7885 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7886 MLX5_COUNTERS_PER_POOL,
7888 pool->raw_hw->mem_mng->dm->id,
7892 (uint64_t)(uintptr_t)pool);
7894 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7895 " %d", pool->min_dcs->id);
7896 pool->raw_hw = NULL;
7899 LIST_REMOVE(pool->raw_hw, next);
7900 sh->cmng.pending_queries++;
7902 if (pool_index >= n_valid)
7905 sh->cmng.pool_index = pool_index;
7906 mlx5_set_query_alarm(sh);
7910 * Check and callback event for new aged flow in the counter pool
7913 * Pointer to mlx5_dev_ctx_shared object.
7915 * Pointer to Current counter pool.
7918 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7919 struct mlx5_flow_counter_pool *pool)
7921 struct mlx5_priv *priv;
7922 struct mlx5_flow_counter *cnt;
7923 struct mlx5_age_info *age_info;
7924 struct mlx5_age_param *age_param;
7925 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7926 struct mlx5_counter_stats_raw *prev = pool->raw;
7927 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7928 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7929 uint16_t expected = AGE_CANDIDATE;
7932 pool->time_of_last_age_check = curr_time;
7933 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7934 cnt = MLX5_POOL_GET_CNT(pool, i);
7935 age_param = MLX5_CNT_TO_AGE(cnt);
7936 if (__atomic_load_n(&age_param->state,
7937 __ATOMIC_RELAXED) != AGE_CANDIDATE)
7939 if (cur->data[i].hits != prev->data[i].hits) {
7940 __atomic_store_n(&age_param->sec_since_last_hit, 0,
7944 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7946 __ATOMIC_RELAXED) <= age_param->timeout)
7949 * Hold the lock first, or if between the
7950 * state AGE_TMOUT and tailq operation the
7951 * release happened, the release procedure
7952 * may delete a non-existent tailq node.
7954 priv = rte_eth_devices[age_param->port_id].data->dev_private;
7955 age_info = GET_PORT_AGE_INFO(priv);
7956 rte_spinlock_lock(&age_info->aged_sl);
7957 if (__atomic_compare_exchange_n(&age_param->state, &expected,
7960 __ATOMIC_RELAXED)) {
7961 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7962 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7964 rte_spinlock_unlock(&age_info->aged_sl);
7966 mlx5_age_event_prepare(sh);
7970 * Handler for the HW respond about ready values from an asynchronous batch
7971 * query. This function is probably called by the host thread.
7974 * The pointer to the shared device context.
7975 * @param[in] async_id
7976 * The Devx async ID.
7978 * The status of the completion.
7981 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7982 uint64_t async_id, int status)
7984 struct mlx5_flow_counter_pool *pool =
7985 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7986 struct mlx5_counter_stats_raw *raw_to_free;
7987 uint8_t query_gen = pool->query_gen ^ 1;
7988 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7989 enum mlx5_counter_type cnt_type =
7990 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7991 MLX5_COUNTER_TYPE_ORIGIN;
7993 if (unlikely(status)) {
7994 raw_to_free = pool->raw_hw;
7996 raw_to_free = pool->raw;
7998 mlx5_flow_aging_check(sh, pool);
7999 rte_spinlock_lock(&pool->sl);
8000 pool->raw = pool->raw_hw;
8001 rte_spinlock_unlock(&pool->sl);
8002 /* Be sure the new raw counters data is updated in memory. */
8004 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8005 rte_spinlock_lock(&cmng->csl[cnt_type]);
8006 TAILQ_CONCAT(&cmng->counters[cnt_type],
8007 &pool->counters[query_gen], next);
8008 rte_spinlock_unlock(&cmng->csl[cnt_type]);
8011 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8012 pool->raw_hw = NULL;
8013 sh->cmng.pending_queries--;
8017 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8018 const struct flow_grp_info *grp_info,
8019 struct rte_flow_error *error)
8021 if (grp_info->transfer && grp_info->external &&
8022 grp_info->fdb_def_rule) {
8023 if (group == UINT32_MAX)
8024 return rte_flow_error_set
8026 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8028 "group index not supported");
8033 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8038 * Translate the rte_flow group index to HW table value.
8040 * If tunnel offload is disabled, all group ids converted to flow table
8041 * id using the standard method.
8042 * If tunnel offload is enabled, group id can be converted using the
8043 * standard or tunnel conversion method. Group conversion method
8044 * selection depends on flags in `grp_info` parameter:
8045 * - Internal (grp_info.external == 0) groups conversion uses the
8047 * - Group ids in JUMP action converted with the tunnel conversion.
8048 * - Group id in rule attribute conversion depends on a rule type and
8050 * ** non zero group attributes converted with the tunnel method
8051 * ** zero group attribute in non-tunnel rule is converted using the
8052 * standard method - there's only one root table
8053 * ** zero group attribute in steer tunnel rule is converted with the
8054 * standard method - single root table
8055 * ** zero group attribute in match tunnel rule is a special OvS
8056 * case: that value is used for portability reasons. That group
8057 * id is converted with the tunnel conversion method.
8062 * PMD tunnel offload object
8064 * rte_flow group index value.
8067 * @param[in] grp_info
8068 * flags used for conversion
8070 * Pointer to error structure.
8073 * 0 on success, a negative errno value otherwise and rte_errno is set.
8076 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8077 const struct mlx5_flow_tunnel *tunnel,
8078 uint32_t group, uint32_t *table,
8079 const struct flow_grp_info *grp_info,
8080 struct rte_flow_error *error)
8083 bool standard_translation;
8085 if (!grp_info->skip_scale && grp_info->external &&
8086 group < MLX5_MAX_TABLES_EXTERNAL)
8087 group *= MLX5_FLOW_TABLE_FACTOR;
8088 if (is_tunnel_offload_active(dev)) {
8089 standard_translation = !grp_info->external ||
8090 grp_info->std_tbl_fix;
8092 standard_translation = true;
8095 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8096 dev->data->port_id, group, grp_info->transfer,
8097 grp_info->external, grp_info->fdb_def_rule,
8098 standard_translation ? "STANDARD" : "TUNNEL");
8099 if (standard_translation)
8100 ret = flow_group_to_table(dev->data->port_id, group, table,
8103 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8110 * Discover availability of metadata reg_c's.
8112 * Iteratively use test flows to check availability.
8115 * Pointer to the Ethernet device structure.
8118 * 0 on success, a negative errno value otherwise and rte_errno is set.
8121 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8123 struct mlx5_priv *priv = dev->data->dev_private;
8124 enum modify_reg idx;
8127 /* reg_c[0] and reg_c[1] are reserved. */
8128 priv->sh->flow_mreg_c[n++] = REG_C_0;
8129 priv->sh->flow_mreg_c[n++] = REG_C_1;
8130 /* Discover availability of other reg_c's. */
8131 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8132 struct rte_flow_attr attr = {
8133 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8134 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8137 struct rte_flow_item items[] = {
8139 .type = RTE_FLOW_ITEM_TYPE_END,
8142 struct rte_flow_action actions[] = {
8144 .type = (enum rte_flow_action_type)
8145 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8146 .conf = &(struct mlx5_flow_action_copy_mreg){
8152 .type = RTE_FLOW_ACTION_TYPE_JUMP,
8153 .conf = &(struct rte_flow_action_jump){
8154 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8158 .type = RTE_FLOW_ACTION_TYPE_END,
8162 struct rte_flow *flow;
8163 struct rte_flow_error error;
8165 if (!priv->config.dv_flow_en)
8167 /* Create internal flow, validation skips copy action. */
8168 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8169 items, actions, false, &error);
8170 flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8174 priv->sh->flow_mreg_c[n++] = idx;
8175 flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8177 for (; n < MLX5_MREG_C_NUM; ++n)
8178 priv->sh->flow_mreg_c[n] = REG_NON;
8179 priv->sh->metadata_regc_check_flag = 1;
8184 save_dump_file(const uint8_t *data, uint32_t size,
8185 uint32_t type, uint64_t id, void *arg, FILE *file)
8187 char line[BUF_SIZE];
8190 uint32_t actions_num;
8191 struct rte_flow_query_count *count;
8193 memset(line, 0, BUF_SIZE);
8195 case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8196 actions_num = *(uint32_t *)(arg);
8197 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8198 type, id, actions_num);
8200 case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8201 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8204 case DR_DUMP_REC_TYPE_PMD_COUNTER:
8205 count = (struct rte_flow_query_count *)arg;
8207 "%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8208 type, id, count->hits, count->bytes);
8214 for (k = 0; k < size; k++) {
8215 /* Make sure we do not overrun the line buffer length. */
8216 if (out >= BUF_SIZE - 4) {
8220 out += snprintf(line + out, BUF_SIZE - out, "%02x",
8223 fprintf(file, "%s\n", line);
8228 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8229 struct rte_flow_query_count *count, struct rte_flow_error *error)
8231 struct rte_flow_action action[2];
8232 enum mlx5_flow_drv_type ftype;
8233 const struct mlx5_flow_driver_ops *fops;
8236 return rte_flow_error_set(error, ENOENT,
8237 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8239 "invalid flow handle");
8241 action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8242 action[1].type = RTE_FLOW_ACTION_TYPE_END;
8243 if (flow->counter) {
8244 memset(count, 0, sizeof(struct rte_flow_query_count));
8245 ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8246 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8247 ftype < MLX5_FLOW_TYPE_MAX);
8248 fops = flow_get_drv_ops(ftype);
8249 return fops->query(dev, flow, action, count, error);
8254 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8256 * Dump flow ipool data to file
8259 * The pointer to Ethernet device.
8261 * A pointer to a file for output.
8263 * Perform verbose error reporting if not NULL. PMDs initialize this
8264 * structure in case of error only.
8266 * 0 on success, a negative value otherwise.
8269 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8270 struct rte_flow *flow, FILE *file,
8271 struct rte_flow_error *error)
8273 struct mlx5_priv *priv = dev->data->dev_private;
8274 struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
8275 struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8276 uint32_t handle_idx;
8277 struct mlx5_flow_handle *dh;
8278 struct rte_flow_query_count count;
8279 uint32_t actions_num;
8280 const uint8_t *data;
8284 void *action = NULL;
8287 return rte_flow_error_set(error, ENOENT,
8288 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8290 "invalid flow handle");
8292 handle_idx = flow->dev_handles;
8293 while (handle_idx) {
8294 dh = mlx5_ipool_get(priv->sh->ipool
8295 [MLX5_IPOOL_MLX5_FLOW], handle_idx);
8298 handle_idx = dh->next.next;
8301 type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8302 flow_dv_query_count_ptr(dev, flow->counter,
8305 id = (uint64_t)(uintptr_t)action;
8306 if (!mlx5_flow_query_counter(dev, flow, &count, error))
8307 save_dump_file(NULL, 0, type,
8308 id, (void *)&count, file);
8310 /* Get modify_hdr and encap_decap buf from ipools. */
8312 modify_hdr = dh->dvh.modify_hdr;
8314 if (dh->dvh.rix_encap_decap) {
8315 encap_decap = mlx5_ipool_get(priv->sh->ipool
8316 [MLX5_IPOOL_DECAP_ENCAP],
8317 dh->dvh.rix_encap_decap);
8320 data = (const uint8_t *)modify_hdr->actions;
8321 size = (size_t)(modify_hdr->actions_num) * 8;
8322 id = (uint64_t)(uintptr_t)modify_hdr->action;
8323 actions_num = modify_hdr->actions_num;
8324 type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8325 save_dump_file(data, size, type, id,
8326 (void *)(&actions_num), file);
8329 data = encap_decap->buf;
8330 size = encap_decap->size;
8331 id = (uint64_t)(uintptr_t)encap_decap->action;
8332 type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8333 save_dump_file(data, size, type,
8341 * Dump all flow's encap_decap/modify_hdr/counter data to file
8344 * The pointer to Ethernet device.
8346 * A pointer to a file for output.
8348 * Perform verbose error reporting if not NULL. PMDs initialize this
8349 * structure in case of error only.
8351 * 0 on success, a negative value otherwise.
8354 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8355 FILE *file, struct rte_flow_error *error)
8357 struct mlx5_priv *priv = dev->data->dev_private;
8358 struct mlx5_dev_ctx_shared *sh = priv->sh;
8359 struct mlx5_hlist *h;
8360 struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
8361 struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8362 struct rte_flow_query_count count;
8363 uint32_t actions_num;
8364 const uint8_t *data;
8370 struct mlx5_list_inconst *l_inconst;
8371 struct mlx5_list_entry *e;
8373 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8377 /* encap_decap hlist is lcore_share, get global core cache. */
8378 i = MLX5_LIST_GLOBAL;
8379 h = sh->encaps_decaps;
8381 for (j = 0; j <= h->mask; j++) {
8382 l_inconst = &h->buckets[j].l;
8383 if (!l_inconst || !l_inconst->cache[i])
8386 e = LIST_FIRST(&l_inconst->cache[i]->h);
8389 (struct mlx5_flow_dv_encap_decap_resource *)e;
8390 data = encap_decap->buf;
8391 size = encap_decap->size;
8392 id = (uint64_t)(uintptr_t)encap_decap->action;
8393 type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8394 save_dump_file(data, size, type,
8396 e = LIST_NEXT(e, next);
8401 /* get modify_hdr */
8402 h = sh->modify_cmds;
8404 lcore_index = rte_lcore_index(rte_lcore_id());
8405 if (unlikely(lcore_index == -1)) {
8406 lcore_index = MLX5_LIST_NLCORE;
8407 rte_spinlock_lock(&h->l_const.lcore_lock);
8411 for (j = 0; j <= h->mask; j++) {
8412 l_inconst = &h->buckets[j].l;
8413 if (!l_inconst || !l_inconst->cache[i])
8416 e = LIST_FIRST(&l_inconst->cache[i]->h);
8419 (struct mlx5_flow_dv_modify_hdr_resource *)e;
8420 data = (const uint8_t *)modify_hdr->actions;
8421 size = (size_t)(modify_hdr->actions_num) * 8;
8422 actions_num = modify_hdr->actions_num;
8423 id = (uint64_t)(uintptr_t)modify_hdr->action;
8424 type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8425 save_dump_file(data, size, type, id,
8426 (void *)(&actions_num), file);
8427 e = LIST_NEXT(e, next);
8431 if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8432 rte_spinlock_unlock(&h->l_const.lcore_lock);
8436 MLX5_ASSERT(cmng->n_valid <= cmng->n);
8437 max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8438 for (j = 1; j <= max; j++) {
8440 flow_dv_query_count_ptr(dev, j, &action, error);
8442 if (!flow_dv_query_count(dev, j, &count, error)) {
8443 type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8444 id = (uint64_t)(uintptr_t)action;
8445 save_dump_file(NULL, 0, type,
8446 id, (void *)&count, file);
8455 * Dump flow raw hw data to file
8458 * The pointer to Ethernet device.
8460 * A pointer to a file for output.
8462 * Perform verbose error reporting if not NULL. PMDs initialize this
8463 * structure in case of error only.
8465 * 0 on success, a nagative value otherwise.
8468 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8470 struct rte_flow_error *error __rte_unused)
8472 struct mlx5_priv *priv = dev->data->dev_private;
8473 struct mlx5_dev_ctx_shared *sh = priv->sh;
8474 uint32_t handle_idx;
8476 struct mlx5_flow_handle *dh;
8477 struct rte_flow *flow;
8479 if (!priv->config.dv_flow_en) {
8480 if (fputs("device dv flow disabled\n", file) <= 0)
8487 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8488 if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8491 return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8493 sh->tx_domain, file);
8496 flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8497 (uintptr_t)(void *)flow_idx);
8501 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8502 mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8504 handle_idx = flow->dev_handles;
8505 while (handle_idx) {
8506 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8511 ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8516 handle_idx = dh->next.next;
8522 * Get aged-out flows.
8525 * Pointer to the Ethernet device structure.
8526 * @param[in] context
8527 * The address of an array of pointers to the aged-out flows contexts.
8528 * @param[in] nb_countexts
8529 * The length of context array pointers.
8531 * Perform verbose error reporting if not NULL. Initialized in case of
8535 * how many contexts get in success, otherwise negative errno value.
8536 * if nb_contexts is 0, return the amount of all aged contexts.
8537 * if nb_contexts is not 0 , return the amount of aged flows reported
8538 * in the context array.
8541 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8542 uint32_t nb_contexts, struct rte_flow_error *error)
8544 const struct mlx5_flow_driver_ops *fops;
8545 struct rte_flow_attr attr = { .transfer = 0 };
8547 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8548 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8549 return fops->get_aged_flows(dev, contexts, nb_contexts,
8553 "port %u get aged flows is not supported.",
8554 dev->data->port_id);
8558 /* Wrapper for driver action_validate op callback */
8560 flow_drv_action_validate(struct rte_eth_dev *dev,
8561 const struct rte_flow_indir_action_conf *conf,
8562 const struct rte_flow_action *action,
8563 const struct mlx5_flow_driver_ops *fops,
8564 struct rte_flow_error *error)
8566 static const char err_msg[] = "indirect action validation unsupported";
8568 if (!fops->action_validate) {
8569 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8570 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8574 return fops->action_validate(dev, conf, action, error);
8578 * Destroys the shared action by handle.
8581 * Pointer to Ethernet device structure.
8583 * Handle for the indirect action object to be destroyed.
8585 * Perform verbose error reporting if not NULL. PMDs initialize this
8586 * structure in case of error only.
8589 * 0 on success, a negative errno value otherwise and rte_errno is set.
8591 * @note: wrapper for driver action_create op callback.
8594 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8595 struct rte_flow_action_handle *handle,
8596 struct rte_flow_error *error)
8598 static const char err_msg[] = "indirect action destruction unsupported";
8599 struct rte_flow_attr attr = { .transfer = 0 };
8600 const struct mlx5_flow_driver_ops *fops =
8601 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8603 if (!fops->action_destroy) {
8604 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8605 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8609 return fops->action_destroy(dev, handle, error);
8612 /* Wrapper for driver action_destroy op callback */
8614 flow_drv_action_update(struct rte_eth_dev *dev,
8615 struct rte_flow_action_handle *handle,
8617 const struct mlx5_flow_driver_ops *fops,
8618 struct rte_flow_error *error)
8620 static const char err_msg[] = "indirect action update unsupported";
8622 if (!fops->action_update) {
8623 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8624 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8628 return fops->action_update(dev, handle, update, error);
8631 /* Wrapper for driver action_destroy op callback */
8633 flow_drv_action_query(struct rte_eth_dev *dev,
8634 const struct rte_flow_action_handle *handle,
8636 const struct mlx5_flow_driver_ops *fops,
8637 struct rte_flow_error *error)
8639 static const char err_msg[] = "indirect action query unsupported";
8641 if (!fops->action_query) {
8642 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8643 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8647 return fops->action_query(dev, handle, data, error);
8651 * Create indirect action for reuse in multiple flow rules.
8654 * Pointer to Ethernet device structure.
8656 * Pointer to indirect action object configuration.
8658 * Action configuration for indirect action object creation.
8660 * Perform verbose error reporting if not NULL. PMDs initialize this
8661 * structure in case of error only.
8663 * A valid handle in case of success, NULL otherwise and rte_errno is set.
8665 static struct rte_flow_action_handle *
8666 mlx5_action_handle_create(struct rte_eth_dev *dev,
8667 const struct rte_flow_indir_action_conf *conf,
8668 const struct rte_flow_action *action,
8669 struct rte_flow_error *error)
8671 static const char err_msg[] = "indirect action creation unsupported";
8672 struct rte_flow_attr attr = { .transfer = 0 };
8673 const struct mlx5_flow_driver_ops *fops =
8674 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8676 if (flow_drv_action_validate(dev, conf, action, fops, error))
8678 if (!fops->action_create) {
8679 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8680 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8684 return fops->action_create(dev, conf, action, error);
8688 * Updates inplace the indirect action configuration pointed by *handle*
8689 * with the configuration provided as *update* argument.
8690 * The update of the indirect action configuration effects all flow rules
8691 * reusing the action via handle.
8694 * Pointer to Ethernet device structure.
8696 * Handle for the indirect action to be updated.
8698 * Action specification used to modify the action pointed by handle.
8699 * *update* could be of same type with the action pointed by the *handle*
8700 * handle argument, or some other structures like a wrapper, depending on
8701 * the indirect action type.
8703 * Perform verbose error reporting if not NULL. PMDs initialize this
8704 * structure in case of error only.
8707 * 0 on success, a negative errno value otherwise and rte_errno is set.
8710 mlx5_action_handle_update(struct rte_eth_dev *dev,
8711 struct rte_flow_action_handle *handle,
8713 struct rte_flow_error *error)
8715 struct rte_flow_attr attr = { .transfer = 0 };
8716 const struct mlx5_flow_driver_ops *fops =
8717 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8720 ret = flow_drv_action_validate(dev, NULL,
8721 (const struct rte_flow_action *)update, fops, error);
8724 return flow_drv_action_update(dev, handle, update, fops,
8729 * Query the indirect action by handle.
8731 * This function allows retrieving action-specific data such as counters.
8732 * Data is gathered by special action which may be present/referenced in
8733 * more than one flow rule definition.
8735 * see @RTE_FLOW_ACTION_TYPE_COUNT
8738 * Pointer to Ethernet device structure.
8740 * Handle for the indirect action to query.
8741 * @param[in, out] data
8742 * Pointer to storage for the associated query data type.
8744 * Perform verbose error reporting if not NULL. PMDs initialize this
8745 * structure in case of error only.
8748 * 0 on success, a negative errno value otherwise and rte_errno is set.
8751 mlx5_action_handle_query(struct rte_eth_dev *dev,
8752 const struct rte_flow_action_handle *handle,
8754 struct rte_flow_error *error)
8756 struct rte_flow_attr attr = { .transfer = 0 };
8757 const struct mlx5_flow_driver_ops *fops =
8758 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8760 return flow_drv_action_query(dev, handle, data, fops, error);
8764 * Destroy all indirect actions (shared RSS).
8767 * Pointer to Ethernet device.
8770 * 0 on success, a negative errno value otherwise and rte_errno is set.
8773 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8775 struct rte_flow_error error;
8776 struct mlx5_priv *priv = dev->data->dev_private;
8777 struct mlx5_shared_action_rss *shared_rss;
8781 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8782 priv->rss_shared_actions, idx, shared_rss, next) {
8783 ret |= mlx5_action_handle_destroy(dev,
8784 (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8789 #ifndef HAVE_MLX5DV_DR
8790 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8792 #define MLX5_DOMAIN_SYNC_FLOW \
8793 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8796 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8798 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8799 const struct mlx5_flow_driver_ops *fops;
8801 struct rte_flow_attr attr = { .transfer = 0 };
8803 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8804 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8810 const struct mlx5_flow_tunnel *
8811 mlx5_get_tof(const struct rte_flow_item *item,
8812 const struct rte_flow_action *action,
8813 enum mlx5_tof_rule_type *rule_type)
8815 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8816 if (item->type == (typeof(item->type))
8817 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8818 *rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8819 return flow_items_to_tunnel(item);
8822 for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8823 if (action->type == (typeof(action->type))
8824 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8825 *rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8826 return flow_actions_to_tunnel(action);
8833 * tunnel offload functionalilty is defined for DV environment only
8835 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8837 union tunnel_offload_mark {
8840 uint32_t app_reserve:8;
8841 uint32_t table_id:15;
8842 uint32_t transfer:1;
8843 uint32_t _unused_:8;
8848 mlx5_access_tunnel_offload_db
8849 (struct rte_eth_dev *dev,
8850 bool (*match)(struct rte_eth_dev *,
8851 struct mlx5_flow_tunnel *, const void *),
8852 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8853 void (*miss)(struct rte_eth_dev *, void *),
8854 void *ctx, bool lock_op);
8857 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8858 struct rte_flow *flow,
8859 const struct rte_flow_attr *attr,
8860 const struct rte_flow_action *app_actions,
8862 const struct mlx5_flow_tunnel *tunnel,
8863 struct tunnel_default_miss_ctx *ctx,
8864 struct rte_flow_error *error)
8866 struct mlx5_priv *priv = dev->data->dev_private;
8867 struct mlx5_flow *dev_flow;
8868 struct rte_flow_attr miss_attr = *attr;
8869 const struct rte_flow_item miss_items[2] = {
8871 .type = RTE_FLOW_ITEM_TYPE_ETH,
8877 .type = RTE_FLOW_ITEM_TYPE_END,
8883 union tunnel_offload_mark mark_id;
8884 struct rte_flow_action_mark miss_mark;
8885 struct rte_flow_action miss_actions[3] = {
8886 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8887 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
8889 const struct rte_flow_action_jump *jump_data;
8890 uint32_t i, flow_table = 0; /* prevent compilation warning */
8891 struct flow_grp_info grp_info = {
8893 .transfer = attr->transfer,
8894 .fdb_def_rule = !!priv->fdb_def_rule,
8899 if (!attr->transfer) {
8902 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8903 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8904 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8907 return rte_flow_error_set
8909 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8910 NULL, "invalid default miss RSS");
8911 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8912 ctx->action_rss.level = 0,
8913 ctx->action_rss.types = priv->rss_conf.rss_hf,
8914 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8915 ctx->action_rss.queue_num = priv->reta_idx_n,
8916 ctx->action_rss.key = priv->rss_conf.rss_key,
8917 ctx->action_rss.queue = ctx->queue;
8918 if (!priv->reta_idx_n || !priv->rxqs_n)
8919 return rte_flow_error_set
8921 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8922 NULL, "invalid port configuration");
8923 if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
8924 ctx->action_rss.types = 0;
8925 for (i = 0; i != priv->reta_idx_n; ++i)
8926 ctx->queue[i] = (*priv->reta_idx)[i];
8928 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8929 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8931 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8932 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8933 jump_data = app_actions->conf;
8934 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8935 miss_attr.group = jump_data->group;
8936 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8937 &flow_table, &grp_info, error);
8939 return rte_flow_error_set(error, EINVAL,
8940 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8941 NULL, "invalid tunnel id");
8942 mark_id.app_reserve = 0;
8943 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8944 mark_id.transfer = !!attr->transfer;
8945 mark_id._unused_ = 0;
8946 miss_mark.id = mark_id.val;
8947 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8948 miss_items, miss_actions, flow_idx, error);
8951 dev_flow->flow = flow;
8952 dev_flow->external = true;
8953 dev_flow->tunnel = tunnel;
8954 dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8955 /* Subflow object was created, we must include one in the list. */
8956 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8957 dev_flow->handle, next);
8959 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8960 dev->data->port_id, tunnel->app_tunnel.type,
8961 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8962 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8963 miss_actions, error);
8965 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8971 static const struct mlx5_flow_tbl_data_entry *
8972 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8974 struct mlx5_priv *priv = dev->data->dev_private;
8975 struct mlx5_dev_ctx_shared *sh = priv->sh;
8976 struct mlx5_list_entry *he;
8977 union tunnel_offload_mark mbits = { .val = mark };
8978 union mlx5_flow_tbl_key table_key = {
8980 .level = tunnel_id_to_flow_tbl(mbits.table_id),
8984 .is_fdb = !!mbits.transfer,
8988 struct mlx5_flow_cb_ctx ctx = {
8989 .data = &table_key.v64,
8992 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
8994 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8998 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
8999 struct mlx5_list_entry *entry)
9001 struct mlx5_dev_ctx_shared *sh = tool_ctx;
9002 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9004 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9005 tunnel_flow_tbl_to_id(tte->flow_table));
9010 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9011 struct mlx5_list_entry *entry, void *cb_ctx)
9013 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9014 union tunnel_tbl_key tbl = {
9015 .val = *(uint64_t *)(ctx->data),
9017 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9019 return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9022 static struct mlx5_list_entry *
9023 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9025 struct mlx5_dev_ctx_shared *sh = tool_ctx;
9026 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9027 struct tunnel_tbl_entry *tte;
9028 union tunnel_tbl_key tbl = {
9029 .val = *(uint64_t *)(ctx->data),
9032 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9037 mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9039 if (tte->flow_table >= MLX5_MAX_TABLES) {
9040 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9042 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9045 } else if (!tte->flow_table) {
9048 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9049 tte->tunnel_id = tbl.tunnel_id;
9050 tte->group = tbl.group;
9058 static struct mlx5_list_entry *
9059 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9060 struct mlx5_list_entry *oentry,
9061 void *cb_ctx __rte_unused)
9063 struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9068 memcpy(tte, oentry, sizeof(*tte));
9073 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9074 struct mlx5_list_entry *entry)
9076 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9082 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9083 const struct mlx5_flow_tunnel *tunnel,
9084 uint32_t group, uint32_t *table,
9085 struct rte_flow_error *error)
9087 struct mlx5_list_entry *he;
9088 struct tunnel_tbl_entry *tte;
9089 union tunnel_tbl_key key = {
9090 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9093 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9094 struct mlx5_hlist *group_hash;
9095 struct mlx5_flow_cb_ctx ctx = {
9099 group_hash = tunnel ? tunnel->groups : thub->groups;
9100 he = mlx5_hlist_register(group_hash, key.val, &ctx);
9102 return rte_flow_error_set(error, EINVAL,
9103 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9105 "tunnel group index not supported");
9106 tte = container_of(he, typeof(*tte), hash);
9107 *table = tte->flow_table;
9108 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9109 dev->data->port_id, key.tunnel_id, group, *table);
9114 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9115 struct mlx5_flow_tunnel *tunnel)
9117 struct mlx5_priv *priv = dev->data->dev_private;
9118 struct mlx5_indexed_pool *ipool;
9120 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9121 dev->data->port_id, tunnel->tunnel_id);
9122 LIST_REMOVE(tunnel, chain);
9123 mlx5_hlist_destroy(tunnel->groups);
9124 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9125 mlx5_ipool_free(ipool, tunnel->tunnel_id);
9129 mlx5_access_tunnel_offload_db
9130 (struct rte_eth_dev *dev,
9131 bool (*match)(struct rte_eth_dev *,
9132 struct mlx5_flow_tunnel *, const void *),
9133 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9134 void (*miss)(struct rte_eth_dev *, void *),
9135 void *ctx, bool lock_op)
9137 bool verdict = false;
9138 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9139 struct mlx5_flow_tunnel *tunnel;
9141 rte_spinlock_lock(&thub->sl);
9142 LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9143 verdict = match(dev, tunnel, (const void *)ctx);
9148 rte_spinlock_unlock(&thub->sl);
9150 hit(dev, tunnel, ctx);
9151 if (!verdict && miss)
9154 rte_spinlock_unlock(&thub->sl);
9159 struct tunnel_db_find_tunnel_id_ctx {
9161 struct mlx5_flow_tunnel *tunnel;
9165 find_tunnel_id_match(struct rte_eth_dev *dev,
9166 struct mlx5_flow_tunnel *tunnel, const void *x)
9168 const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9171 return tunnel->tunnel_id == ctx->tunnel_id;
9175 find_tunnel_id_hit(struct rte_eth_dev *dev,
9176 struct mlx5_flow_tunnel *tunnel, void *x)
9178 struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9180 ctx->tunnel = tunnel;
9183 static struct mlx5_flow_tunnel *
9184 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9186 struct tunnel_db_find_tunnel_id_ctx ctx = {
9190 mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9191 find_tunnel_id_hit, NULL, &ctx, true);
9196 static struct mlx5_flow_tunnel *
9197 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9198 const struct rte_flow_tunnel *app_tunnel)
9200 struct mlx5_priv *priv = dev->data->dev_private;
9201 struct mlx5_indexed_pool *ipool;
9202 struct mlx5_flow_tunnel *tunnel;
9205 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9206 tunnel = mlx5_ipool_zmalloc(ipool, &id);
9209 if (id >= MLX5_MAX_TUNNELS) {
9210 mlx5_ipool_free(ipool, id);
9211 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9214 tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9216 mlx5_flow_tunnel_grp2tbl_create_cb,
9217 mlx5_flow_tunnel_grp2tbl_match_cb,
9218 mlx5_flow_tunnel_grp2tbl_remove_cb,
9219 mlx5_flow_tunnel_grp2tbl_clone_cb,
9220 mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9221 if (!tunnel->groups) {
9222 mlx5_ipool_free(ipool, id);
9225 /* initiate new PMD tunnel */
9226 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9227 tunnel->tunnel_id = id;
9228 tunnel->action.type = (typeof(tunnel->action.type))
9229 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9230 tunnel->action.conf = tunnel;
9231 tunnel->item.type = (typeof(tunnel->item.type))
9232 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9233 tunnel->item.spec = tunnel;
9234 tunnel->item.last = NULL;
9235 tunnel->item.mask = NULL;
9237 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9238 dev->data->port_id, tunnel->tunnel_id);
9243 struct tunnel_db_get_tunnel_ctx {
9244 const struct rte_flow_tunnel *app_tunnel;
9245 struct mlx5_flow_tunnel *tunnel;
9248 static bool get_tunnel_match(struct rte_eth_dev *dev,
9249 struct mlx5_flow_tunnel *tunnel, const void *x)
9251 const struct tunnel_db_get_tunnel_ctx *ctx = x;
9254 return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9255 sizeof(*ctx->app_tunnel));
9258 static void get_tunnel_hit(struct rte_eth_dev *dev,
9259 struct mlx5_flow_tunnel *tunnel, void *x)
9261 /* called under tunnel spinlock protection */
9262 struct tunnel_db_get_tunnel_ctx *ctx = x;
9266 ctx->tunnel = tunnel;
9269 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9271 /* called under tunnel spinlock protection */
9272 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9273 struct tunnel_db_get_tunnel_ctx *ctx = x;
9275 rte_spinlock_unlock(&thub->sl);
9276 ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9277 rte_spinlock_lock(&thub->sl);
9279 ctx->tunnel->refctn = 1;
9280 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9286 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9287 const struct rte_flow_tunnel *app_tunnel,
9288 struct mlx5_flow_tunnel **tunnel)
9290 struct tunnel_db_get_tunnel_ctx ctx = {
9291 .app_tunnel = app_tunnel,
9294 mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9295 get_tunnel_miss, &ctx, true);
9296 *tunnel = ctx.tunnel;
9297 return ctx.tunnel ? 0 : -ENOMEM;
9300 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9302 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9306 if (!LIST_EMPTY(&thub->tunnels))
9307 DRV_LOG(WARNING, "port %u tunnels present", port_id);
9308 mlx5_hlist_destroy(thub->groups);
9312 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9315 struct mlx5_flow_tunnel_hub *thub;
9317 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9321 LIST_INIT(&thub->tunnels);
9322 rte_spinlock_init(&thub->sl);
9323 thub->groups = mlx5_hlist_create("flow groups", 64,
9325 mlx5_flow_tunnel_grp2tbl_create_cb,
9326 mlx5_flow_tunnel_grp2tbl_match_cb,
9327 mlx5_flow_tunnel_grp2tbl_remove_cb,
9328 mlx5_flow_tunnel_grp2tbl_clone_cb,
9329 mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9330 if (!thub->groups) {
9334 sh->tunnel_hub = thub;
9340 mlx5_hlist_destroy(thub->groups);
9347 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9348 struct rte_flow_tunnel *tunnel,
9349 const char *err_msg)
9352 if (!is_tunnel_offload_active(dev)) {
9353 err_msg = "tunnel offload was not activated";
9355 } else if (!tunnel) {
9356 err_msg = "no application tunnel";
9360 switch (tunnel->type) {
9362 err_msg = "unsupported tunnel type";
9364 case RTE_FLOW_ITEM_TYPE_VXLAN:
9365 case RTE_FLOW_ITEM_TYPE_GRE:
9366 case RTE_FLOW_ITEM_TYPE_NVGRE:
9367 case RTE_FLOW_ITEM_TYPE_GENEVE:
9376 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9377 struct rte_flow_tunnel *app_tunnel,
9378 struct rte_flow_action **actions,
9379 uint32_t *num_of_actions,
9380 struct rte_flow_error *error)
9383 struct mlx5_flow_tunnel *tunnel;
9384 const char *err_msg = NULL;
9385 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9388 return rte_flow_error_set(error, EINVAL,
9389 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9391 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9393 return rte_flow_error_set(error, ret,
9394 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9395 "failed to initialize pmd tunnel");
9397 *actions = &tunnel->action;
9398 *num_of_actions = 1;
9403 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9404 struct rte_flow_tunnel *app_tunnel,
9405 struct rte_flow_item **items,
9406 uint32_t *num_of_items,
9407 struct rte_flow_error *error)
9410 struct mlx5_flow_tunnel *tunnel;
9411 const char *err_msg = NULL;
9412 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9415 return rte_flow_error_set(error, EINVAL,
9416 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9418 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9420 return rte_flow_error_set(error, ret,
9421 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9422 "failed to initialize pmd tunnel");
9424 *items = &tunnel->item;
9429 struct tunnel_db_element_release_ctx {
9430 struct rte_flow_item *items;
9431 struct rte_flow_action *actions;
9432 uint32_t num_elements;
9433 struct rte_flow_error *error;
9438 tunnel_element_release_match(struct rte_eth_dev *dev,
9439 struct mlx5_flow_tunnel *tunnel, const void *x)
9441 const struct tunnel_db_element_release_ctx *ctx = x;
9444 if (ctx->num_elements != 1)
9446 else if (ctx->items)
9447 return ctx->items == &tunnel->item;
9448 else if (ctx->actions)
9449 return ctx->actions == &tunnel->action;
9455 tunnel_element_release_hit(struct rte_eth_dev *dev,
9456 struct mlx5_flow_tunnel *tunnel, void *x)
9458 struct tunnel_db_element_release_ctx *ctx = x;
9460 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9461 mlx5_flow_tunnel_free(dev, tunnel);
9465 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9467 struct tunnel_db_element_release_ctx *ctx = x;
9469 ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9470 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9471 "invalid argument");
9475 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9476 struct rte_flow_item *pmd_items,
9477 uint32_t num_items, struct rte_flow_error *err)
9479 struct tunnel_db_element_release_ctx ctx = {
9482 .num_elements = num_items,
9486 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9487 tunnel_element_release_hit,
9488 tunnel_element_release_miss, &ctx, false);
9494 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9495 struct rte_flow_action *pmd_actions,
9496 uint32_t num_actions, struct rte_flow_error *err)
9498 struct tunnel_db_element_release_ctx ctx = {
9500 .actions = pmd_actions,
9501 .num_elements = num_actions,
9505 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9506 tunnel_element_release_hit,
9507 tunnel_element_release_miss, &ctx, false);
9513 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9515 struct rte_flow_restore_info *info,
9516 struct rte_flow_error *err)
9518 uint64_t ol_flags = m->ol_flags;
9519 const struct mlx5_flow_tbl_data_entry *tble;
9520 const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9522 if (!is_tunnel_offload_active(dev)) {
9527 if ((ol_flags & mask) != mask)
9529 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9531 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9532 dev->data->port_id, m->hash.fdir.hi);
9535 MLX5_ASSERT(tble->tunnel);
9536 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9537 info->group_id = tble->group_id;
9538 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9539 RTE_FLOW_RESTORE_INFO_GROUP_ID |
9540 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9545 return rte_flow_error_set(err, EINVAL,
9546 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9547 "failed to get restore info");
9550 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9552 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9553 __rte_unused struct rte_flow_tunnel *app_tunnel,
9554 __rte_unused struct rte_flow_action **actions,
9555 __rte_unused uint32_t *num_of_actions,
9556 __rte_unused struct rte_flow_error *error)
9562 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9563 __rte_unused struct rte_flow_tunnel *app_tunnel,
9564 __rte_unused struct rte_flow_item **items,
9565 __rte_unused uint32_t *num_of_items,
9566 __rte_unused struct rte_flow_error *error)
9572 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9573 __rte_unused struct rte_flow_item *pmd_items,
9574 __rte_unused uint32_t num_items,
9575 __rte_unused struct rte_flow_error *err)
9581 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9582 __rte_unused struct rte_flow_action *pmd_action,
9583 __rte_unused uint32_t num_actions,
9584 __rte_unused struct rte_flow_error *err)
9590 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9591 __rte_unused struct rte_mbuf *m,
9592 __rte_unused struct rte_flow_restore_info *i,
9593 __rte_unused struct rte_flow_error *err)
9599 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9600 __rte_unused struct rte_flow *flow,
9601 __rte_unused const struct rte_flow_attr *attr,
9602 __rte_unused const struct rte_flow_action *actions,
9603 __rte_unused uint32_t flow_idx,
9604 __rte_unused const struct mlx5_flow_tunnel *tunnel,
9605 __rte_unused struct tunnel_default_miss_ctx *ctx,
9606 __rte_unused struct rte_flow_error *error)
9611 static struct mlx5_flow_tunnel *
9612 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9613 __rte_unused uint32_t id)
9619 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9620 __rte_unused struct mlx5_flow_tunnel *tunnel)
9625 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9626 __rte_unused const struct mlx5_flow_tunnel *t,
9627 __rte_unused uint32_t group,
9628 __rte_unused uint32_t *table,
9629 struct rte_flow_error *error)
9631 return rte_flow_error_set(error, ENOTSUP,
9632 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9633 "tunnel offload requires DV support");
9637 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9638 __rte_unused uint16_t port_id)
9641 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9644 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9647 struct rte_flow_error error;
9649 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9651 ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9653 (void *)(uintptr_t)item->type, &error);
9655 printf("%s ", item_name);
9657 printf("%d\n", (int)item->type);
9663 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9665 const struct rte_flow_item_udp *spec = udp_item->spec;
9666 const struct rte_flow_item_udp *mask = udp_item->mask;
9667 uint16_t udp_dport = 0;
9671 mask = &rte_flow_item_udp_mask;
9672 udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9673 mask->hdr.dst_port);
9675 return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9678 static const struct mlx5_flow_expand_node *
9679 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9680 unsigned int item_idx,
9681 const struct mlx5_flow_expand_node graph[],
9682 const struct mlx5_flow_expand_node *node)
9684 const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9686 if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
9688 node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
9690 * The expansion node is VXLAN and it is also the last
9691 * expandable item in the pattern, so need to continue
9692 * expansion of the inner tunnel.
9694 MLX5_ASSERT(item_idx > 0);
9695 prev_item = pattern + item_idx - 1;
9696 MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9697 if (mlx5_flow_is_std_vxlan_port(prev_item))
9698 return &graph[MLX5_EXPANSION_STD_VXLAN];
9699 return &graph[MLX5_EXPANSION_L3_VXLAN];
9704 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
9705 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
9706 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
9709 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
9710 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
9711 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
9712 { 9, 10, 11 }, { 12, 13, 14 },
9716 * Discover the number of available flow priorities.
9722 * On success, number of available flow priorities.
9723 * On failure, a negative errno-style code and rte_errno is set.
9726 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
9728 static const uint16_t vprio[] = {8, 16};
9729 const struct mlx5_priv *priv = dev->data->dev_private;
9730 const struct mlx5_flow_driver_ops *fops;
9731 enum mlx5_flow_drv_type type;
9734 type = mlx5_flow_os_get_type();
9735 if (type == MLX5_FLOW_TYPE_MAX) {
9736 type = MLX5_FLOW_TYPE_VERBS;
9737 if (priv->sh->devx && priv->config.dv_flow_en)
9738 type = MLX5_FLOW_TYPE_DV;
9740 fops = flow_get_drv_ops(type);
9741 if (fops->discover_priorities == NULL) {
9742 DRV_LOG(ERR, "Priority discovery not supported");
9743 rte_errno = ENOTSUP;
9746 ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
9751 ret = RTE_DIM(priority_map_3);
9754 ret = RTE_DIM(priority_map_5);
9757 rte_errno = ENOTSUP;
9759 "port %u maximum priority: %d expected 8/16",
9760 dev->data->port_id, ret);
9763 DRV_LOG(INFO, "port %u supported flow priorities:"
9764 " 0-%d for ingress or egress root table,"
9765 " 0-%d for non-root table or transfer root table.",
9766 dev->data->port_id, ret - 2,
9767 MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
9772 * Adjust flow priority based on the highest layer and the request priority.
9775 * Pointer to the Ethernet device structure.
9776 * @param[in] priority
9777 * The rule base priority.
9778 * @param[in] subpriority
9779 * The priority based on the items.
9785 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
9786 uint32_t subpriority)
9789 struct mlx5_priv *priv = dev->data->dev_private;
9791 switch (priv->sh->flow_max_priority) {
9792 case RTE_DIM(priority_map_3):
9793 res = priority_map_3[priority][subpriority];
9795 case RTE_DIM(priority_map_5):
9796 res = priority_map_5[priority][subpriority];