1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
10 #include <sys/queue.h>
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
25 #include <mlx5_malloc.h>
27 #include "mlx5_defs.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
36 struct tunnel_default_miss_ctx {
40 struct rte_flow_action_rss action_rss;
41 struct rte_flow_action_queue miss_queue;
42 struct rte_flow_action_jump miss_jump;
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 struct rte_flow *flow,
50 const struct rte_flow_attr *attr,
51 const struct rte_flow_action *app_actions,
53 const struct mlx5_flow_tunnel *tunnel,
54 struct tunnel_default_miss_ctx *ctx,
55 struct rte_flow_error *error);
56 static struct mlx5_flow_tunnel *
57 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
59 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
61 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
62 const struct mlx5_flow_tunnel *tunnel,
63 uint32_t group, uint32_t *table,
64 struct rte_flow_error *error);
66 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
67 static void mlx5_flow_pop_thread_workspace(void);
70 /** Device flow drivers. */
71 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
73 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
75 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
76 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
77 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
78 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
80 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
81 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
84 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
85 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
90 /** Node object of input graph for mlx5_flow_expand_rss(). */
91 struct mlx5_flow_expand_node {
92 const int *const next;
94 * List of next node indexes. Index 0 is interpreted as a terminator.
96 const enum rte_flow_item_type type;
97 /**< Pattern item type of current node. */
100 * RSS types bit-field associated with this node
101 * (see ETH_RSS_* definitions).
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 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) {
394 for (i = 0; node->next && node->next[i]; ++i) {
395 next = &graph[node->next[i]];
396 if (next->type == missed_item.type) {
397 flow_items[0].type = missed_item.type;
398 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
404 if (next && missed) {
405 elt = 2; /* missed item + item end. */
407 lsize += elt * sizeof(*item) + user_pattern_size;
410 if (node->rss_types & types) {
411 buf->entry[buf->entries].priority = 1;
412 buf->entry[buf->entries].pattern = addr;
414 rte_memcpy(addr, buf->entry[0].pattern,
416 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
417 rte_memcpy(addr, flow_items, elt * sizeof(*item));
418 addr = (void *)(((uintptr_t)addr) +
419 elt * sizeof(*item));
421 } else if (last_expand_item != NULL) {
422 node = mlx5_flow_expand_rss_adjust_node(pattern,
423 last_expand_item_idx, graph, node);
425 memset(flow_items, 0, sizeof(flow_items));
426 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
428 stack[stack_pos] = next_node;
429 node = next_node ? &graph[*next_node] : NULL;
431 flow_items[stack_pos].type = node->type;
432 if (node->rss_types & types) {
435 * compute the number of items to copy from the
436 * expansion and copy it.
437 * When the stack_pos is 0, there are 1 element in it,
438 * plus the addition END item.
441 flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
442 lsize += elt * sizeof(*item) + user_pattern_size;
445 n = elt * sizeof(*item);
446 buf->entry[buf->entries].priority =
447 stack_pos + 1 + missed;
448 buf->entry[buf->entries].pattern = addr;
450 rte_memcpy(addr, buf->entry[0].pattern,
452 addr = (void *)(((uintptr_t)addr) +
454 rte_memcpy(addr, &missed_item,
455 missed * sizeof(*item));
456 addr = (void *)(((uintptr_t)addr) +
457 missed * sizeof(*item));
458 rte_memcpy(addr, flow_items, n);
459 addr = (void *)(((uintptr_t)addr) + n);
462 if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
464 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
466 if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
470 stack[stack_pos] = next_node;
471 } else if (*(next_node + 1)) {
472 /* Follow up with the next possibility. */
473 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
475 } else if (!stack_pos) {
477 * Completing the traverse over the different paths.
478 * The next_node is advanced to the terminator.
482 /* Move to the next path. */
484 next_node = stack[--stack_pos];
489 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
491 stack[stack_pos] = next_node;
493 node = next_node && *next_node ? &graph[*next_node] : NULL;
498 enum mlx5_expansion {
500 MLX5_EXPANSION_ROOT_OUTER,
501 MLX5_EXPANSION_OUTER_ETH,
502 MLX5_EXPANSION_OUTER_VLAN,
503 MLX5_EXPANSION_OUTER_IPV4,
504 MLX5_EXPANSION_OUTER_IPV4_UDP,
505 MLX5_EXPANSION_OUTER_IPV4_TCP,
506 MLX5_EXPANSION_OUTER_IPV6,
507 MLX5_EXPANSION_OUTER_IPV6_UDP,
508 MLX5_EXPANSION_OUTER_IPV6_TCP,
509 MLX5_EXPANSION_VXLAN,
510 MLX5_EXPANSION_STD_VXLAN,
511 MLX5_EXPANSION_L3_VXLAN,
512 MLX5_EXPANSION_VXLAN_GPE,
514 MLX5_EXPANSION_NVGRE,
515 MLX5_EXPANSION_GRE_KEY,
520 MLX5_EXPANSION_IPV4_UDP,
521 MLX5_EXPANSION_IPV4_TCP,
523 MLX5_EXPANSION_IPV6_UDP,
524 MLX5_EXPANSION_IPV6_TCP,
525 MLX5_EXPANSION_IPV6_FRAG_EXT,
529 /** Supported expansion of items. */
530 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
531 [MLX5_EXPANSION_ROOT] = {
532 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
534 MLX5_EXPANSION_IPV6),
535 .type = RTE_FLOW_ITEM_TYPE_END,
537 [MLX5_EXPANSION_ROOT_OUTER] = {
538 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
539 MLX5_EXPANSION_OUTER_IPV4,
540 MLX5_EXPANSION_OUTER_IPV6),
541 .type = RTE_FLOW_ITEM_TYPE_END,
543 [MLX5_EXPANSION_OUTER_ETH] = {
544 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
545 .type = RTE_FLOW_ITEM_TYPE_ETH,
548 [MLX5_EXPANSION_OUTER_VLAN] = {
549 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
550 MLX5_EXPANSION_OUTER_IPV6),
551 .type = RTE_FLOW_ITEM_TYPE_VLAN,
552 .node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
554 [MLX5_EXPANSION_OUTER_IPV4] = {
555 .next = MLX5_FLOW_EXPAND_RSS_NEXT
556 (MLX5_EXPANSION_OUTER_IPV4_UDP,
557 MLX5_EXPANSION_OUTER_IPV4_TCP,
559 MLX5_EXPANSION_NVGRE,
561 MLX5_EXPANSION_IPV6),
562 .type = RTE_FLOW_ITEM_TYPE_IPV4,
563 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
564 ETH_RSS_NONFRAG_IPV4_OTHER,
566 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
567 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
568 MLX5_EXPANSION_VXLAN_GPE,
571 .type = RTE_FLOW_ITEM_TYPE_UDP,
572 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
574 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
575 .type = RTE_FLOW_ITEM_TYPE_TCP,
576 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
578 [MLX5_EXPANSION_OUTER_IPV6] = {
579 .next = MLX5_FLOW_EXPAND_RSS_NEXT
580 (MLX5_EXPANSION_OUTER_IPV6_UDP,
581 MLX5_EXPANSION_OUTER_IPV6_TCP,
585 MLX5_EXPANSION_NVGRE),
586 .type = RTE_FLOW_ITEM_TYPE_IPV6,
587 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
588 ETH_RSS_NONFRAG_IPV6_OTHER,
590 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
591 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
592 MLX5_EXPANSION_VXLAN_GPE,
595 .type = RTE_FLOW_ITEM_TYPE_UDP,
596 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
598 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
599 .type = RTE_FLOW_ITEM_TYPE_TCP,
600 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
602 [MLX5_EXPANSION_VXLAN] = {
603 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
605 MLX5_EXPANSION_IPV6),
606 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
608 [MLX5_EXPANSION_STD_VXLAN] = {
609 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
610 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
612 [MLX5_EXPANSION_L3_VXLAN] = {
613 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
614 MLX5_EXPANSION_IPV6),
615 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
617 [MLX5_EXPANSION_VXLAN_GPE] = {
618 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
620 MLX5_EXPANSION_IPV6),
621 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
623 [MLX5_EXPANSION_GRE] = {
624 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
626 MLX5_EXPANSION_GRE_KEY,
627 MLX5_EXPANSION_MPLS),
628 .type = RTE_FLOW_ITEM_TYPE_GRE,
630 [MLX5_EXPANSION_GRE_KEY] = {
631 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
633 MLX5_EXPANSION_MPLS),
634 .type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
635 .node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
637 [MLX5_EXPANSION_NVGRE] = {
638 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
639 .type = RTE_FLOW_ITEM_TYPE_NVGRE,
641 [MLX5_EXPANSION_MPLS] = {
642 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
645 .type = RTE_FLOW_ITEM_TYPE_MPLS,
646 .node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
648 [MLX5_EXPANSION_ETH] = {
649 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
650 .type = RTE_FLOW_ITEM_TYPE_ETH,
652 [MLX5_EXPANSION_VLAN] = {
653 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
654 MLX5_EXPANSION_IPV6),
655 .type = RTE_FLOW_ITEM_TYPE_VLAN,
656 .node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
658 [MLX5_EXPANSION_IPV4] = {
659 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
660 MLX5_EXPANSION_IPV4_TCP),
661 .type = RTE_FLOW_ITEM_TYPE_IPV4,
662 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
663 ETH_RSS_NONFRAG_IPV4_OTHER,
665 [MLX5_EXPANSION_IPV4_UDP] = {
666 .type = RTE_FLOW_ITEM_TYPE_UDP,
667 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
669 [MLX5_EXPANSION_IPV4_TCP] = {
670 .type = RTE_FLOW_ITEM_TYPE_TCP,
671 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
673 [MLX5_EXPANSION_IPV6] = {
674 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
675 MLX5_EXPANSION_IPV6_TCP,
676 MLX5_EXPANSION_IPV6_FRAG_EXT),
677 .type = RTE_FLOW_ITEM_TYPE_IPV6,
678 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
679 ETH_RSS_NONFRAG_IPV6_OTHER,
681 [MLX5_EXPANSION_IPV6_UDP] = {
682 .type = RTE_FLOW_ITEM_TYPE_UDP,
683 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
685 [MLX5_EXPANSION_IPV6_TCP] = {
686 .type = RTE_FLOW_ITEM_TYPE_TCP,
687 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
689 [MLX5_EXPANSION_IPV6_FRAG_EXT] = {
690 .type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
692 [MLX5_EXPANSION_GTP] = {
693 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
694 MLX5_EXPANSION_IPV6),
695 .type = RTE_FLOW_ITEM_TYPE_GTP,
699 static struct rte_flow_action_handle *
700 mlx5_action_handle_create(struct rte_eth_dev *dev,
701 const struct rte_flow_indir_action_conf *conf,
702 const struct rte_flow_action *action,
703 struct rte_flow_error *error);
704 static int mlx5_action_handle_destroy
705 (struct rte_eth_dev *dev,
706 struct rte_flow_action_handle *handle,
707 struct rte_flow_error *error);
708 static int mlx5_action_handle_update
709 (struct rte_eth_dev *dev,
710 struct rte_flow_action_handle *handle,
712 struct rte_flow_error *error);
713 static int mlx5_action_handle_query
714 (struct rte_eth_dev *dev,
715 const struct rte_flow_action_handle *handle,
717 struct rte_flow_error *error);
719 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
720 struct rte_flow_tunnel *app_tunnel,
721 struct rte_flow_action **actions,
722 uint32_t *num_of_actions,
723 struct rte_flow_error *error);
725 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
726 struct rte_flow_tunnel *app_tunnel,
727 struct rte_flow_item **items,
728 uint32_t *num_of_items,
729 struct rte_flow_error *error);
731 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
732 struct rte_flow_item *pmd_items,
733 uint32_t num_items, struct rte_flow_error *err);
735 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
736 struct rte_flow_action *pmd_actions,
737 uint32_t num_actions,
738 struct rte_flow_error *err);
740 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
742 struct rte_flow_restore_info *info,
743 struct rte_flow_error *err);
745 static const struct rte_flow_ops mlx5_flow_ops = {
746 .validate = mlx5_flow_validate,
747 .create = mlx5_flow_create,
748 .destroy = mlx5_flow_destroy,
749 .flush = mlx5_flow_flush,
750 .isolate = mlx5_flow_isolate,
751 .query = mlx5_flow_query,
752 .dev_dump = mlx5_flow_dev_dump,
753 .get_aged_flows = mlx5_flow_get_aged_flows,
754 .action_handle_create = mlx5_action_handle_create,
755 .action_handle_destroy = mlx5_action_handle_destroy,
756 .action_handle_update = mlx5_action_handle_update,
757 .action_handle_query = mlx5_action_handle_query,
758 .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
759 .tunnel_match = mlx5_flow_tunnel_match,
760 .tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
761 .tunnel_item_release = mlx5_flow_tunnel_item_release,
762 .get_restore_info = mlx5_flow_tunnel_get_restore_info,
765 /* Tunnel information. */
766 struct mlx5_flow_tunnel_info {
767 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
768 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
771 static struct mlx5_flow_tunnel_info tunnels_info[] = {
773 .tunnel = MLX5_FLOW_LAYER_VXLAN,
774 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
777 .tunnel = MLX5_FLOW_LAYER_GENEVE,
778 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
781 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
782 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
785 .tunnel = MLX5_FLOW_LAYER_GRE,
786 .ptype = RTE_PTYPE_TUNNEL_GRE,
789 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
790 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
793 .tunnel = MLX5_FLOW_LAYER_MPLS,
794 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
797 .tunnel = MLX5_FLOW_LAYER_NVGRE,
798 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
801 .tunnel = MLX5_FLOW_LAYER_IPIP,
802 .ptype = RTE_PTYPE_TUNNEL_IP,
805 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
806 .ptype = RTE_PTYPE_TUNNEL_IP,
809 .tunnel = MLX5_FLOW_LAYER_GTP,
810 .ptype = RTE_PTYPE_TUNNEL_GTPU,
817 * Translate tag ID to register.
820 * Pointer to the Ethernet device structure.
822 * The feature that request the register.
824 * The request register ID.
826 * Error description in case of any.
829 * The request register on success, a negative errno
830 * value otherwise and rte_errno is set.
833 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
834 enum mlx5_feature_name feature,
836 struct rte_flow_error *error)
838 struct mlx5_priv *priv = dev->data->dev_private;
839 struct mlx5_dev_config *config = &priv->config;
840 enum modify_reg start_reg;
841 bool skip_mtr_reg = false;
844 case MLX5_HAIRPIN_RX:
846 case MLX5_HAIRPIN_TX:
848 case MLX5_METADATA_RX:
849 switch (config->dv_xmeta_en) {
850 case MLX5_XMETA_MODE_LEGACY:
852 case MLX5_XMETA_MODE_META16:
854 case MLX5_XMETA_MODE_META32:
858 case MLX5_METADATA_TX:
860 case MLX5_METADATA_FDB:
861 switch (config->dv_xmeta_en) {
862 case MLX5_XMETA_MODE_LEGACY:
864 case MLX5_XMETA_MODE_META16:
866 case MLX5_XMETA_MODE_META32:
871 switch (config->dv_xmeta_en) {
872 case MLX5_XMETA_MODE_LEGACY:
874 case MLX5_XMETA_MODE_META16:
876 case MLX5_XMETA_MODE_META32:
882 * If meter color and meter id share one register, flow match
883 * should use the meter color register for match.
885 if (priv->mtr_reg_share)
886 return priv->mtr_color_reg;
888 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
891 case MLX5_ASO_FLOW_HIT:
892 case MLX5_ASO_CONNTRACK:
893 /* All features use the same REG_C. */
894 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
895 return priv->mtr_color_reg;
898 * Metadata COPY_MARK register using is in meter suffix sub
899 * flow while with meter. It's safe to share the same register.
901 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
904 * If meter is enable, it will engage the register for color
905 * match and flow match. If meter color match is not using the
906 * REG_C_2, need to skip the REG_C_x be used by meter color
908 * If meter is disable, free to use all available registers.
910 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
911 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
912 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
913 if (id > (uint32_t)(REG_C_7 - start_reg))
914 return rte_flow_error_set(error, EINVAL,
915 RTE_FLOW_ERROR_TYPE_ITEM,
916 NULL, "invalid tag id");
917 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
918 return rte_flow_error_set(error, ENOTSUP,
919 RTE_FLOW_ERROR_TYPE_ITEM,
920 NULL, "unsupported tag id");
922 * This case means meter is using the REG_C_x great than 2.
923 * Take care not to conflict with meter color REG_C_x.
924 * If the available index REG_C_y >= REG_C_x, skip the
927 if (skip_mtr_reg && config->flow_mreg_c
928 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
929 if (id >= (uint32_t)(REG_C_7 - start_reg))
930 return rte_flow_error_set(error, EINVAL,
931 RTE_FLOW_ERROR_TYPE_ITEM,
932 NULL, "invalid tag id");
933 if (config->flow_mreg_c
934 [id + 1 + start_reg - REG_C_0] != REG_NON)
935 return config->flow_mreg_c
936 [id + 1 + start_reg - REG_C_0];
937 return rte_flow_error_set(error, ENOTSUP,
938 RTE_FLOW_ERROR_TYPE_ITEM,
939 NULL, "unsupported tag id");
941 return config->flow_mreg_c[id + start_reg - REG_C_0];
944 return rte_flow_error_set(error, EINVAL,
945 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
946 NULL, "invalid feature name");
950 * Check extensive flow metadata register support.
953 * Pointer to rte_eth_dev structure.
956 * True if device supports extensive flow metadata register, otherwise false.
959 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
961 struct mlx5_priv *priv = dev->data->dev_private;
962 struct mlx5_dev_config *config = &priv->config;
965 * Having available reg_c can be regarded inclusively as supporting
966 * extensive flow metadata register, which could mean,
967 * - metadata register copy action by modify header.
968 * - 16 modify header actions is supported.
969 * - reg_c's are preserved across different domain (FDB and NIC) on
970 * packet loopback by flow lookup miss.
972 return config->flow_mreg_c[2] != REG_NON;
976 * Get the lowest priority.
979 * Pointer to the Ethernet device structure.
980 * @param[in] attributes
981 * Pointer to device flow rule attributes.
984 * The value of lowest priority of flow.
987 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
988 const struct rte_flow_attr *attr)
990 struct mlx5_priv *priv = dev->data->dev_private;
992 if (!attr->group && !attr->transfer)
993 return priv->config.flow_prio - 2;
994 return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
998 * Calculate matcher priority of the flow.
1001 * Pointer to the Ethernet device structure.
1003 * Pointer to device flow rule attributes.
1004 * @param[in] subpriority
1005 * The priority based on the items.
1007 * The matcher priority of the flow.
1010 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1011 const struct rte_flow_attr *attr,
1012 uint32_t subpriority)
1014 uint16_t priority = (uint16_t)attr->priority;
1015 struct mlx5_priv *priv = dev->data->dev_private;
1017 if (!attr->group && !attr->transfer) {
1018 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1019 priority = priv->config.flow_prio - 1;
1020 return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1022 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1023 priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1024 return priority * 3 + subpriority;
1028 * Verify the @p item specifications (spec, last, mask) are compatible with the
1032 * Item specification.
1034 * @p item->mask or flow default bit-masks.
1035 * @param[in] nic_mask
1036 * Bit-masks covering supported fields by the NIC to compare with user mask.
1038 * Bit-masks size in bytes.
1039 * @param[in] range_accepted
1040 * True if range of values is accepted for specific fields, false otherwise.
1042 * Pointer to error structure.
1045 * 0 on success, a negative errno value otherwise and rte_errno is set.
1048 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1049 const uint8_t *mask,
1050 const uint8_t *nic_mask,
1052 bool range_accepted,
1053 struct rte_flow_error *error)
1057 MLX5_ASSERT(nic_mask);
1058 for (i = 0; i < size; ++i)
1059 if ((nic_mask[i] | mask[i]) != nic_mask[i])
1060 return rte_flow_error_set(error, ENOTSUP,
1061 RTE_FLOW_ERROR_TYPE_ITEM,
1063 "mask enables non supported"
1065 if (!item->spec && (item->mask || item->last))
1066 return rte_flow_error_set(error, EINVAL,
1067 RTE_FLOW_ERROR_TYPE_ITEM, item,
1068 "mask/last without a spec is not"
1070 if (item->spec && item->last && !range_accepted) {
1076 for (i = 0; i < size; ++i) {
1077 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1078 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1080 ret = memcmp(spec, last, size);
1082 return rte_flow_error_set(error, EINVAL,
1083 RTE_FLOW_ERROR_TYPE_ITEM,
1085 "range is not valid");
1091 * Adjust the hash fields according to the @p flow information.
1093 * @param[in] dev_flow.
1094 * Pointer to the mlx5_flow.
1096 * 1 when the hash field is for a tunnel item.
1097 * @param[in] layer_types
1099 * @param[in] hash_fields
1103 * The hash fields that should be used.
1106 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1107 int tunnel __rte_unused, uint64_t layer_types,
1108 uint64_t hash_fields)
1110 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1111 int rss_request_inner = rss_desc->level >= 2;
1113 /* Check RSS hash level for tunnel. */
1114 if (tunnel && rss_request_inner)
1115 hash_fields |= IBV_RX_HASH_INNER;
1116 else if (tunnel || rss_request_inner)
1119 /* Check if requested layer matches RSS hash fields. */
1120 if (!(rss_desc->types & layer_types))
1126 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
1127 * if several tunnel rules are used on this queue, the tunnel ptype will be
1131 * Rx queue to update.
1134 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1137 uint32_t tunnel_ptype = 0;
1139 /* Look up for the ptype to use. */
1140 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1141 if (!rxq_ctrl->flow_tunnels_n[i])
1143 if (!tunnel_ptype) {
1144 tunnel_ptype = tunnels_info[i].ptype;
1150 rxq_ctrl->rxq.tunnel = tunnel_ptype;
1154 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1158 * Pointer to the Ethernet device structure.
1159 * @param[in] dev_handle
1160 * Pointer to device flow handle structure.
1163 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1164 struct mlx5_flow_handle *dev_handle)
1166 struct mlx5_priv *priv = dev->data->dev_private;
1167 const int mark = dev_handle->mark;
1168 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1169 struct mlx5_ind_table_obj *ind_tbl = NULL;
1172 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1173 struct mlx5_hrxq *hrxq;
1175 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1176 dev_handle->rix_hrxq);
1178 ind_tbl = hrxq->ind_table;
1179 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1180 struct mlx5_shared_action_rss *shared_rss;
1182 shared_rss = mlx5_ipool_get
1183 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1184 dev_handle->rix_srss);
1186 ind_tbl = shared_rss->ind_tbl;
1190 for (i = 0; i != ind_tbl->queues_n; ++i) {
1191 int idx = ind_tbl->queues[i];
1192 struct mlx5_rxq_ctrl *rxq_ctrl =
1193 container_of((*priv->rxqs)[idx],
1194 struct mlx5_rxq_ctrl, rxq);
1197 * To support metadata register copy on Tx loopback,
1198 * this must be always enabled (metadata may arive
1199 * from other port - not from local flows only.
1201 if (priv->config.dv_flow_en &&
1202 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1203 mlx5_flow_ext_mreg_supported(dev)) {
1204 rxq_ctrl->rxq.mark = 1;
1205 rxq_ctrl->flow_mark_n = 1;
1207 rxq_ctrl->rxq.mark = 1;
1208 rxq_ctrl->flow_mark_n++;
1213 /* Increase the counter matching the flow. */
1214 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1215 if ((tunnels_info[j].tunnel &
1216 dev_handle->layers) ==
1217 tunnels_info[j].tunnel) {
1218 rxq_ctrl->flow_tunnels_n[j]++;
1222 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1228 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1231 * Pointer to the Ethernet device structure.
1233 * Pointer to flow structure.
1236 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1238 struct mlx5_priv *priv = dev->data->dev_private;
1239 uint32_t handle_idx;
1240 struct mlx5_flow_handle *dev_handle;
1242 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1243 handle_idx, dev_handle, next)
1244 flow_drv_rxq_flags_set(dev, dev_handle);
1248 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1249 * device flow if no other flow uses it with the same kind of request.
1252 * Pointer to Ethernet device.
1253 * @param[in] dev_handle
1254 * Pointer to the device flow handle structure.
1257 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1258 struct mlx5_flow_handle *dev_handle)
1260 struct mlx5_priv *priv = dev->data->dev_private;
1261 const int mark = dev_handle->mark;
1262 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1263 struct mlx5_ind_table_obj *ind_tbl = NULL;
1266 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1267 struct mlx5_hrxq *hrxq;
1269 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1270 dev_handle->rix_hrxq);
1272 ind_tbl = hrxq->ind_table;
1273 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1274 struct mlx5_shared_action_rss *shared_rss;
1276 shared_rss = mlx5_ipool_get
1277 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1278 dev_handle->rix_srss);
1280 ind_tbl = shared_rss->ind_tbl;
1284 MLX5_ASSERT(dev->data->dev_started);
1285 for (i = 0; i != ind_tbl->queues_n; ++i) {
1286 int idx = ind_tbl->queues[i];
1287 struct mlx5_rxq_ctrl *rxq_ctrl =
1288 container_of((*priv->rxqs)[idx],
1289 struct mlx5_rxq_ctrl, rxq);
1291 if (priv->config.dv_flow_en &&
1292 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1293 mlx5_flow_ext_mreg_supported(dev)) {
1294 rxq_ctrl->rxq.mark = 1;
1295 rxq_ctrl->flow_mark_n = 1;
1297 rxq_ctrl->flow_mark_n--;
1298 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1303 /* Decrease the counter matching the flow. */
1304 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1305 if ((tunnels_info[j].tunnel &
1306 dev_handle->layers) ==
1307 tunnels_info[j].tunnel) {
1308 rxq_ctrl->flow_tunnels_n[j]--;
1312 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1318 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1319 * @p flow if no other flow uses it with the same kind of request.
1322 * Pointer to Ethernet device.
1324 * Pointer to the flow.
1327 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1329 struct mlx5_priv *priv = dev->data->dev_private;
1330 uint32_t handle_idx;
1331 struct mlx5_flow_handle *dev_handle;
1333 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1334 handle_idx, dev_handle, next)
1335 flow_drv_rxq_flags_trim(dev, dev_handle);
1339 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1342 * Pointer to Ethernet device.
1345 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1347 struct mlx5_priv *priv = dev->data->dev_private;
1350 for (i = 0; i != priv->rxqs_n; ++i) {
1351 struct mlx5_rxq_ctrl *rxq_ctrl;
1354 if (!(*priv->rxqs)[i])
1356 rxq_ctrl = container_of((*priv->rxqs)[i],
1357 struct mlx5_rxq_ctrl, rxq);
1358 rxq_ctrl->flow_mark_n = 0;
1359 rxq_ctrl->rxq.mark = 0;
1360 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1361 rxq_ctrl->flow_tunnels_n[j] = 0;
1362 rxq_ctrl->rxq.tunnel = 0;
1367 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1370 * Pointer to the Ethernet device structure.
1373 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1375 struct mlx5_priv *priv = dev->data->dev_private;
1376 struct mlx5_rxq_data *data;
1379 for (i = 0; i != priv->rxqs_n; ++i) {
1380 if (!(*priv->rxqs)[i])
1382 data = (*priv->rxqs)[i];
1383 if (!rte_flow_dynf_metadata_avail()) {
1384 data->dynf_meta = 0;
1385 data->flow_meta_mask = 0;
1386 data->flow_meta_offset = -1;
1387 data->flow_meta_port_mask = 0;
1389 data->dynf_meta = 1;
1390 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1391 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1392 data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1398 * return a pointer to the desired action in the list of actions.
1400 * @param[in] actions
1401 * The list of actions to search the action in.
1403 * The action to find.
1406 * Pointer to the action in the list, if found. NULL otherwise.
1408 const struct rte_flow_action *
1409 mlx5_flow_find_action(const struct rte_flow_action *actions,
1410 enum rte_flow_action_type action)
1412 if (actions == NULL)
1414 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1415 if (actions->type == action)
1421 * Validate the flag action.
1423 * @param[in] action_flags
1424 * Bit-fields that holds the actions detected until now.
1426 * Attributes of flow that includes this action.
1428 * Pointer to error structure.
1431 * 0 on success, a negative errno value otherwise and rte_errno is set.
1434 mlx5_flow_validate_action_flag(uint64_t action_flags,
1435 const struct rte_flow_attr *attr,
1436 struct rte_flow_error *error)
1438 if (action_flags & MLX5_FLOW_ACTION_MARK)
1439 return rte_flow_error_set(error, EINVAL,
1440 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1441 "can't mark and flag in same flow");
1442 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1443 return rte_flow_error_set(error, EINVAL,
1444 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1446 " actions in same flow");
1448 return rte_flow_error_set(error, ENOTSUP,
1449 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1450 "flag action not supported for "
1456 * Validate the mark action.
1459 * Pointer to the queue action.
1460 * @param[in] action_flags
1461 * Bit-fields that holds the actions detected until now.
1463 * Attributes of flow that includes this action.
1465 * Pointer to error structure.
1468 * 0 on success, a negative errno value otherwise and rte_errno is set.
1471 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1472 uint64_t action_flags,
1473 const struct rte_flow_attr *attr,
1474 struct rte_flow_error *error)
1476 const struct rte_flow_action_mark *mark = action->conf;
1479 return rte_flow_error_set(error, EINVAL,
1480 RTE_FLOW_ERROR_TYPE_ACTION,
1482 "configuration cannot be null");
1483 if (mark->id >= MLX5_FLOW_MARK_MAX)
1484 return rte_flow_error_set(error, EINVAL,
1485 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1487 "mark id must in 0 <= id < "
1488 RTE_STR(MLX5_FLOW_MARK_MAX));
1489 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1490 return rte_flow_error_set(error, EINVAL,
1491 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1492 "can't flag and mark in same flow");
1493 if (action_flags & MLX5_FLOW_ACTION_MARK)
1494 return rte_flow_error_set(error, EINVAL,
1495 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1496 "can't have 2 mark actions in same"
1499 return rte_flow_error_set(error, ENOTSUP,
1500 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1501 "mark action not supported for "
1507 * Validate the drop action.
1509 * @param[in] action_flags
1510 * Bit-fields that holds the actions detected until now.
1512 * Attributes of flow that includes this action.
1514 * Pointer to error structure.
1517 * 0 on success, a negative errno value otherwise and rte_errno is set.
1520 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1521 const struct rte_flow_attr *attr,
1522 struct rte_flow_error *error)
1525 return rte_flow_error_set(error, ENOTSUP,
1526 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1527 "drop action not supported for "
1533 * Validate the queue action.
1536 * Pointer to the queue action.
1537 * @param[in] action_flags
1538 * Bit-fields that holds the actions detected until now.
1540 * Pointer to the Ethernet device structure.
1542 * Attributes of flow that includes this action.
1544 * Pointer to error structure.
1547 * 0 on success, a negative errno value otherwise and rte_errno is set.
1550 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1551 uint64_t action_flags,
1552 struct rte_eth_dev *dev,
1553 const struct rte_flow_attr *attr,
1554 struct rte_flow_error *error)
1556 struct mlx5_priv *priv = dev->data->dev_private;
1557 const struct rte_flow_action_queue *queue = action->conf;
1559 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1560 return rte_flow_error_set(error, EINVAL,
1561 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1562 "can't have 2 fate actions in"
1565 return rte_flow_error_set(error, EINVAL,
1566 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1567 NULL, "No Rx queues configured");
1568 if (queue->index >= priv->rxqs_n)
1569 return rte_flow_error_set(error, EINVAL,
1570 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1572 "queue index out of range");
1573 if (!(*priv->rxqs)[queue->index])
1574 return rte_flow_error_set(error, EINVAL,
1575 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1577 "queue is not configured");
1579 return rte_flow_error_set(error, ENOTSUP,
1580 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1581 "queue action not supported for "
1587 * Validate the rss action.
1590 * Pointer to the Ethernet device structure.
1592 * Pointer to the queue action.
1594 * Pointer to error structure.
1597 * 0 on success, a negative errno value otherwise and rte_errno is set.
1600 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1601 const struct rte_flow_action *action,
1602 struct rte_flow_error *error)
1604 struct mlx5_priv *priv = dev->data->dev_private;
1605 const struct rte_flow_action_rss *rss = action->conf;
1606 enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1609 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1610 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1611 return rte_flow_error_set(error, ENOTSUP,
1612 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1614 "RSS hash function not supported");
1615 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1620 return rte_flow_error_set(error, ENOTSUP,
1621 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1623 "tunnel RSS is not supported");
1624 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1625 if (rss->key_len == 0 && rss->key != NULL)
1626 return rte_flow_error_set(error, ENOTSUP,
1627 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1629 "RSS hash key length 0");
1630 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1631 return rte_flow_error_set(error, ENOTSUP,
1632 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1634 "RSS hash key too small");
1635 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1636 return rte_flow_error_set(error, ENOTSUP,
1637 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1639 "RSS hash key too large");
1640 if (rss->queue_num > priv->config.ind_table_max_size)
1641 return rte_flow_error_set(error, ENOTSUP,
1642 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1644 "number of queues too large");
1645 if (rss->types & MLX5_RSS_HF_MASK)
1646 return rte_flow_error_set(error, ENOTSUP,
1647 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1649 "some RSS protocols are not"
1651 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1652 !(rss->types & ETH_RSS_IP))
1653 return rte_flow_error_set(error, EINVAL,
1654 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1655 "L3 partial RSS requested but L3 RSS"
1656 " type not specified");
1657 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1658 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1659 return rte_flow_error_set(error, EINVAL,
1660 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1661 "L4 partial RSS requested but L4 RSS"
1662 " type not specified");
1664 return rte_flow_error_set(error, EINVAL,
1665 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1666 NULL, "No Rx queues configured");
1667 if (!rss->queue_num)
1668 return rte_flow_error_set(error, EINVAL,
1669 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1670 NULL, "No queues configured");
1671 for (i = 0; i != rss->queue_num; ++i) {
1672 struct mlx5_rxq_ctrl *rxq_ctrl;
1674 if (rss->queue[i] >= priv->rxqs_n)
1675 return rte_flow_error_set
1677 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1678 &rss->queue[i], "queue index out of range");
1679 if (!(*priv->rxqs)[rss->queue[i]])
1680 return rte_flow_error_set
1681 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1682 &rss->queue[i], "queue is not configured");
1683 rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1684 struct mlx5_rxq_ctrl, rxq);
1686 rxq_type = rxq_ctrl->type;
1687 if (rxq_type != rxq_ctrl->type)
1688 return rte_flow_error_set
1689 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1691 "combining hairpin and regular RSS queues is not supported");
1697 * Validate the rss action.
1700 * Pointer to the queue action.
1701 * @param[in] action_flags
1702 * Bit-fields that holds the actions detected until now.
1704 * Pointer to the Ethernet device structure.
1706 * Attributes of flow that includes this action.
1707 * @param[in] item_flags
1708 * Items that were detected.
1710 * Pointer to error structure.
1713 * 0 on success, a negative errno value otherwise and rte_errno is set.
1716 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1717 uint64_t action_flags,
1718 struct rte_eth_dev *dev,
1719 const struct rte_flow_attr *attr,
1720 uint64_t item_flags,
1721 struct rte_flow_error *error)
1723 const struct rte_flow_action_rss *rss = action->conf;
1724 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1727 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1728 return rte_flow_error_set(error, EINVAL,
1729 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1730 "can't have 2 fate actions"
1732 ret = mlx5_validate_action_rss(dev, action, error);
1736 return rte_flow_error_set(error, ENOTSUP,
1737 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1738 "rss action not supported for "
1740 if (rss->level > 1 && !tunnel)
1741 return rte_flow_error_set(error, EINVAL,
1742 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1743 "inner RSS is not supported for "
1744 "non-tunnel flows");
1745 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1746 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1747 return rte_flow_error_set(error, EINVAL,
1748 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1749 "RSS on eCPRI is not supported now");
1751 if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1753 (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1755 return rte_flow_error_set(error, EINVAL,
1756 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1757 "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1762 * Validate the default miss action.
1764 * @param[in] action_flags
1765 * Bit-fields that holds the actions detected until now.
1767 * Pointer to error structure.
1770 * 0 on success, a negative errno value otherwise and rte_errno is set.
1773 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1774 const struct rte_flow_attr *attr,
1775 struct rte_flow_error *error)
1777 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1778 return rte_flow_error_set(error, EINVAL,
1779 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1780 "can't have 2 fate actions in"
1783 return rte_flow_error_set(error, ENOTSUP,
1784 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1785 "default miss action not supported "
1788 return rte_flow_error_set(error, ENOTSUP,
1789 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1790 "only group 0 is supported");
1792 return rte_flow_error_set(error, ENOTSUP,
1793 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1794 NULL, "transfer is not supported");
1799 * Validate the count action.
1802 * Pointer to the Ethernet device structure.
1804 * Attributes of flow that includes this action.
1806 * Pointer to error structure.
1809 * 0 on success, a negative errno value otherwise and rte_errno is set.
1812 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1813 const struct rte_flow_attr *attr,
1814 struct rte_flow_error *error)
1817 return rte_flow_error_set(error, ENOTSUP,
1818 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1819 "count action not supported for "
1825 * Validate the ASO CT action.
1828 * Pointer to the Ethernet device structure.
1829 * @param[in] conntrack
1830 * Pointer to the CT action profile.
1832 * Pointer to error structure.
1835 * 0 on success, a negative errno value otherwise and rte_errno is set.
1838 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1839 const struct rte_flow_action_conntrack *conntrack,
1840 struct rte_flow_error *error)
1844 if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1845 return rte_flow_error_set(error, EINVAL,
1846 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1847 "Invalid CT state");
1848 if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1849 return rte_flow_error_set(error, EINVAL,
1850 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1851 "Invalid last TCP packet flag");
1856 * Verify the @p attributes will be correctly understood by the NIC and store
1857 * them in the @p flow if everything is correct.
1860 * Pointer to the Ethernet device structure.
1861 * @param[in] attributes
1862 * Pointer to flow attributes
1864 * Pointer to error structure.
1867 * 0 on success, a negative errno value otherwise and rte_errno is set.
1870 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1871 const struct rte_flow_attr *attributes,
1872 struct rte_flow_error *error)
1874 struct mlx5_priv *priv = dev->data->dev_private;
1875 uint32_t priority_max = priv->config.flow_prio - 1;
1877 if (attributes->group)
1878 return rte_flow_error_set(error, ENOTSUP,
1879 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1880 NULL, "groups is not supported");
1881 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1882 attributes->priority >= priority_max)
1883 return rte_flow_error_set(error, ENOTSUP,
1884 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1885 NULL, "priority out of range");
1886 if (attributes->egress)
1887 return rte_flow_error_set(error, ENOTSUP,
1888 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1889 "egress is not supported");
1890 if (attributes->transfer && !priv->config.dv_esw_en)
1891 return rte_flow_error_set(error, ENOTSUP,
1892 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1893 NULL, "transfer is not supported");
1894 if (!attributes->ingress)
1895 return rte_flow_error_set(error, EINVAL,
1896 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1898 "ingress attribute is mandatory");
1903 * Validate ICMP6 item.
1906 * Item specification.
1907 * @param[in] item_flags
1908 * Bit-fields that holds the items detected until now.
1909 * @param[in] ext_vlan_sup
1910 * Whether extended VLAN features are supported or not.
1912 * Pointer to error structure.
1915 * 0 on success, a negative errno value otherwise and rte_errno is set.
1918 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1919 uint64_t item_flags,
1920 uint8_t target_protocol,
1921 struct rte_flow_error *error)
1923 const struct rte_flow_item_icmp6 *mask = item->mask;
1924 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1925 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1926 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1927 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1928 MLX5_FLOW_LAYER_OUTER_L4;
1931 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1932 return rte_flow_error_set(error, EINVAL,
1933 RTE_FLOW_ERROR_TYPE_ITEM, item,
1934 "protocol filtering not compatible"
1935 " with ICMP6 layer");
1936 if (!(item_flags & l3m))
1937 return rte_flow_error_set(error, EINVAL,
1938 RTE_FLOW_ERROR_TYPE_ITEM, item,
1939 "IPv6 is mandatory to filter on"
1941 if (item_flags & l4m)
1942 return rte_flow_error_set(error, EINVAL,
1943 RTE_FLOW_ERROR_TYPE_ITEM, item,
1944 "multiple L4 layers not supported");
1946 mask = &rte_flow_item_icmp6_mask;
1947 ret = mlx5_flow_item_acceptable
1948 (item, (const uint8_t *)mask,
1949 (const uint8_t *)&rte_flow_item_icmp6_mask,
1950 sizeof(struct rte_flow_item_icmp6),
1951 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1958 * Validate ICMP item.
1961 * Item specification.
1962 * @param[in] item_flags
1963 * Bit-fields that holds the items detected until now.
1965 * Pointer to error structure.
1968 * 0 on success, a negative errno value otherwise and rte_errno is set.
1971 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1972 uint64_t item_flags,
1973 uint8_t target_protocol,
1974 struct rte_flow_error *error)
1976 const struct rte_flow_item_icmp *mask = item->mask;
1977 const struct rte_flow_item_icmp nic_mask = {
1978 .hdr.icmp_type = 0xff,
1979 .hdr.icmp_code = 0xff,
1980 .hdr.icmp_ident = RTE_BE16(0xffff),
1981 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1983 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1984 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1985 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1986 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1987 MLX5_FLOW_LAYER_OUTER_L4;
1990 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1991 return rte_flow_error_set(error, EINVAL,
1992 RTE_FLOW_ERROR_TYPE_ITEM, item,
1993 "protocol filtering not compatible"
1994 " with ICMP layer");
1995 if (!(item_flags & l3m))
1996 return rte_flow_error_set(error, EINVAL,
1997 RTE_FLOW_ERROR_TYPE_ITEM, item,
1998 "IPv4 is mandatory to filter"
2000 if (item_flags & l4m)
2001 return rte_flow_error_set(error, EINVAL,
2002 RTE_FLOW_ERROR_TYPE_ITEM, item,
2003 "multiple L4 layers not supported");
2006 ret = mlx5_flow_item_acceptable
2007 (item, (const uint8_t *)mask,
2008 (const uint8_t *)&nic_mask,
2009 sizeof(struct rte_flow_item_icmp),
2010 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2017 * Validate Ethernet item.
2020 * Item specification.
2021 * @param[in] item_flags
2022 * Bit-fields that holds the items detected until now.
2024 * Pointer to error structure.
2027 * 0 on success, a negative errno value otherwise and rte_errno is set.
2030 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2031 uint64_t item_flags, bool ext_vlan_sup,
2032 struct rte_flow_error *error)
2034 const struct rte_flow_item_eth *mask = item->mask;
2035 const struct rte_flow_item_eth nic_mask = {
2036 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2037 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2038 .type = RTE_BE16(0xffff),
2039 .has_vlan = ext_vlan_sup ? 1 : 0,
2042 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2043 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
2044 MLX5_FLOW_LAYER_OUTER_L2;
2046 if (item_flags & ethm)
2047 return rte_flow_error_set(error, ENOTSUP,
2048 RTE_FLOW_ERROR_TYPE_ITEM, item,
2049 "multiple L2 layers not supported");
2050 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2051 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2052 return rte_flow_error_set(error, EINVAL,
2053 RTE_FLOW_ERROR_TYPE_ITEM, item,
2054 "L2 layer should not follow "
2056 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2057 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2058 return rte_flow_error_set(error, EINVAL,
2059 RTE_FLOW_ERROR_TYPE_ITEM, item,
2060 "L2 layer should not follow VLAN");
2061 if (item_flags & MLX5_FLOW_LAYER_GTP)
2062 return rte_flow_error_set(error, EINVAL,
2063 RTE_FLOW_ERROR_TYPE_ITEM, item,
2064 "L2 layer should not follow GTP");
2066 mask = &rte_flow_item_eth_mask;
2067 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2068 (const uint8_t *)&nic_mask,
2069 sizeof(struct rte_flow_item_eth),
2070 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2075 * Validate VLAN item.
2078 * Item specification.
2079 * @param[in] item_flags
2080 * Bit-fields that holds the items detected until now.
2082 * Ethernet device flow is being created on.
2084 * Pointer to error structure.
2087 * 0 on success, a negative errno value otherwise and rte_errno is set.
2090 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2091 uint64_t item_flags,
2092 struct rte_eth_dev *dev,
2093 struct rte_flow_error *error)
2095 const struct rte_flow_item_vlan *spec = item->spec;
2096 const struct rte_flow_item_vlan *mask = item->mask;
2097 const struct rte_flow_item_vlan nic_mask = {
2098 .tci = RTE_BE16(UINT16_MAX),
2099 .inner_type = RTE_BE16(UINT16_MAX),
2101 uint16_t vlan_tag = 0;
2102 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2104 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2105 MLX5_FLOW_LAYER_INNER_L4) :
2106 (MLX5_FLOW_LAYER_OUTER_L3 |
2107 MLX5_FLOW_LAYER_OUTER_L4);
2108 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2109 MLX5_FLOW_LAYER_OUTER_VLAN;
2111 if (item_flags & vlanm)
2112 return rte_flow_error_set(error, EINVAL,
2113 RTE_FLOW_ERROR_TYPE_ITEM, item,
2114 "multiple VLAN layers not supported");
2115 else if ((item_flags & l34m) != 0)
2116 return rte_flow_error_set(error, EINVAL,
2117 RTE_FLOW_ERROR_TYPE_ITEM, item,
2118 "VLAN cannot follow L3/L4 layer");
2120 mask = &rte_flow_item_vlan_mask;
2121 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2122 (const uint8_t *)&nic_mask,
2123 sizeof(struct rte_flow_item_vlan),
2124 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2127 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2128 struct mlx5_priv *priv = dev->data->dev_private;
2130 if (priv->vmwa_context) {
2132 * Non-NULL context means we have a virtual machine
2133 * and SR-IOV enabled, we have to create VLAN interface
2134 * to make hypervisor to setup E-Switch vport
2135 * context correctly. We avoid creating the multiple
2136 * VLAN interfaces, so we cannot support VLAN tag mask.
2138 return rte_flow_error_set(error, EINVAL,
2139 RTE_FLOW_ERROR_TYPE_ITEM,
2141 "VLAN tag mask is not"
2142 " supported in virtual"
2147 vlan_tag = spec->tci;
2148 vlan_tag &= mask->tci;
2151 * From verbs perspective an empty VLAN is equivalent
2152 * to a packet without VLAN layer.
2155 return rte_flow_error_set(error, EINVAL,
2156 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2158 "VLAN cannot be empty");
2163 * Validate IPV4 item.
2166 * Item specification.
2167 * @param[in] item_flags
2168 * Bit-fields that holds the items detected until now.
2169 * @param[in] last_item
2170 * Previous validated item in the pattern items.
2171 * @param[in] ether_type
2172 * Type in the ethernet layer header (including dot1q).
2173 * @param[in] acc_mask
2174 * Acceptable mask, if NULL default internal default mask
2175 * will be used to check whether item fields are supported.
2176 * @param[in] range_accepted
2177 * True if range of values is accepted for specific fields, false otherwise.
2179 * Pointer to error structure.
2182 * 0 on success, a negative errno value otherwise and rte_errno is set.
2185 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2186 uint64_t item_flags,
2188 uint16_t ether_type,
2189 const struct rte_flow_item_ipv4 *acc_mask,
2190 bool range_accepted,
2191 struct rte_flow_error *error)
2193 const struct rte_flow_item_ipv4 *mask = item->mask;
2194 const struct rte_flow_item_ipv4 *spec = item->spec;
2195 const struct rte_flow_item_ipv4 nic_mask = {
2197 .src_addr = RTE_BE32(0xffffffff),
2198 .dst_addr = RTE_BE32(0xffffffff),
2199 .type_of_service = 0xff,
2200 .next_proto_id = 0xff,
2203 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2204 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2205 MLX5_FLOW_LAYER_OUTER_L3;
2206 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2207 MLX5_FLOW_LAYER_OUTER_L4;
2209 uint8_t next_proto = 0xFF;
2210 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2211 MLX5_FLOW_LAYER_OUTER_VLAN |
2212 MLX5_FLOW_LAYER_INNER_VLAN);
2214 if ((last_item & l2_vlan) && ether_type &&
2215 ether_type != RTE_ETHER_TYPE_IPV4)
2216 return rte_flow_error_set(error, EINVAL,
2217 RTE_FLOW_ERROR_TYPE_ITEM, item,
2218 "IPv4 cannot follow L2/VLAN layer "
2219 "which ether type is not IPv4");
2220 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2222 next_proto = mask->hdr.next_proto_id &
2223 spec->hdr.next_proto_id;
2224 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2225 return rte_flow_error_set(error, EINVAL,
2226 RTE_FLOW_ERROR_TYPE_ITEM,
2231 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2232 return rte_flow_error_set(error, EINVAL,
2233 RTE_FLOW_ERROR_TYPE_ITEM, item,
2234 "wrong tunnel type - IPv6 specified "
2235 "but IPv4 item provided");
2236 if (item_flags & l3m)
2237 return rte_flow_error_set(error, ENOTSUP,
2238 RTE_FLOW_ERROR_TYPE_ITEM, item,
2239 "multiple L3 layers not supported");
2240 else if (item_flags & l4m)
2241 return rte_flow_error_set(error, EINVAL,
2242 RTE_FLOW_ERROR_TYPE_ITEM, item,
2243 "L3 cannot follow an L4 layer.");
2244 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2245 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2246 return rte_flow_error_set(error, EINVAL,
2247 RTE_FLOW_ERROR_TYPE_ITEM, item,
2248 "L3 cannot follow an NVGRE layer.");
2250 mask = &rte_flow_item_ipv4_mask;
2251 else if (mask->hdr.next_proto_id != 0 &&
2252 mask->hdr.next_proto_id != 0xff)
2253 return rte_flow_error_set(error, EINVAL,
2254 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2255 "partial mask is not supported"
2257 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2258 acc_mask ? (const uint8_t *)acc_mask
2259 : (const uint8_t *)&nic_mask,
2260 sizeof(struct rte_flow_item_ipv4),
2261 range_accepted, error);
2268 * Validate IPV6 item.
2271 * Item specification.
2272 * @param[in] item_flags
2273 * Bit-fields that holds the items detected until now.
2274 * @param[in] last_item
2275 * Previous validated item in the pattern items.
2276 * @param[in] ether_type
2277 * Type in the ethernet layer header (including dot1q).
2278 * @param[in] acc_mask
2279 * Acceptable mask, if NULL default internal default mask
2280 * will be used to check whether item fields are supported.
2282 * Pointer to error structure.
2285 * 0 on success, a negative errno value otherwise and rte_errno is set.
2288 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2289 uint64_t item_flags,
2291 uint16_t ether_type,
2292 const struct rte_flow_item_ipv6 *acc_mask,
2293 struct rte_flow_error *error)
2295 const struct rte_flow_item_ipv6 *mask = item->mask;
2296 const struct rte_flow_item_ipv6 *spec = item->spec;
2297 const struct rte_flow_item_ipv6 nic_mask = {
2300 "\xff\xff\xff\xff\xff\xff\xff\xff"
2301 "\xff\xff\xff\xff\xff\xff\xff\xff",
2303 "\xff\xff\xff\xff\xff\xff\xff\xff"
2304 "\xff\xff\xff\xff\xff\xff\xff\xff",
2305 .vtc_flow = RTE_BE32(0xffffffff),
2309 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2310 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2311 MLX5_FLOW_LAYER_OUTER_L3;
2312 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2313 MLX5_FLOW_LAYER_OUTER_L4;
2315 uint8_t next_proto = 0xFF;
2316 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2317 MLX5_FLOW_LAYER_OUTER_VLAN |
2318 MLX5_FLOW_LAYER_INNER_VLAN);
2320 if ((last_item & l2_vlan) && ether_type &&
2321 ether_type != RTE_ETHER_TYPE_IPV6)
2322 return rte_flow_error_set(error, EINVAL,
2323 RTE_FLOW_ERROR_TYPE_ITEM, item,
2324 "IPv6 cannot follow L2/VLAN layer "
2325 "which ether type is not IPv6");
2326 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2327 next_proto = spec->hdr.proto;
2328 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2329 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2330 return rte_flow_error_set(error, EINVAL,
2331 RTE_FLOW_ERROR_TYPE_ITEM,
2336 if (next_proto == IPPROTO_HOPOPTS ||
2337 next_proto == IPPROTO_ROUTING ||
2338 next_proto == IPPROTO_FRAGMENT ||
2339 next_proto == IPPROTO_ESP ||
2340 next_proto == IPPROTO_AH ||
2341 next_proto == IPPROTO_DSTOPTS)
2342 return rte_flow_error_set(error, EINVAL,
2343 RTE_FLOW_ERROR_TYPE_ITEM, item,
2344 "IPv6 proto (next header) should "
2345 "not be set as extension header");
2346 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2347 return rte_flow_error_set(error, EINVAL,
2348 RTE_FLOW_ERROR_TYPE_ITEM, item,
2349 "wrong tunnel type - IPv4 specified "
2350 "but IPv6 item provided");
2351 if (item_flags & l3m)
2352 return rte_flow_error_set(error, ENOTSUP,
2353 RTE_FLOW_ERROR_TYPE_ITEM, item,
2354 "multiple L3 layers not supported");
2355 else if (item_flags & l4m)
2356 return rte_flow_error_set(error, EINVAL,
2357 RTE_FLOW_ERROR_TYPE_ITEM, item,
2358 "L3 cannot follow an L4 layer.");
2359 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2360 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2361 return rte_flow_error_set(error, EINVAL,
2362 RTE_FLOW_ERROR_TYPE_ITEM, item,
2363 "L3 cannot follow an NVGRE layer.");
2365 mask = &rte_flow_item_ipv6_mask;
2366 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2367 acc_mask ? (const uint8_t *)acc_mask
2368 : (const uint8_t *)&nic_mask,
2369 sizeof(struct rte_flow_item_ipv6),
2370 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2377 * Validate UDP item.
2380 * Item specification.
2381 * @param[in] item_flags
2382 * Bit-fields that holds the items detected until now.
2383 * @param[in] target_protocol
2384 * The next protocol in the previous item.
2385 * @param[in] flow_mask
2386 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2388 * Pointer to error structure.
2391 * 0 on success, a negative errno value otherwise and rte_errno is set.
2394 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2395 uint64_t item_flags,
2396 uint8_t target_protocol,
2397 struct rte_flow_error *error)
2399 const struct rte_flow_item_udp *mask = item->mask;
2400 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2401 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2402 MLX5_FLOW_LAYER_OUTER_L3;
2403 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2404 MLX5_FLOW_LAYER_OUTER_L4;
2407 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2408 return rte_flow_error_set(error, EINVAL,
2409 RTE_FLOW_ERROR_TYPE_ITEM, item,
2410 "protocol filtering not compatible"
2412 if (!(item_flags & l3m))
2413 return rte_flow_error_set(error, EINVAL,
2414 RTE_FLOW_ERROR_TYPE_ITEM, item,
2415 "L3 is mandatory to filter on L4");
2416 if (item_flags & l4m)
2417 return rte_flow_error_set(error, EINVAL,
2418 RTE_FLOW_ERROR_TYPE_ITEM, item,
2419 "multiple L4 layers not supported");
2421 mask = &rte_flow_item_udp_mask;
2422 ret = mlx5_flow_item_acceptable
2423 (item, (const uint8_t *)mask,
2424 (const uint8_t *)&rte_flow_item_udp_mask,
2425 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2433 * Validate TCP item.
2436 * Item specification.
2437 * @param[in] item_flags
2438 * Bit-fields that holds the items detected until now.
2439 * @param[in] target_protocol
2440 * The next protocol in the previous item.
2442 * Pointer to error structure.
2445 * 0 on success, a negative errno value otherwise and rte_errno is set.
2448 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2449 uint64_t item_flags,
2450 uint8_t target_protocol,
2451 const struct rte_flow_item_tcp *flow_mask,
2452 struct rte_flow_error *error)
2454 const struct rte_flow_item_tcp *mask = item->mask;
2455 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2456 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2457 MLX5_FLOW_LAYER_OUTER_L3;
2458 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2459 MLX5_FLOW_LAYER_OUTER_L4;
2462 MLX5_ASSERT(flow_mask);
2463 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2464 return rte_flow_error_set(error, EINVAL,
2465 RTE_FLOW_ERROR_TYPE_ITEM, item,
2466 "protocol filtering not compatible"
2468 if (!(item_flags & l3m))
2469 return rte_flow_error_set(error, EINVAL,
2470 RTE_FLOW_ERROR_TYPE_ITEM, item,
2471 "L3 is mandatory to filter on L4");
2472 if (item_flags & l4m)
2473 return rte_flow_error_set(error, EINVAL,
2474 RTE_FLOW_ERROR_TYPE_ITEM, item,
2475 "multiple L4 layers not supported");
2477 mask = &rte_flow_item_tcp_mask;
2478 ret = mlx5_flow_item_acceptable
2479 (item, (const uint8_t *)mask,
2480 (const uint8_t *)flow_mask,
2481 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2489 * Validate VXLAN item.
2492 * Pointer to the Ethernet device structure.
2493 * @param[in] udp_dport
2494 * UDP destination port
2496 * Item specification.
2497 * @param[in] item_flags
2498 * Bit-fields that holds the items detected until now.
2500 * Flow rule attributes.
2502 * Pointer to error structure.
2505 * 0 on success, a negative errno value otherwise and rte_errno is set.
2508 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2510 const struct rte_flow_item *item,
2511 uint64_t item_flags,
2512 const struct rte_flow_attr *attr,
2513 struct rte_flow_error *error)
2515 const struct rte_flow_item_vxlan *spec = item->spec;
2516 const struct rte_flow_item_vxlan *mask = item->mask;
2518 struct mlx5_priv *priv = dev->data->dev_private;
2522 } id = { .vlan_id = 0, };
2523 const struct rte_flow_item_vxlan nic_mask = {
2524 .vni = "\xff\xff\xff",
2527 const struct rte_flow_item_vxlan *valid_mask;
2529 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2530 return rte_flow_error_set(error, ENOTSUP,
2531 RTE_FLOW_ERROR_TYPE_ITEM, item,
2532 "multiple tunnel layers not"
2534 valid_mask = &rte_flow_item_vxlan_mask;
2536 * Verify only UDPv4 is present as defined in
2537 * https://tools.ietf.org/html/rfc7348
2539 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2540 return rte_flow_error_set(error, EINVAL,
2541 RTE_FLOW_ERROR_TYPE_ITEM, item,
2542 "no outer UDP layer found");
2544 mask = &rte_flow_item_vxlan_mask;
2546 if (priv->sh->steering_format_version !=
2547 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2548 !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2549 /* FDB domain & NIC domain non-zero group */
2550 if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2551 valid_mask = &nic_mask;
2552 /* Group zero in NIC domain */
2553 if (!attr->group && !attr->transfer &&
2554 priv->sh->tunnel_header_0_1)
2555 valid_mask = &nic_mask;
2557 ret = mlx5_flow_item_acceptable
2558 (item, (const uint8_t *)mask,
2559 (const uint8_t *)valid_mask,
2560 sizeof(struct rte_flow_item_vxlan),
2561 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2565 memcpy(&id.vni[1], spec->vni, 3);
2566 memcpy(&id.vni[1], mask->vni, 3);
2568 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2569 return rte_flow_error_set(error, ENOTSUP,
2570 RTE_FLOW_ERROR_TYPE_ITEM, item,
2571 "VXLAN tunnel must be fully defined");
2576 * Validate VXLAN_GPE item.
2579 * Item specification.
2580 * @param[in] item_flags
2581 * Bit-fields that holds the items detected until now.
2583 * Pointer to the private data structure.
2584 * @param[in] target_protocol
2585 * The next protocol in the previous item.
2587 * Pointer to error structure.
2590 * 0 on success, a negative errno value otherwise and rte_errno is set.
2593 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2594 uint64_t item_flags,
2595 struct rte_eth_dev *dev,
2596 struct rte_flow_error *error)
2598 struct mlx5_priv *priv = dev->data->dev_private;
2599 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2600 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2605 } id = { .vlan_id = 0, };
2607 if (!priv->config.l3_vxlan_en)
2608 return rte_flow_error_set(error, ENOTSUP,
2609 RTE_FLOW_ERROR_TYPE_ITEM, item,
2610 "L3 VXLAN is not enabled by device"
2611 " parameter and/or not configured in"
2613 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2614 return rte_flow_error_set(error, ENOTSUP,
2615 RTE_FLOW_ERROR_TYPE_ITEM, item,
2616 "multiple tunnel layers not"
2619 * Verify only UDPv4 is present as defined in
2620 * https://tools.ietf.org/html/rfc7348
2622 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2623 return rte_flow_error_set(error, EINVAL,
2624 RTE_FLOW_ERROR_TYPE_ITEM, item,
2625 "no outer UDP layer found");
2627 mask = &rte_flow_item_vxlan_gpe_mask;
2628 ret = mlx5_flow_item_acceptable
2629 (item, (const uint8_t *)mask,
2630 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2631 sizeof(struct rte_flow_item_vxlan_gpe),
2632 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2637 return rte_flow_error_set(error, ENOTSUP,
2638 RTE_FLOW_ERROR_TYPE_ITEM,
2640 "VxLAN-GPE protocol"
2642 memcpy(&id.vni[1], spec->vni, 3);
2643 memcpy(&id.vni[1], mask->vni, 3);
2645 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2646 return rte_flow_error_set(error, ENOTSUP,
2647 RTE_FLOW_ERROR_TYPE_ITEM, item,
2648 "VXLAN-GPE tunnel must be fully"
2653 * Validate GRE Key item.
2656 * Item specification.
2657 * @param[in] item_flags
2658 * Bit flags to mark detected items.
2659 * @param[in] gre_item
2660 * Pointer to gre_item
2662 * Pointer to error structure.
2665 * 0 on success, a negative errno value otherwise and rte_errno is set.
2668 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2669 uint64_t item_flags,
2670 const struct rte_flow_item *gre_item,
2671 struct rte_flow_error *error)
2673 const rte_be32_t *mask = item->mask;
2675 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2676 const struct rte_flow_item_gre *gre_spec;
2677 const struct rte_flow_item_gre *gre_mask;
2679 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2680 return rte_flow_error_set(error, ENOTSUP,
2681 RTE_FLOW_ERROR_TYPE_ITEM, item,
2682 "Multiple GRE key not support");
2683 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2684 return rte_flow_error_set(error, ENOTSUP,
2685 RTE_FLOW_ERROR_TYPE_ITEM, item,
2686 "No preceding GRE header");
2687 if (item_flags & MLX5_FLOW_LAYER_INNER)
2688 return rte_flow_error_set(error, ENOTSUP,
2689 RTE_FLOW_ERROR_TYPE_ITEM, item,
2690 "GRE key following a wrong item");
2691 gre_mask = gre_item->mask;
2693 gre_mask = &rte_flow_item_gre_mask;
2694 gre_spec = gre_item->spec;
2695 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2696 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2697 return rte_flow_error_set(error, EINVAL,
2698 RTE_FLOW_ERROR_TYPE_ITEM, item,
2699 "Key bit must be on");
2702 mask = &gre_key_default_mask;
2703 ret = mlx5_flow_item_acceptable
2704 (item, (const uint8_t *)mask,
2705 (const uint8_t *)&gre_key_default_mask,
2706 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2711 * Validate GRE item.
2714 * Item specification.
2715 * @param[in] item_flags
2716 * Bit flags to mark detected items.
2717 * @param[in] target_protocol
2718 * The next protocol in the previous item.
2720 * Pointer to error structure.
2723 * 0 on success, a negative errno value otherwise and rte_errno is set.
2726 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2727 uint64_t item_flags,
2728 uint8_t target_protocol,
2729 struct rte_flow_error *error)
2731 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2732 const struct rte_flow_item_gre *mask = item->mask;
2734 const struct rte_flow_item_gre nic_mask = {
2735 .c_rsvd0_ver = RTE_BE16(0xB000),
2736 .protocol = RTE_BE16(UINT16_MAX),
2739 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2740 return rte_flow_error_set(error, EINVAL,
2741 RTE_FLOW_ERROR_TYPE_ITEM, item,
2742 "protocol filtering not compatible"
2743 " with this GRE layer");
2744 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2745 return rte_flow_error_set(error, ENOTSUP,
2746 RTE_FLOW_ERROR_TYPE_ITEM, item,
2747 "multiple tunnel layers not"
2749 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2750 return rte_flow_error_set(error, ENOTSUP,
2751 RTE_FLOW_ERROR_TYPE_ITEM, item,
2752 "L3 Layer is missing");
2754 mask = &rte_flow_item_gre_mask;
2755 ret = mlx5_flow_item_acceptable
2756 (item, (const uint8_t *)mask,
2757 (const uint8_t *)&nic_mask,
2758 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2762 #ifndef HAVE_MLX5DV_DR
2763 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2764 if (spec && (spec->protocol & mask->protocol))
2765 return rte_flow_error_set(error, ENOTSUP,
2766 RTE_FLOW_ERROR_TYPE_ITEM, item,
2767 "without MPLS support the"
2768 " specification cannot be used for"
2776 * Validate Geneve item.
2779 * Item specification.
2780 * @param[in] itemFlags
2781 * Bit-fields that holds the items detected until now.
2783 * Pointer to the private data structure.
2785 * Pointer to error structure.
2788 * 0 on success, a negative errno value otherwise and rte_errno is set.
2792 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2793 uint64_t item_flags,
2794 struct rte_eth_dev *dev,
2795 struct rte_flow_error *error)
2797 struct mlx5_priv *priv = dev->data->dev_private;
2798 const struct rte_flow_item_geneve *spec = item->spec;
2799 const struct rte_flow_item_geneve *mask = item->mask;
2802 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2803 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2804 const struct rte_flow_item_geneve nic_mask = {
2805 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2806 .vni = "\xff\xff\xff",
2807 .protocol = RTE_BE16(UINT16_MAX),
2810 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2811 return rte_flow_error_set(error, ENOTSUP,
2812 RTE_FLOW_ERROR_TYPE_ITEM, item,
2813 "L3 Geneve is not enabled by device"
2814 " parameter and/or not configured in"
2816 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2817 return rte_flow_error_set(error, ENOTSUP,
2818 RTE_FLOW_ERROR_TYPE_ITEM, item,
2819 "multiple tunnel layers not"
2822 * Verify only UDPv4 is present as defined in
2823 * https://tools.ietf.org/html/rfc7348
2825 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2826 return rte_flow_error_set(error, EINVAL,
2827 RTE_FLOW_ERROR_TYPE_ITEM, item,
2828 "no outer UDP layer found");
2830 mask = &rte_flow_item_geneve_mask;
2831 ret = mlx5_flow_item_acceptable
2832 (item, (const uint8_t *)mask,
2833 (const uint8_t *)&nic_mask,
2834 sizeof(struct rte_flow_item_geneve),
2835 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2839 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2840 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2841 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2842 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2843 return rte_flow_error_set(error, ENOTSUP,
2844 RTE_FLOW_ERROR_TYPE_ITEM,
2846 "Geneve protocol unsupported"
2847 " fields are being used");
2848 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2849 return rte_flow_error_set
2851 RTE_FLOW_ERROR_TYPE_ITEM,
2853 "Unsupported Geneve options length");
2855 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2856 return rte_flow_error_set
2858 RTE_FLOW_ERROR_TYPE_ITEM, item,
2859 "Geneve tunnel must be fully defined");
2864 * Validate Geneve TLV option item.
2867 * Item specification.
2868 * @param[in] last_item
2869 * Previous validated item in the pattern items.
2870 * @param[in] geneve_item
2871 * Previous GENEVE item specification.
2873 * Pointer to the rte_eth_dev structure.
2875 * Pointer to error structure.
2878 * 0 on success, a negative errno value otherwise and rte_errno is set.
2881 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2883 const struct rte_flow_item *geneve_item,
2884 struct rte_eth_dev *dev,
2885 struct rte_flow_error *error)
2887 struct mlx5_priv *priv = dev->data->dev_private;
2888 struct mlx5_dev_ctx_shared *sh = priv->sh;
2889 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2890 struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2891 uint8_t data_max_supported =
2892 hca_attr->max_geneve_tlv_option_data_len * 4;
2893 struct mlx5_dev_config *config = &priv->config;
2894 const struct rte_flow_item_geneve *geneve_spec;
2895 const struct rte_flow_item_geneve *geneve_mask;
2896 const struct rte_flow_item_geneve_opt *spec = item->spec;
2897 const struct rte_flow_item_geneve_opt *mask = item->mask;
2899 unsigned int data_len;
2900 uint8_t tlv_option_len;
2901 uint16_t optlen_m, optlen_v;
2902 const struct rte_flow_item_geneve_opt full_mask = {
2903 .option_class = RTE_BE16(0xffff),
2904 .option_type = 0xff,
2909 mask = &rte_flow_item_geneve_opt_mask;
2911 return rte_flow_error_set
2912 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2913 "Geneve TLV opt class/type/length must be specified");
2914 if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2915 return rte_flow_error_set
2916 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2917 "Geneve TLV opt length exceeeds the limit (31)");
2918 /* Check if class type and length masks are full. */
2919 if (full_mask.option_class != mask->option_class ||
2920 full_mask.option_type != mask->option_type ||
2921 full_mask.option_len != (mask->option_len & full_mask.option_len))
2922 return rte_flow_error_set
2923 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2924 "Geneve TLV opt class/type/length masks must be full");
2925 /* Check if length is supported */
2926 if ((uint32_t)spec->option_len >
2927 config->hca_attr.max_geneve_tlv_option_data_len)
2928 return rte_flow_error_set
2929 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2930 "Geneve TLV opt length not supported");
2931 if (config->hca_attr.max_geneve_tlv_options > 1)
2933 "max_geneve_tlv_options supports more than 1 option");
2934 /* Check GENEVE item preceding. */
2935 if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2936 return rte_flow_error_set
2937 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2938 "Geneve opt item must be preceded with Geneve item");
2939 geneve_spec = geneve_item->spec;
2940 geneve_mask = geneve_item->mask ? geneve_item->mask :
2941 &rte_flow_item_geneve_mask;
2942 /* Check if GENEVE TLV option size doesn't exceed option length */
2943 if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2944 geneve_spec->ver_opt_len_o_c_rsvd0)) {
2945 tlv_option_len = spec->option_len & mask->option_len;
2946 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2947 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2948 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2949 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2950 if ((optlen_v & optlen_m) <= tlv_option_len)
2951 return rte_flow_error_set
2952 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2953 "GENEVE TLV option length exceeds optlen");
2955 /* Check if length is 0 or data is 0. */
2956 if (spec->data == NULL || spec->option_len == 0)
2957 return rte_flow_error_set
2958 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2959 "Geneve TLV opt with zero data/length not supported");
2960 /* Check not all data & mask are 0. */
2961 data_len = spec->option_len * 4;
2962 if (mask->data == NULL) {
2963 for (i = 0; i < data_len; i++)
2967 return rte_flow_error_set(error, ENOTSUP,
2968 RTE_FLOW_ERROR_TYPE_ITEM, item,
2969 "Can't match on Geneve option data 0");
2971 for (i = 0; i < data_len; i++)
2972 if (spec->data[i] & mask->data[i])
2975 return rte_flow_error_set(error, ENOTSUP,
2976 RTE_FLOW_ERROR_TYPE_ITEM, item,
2977 "Can't match on Geneve option data and mask 0");
2978 /* Check data mask supported. */
2979 for (i = data_max_supported; i < data_len ; i++)
2981 return rte_flow_error_set(error, ENOTSUP,
2982 RTE_FLOW_ERROR_TYPE_ITEM, item,
2983 "Data mask is of unsupported size");
2985 /* Check GENEVE option is supported in NIC. */
2986 if (!config->hca_attr.geneve_tlv_opt)
2987 return rte_flow_error_set
2988 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2989 "Geneve TLV opt not supported");
2990 /* Check if we already have geneve option with different type/class. */
2991 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2992 geneve_opt_resource = sh->geneve_tlv_option_resource;
2993 if (geneve_opt_resource != NULL)
2994 if (geneve_opt_resource->option_class != spec->option_class ||
2995 geneve_opt_resource->option_type != spec->option_type ||
2996 geneve_opt_resource->length != spec->option_len) {
2997 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2998 return rte_flow_error_set(error, ENOTSUP,
2999 RTE_FLOW_ERROR_TYPE_ITEM, item,
3000 "Only one Geneve TLV option supported");
3002 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3007 * Validate MPLS item.
3010 * Pointer to the rte_eth_dev structure.
3012 * Item specification.
3013 * @param[in] item_flags
3014 * Bit-fields that holds the items detected until now.
3015 * @param[in] prev_layer
3016 * The protocol layer indicated in previous item.
3018 * Pointer to error structure.
3021 * 0 on success, a negative errno value otherwise and rte_errno is set.
3024 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3025 const struct rte_flow_item *item __rte_unused,
3026 uint64_t item_flags __rte_unused,
3027 uint64_t prev_layer __rte_unused,
3028 struct rte_flow_error *error)
3030 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3031 const struct rte_flow_item_mpls *mask = item->mask;
3032 struct mlx5_priv *priv = dev->data->dev_private;
3035 if (!priv->config.mpls_en)
3036 return rte_flow_error_set(error, ENOTSUP,
3037 RTE_FLOW_ERROR_TYPE_ITEM, item,
3038 "MPLS not supported or"
3039 " disabled in firmware"
3041 /* MPLS over UDP, GRE is allowed */
3042 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3043 MLX5_FLOW_LAYER_GRE |
3044 MLX5_FLOW_LAYER_GRE_KEY)))
3045 return rte_flow_error_set(error, EINVAL,
3046 RTE_FLOW_ERROR_TYPE_ITEM, item,
3047 "protocol filtering not compatible"
3048 " with MPLS layer");
3049 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3050 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3051 !(item_flags & MLX5_FLOW_LAYER_GRE))
3052 return rte_flow_error_set(error, ENOTSUP,
3053 RTE_FLOW_ERROR_TYPE_ITEM, item,
3054 "multiple tunnel layers not"
3057 mask = &rte_flow_item_mpls_mask;
3058 ret = mlx5_flow_item_acceptable
3059 (item, (const uint8_t *)mask,
3060 (const uint8_t *)&rte_flow_item_mpls_mask,
3061 sizeof(struct rte_flow_item_mpls),
3062 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3067 return rte_flow_error_set(error, ENOTSUP,
3068 RTE_FLOW_ERROR_TYPE_ITEM, item,
3069 "MPLS is not supported by Verbs, please"
3075 * Validate NVGRE item.
3078 * Item specification.
3079 * @param[in] item_flags
3080 * Bit flags to mark detected items.
3081 * @param[in] target_protocol
3082 * The next protocol in the previous item.
3084 * Pointer to error structure.
3087 * 0 on success, a negative errno value otherwise and rte_errno is set.
3090 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3091 uint64_t item_flags,
3092 uint8_t target_protocol,
3093 struct rte_flow_error *error)
3095 const struct rte_flow_item_nvgre *mask = item->mask;
3098 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3099 return rte_flow_error_set(error, EINVAL,
3100 RTE_FLOW_ERROR_TYPE_ITEM, item,
3101 "protocol filtering not compatible"
3102 " with this GRE layer");
3103 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3104 return rte_flow_error_set(error, ENOTSUP,
3105 RTE_FLOW_ERROR_TYPE_ITEM, item,
3106 "multiple tunnel layers not"
3108 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3109 return rte_flow_error_set(error, ENOTSUP,
3110 RTE_FLOW_ERROR_TYPE_ITEM, item,
3111 "L3 Layer is missing");
3113 mask = &rte_flow_item_nvgre_mask;
3114 ret = mlx5_flow_item_acceptable
3115 (item, (const uint8_t *)mask,
3116 (const uint8_t *)&rte_flow_item_nvgre_mask,
3117 sizeof(struct rte_flow_item_nvgre),
3118 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3125 * Validate eCPRI item.
3128 * Item specification.
3129 * @param[in] item_flags
3130 * Bit-fields that holds the items detected until now.
3131 * @param[in] last_item
3132 * Previous validated item in the pattern items.
3133 * @param[in] ether_type
3134 * Type in the ethernet layer header (including dot1q).
3135 * @param[in] acc_mask
3136 * Acceptable mask, if NULL default internal default mask
3137 * will be used to check whether item fields are supported.
3139 * Pointer to error structure.
3142 * 0 on success, a negative errno value otherwise and rte_errno is set.
3145 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3146 uint64_t item_flags,
3148 uint16_t ether_type,
3149 const struct rte_flow_item_ecpri *acc_mask,
3150 struct rte_flow_error *error)
3152 const struct rte_flow_item_ecpri *mask = item->mask;
3153 const struct rte_flow_item_ecpri nic_mask = {
3157 RTE_BE32(((const struct rte_ecpri_common_hdr) {
3161 .dummy[0] = 0xFFFFFFFF,
3164 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3165 MLX5_FLOW_LAYER_OUTER_VLAN);
3166 struct rte_flow_item_ecpri mask_lo;
3168 if (!(last_item & outer_l2_vlan) &&
3169 last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3170 return rte_flow_error_set(error, EINVAL,
3171 RTE_FLOW_ERROR_TYPE_ITEM, item,
3172 "eCPRI can only follow L2/VLAN layer or UDP layer");
3173 if ((last_item & outer_l2_vlan) && ether_type &&
3174 ether_type != RTE_ETHER_TYPE_ECPRI)
3175 return rte_flow_error_set(error, EINVAL,
3176 RTE_FLOW_ERROR_TYPE_ITEM, item,
3177 "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3178 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3179 return rte_flow_error_set(error, EINVAL,
3180 RTE_FLOW_ERROR_TYPE_ITEM, item,
3181 "eCPRI with tunnel is not supported right now");
3182 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3183 return rte_flow_error_set(error, ENOTSUP,
3184 RTE_FLOW_ERROR_TYPE_ITEM, item,
3185 "multiple L3 layers not supported");
3186 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3187 return rte_flow_error_set(error, EINVAL,
3188 RTE_FLOW_ERROR_TYPE_ITEM, item,
3189 "eCPRI cannot coexist with a TCP layer");
3190 /* In specification, eCPRI could be over UDP layer. */
3191 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3192 return rte_flow_error_set(error, EINVAL,
3193 RTE_FLOW_ERROR_TYPE_ITEM, item,
3194 "eCPRI over UDP layer is not yet supported right now");
3195 /* Mask for type field in common header could be zero. */
3197 mask = &rte_flow_item_ecpri_mask;
3198 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3199 /* Input mask is in big-endian format. */
3200 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3201 return rte_flow_error_set(error, EINVAL,
3202 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3203 "partial mask is not supported for protocol");
3204 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3205 return rte_flow_error_set(error, EINVAL,
3206 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3207 "message header mask must be after a type mask");
3208 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3209 acc_mask ? (const uint8_t *)acc_mask
3210 : (const uint8_t *)&nic_mask,
3211 sizeof(struct rte_flow_item_ecpri),
3212 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3216 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3217 const struct rte_flow_attr *attr __rte_unused,
3218 const struct rte_flow_item items[] __rte_unused,
3219 const struct rte_flow_action actions[] __rte_unused,
3220 bool external __rte_unused,
3221 int hairpin __rte_unused,
3222 struct rte_flow_error *error)
3224 return rte_flow_error_set(error, ENOTSUP,
3225 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3228 static struct mlx5_flow *
3229 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3230 const struct rte_flow_attr *attr __rte_unused,
3231 const struct rte_flow_item items[] __rte_unused,
3232 const struct rte_flow_action actions[] __rte_unused,
3233 struct rte_flow_error *error)
3235 rte_flow_error_set(error, ENOTSUP,
3236 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3241 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3242 struct mlx5_flow *dev_flow __rte_unused,
3243 const struct rte_flow_attr *attr __rte_unused,
3244 const struct rte_flow_item items[] __rte_unused,
3245 const struct rte_flow_action actions[] __rte_unused,
3246 struct rte_flow_error *error)
3248 return rte_flow_error_set(error, ENOTSUP,
3249 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3253 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3254 struct rte_flow *flow __rte_unused,
3255 struct rte_flow_error *error)
3257 return rte_flow_error_set(error, ENOTSUP,
3258 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3262 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3263 struct rte_flow *flow __rte_unused)
3268 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3269 struct rte_flow *flow __rte_unused)
3274 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3275 struct rte_flow *flow __rte_unused,
3276 const struct rte_flow_action *actions __rte_unused,
3277 void *data __rte_unused,
3278 struct rte_flow_error *error)
3280 return rte_flow_error_set(error, ENOTSUP,
3281 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3285 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3286 uint32_t domains __rte_unused,
3287 uint32_t flags __rte_unused)
3292 /* Void driver to protect from null pointer reference. */
3293 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3294 .validate = flow_null_validate,
3295 .prepare = flow_null_prepare,
3296 .translate = flow_null_translate,
3297 .apply = flow_null_apply,
3298 .remove = flow_null_remove,
3299 .destroy = flow_null_destroy,
3300 .query = flow_null_query,
3301 .sync_domain = flow_null_sync_domain,
3305 * Select flow driver type according to flow attributes and device
3309 * Pointer to the dev structure.
3311 * Pointer to the flow attributes.
3314 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3316 static enum mlx5_flow_drv_type
3317 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3319 struct mlx5_priv *priv = dev->data->dev_private;
3320 /* The OS can determine first a specific flow type (DV, VERBS) */
3321 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3323 if (type != MLX5_FLOW_TYPE_MAX)
3325 /* If no OS specific type - continue with DV/VERBS selection */
3326 if (attr->transfer && priv->config.dv_esw_en)
3327 type = MLX5_FLOW_TYPE_DV;
3328 if (!attr->transfer)
3329 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3330 MLX5_FLOW_TYPE_VERBS;
3334 #define flow_get_drv_ops(type) flow_drv_ops[type]
3337 * Flow driver validation API. This abstracts calling driver specific functions.
3338 * The type of flow driver is determined according to flow attributes.
3341 * Pointer to the dev structure.
3343 * Pointer to the flow attributes.
3345 * Pointer to the list of items.
3346 * @param[in] actions
3347 * Pointer to the list of actions.
3348 * @param[in] external
3349 * This flow rule is created by request external to PMD.
3350 * @param[in] hairpin
3351 * Number of hairpin TX actions, 0 means classic flow.
3353 * Pointer to the error structure.
3356 * 0 on success, a negative errno value otherwise and rte_errno is set.
3359 flow_drv_validate(struct rte_eth_dev *dev,
3360 const struct rte_flow_attr *attr,
3361 const struct rte_flow_item items[],
3362 const struct rte_flow_action actions[],
3363 bool external, int hairpin, struct rte_flow_error *error)
3365 const struct mlx5_flow_driver_ops *fops;
3366 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3368 fops = flow_get_drv_ops(type);
3369 return fops->validate(dev, attr, items, actions, external,
3374 * Flow driver preparation API. This abstracts calling driver specific
3375 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3376 * calculates the size of memory required for device flow, allocates the memory,
3377 * initializes the device flow and returns the pointer.
3380 * This function initializes device flow structure such as dv or verbs in
3381 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3382 * rest. For example, adding returning device flow to flow->dev_flow list and
3383 * setting backward reference to the flow should be done out of this function.
3384 * layers field is not filled either.
3387 * Pointer to the dev structure.
3389 * Pointer to the flow attributes.
3391 * Pointer to the list of items.
3392 * @param[in] actions
3393 * Pointer to the list of actions.
3394 * @param[in] flow_idx
3395 * This memory pool index to the flow.
3397 * Pointer to the error structure.
3400 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3402 static inline struct mlx5_flow *
3403 flow_drv_prepare(struct rte_eth_dev *dev,
3404 const struct rte_flow *flow,
3405 const struct rte_flow_attr *attr,
3406 const struct rte_flow_item items[],
3407 const struct rte_flow_action actions[],
3409 struct rte_flow_error *error)
3411 const struct mlx5_flow_driver_ops *fops;
3412 enum mlx5_flow_drv_type type = flow->drv_type;
3413 struct mlx5_flow *mlx5_flow = NULL;
3415 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3416 fops = flow_get_drv_ops(type);
3417 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3419 mlx5_flow->flow_idx = flow_idx;
3424 * Flow driver translation API. This abstracts calling driver specific
3425 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3426 * translates a generic flow into a driver flow. flow_drv_prepare() must
3430 * dev_flow->layers could be filled as a result of parsing during translation
3431 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3432 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3433 * flow->actions could be overwritten even though all the expanded dev_flows
3434 * have the same actions.
3437 * Pointer to the rte dev structure.
3438 * @param[in, out] dev_flow
3439 * Pointer to the mlx5 flow.
3441 * Pointer to the flow attributes.
3443 * Pointer to the list of items.
3444 * @param[in] actions
3445 * Pointer to the list of actions.
3447 * Pointer to the error structure.
3450 * 0 on success, a negative errno value otherwise and rte_errno is set.
3453 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3454 const struct rte_flow_attr *attr,
3455 const struct rte_flow_item items[],
3456 const struct rte_flow_action actions[],
3457 struct rte_flow_error *error)
3459 const struct mlx5_flow_driver_ops *fops;
3460 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3462 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3463 fops = flow_get_drv_ops(type);
3464 return fops->translate(dev, dev_flow, attr, items, actions, error);
3468 * Flow driver apply API. This abstracts calling driver specific functions.
3469 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3470 * translated driver flows on to device. flow_drv_translate() must precede.
3473 * Pointer to Ethernet device structure.
3474 * @param[in, out] flow
3475 * Pointer to flow structure.
3477 * Pointer to error structure.
3480 * 0 on success, a negative errno value otherwise and rte_errno is set.
3483 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3484 struct rte_flow_error *error)
3486 const struct mlx5_flow_driver_ops *fops;
3487 enum mlx5_flow_drv_type type = flow->drv_type;
3489 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3490 fops = flow_get_drv_ops(type);
3491 return fops->apply(dev, flow, error);
3495 * Flow driver destroy API. This abstracts calling driver specific functions.
3496 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3497 * on device and releases resources of the flow.
3500 * Pointer to Ethernet device.
3501 * @param[in, out] flow
3502 * Pointer to flow structure.
3505 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3507 const struct mlx5_flow_driver_ops *fops;
3508 enum mlx5_flow_drv_type type = flow->drv_type;
3510 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3511 fops = flow_get_drv_ops(type);
3512 fops->destroy(dev, flow);
3516 * Flow driver find RSS policy tbl API. This abstracts calling driver
3517 * specific functions. Parent flow (rte_flow) should have driver
3518 * type (drv_type). It will find the RSS policy table that has the rss_desc.
3521 * Pointer to Ethernet device.
3522 * @param[in, out] flow
3523 * Pointer to flow structure.
3525 * Pointer to meter policy table.
3526 * @param[in] rss_desc
3527 * Pointer to rss_desc
3529 static struct mlx5_flow_meter_sub_policy *
3530 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3531 struct rte_flow *flow,
3532 struct mlx5_flow_meter_policy *policy,
3533 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3535 const struct mlx5_flow_driver_ops *fops;
3536 enum mlx5_flow_drv_type type = flow->drv_type;
3538 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3539 fops = flow_get_drv_ops(type);
3540 return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3544 * Flow driver color tag rule API. This abstracts calling driver
3545 * specific functions. Parent flow (rte_flow) should have driver
3546 * type (drv_type). It will create the color tag rules in hierarchy meter.
3549 * Pointer to Ethernet device.
3550 * @param[in, out] flow
3551 * Pointer to flow structure.
3553 * Pointer to flow meter structure.
3554 * @param[in] src_port
3555 * The src port this extra rule should use.
3557 * The src port id match item.
3559 * Pointer to error structure.
3562 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3563 struct rte_flow *flow,
3564 struct mlx5_flow_meter_info *fm,
3566 const struct rte_flow_item *item,
3567 struct rte_flow_error *error)
3569 const struct mlx5_flow_driver_ops *fops;
3570 enum mlx5_flow_drv_type type = flow->drv_type;
3572 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3573 fops = flow_get_drv_ops(type);
3574 return fops->meter_hierarchy_rule_create(dev, fm,
3575 src_port, item, error);
3579 * Get RSS action from the action list.
3582 * Pointer to Ethernet device.
3583 * @param[in] actions
3584 * Pointer to the list of actions.
3586 * Parent flow structure pointer.
3589 * Pointer to the RSS action if exist, else return NULL.
3591 static const struct rte_flow_action_rss*
3592 flow_get_rss_action(struct rte_eth_dev *dev,
3593 const struct rte_flow_action actions[])
3595 struct mlx5_priv *priv = dev->data->dev_private;
3596 const struct rte_flow_action_rss *rss = NULL;
3598 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3599 switch (actions->type) {
3600 case RTE_FLOW_ACTION_TYPE_RSS:
3601 rss = actions->conf;
3603 case RTE_FLOW_ACTION_TYPE_SAMPLE:
3605 const struct rte_flow_action_sample *sample =
3607 const struct rte_flow_action *act = sample->actions;
3608 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3609 if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3613 case RTE_FLOW_ACTION_TYPE_METER:
3616 struct mlx5_flow_meter_info *fm;
3617 struct mlx5_flow_meter_policy *policy;
3618 const struct rte_flow_action_meter *mtr = actions->conf;
3620 fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3621 if (fm && !fm->def_policy) {
3622 policy = mlx5_flow_meter_policy_find(dev,
3623 fm->policy_id, NULL);
3624 MLX5_ASSERT(policy);
3625 if (policy->is_hierarchy) {
3627 mlx5_flow_meter_hierarchy_get_final_policy(dev,
3634 policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3646 * Get ASO age action by index.
3649 * Pointer to the Ethernet device structure.
3650 * @param[in] age_idx
3651 * Index to the ASO age action.
3654 * The specified ASO age action.
3656 struct mlx5_aso_age_action*
3657 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3659 uint16_t pool_idx = age_idx & UINT16_MAX;
3660 uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3661 struct mlx5_priv *priv = dev->data->dev_private;
3662 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3663 struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3665 return &pool->actions[offset - 1];
3668 /* maps indirect action to translated direct in some actions array */
3669 struct mlx5_translated_action_handle {
3670 struct rte_flow_action_handle *action; /**< Indirect action handle. */
3671 int index; /**< Index in related array of rte_flow_action. */
3675 * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3676 * direct action if translation possible.
3677 * This functionality used to run same execution path for both direct and
3678 * indirect actions on flow create. All necessary preparations for indirect
3679 * action handling should be performed on *handle* actions list returned
3683 * Pointer to Ethernet device.
3684 * @param[in] actions
3685 * List of actions to translate.
3686 * @param[out] handle
3687 * List to store translated indirect action object handles.
3688 * @param[in, out] indir_n
3689 * Size of *handle* array. On return should be updated with number of
3690 * indirect actions retrieved from the *actions* list.
3691 * @param[out] translated_actions
3692 * List of actions where all indirect actions were translated to direct
3693 * if possible. NULL if no translation took place.
3695 * Pointer to the error structure.
3698 * 0 on success, a negative errno value otherwise and rte_errno is set.
3701 flow_action_handles_translate(struct rte_eth_dev *dev,
3702 const struct rte_flow_action actions[],
3703 struct mlx5_translated_action_handle *handle,
3705 struct rte_flow_action **translated_actions,
3706 struct rte_flow_error *error)
3708 struct mlx5_priv *priv = dev->data->dev_private;
3709 struct rte_flow_action *translated = NULL;
3710 size_t actions_size;
3713 struct mlx5_translated_action_handle *handle_end = NULL;
3715 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3716 if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3718 if (copied_n == *indir_n) {
3719 return rte_flow_error_set
3720 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3721 NULL, "too many shared actions");
3723 rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3724 sizeof(actions[n].conf));
3725 handle[copied_n].index = n;
3729 *indir_n = copied_n;
3732 actions_size = sizeof(struct rte_flow_action) * n;
3733 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3738 memcpy(translated, actions, actions_size);
3739 for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3740 struct mlx5_shared_action_rss *shared_rss;
3741 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3742 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3743 uint32_t idx = act_idx &
3744 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3747 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3748 shared_rss = mlx5_ipool_get
3749 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3750 translated[handle->index].type =
3751 RTE_FLOW_ACTION_TYPE_RSS;
3752 translated[handle->index].conf =
3753 &shared_rss->origin;
3755 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3756 translated[handle->index].type =
3757 (enum rte_flow_action_type)
3758 MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3759 translated[handle->index].conf = (void *)(uintptr_t)idx;
3761 case MLX5_INDIRECT_ACTION_TYPE_AGE:
3762 if (priv->sh->flow_hit_aso_en) {
3763 translated[handle->index].type =
3764 (enum rte_flow_action_type)
3765 MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3766 translated[handle->index].conf =
3767 (void *)(uintptr_t)idx;
3771 case MLX5_INDIRECT_ACTION_TYPE_CT:
3772 if (priv->sh->ct_aso_en) {
3773 translated[handle->index].type =
3774 RTE_FLOW_ACTION_TYPE_CONNTRACK;
3775 translated[handle->index].conf =
3776 (void *)(uintptr_t)idx;
3781 mlx5_free(translated);
3782 return rte_flow_error_set
3783 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3784 NULL, "invalid indirect action type");
3787 *translated_actions = translated;
3792 * Get Shared RSS action from the action list.
3795 * Pointer to Ethernet device.
3797 * Pointer to the list of actions.
3798 * @param[in] shared_n
3799 * Actions list length.
3802 * The MLX5 RSS action ID if exists, otherwise return 0.
3805 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3806 struct mlx5_translated_action_handle *handle,
3809 struct mlx5_translated_action_handle *handle_end;
3810 struct mlx5_priv *priv = dev->data->dev_private;
3811 struct mlx5_shared_action_rss *shared_rss;
3814 for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3815 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3816 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3817 uint32_t idx = act_idx &
3818 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3820 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3821 shared_rss = mlx5_ipool_get
3822 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3824 __atomic_add_fetch(&shared_rss->refcnt, 1,
3835 find_graph_root(uint32_t rss_level)
3837 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3838 MLX5_EXPANSION_ROOT_OUTER;
3842 * Get layer flags from the prefix flow.
3844 * Some flows may be split to several subflows, the prefix subflow gets the
3845 * match items and the suffix sub flow gets the actions.
3846 * Some actions need the user defined match item flags to get the detail for
3848 * This function helps the suffix flow to get the item layer flags from prefix
3851 * @param[in] dev_flow
3852 * Pointer the created preifx subflow.
3855 * The layers get from prefix subflow.
3857 static inline uint64_t
3858 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3860 uint64_t layers = 0;
3863 * Layers bits could be localization, but usually the compiler will
3864 * help to do the optimization work for source code.
3865 * If no decap actions, use the layers directly.
3867 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3868 return dev_flow->handle->layers;
3869 /* Convert L3 layers with decap action. */
3870 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3871 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3872 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3873 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3874 /* Convert L4 layers with decap action. */
3875 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3876 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3877 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3878 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3883 * Get metadata split action information.
3885 * @param[in] actions
3886 * Pointer to the list of actions.
3888 * Pointer to the return pointer.
3889 * @param[out] qrss_type
3890 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3891 * if no QUEUE/RSS is found.
3892 * @param[out] encap_idx
3893 * Pointer to the index of the encap action if exists, otherwise the last
3897 * Total number of actions.
3900 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3901 const struct rte_flow_action **qrss,
3904 const struct rte_flow_action_raw_encap *raw_encap;
3906 int raw_decap_idx = -1;
3909 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3910 switch (actions->type) {
3911 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3912 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3913 *encap_idx = actions_n;
3915 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3916 raw_decap_idx = actions_n;
3918 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3919 raw_encap = actions->conf;
3920 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3921 *encap_idx = raw_decap_idx != -1 ?
3922 raw_decap_idx : actions_n;
3924 case RTE_FLOW_ACTION_TYPE_QUEUE:
3925 case RTE_FLOW_ACTION_TYPE_RSS:
3933 if (*encap_idx == -1)
3934 *encap_idx = actions_n;
3935 /* Count RTE_FLOW_ACTION_TYPE_END. */
3936 return actions_n + 1;
3940 * Check if the action will change packet.
3943 * Pointer to Ethernet device.
3948 * true if action will change packet, false otherwise.
3950 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3951 enum rte_flow_action_type type)
3953 struct mlx5_priv *priv = dev->data->dev_private;
3956 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3957 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3958 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3959 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3960 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3961 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3962 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3963 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3964 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3965 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3966 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3967 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3968 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3969 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3970 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3971 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3972 case RTE_FLOW_ACTION_TYPE_SET_META:
3973 case RTE_FLOW_ACTION_TYPE_SET_TAG:
3974 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3975 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3976 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3977 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3978 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3979 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3980 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3981 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3982 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3983 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3984 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3986 case RTE_FLOW_ACTION_TYPE_FLAG:
3987 case RTE_FLOW_ACTION_TYPE_MARK:
3988 if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3998 * Check meter action from the action list.
4001 * Pointer to Ethernet device.
4002 * @param[in] actions
4003 * Pointer to the list of actions.
4004 * @param[out] has_mtr
4005 * Pointer to the meter exist flag.
4006 * @param[out] has_modify
4007 * Pointer to the flag showing there's packet change action.
4008 * @param[out] meter_id
4009 * Pointer to the meter id.
4012 * Total number of actions.
4015 flow_check_meter_action(struct rte_eth_dev *dev,
4016 const struct rte_flow_action actions[],
4017 bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4019 const struct rte_flow_action_meter *mtr = NULL;
4022 MLX5_ASSERT(has_mtr);
4024 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4025 switch (actions->type) {
4026 case RTE_FLOW_ACTION_TYPE_METER:
4027 mtr = actions->conf;
4028 *meter_id = mtr->mtr_id;
4035 *has_modify |= flow_check_modify_action_type(dev,
4039 /* Count RTE_FLOW_ACTION_TYPE_END. */
4040 return actions_n + 1;
4044 * Check if the flow should be split due to hairpin.
4045 * The reason for the split is that in current HW we can't
4046 * support encap and push-vlan on Rx, so if a flow contains
4047 * these actions we move it to Tx.
4050 * Pointer to Ethernet device.
4052 * Flow rule attributes.
4053 * @param[in] actions
4054 * Associated actions (list terminated by the END action).
4057 * > 0 the number of actions and the flow should be split,
4058 * 0 when no split required.
4061 flow_check_hairpin_split(struct rte_eth_dev *dev,
4062 const struct rte_flow_attr *attr,
4063 const struct rte_flow_action actions[])
4065 int queue_action = 0;
4068 const struct rte_flow_action_queue *queue;
4069 const struct rte_flow_action_rss *rss;
4070 const struct rte_flow_action_raw_encap *raw_encap;
4071 const struct rte_eth_hairpin_conf *conf;
4075 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4076 switch (actions->type) {
4077 case RTE_FLOW_ACTION_TYPE_QUEUE:
4078 queue = actions->conf;
4081 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4082 if (conf == NULL || conf->tx_explicit != 0)
4087 case RTE_FLOW_ACTION_TYPE_RSS:
4088 rss = actions->conf;
4089 if (rss == NULL || rss->queue_num == 0)
4091 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4092 if (conf == NULL || conf->tx_explicit != 0)
4097 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4098 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4099 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4100 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4101 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4105 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4106 raw_encap = actions->conf;
4107 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4116 if (split && queue_action)
4121 /* Declare flow create/destroy prototype in advance. */
4123 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4124 const struct rte_flow_attr *attr,
4125 const struct rte_flow_item items[],
4126 const struct rte_flow_action actions[],
4127 bool external, struct rte_flow_error *error);
4130 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4134 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4135 struct mlx5_list_entry *entry, void *cb_ctx)
4137 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4138 struct mlx5_flow_mreg_copy_resource *mcp_res =
4139 container_of(entry, typeof(*mcp_res), hlist_ent);
4141 return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4144 struct mlx5_list_entry *
4145 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4147 struct rte_eth_dev *dev = tool_ctx;
4148 struct mlx5_priv *priv = dev->data->dev_private;
4149 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4150 struct mlx5_flow_mreg_copy_resource *mcp_res;
4151 struct rte_flow_error *error = ctx->error;
4154 uint32_t mark_id = *(uint32_t *)(ctx->data);
4155 struct rte_flow_attr attr = {
4156 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4159 struct mlx5_rte_flow_item_tag tag_spec = {
4162 struct rte_flow_item items[] = {
4163 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4165 struct rte_flow_action_mark ftag = {
4168 struct mlx5_flow_action_copy_mreg cp_mreg = {
4172 struct rte_flow_action_jump jump = {
4173 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4175 struct rte_flow_action actions[] = {
4176 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4179 /* Fill the register fileds in the flow. */
4180 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4184 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4188 /* Provide the full width of FLAG specific value. */
4189 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4190 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4191 /* Build a new flow. */
4192 if (mark_id != MLX5_DEFAULT_COPY_ID) {
4193 items[0] = (struct rte_flow_item){
4194 .type = (enum rte_flow_item_type)
4195 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4198 items[1] = (struct rte_flow_item){
4199 .type = RTE_FLOW_ITEM_TYPE_END,
4201 actions[0] = (struct rte_flow_action){
4202 .type = (enum rte_flow_action_type)
4203 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4206 actions[1] = (struct rte_flow_action){
4207 .type = (enum rte_flow_action_type)
4208 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4211 actions[2] = (struct rte_flow_action){
4212 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4215 actions[3] = (struct rte_flow_action){
4216 .type = RTE_FLOW_ACTION_TYPE_END,
4219 /* Default rule, wildcard match. */
4220 attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4221 items[0] = (struct rte_flow_item){
4222 .type = RTE_FLOW_ITEM_TYPE_END,
4224 actions[0] = (struct rte_flow_action){
4225 .type = (enum rte_flow_action_type)
4226 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4229 actions[1] = (struct rte_flow_action){
4230 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4233 actions[2] = (struct rte_flow_action){
4234 .type = RTE_FLOW_ACTION_TYPE_END,
4237 /* Build a new entry. */
4238 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4244 mcp_res->mark_id = mark_id;
4246 * The copy Flows are not included in any list. There
4247 * ones are referenced from other Flows and can not
4248 * be applied, removed, deleted in ardbitrary order
4249 * by list traversing.
4251 mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4252 &attr, items, actions, false, error);
4253 if (!mcp_res->rix_flow) {
4254 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4257 return &mcp_res->hlist_ent;
4260 struct mlx5_list_entry *
4261 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4262 void *cb_ctx __rte_unused)
4264 struct rte_eth_dev *dev = tool_ctx;
4265 struct mlx5_priv *priv = dev->data->dev_private;
4266 struct mlx5_flow_mreg_copy_resource *mcp_res;
4269 mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4274 memcpy(mcp_res, oentry, sizeof(*mcp_res));
4276 return &mcp_res->hlist_ent;
4280 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4282 struct mlx5_flow_mreg_copy_resource *mcp_res =
4283 container_of(entry, typeof(*mcp_res), hlist_ent);
4284 struct rte_eth_dev *dev = tool_ctx;
4285 struct mlx5_priv *priv = dev->data->dev_private;
4287 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4291 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4293 * As mark_id is unique, if there's already a registered flow for the mark_id,
4294 * return by increasing the reference counter of the resource. Otherwise, create
4295 * the resource (mcp_res) and flow.
4298 * - If ingress port is ANY and reg_c[1] is mark_id,
4299 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4301 * For default flow (zero mark_id), flow is like,
4302 * - If ingress port is ANY,
4303 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
4306 * Pointer to Ethernet device.
4308 * ID of MARK action, zero means default flow for META.
4310 * Perform verbose error reporting if not NULL.
4313 * Associated resource on success, NULL otherwise and rte_errno is set.
4315 static struct mlx5_flow_mreg_copy_resource *
4316 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4317 struct rte_flow_error *error)
4319 struct mlx5_priv *priv = dev->data->dev_private;
4320 struct mlx5_list_entry *entry;
4321 struct mlx5_flow_cb_ctx ctx = {
4327 /* Check if already registered. */
4328 MLX5_ASSERT(priv->mreg_cp_tbl);
4329 entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4332 return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4337 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4339 struct mlx5_flow_mreg_copy_resource *mcp_res =
4340 container_of(entry, typeof(*mcp_res), hlist_ent);
4341 struct rte_eth_dev *dev = tool_ctx;
4342 struct mlx5_priv *priv = dev->data->dev_private;
4344 MLX5_ASSERT(mcp_res->rix_flow);
4345 flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4346 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4350 * Release flow in RX_CP_TBL.
4353 * Pointer to Ethernet device.
4355 * Parent flow for wich copying is provided.
4358 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4359 struct rte_flow *flow)
4361 struct mlx5_flow_mreg_copy_resource *mcp_res;
4362 struct mlx5_priv *priv = dev->data->dev_private;
4364 if (!flow->rix_mreg_copy)
4366 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4367 flow->rix_mreg_copy);
4368 if (!mcp_res || !priv->mreg_cp_tbl)
4370 MLX5_ASSERT(mcp_res->rix_flow);
4371 mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4372 flow->rix_mreg_copy = 0;
4376 * Remove the default copy action from RX_CP_TBL.
4378 * This functions is called in the mlx5_dev_start(). No thread safe
4382 * Pointer to Ethernet device.
4385 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4387 struct mlx5_list_entry *entry;
4388 struct mlx5_priv *priv = dev->data->dev_private;
4389 struct mlx5_flow_cb_ctx ctx;
4392 /* Check if default flow is registered. */
4393 if (!priv->mreg_cp_tbl)
4395 mark_id = MLX5_DEFAULT_COPY_ID;
4396 ctx.data = &mark_id;
4397 entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4400 mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4404 * Add the default copy action in in RX_CP_TBL.
4406 * This functions is called in the mlx5_dev_start(). No thread safe
4410 * Pointer to Ethernet device.
4412 * Perform verbose error reporting if not NULL.
4415 * 0 for success, negative value otherwise and rte_errno is set.
4418 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4419 struct rte_flow_error *error)
4421 struct mlx5_priv *priv = dev->data->dev_private;
4422 struct mlx5_flow_mreg_copy_resource *mcp_res;
4423 struct mlx5_flow_cb_ctx ctx;
4426 /* Check whether extensive metadata feature is engaged. */
4427 if (!priv->config.dv_flow_en ||
4428 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4429 !mlx5_flow_ext_mreg_supported(dev) ||
4430 !priv->sh->dv_regc0_mask)
4433 * Add default mreg copy flow may be called multiple time, but
4434 * only be called once in stop. Avoid register it twice.
4436 mark_id = MLX5_DEFAULT_COPY_ID;
4437 ctx.data = &mark_id;
4438 if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4440 mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4447 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4449 * All the flow having Q/RSS action should be split by
4450 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4451 * performs the following,
4452 * - CQE->flow_tag := reg_c[1] (MARK)
4453 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4454 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4455 * but there should be a flow per each MARK ID set by MARK action.
4457 * For the aforementioned reason, if there's a MARK action in flow's action
4458 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4459 * the MARK ID to CQE's flow_tag like,
4460 * - If reg_c[1] is mark_id,
4461 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4463 * For SET_META action which stores value in reg_c[0], as the destination is
4464 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4465 * MARK ID means the default flow. The default flow looks like,
4466 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4469 * Pointer to Ethernet device.
4471 * Pointer to flow structure.
4472 * @param[in] actions
4473 * Pointer to the list of actions.
4475 * Perform verbose error reporting if not NULL.
4478 * 0 on success, negative value otherwise and rte_errno is set.
4481 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4482 struct rte_flow *flow,
4483 const struct rte_flow_action *actions,
4484 struct rte_flow_error *error)
4486 struct mlx5_priv *priv = dev->data->dev_private;
4487 struct mlx5_dev_config *config = &priv->config;
4488 struct mlx5_flow_mreg_copy_resource *mcp_res;
4489 const struct rte_flow_action_mark *mark;
4491 /* Check whether extensive metadata feature is engaged. */
4492 if (!config->dv_flow_en ||
4493 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4494 !mlx5_flow_ext_mreg_supported(dev) ||
4495 !priv->sh->dv_regc0_mask)
4497 /* Find MARK action. */
4498 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4499 switch (actions->type) {
4500 case RTE_FLOW_ACTION_TYPE_FLAG:
4501 mcp_res = flow_mreg_add_copy_action
4502 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4505 flow->rix_mreg_copy = mcp_res->idx;
4507 case RTE_FLOW_ACTION_TYPE_MARK:
4508 mark = (const struct rte_flow_action_mark *)
4511 flow_mreg_add_copy_action(dev, mark->id, error);
4514 flow->rix_mreg_copy = mcp_res->idx;
4523 #define MLX5_MAX_SPLIT_ACTIONS 24
4524 #define MLX5_MAX_SPLIT_ITEMS 24
4527 * Split the hairpin flow.
4528 * Since HW can't support encap and push-vlan on Rx, we move these
4530 * If the count action is after the encap then we also
4531 * move the count action. in this case the count will also measure
4535 * Pointer to Ethernet device.
4536 * @param[in] actions
4537 * Associated actions (list terminated by the END action).
4538 * @param[out] actions_rx
4540 * @param[out] actions_tx
4542 * @param[out] pattern_tx
4543 * The pattern items for the Tx flow.
4544 * @param[out] flow_id
4545 * The flow ID connected to this flow.
4551 flow_hairpin_split(struct rte_eth_dev *dev,
4552 const struct rte_flow_action actions[],
4553 struct rte_flow_action actions_rx[],
4554 struct rte_flow_action actions_tx[],
4555 struct rte_flow_item pattern_tx[],
4558 const struct rte_flow_action_raw_encap *raw_encap;
4559 const struct rte_flow_action_raw_decap *raw_decap;
4560 struct mlx5_rte_flow_action_set_tag *set_tag;
4561 struct rte_flow_action *tag_action;
4562 struct mlx5_rte_flow_item_tag *tag_item;
4563 struct rte_flow_item *item;
4567 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4568 switch (actions->type) {
4569 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4570 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4571 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4572 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4573 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4574 rte_memcpy(actions_tx, actions,
4575 sizeof(struct rte_flow_action));
4578 case RTE_FLOW_ACTION_TYPE_COUNT:
4580 rte_memcpy(actions_tx, actions,
4581 sizeof(struct rte_flow_action));
4584 rte_memcpy(actions_rx, actions,
4585 sizeof(struct rte_flow_action));
4589 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4590 raw_encap = actions->conf;
4591 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4592 memcpy(actions_tx, actions,
4593 sizeof(struct rte_flow_action));
4597 rte_memcpy(actions_rx, actions,
4598 sizeof(struct rte_flow_action));
4602 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4603 raw_decap = actions->conf;
4604 if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4605 memcpy(actions_tx, actions,
4606 sizeof(struct rte_flow_action));
4609 rte_memcpy(actions_rx, actions,
4610 sizeof(struct rte_flow_action));
4615 rte_memcpy(actions_rx, actions,
4616 sizeof(struct rte_flow_action));
4621 /* Add set meta action and end action for the Rx flow. */
4622 tag_action = actions_rx;
4623 tag_action->type = (enum rte_flow_action_type)
4624 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4626 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4628 set_tag = (void *)actions_rx;
4629 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4630 .id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4633 MLX5_ASSERT(set_tag->id > REG_NON);
4634 tag_action->conf = set_tag;
4635 /* Create Tx item list. */
4636 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4637 addr = (void *)&pattern_tx[2];
4639 item->type = (enum rte_flow_item_type)
4640 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4641 tag_item = (void *)addr;
4642 tag_item->data = flow_id;
4643 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4644 MLX5_ASSERT(set_tag->id > REG_NON);
4645 item->spec = tag_item;
4646 addr += sizeof(struct mlx5_rte_flow_item_tag);
4647 tag_item = (void *)addr;
4648 tag_item->data = UINT32_MAX;
4649 tag_item->id = UINT16_MAX;
4650 item->mask = tag_item;
4653 item->type = RTE_FLOW_ITEM_TYPE_END;
4658 * The last stage of splitting chain, just creates the subflow
4659 * without any modification.
4662 * Pointer to Ethernet device.
4664 * Parent flow structure pointer.
4665 * @param[in, out] sub_flow
4666 * Pointer to return the created subflow, may be NULL.
4668 * Flow rule attributes.
4670 * Pattern specification (list terminated by the END pattern item).
4671 * @param[in] actions
4672 * Associated actions (list terminated by the END action).
4673 * @param[in] flow_split_info
4674 * Pointer to flow split info structure.
4676 * Perform verbose error reporting if not NULL.
4678 * 0 on success, negative value otherwise
4681 flow_create_split_inner(struct rte_eth_dev *dev,
4682 struct rte_flow *flow,
4683 struct mlx5_flow **sub_flow,
4684 const struct rte_flow_attr *attr,
4685 const struct rte_flow_item items[],
4686 const struct rte_flow_action actions[],
4687 struct mlx5_flow_split_info *flow_split_info,
4688 struct rte_flow_error *error)
4690 struct mlx5_flow *dev_flow;
4692 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4693 flow_split_info->flow_idx, error);
4696 dev_flow->flow = flow;
4697 dev_flow->external = flow_split_info->external;
4698 dev_flow->skip_scale = flow_split_info->skip_scale;
4699 /* Subflow object was created, we must include one in the list. */
4700 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4701 dev_flow->handle, next);
4703 * If dev_flow is as one of the suffix flow, some actions in suffix
4704 * flow may need some user defined item layer flags, and pass the
4705 * Metadate rxq mark flag to suffix flow as well.
4707 if (flow_split_info->prefix_layers)
4708 dev_flow->handle->layers = flow_split_info->prefix_layers;
4709 if (flow_split_info->prefix_mark)
4710 dev_flow->handle->mark = 1;
4712 *sub_flow = dev_flow;
4713 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4714 dev_flow->dv.table_id = flow_split_info->table_id;
4716 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4720 * Get the sub policy of a meter.
4723 * Pointer to Ethernet device.
4725 * Parent flow structure pointer.
4727 * Pointer to thread flow work space.
4729 * Flow rule attributes.
4731 * Pattern specification (list terminated by the END pattern item).
4733 * Perform verbose error reporting if not NULL.
4736 * Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4738 static struct mlx5_flow_meter_sub_policy *
4739 get_meter_sub_policy(struct rte_eth_dev *dev,
4740 struct rte_flow *flow,
4741 struct mlx5_flow_workspace *wks,
4742 const struct rte_flow_attr *attr,
4743 const struct rte_flow_item items[],
4744 struct rte_flow_error *error)
4746 struct mlx5_flow_meter_policy *policy;
4747 struct mlx5_flow_meter_policy *final_policy;
4748 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4750 policy = wks->policy;
4751 final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4752 if (final_policy->is_rss || final_policy->is_queue) {
4753 struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4754 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4758 * This is a tmp dev_flow,
4759 * no need to register any matcher for it in translate.
4761 wks->skip_matcher_reg = 1;
4762 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4763 struct mlx5_flow dev_flow = {0};
4764 struct mlx5_flow_handle dev_handle = { {0} };
4765 uint8_t fate = final_policy->act_cnt[i].fate_action;
4767 if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4768 const struct rte_flow_action_rss *rss_act =
4769 final_policy->act_cnt[i].rss->conf;
4770 struct rte_flow_action rss_actions[2] = {
4772 .type = RTE_FLOW_ACTION_TYPE_RSS,
4776 .type = RTE_FLOW_ACTION_TYPE_END,
4781 dev_flow.handle = &dev_handle;
4782 dev_flow.ingress = attr->ingress;
4783 dev_flow.flow = flow;
4784 dev_flow.external = 0;
4785 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4786 dev_flow.dv.transfer = attr->transfer;
4789 * Translate RSS action to get rss hash fields.
4791 if (flow_drv_translate(dev, &dev_flow, attr,
4792 items, rss_actions, error))
4794 rss_desc_v[i] = wks->rss_desc;
4795 rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4796 rss_desc_v[i].hash_fields =
4797 dev_flow.hash_fields;
4798 rss_desc_v[i].queue_num =
4799 rss_desc_v[i].hash_fields ?
4800 rss_desc_v[i].queue_num : 1;
4801 rss_desc_v[i].tunnel =
4802 !!(dev_flow.handle->layers &
4803 MLX5_FLOW_LAYER_TUNNEL);
4804 /* Use the RSS queues in the containers. */
4805 rss_desc_v[i].queue =
4806 (uint16_t *)(uintptr_t)rss_act->queue;
4807 rss_desc[i] = &rss_desc_v[i];
4808 } else if (fate == MLX5_FLOW_FATE_QUEUE) {
4809 /* This is queue action. */
4810 rss_desc_v[i] = wks->rss_desc;
4811 rss_desc_v[i].key_len = 0;
4812 rss_desc_v[i].hash_fields = 0;
4813 rss_desc_v[i].queue =
4814 &final_policy->act_cnt[i].queue;
4815 rss_desc_v[i].queue_num = 1;
4816 rss_desc[i] = &rss_desc_v[i];
4821 sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4822 flow, policy, rss_desc);
4824 enum mlx5_meter_domain mtr_domain =
4825 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4826 (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4827 MLX5_MTR_DOMAIN_INGRESS);
4828 sub_policy = policy->sub_policys[mtr_domain][0];
4831 rte_flow_error_set(error, EINVAL,
4832 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4833 "Failed to get meter sub-policy.");
4839 * Split the meter flow.
4841 * As meter flow will split to three sub flow, other than meter
4842 * action, the other actions make sense to only meter accepts
4843 * the packet. If it need to be dropped, no other additional
4844 * actions should be take.
4846 * One kind of special action which decapsulates the L3 tunnel
4847 * header will be in the prefix sub flow, as not to take the
4848 * L3 tunnel header into account.
4851 * Pointer to Ethernet device.
4853 * Parent flow structure pointer.
4855 * Pointer to thread flow work space.
4857 * Flow rule attributes.
4859 * Pattern specification (list terminated by the END pattern item).
4860 * @param[out] sfx_items
4861 * Suffix flow match items (list terminated by the END pattern item).
4862 * @param[in] actions
4863 * Associated actions (list terminated by the END action).
4864 * @param[out] actions_sfx
4865 * Suffix flow actions.
4866 * @param[out] actions_pre
4867 * Prefix flow actions.
4868 * @param[out] mtr_flow_id
4869 * Pointer to meter flow id.
4871 * Perform verbose error reporting if not NULL.
4874 * 0 on success, a negative errno value otherwise and rte_errno is set.
4877 flow_meter_split_prep(struct rte_eth_dev *dev,
4878 struct rte_flow *flow,
4879 struct mlx5_flow_workspace *wks,
4880 const struct rte_flow_attr *attr,
4881 const struct rte_flow_item items[],
4882 struct rte_flow_item sfx_items[],
4883 const struct rte_flow_action actions[],
4884 struct rte_flow_action actions_sfx[],
4885 struct rte_flow_action actions_pre[],
4886 uint32_t *mtr_flow_id,
4887 struct rte_flow_error *error)
4889 struct mlx5_priv *priv = dev->data->dev_private;
4890 struct mlx5_flow_meter_info *fm = wks->fm;
4891 struct rte_flow_action *tag_action = NULL;
4892 struct rte_flow_item *tag_item;
4893 struct mlx5_rte_flow_action_set_tag *set_tag;
4894 const struct rte_flow_action_raw_encap *raw_encap;
4895 const struct rte_flow_action_raw_decap *raw_decap;
4896 struct mlx5_rte_flow_item_tag *tag_item_spec;
4897 struct mlx5_rte_flow_item_tag *tag_item_mask;
4898 uint32_t tag_id = 0;
4899 struct rte_flow_item *vlan_item_dst = NULL;
4900 const struct rte_flow_item *vlan_item_src = NULL;
4901 struct rte_flow_action *hw_mtr_action;
4902 struct rte_flow_action *action_pre_head = NULL;
4903 int32_t flow_src_port = priv->representor_id;
4905 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4906 uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4907 MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4908 uint32_t flow_id = 0;
4909 uint32_t flow_id_reversed = 0;
4910 uint8_t flow_id_bits = 0;
4913 /* Prepare the suffix subflow items. */
4914 tag_item = sfx_items++;
4915 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4916 struct mlx5_priv *port_priv;
4917 const struct rte_flow_item_port_id *pid_v;
4918 int item_type = items->type;
4920 switch (item_type) {
4921 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4922 pid_v = items->spec;
4924 port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4926 return rte_flow_error_set(error,
4928 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
4930 "Failed to get port info.");
4931 flow_src_port = port_priv->representor_id;
4932 if (!fm->def_policy && wks->policy->is_hierarchy &&
4933 flow_src_port != priv->representor_id) {
4934 if (flow_drv_mtr_hierarchy_rule_create(dev,
4941 memcpy(sfx_items, items, sizeof(*sfx_items));
4944 case RTE_FLOW_ITEM_TYPE_VLAN:
4945 /* Determine if copy vlan item below. */
4946 vlan_item_src = items;
4947 vlan_item_dst = sfx_items++;
4948 vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
4954 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4956 mtr_first = priv->sh->meter_aso_en &&
4957 (attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
4958 /* For ASO meter, meter must be before tag in TX direction. */
4960 action_pre_head = actions_pre++;
4961 /* Leave space for tag action. */
4962 tag_action = actions_pre++;
4964 /* Prepare the actions for prefix and suffix flow. */
4965 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4966 struct rte_flow_action *action_cur = NULL;
4968 switch (actions->type) {
4969 case RTE_FLOW_ACTION_TYPE_METER:
4971 action_cur = action_pre_head;
4973 /* Leave space for tag action. */
4974 tag_action = actions_pre++;
4975 action_cur = actions_pre++;
4978 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4979 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4980 action_cur = actions_pre++;
4982 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4983 raw_encap = actions->conf;
4984 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4985 action_cur = actions_pre++;
4987 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4988 raw_decap = actions->conf;
4989 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4990 action_cur = actions_pre++;
4992 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4993 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4994 if (vlan_item_dst && vlan_item_src) {
4995 memcpy(vlan_item_dst, vlan_item_src,
4996 sizeof(*vlan_item_dst));
4998 * Convert to internal match item, it is used
4999 * for vlan push and set vid.
5001 vlan_item_dst->type = (enum rte_flow_item_type)
5002 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5009 action_cur = (fm->def_policy) ?
5010 actions_sfx++ : actions_pre++;
5011 memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5013 /* Add end action to the actions. */
5014 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5015 if (priv->sh->meter_aso_en) {
5017 * For ASO meter, need to add an extra jump action explicitly,
5018 * to jump from meter to policer table.
5020 struct mlx5_flow_meter_sub_policy *sub_policy;
5021 struct mlx5_flow_tbl_data_entry *tbl_data;
5023 if (!fm->def_policy) {
5024 sub_policy = get_meter_sub_policy(dev, flow, wks,
5025 attr, items, error);
5029 enum mlx5_meter_domain mtr_domain =
5030 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5031 (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5032 MLX5_MTR_DOMAIN_INGRESS);
5035 &priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5037 tbl_data = container_of(sub_policy->tbl_rsc,
5038 struct mlx5_flow_tbl_data_entry, tbl);
5039 hw_mtr_action = actions_pre++;
5040 hw_mtr_action->type = (enum rte_flow_action_type)
5041 MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5042 hw_mtr_action->conf = tbl_data->jump.action;
5044 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5047 return rte_flow_error_set(error, ENOMEM,
5048 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5049 NULL, "No tag action space.");
5051 tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5054 /* Only default-policy Meter creates mtr flow id. */
5055 if (fm->def_policy) {
5056 mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5058 return rte_flow_error_set(error, ENOMEM,
5059 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5060 "Failed to allocate meter flow id.");
5061 flow_id = tag_id - 1;
5062 flow_id_bits = (!flow_id) ? 1 :
5063 (MLX5_REG_BITS - __builtin_clz(flow_id));
5064 if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5066 mlx5_ipool_free(fm->flow_ipool, tag_id);
5067 return rte_flow_error_set(error, EINVAL,
5068 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5069 "Meter flow id exceeds max limit.");
5071 if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5072 priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5074 /* Build tag actions and items for meter_id/meter flow_id. */
5075 set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5076 tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5077 tag_item_mask = tag_item_spec + 1;
5078 /* Both flow_id and meter_id share the same register. */
5079 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5080 .id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5082 .offset = mtr_id_offset,
5083 .length = mtr_reg_bits,
5084 .data = flow->meter,
5087 * The color Reg bits used by flow_id are growing from
5088 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5090 for (shift = 0; shift < flow_id_bits; shift++)
5091 flow_id_reversed = (flow_id_reversed << 1) |
5092 ((flow_id >> shift) & 0x1);
5094 flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5095 tag_item_spec->id = set_tag->id;
5096 tag_item_spec->data = set_tag->data << mtr_id_offset;
5097 tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5098 tag_action->type = (enum rte_flow_action_type)
5099 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5100 tag_action->conf = set_tag;
5101 tag_item->type = (enum rte_flow_item_type)
5102 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5103 tag_item->spec = tag_item_spec;
5104 tag_item->last = NULL;
5105 tag_item->mask = tag_item_mask;
5108 *mtr_flow_id = tag_id;
5113 * Split action list having QUEUE/RSS for metadata register copy.
5115 * Once Q/RSS action is detected in user's action list, the flow action
5116 * should be split in order to copy metadata registers, which will happen in
5118 * - CQE->flow_tag := reg_c[1] (MARK)
5119 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5120 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5121 * This is because the last action of each flow must be a terminal action
5122 * (QUEUE, RSS or DROP).
5124 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5125 * stored and kept in the mlx5_flow structure per each sub_flow.
5127 * The Q/RSS action is replaced with,
5128 * - SET_TAG, setting the allocated flow ID to reg_c[2].
5129 * And the following JUMP action is added at the end,
5130 * - JUMP, to RX_CP_TBL.
5132 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5133 * flow_create_split_metadata() routine. The flow will look like,
5134 * - If flow ID matches (reg_c[2]), perform Q/RSS.
5137 * Pointer to Ethernet device.
5138 * @param[out] split_actions
5139 * Pointer to store split actions to jump to CP_TBL.
5140 * @param[in] actions
5141 * Pointer to the list of original flow actions.
5143 * Pointer to the Q/RSS action.
5144 * @param[in] actions_n
5145 * Number of original actions.
5147 * Perform verbose error reporting if not NULL.
5150 * non-zero unique flow_id on success, otherwise 0 and
5151 * error/rte_error are set.
5154 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5155 struct rte_flow_action *split_actions,
5156 const struct rte_flow_action *actions,
5157 const struct rte_flow_action *qrss,
5158 int actions_n, struct rte_flow_error *error)
5160 struct mlx5_priv *priv = dev->data->dev_private;
5161 struct mlx5_rte_flow_action_set_tag *set_tag;
5162 struct rte_flow_action_jump *jump;
5163 const int qrss_idx = qrss - actions;
5164 uint32_t flow_id = 0;
5168 * Given actions will be split
5169 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5170 * - Add jump to mreg CP_TBL.
5171 * As a result, there will be one more action.
5174 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5175 set_tag = (void *)(split_actions + actions_n);
5177 * If tag action is not set to void(it means we are not the meter
5178 * suffix flow), add the tag action. Since meter suffix flow already
5179 * has the tag added.
5181 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5183 * Allocate the new subflow ID. This one is unique within
5184 * device and not shared with representors. Otherwise,
5185 * we would have to resolve multi-thread access synch
5186 * issue. Each flow on the shared device is appended
5187 * with source vport identifier, so the resulting
5188 * flows will be unique in the shared (by master and
5189 * representors) domain even if they have coinciding
5192 mlx5_ipool_malloc(priv->sh->ipool
5193 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5195 return rte_flow_error_set(error, ENOMEM,
5196 RTE_FLOW_ERROR_TYPE_ACTION,
5197 NULL, "can't allocate id "
5198 "for split Q/RSS subflow");
5199 /* Internal SET_TAG action to set flow ID. */
5200 *set_tag = (struct mlx5_rte_flow_action_set_tag){
5203 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5207 /* Construct new actions array. */
5208 /* Replace QUEUE/RSS action. */
5209 split_actions[qrss_idx] = (struct rte_flow_action){
5210 .type = (enum rte_flow_action_type)
5211 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5215 /* JUMP action to jump to mreg copy table (CP_TBL). */
5216 jump = (void *)(set_tag + 1);
5217 *jump = (struct rte_flow_action_jump){
5218 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5220 split_actions[actions_n - 2] = (struct rte_flow_action){
5221 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5224 split_actions[actions_n - 1] = (struct rte_flow_action){
5225 .type = RTE_FLOW_ACTION_TYPE_END,
5231 * Extend the given action list for Tx metadata copy.
5233 * Copy the given action list to the ext_actions and add flow metadata register
5234 * copy action in order to copy reg_a set by WQE to reg_c[0].
5236 * @param[out] ext_actions
5237 * Pointer to the extended action list.
5238 * @param[in] actions
5239 * Pointer to the list of actions.
5240 * @param[in] actions_n
5241 * Number of actions in the list.
5243 * Perform verbose error reporting if not NULL.
5244 * @param[in] encap_idx
5245 * The encap action inndex.
5248 * 0 on success, negative value otherwise
5251 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5252 struct rte_flow_action *ext_actions,
5253 const struct rte_flow_action *actions,
5254 int actions_n, struct rte_flow_error *error,
5257 struct mlx5_flow_action_copy_mreg *cp_mreg =
5258 (struct mlx5_flow_action_copy_mreg *)
5259 (ext_actions + actions_n + 1);
5262 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5266 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5271 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5272 if (encap_idx == actions_n - 1) {
5273 ext_actions[actions_n - 1] = (struct rte_flow_action){
5274 .type = (enum rte_flow_action_type)
5275 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5278 ext_actions[actions_n] = (struct rte_flow_action){
5279 .type = RTE_FLOW_ACTION_TYPE_END,
5282 ext_actions[encap_idx] = (struct rte_flow_action){
5283 .type = (enum rte_flow_action_type)
5284 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5287 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5288 sizeof(*ext_actions) * (actions_n - encap_idx));
5294 * Check the match action from the action list.
5296 * @param[in] actions
5297 * Pointer to the list of actions.
5299 * Flow rule attributes.
5301 * The action to be check if exist.
5302 * @param[out] match_action_pos
5303 * Pointer to the position of the matched action if exists, otherwise is -1.
5304 * @param[out] qrss_action_pos
5305 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5306 * @param[out] modify_after_mirror
5307 * Pointer to the flag of modify action after FDB mirroring.
5310 * > 0 the total number of actions.
5311 * 0 if not found match action in action list.
5314 flow_check_match_action(const struct rte_flow_action actions[],
5315 const struct rte_flow_attr *attr,
5316 enum rte_flow_action_type action,
5317 int *match_action_pos, int *qrss_action_pos,
5318 int *modify_after_mirror)
5320 const struct rte_flow_action_sample *sample;
5321 const struct rte_flow_action_raw_decap *decap;
5328 *match_action_pos = -1;
5329 *qrss_action_pos = -1;
5330 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5331 if (actions->type == action) {
5333 *match_action_pos = actions_n;
5335 switch (actions->type) {
5336 case RTE_FLOW_ACTION_TYPE_QUEUE:
5337 case RTE_FLOW_ACTION_TYPE_RSS:
5338 *qrss_action_pos = actions_n;
5340 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5341 sample = actions->conf;
5342 ratio = sample->ratio;
5343 sub_type = ((const struct rte_flow_action *)
5344 (sample->actions))->type;
5345 if (ratio == 1 && attr->transfer)
5348 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5349 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5350 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5351 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5352 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5353 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5354 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5355 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5356 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5357 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5358 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5359 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5360 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5361 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5362 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5363 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5364 case RTE_FLOW_ACTION_TYPE_FLAG:
5365 case RTE_FLOW_ACTION_TYPE_MARK:
5366 case RTE_FLOW_ACTION_TYPE_SET_META:
5367 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5368 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5369 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5370 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5371 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5372 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5373 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5374 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5375 case RTE_FLOW_ACTION_TYPE_METER:
5377 *modify_after_mirror = 1;
5379 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5380 decap = actions->conf;
5381 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5384 if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5385 const struct rte_flow_action_raw_encap *encap =
5388 MLX5_ENCAPSULATION_DECISION_SIZE &&
5390 MLX5_ENCAPSULATION_DECISION_SIZE)
5395 *modify_after_mirror = 1;
5402 if (flag && fdb_mirror && !*modify_after_mirror) {
5403 /* FDB mirroring uses the destination array to implement
5404 * instead of FLOW_SAMPLER object.
5406 if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5409 /* Count RTE_FLOW_ACTION_TYPE_END. */
5410 return flag ? actions_n + 1 : 0;
5413 #define SAMPLE_SUFFIX_ITEM 2
5416 * Split the sample flow.
5418 * As sample flow will split to two sub flow, sample flow with
5419 * sample action, the other actions will move to new suffix flow.
5421 * Also add unique tag id with tag action in the sample flow,
5422 * the same tag id will be as match in the suffix flow.
5425 * Pointer to Ethernet device.
5426 * @param[in] add_tag
5427 * Add extra tag action flag.
5428 * @param[out] sfx_items
5429 * Suffix flow match items (list terminated by the END pattern item).
5430 * @param[in] actions
5431 * Associated actions (list terminated by the END action).
5432 * @param[out] actions_sfx
5433 * Suffix flow actions.
5434 * @param[out] actions_pre
5435 * Prefix flow actions.
5436 * @param[in] actions_n
5437 * The total number of actions.
5438 * @param[in] sample_action_pos
5439 * The sample action position.
5440 * @param[in] qrss_action_pos
5441 * The Queue/RSS action position.
5442 * @param[in] jump_table
5443 * Add extra jump action flag.
5445 * Perform verbose error reporting if not NULL.
5448 * 0 on success, or unique flow_id, a negative errno value
5449 * otherwise and rte_errno is set.
5452 flow_sample_split_prep(struct rte_eth_dev *dev,
5454 struct rte_flow_item sfx_items[],
5455 const struct rte_flow_action actions[],
5456 struct rte_flow_action actions_sfx[],
5457 struct rte_flow_action actions_pre[],
5459 int sample_action_pos,
5460 int qrss_action_pos,
5462 struct rte_flow_error *error)
5464 struct mlx5_priv *priv = dev->data->dev_private;
5465 struct mlx5_rte_flow_action_set_tag *set_tag;
5466 struct mlx5_rte_flow_item_tag *tag_spec;
5467 struct mlx5_rte_flow_item_tag *tag_mask;
5468 struct rte_flow_action_jump *jump_action;
5469 uint32_t tag_id = 0;
5471 int append_index = 0;
5474 if (sample_action_pos < 0)
5475 return rte_flow_error_set(error, EINVAL,
5476 RTE_FLOW_ERROR_TYPE_ACTION,
5477 NULL, "invalid position of sample "
5479 /* Prepare the actions for prefix and suffix flow. */
5480 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5481 index = qrss_action_pos;
5482 /* Put the preceding the Queue/RSS action into prefix flow. */
5484 memcpy(actions_pre, actions,
5485 sizeof(struct rte_flow_action) * index);
5486 /* Put others preceding the sample action into prefix flow. */
5487 if (sample_action_pos > index + 1)
5488 memcpy(actions_pre + index, actions + index + 1,
5489 sizeof(struct rte_flow_action) *
5490 (sample_action_pos - index - 1));
5491 index = sample_action_pos - 1;
5492 /* Put Queue/RSS action into Suffix flow. */
5493 memcpy(actions_sfx, actions + qrss_action_pos,
5494 sizeof(struct rte_flow_action));
5497 index = sample_action_pos;
5499 memcpy(actions_pre, actions,
5500 sizeof(struct rte_flow_action) * index);
5502 /* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5503 * For CX6DX and above, metadata registers Cx preserve their value,
5504 * add an extra tag action for NIC-RX and E-Switch Domain.
5507 /* Prepare the prefix tag action. */
5509 set_tag = (void *)(actions_pre + actions_n + append_index);
5510 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5513 mlx5_ipool_malloc(priv->sh->ipool
5514 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5515 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5519 /* Prepare the suffix subflow items. */
5520 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5521 tag_spec->data = tag_id;
5522 tag_spec->id = set_tag->id;
5523 tag_mask = tag_spec + 1;
5524 tag_mask->data = UINT32_MAX;
5525 sfx_items[0] = (struct rte_flow_item){
5526 .type = (enum rte_flow_item_type)
5527 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5532 sfx_items[1] = (struct rte_flow_item){
5533 .type = (enum rte_flow_item_type)
5534 RTE_FLOW_ITEM_TYPE_END,
5536 /* Prepare the tag action in prefix subflow. */
5537 actions_pre[index++] =
5538 (struct rte_flow_action){
5539 .type = (enum rte_flow_action_type)
5540 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5544 memcpy(actions_pre + index, actions + sample_action_pos,
5545 sizeof(struct rte_flow_action));
5547 /* For the modify action after the sample action in E-Switch mirroring,
5548 * Add the extra jump action in prefix subflow and jump into the next
5549 * table, then do the modify action in the new table.
5552 /* Prepare the prefix jump action. */
5554 jump_action = (void *)(actions_pre + actions_n + append_index);
5555 jump_action->group = jump_table;
5556 actions_pre[index++] =
5557 (struct rte_flow_action){
5558 .type = (enum rte_flow_action_type)
5559 RTE_FLOW_ACTION_TYPE_JUMP,
5560 .conf = jump_action,
5563 actions_pre[index] = (struct rte_flow_action){
5564 .type = (enum rte_flow_action_type)
5565 RTE_FLOW_ACTION_TYPE_END,
5567 /* Put the actions after sample into Suffix flow. */
5568 memcpy(actions_sfx, actions + sample_action_pos + 1,
5569 sizeof(struct rte_flow_action) *
5570 (actions_n - sample_action_pos - 1));
5575 * The splitting for metadata feature.
5577 * - Q/RSS action on NIC Rx should be split in order to pass by
5578 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5579 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5581 * - All the actions on NIC Tx should have a mreg copy action to
5582 * copy reg_a from WQE to reg_c[0].
5585 * Pointer to Ethernet device.
5587 * Parent flow structure pointer.
5589 * Flow rule attributes.
5591 * Pattern specification (list terminated by the END pattern item).
5592 * @param[in] actions
5593 * Associated actions (list terminated by the END action).
5594 * @param[in] flow_split_info
5595 * Pointer to flow split info structure.
5597 * Perform verbose error reporting if not NULL.
5599 * 0 on success, negative value otherwise
5602 flow_create_split_metadata(struct rte_eth_dev *dev,
5603 struct rte_flow *flow,
5604 const struct rte_flow_attr *attr,
5605 const struct rte_flow_item items[],
5606 const struct rte_flow_action actions[],
5607 struct mlx5_flow_split_info *flow_split_info,
5608 struct rte_flow_error *error)
5610 struct mlx5_priv *priv = dev->data->dev_private;
5611 struct mlx5_dev_config *config = &priv->config;
5612 const struct rte_flow_action *qrss = NULL;
5613 struct rte_flow_action *ext_actions = NULL;
5614 struct mlx5_flow *dev_flow = NULL;
5615 uint32_t qrss_id = 0;
5622 /* Check whether extensive metadata feature is engaged. */
5623 if (!config->dv_flow_en ||
5624 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5625 !mlx5_flow_ext_mreg_supported(dev))
5626 return flow_create_split_inner(dev, flow, NULL, attr, items,
5627 actions, flow_split_info, error);
5628 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5631 /* Exclude hairpin flows from splitting. */
5632 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5633 const struct rte_flow_action_queue *queue;
5636 if (mlx5_rxq_get_type(dev, queue->index) ==
5637 MLX5_RXQ_TYPE_HAIRPIN)
5639 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5640 const struct rte_flow_action_rss *rss;
5643 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5644 MLX5_RXQ_TYPE_HAIRPIN)
5649 /* Check if it is in meter suffix table. */
5650 mtr_sfx = attr->group == (attr->transfer ?
5651 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5652 MLX5_FLOW_TABLE_LEVEL_METER);
5654 * Q/RSS action on NIC Rx should be split in order to pass by
5655 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5656 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5658 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5659 sizeof(struct rte_flow_action_set_tag) +
5660 sizeof(struct rte_flow_action_jump);
5661 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5664 return rte_flow_error_set(error, ENOMEM,
5665 RTE_FLOW_ERROR_TYPE_ACTION,
5666 NULL, "no memory to split "
5669 * If we are the suffix flow of meter, tag already exist.
5670 * Set the tag action to void.
5673 ext_actions[qrss - actions].type =
5674 RTE_FLOW_ACTION_TYPE_VOID;
5676 ext_actions[qrss - actions].type =
5677 (enum rte_flow_action_type)
5678 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5680 * Create the new actions list with removed Q/RSS action
5681 * and appended set tag and jump to register copy table
5682 * (RX_CP_TBL). We should preallocate unique tag ID here
5683 * in advance, because it is needed for set tag action.
5685 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5686 qrss, actions_n, error);
5687 if (!mtr_sfx && !qrss_id) {
5691 } else if (attr->egress && !attr->transfer) {
5693 * All the actions on NIC Tx should have a metadata register
5694 * copy action to copy reg_a from WQE to reg_c[meta]
5696 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5697 sizeof(struct mlx5_flow_action_copy_mreg);
5698 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5701 return rte_flow_error_set(error, ENOMEM,
5702 RTE_FLOW_ERROR_TYPE_ACTION,
5703 NULL, "no memory to split "
5705 /* Create the action list appended with copy register. */
5706 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5707 actions_n, error, encap_idx);
5711 /* Add the unmodified original or prefix subflow. */
5712 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5713 items, ext_actions ? ext_actions :
5714 actions, flow_split_info, error);
5717 MLX5_ASSERT(dev_flow);
5719 const struct rte_flow_attr q_attr = {
5720 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5723 /* Internal PMD action to set register. */
5724 struct mlx5_rte_flow_item_tag q_tag_spec = {
5728 struct rte_flow_item q_items[] = {
5730 .type = (enum rte_flow_item_type)
5731 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5732 .spec = &q_tag_spec,
5737 .type = RTE_FLOW_ITEM_TYPE_END,
5740 struct rte_flow_action q_actions[] = {
5746 .type = RTE_FLOW_ACTION_TYPE_END,
5749 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5752 * Configure the tag item only if there is no meter subflow.
5753 * Since tag is already marked in the meter suffix subflow
5754 * we can just use the meter suffix items as is.
5757 /* Not meter subflow. */
5758 MLX5_ASSERT(!mtr_sfx);
5760 * Put unique id in prefix flow due to it is destroyed
5761 * after suffix flow and id will be freed after there
5762 * is no actual flows with this id and identifier
5763 * reallocation becomes possible (for example, for
5764 * other flows in other threads).
5766 dev_flow->handle->split_flow_id = qrss_id;
5767 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5771 q_tag_spec.id = ret;
5774 /* Add suffix subflow to execute Q/RSS. */
5775 flow_split_info->prefix_layers = layers;
5776 flow_split_info->prefix_mark = 0;
5777 ret = flow_create_split_inner(dev, flow, &dev_flow,
5778 &q_attr, mtr_sfx ? items :
5780 flow_split_info, error);
5783 /* qrss ID should be freed if failed. */
5785 MLX5_ASSERT(dev_flow);
5790 * We do not destroy the partially created sub_flows in case of error.
5791 * These ones are included into parent flow list and will be destroyed
5792 * by flow_drv_destroy.
5794 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5796 mlx5_free(ext_actions);
5801 * Create meter internal drop flow with the original pattern.
5804 * Pointer to Ethernet device.
5806 * Parent flow structure pointer.
5808 * Flow rule attributes.
5810 * Pattern specification (list terminated by the END pattern item).
5811 * @param[in] flow_split_info
5812 * Pointer to flow split info structure.
5814 * Pointer to flow meter structure.
5816 * Perform verbose error reporting if not NULL.
5818 * 0 on success, negative value otherwise
5821 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5822 struct rte_flow *flow,
5823 const struct rte_flow_attr *attr,
5824 const struct rte_flow_item items[],
5825 struct mlx5_flow_split_info *flow_split_info,
5826 struct mlx5_flow_meter_info *fm,
5827 struct rte_flow_error *error)
5829 struct mlx5_flow *dev_flow = NULL;
5830 struct rte_flow_attr drop_attr = *attr;
5831 struct rte_flow_action drop_actions[3];
5832 struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5834 MLX5_ASSERT(fm->drop_cnt);
5835 drop_actions[0].type =
5836 (enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5837 drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5838 drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5839 drop_actions[1].conf = NULL;
5840 drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5841 drop_actions[2].conf = NULL;
5842 drop_split_info.external = false;
5843 drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5844 drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5845 drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5846 return flow_create_split_inner(dev, flow, &dev_flow,
5847 &drop_attr, items, drop_actions,
5848 &drop_split_info, error);
5852 * The splitting for meter feature.
5854 * - The meter flow will be split to two flows as prefix and
5855 * suffix flow. The packets make sense only it pass the prefix
5858 * - Reg_C_5 is used for the packet to match betweend prefix and
5862 * Pointer to Ethernet device.
5864 * Parent flow structure pointer.
5866 * Flow rule attributes.
5868 * Pattern specification (list terminated by the END pattern item).
5869 * @param[in] actions
5870 * Associated actions (list terminated by the END action).
5871 * @param[in] flow_split_info
5872 * Pointer to flow split info structure.
5874 * Perform verbose error reporting if not NULL.
5876 * 0 on success, negative value otherwise
5879 flow_create_split_meter(struct rte_eth_dev *dev,
5880 struct rte_flow *flow,
5881 const struct rte_flow_attr *attr,
5882 const struct rte_flow_item items[],
5883 const struct rte_flow_action actions[],
5884 struct mlx5_flow_split_info *flow_split_info,
5885 struct rte_flow_error *error)
5887 struct mlx5_priv *priv = dev->data->dev_private;
5888 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5889 struct rte_flow_action *sfx_actions = NULL;
5890 struct rte_flow_action *pre_actions = NULL;
5891 struct rte_flow_item *sfx_items = NULL;
5892 struct mlx5_flow *dev_flow = NULL;
5893 struct rte_flow_attr sfx_attr = *attr;
5894 struct mlx5_flow_meter_info *fm = NULL;
5895 uint8_t skip_scale_restore;
5896 bool has_mtr = false;
5897 bool has_modify = false;
5898 bool set_mtr_reg = true;
5899 bool is_mtr_hierarchy = false;
5900 uint32_t meter_id = 0;
5901 uint32_t mtr_idx = 0;
5902 uint32_t mtr_flow_id = 0;
5909 actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5910 &has_modify, &meter_id);
5913 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5915 return rte_flow_error_set(error, EINVAL,
5916 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5917 NULL, "Meter not found.");
5919 fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5921 return rte_flow_error_set(error, EINVAL,
5922 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5923 NULL, "Meter not found.");
5924 ret = mlx5_flow_meter_attach(priv, fm,
5928 flow->meter = mtr_idx;
5932 if (!fm->def_policy) {
5933 wks->policy = mlx5_flow_meter_policy_find(dev,
5936 MLX5_ASSERT(wks->policy);
5937 if (wks->policy->is_hierarchy) {
5939 mlx5_flow_meter_hierarchy_get_final_policy(dev,
5941 if (!wks->final_policy)
5942 return rte_flow_error_set(error,
5944 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5945 "Failed to find terminal policy of hierarchy.");
5946 is_mtr_hierarchy = true;
5950 * If it isn't default-policy Meter, and
5951 * 1. There's no action in flow to change
5952 * packet (modify/encap/decap etc.), OR
5953 * 2. No drop count needed for this meter.
5954 * 3. It's not meter hierarchy.
5955 * Then no need to use regC to save meter id anymore.
5957 if (!fm->def_policy && !is_mtr_hierarchy &&
5958 (!has_modify || !fm->drop_cnt))
5959 set_mtr_reg = false;
5960 /* Prefix actions: meter, decap, encap, tag, jump, end. */
5961 act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5962 sizeof(struct mlx5_rte_flow_action_set_tag);
5963 /* Suffix items: tag, vlan, port id, end. */
5964 #define METER_SUFFIX_ITEM 4
5965 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5966 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5967 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5970 return rte_flow_error_set(error, ENOMEM,
5971 RTE_FLOW_ERROR_TYPE_ACTION,
5972 NULL, "no memory to split "
5974 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5976 /* There's no suffix flow for meter of non-default policy. */
5977 if (!fm->def_policy)
5978 pre_actions = sfx_actions + 1;
5980 pre_actions = sfx_actions + actions_n;
5981 ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
5982 items, sfx_items, actions,
5983 sfx_actions, pre_actions,
5984 (set_mtr_reg ? &mtr_flow_id : NULL),
5990 /* Add the prefix subflow. */
5991 flow_split_info->prefix_mark = 0;
5992 skip_scale_restore = flow_split_info->skip_scale;
5993 flow_split_info->skip_scale |=
5994 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5995 ret = flow_create_split_inner(dev, flow, &dev_flow,
5996 attr, items, pre_actions,
5997 flow_split_info, error);
5998 flow_split_info->skip_scale = skip_scale_restore;
6001 mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6006 dev_flow->handle->split_flow_id = mtr_flow_id;
6007 dev_flow->handle->is_meter_flow_id = 1;
6009 if (!fm->def_policy) {
6010 if (!set_mtr_reg && fm->drop_cnt)
6012 flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6018 /* Setting the sfx group atrr. */
6019 sfx_attr.group = sfx_attr.transfer ?
6020 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6021 MLX5_FLOW_TABLE_LEVEL_METER;
6022 flow_split_info->prefix_layers =
6023 flow_get_prefix_layer_flags(dev_flow);
6024 flow_split_info->prefix_mark = dev_flow->handle->mark;
6025 flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6027 /* Add the prefix subflow. */
6028 ret = flow_create_split_metadata(dev, flow,
6029 &sfx_attr, sfx_items ?
6031 sfx_actions ? sfx_actions : actions,
6032 flow_split_info, error);
6035 mlx5_free(sfx_actions);
6040 * The splitting for sample feature.
6042 * Once Sample action is detected in the action list, the flow actions should
6043 * be split into prefix sub flow and suffix sub flow.
6045 * The original items remain in the prefix sub flow, all actions preceding the
6046 * sample action and the sample action itself will be copied to the prefix
6047 * sub flow, the actions following the sample action will be copied to the
6048 * suffix sub flow, Queue action always be located in the suffix sub flow.
6050 * In order to make the packet from prefix sub flow matches with suffix sub
6051 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6052 * flow uses tag item with the unique flow id.
6055 * Pointer to Ethernet device.
6057 * Parent flow structure pointer.
6059 * Flow rule attributes.
6061 * Pattern specification (list terminated by the END pattern item).
6062 * @param[in] actions
6063 * Associated actions (list terminated by the END action).
6064 * @param[in] flow_split_info
6065 * Pointer to flow split info structure.
6067 * Perform verbose error reporting if not NULL.
6069 * 0 on success, negative value otherwise
6072 flow_create_split_sample(struct rte_eth_dev *dev,
6073 struct rte_flow *flow,
6074 const struct rte_flow_attr *attr,
6075 const struct rte_flow_item items[],
6076 const struct rte_flow_action actions[],
6077 struct mlx5_flow_split_info *flow_split_info,
6078 struct rte_flow_error *error)
6080 struct mlx5_priv *priv = dev->data->dev_private;
6081 struct rte_flow_action *sfx_actions = NULL;
6082 struct rte_flow_action *pre_actions = NULL;
6083 struct rte_flow_item *sfx_items = NULL;
6084 struct mlx5_flow *dev_flow = NULL;
6085 struct rte_flow_attr sfx_attr = *attr;
6086 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6087 struct mlx5_flow_dv_sample_resource *sample_res;
6088 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6089 struct mlx5_flow_tbl_resource *sfx_tbl;
6093 uint32_t fdb_tx = 0;
6096 int sample_action_pos;
6097 int qrss_action_pos;
6099 int modify_after_mirror = 0;
6100 uint16_t jump_table = 0;
6101 const uint32_t next_ft_step = 1;
6104 if (priv->sampler_en)
6105 actions_n = flow_check_match_action(actions, attr,
6106 RTE_FLOW_ACTION_TYPE_SAMPLE,
6107 &sample_action_pos, &qrss_action_pos,
6108 &modify_after_mirror);
6110 /* The prefix actions must includes sample, tag, end. */
6111 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6112 + sizeof(struct mlx5_rte_flow_action_set_tag);
6113 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6114 sizeof(struct mlx5_rte_flow_item_tag) * 2;
6115 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6116 item_size), 0, SOCKET_ID_ANY);
6118 return rte_flow_error_set(error, ENOMEM,
6119 RTE_FLOW_ERROR_TYPE_ACTION,
6120 NULL, "no memory to split "
6122 /* The representor_id is UINT16_MAX for uplink. */
6123 fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6125 * When reg_c_preserve is set, metadata registers Cx preserve
6126 * their value even through packet duplication.
6128 add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6130 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6132 if (modify_after_mirror)
6133 jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6135 pre_actions = sfx_actions + actions_n;
6136 tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6137 actions, sfx_actions,
6138 pre_actions, actions_n,
6140 qrss_action_pos, jump_table,
6142 if (tag_id < 0 || (add_tag && !tag_id)) {
6146 if (modify_after_mirror)
6147 flow_split_info->skip_scale =
6148 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6149 /* Add the prefix subflow. */
6150 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6152 flow_split_info, error);
6157 dev_flow->handle->split_flow_id = tag_id;
6158 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6159 if (!modify_after_mirror) {
6160 /* Set the sfx group attr. */
6161 sample_res = (struct mlx5_flow_dv_sample_resource *)
6162 dev_flow->dv.sample_res;
6163 sfx_tbl = (struct mlx5_flow_tbl_resource *)
6164 sample_res->normal_path_tbl;
6165 sfx_tbl_data = container_of(sfx_tbl,
6166 struct mlx5_flow_tbl_data_entry,
6168 sfx_attr.group = sfx_attr.transfer ?
6169 (sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6171 MLX5_ASSERT(attr->transfer);
6172 sfx_attr.group = jump_table;
6174 flow_split_info->prefix_layers =
6175 flow_get_prefix_layer_flags(dev_flow);
6176 flow_split_info->prefix_mark = dev_flow->handle->mark;
6177 /* Suffix group level already be scaled with factor, set
6178 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6179 * again in translation.
6181 flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6184 /* Add the suffix subflow. */
6185 ret = flow_create_split_meter(dev, flow, &sfx_attr,
6186 sfx_items ? sfx_items : items,
6187 sfx_actions ? sfx_actions : actions,
6188 flow_split_info, error);
6191 mlx5_free(sfx_actions);
6196 * Split the flow to subflow set. The splitters might be linked
6197 * in the chain, like this:
6198 * flow_create_split_outer() calls:
6199 * flow_create_split_meter() calls:
6200 * flow_create_split_metadata(meter_subflow_0) calls:
6201 * flow_create_split_inner(metadata_subflow_0)
6202 * flow_create_split_inner(metadata_subflow_1)
6203 * flow_create_split_inner(metadata_subflow_2)
6204 * flow_create_split_metadata(meter_subflow_1) calls:
6205 * flow_create_split_inner(metadata_subflow_0)
6206 * flow_create_split_inner(metadata_subflow_1)
6207 * flow_create_split_inner(metadata_subflow_2)
6209 * This provide flexible way to add new levels of flow splitting.
6210 * The all of successfully created subflows are included to the
6211 * parent flow dev_flow list.
6214 * Pointer to Ethernet device.
6216 * Parent flow structure pointer.
6218 * Flow rule attributes.
6220 * Pattern specification (list terminated by the END pattern item).
6221 * @param[in] actions
6222 * Associated actions (list terminated by the END action).
6223 * @param[in] flow_split_info
6224 * Pointer to flow split info structure.
6226 * Perform verbose error reporting if not NULL.
6228 * 0 on success, negative value otherwise
6231 flow_create_split_outer(struct rte_eth_dev *dev,
6232 struct rte_flow *flow,
6233 const struct rte_flow_attr *attr,
6234 const struct rte_flow_item items[],
6235 const struct rte_flow_action actions[],
6236 struct mlx5_flow_split_info *flow_split_info,
6237 struct rte_flow_error *error)
6241 ret = flow_create_split_sample(dev, flow, attr, items,
6242 actions, flow_split_info, error);
6243 MLX5_ASSERT(ret <= 0);
6247 static inline struct mlx5_flow_tunnel *
6248 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6250 struct mlx5_flow_tunnel *tunnel;
6252 #pragma GCC diagnostic push
6253 #pragma GCC diagnostic ignored "-Wcast-qual"
6254 tunnel = (typeof(tunnel))flow->tunnel;
6255 #pragma GCC diagnostic pop
6261 * Adjust flow RSS workspace if needed.
6264 * Pointer to thread flow work space.
6266 * Pointer to RSS descriptor.
6267 * @param[in] nrssq_num
6268 * New RSS queue number.
6271 * 0 on success, -1 otherwise and rte_errno is set.
6274 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6275 struct mlx5_flow_rss_desc *rss_desc,
6278 if (likely(nrssq_num <= wks->rssq_num))
6280 rss_desc->queue = realloc(rss_desc->queue,
6281 sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6282 if (!rss_desc->queue) {
6286 wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6291 * Create a flow and add it to @p list.
6294 * Pointer to Ethernet device.
6296 * Pointer to a TAILQ flow list. If this parameter NULL,
6297 * no list insertion occurred, flow is just created,
6298 * this is caller's responsibility to track the
6301 * Flow rule attributes.
6303 * Pattern specification (list terminated by the END pattern item).
6304 * @param[in] actions
6305 * Associated actions (list terminated by the END action).
6306 * @param[in] external
6307 * This flow rule is created by request external to PMD.
6309 * Perform verbose error reporting if not NULL.
6312 * A flow index on success, 0 otherwise and rte_errno is set.
6315 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6316 const struct rte_flow_attr *attr,
6317 const struct rte_flow_item items[],
6318 const struct rte_flow_action original_actions[],
6319 bool external, struct rte_flow_error *error)
6321 struct mlx5_priv *priv = dev->data->dev_private;
6322 struct rte_flow *flow = NULL;
6323 struct mlx5_flow *dev_flow;
6324 const struct rte_flow_action_rss *rss = NULL;
6325 struct mlx5_translated_action_handle
6326 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6327 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6329 struct mlx5_flow_expand_rss buf;
6330 uint8_t buffer[4096];
6333 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6334 uint8_t buffer[2048];
6337 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6338 uint8_t buffer[2048];
6339 } actions_hairpin_tx;
6341 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6342 uint8_t buffer[2048];
6344 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6345 struct mlx5_flow_rss_desc *rss_desc;
6346 const struct rte_flow_action *p_actions_rx;
6350 struct rte_flow_attr attr_tx = { .priority = 0 };
6351 const struct rte_flow_action *actions;
6352 struct rte_flow_action *translated_actions = NULL;
6353 struct mlx5_flow_tunnel *tunnel;
6354 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6355 struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6356 struct mlx5_flow_split_info flow_split_info = {
6357 .external = !!external,
6367 rss_desc = &wks->rss_desc;
6368 ret = flow_action_handles_translate(dev, original_actions,
6371 &translated_actions, error);
6373 MLX5_ASSERT(translated_actions == NULL);
6376 actions = translated_actions ? translated_actions : original_actions;
6377 p_actions_rx = actions;
6378 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6379 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6380 external, hairpin_flow, error);
6382 goto error_before_hairpin_split;
6383 flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6386 goto error_before_hairpin_split;
6388 if (hairpin_flow > 0) {
6389 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6391 goto error_before_hairpin_split;
6393 flow_hairpin_split(dev, actions, actions_rx.actions,
6394 actions_hairpin_tx.actions, items_tx.items,
6396 p_actions_rx = actions_rx.actions;
6398 flow_split_info.flow_idx = idx;
6399 flow->drv_type = flow_get_drv_type(dev, attr);
6400 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6401 flow->drv_type < MLX5_FLOW_TYPE_MAX);
6402 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6403 /* RSS Action only works on NIC RX domain */
6404 if (attr->ingress && !attr->transfer)
6405 rss = flow_get_rss_action(dev, p_actions_rx);
6407 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6410 * The following information is required by
6411 * mlx5_flow_hashfields_adjust() in advance.
6413 rss_desc->level = rss->level;
6414 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6415 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6417 flow->dev_handles = 0;
6418 if (rss && rss->types) {
6419 unsigned int graph_root;
6421 graph_root = find_graph_root(rss->level);
6422 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6424 mlx5_support_expansion, graph_root);
6425 MLX5_ASSERT(ret > 0 &&
6426 (unsigned int)ret < sizeof(expand_buffer.buffer));
6427 if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6428 for (i = 0; i < buf->entries; ++i)
6429 mlx5_dbg__print_pattern(buf->entry[i].pattern);
6433 buf->entry[0].pattern = (void *)(uintptr_t)items;
6435 rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6437 for (i = 0; i < buf->entries; ++i) {
6438 /* Initialize flow split data. */
6439 flow_split_info.prefix_layers = 0;
6440 flow_split_info.prefix_mark = 0;
6441 flow_split_info.skip_scale = 0;
6443 * The splitter may create multiple dev_flows,
6444 * depending on configuration. In the simplest
6445 * case it just creates unmodified original flow.
6447 ret = flow_create_split_outer(dev, flow, attr,
6448 buf->entry[i].pattern,
6449 p_actions_rx, &flow_split_info,
6453 if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6454 ret = flow_tunnel_add_default_miss(dev, flow, attr,
6457 wks->flows[0].tunnel,
6461 mlx5_free(default_miss_ctx.queue);
6466 /* Create the tx flow. */
6468 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6469 attr_tx.ingress = 0;
6471 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6472 actions_hairpin_tx.actions,
6476 dev_flow->flow = flow;
6477 dev_flow->external = 0;
6478 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6479 dev_flow->handle, next);
6480 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6482 actions_hairpin_tx.actions, error);
6487 * Update the metadata register copy table. If extensive
6488 * metadata feature is enabled and registers are supported
6489 * we might create the extra rte_flow for each unique
6490 * MARK/FLAG action ID.
6492 * The table is updated for ingress Flows only, because
6493 * the egress Flows belong to the different device and
6494 * copy table should be updated in peer NIC Rx domain.
6496 if (attr->ingress &&
6497 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6498 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6503 * If the flow is external (from application) OR device is started,
6504 * OR mreg discover, then apply immediately.
6506 if (external || dev->data->dev_started ||
6507 (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6508 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6509 ret = flow_drv_apply(dev, flow, error);
6514 flow_rxq_flags_set(dev, flow);
6515 rte_free(translated_actions);
6516 tunnel = flow_tunnel_from_rule(wks->flows);
6519 flow->tunnel_id = tunnel->tunnel_id;
6520 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6521 mlx5_free(default_miss_ctx.queue);
6523 mlx5_flow_pop_thread_workspace();
6527 ret = rte_errno; /* Save rte_errno before cleanup. */
6528 flow_mreg_del_copy_action(dev, flow);
6529 flow_drv_destroy(dev, flow);
6530 if (rss_desc->shared_rss)
6531 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6533 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6534 rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6535 mlx5_ipool_free(priv->flows[type], idx);
6536 rte_errno = ret; /* Restore rte_errno. */
6539 mlx5_flow_pop_thread_workspace();
6540 error_before_hairpin_split:
6541 rte_free(translated_actions);
6546 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6547 * incoming packets to table 1.
6549 * Other flow rules, requested for group n, will be created in
6550 * e-switch table n+1.
6551 * Jump action to e-switch group n will be created to group n+1.
6553 * Used when working in switchdev mode, to utilise advantages of table 1
6557 * Pointer to Ethernet device.
6560 * Pointer to flow on success, NULL otherwise and rte_errno is set.
6563 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6565 const struct rte_flow_attr attr = {
6572 const struct rte_flow_item pattern = {
6573 .type = RTE_FLOW_ITEM_TYPE_END,
6575 struct rte_flow_action_jump jump = {
6578 const struct rte_flow_action actions[] = {
6580 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6584 .type = RTE_FLOW_ACTION_TYPE_END,
6587 struct rte_flow_error error;
6589 return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6591 actions, false, &error);
6595 * Validate a flow supported by the NIC.
6597 * @see rte_flow_validate()
6601 mlx5_flow_validate(struct rte_eth_dev *dev,
6602 const struct rte_flow_attr *attr,
6603 const struct rte_flow_item items[],
6604 const struct rte_flow_action original_actions[],
6605 struct rte_flow_error *error)
6608 struct mlx5_translated_action_handle
6609 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6610 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6611 const struct rte_flow_action *actions;
6612 struct rte_flow_action *translated_actions = NULL;
6613 int ret = flow_action_handles_translate(dev, original_actions,
6616 &translated_actions, error);
6620 actions = translated_actions ? translated_actions : original_actions;
6621 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6622 ret = flow_drv_validate(dev, attr, items, actions,
6623 true, hairpin_flow, error);
6624 rte_free(translated_actions);
6631 * @see rte_flow_create()
6635 mlx5_flow_create(struct rte_eth_dev *dev,
6636 const struct rte_flow_attr *attr,
6637 const struct rte_flow_item items[],
6638 const struct rte_flow_action actions[],
6639 struct rte_flow_error *error)
6642 * If the device is not started yet, it is not allowed to created a
6643 * flow from application. PMD default flows and traffic control flows
6646 if (unlikely(!dev->data->dev_started)) {
6647 DRV_LOG(DEBUG, "port %u is not started when "
6648 "inserting a flow", dev->data->port_id);
6649 rte_flow_error_set(error, ENODEV,
6650 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6652 "port not started");
6656 return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6657 attr, items, actions,
6662 * Destroy a flow in a list.
6665 * Pointer to Ethernet device.
6666 * @param[in] flow_idx
6667 * Index of flow to destroy.
6670 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6673 struct mlx5_priv *priv = dev->data->dev_private;
6674 struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6678 MLX5_ASSERT(flow->type == type);
6680 * Update RX queue flags only if port is started, otherwise it is
6683 if (dev->data->dev_started)
6684 flow_rxq_flags_trim(dev, flow);
6685 flow_drv_destroy(dev, flow);
6687 struct mlx5_flow_tunnel *tunnel;
6689 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6691 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6692 mlx5_flow_tunnel_free(dev, tunnel);
6694 flow_mreg_del_copy_action(dev, flow);
6695 mlx5_ipool_free(priv->flows[type], flow_idx);
6699 * Destroy all flows.
6702 * Pointer to Ethernet device.
6704 * Flow type to be flushed.
6706 * If flushing is called avtively.
6709 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6712 struct mlx5_priv *priv = dev->data->dev_private;
6713 uint32_t num_flushed = 0, fidx = 1;
6714 struct rte_flow *flow;
6716 MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6717 flow_list_destroy(dev, type, fidx);
6721 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6722 dev->data->port_id, num_flushed);
6727 * Stop all default actions for flows.
6730 * Pointer to Ethernet device.
6733 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6735 flow_mreg_del_default_copy_action(dev);
6736 flow_rxq_flags_clear(dev);
6740 * Start all default actions for flows.
6743 * Pointer to Ethernet device.
6745 * 0 on success, a negative errno value otherwise and rte_errno is set.
6748 mlx5_flow_start_default(struct rte_eth_dev *dev)
6750 struct rte_flow_error error;
6752 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6753 return flow_mreg_add_default_copy_action(dev, &error);
6757 * Release key of thread specific flow workspace data.
6760 flow_release_workspace(void *data)
6762 struct mlx5_flow_workspace *wks = data;
6763 struct mlx5_flow_workspace *next;
6767 free(wks->rss_desc.queue);
6774 * Get thread specific current flow workspace.
6776 * @return pointer to thread specific flow workspace data, NULL on error.
6778 struct mlx5_flow_workspace*
6779 mlx5_flow_get_thread_workspace(void)
6781 struct mlx5_flow_workspace *data;
6783 data = mlx5_flow_os_get_specific_workspace();
6784 MLX5_ASSERT(data && data->inuse);
6785 if (!data || !data->inuse)
6786 DRV_LOG(ERR, "flow workspace not initialized.");
6791 * Allocate and init new flow workspace.
6793 * @return pointer to flow workspace data, NULL on error.
6795 static struct mlx5_flow_workspace*
6796 flow_alloc_thread_workspace(void)
6798 struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6801 DRV_LOG(ERR, "Failed to allocate flow workspace "
6805 data->rss_desc.queue = calloc(1,
6806 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6807 if (!data->rss_desc.queue)
6809 data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6812 if (data->rss_desc.queue)
6813 free(data->rss_desc.queue);
6819 * Get new thread specific flow workspace.
6821 * If current workspace inuse, create new one and set as current.
6823 * @return pointer to thread specific flow workspace data, NULL on error.
6825 static struct mlx5_flow_workspace*
6826 mlx5_flow_push_thread_workspace(void)
6828 struct mlx5_flow_workspace *curr;
6829 struct mlx5_flow_workspace *data;
6831 curr = mlx5_flow_os_get_specific_workspace();
6833 data = flow_alloc_thread_workspace();
6836 } else if (!curr->inuse) {
6838 } else if (curr->next) {
6841 data = flow_alloc_thread_workspace();
6849 /* Set as current workspace */
6850 if (mlx5_flow_os_set_specific_workspace(data))
6851 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6856 * Close current thread specific flow workspace.
6858 * If previous workspace available, set it as current.
6860 * @return pointer to thread specific flow workspace data, NULL on error.
6863 mlx5_flow_pop_thread_workspace(void)
6865 struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6870 DRV_LOG(ERR, "Failed to close unused flow workspace.");
6876 if (mlx5_flow_os_set_specific_workspace(data->prev))
6877 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6881 * Verify the flow list is empty
6884 * Pointer to Ethernet device.
6886 * @return the number of flows not released.
6889 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
6891 struct mlx5_priv *priv = dev->data->dev_private;
6892 struct rte_flow *flow;
6896 for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
6897 MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
6898 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6899 dev->data->port_id, (void *)flow);
6907 * Enable default hairpin egress flow.
6910 * Pointer to Ethernet device.
6915 * 0 on success, a negative errno value otherwise and rte_errno is set.
6918 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6921 const struct rte_flow_attr attr = {
6925 struct mlx5_rte_flow_item_tx_queue queue_spec = {
6928 struct mlx5_rte_flow_item_tx_queue queue_mask = {
6929 .queue = UINT32_MAX,
6931 struct rte_flow_item items[] = {
6933 .type = (enum rte_flow_item_type)
6934 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6935 .spec = &queue_spec,
6937 .mask = &queue_mask,
6940 .type = RTE_FLOW_ITEM_TYPE_END,
6943 struct rte_flow_action_jump jump = {
6944 .group = MLX5_HAIRPIN_TX_TABLE,
6946 struct rte_flow_action actions[2];
6948 struct rte_flow_error error;
6950 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6951 actions[0].conf = &jump;
6952 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6953 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6954 &attr, items, actions, false, &error);
6957 "Failed to create ctrl flow: rte_errno(%d),"
6958 " type(%d), message(%s)",
6959 rte_errno, error.type,
6960 error.message ? error.message : " (no stated reason)");
6967 * Enable a control flow configured from the control plane.
6970 * Pointer to Ethernet device.
6972 * An Ethernet flow spec to apply.
6974 * An Ethernet flow mask to apply.
6976 * A VLAN flow spec to apply.
6978 * A VLAN flow mask to apply.
6981 * 0 on success, a negative errno value otherwise and rte_errno is set.
6984 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6985 struct rte_flow_item_eth *eth_spec,
6986 struct rte_flow_item_eth *eth_mask,
6987 struct rte_flow_item_vlan *vlan_spec,
6988 struct rte_flow_item_vlan *vlan_mask)
6990 struct mlx5_priv *priv = dev->data->dev_private;
6991 const struct rte_flow_attr attr = {
6993 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6995 struct rte_flow_item items[] = {
6997 .type = RTE_FLOW_ITEM_TYPE_ETH,
7003 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7004 RTE_FLOW_ITEM_TYPE_END,
7010 .type = RTE_FLOW_ITEM_TYPE_END,
7013 uint16_t queue[priv->reta_idx_n];
7014 struct rte_flow_action_rss action_rss = {
7015 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7017 .types = priv->rss_conf.rss_hf,
7018 .key_len = priv->rss_conf.rss_key_len,
7019 .queue_num = priv->reta_idx_n,
7020 .key = priv->rss_conf.rss_key,
7023 struct rte_flow_action actions[] = {
7025 .type = RTE_FLOW_ACTION_TYPE_RSS,
7026 .conf = &action_rss,
7029 .type = RTE_FLOW_ACTION_TYPE_END,
7033 struct rte_flow_error error;
7036 if (!priv->reta_idx_n || !priv->rxqs_n) {
7039 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7040 action_rss.types = 0;
7041 for (i = 0; i != priv->reta_idx_n; ++i)
7042 queue[i] = (*priv->reta_idx)[i];
7043 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7044 &attr, items, actions, false, &error);
7051 * Enable a flow control configured from the control plane.
7054 * Pointer to Ethernet device.
7056 * An Ethernet flow spec to apply.
7058 * An Ethernet flow mask to apply.
7061 * 0 on success, a negative errno value otherwise and rte_errno is set.
7064 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7065 struct rte_flow_item_eth *eth_spec,
7066 struct rte_flow_item_eth *eth_mask)
7068 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7072 * Create default miss flow rule matching lacp traffic
7075 * Pointer to Ethernet device.
7077 * An Ethernet flow spec to apply.
7080 * 0 on success, a negative errno value otherwise and rte_errno is set.
7083 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7086 * The LACP matching is done by only using ether type since using
7087 * a multicast dst mac causes kernel to give low priority to this flow.
7089 static const struct rte_flow_item_eth lacp_spec = {
7090 .type = RTE_BE16(0x8809),
7092 static const struct rte_flow_item_eth lacp_mask = {
7095 const struct rte_flow_attr attr = {
7098 struct rte_flow_item items[] = {
7100 .type = RTE_FLOW_ITEM_TYPE_ETH,
7105 .type = RTE_FLOW_ITEM_TYPE_END,
7108 struct rte_flow_action actions[] = {
7110 .type = (enum rte_flow_action_type)
7111 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7114 .type = RTE_FLOW_ACTION_TYPE_END,
7117 struct rte_flow_error error;
7118 uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7119 &attr, items, actions,
7130 * @see rte_flow_destroy()
7134 mlx5_flow_destroy(struct rte_eth_dev *dev,
7135 struct rte_flow *flow,
7136 struct rte_flow_error *error __rte_unused)
7138 flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7139 (uintptr_t)(void *)flow);
7144 * Destroy all flows.
7146 * @see rte_flow_flush()
7150 mlx5_flow_flush(struct rte_eth_dev *dev,
7151 struct rte_flow_error *error __rte_unused)
7153 mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7160 * @see rte_flow_isolate()
7164 mlx5_flow_isolate(struct rte_eth_dev *dev,
7166 struct rte_flow_error *error)
7168 struct mlx5_priv *priv = dev->data->dev_private;
7170 if (dev->data->dev_started) {
7171 rte_flow_error_set(error, EBUSY,
7172 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7174 "port must be stopped first");
7177 priv->isolated = !!enable;
7179 dev->dev_ops = &mlx5_dev_ops_isolate;
7181 dev->dev_ops = &mlx5_dev_ops;
7183 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7184 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7192 * @see rte_flow_query()
7196 flow_drv_query(struct rte_eth_dev *dev,
7198 const struct rte_flow_action *actions,
7200 struct rte_flow_error *error)
7202 struct mlx5_priv *priv = dev->data->dev_private;
7203 const struct mlx5_flow_driver_ops *fops;
7204 struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7206 enum mlx5_flow_drv_type ftype;
7209 return rte_flow_error_set(error, ENOENT,
7210 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7212 "invalid flow handle");
7214 ftype = flow->drv_type;
7215 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7216 fops = flow_get_drv_ops(ftype);
7218 return fops->query(dev, flow, actions, data, error);
7224 * @see rte_flow_query()
7228 mlx5_flow_query(struct rte_eth_dev *dev,
7229 struct rte_flow *flow,
7230 const struct rte_flow_action *actions,
7232 struct rte_flow_error *error)
7236 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7244 * Get rte_flow callbacks.
7247 * Pointer to Ethernet device structure.
7249 * Pointer to operation-specific structure.
7254 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7255 const struct rte_flow_ops **ops)
7257 *ops = &mlx5_flow_ops;
7262 * Validate meter policy actions.
7263 * Dispatcher for action type specific validation.
7266 * Pointer to the Ethernet device structure.
7268 * The meter policy action object to validate.
7270 * Attributes of flow to determine steering domain.
7271 * @param[out] is_rss
7273 * @param[out] domain_bitmap
7275 * @param[out] is_def_policy
7276 * Is default policy or not.
7278 * Perform verbose error reporting if not NULL. Initialized in case of
7282 * 0 on success, otherwise negative errno value.
7285 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7286 const struct rte_flow_action *actions[RTE_COLORS],
7287 struct rte_flow_attr *attr,
7289 uint8_t *domain_bitmap,
7290 uint8_t *policy_mode,
7291 struct rte_mtr_error *error)
7293 const struct mlx5_flow_driver_ops *fops;
7295 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7296 return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7297 domain_bitmap, policy_mode, error);
7301 * Destroy the meter table set.
7304 * Pointer to Ethernet device.
7305 * @param[in] mtr_policy
7306 * Meter policy struct.
7309 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7310 struct mlx5_flow_meter_policy *mtr_policy)
7312 const struct mlx5_flow_driver_ops *fops;
7314 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7315 fops->destroy_mtr_acts(dev, mtr_policy);
7319 * Create policy action, lock free,
7320 * (mutex should be acquired by caller).
7321 * Dispatcher for action type specific call.
7324 * Pointer to the Ethernet device structure.
7325 * @param[in] mtr_policy
7326 * Meter policy struct.
7328 * Action specification used to create meter actions.
7330 * Perform verbose error reporting if not NULL. Initialized in case of
7334 * 0 on success, otherwise negative errno value.
7337 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7338 struct mlx5_flow_meter_policy *mtr_policy,
7339 const struct rte_flow_action *actions[RTE_COLORS],
7340 struct rte_mtr_error *error)
7342 const struct mlx5_flow_driver_ops *fops;
7344 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7345 return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7349 * Create policy rules, lock free,
7350 * (mutex should be acquired by caller).
7351 * Dispatcher for action type specific call.
7354 * Pointer to the Ethernet device structure.
7355 * @param[in] mtr_policy
7356 * Meter policy struct.
7359 * 0 on success, -1 otherwise.
7362 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7363 struct mlx5_flow_meter_policy *mtr_policy)
7365 const struct mlx5_flow_driver_ops *fops;
7367 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7368 return fops->create_policy_rules(dev, mtr_policy);
7372 * Destroy policy rules, lock free,
7373 * (mutex should be acquired by caller).
7374 * Dispatcher for action type specific call.
7377 * Pointer to the Ethernet device structure.
7378 * @param[in] mtr_policy
7379 * Meter policy struct.
7382 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7383 struct mlx5_flow_meter_policy *mtr_policy)
7385 const struct mlx5_flow_driver_ops *fops;
7387 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7388 fops->destroy_policy_rules(dev, mtr_policy);
7392 * Destroy the default policy table set.
7395 * Pointer to Ethernet device.
7398 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7400 const struct mlx5_flow_driver_ops *fops;
7402 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7403 fops->destroy_def_policy(dev);
7407 * Destroy the default policy table set.
7410 * Pointer to Ethernet device.
7413 * 0 on success, -1 otherwise.
7416 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7418 const struct mlx5_flow_driver_ops *fops;
7420 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7421 return fops->create_def_policy(dev);
7425 * Create the needed meter and suffix tables.
7428 * Pointer to Ethernet device.
7431 * 0 on success, -1 otherwise.
7434 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7435 struct mlx5_flow_meter_info *fm,
7437 uint8_t domain_bitmap)
7439 const struct mlx5_flow_driver_ops *fops;
7441 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7442 return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7446 * Destroy the meter table set.
7449 * Pointer to Ethernet device.
7451 * Pointer to the meter table set.
7454 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7455 struct mlx5_flow_meter_info *fm)
7457 const struct mlx5_flow_driver_ops *fops;
7459 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7460 fops->destroy_mtr_tbls(dev, fm);
7464 * Destroy the global meter drop table.
7467 * Pointer to Ethernet device.
7470 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7472 const struct mlx5_flow_driver_ops *fops;
7474 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7475 fops->destroy_mtr_drop_tbls(dev);
7479 * Destroy the sub policy table with RX queue.
7482 * Pointer to Ethernet device.
7483 * @param[in] mtr_policy
7484 * Pointer to meter policy table.
7487 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7488 struct mlx5_flow_meter_policy *mtr_policy)
7490 const struct mlx5_flow_driver_ops *fops;
7492 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7493 fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7497 * Allocate the needed aso flow meter id.
7500 * Pointer to Ethernet device.
7503 * Index to aso flow meter on success, NULL otherwise.
7506 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7508 const struct mlx5_flow_driver_ops *fops;
7510 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7511 return fops->create_meter(dev);
7515 * Free the aso flow meter id.
7518 * Pointer to Ethernet device.
7519 * @param[in] mtr_idx
7520 * Index to aso flow meter to be free.
7526 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7528 const struct mlx5_flow_driver_ops *fops;
7530 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7531 fops->free_meter(dev, mtr_idx);
7535 * Allocate a counter.
7538 * Pointer to Ethernet device structure.
7541 * Index to allocated counter on success, 0 otherwise.
7544 mlx5_counter_alloc(struct rte_eth_dev *dev)
7546 const struct mlx5_flow_driver_ops *fops;
7547 struct rte_flow_attr attr = { .transfer = 0 };
7549 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7550 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7551 return fops->counter_alloc(dev);
7554 "port %u counter allocate is not supported.",
7555 dev->data->port_id);
7563 * Pointer to Ethernet device structure.
7565 * Index to counter to be free.
7568 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7570 const struct mlx5_flow_driver_ops *fops;
7571 struct rte_flow_attr attr = { .transfer = 0 };
7573 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7574 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7575 fops->counter_free(dev, cnt);
7579 "port %u counter free is not supported.",
7580 dev->data->port_id);
7584 * Query counter statistics.
7587 * Pointer to Ethernet device structure.
7589 * Index to counter to query.
7591 * Set to clear counter statistics.
7593 * The counter hits packets number to save.
7595 * The counter hits bytes number to save.
7598 * 0 on success, a negative errno value otherwise.
7601 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7602 bool clear, uint64_t *pkts, uint64_t *bytes)
7604 const struct mlx5_flow_driver_ops *fops;
7605 struct rte_flow_attr attr = { .transfer = 0 };
7607 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7608 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7609 return fops->counter_query(dev, cnt, clear, pkts, bytes);
7612 "port %u counter query is not supported.",
7613 dev->data->port_id);
7618 * Allocate a new memory for the counter values wrapped by all the needed
7622 * Pointer to mlx5_dev_ctx_shared object.
7625 * 0 on success, a negative errno value otherwise.
7628 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7630 struct mlx5_devx_mkey_attr mkey_attr;
7631 struct mlx5_counter_stats_mem_mng *mem_mng;
7632 volatile struct flow_counter_stats *raw_data;
7633 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7634 int size = (sizeof(struct flow_counter_stats) *
7635 MLX5_COUNTERS_PER_POOL +
7636 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7637 sizeof(struct mlx5_counter_stats_mem_mng);
7638 size_t pgsize = rte_mem_page_size();
7642 if (pgsize == (size_t)-1) {
7643 DRV_LOG(ERR, "Failed to get mem page size");
7647 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7652 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7653 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7654 mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7655 IBV_ACCESS_LOCAL_WRITE);
7656 if (!mem_mng->umem) {
7661 memset(&mkey_attr, 0, sizeof(mkey_attr));
7662 mkey_attr.addr = (uintptr_t)mem;
7663 mkey_attr.size = size;
7664 mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7665 mkey_attr.pd = sh->pdn;
7666 mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7667 mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7668 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7670 mlx5_os_umem_dereg(mem_mng->umem);
7675 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7676 raw_data = (volatile struct flow_counter_stats *)mem;
7677 for (i = 0; i < raws_n; ++i) {
7678 mem_mng->raws[i].mem_mng = mem_mng;
7679 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7681 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7682 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7683 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7685 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7686 sh->cmng.mem_mng = mem_mng;
7691 * Set the statistic memory to the new counter pool.
7694 * Pointer to mlx5_dev_ctx_shared object.
7696 * Pointer to the pool to set the statistic memory.
7699 * 0 on success, a negative errno value otherwise.
7702 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7703 struct mlx5_flow_counter_pool *pool)
7705 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7706 /* Resize statistic memory once used out. */
7707 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7708 mlx5_flow_create_counter_stat_mem_mng(sh)) {
7709 DRV_LOG(ERR, "Cannot resize counter stat mem.");
7712 rte_spinlock_lock(&pool->sl);
7713 pool->raw = cmng->mem_mng->raws + pool->index %
7714 MLX5_CNT_CONTAINER_RESIZE;
7715 rte_spinlock_unlock(&pool->sl);
7716 pool->raw_hw = NULL;
7720 #define MLX5_POOL_QUERY_FREQ_US 1000000
7723 * Set the periodic procedure for triggering asynchronous batch queries for all
7724 * the counter pools.
7727 * Pointer to mlx5_dev_ctx_shared object.
7730 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7732 uint32_t pools_n, us;
7734 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7735 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7736 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7737 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7738 sh->cmng.query_thread_on = 0;
7739 DRV_LOG(ERR, "Cannot reinitialize query alarm");
7741 sh->cmng.query_thread_on = 1;
7746 * The periodic procedure for triggering asynchronous batch queries for all the
7747 * counter pools. This function is probably called by the host thread.
7750 * The parameter for the alarm process.
7753 mlx5_flow_query_alarm(void *arg)
7755 struct mlx5_dev_ctx_shared *sh = arg;
7757 uint16_t pool_index = sh->cmng.pool_index;
7758 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7759 struct mlx5_flow_counter_pool *pool;
7762 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7764 rte_spinlock_lock(&cmng->pool_update_sl);
7765 pool = cmng->pools[pool_index];
7766 n_valid = cmng->n_valid;
7767 rte_spinlock_unlock(&cmng->pool_update_sl);
7768 /* Set the statistic memory to the new created pool. */
7769 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7772 /* There is a pool query in progress. */
7775 LIST_FIRST(&sh->cmng.free_stat_raws);
7777 /* No free counter statistics raw memory. */
7780 * Identify the counters released between query trigger and query
7781 * handle more efficiently. The counter released in this gap period
7782 * should wait for a new round of query as the new arrived packets
7783 * will not be taken into account.
7786 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7787 MLX5_COUNTERS_PER_POOL,
7789 pool->raw_hw->mem_mng->dm->id,
7793 (uint64_t)(uintptr_t)pool);
7795 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7796 " %d", pool->min_dcs->id);
7797 pool->raw_hw = NULL;
7800 LIST_REMOVE(pool->raw_hw, next);
7801 sh->cmng.pending_queries++;
7803 if (pool_index >= n_valid)
7806 sh->cmng.pool_index = pool_index;
7807 mlx5_set_query_alarm(sh);
7811 * Check and callback event for new aged flow in the counter pool
7814 * Pointer to mlx5_dev_ctx_shared object.
7816 * Pointer to Current counter pool.
7819 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7820 struct mlx5_flow_counter_pool *pool)
7822 struct mlx5_priv *priv;
7823 struct mlx5_flow_counter *cnt;
7824 struct mlx5_age_info *age_info;
7825 struct mlx5_age_param *age_param;
7826 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7827 struct mlx5_counter_stats_raw *prev = pool->raw;
7828 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7829 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7830 uint16_t expected = AGE_CANDIDATE;
7833 pool->time_of_last_age_check = curr_time;
7834 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7835 cnt = MLX5_POOL_GET_CNT(pool, i);
7836 age_param = MLX5_CNT_TO_AGE(cnt);
7837 if (__atomic_load_n(&age_param->state,
7838 __ATOMIC_RELAXED) != AGE_CANDIDATE)
7840 if (cur->data[i].hits != prev->data[i].hits) {
7841 __atomic_store_n(&age_param->sec_since_last_hit, 0,
7845 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7847 __ATOMIC_RELAXED) <= age_param->timeout)
7850 * Hold the lock first, or if between the
7851 * state AGE_TMOUT and tailq operation the
7852 * release happened, the release procedure
7853 * may delete a non-existent tailq node.
7855 priv = rte_eth_devices[age_param->port_id].data->dev_private;
7856 age_info = GET_PORT_AGE_INFO(priv);
7857 rte_spinlock_lock(&age_info->aged_sl);
7858 if (__atomic_compare_exchange_n(&age_param->state, &expected,
7861 __ATOMIC_RELAXED)) {
7862 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7863 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7865 rte_spinlock_unlock(&age_info->aged_sl);
7867 mlx5_age_event_prepare(sh);
7871 * Handler for the HW respond about ready values from an asynchronous batch
7872 * query. This function is probably called by the host thread.
7875 * The pointer to the shared device context.
7876 * @param[in] async_id
7877 * The Devx async ID.
7879 * The status of the completion.
7882 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7883 uint64_t async_id, int status)
7885 struct mlx5_flow_counter_pool *pool =
7886 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7887 struct mlx5_counter_stats_raw *raw_to_free;
7888 uint8_t query_gen = pool->query_gen ^ 1;
7889 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7890 enum mlx5_counter_type cnt_type =
7891 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7892 MLX5_COUNTER_TYPE_ORIGIN;
7894 if (unlikely(status)) {
7895 raw_to_free = pool->raw_hw;
7897 raw_to_free = pool->raw;
7899 mlx5_flow_aging_check(sh, pool);
7900 rte_spinlock_lock(&pool->sl);
7901 pool->raw = pool->raw_hw;
7902 rte_spinlock_unlock(&pool->sl);
7903 /* Be sure the new raw counters data is updated in memory. */
7905 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7906 rte_spinlock_lock(&cmng->csl[cnt_type]);
7907 TAILQ_CONCAT(&cmng->counters[cnt_type],
7908 &pool->counters[query_gen], next);
7909 rte_spinlock_unlock(&cmng->csl[cnt_type]);
7912 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7913 pool->raw_hw = NULL;
7914 sh->cmng.pending_queries--;
7918 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7919 const struct flow_grp_info *grp_info,
7920 struct rte_flow_error *error)
7922 if (grp_info->transfer && grp_info->external &&
7923 grp_info->fdb_def_rule) {
7924 if (group == UINT32_MAX)
7925 return rte_flow_error_set
7927 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7929 "group index not supported");
7934 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7939 * Translate the rte_flow group index to HW table value.
7941 * If tunnel offload is disabled, all group ids converted to flow table
7942 * id using the standard method.
7943 * If tunnel offload is enabled, group id can be converted using the
7944 * standard or tunnel conversion method. Group conversion method
7945 * selection depends on flags in `grp_info` parameter:
7946 * - Internal (grp_info.external == 0) groups conversion uses the
7948 * - Group ids in JUMP action converted with the tunnel conversion.
7949 * - Group id in rule attribute conversion depends on a rule type and
7951 * ** non zero group attributes converted with the tunnel method
7952 * ** zero group attribute in non-tunnel rule is converted using the
7953 * standard method - there's only one root table
7954 * ** zero group attribute in steer tunnel rule is converted with the
7955 * standard method - single root table
7956 * ** zero group attribute in match tunnel rule is a special OvS
7957 * case: that value is used for portability reasons. That group
7958 * id is converted with the tunnel conversion method.
7963 * PMD tunnel offload object
7965 * rte_flow group index value.
7968 * @param[in] grp_info
7969 * flags used for conversion
7971 * Pointer to error structure.
7974 * 0 on success, a negative errno value otherwise and rte_errno is set.
7977 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7978 const struct mlx5_flow_tunnel *tunnel,
7979 uint32_t group, uint32_t *table,
7980 const struct flow_grp_info *grp_info,
7981 struct rte_flow_error *error)
7984 bool standard_translation;
7986 if (!grp_info->skip_scale && grp_info->external &&
7987 group < MLX5_MAX_TABLES_EXTERNAL)
7988 group *= MLX5_FLOW_TABLE_FACTOR;
7989 if (is_tunnel_offload_active(dev)) {
7990 standard_translation = !grp_info->external ||
7991 grp_info->std_tbl_fix;
7993 standard_translation = true;
7996 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7997 dev->data->port_id, group, grp_info->transfer,
7998 grp_info->external, grp_info->fdb_def_rule,
7999 standard_translation ? "STANDARD" : "TUNNEL");
8000 if (standard_translation)
8001 ret = flow_group_to_table(dev->data->port_id, group, table,
8004 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8011 * Discover availability of metadata reg_c's.
8013 * Iteratively use test flows to check availability.
8016 * Pointer to the Ethernet device structure.
8019 * 0 on success, a negative errno value otherwise and rte_errno is set.
8022 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8024 struct mlx5_priv *priv = dev->data->dev_private;
8025 struct mlx5_dev_config *config = &priv->config;
8026 enum modify_reg idx;
8029 /* reg_c[0] and reg_c[1] are reserved. */
8030 config->flow_mreg_c[n++] = REG_C_0;
8031 config->flow_mreg_c[n++] = REG_C_1;
8032 /* Discover availability of other reg_c's. */
8033 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8034 struct rte_flow_attr attr = {
8035 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8036 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8039 struct rte_flow_item items[] = {
8041 .type = RTE_FLOW_ITEM_TYPE_END,
8044 struct rte_flow_action actions[] = {
8046 .type = (enum rte_flow_action_type)
8047 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8048 .conf = &(struct mlx5_flow_action_copy_mreg){
8054 .type = RTE_FLOW_ACTION_TYPE_JUMP,
8055 .conf = &(struct rte_flow_action_jump){
8056 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8060 .type = RTE_FLOW_ACTION_TYPE_END,
8064 struct rte_flow *flow;
8065 struct rte_flow_error error;
8067 if (!config->dv_flow_en)
8069 /* Create internal flow, validation skips copy action. */
8070 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8071 items, actions, false, &error);
8072 flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8076 config->flow_mreg_c[n++] = idx;
8077 flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8079 for (; n < MLX5_MREG_C_NUM; ++n)
8080 config->flow_mreg_c[n] = REG_NON;
8085 save_dump_file(const uint8_t *data, uint32_t size,
8086 uint32_t type, uint32_t id, void *arg, FILE *file)
8088 char line[BUF_SIZE];
8091 uint32_t actions_num;
8092 struct rte_flow_query_count *count;
8094 memset(line, 0, BUF_SIZE);
8096 case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8097 actions_num = *(uint32_t *)(arg);
8098 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,%d,",
8099 type, id, actions_num);
8101 case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8102 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,",
8105 case DR_DUMP_REC_TYPE_PMD_COUNTER:
8106 count = (struct rte_flow_query_count *)arg;
8107 fprintf(file, "%d,0x%x,%" PRIu64 ",%" PRIu64 "\n", type,
8108 id, count->hits, count->bytes);
8114 for (k = 0; k < size; k++) {
8115 /* Make sure we do not overrun the line buffer length. */
8116 if (out >= BUF_SIZE - 4) {
8120 out += snprintf(line + out, BUF_SIZE - out, "%02x",
8123 fprintf(file, "%s\n", line);
8128 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8129 struct rte_flow_query_count *count, struct rte_flow_error *error)
8131 struct rte_flow_action action[2];
8132 enum mlx5_flow_drv_type ftype;
8133 const struct mlx5_flow_driver_ops *fops;
8136 return rte_flow_error_set(error, ENOENT,
8137 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8139 "invalid flow handle");
8141 action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8142 action[1].type = RTE_FLOW_ACTION_TYPE_END;
8143 if (flow->counter) {
8144 memset(count, 0, sizeof(struct rte_flow_query_count));
8145 ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8146 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8147 ftype < MLX5_FLOW_TYPE_MAX);
8148 fops = flow_get_drv_ops(ftype);
8149 return fops->query(dev, flow, action, count, error);
8154 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8156 * Dump flow ipool data to file
8159 * The pointer to Ethernet device.
8161 * A pointer to a file for output.
8163 * Perform verbose error reporting if not NULL. PMDs initialize this
8164 * structure in case of error only.
8166 * 0 on success, a negative value otherwise.
8169 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8170 struct rte_flow *flow, FILE *file,
8171 struct rte_flow_error *error)
8173 struct mlx5_priv *priv = dev->data->dev_private;
8174 struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
8175 struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8176 uint32_t handle_idx;
8177 struct mlx5_flow_handle *dh;
8178 struct rte_flow_query_count count;
8179 uint32_t actions_num;
8180 const uint8_t *data;
8186 return rte_flow_error_set(error, ENOENT,
8187 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8189 "invalid flow handle");
8191 handle_idx = flow->dev_handles;
8192 while (handle_idx) {
8193 dh = mlx5_ipool_get(priv->sh->ipool
8194 [MLX5_IPOOL_MLX5_FLOW], handle_idx);
8197 handle_idx = dh->next.next;
8198 id = (uint32_t)(uintptr_t)dh->drv_flow;
8201 type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8202 if (!mlx5_flow_query_counter(dev, flow, &count, error))
8203 save_dump_file(NULL, 0, type,
8204 id, (void *)&count, file);
8206 /* Get modify_hdr and encap_decap buf from ipools. */
8208 modify_hdr = dh->dvh.modify_hdr;
8210 if (dh->dvh.rix_encap_decap) {
8211 encap_decap = mlx5_ipool_get(priv->sh->ipool
8212 [MLX5_IPOOL_DECAP_ENCAP],
8213 dh->dvh.rix_encap_decap);
8216 data = (const uint8_t *)modify_hdr->actions;
8217 size = (size_t)(modify_hdr->actions_num) * 8;
8218 actions_num = modify_hdr->actions_num;
8219 type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8220 save_dump_file(data, size, type, id,
8221 (void *)(&actions_num), file);
8224 data = encap_decap->buf;
8225 size = encap_decap->size;
8226 type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8227 save_dump_file(data, size, type,
8236 * Dump flow raw hw data to file
8239 * The pointer to Ethernet device.
8241 * A pointer to a file for output.
8243 * Perform verbose error reporting if not NULL. PMDs initialize this
8244 * structure in case of error only.
8246 * 0 on success, a nagative value otherwise.
8249 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8251 struct rte_flow_error *error __rte_unused)
8253 struct mlx5_priv *priv = dev->data->dev_private;
8254 struct mlx5_dev_ctx_shared *sh = priv->sh;
8255 uint32_t handle_idx;
8257 struct mlx5_flow_handle *dh;
8258 struct rte_flow *flow;
8259 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8263 if (!priv->config.dv_flow_en) {
8264 if (fputs("device dv flow disabled\n", file) <= 0)
8271 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8272 MLX5_IPOOL_FOREACH(priv->flows[MLX5_FLOW_TYPE_GEN], idx, flow)
8273 mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8275 return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8277 sh->tx_domain, file);
8280 flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8281 (uintptr_t)(void *)flow_idx);
8285 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8286 mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8288 handle_idx = flow->dev_handles;
8289 while (handle_idx) {
8290 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8295 ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8300 handle_idx = dh->next.next;
8306 * Get aged-out flows.
8309 * Pointer to the Ethernet device structure.
8310 * @param[in] context
8311 * The address of an array of pointers to the aged-out flows contexts.
8312 * @param[in] nb_countexts
8313 * The length of context array pointers.
8315 * Perform verbose error reporting if not NULL. Initialized in case of
8319 * how many contexts get in success, otherwise negative errno value.
8320 * if nb_contexts is 0, return the amount of all aged contexts.
8321 * if nb_contexts is not 0 , return the amount of aged flows reported
8322 * in the context array.
8325 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8326 uint32_t nb_contexts, struct rte_flow_error *error)
8328 const struct mlx5_flow_driver_ops *fops;
8329 struct rte_flow_attr attr = { .transfer = 0 };
8331 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8332 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8333 return fops->get_aged_flows(dev, contexts, nb_contexts,
8337 "port %u get aged flows is not supported.",
8338 dev->data->port_id);
8342 /* Wrapper for driver action_validate op callback */
8344 flow_drv_action_validate(struct rte_eth_dev *dev,
8345 const struct rte_flow_indir_action_conf *conf,
8346 const struct rte_flow_action *action,
8347 const struct mlx5_flow_driver_ops *fops,
8348 struct rte_flow_error *error)
8350 static const char err_msg[] = "indirect action validation unsupported";
8352 if (!fops->action_validate) {
8353 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8354 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8358 return fops->action_validate(dev, conf, action, error);
8362 * Destroys the shared action by handle.
8365 * Pointer to Ethernet device structure.
8367 * Handle for the indirect action object to be destroyed.
8369 * Perform verbose error reporting if not NULL. PMDs initialize this
8370 * structure in case of error only.
8373 * 0 on success, a negative errno value otherwise and rte_errno is set.
8375 * @note: wrapper for driver action_create op callback.
8378 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8379 struct rte_flow_action_handle *handle,
8380 struct rte_flow_error *error)
8382 static const char err_msg[] = "indirect action destruction unsupported";
8383 struct rte_flow_attr attr = { .transfer = 0 };
8384 const struct mlx5_flow_driver_ops *fops =
8385 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8387 if (!fops->action_destroy) {
8388 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8389 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8393 return fops->action_destroy(dev, handle, error);
8396 /* Wrapper for driver action_destroy op callback */
8398 flow_drv_action_update(struct rte_eth_dev *dev,
8399 struct rte_flow_action_handle *handle,
8401 const struct mlx5_flow_driver_ops *fops,
8402 struct rte_flow_error *error)
8404 static const char err_msg[] = "indirect action update unsupported";
8406 if (!fops->action_update) {
8407 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8408 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8412 return fops->action_update(dev, handle, update, error);
8415 /* Wrapper for driver action_destroy op callback */
8417 flow_drv_action_query(struct rte_eth_dev *dev,
8418 const struct rte_flow_action_handle *handle,
8420 const struct mlx5_flow_driver_ops *fops,
8421 struct rte_flow_error *error)
8423 static const char err_msg[] = "indirect action query unsupported";
8425 if (!fops->action_query) {
8426 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8427 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8431 return fops->action_query(dev, handle, data, error);
8435 * Create indirect action for reuse in multiple flow rules.
8438 * Pointer to Ethernet device structure.
8440 * Pointer to indirect action object configuration.
8442 * Action configuration for indirect action object creation.
8444 * Perform verbose error reporting if not NULL. PMDs initialize this
8445 * structure in case of error only.
8447 * A valid handle in case of success, NULL otherwise and rte_errno is set.
8449 static struct rte_flow_action_handle *
8450 mlx5_action_handle_create(struct rte_eth_dev *dev,
8451 const struct rte_flow_indir_action_conf *conf,
8452 const struct rte_flow_action *action,
8453 struct rte_flow_error *error)
8455 static const char err_msg[] = "indirect action creation unsupported";
8456 struct rte_flow_attr attr = { .transfer = 0 };
8457 const struct mlx5_flow_driver_ops *fops =
8458 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8460 if (flow_drv_action_validate(dev, conf, action, fops, error))
8462 if (!fops->action_create) {
8463 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8464 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8468 return fops->action_create(dev, conf, action, error);
8472 * Updates inplace the indirect action configuration pointed by *handle*
8473 * with the configuration provided as *update* argument.
8474 * The update of the indirect action configuration effects all flow rules
8475 * reusing the action via handle.
8478 * Pointer to Ethernet device structure.
8480 * Handle for the indirect action to be updated.
8482 * Action specification used to modify the action pointed by handle.
8483 * *update* could be of same type with the action pointed by the *handle*
8484 * handle argument, or some other structures like a wrapper, depending on
8485 * the indirect action type.
8487 * Perform verbose error reporting if not NULL. PMDs initialize this
8488 * structure in case of error only.
8491 * 0 on success, a negative errno value otherwise and rte_errno is set.
8494 mlx5_action_handle_update(struct rte_eth_dev *dev,
8495 struct rte_flow_action_handle *handle,
8497 struct rte_flow_error *error)
8499 struct rte_flow_attr attr = { .transfer = 0 };
8500 const struct mlx5_flow_driver_ops *fops =
8501 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8504 ret = flow_drv_action_validate(dev, NULL,
8505 (const struct rte_flow_action *)update, fops, error);
8508 return flow_drv_action_update(dev, handle, update, fops,
8513 * Query the indirect action by handle.
8515 * This function allows retrieving action-specific data such as counters.
8516 * Data is gathered by special action which may be present/referenced in
8517 * more than one flow rule definition.
8519 * see @RTE_FLOW_ACTION_TYPE_COUNT
8522 * Pointer to Ethernet device structure.
8524 * Handle for the indirect action to query.
8525 * @param[in, out] data
8526 * Pointer to storage for the associated query data type.
8528 * Perform verbose error reporting if not NULL. PMDs initialize this
8529 * structure in case of error only.
8532 * 0 on success, a negative errno value otherwise and rte_errno is set.
8535 mlx5_action_handle_query(struct rte_eth_dev *dev,
8536 const struct rte_flow_action_handle *handle,
8538 struct rte_flow_error *error)
8540 struct rte_flow_attr attr = { .transfer = 0 };
8541 const struct mlx5_flow_driver_ops *fops =
8542 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8544 return flow_drv_action_query(dev, handle, data, fops, error);
8548 * Destroy all indirect actions (shared RSS).
8551 * Pointer to Ethernet device.
8554 * 0 on success, a negative errno value otherwise and rte_errno is set.
8557 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8559 struct rte_flow_error error;
8560 struct mlx5_priv *priv = dev->data->dev_private;
8561 struct mlx5_shared_action_rss *shared_rss;
8565 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8566 priv->rss_shared_actions, idx, shared_rss, next) {
8567 ret |= mlx5_action_handle_destroy(dev,
8568 (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8573 #ifndef HAVE_MLX5DV_DR
8574 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8576 #define MLX5_DOMAIN_SYNC_FLOW \
8577 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8580 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8582 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8583 const struct mlx5_flow_driver_ops *fops;
8585 struct rte_flow_attr attr = { .transfer = 0 };
8587 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8588 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8594 const struct mlx5_flow_tunnel *
8595 mlx5_get_tof(const struct rte_flow_item *item,
8596 const struct rte_flow_action *action,
8597 enum mlx5_tof_rule_type *rule_type)
8599 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8600 if (item->type == (typeof(item->type))
8601 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8602 *rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8603 return flow_items_to_tunnel(item);
8606 for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8607 if (action->type == (typeof(action->type))
8608 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8609 *rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8610 return flow_actions_to_tunnel(action);
8617 * tunnel offload functionalilty is defined for DV environment only
8619 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8621 union tunnel_offload_mark {
8624 uint32_t app_reserve:8;
8625 uint32_t table_id:15;
8626 uint32_t transfer:1;
8627 uint32_t _unused_:8;
8632 mlx5_access_tunnel_offload_db
8633 (struct rte_eth_dev *dev,
8634 bool (*match)(struct rte_eth_dev *,
8635 struct mlx5_flow_tunnel *, const void *),
8636 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8637 void (*miss)(struct rte_eth_dev *, void *),
8638 void *ctx, bool lock_op);
8641 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8642 struct rte_flow *flow,
8643 const struct rte_flow_attr *attr,
8644 const struct rte_flow_action *app_actions,
8646 const struct mlx5_flow_tunnel *tunnel,
8647 struct tunnel_default_miss_ctx *ctx,
8648 struct rte_flow_error *error)
8650 struct mlx5_priv *priv = dev->data->dev_private;
8651 struct mlx5_flow *dev_flow;
8652 struct rte_flow_attr miss_attr = *attr;
8653 const struct rte_flow_item miss_items[2] = {
8655 .type = RTE_FLOW_ITEM_TYPE_ETH,
8661 .type = RTE_FLOW_ITEM_TYPE_END,
8667 union tunnel_offload_mark mark_id;
8668 struct rte_flow_action_mark miss_mark;
8669 struct rte_flow_action miss_actions[3] = {
8670 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8671 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
8673 const struct rte_flow_action_jump *jump_data;
8674 uint32_t i, flow_table = 0; /* prevent compilation warning */
8675 struct flow_grp_info grp_info = {
8677 .transfer = attr->transfer,
8678 .fdb_def_rule = !!priv->fdb_def_rule,
8683 if (!attr->transfer) {
8686 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8687 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8688 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8691 return rte_flow_error_set
8693 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8694 NULL, "invalid default miss RSS");
8695 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8696 ctx->action_rss.level = 0,
8697 ctx->action_rss.types = priv->rss_conf.rss_hf,
8698 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8699 ctx->action_rss.queue_num = priv->reta_idx_n,
8700 ctx->action_rss.key = priv->rss_conf.rss_key,
8701 ctx->action_rss.queue = ctx->queue;
8702 if (!priv->reta_idx_n || !priv->rxqs_n)
8703 return rte_flow_error_set
8705 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8706 NULL, "invalid port configuration");
8707 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8708 ctx->action_rss.types = 0;
8709 for (i = 0; i != priv->reta_idx_n; ++i)
8710 ctx->queue[i] = (*priv->reta_idx)[i];
8712 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8713 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8715 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8716 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8717 jump_data = app_actions->conf;
8718 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8719 miss_attr.group = jump_data->group;
8720 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8721 &flow_table, &grp_info, error);
8723 return rte_flow_error_set(error, EINVAL,
8724 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8725 NULL, "invalid tunnel id");
8726 mark_id.app_reserve = 0;
8727 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8728 mark_id.transfer = !!attr->transfer;
8729 mark_id._unused_ = 0;
8730 miss_mark.id = mark_id.val;
8731 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8732 miss_items, miss_actions, flow_idx, error);
8735 dev_flow->flow = flow;
8736 dev_flow->external = true;
8737 dev_flow->tunnel = tunnel;
8738 dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8739 /* Subflow object was created, we must include one in the list. */
8740 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8741 dev_flow->handle, next);
8743 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8744 dev->data->port_id, tunnel->app_tunnel.type,
8745 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8746 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8747 miss_actions, error);
8749 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8755 static const struct mlx5_flow_tbl_data_entry *
8756 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8758 struct mlx5_priv *priv = dev->data->dev_private;
8759 struct mlx5_dev_ctx_shared *sh = priv->sh;
8760 struct mlx5_list_entry *he;
8761 union tunnel_offload_mark mbits = { .val = mark };
8762 union mlx5_flow_tbl_key table_key = {
8764 .level = tunnel_id_to_flow_tbl(mbits.table_id),
8768 .is_fdb = !!mbits.transfer,
8772 struct mlx5_flow_cb_ctx ctx = {
8773 .data = &table_key.v64,
8776 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
8778 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8782 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
8783 struct mlx5_list_entry *entry)
8785 struct mlx5_dev_ctx_shared *sh = tool_ctx;
8786 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8788 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8789 tunnel_flow_tbl_to_id(tte->flow_table));
8794 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
8795 struct mlx5_list_entry *entry, void *cb_ctx)
8797 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8798 union tunnel_tbl_key tbl = {
8799 .val = *(uint64_t *)(ctx->data),
8801 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8803 return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8806 static struct mlx5_list_entry *
8807 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
8809 struct mlx5_dev_ctx_shared *sh = tool_ctx;
8810 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8811 struct tunnel_tbl_entry *tte;
8812 union tunnel_tbl_key tbl = {
8813 .val = *(uint64_t *)(ctx->data),
8816 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8821 mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8823 if (tte->flow_table >= MLX5_MAX_TABLES) {
8824 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8826 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8829 } else if (!tte->flow_table) {
8832 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8833 tte->tunnel_id = tbl.tunnel_id;
8834 tte->group = tbl.group;
8842 static struct mlx5_list_entry *
8843 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
8844 struct mlx5_list_entry *oentry,
8845 void *cb_ctx __rte_unused)
8847 struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
8852 memcpy(tte, oentry, sizeof(*tte));
8857 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
8858 struct mlx5_list_entry *entry)
8860 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8866 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8867 const struct mlx5_flow_tunnel *tunnel,
8868 uint32_t group, uint32_t *table,
8869 struct rte_flow_error *error)
8871 struct mlx5_list_entry *he;
8872 struct tunnel_tbl_entry *tte;
8873 union tunnel_tbl_key key = {
8874 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8877 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8878 struct mlx5_hlist *group_hash;
8879 struct mlx5_flow_cb_ctx ctx = {
8883 group_hash = tunnel ? tunnel->groups : thub->groups;
8884 he = mlx5_hlist_register(group_hash, key.val, &ctx);
8886 return rte_flow_error_set(error, EINVAL,
8887 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8889 "tunnel group index not supported");
8890 tte = container_of(he, typeof(*tte), hash);
8891 *table = tte->flow_table;
8892 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8893 dev->data->port_id, key.tunnel_id, group, *table);
8898 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8899 struct mlx5_flow_tunnel *tunnel)
8901 struct mlx5_priv *priv = dev->data->dev_private;
8902 struct mlx5_indexed_pool *ipool;
8904 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8905 dev->data->port_id, tunnel->tunnel_id);
8906 LIST_REMOVE(tunnel, chain);
8907 mlx5_hlist_destroy(tunnel->groups);
8908 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8909 mlx5_ipool_free(ipool, tunnel->tunnel_id);
8913 mlx5_access_tunnel_offload_db
8914 (struct rte_eth_dev *dev,
8915 bool (*match)(struct rte_eth_dev *,
8916 struct mlx5_flow_tunnel *, const void *),
8917 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8918 void (*miss)(struct rte_eth_dev *, void *),
8919 void *ctx, bool lock_op)
8921 bool verdict = false;
8922 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8923 struct mlx5_flow_tunnel *tunnel;
8925 rte_spinlock_lock(&thub->sl);
8926 LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8927 verdict = match(dev, tunnel, (const void *)ctx);
8932 rte_spinlock_unlock(&thub->sl);
8934 hit(dev, tunnel, ctx);
8935 if (!verdict && miss)
8938 rte_spinlock_unlock(&thub->sl);
8943 struct tunnel_db_find_tunnel_id_ctx {
8945 struct mlx5_flow_tunnel *tunnel;
8949 find_tunnel_id_match(struct rte_eth_dev *dev,
8950 struct mlx5_flow_tunnel *tunnel, const void *x)
8952 const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8955 return tunnel->tunnel_id == ctx->tunnel_id;
8959 find_tunnel_id_hit(struct rte_eth_dev *dev,
8960 struct mlx5_flow_tunnel *tunnel, void *x)
8962 struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8964 ctx->tunnel = tunnel;
8967 static struct mlx5_flow_tunnel *
8968 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8970 struct tunnel_db_find_tunnel_id_ctx ctx = {
8974 mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8975 find_tunnel_id_hit, NULL, &ctx, true);
8980 static struct mlx5_flow_tunnel *
8981 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8982 const struct rte_flow_tunnel *app_tunnel)
8984 struct mlx5_priv *priv = dev->data->dev_private;
8985 struct mlx5_indexed_pool *ipool;
8986 struct mlx5_flow_tunnel *tunnel;
8989 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8990 tunnel = mlx5_ipool_zmalloc(ipool, &id);
8993 if (id >= MLX5_MAX_TUNNELS) {
8994 mlx5_ipool_free(ipool, id);
8995 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8998 tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9000 mlx5_flow_tunnel_grp2tbl_create_cb,
9001 mlx5_flow_tunnel_grp2tbl_match_cb,
9002 mlx5_flow_tunnel_grp2tbl_remove_cb,
9003 mlx5_flow_tunnel_grp2tbl_clone_cb,
9004 mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9005 if (!tunnel->groups) {
9006 mlx5_ipool_free(ipool, id);
9009 /* initiate new PMD tunnel */
9010 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9011 tunnel->tunnel_id = id;
9012 tunnel->action.type = (typeof(tunnel->action.type))
9013 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9014 tunnel->action.conf = tunnel;
9015 tunnel->item.type = (typeof(tunnel->item.type))
9016 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9017 tunnel->item.spec = tunnel;
9018 tunnel->item.last = NULL;
9019 tunnel->item.mask = NULL;
9021 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9022 dev->data->port_id, tunnel->tunnel_id);
9027 struct tunnel_db_get_tunnel_ctx {
9028 const struct rte_flow_tunnel *app_tunnel;
9029 struct mlx5_flow_tunnel *tunnel;
9032 static bool get_tunnel_match(struct rte_eth_dev *dev,
9033 struct mlx5_flow_tunnel *tunnel, const void *x)
9035 const struct tunnel_db_get_tunnel_ctx *ctx = x;
9038 return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9039 sizeof(*ctx->app_tunnel));
9042 static void get_tunnel_hit(struct rte_eth_dev *dev,
9043 struct mlx5_flow_tunnel *tunnel, void *x)
9045 /* called under tunnel spinlock protection */
9046 struct tunnel_db_get_tunnel_ctx *ctx = x;
9050 ctx->tunnel = tunnel;
9053 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9055 /* called under tunnel spinlock protection */
9056 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9057 struct tunnel_db_get_tunnel_ctx *ctx = x;
9059 rte_spinlock_unlock(&thub->sl);
9060 ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9061 rte_spinlock_lock(&thub->sl);
9063 ctx->tunnel->refctn = 1;
9064 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9070 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9071 const struct rte_flow_tunnel *app_tunnel,
9072 struct mlx5_flow_tunnel **tunnel)
9074 struct tunnel_db_get_tunnel_ctx ctx = {
9075 .app_tunnel = app_tunnel,
9078 mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9079 get_tunnel_miss, &ctx, true);
9080 *tunnel = ctx.tunnel;
9081 return ctx.tunnel ? 0 : -ENOMEM;
9084 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9086 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9090 if (!LIST_EMPTY(&thub->tunnels))
9091 DRV_LOG(WARNING, "port %u tunnels present", port_id);
9092 mlx5_hlist_destroy(thub->groups);
9096 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9099 struct mlx5_flow_tunnel_hub *thub;
9101 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9105 LIST_INIT(&thub->tunnels);
9106 rte_spinlock_init(&thub->sl);
9107 thub->groups = mlx5_hlist_create("flow groups", 64,
9109 mlx5_flow_tunnel_grp2tbl_create_cb,
9110 mlx5_flow_tunnel_grp2tbl_match_cb,
9111 mlx5_flow_tunnel_grp2tbl_remove_cb,
9112 mlx5_flow_tunnel_grp2tbl_clone_cb,
9113 mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9114 if (!thub->groups) {
9118 sh->tunnel_hub = thub;
9124 mlx5_hlist_destroy(thub->groups);
9131 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9132 struct rte_flow_tunnel *tunnel,
9133 const char *err_msg)
9136 if (!is_tunnel_offload_active(dev)) {
9137 err_msg = "tunnel offload was not activated";
9139 } else if (!tunnel) {
9140 err_msg = "no application tunnel";
9144 switch (tunnel->type) {
9146 err_msg = "unsupported tunnel type";
9148 case RTE_FLOW_ITEM_TYPE_VXLAN:
9149 case RTE_FLOW_ITEM_TYPE_GRE:
9150 case RTE_FLOW_ITEM_TYPE_NVGRE:
9151 case RTE_FLOW_ITEM_TYPE_GENEVE:
9160 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9161 struct rte_flow_tunnel *app_tunnel,
9162 struct rte_flow_action **actions,
9163 uint32_t *num_of_actions,
9164 struct rte_flow_error *error)
9167 struct mlx5_flow_tunnel *tunnel;
9168 const char *err_msg = NULL;
9169 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9172 return rte_flow_error_set(error, EINVAL,
9173 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9175 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9177 return rte_flow_error_set(error, ret,
9178 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9179 "failed to initialize pmd tunnel");
9181 *actions = &tunnel->action;
9182 *num_of_actions = 1;
9187 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9188 struct rte_flow_tunnel *app_tunnel,
9189 struct rte_flow_item **items,
9190 uint32_t *num_of_items,
9191 struct rte_flow_error *error)
9194 struct mlx5_flow_tunnel *tunnel;
9195 const char *err_msg = NULL;
9196 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9199 return rte_flow_error_set(error, EINVAL,
9200 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9202 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9204 return rte_flow_error_set(error, ret,
9205 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9206 "failed to initialize pmd tunnel");
9208 *items = &tunnel->item;
9213 struct tunnel_db_element_release_ctx {
9214 struct rte_flow_item *items;
9215 struct rte_flow_action *actions;
9216 uint32_t num_elements;
9217 struct rte_flow_error *error;
9222 tunnel_element_release_match(struct rte_eth_dev *dev,
9223 struct mlx5_flow_tunnel *tunnel, const void *x)
9225 const struct tunnel_db_element_release_ctx *ctx = x;
9228 if (ctx->num_elements != 1)
9230 else if (ctx->items)
9231 return ctx->items == &tunnel->item;
9232 else if (ctx->actions)
9233 return ctx->actions == &tunnel->action;
9239 tunnel_element_release_hit(struct rte_eth_dev *dev,
9240 struct mlx5_flow_tunnel *tunnel, void *x)
9242 struct tunnel_db_element_release_ctx *ctx = x;
9244 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9245 mlx5_flow_tunnel_free(dev, tunnel);
9249 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9251 struct tunnel_db_element_release_ctx *ctx = x;
9253 ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9254 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9255 "invalid argument");
9259 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9260 struct rte_flow_item *pmd_items,
9261 uint32_t num_items, struct rte_flow_error *err)
9263 struct tunnel_db_element_release_ctx ctx = {
9266 .num_elements = num_items,
9270 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9271 tunnel_element_release_hit,
9272 tunnel_element_release_miss, &ctx, false);
9278 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9279 struct rte_flow_action *pmd_actions,
9280 uint32_t num_actions, struct rte_flow_error *err)
9282 struct tunnel_db_element_release_ctx ctx = {
9284 .actions = pmd_actions,
9285 .num_elements = num_actions,
9289 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9290 tunnel_element_release_hit,
9291 tunnel_element_release_miss, &ctx, false);
9297 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9299 struct rte_flow_restore_info *info,
9300 struct rte_flow_error *err)
9302 uint64_t ol_flags = m->ol_flags;
9303 const struct mlx5_flow_tbl_data_entry *tble;
9304 const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
9306 if (!is_tunnel_offload_active(dev)) {
9311 if ((ol_flags & mask) != mask)
9313 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9315 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9316 dev->data->port_id, m->hash.fdir.hi);
9319 MLX5_ASSERT(tble->tunnel);
9320 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9321 info->group_id = tble->group_id;
9322 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9323 RTE_FLOW_RESTORE_INFO_GROUP_ID |
9324 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9329 return rte_flow_error_set(err, EINVAL,
9330 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9331 "failed to get restore info");
9334 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9336 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9337 __rte_unused struct rte_flow_tunnel *app_tunnel,
9338 __rte_unused struct rte_flow_action **actions,
9339 __rte_unused uint32_t *num_of_actions,
9340 __rte_unused struct rte_flow_error *error)
9346 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9347 __rte_unused struct rte_flow_tunnel *app_tunnel,
9348 __rte_unused struct rte_flow_item **items,
9349 __rte_unused uint32_t *num_of_items,
9350 __rte_unused struct rte_flow_error *error)
9356 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9357 __rte_unused struct rte_flow_item *pmd_items,
9358 __rte_unused uint32_t num_items,
9359 __rte_unused struct rte_flow_error *err)
9365 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9366 __rte_unused struct rte_flow_action *pmd_action,
9367 __rte_unused uint32_t num_actions,
9368 __rte_unused struct rte_flow_error *err)
9374 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9375 __rte_unused struct rte_mbuf *m,
9376 __rte_unused struct rte_flow_restore_info *i,
9377 __rte_unused struct rte_flow_error *err)
9383 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9384 __rte_unused struct rte_flow *flow,
9385 __rte_unused const struct rte_flow_attr *attr,
9386 __rte_unused const struct rte_flow_action *actions,
9387 __rte_unused uint32_t flow_idx,
9388 __rte_unused const struct mlx5_flow_tunnel *tunnel,
9389 __rte_unused struct tunnel_default_miss_ctx *ctx,
9390 __rte_unused struct rte_flow_error *error)
9395 static struct mlx5_flow_tunnel *
9396 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9397 __rte_unused uint32_t id)
9403 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9404 __rte_unused struct mlx5_flow_tunnel *tunnel)
9409 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9410 __rte_unused const struct mlx5_flow_tunnel *t,
9411 __rte_unused uint32_t group,
9412 __rte_unused uint32_t *table,
9413 struct rte_flow_error *error)
9415 return rte_flow_error_set(error, ENOTSUP,
9416 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9417 "tunnel offload requires DV support");
9421 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9422 __rte_unused uint16_t port_id)
9425 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9428 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9431 struct rte_flow_error error;
9433 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9435 ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9437 (void *)(uintptr_t)item->type, &error);
9439 printf("%s ", item_name);
9441 printf("%d\n", (int)item->type);
9447 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9449 const struct rte_flow_item_udp *spec = udp_item->spec;
9450 const struct rte_flow_item_udp *mask = udp_item->mask;
9451 uint16_t udp_dport = 0;
9455 mask = &rte_flow_item_udp_mask;
9456 udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9457 mask->hdr.dst_port);
9459 return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9462 static const struct mlx5_flow_expand_node *
9463 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9464 unsigned int item_idx,
9465 const struct mlx5_flow_expand_node graph[],
9466 const struct mlx5_flow_expand_node *node)
9468 const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9469 switch (item->type) {
9470 case RTE_FLOW_ITEM_TYPE_VXLAN:
9471 MLX5_ASSERT(item_idx > 0);
9472 prev_item = pattern + item_idx - 1;
9473 MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9474 if (mlx5_flow_is_std_vxlan_port(prev_item))
9475 return &graph[MLX5_EXPANSION_STD_VXLAN];
9477 return &graph[MLX5_EXPANSION_L3_VXLAN];