1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
10 #include <sys/queue.h>
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
25 #include <mlx5_malloc.h>
27 #include "mlx5_defs.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
36 struct tunnel_default_miss_ctx {
40 struct rte_flow_action_rss action_rss;
41 struct rte_flow_action_queue miss_queue;
42 struct rte_flow_action_jump miss_jump;
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 struct rte_flow *flow,
50 const struct rte_flow_attr *attr,
51 const struct rte_flow_action *app_actions,
53 const struct mlx5_flow_tunnel *tunnel,
54 struct tunnel_default_miss_ctx *ctx,
55 struct rte_flow_error *error);
56 static struct mlx5_flow_tunnel *
57 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
59 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
61 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
62 const struct mlx5_flow_tunnel *tunnel,
63 uint32_t group, uint32_t *table,
64 struct rte_flow_error *error);
66 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
67 static void mlx5_flow_pop_thread_workspace(void);
70 /** Device flow drivers. */
71 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
73 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
75 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
76 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
77 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
78 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
80 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
81 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
84 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
85 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
90 /** Node object of input graph for mlx5_flow_expand_rss(). */
91 struct mlx5_flow_expand_node {
92 const int *const next;
94 * List of next node indexes. Index 0 is interpreted as a terminator.
96 const enum rte_flow_item_type type;
97 /**< Pattern item type of current node. */
100 * RSS types bit-field associated with this node
101 * (see RTE_ETH_RSS_* definitions).
105 * Bit-fields that define how the node is used in the expansion.
106 * (see MLX5_EXPANSION_NODE_* definitions).
110 /* Optional expand field. The expansion alg will not go deeper. */
111 #define MLX5_EXPANSION_NODE_OPTIONAL (UINT64_C(1) << 0)
113 /* The node is not added implicitly as expansion to the flow pattern.
114 * If the node type does not match the flow pattern item type, the
115 * expansion alg will go deeper to its next items.
116 * In the current implementation, the list of next nodes indexes can
117 * have up to one node with this flag set and it has to be the last
118 * node index (before the list terminator).
120 #define MLX5_EXPANSION_NODE_EXPLICIT (UINT64_C(1) << 1)
122 /** Object returned by mlx5_flow_expand_rss(). */
123 struct mlx5_flow_expand_rss {
125 /**< Number of entries @p patterns and @p priorities. */
127 struct rte_flow_item *pattern; /**< Expanded pattern array. */
128 uint32_t priority; /**< Priority offset for each expansion. */
133 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
135 static const struct mlx5_flow_expand_node *
136 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
137 unsigned int item_idx,
138 const struct mlx5_flow_expand_node graph[],
139 const struct mlx5_flow_expand_node *node);
142 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
144 switch (item->type) {
145 case RTE_FLOW_ITEM_TYPE_ETH:
146 case RTE_FLOW_ITEM_TYPE_VLAN:
147 case RTE_FLOW_ITEM_TYPE_IPV4:
148 case RTE_FLOW_ITEM_TYPE_IPV6:
149 case RTE_FLOW_ITEM_TYPE_UDP:
150 case RTE_FLOW_ITEM_TYPE_TCP:
151 case RTE_FLOW_ITEM_TYPE_VXLAN:
152 case RTE_FLOW_ITEM_TYPE_NVGRE:
153 case RTE_FLOW_ITEM_TYPE_GRE:
154 case RTE_FLOW_ITEM_TYPE_GENEVE:
155 case RTE_FLOW_ITEM_TYPE_MPLS:
156 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
157 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
158 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
159 case RTE_FLOW_ITEM_TYPE_GTP:
167 static enum rte_flow_item_type
168 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
170 enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
171 uint16_t ether_type = 0;
172 uint16_t ether_type_m;
173 uint8_t ip_next_proto = 0;
174 uint8_t ip_next_proto_m;
176 if (item == NULL || item->spec == NULL)
178 switch (item->type) {
179 case RTE_FLOW_ITEM_TYPE_ETH:
181 ether_type_m = ((const struct rte_flow_item_eth *)
184 ether_type_m = rte_flow_item_eth_mask.type;
185 if (ether_type_m != RTE_BE16(0xFFFF))
187 ether_type = ((const struct rte_flow_item_eth *)
189 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
190 ret = RTE_FLOW_ITEM_TYPE_IPV4;
191 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
192 ret = RTE_FLOW_ITEM_TYPE_IPV6;
193 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
194 ret = RTE_FLOW_ITEM_TYPE_VLAN;
196 ret = RTE_FLOW_ITEM_TYPE_END;
198 case RTE_FLOW_ITEM_TYPE_VLAN:
200 ether_type_m = ((const struct rte_flow_item_vlan *)
201 (item->mask))->inner_type;
203 ether_type_m = rte_flow_item_vlan_mask.inner_type;
204 if (ether_type_m != RTE_BE16(0xFFFF))
206 ether_type = ((const struct rte_flow_item_vlan *)
207 (item->spec))->inner_type;
208 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
209 ret = RTE_FLOW_ITEM_TYPE_IPV4;
210 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
211 ret = RTE_FLOW_ITEM_TYPE_IPV6;
212 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
213 ret = RTE_FLOW_ITEM_TYPE_VLAN;
215 ret = RTE_FLOW_ITEM_TYPE_END;
217 case RTE_FLOW_ITEM_TYPE_IPV4:
219 ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
220 (item->mask))->hdr.next_proto_id;
223 rte_flow_item_ipv4_mask.hdr.next_proto_id;
224 if (ip_next_proto_m != 0xFF)
226 ip_next_proto = ((const struct rte_flow_item_ipv4 *)
227 (item->spec))->hdr.next_proto_id;
228 if (ip_next_proto == IPPROTO_UDP)
229 ret = RTE_FLOW_ITEM_TYPE_UDP;
230 else if (ip_next_proto == IPPROTO_TCP)
231 ret = RTE_FLOW_ITEM_TYPE_TCP;
232 else if (ip_next_proto == IPPROTO_IP)
233 ret = RTE_FLOW_ITEM_TYPE_IPV4;
234 else if (ip_next_proto == IPPROTO_IPV6)
235 ret = RTE_FLOW_ITEM_TYPE_IPV6;
237 ret = RTE_FLOW_ITEM_TYPE_END;
239 case RTE_FLOW_ITEM_TYPE_IPV6:
241 ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
242 (item->mask))->hdr.proto;
245 rte_flow_item_ipv6_mask.hdr.proto;
246 if (ip_next_proto_m != 0xFF)
248 ip_next_proto = ((const struct rte_flow_item_ipv6 *)
249 (item->spec))->hdr.proto;
250 if (ip_next_proto == IPPROTO_UDP)
251 ret = RTE_FLOW_ITEM_TYPE_UDP;
252 else if (ip_next_proto == IPPROTO_TCP)
253 ret = RTE_FLOW_ITEM_TYPE_TCP;
254 else if (ip_next_proto == IPPROTO_IP)
255 ret = RTE_FLOW_ITEM_TYPE_IPV4;
256 else if (ip_next_proto == IPPROTO_IPV6)
257 ret = RTE_FLOW_ITEM_TYPE_IPV6;
259 ret = RTE_FLOW_ITEM_TYPE_END;
262 ret = RTE_FLOW_ITEM_TYPE_VOID;
269 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
270 const int *next_node)
272 const struct mlx5_flow_expand_node *node = NULL;
273 const int *next = next_node;
275 while (next && *next) {
277 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
278 * flag set, because they were not found in the flow pattern.
280 node = &graph[*next];
281 if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
288 #define MLX5_RSS_EXP_ELT_N 16
291 * Expand RSS flows into several possible flows according to the RSS hash
292 * fields requested and the driver capabilities.
295 * Buffer to store the result expansion.
297 * Buffer size in bytes. If 0, @p buf can be NULL.
301 * RSS types to expand (see RTE_ETH_RSS_* definitions).
303 * Input graph to expand @p pattern according to @p types.
304 * @param[in] graph_root_index
305 * Index of root node in @p graph, typically 0.
308 * A positive value representing the size of @p buf in bytes regardless of
309 * @p size on success, a negative errno value otherwise and rte_errno is
310 * set, the following errors are defined:
312 * -E2BIG: graph-depth @p graph is too deep.
313 * -EINVAL: @p size has not enough space for expanded pattern.
316 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
317 const struct rte_flow_item *pattern, uint64_t types,
318 const struct mlx5_flow_expand_node graph[],
319 int graph_root_index)
321 const struct rte_flow_item *item;
322 const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
323 const int *next_node;
324 const int *stack[MLX5_RSS_EXP_ELT_N];
326 struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
327 unsigned int i, item_idx, last_expand_item_idx = 0;
329 size_t user_pattern_size = 0;
331 const struct mlx5_flow_expand_node *next = NULL;
332 struct rte_flow_item missed_item;
335 const struct rte_flow_item *last_expand_item = NULL;
337 memset(&missed_item, 0, sizeof(missed_item));
338 lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
339 MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
342 buf->entry[0].priority = 0;
343 buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
345 addr = buf->entry[0].pattern;
346 for (item = pattern, item_idx = 0;
347 item->type != RTE_FLOW_ITEM_TYPE_END;
348 item++, item_idx++) {
349 if (!mlx5_flow_is_rss_expandable_item(item)) {
350 user_pattern_size += sizeof(*item);
353 last_expand_item = item;
354 last_expand_item_idx = item_idx;
356 while (node->next && node->next[i]) {
357 next = &graph[node->next[i]];
358 if (next->type == item->type)
360 if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
369 user_pattern_size += sizeof(*item);
371 user_pattern_size += sizeof(*item); /* Handle END item. */
372 lsize += user_pattern_size;
375 /* Copy the user pattern in the first entry of the buffer. */
376 rte_memcpy(addr, pattern, user_pattern_size);
377 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
379 /* Start expanding. */
380 memset(flow_items, 0, sizeof(flow_items));
381 user_pattern_size -= sizeof(*item);
383 * Check if the last valid item has spec set, need complete pattern,
384 * and the pattern can be used for expansion.
386 missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
387 if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
388 /* Item type END indicates expansion is not required. */
391 if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
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 = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
564 RTE_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 = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
574 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
575 .type = RTE_FLOW_ITEM_TYPE_TCP,
576 .rss_types = RTE_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 = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
588 RTE_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 = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
598 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
599 .type = RTE_FLOW_ITEM_TYPE_TCP,
600 .rss_types = RTE_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 = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
663 RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
665 [MLX5_EXPANSION_IPV4_UDP] = {
666 .type = RTE_FLOW_ITEM_TYPE_UDP,
667 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
669 [MLX5_EXPANSION_IPV4_TCP] = {
670 .type = RTE_FLOW_ITEM_TYPE_TCP,
671 .rss_types = RTE_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 = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
679 RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
681 [MLX5_EXPANSION_IPV6_UDP] = {
682 .type = RTE_FLOW_ITEM_TYPE_UDP,
683 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
685 [MLX5_EXPANSION_IPV6_TCP] = {
686 .type = RTE_FLOW_ITEM_TYPE_TCP,
687 .rss_types = RTE_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 (priv->sh->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 && priv->sh->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 (priv->sh->flow_mreg_c
934 [id + 1 + start_reg - REG_C_0] != REG_NON)
935 return priv->sh->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 priv->sh->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;
964 * Having available reg_c can be regarded inclusively as supporting
965 * extensive flow metadata register, which could mean,
966 * - metadata register copy action by modify header.
967 * - 16 modify header actions is supported.
968 * - reg_c's are preserved across different domain (FDB and NIC) on
969 * packet loopback by flow lookup miss.
971 return priv->sh->flow_mreg_c[2] != REG_NON;
975 * Get the lowest priority.
978 * Pointer to the Ethernet device structure.
979 * @param[in] attributes
980 * Pointer to device flow rule attributes.
983 * The value of lowest priority of flow.
986 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
987 const struct rte_flow_attr *attr)
989 struct mlx5_priv *priv = dev->data->dev_private;
991 if (!attr->group && !attr->transfer)
992 return priv->sh->flow_max_priority - 2;
993 return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
997 * Calculate matcher priority of the flow.
1000 * Pointer to the Ethernet device structure.
1002 * Pointer to device flow rule attributes.
1003 * @param[in] subpriority
1004 * The priority based on the items.
1005 * @param[in] external
1006 * Flow is user flow.
1008 * The matcher priority of the flow.
1011 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1012 const struct rte_flow_attr *attr,
1013 uint32_t subpriority, bool external)
1015 uint16_t priority = (uint16_t)attr->priority;
1016 struct mlx5_priv *priv = dev->data->dev_private;
1018 if (!attr->group && !attr->transfer) {
1019 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1020 priority = priv->sh->flow_max_priority - 1;
1021 return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1022 } else if (!external && attr->transfer && attr->group == 0 &&
1023 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1024 return (priv->sh->flow_max_priority - 1) * 3;
1026 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1027 priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1028 return priority * 3 + subpriority;
1032 * Verify the @p item specifications (spec, last, mask) are compatible with the
1036 * Item specification.
1038 * @p item->mask or flow default bit-masks.
1039 * @param[in] nic_mask
1040 * Bit-masks covering supported fields by the NIC to compare with user mask.
1042 * Bit-masks size in bytes.
1043 * @param[in] range_accepted
1044 * True if range of values is accepted for specific fields, false otherwise.
1046 * Pointer to error structure.
1049 * 0 on success, a negative errno value otherwise and rte_errno is set.
1052 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1053 const uint8_t *mask,
1054 const uint8_t *nic_mask,
1056 bool range_accepted,
1057 struct rte_flow_error *error)
1061 MLX5_ASSERT(nic_mask);
1062 for (i = 0; i < size; ++i)
1063 if ((nic_mask[i] | mask[i]) != nic_mask[i])
1064 return rte_flow_error_set(error, ENOTSUP,
1065 RTE_FLOW_ERROR_TYPE_ITEM,
1067 "mask enables non supported"
1069 if (!item->spec && (item->mask || item->last))
1070 return rte_flow_error_set(error, EINVAL,
1071 RTE_FLOW_ERROR_TYPE_ITEM, item,
1072 "mask/last without a spec is not"
1074 if (item->spec && item->last && !range_accepted) {
1080 for (i = 0; i < size; ++i) {
1081 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1082 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1084 ret = memcmp(spec, last, size);
1086 return rte_flow_error_set(error, EINVAL,
1087 RTE_FLOW_ERROR_TYPE_ITEM,
1089 "range is not valid");
1095 * Adjust the hash fields according to the @p flow information.
1097 * @param[in] dev_flow.
1098 * Pointer to the mlx5_flow.
1100 * 1 when the hash field is for a tunnel item.
1101 * @param[in] layer_types
1102 * RTE_ETH_RSS_* types.
1103 * @param[in] hash_fields
1107 * The hash fields that should be used.
1110 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1111 int tunnel __rte_unused, uint64_t layer_types,
1112 uint64_t hash_fields)
1114 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1115 int rss_request_inner = rss_desc->level >= 2;
1117 /* Check RSS hash level for tunnel. */
1118 if (tunnel && rss_request_inner)
1119 hash_fields |= IBV_RX_HASH_INNER;
1120 else if (tunnel || rss_request_inner)
1123 /* Check if requested layer matches RSS hash fields. */
1124 if (!(rss_desc->types & layer_types))
1130 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
1131 * if several tunnel rules are used on this queue, the tunnel ptype will be
1135 * Rx queue to update.
1138 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1141 uint32_t tunnel_ptype = 0;
1143 /* Look up for the ptype to use. */
1144 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1145 if (!rxq_ctrl->flow_tunnels_n[i])
1147 if (!tunnel_ptype) {
1148 tunnel_ptype = tunnels_info[i].ptype;
1154 rxq_ctrl->rxq.tunnel = tunnel_ptype;
1158 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1162 * Pointer to the Ethernet device structure.
1163 * @param[in] dev_handle
1164 * Pointer to device flow handle structure.
1167 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1168 struct mlx5_flow_handle *dev_handle)
1170 struct mlx5_priv *priv = dev->data->dev_private;
1171 const int mark = dev_handle->mark;
1172 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1173 struct mlx5_ind_table_obj *ind_tbl = NULL;
1176 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1177 struct mlx5_hrxq *hrxq;
1179 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1180 dev_handle->rix_hrxq);
1182 ind_tbl = hrxq->ind_table;
1183 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1184 struct mlx5_shared_action_rss *shared_rss;
1186 shared_rss = mlx5_ipool_get
1187 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1188 dev_handle->rix_srss);
1190 ind_tbl = shared_rss->ind_tbl;
1194 for (i = 0; i != ind_tbl->queues_n; ++i) {
1195 int idx = ind_tbl->queues[i];
1196 struct mlx5_rxq_ctrl *rxq_ctrl =
1197 container_of((*priv->rxqs)[idx],
1198 struct mlx5_rxq_ctrl, rxq);
1201 * To support metadata register copy on Tx loopback,
1202 * this must be always enabled (metadata may arive
1203 * from other port - not from local flows only.
1205 if (priv->config.dv_flow_en &&
1206 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1207 mlx5_flow_ext_mreg_supported(dev)) {
1208 rxq_ctrl->rxq.mark = 1;
1209 rxq_ctrl->flow_mark_n = 1;
1211 rxq_ctrl->rxq.mark = 1;
1212 rxq_ctrl->flow_mark_n++;
1217 /* Increase the counter matching the flow. */
1218 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1219 if ((tunnels_info[j].tunnel &
1220 dev_handle->layers) ==
1221 tunnels_info[j].tunnel) {
1222 rxq_ctrl->flow_tunnels_n[j]++;
1226 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1232 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1235 * Pointer to the Ethernet device structure.
1237 * Pointer to flow structure.
1240 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1242 struct mlx5_priv *priv = dev->data->dev_private;
1243 uint32_t handle_idx;
1244 struct mlx5_flow_handle *dev_handle;
1246 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1247 handle_idx, dev_handle, next)
1248 flow_drv_rxq_flags_set(dev, dev_handle);
1252 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1253 * device flow if no other flow uses it with the same kind of request.
1256 * Pointer to Ethernet device.
1257 * @param[in] dev_handle
1258 * Pointer to the device flow handle structure.
1261 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1262 struct mlx5_flow_handle *dev_handle)
1264 struct mlx5_priv *priv = dev->data->dev_private;
1265 const int mark = dev_handle->mark;
1266 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1267 struct mlx5_ind_table_obj *ind_tbl = NULL;
1270 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1271 struct mlx5_hrxq *hrxq;
1273 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1274 dev_handle->rix_hrxq);
1276 ind_tbl = hrxq->ind_table;
1277 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1278 struct mlx5_shared_action_rss *shared_rss;
1280 shared_rss = mlx5_ipool_get
1281 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1282 dev_handle->rix_srss);
1284 ind_tbl = shared_rss->ind_tbl;
1288 MLX5_ASSERT(dev->data->dev_started);
1289 for (i = 0; i != ind_tbl->queues_n; ++i) {
1290 int idx = ind_tbl->queues[i];
1291 struct mlx5_rxq_ctrl *rxq_ctrl =
1292 container_of((*priv->rxqs)[idx],
1293 struct mlx5_rxq_ctrl, rxq);
1295 if (priv->config.dv_flow_en &&
1296 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1297 mlx5_flow_ext_mreg_supported(dev)) {
1298 rxq_ctrl->rxq.mark = 1;
1299 rxq_ctrl->flow_mark_n = 1;
1301 rxq_ctrl->flow_mark_n--;
1302 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1307 /* Decrease the counter matching the flow. */
1308 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1309 if ((tunnels_info[j].tunnel &
1310 dev_handle->layers) ==
1311 tunnels_info[j].tunnel) {
1312 rxq_ctrl->flow_tunnels_n[j]--;
1316 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1322 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1323 * @p flow if no other flow uses it with the same kind of request.
1326 * Pointer to Ethernet device.
1328 * Pointer to the flow.
1331 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1333 struct mlx5_priv *priv = dev->data->dev_private;
1334 uint32_t handle_idx;
1335 struct mlx5_flow_handle *dev_handle;
1337 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1338 handle_idx, dev_handle, next)
1339 flow_drv_rxq_flags_trim(dev, dev_handle);
1343 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1346 * Pointer to Ethernet device.
1349 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1351 struct mlx5_priv *priv = dev->data->dev_private;
1354 for (i = 0; i != priv->rxqs_n; ++i) {
1355 struct mlx5_rxq_ctrl *rxq_ctrl;
1358 if (!(*priv->rxqs)[i])
1360 rxq_ctrl = container_of((*priv->rxqs)[i],
1361 struct mlx5_rxq_ctrl, rxq);
1362 rxq_ctrl->flow_mark_n = 0;
1363 rxq_ctrl->rxq.mark = 0;
1364 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1365 rxq_ctrl->flow_tunnels_n[j] = 0;
1366 rxq_ctrl->rxq.tunnel = 0;
1371 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1374 * Pointer to the Ethernet device structure.
1377 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1379 struct mlx5_priv *priv = dev->data->dev_private;
1380 struct mlx5_rxq_data *data;
1383 for (i = 0; i != priv->rxqs_n; ++i) {
1384 if (!(*priv->rxqs)[i])
1386 data = (*priv->rxqs)[i];
1387 if (!rte_flow_dynf_metadata_avail()) {
1388 data->dynf_meta = 0;
1389 data->flow_meta_mask = 0;
1390 data->flow_meta_offset = -1;
1391 data->flow_meta_port_mask = 0;
1393 data->dynf_meta = 1;
1394 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1395 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1396 data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1402 * return a pointer to the desired action in the list of actions.
1404 * @param[in] actions
1405 * The list of actions to search the action in.
1407 * The action to find.
1410 * Pointer to the action in the list, if found. NULL otherwise.
1412 const struct rte_flow_action *
1413 mlx5_flow_find_action(const struct rte_flow_action *actions,
1414 enum rte_flow_action_type action)
1416 if (actions == NULL)
1418 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1419 if (actions->type == action)
1425 * Validate the flag action.
1427 * @param[in] action_flags
1428 * Bit-fields that holds the actions detected until now.
1430 * Attributes of flow that includes this action.
1432 * Pointer to error structure.
1435 * 0 on success, a negative errno value otherwise and rte_errno is set.
1438 mlx5_flow_validate_action_flag(uint64_t action_flags,
1439 const struct rte_flow_attr *attr,
1440 struct rte_flow_error *error)
1442 if (action_flags & MLX5_FLOW_ACTION_MARK)
1443 return rte_flow_error_set(error, EINVAL,
1444 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1445 "can't mark and flag in same flow");
1446 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1447 return rte_flow_error_set(error, EINVAL,
1448 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1450 " actions in same flow");
1452 return rte_flow_error_set(error, ENOTSUP,
1453 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1454 "flag action not supported for "
1460 * Validate the mark action.
1463 * Pointer to the queue action.
1464 * @param[in] action_flags
1465 * Bit-fields that holds the actions detected until now.
1467 * Attributes of flow that includes this action.
1469 * Pointer to error structure.
1472 * 0 on success, a negative errno value otherwise and rte_errno is set.
1475 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1476 uint64_t action_flags,
1477 const struct rte_flow_attr *attr,
1478 struct rte_flow_error *error)
1480 const struct rte_flow_action_mark *mark = action->conf;
1483 return rte_flow_error_set(error, EINVAL,
1484 RTE_FLOW_ERROR_TYPE_ACTION,
1486 "configuration cannot be null");
1487 if (mark->id >= MLX5_FLOW_MARK_MAX)
1488 return rte_flow_error_set(error, EINVAL,
1489 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1491 "mark id must in 0 <= id < "
1492 RTE_STR(MLX5_FLOW_MARK_MAX));
1493 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1494 return rte_flow_error_set(error, EINVAL,
1495 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1496 "can't flag and mark in same flow");
1497 if (action_flags & MLX5_FLOW_ACTION_MARK)
1498 return rte_flow_error_set(error, EINVAL,
1499 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1500 "can't have 2 mark actions in same"
1503 return rte_flow_error_set(error, ENOTSUP,
1504 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1505 "mark action not supported for "
1511 * Validate the drop action.
1513 * @param[in] action_flags
1514 * Bit-fields that holds the actions detected until now.
1516 * Attributes of flow that includes this action.
1518 * Pointer to error structure.
1521 * 0 on success, a negative errno value otherwise and rte_errno is set.
1524 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1525 const struct rte_flow_attr *attr,
1526 struct rte_flow_error *error)
1529 return rte_flow_error_set(error, ENOTSUP,
1530 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1531 "drop action not supported for "
1537 * Validate the queue action.
1540 * Pointer to the queue action.
1541 * @param[in] action_flags
1542 * Bit-fields that holds the actions detected until now.
1544 * Pointer to the Ethernet device structure.
1546 * Attributes of flow that includes this action.
1548 * Pointer to error structure.
1551 * 0 on success, a negative errno value otherwise and rte_errno is set.
1554 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1555 uint64_t action_flags,
1556 struct rte_eth_dev *dev,
1557 const struct rte_flow_attr *attr,
1558 struct rte_flow_error *error)
1560 struct mlx5_priv *priv = dev->data->dev_private;
1561 const struct rte_flow_action_queue *queue = action->conf;
1563 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1564 return rte_flow_error_set(error, EINVAL,
1565 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1566 "can't have 2 fate actions in"
1569 return rte_flow_error_set(error, EINVAL,
1570 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1571 NULL, "No Rx queues configured");
1572 if (queue->index >= priv->rxqs_n)
1573 return rte_flow_error_set(error, EINVAL,
1574 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1576 "queue index out of range");
1577 if (!(*priv->rxqs)[queue->index])
1578 return rte_flow_error_set(error, EINVAL,
1579 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1581 "queue is not configured");
1583 return rte_flow_error_set(error, ENOTSUP,
1584 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1585 "queue action not supported for "
1591 * Validate the rss action.
1594 * Pointer to the Ethernet device structure.
1596 * Pointer to the queue action.
1598 * Pointer to error structure.
1601 * 0 on success, a negative errno value otherwise and rte_errno is set.
1604 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1605 const struct rte_flow_action *action,
1606 struct rte_flow_error *error)
1608 struct mlx5_priv *priv = dev->data->dev_private;
1609 const struct rte_flow_action_rss *rss = action->conf;
1610 enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1613 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1614 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1615 return rte_flow_error_set(error, ENOTSUP,
1616 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1618 "RSS hash function not supported");
1619 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1624 return rte_flow_error_set(error, ENOTSUP,
1625 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1627 "tunnel RSS is not supported");
1628 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1629 if (rss->key_len == 0 && rss->key != NULL)
1630 return rte_flow_error_set(error, ENOTSUP,
1631 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1633 "RSS hash key length 0");
1634 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1635 return rte_flow_error_set(error, ENOTSUP,
1636 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1638 "RSS hash key too small");
1639 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1640 return rte_flow_error_set(error, ENOTSUP,
1641 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1643 "RSS hash key too large");
1644 if (rss->queue_num > priv->config.ind_table_max_size)
1645 return rte_flow_error_set(error, ENOTSUP,
1646 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1648 "number of queues too large");
1649 if (rss->types & MLX5_RSS_HF_MASK)
1650 return rte_flow_error_set(error, ENOTSUP,
1651 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1653 "some RSS protocols are not"
1655 if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1656 !(rss->types & RTE_ETH_RSS_IP))
1657 return rte_flow_error_set(error, EINVAL,
1658 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1659 "L3 partial RSS requested but L3 RSS"
1660 " type not specified");
1661 if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1662 !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1663 return rte_flow_error_set(error, EINVAL,
1664 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1665 "L4 partial RSS requested but L4 RSS"
1666 " type not specified");
1668 return rte_flow_error_set(error, EINVAL,
1669 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1670 NULL, "No Rx queues configured");
1671 if (!rss->queue_num)
1672 return rte_flow_error_set(error, EINVAL,
1673 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1674 NULL, "No queues configured");
1675 for (i = 0; i != rss->queue_num; ++i) {
1676 struct mlx5_rxq_ctrl *rxq_ctrl;
1678 if (rss->queue[i] >= priv->rxqs_n)
1679 return rte_flow_error_set
1681 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1682 &rss->queue[i], "queue index out of range");
1683 if (!(*priv->rxqs)[rss->queue[i]])
1684 return rte_flow_error_set
1685 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1686 &rss->queue[i], "queue is not configured");
1687 rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1688 struct mlx5_rxq_ctrl, rxq);
1690 rxq_type = rxq_ctrl->type;
1691 if (rxq_type != rxq_ctrl->type)
1692 return rte_flow_error_set
1693 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1695 "combining hairpin and regular RSS queues is not supported");
1701 * Validate the rss action.
1704 * Pointer to the queue action.
1705 * @param[in] action_flags
1706 * Bit-fields that holds the actions detected until now.
1708 * Pointer to the Ethernet device structure.
1710 * Attributes of flow that includes this action.
1711 * @param[in] item_flags
1712 * Items that were detected.
1714 * Pointer to error structure.
1717 * 0 on success, a negative errno value otherwise and rte_errno is set.
1720 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1721 uint64_t action_flags,
1722 struct rte_eth_dev *dev,
1723 const struct rte_flow_attr *attr,
1724 uint64_t item_flags,
1725 struct rte_flow_error *error)
1727 const struct rte_flow_action_rss *rss = action->conf;
1728 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1731 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1732 return rte_flow_error_set(error, EINVAL,
1733 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1734 "can't have 2 fate actions"
1736 ret = mlx5_validate_action_rss(dev, action, error);
1740 return rte_flow_error_set(error, ENOTSUP,
1741 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1742 "rss action not supported for "
1744 if (rss->level > 1 && !tunnel)
1745 return rte_flow_error_set(error, EINVAL,
1746 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1747 "inner RSS is not supported for "
1748 "non-tunnel flows");
1749 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1750 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1751 return rte_flow_error_set(error, EINVAL,
1752 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1753 "RSS on eCPRI is not supported now");
1755 if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1757 (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1759 return rte_flow_error_set(error, EINVAL,
1760 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1761 "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1766 * Validate the default miss action.
1768 * @param[in] action_flags
1769 * Bit-fields that holds the actions detected until now.
1771 * Pointer to error structure.
1774 * 0 on success, a negative errno value otherwise and rte_errno is set.
1777 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1778 const struct rte_flow_attr *attr,
1779 struct rte_flow_error *error)
1781 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1782 return rte_flow_error_set(error, EINVAL,
1783 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1784 "can't have 2 fate actions in"
1787 return rte_flow_error_set(error, ENOTSUP,
1788 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1789 "default miss action not supported "
1792 return rte_flow_error_set(error, ENOTSUP,
1793 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1794 "only group 0 is supported");
1796 return rte_flow_error_set(error, ENOTSUP,
1797 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1798 NULL, "transfer is not supported");
1803 * Validate the count action.
1806 * Pointer to the Ethernet device structure.
1808 * Attributes of flow that includes this action.
1810 * Pointer to error structure.
1813 * 0 on success, a negative errno value otherwise and rte_errno is set.
1816 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1817 const struct rte_flow_attr *attr,
1818 struct rte_flow_error *error)
1821 return rte_flow_error_set(error, ENOTSUP,
1822 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1823 "count action not supported for "
1829 * Validate the ASO CT action.
1832 * Pointer to the Ethernet device structure.
1833 * @param[in] conntrack
1834 * Pointer to the CT action profile.
1836 * Pointer to error structure.
1839 * 0 on success, a negative errno value otherwise and rte_errno is set.
1842 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1843 const struct rte_flow_action_conntrack *conntrack,
1844 struct rte_flow_error *error)
1848 if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1849 return rte_flow_error_set(error, EINVAL,
1850 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1851 "Invalid CT state");
1852 if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1853 return rte_flow_error_set(error, EINVAL,
1854 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1855 "Invalid last TCP packet flag");
1860 * Verify the @p attributes will be correctly understood by the NIC and store
1861 * them in the @p flow if everything is correct.
1864 * Pointer to the Ethernet device structure.
1865 * @param[in] attributes
1866 * Pointer to flow attributes
1868 * Pointer to error structure.
1871 * 0 on success, a negative errno value otherwise and rte_errno is set.
1874 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1875 const struct rte_flow_attr *attributes,
1876 struct rte_flow_error *error)
1878 struct mlx5_priv *priv = dev->data->dev_private;
1879 uint32_t priority_max = priv->sh->flow_max_priority - 1;
1881 if (attributes->group)
1882 return rte_flow_error_set(error, ENOTSUP,
1883 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1884 NULL, "groups is not supported");
1885 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1886 attributes->priority >= priority_max)
1887 return rte_flow_error_set(error, ENOTSUP,
1888 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1889 NULL, "priority out of range");
1890 if (attributes->egress)
1891 return rte_flow_error_set(error, ENOTSUP,
1892 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1893 "egress is not supported");
1894 if (attributes->transfer && !priv->config.dv_esw_en)
1895 return rte_flow_error_set(error, ENOTSUP,
1896 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1897 NULL, "transfer is not supported");
1898 if (!attributes->ingress)
1899 return rte_flow_error_set(error, EINVAL,
1900 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1902 "ingress attribute is mandatory");
1907 * Validate ICMP6 item.
1910 * Item specification.
1911 * @param[in] item_flags
1912 * Bit-fields that holds the items detected until now.
1913 * @param[in] ext_vlan_sup
1914 * Whether extended VLAN features are supported or not.
1916 * Pointer to error structure.
1919 * 0 on success, a negative errno value otherwise and rte_errno is set.
1922 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1923 uint64_t item_flags,
1924 uint8_t target_protocol,
1925 struct rte_flow_error *error)
1927 const struct rte_flow_item_icmp6 *mask = item->mask;
1928 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1929 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1930 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1931 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1932 MLX5_FLOW_LAYER_OUTER_L4;
1935 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1936 return rte_flow_error_set(error, EINVAL,
1937 RTE_FLOW_ERROR_TYPE_ITEM, item,
1938 "protocol filtering not compatible"
1939 " with ICMP6 layer");
1940 if (!(item_flags & l3m))
1941 return rte_flow_error_set(error, EINVAL,
1942 RTE_FLOW_ERROR_TYPE_ITEM, item,
1943 "IPv6 is mandatory to filter on"
1945 if (item_flags & l4m)
1946 return rte_flow_error_set(error, EINVAL,
1947 RTE_FLOW_ERROR_TYPE_ITEM, item,
1948 "multiple L4 layers not supported");
1950 mask = &rte_flow_item_icmp6_mask;
1951 ret = mlx5_flow_item_acceptable
1952 (item, (const uint8_t *)mask,
1953 (const uint8_t *)&rte_flow_item_icmp6_mask,
1954 sizeof(struct rte_flow_item_icmp6),
1955 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1962 * Validate ICMP item.
1965 * Item specification.
1966 * @param[in] item_flags
1967 * Bit-fields that holds the items detected until now.
1969 * Pointer to error structure.
1972 * 0 on success, a negative errno value otherwise and rte_errno is set.
1975 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1976 uint64_t item_flags,
1977 uint8_t target_protocol,
1978 struct rte_flow_error *error)
1980 const struct rte_flow_item_icmp *mask = item->mask;
1981 const struct rte_flow_item_icmp nic_mask = {
1982 .hdr.icmp_type = 0xff,
1983 .hdr.icmp_code = 0xff,
1984 .hdr.icmp_ident = RTE_BE16(0xffff),
1985 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1987 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1988 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1989 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1990 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1991 MLX5_FLOW_LAYER_OUTER_L4;
1994 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1995 return rte_flow_error_set(error, EINVAL,
1996 RTE_FLOW_ERROR_TYPE_ITEM, item,
1997 "protocol filtering not compatible"
1998 " with ICMP layer");
1999 if (!(item_flags & l3m))
2000 return rte_flow_error_set(error, EINVAL,
2001 RTE_FLOW_ERROR_TYPE_ITEM, item,
2002 "IPv4 is mandatory to filter"
2004 if (item_flags & l4m)
2005 return rte_flow_error_set(error, EINVAL,
2006 RTE_FLOW_ERROR_TYPE_ITEM, item,
2007 "multiple L4 layers not supported");
2010 ret = mlx5_flow_item_acceptable
2011 (item, (const uint8_t *)mask,
2012 (const uint8_t *)&nic_mask,
2013 sizeof(struct rte_flow_item_icmp),
2014 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2021 * Validate Ethernet item.
2024 * Item specification.
2025 * @param[in] item_flags
2026 * Bit-fields that holds the items detected until now.
2028 * Pointer to error structure.
2031 * 0 on success, a negative errno value otherwise and rte_errno is set.
2034 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2035 uint64_t item_flags, bool ext_vlan_sup,
2036 struct rte_flow_error *error)
2038 const struct rte_flow_item_eth *mask = item->mask;
2039 const struct rte_flow_item_eth nic_mask = {
2040 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2041 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2042 .type = RTE_BE16(0xffff),
2043 .has_vlan = ext_vlan_sup ? 1 : 0,
2046 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2047 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
2048 MLX5_FLOW_LAYER_OUTER_L2;
2050 if (item_flags & ethm)
2051 return rte_flow_error_set(error, ENOTSUP,
2052 RTE_FLOW_ERROR_TYPE_ITEM, item,
2053 "multiple L2 layers not supported");
2054 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2055 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2056 return rte_flow_error_set(error, EINVAL,
2057 RTE_FLOW_ERROR_TYPE_ITEM, item,
2058 "L2 layer should not follow "
2060 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2061 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2062 return rte_flow_error_set(error, EINVAL,
2063 RTE_FLOW_ERROR_TYPE_ITEM, item,
2064 "L2 layer should not follow VLAN");
2065 if (item_flags & MLX5_FLOW_LAYER_GTP)
2066 return rte_flow_error_set(error, EINVAL,
2067 RTE_FLOW_ERROR_TYPE_ITEM, item,
2068 "L2 layer should not follow GTP");
2070 mask = &rte_flow_item_eth_mask;
2071 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2072 (const uint8_t *)&nic_mask,
2073 sizeof(struct rte_flow_item_eth),
2074 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2079 * Validate VLAN item.
2082 * Item specification.
2083 * @param[in] item_flags
2084 * Bit-fields that holds the items detected until now.
2086 * Ethernet device flow is being created on.
2088 * Pointer to error structure.
2091 * 0 on success, a negative errno value otherwise and rte_errno is set.
2094 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2095 uint64_t item_flags,
2096 struct rte_eth_dev *dev,
2097 struct rte_flow_error *error)
2099 const struct rte_flow_item_vlan *spec = item->spec;
2100 const struct rte_flow_item_vlan *mask = item->mask;
2101 const struct rte_flow_item_vlan nic_mask = {
2102 .tci = RTE_BE16(UINT16_MAX),
2103 .inner_type = RTE_BE16(UINT16_MAX),
2105 uint16_t vlan_tag = 0;
2106 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2108 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2109 MLX5_FLOW_LAYER_INNER_L4) :
2110 (MLX5_FLOW_LAYER_OUTER_L3 |
2111 MLX5_FLOW_LAYER_OUTER_L4);
2112 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2113 MLX5_FLOW_LAYER_OUTER_VLAN;
2115 if (item_flags & vlanm)
2116 return rte_flow_error_set(error, EINVAL,
2117 RTE_FLOW_ERROR_TYPE_ITEM, item,
2118 "multiple VLAN layers not supported");
2119 else if ((item_flags & l34m) != 0)
2120 return rte_flow_error_set(error, EINVAL,
2121 RTE_FLOW_ERROR_TYPE_ITEM, item,
2122 "VLAN cannot follow L3/L4 layer");
2124 mask = &rte_flow_item_vlan_mask;
2125 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2126 (const uint8_t *)&nic_mask,
2127 sizeof(struct rte_flow_item_vlan),
2128 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2131 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2132 struct mlx5_priv *priv = dev->data->dev_private;
2134 if (priv->vmwa_context) {
2136 * Non-NULL context means we have a virtual machine
2137 * and SR-IOV enabled, we have to create VLAN interface
2138 * to make hypervisor to setup E-Switch vport
2139 * context correctly. We avoid creating the multiple
2140 * VLAN interfaces, so we cannot support VLAN tag mask.
2142 return rte_flow_error_set(error, EINVAL,
2143 RTE_FLOW_ERROR_TYPE_ITEM,
2145 "VLAN tag mask is not"
2146 " supported in virtual"
2151 vlan_tag = spec->tci;
2152 vlan_tag &= mask->tci;
2155 * From verbs perspective an empty VLAN is equivalent
2156 * to a packet without VLAN layer.
2159 return rte_flow_error_set(error, EINVAL,
2160 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2162 "VLAN cannot be empty");
2167 * Validate IPV4 item.
2170 * Item specification.
2171 * @param[in] item_flags
2172 * Bit-fields that holds the items detected until now.
2173 * @param[in] last_item
2174 * Previous validated item in the pattern items.
2175 * @param[in] ether_type
2176 * Type in the ethernet layer header (including dot1q).
2177 * @param[in] acc_mask
2178 * Acceptable mask, if NULL default internal default mask
2179 * will be used to check whether item fields are supported.
2180 * @param[in] range_accepted
2181 * True if range of values is accepted for specific fields, false otherwise.
2183 * Pointer to error structure.
2186 * 0 on success, a negative errno value otherwise and rte_errno is set.
2189 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2190 uint64_t item_flags,
2192 uint16_t ether_type,
2193 const struct rte_flow_item_ipv4 *acc_mask,
2194 bool range_accepted,
2195 struct rte_flow_error *error)
2197 const struct rte_flow_item_ipv4 *mask = item->mask;
2198 const struct rte_flow_item_ipv4 *spec = item->spec;
2199 const struct rte_flow_item_ipv4 nic_mask = {
2201 .src_addr = RTE_BE32(0xffffffff),
2202 .dst_addr = RTE_BE32(0xffffffff),
2203 .type_of_service = 0xff,
2204 .next_proto_id = 0xff,
2207 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2208 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2209 MLX5_FLOW_LAYER_OUTER_L3;
2210 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2211 MLX5_FLOW_LAYER_OUTER_L4;
2213 uint8_t next_proto = 0xFF;
2214 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2215 MLX5_FLOW_LAYER_OUTER_VLAN |
2216 MLX5_FLOW_LAYER_INNER_VLAN);
2218 if ((last_item & l2_vlan) && ether_type &&
2219 ether_type != RTE_ETHER_TYPE_IPV4)
2220 return rte_flow_error_set(error, EINVAL,
2221 RTE_FLOW_ERROR_TYPE_ITEM, item,
2222 "IPv4 cannot follow L2/VLAN layer "
2223 "which ether type is not IPv4");
2224 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2226 next_proto = mask->hdr.next_proto_id &
2227 spec->hdr.next_proto_id;
2228 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2229 return rte_flow_error_set(error, EINVAL,
2230 RTE_FLOW_ERROR_TYPE_ITEM,
2235 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2236 return rte_flow_error_set(error, EINVAL,
2237 RTE_FLOW_ERROR_TYPE_ITEM, item,
2238 "wrong tunnel type - IPv6 specified "
2239 "but IPv4 item provided");
2240 if (item_flags & l3m)
2241 return rte_flow_error_set(error, ENOTSUP,
2242 RTE_FLOW_ERROR_TYPE_ITEM, item,
2243 "multiple L3 layers not supported");
2244 else if (item_flags & l4m)
2245 return rte_flow_error_set(error, EINVAL,
2246 RTE_FLOW_ERROR_TYPE_ITEM, item,
2247 "L3 cannot follow an L4 layer.");
2248 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2249 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2250 return rte_flow_error_set(error, EINVAL,
2251 RTE_FLOW_ERROR_TYPE_ITEM, item,
2252 "L3 cannot follow an NVGRE layer.");
2254 mask = &rte_flow_item_ipv4_mask;
2255 else if (mask->hdr.next_proto_id != 0 &&
2256 mask->hdr.next_proto_id != 0xff)
2257 return rte_flow_error_set(error, EINVAL,
2258 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2259 "partial mask is not supported"
2261 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2262 acc_mask ? (const uint8_t *)acc_mask
2263 : (const uint8_t *)&nic_mask,
2264 sizeof(struct rte_flow_item_ipv4),
2265 range_accepted, error);
2272 * Validate IPV6 item.
2275 * Item specification.
2276 * @param[in] item_flags
2277 * Bit-fields that holds the items detected until now.
2278 * @param[in] last_item
2279 * Previous validated item in the pattern items.
2280 * @param[in] ether_type
2281 * Type in the ethernet layer header (including dot1q).
2282 * @param[in] acc_mask
2283 * Acceptable mask, if NULL default internal default mask
2284 * will be used to check whether item fields are supported.
2286 * Pointer to error structure.
2289 * 0 on success, a negative errno value otherwise and rte_errno is set.
2292 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2293 uint64_t item_flags,
2295 uint16_t ether_type,
2296 const struct rte_flow_item_ipv6 *acc_mask,
2297 struct rte_flow_error *error)
2299 const struct rte_flow_item_ipv6 *mask = item->mask;
2300 const struct rte_flow_item_ipv6 *spec = item->spec;
2301 const struct rte_flow_item_ipv6 nic_mask = {
2304 "\xff\xff\xff\xff\xff\xff\xff\xff"
2305 "\xff\xff\xff\xff\xff\xff\xff\xff",
2307 "\xff\xff\xff\xff\xff\xff\xff\xff"
2308 "\xff\xff\xff\xff\xff\xff\xff\xff",
2309 .vtc_flow = RTE_BE32(0xffffffff),
2313 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2314 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2315 MLX5_FLOW_LAYER_OUTER_L3;
2316 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2317 MLX5_FLOW_LAYER_OUTER_L4;
2319 uint8_t next_proto = 0xFF;
2320 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2321 MLX5_FLOW_LAYER_OUTER_VLAN |
2322 MLX5_FLOW_LAYER_INNER_VLAN);
2324 if ((last_item & l2_vlan) && ether_type &&
2325 ether_type != RTE_ETHER_TYPE_IPV6)
2326 return rte_flow_error_set(error, EINVAL,
2327 RTE_FLOW_ERROR_TYPE_ITEM, item,
2328 "IPv6 cannot follow L2/VLAN layer "
2329 "which ether type is not IPv6");
2330 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2331 next_proto = spec->hdr.proto;
2332 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2333 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2334 return rte_flow_error_set(error, EINVAL,
2335 RTE_FLOW_ERROR_TYPE_ITEM,
2340 if (next_proto == IPPROTO_HOPOPTS ||
2341 next_proto == IPPROTO_ROUTING ||
2342 next_proto == IPPROTO_FRAGMENT ||
2343 next_proto == IPPROTO_ESP ||
2344 next_proto == IPPROTO_AH ||
2345 next_proto == IPPROTO_DSTOPTS)
2346 return rte_flow_error_set(error, EINVAL,
2347 RTE_FLOW_ERROR_TYPE_ITEM, item,
2348 "IPv6 proto (next header) should "
2349 "not be set as extension header");
2350 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2351 return rte_flow_error_set(error, EINVAL,
2352 RTE_FLOW_ERROR_TYPE_ITEM, item,
2353 "wrong tunnel type - IPv4 specified "
2354 "but IPv6 item provided");
2355 if (item_flags & l3m)
2356 return rte_flow_error_set(error, ENOTSUP,
2357 RTE_FLOW_ERROR_TYPE_ITEM, item,
2358 "multiple L3 layers not supported");
2359 else if (item_flags & l4m)
2360 return rte_flow_error_set(error, EINVAL,
2361 RTE_FLOW_ERROR_TYPE_ITEM, item,
2362 "L3 cannot follow an L4 layer.");
2363 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2364 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2365 return rte_flow_error_set(error, EINVAL,
2366 RTE_FLOW_ERROR_TYPE_ITEM, item,
2367 "L3 cannot follow an NVGRE layer.");
2369 mask = &rte_flow_item_ipv6_mask;
2370 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2371 acc_mask ? (const uint8_t *)acc_mask
2372 : (const uint8_t *)&nic_mask,
2373 sizeof(struct rte_flow_item_ipv6),
2374 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2381 * Validate UDP item.
2384 * Item specification.
2385 * @param[in] item_flags
2386 * Bit-fields that holds the items detected until now.
2387 * @param[in] target_protocol
2388 * The next protocol in the previous item.
2389 * @param[in] flow_mask
2390 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2392 * Pointer to error structure.
2395 * 0 on success, a negative errno value otherwise and rte_errno is set.
2398 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2399 uint64_t item_flags,
2400 uint8_t target_protocol,
2401 struct rte_flow_error *error)
2403 const struct rte_flow_item_udp *mask = item->mask;
2404 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2405 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2406 MLX5_FLOW_LAYER_OUTER_L3;
2407 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2408 MLX5_FLOW_LAYER_OUTER_L4;
2411 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2412 return rte_flow_error_set(error, EINVAL,
2413 RTE_FLOW_ERROR_TYPE_ITEM, item,
2414 "protocol filtering not compatible"
2416 if (!(item_flags & l3m))
2417 return rte_flow_error_set(error, EINVAL,
2418 RTE_FLOW_ERROR_TYPE_ITEM, item,
2419 "L3 is mandatory to filter on L4");
2420 if (item_flags & l4m)
2421 return rte_flow_error_set(error, EINVAL,
2422 RTE_FLOW_ERROR_TYPE_ITEM, item,
2423 "multiple L4 layers not supported");
2425 mask = &rte_flow_item_udp_mask;
2426 ret = mlx5_flow_item_acceptable
2427 (item, (const uint8_t *)mask,
2428 (const uint8_t *)&rte_flow_item_udp_mask,
2429 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2437 * Validate TCP item.
2440 * Item specification.
2441 * @param[in] item_flags
2442 * Bit-fields that holds the items detected until now.
2443 * @param[in] target_protocol
2444 * The next protocol in the previous item.
2446 * Pointer to error structure.
2449 * 0 on success, a negative errno value otherwise and rte_errno is set.
2452 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2453 uint64_t item_flags,
2454 uint8_t target_protocol,
2455 const struct rte_flow_item_tcp *flow_mask,
2456 struct rte_flow_error *error)
2458 const struct rte_flow_item_tcp *mask = item->mask;
2459 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2460 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2461 MLX5_FLOW_LAYER_OUTER_L3;
2462 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2463 MLX5_FLOW_LAYER_OUTER_L4;
2466 MLX5_ASSERT(flow_mask);
2467 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2468 return rte_flow_error_set(error, EINVAL,
2469 RTE_FLOW_ERROR_TYPE_ITEM, item,
2470 "protocol filtering not compatible"
2472 if (!(item_flags & l3m))
2473 return rte_flow_error_set(error, EINVAL,
2474 RTE_FLOW_ERROR_TYPE_ITEM, item,
2475 "L3 is mandatory to filter on L4");
2476 if (item_flags & l4m)
2477 return rte_flow_error_set(error, EINVAL,
2478 RTE_FLOW_ERROR_TYPE_ITEM, item,
2479 "multiple L4 layers not supported");
2481 mask = &rte_flow_item_tcp_mask;
2482 ret = mlx5_flow_item_acceptable
2483 (item, (const uint8_t *)mask,
2484 (const uint8_t *)flow_mask,
2485 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2493 * Validate VXLAN item.
2496 * Pointer to the Ethernet device structure.
2497 * @param[in] udp_dport
2498 * UDP destination port
2500 * Item specification.
2501 * @param[in] item_flags
2502 * Bit-fields that holds the items detected until now.
2504 * Flow rule attributes.
2506 * Pointer to error structure.
2509 * 0 on success, a negative errno value otherwise and rte_errno is set.
2512 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2514 const struct rte_flow_item *item,
2515 uint64_t item_flags,
2516 const struct rte_flow_attr *attr,
2517 struct rte_flow_error *error)
2519 const struct rte_flow_item_vxlan *spec = item->spec;
2520 const struct rte_flow_item_vxlan *mask = item->mask;
2522 struct mlx5_priv *priv = dev->data->dev_private;
2526 } id = { .vlan_id = 0, };
2527 const struct rte_flow_item_vxlan nic_mask = {
2528 .vni = "\xff\xff\xff",
2531 const struct rte_flow_item_vxlan *valid_mask;
2533 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2534 return rte_flow_error_set(error, ENOTSUP,
2535 RTE_FLOW_ERROR_TYPE_ITEM, item,
2536 "multiple tunnel layers not"
2538 valid_mask = &rte_flow_item_vxlan_mask;
2540 * Verify only UDPv4 is present as defined in
2541 * https://tools.ietf.org/html/rfc7348
2543 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2544 return rte_flow_error_set(error, EINVAL,
2545 RTE_FLOW_ERROR_TYPE_ITEM, item,
2546 "no outer UDP layer found");
2548 mask = &rte_flow_item_vxlan_mask;
2550 if (priv->sh->steering_format_version !=
2551 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2552 !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2553 /* FDB domain & NIC domain non-zero group */
2554 if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2555 valid_mask = &nic_mask;
2556 /* Group zero in NIC domain */
2557 if (!attr->group && !attr->transfer &&
2558 priv->sh->tunnel_header_0_1)
2559 valid_mask = &nic_mask;
2561 ret = mlx5_flow_item_acceptable
2562 (item, (const uint8_t *)mask,
2563 (const uint8_t *)valid_mask,
2564 sizeof(struct rte_flow_item_vxlan),
2565 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2569 memcpy(&id.vni[1], spec->vni, 3);
2570 memcpy(&id.vni[1], mask->vni, 3);
2572 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2573 return rte_flow_error_set(error, ENOTSUP,
2574 RTE_FLOW_ERROR_TYPE_ITEM, item,
2575 "VXLAN tunnel must be fully defined");
2580 * Validate VXLAN_GPE item.
2583 * Item specification.
2584 * @param[in] item_flags
2585 * Bit-fields that holds the items detected until now.
2587 * Pointer to the private data structure.
2588 * @param[in] target_protocol
2589 * The next protocol in the previous item.
2591 * Pointer to error structure.
2594 * 0 on success, a negative errno value otherwise and rte_errno is set.
2597 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2598 uint64_t item_flags,
2599 struct rte_eth_dev *dev,
2600 struct rte_flow_error *error)
2602 struct mlx5_priv *priv = dev->data->dev_private;
2603 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2604 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2609 } id = { .vlan_id = 0, };
2611 if (!priv->config.l3_vxlan_en)
2612 return rte_flow_error_set(error, ENOTSUP,
2613 RTE_FLOW_ERROR_TYPE_ITEM, item,
2614 "L3 VXLAN is not enabled by device"
2615 " parameter and/or not configured in"
2617 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2618 return rte_flow_error_set(error, ENOTSUP,
2619 RTE_FLOW_ERROR_TYPE_ITEM, item,
2620 "multiple tunnel layers not"
2623 * Verify only UDPv4 is present as defined in
2624 * https://tools.ietf.org/html/rfc7348
2626 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2627 return rte_flow_error_set(error, EINVAL,
2628 RTE_FLOW_ERROR_TYPE_ITEM, item,
2629 "no outer UDP layer found");
2631 mask = &rte_flow_item_vxlan_gpe_mask;
2632 ret = mlx5_flow_item_acceptable
2633 (item, (const uint8_t *)mask,
2634 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2635 sizeof(struct rte_flow_item_vxlan_gpe),
2636 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2641 return rte_flow_error_set(error, ENOTSUP,
2642 RTE_FLOW_ERROR_TYPE_ITEM,
2644 "VxLAN-GPE protocol"
2646 memcpy(&id.vni[1], spec->vni, 3);
2647 memcpy(&id.vni[1], mask->vni, 3);
2649 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2650 return rte_flow_error_set(error, ENOTSUP,
2651 RTE_FLOW_ERROR_TYPE_ITEM, item,
2652 "VXLAN-GPE tunnel must be fully"
2657 * Validate GRE Key item.
2660 * Item specification.
2661 * @param[in] item_flags
2662 * Bit flags to mark detected items.
2663 * @param[in] gre_item
2664 * Pointer to gre_item
2666 * Pointer to error structure.
2669 * 0 on success, a negative errno value otherwise and rte_errno is set.
2672 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2673 uint64_t item_flags,
2674 const struct rte_flow_item *gre_item,
2675 struct rte_flow_error *error)
2677 const rte_be32_t *mask = item->mask;
2679 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2680 const struct rte_flow_item_gre *gre_spec;
2681 const struct rte_flow_item_gre *gre_mask;
2683 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2684 return rte_flow_error_set(error, ENOTSUP,
2685 RTE_FLOW_ERROR_TYPE_ITEM, item,
2686 "Multiple GRE key not support");
2687 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2688 return rte_flow_error_set(error, ENOTSUP,
2689 RTE_FLOW_ERROR_TYPE_ITEM, item,
2690 "No preceding GRE header");
2691 if (item_flags & MLX5_FLOW_LAYER_INNER)
2692 return rte_flow_error_set(error, ENOTSUP,
2693 RTE_FLOW_ERROR_TYPE_ITEM, item,
2694 "GRE key following a wrong item");
2695 gre_mask = gre_item->mask;
2697 gre_mask = &rte_flow_item_gre_mask;
2698 gre_spec = gre_item->spec;
2699 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2700 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2701 return rte_flow_error_set(error, EINVAL,
2702 RTE_FLOW_ERROR_TYPE_ITEM, item,
2703 "Key bit must be on");
2706 mask = &gre_key_default_mask;
2707 ret = mlx5_flow_item_acceptable
2708 (item, (const uint8_t *)mask,
2709 (const uint8_t *)&gre_key_default_mask,
2710 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2715 * Validate GRE item.
2718 * Item specification.
2719 * @param[in] item_flags
2720 * Bit flags to mark detected items.
2721 * @param[in] target_protocol
2722 * The next protocol in the previous item.
2724 * Pointer to error structure.
2727 * 0 on success, a negative errno value otherwise and rte_errno is set.
2730 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2731 uint64_t item_flags,
2732 uint8_t target_protocol,
2733 struct rte_flow_error *error)
2735 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2736 const struct rte_flow_item_gre *mask = item->mask;
2738 const struct rte_flow_item_gre nic_mask = {
2739 .c_rsvd0_ver = RTE_BE16(0xB000),
2740 .protocol = RTE_BE16(UINT16_MAX),
2743 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2744 return rte_flow_error_set(error, EINVAL,
2745 RTE_FLOW_ERROR_TYPE_ITEM, item,
2746 "protocol filtering not compatible"
2747 " with this GRE layer");
2748 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2749 return rte_flow_error_set(error, ENOTSUP,
2750 RTE_FLOW_ERROR_TYPE_ITEM, item,
2751 "multiple tunnel layers not"
2753 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2754 return rte_flow_error_set(error, ENOTSUP,
2755 RTE_FLOW_ERROR_TYPE_ITEM, item,
2756 "L3 Layer is missing");
2758 mask = &rte_flow_item_gre_mask;
2759 ret = mlx5_flow_item_acceptable
2760 (item, (const uint8_t *)mask,
2761 (const uint8_t *)&nic_mask,
2762 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2766 #ifndef HAVE_MLX5DV_DR
2767 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2768 if (spec && (spec->protocol & mask->protocol))
2769 return rte_flow_error_set(error, ENOTSUP,
2770 RTE_FLOW_ERROR_TYPE_ITEM, item,
2771 "without MPLS support the"
2772 " specification cannot be used for"
2780 * Validate Geneve item.
2783 * Item specification.
2784 * @param[in] itemFlags
2785 * Bit-fields that holds the items detected until now.
2787 * Pointer to the private data structure.
2789 * Pointer to error structure.
2792 * 0 on success, a negative errno value otherwise and rte_errno is set.
2796 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2797 uint64_t item_flags,
2798 struct rte_eth_dev *dev,
2799 struct rte_flow_error *error)
2801 struct mlx5_priv *priv = dev->data->dev_private;
2802 const struct rte_flow_item_geneve *spec = item->spec;
2803 const struct rte_flow_item_geneve *mask = item->mask;
2806 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2807 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2808 const struct rte_flow_item_geneve nic_mask = {
2809 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2810 .vni = "\xff\xff\xff",
2811 .protocol = RTE_BE16(UINT16_MAX),
2814 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2815 return rte_flow_error_set(error, ENOTSUP,
2816 RTE_FLOW_ERROR_TYPE_ITEM, item,
2817 "L3 Geneve is not enabled by device"
2818 " parameter and/or not configured in"
2820 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2821 return rte_flow_error_set(error, ENOTSUP,
2822 RTE_FLOW_ERROR_TYPE_ITEM, item,
2823 "multiple tunnel layers not"
2826 * Verify only UDPv4 is present as defined in
2827 * https://tools.ietf.org/html/rfc7348
2829 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2830 return rte_flow_error_set(error, EINVAL,
2831 RTE_FLOW_ERROR_TYPE_ITEM, item,
2832 "no outer UDP layer found");
2834 mask = &rte_flow_item_geneve_mask;
2835 ret = mlx5_flow_item_acceptable
2836 (item, (const uint8_t *)mask,
2837 (const uint8_t *)&nic_mask,
2838 sizeof(struct rte_flow_item_geneve),
2839 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2843 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2844 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2845 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2846 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2847 return rte_flow_error_set(error, ENOTSUP,
2848 RTE_FLOW_ERROR_TYPE_ITEM,
2850 "Geneve protocol unsupported"
2851 " fields are being used");
2852 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2853 return rte_flow_error_set
2855 RTE_FLOW_ERROR_TYPE_ITEM,
2857 "Unsupported Geneve options length");
2859 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2860 return rte_flow_error_set
2862 RTE_FLOW_ERROR_TYPE_ITEM, item,
2863 "Geneve tunnel must be fully defined");
2868 * Validate Geneve TLV option item.
2871 * Item specification.
2872 * @param[in] last_item
2873 * Previous validated item in the pattern items.
2874 * @param[in] geneve_item
2875 * Previous GENEVE item specification.
2877 * Pointer to the rte_eth_dev structure.
2879 * Pointer to error structure.
2882 * 0 on success, a negative errno value otherwise and rte_errno is set.
2885 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2887 const struct rte_flow_item *geneve_item,
2888 struct rte_eth_dev *dev,
2889 struct rte_flow_error *error)
2891 struct mlx5_priv *priv = dev->data->dev_private;
2892 struct mlx5_dev_ctx_shared *sh = priv->sh;
2893 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2894 struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2895 uint8_t data_max_supported =
2896 hca_attr->max_geneve_tlv_option_data_len * 4;
2897 struct mlx5_dev_config *config = &priv->config;
2898 const struct rte_flow_item_geneve *geneve_spec;
2899 const struct rte_flow_item_geneve *geneve_mask;
2900 const struct rte_flow_item_geneve_opt *spec = item->spec;
2901 const struct rte_flow_item_geneve_opt *mask = item->mask;
2903 unsigned int data_len;
2904 uint8_t tlv_option_len;
2905 uint16_t optlen_m, optlen_v;
2906 const struct rte_flow_item_geneve_opt full_mask = {
2907 .option_class = RTE_BE16(0xffff),
2908 .option_type = 0xff,
2913 mask = &rte_flow_item_geneve_opt_mask;
2915 return rte_flow_error_set
2916 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2917 "Geneve TLV opt class/type/length must be specified");
2918 if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2919 return rte_flow_error_set
2920 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2921 "Geneve TLV opt length exceeeds the limit (31)");
2922 /* Check if class type and length masks are full. */
2923 if (full_mask.option_class != mask->option_class ||
2924 full_mask.option_type != mask->option_type ||
2925 full_mask.option_len != (mask->option_len & full_mask.option_len))
2926 return rte_flow_error_set
2927 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2928 "Geneve TLV opt class/type/length masks must be full");
2929 /* Check if length is supported */
2930 if ((uint32_t)spec->option_len >
2931 config->hca_attr.max_geneve_tlv_option_data_len)
2932 return rte_flow_error_set
2933 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2934 "Geneve TLV opt length not supported");
2935 if (config->hca_attr.max_geneve_tlv_options > 1)
2937 "max_geneve_tlv_options supports more than 1 option");
2938 /* Check GENEVE item preceding. */
2939 if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2940 return rte_flow_error_set
2941 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2942 "Geneve opt item must be preceded with Geneve item");
2943 geneve_spec = geneve_item->spec;
2944 geneve_mask = geneve_item->mask ? geneve_item->mask :
2945 &rte_flow_item_geneve_mask;
2946 /* Check if GENEVE TLV option size doesn't exceed option length */
2947 if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2948 geneve_spec->ver_opt_len_o_c_rsvd0)) {
2949 tlv_option_len = spec->option_len & mask->option_len;
2950 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2951 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2952 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2953 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2954 if ((optlen_v & optlen_m) <= tlv_option_len)
2955 return rte_flow_error_set
2956 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2957 "GENEVE TLV option length exceeds optlen");
2959 /* Check if length is 0 or data is 0. */
2960 if (spec->data == NULL || spec->option_len == 0)
2961 return rte_flow_error_set
2962 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2963 "Geneve TLV opt with zero data/length not supported");
2964 /* Check not all data & mask are 0. */
2965 data_len = spec->option_len * 4;
2966 if (mask->data == NULL) {
2967 for (i = 0; i < data_len; i++)
2971 return rte_flow_error_set(error, ENOTSUP,
2972 RTE_FLOW_ERROR_TYPE_ITEM, item,
2973 "Can't match on Geneve option data 0");
2975 for (i = 0; i < data_len; i++)
2976 if (spec->data[i] & mask->data[i])
2979 return rte_flow_error_set(error, ENOTSUP,
2980 RTE_FLOW_ERROR_TYPE_ITEM, item,
2981 "Can't match on Geneve option data and mask 0");
2982 /* Check data mask supported. */
2983 for (i = data_max_supported; i < data_len ; i++)
2985 return rte_flow_error_set(error, ENOTSUP,
2986 RTE_FLOW_ERROR_TYPE_ITEM, item,
2987 "Data mask is of unsupported size");
2989 /* Check GENEVE option is supported in NIC. */
2990 if (!config->hca_attr.geneve_tlv_opt)
2991 return rte_flow_error_set
2992 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2993 "Geneve TLV opt not supported");
2994 /* Check if we already have geneve option with different type/class. */
2995 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2996 geneve_opt_resource = sh->geneve_tlv_option_resource;
2997 if (geneve_opt_resource != NULL)
2998 if (geneve_opt_resource->option_class != spec->option_class ||
2999 geneve_opt_resource->option_type != spec->option_type ||
3000 geneve_opt_resource->length != spec->option_len) {
3001 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3002 return rte_flow_error_set(error, ENOTSUP,
3003 RTE_FLOW_ERROR_TYPE_ITEM, item,
3004 "Only one Geneve TLV option supported");
3006 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3011 * Validate MPLS item.
3014 * Pointer to the rte_eth_dev structure.
3016 * Item specification.
3017 * @param[in] item_flags
3018 * Bit-fields that holds the items detected until now.
3019 * @param[in] prev_layer
3020 * The protocol layer indicated in previous item.
3022 * Pointer to error structure.
3025 * 0 on success, a negative errno value otherwise and rte_errno is set.
3028 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3029 const struct rte_flow_item *item __rte_unused,
3030 uint64_t item_flags __rte_unused,
3031 uint64_t prev_layer __rte_unused,
3032 struct rte_flow_error *error)
3034 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3035 const struct rte_flow_item_mpls *mask = item->mask;
3036 struct mlx5_priv *priv = dev->data->dev_private;
3039 if (!priv->config.mpls_en)
3040 return rte_flow_error_set(error, ENOTSUP,
3041 RTE_FLOW_ERROR_TYPE_ITEM, item,
3042 "MPLS not supported or"
3043 " disabled in firmware"
3045 /* MPLS over UDP, GRE is allowed */
3046 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3047 MLX5_FLOW_LAYER_GRE |
3048 MLX5_FLOW_LAYER_GRE_KEY)))
3049 return rte_flow_error_set(error, EINVAL,
3050 RTE_FLOW_ERROR_TYPE_ITEM, item,
3051 "protocol filtering not compatible"
3052 " with MPLS layer");
3053 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3054 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3055 !(item_flags & MLX5_FLOW_LAYER_GRE))
3056 return rte_flow_error_set(error, ENOTSUP,
3057 RTE_FLOW_ERROR_TYPE_ITEM, item,
3058 "multiple tunnel layers not"
3061 mask = &rte_flow_item_mpls_mask;
3062 ret = mlx5_flow_item_acceptable
3063 (item, (const uint8_t *)mask,
3064 (const uint8_t *)&rte_flow_item_mpls_mask,
3065 sizeof(struct rte_flow_item_mpls),
3066 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3071 return rte_flow_error_set(error, ENOTSUP,
3072 RTE_FLOW_ERROR_TYPE_ITEM, item,
3073 "MPLS is not supported by Verbs, please"
3079 * Validate NVGRE item.
3082 * Item specification.
3083 * @param[in] item_flags
3084 * Bit flags to mark detected items.
3085 * @param[in] target_protocol
3086 * The next protocol in the previous item.
3088 * Pointer to error structure.
3091 * 0 on success, a negative errno value otherwise and rte_errno is set.
3094 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3095 uint64_t item_flags,
3096 uint8_t target_protocol,
3097 struct rte_flow_error *error)
3099 const struct rte_flow_item_nvgre *mask = item->mask;
3102 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3103 return rte_flow_error_set(error, EINVAL,
3104 RTE_FLOW_ERROR_TYPE_ITEM, item,
3105 "protocol filtering not compatible"
3106 " with this GRE layer");
3107 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3108 return rte_flow_error_set(error, ENOTSUP,
3109 RTE_FLOW_ERROR_TYPE_ITEM, item,
3110 "multiple tunnel layers not"
3112 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3113 return rte_flow_error_set(error, ENOTSUP,
3114 RTE_FLOW_ERROR_TYPE_ITEM, item,
3115 "L3 Layer is missing");
3117 mask = &rte_flow_item_nvgre_mask;
3118 ret = mlx5_flow_item_acceptable
3119 (item, (const uint8_t *)mask,
3120 (const uint8_t *)&rte_flow_item_nvgre_mask,
3121 sizeof(struct rte_flow_item_nvgre),
3122 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3129 * Validate eCPRI item.
3132 * Item specification.
3133 * @param[in] item_flags
3134 * Bit-fields that holds the items detected until now.
3135 * @param[in] last_item
3136 * Previous validated item in the pattern items.
3137 * @param[in] ether_type
3138 * Type in the ethernet layer header (including dot1q).
3139 * @param[in] acc_mask
3140 * Acceptable mask, if NULL default internal default mask
3141 * will be used to check whether item fields are supported.
3143 * Pointer to error structure.
3146 * 0 on success, a negative errno value otherwise and rte_errno is set.
3149 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3150 uint64_t item_flags,
3152 uint16_t ether_type,
3153 const struct rte_flow_item_ecpri *acc_mask,
3154 struct rte_flow_error *error)
3156 const struct rte_flow_item_ecpri *mask = item->mask;
3157 const struct rte_flow_item_ecpri nic_mask = {
3161 RTE_BE32(((const struct rte_ecpri_common_hdr) {
3165 .dummy[0] = 0xFFFFFFFF,
3168 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3169 MLX5_FLOW_LAYER_OUTER_VLAN);
3170 struct rte_flow_item_ecpri mask_lo;
3172 if (!(last_item & outer_l2_vlan) &&
3173 last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3174 return rte_flow_error_set(error, EINVAL,
3175 RTE_FLOW_ERROR_TYPE_ITEM, item,
3176 "eCPRI can only follow L2/VLAN layer or UDP layer");
3177 if ((last_item & outer_l2_vlan) && ether_type &&
3178 ether_type != RTE_ETHER_TYPE_ECPRI)
3179 return rte_flow_error_set(error, EINVAL,
3180 RTE_FLOW_ERROR_TYPE_ITEM, item,
3181 "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3182 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3183 return rte_flow_error_set(error, EINVAL,
3184 RTE_FLOW_ERROR_TYPE_ITEM, item,
3185 "eCPRI with tunnel is not supported right now");
3186 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3187 return rte_flow_error_set(error, ENOTSUP,
3188 RTE_FLOW_ERROR_TYPE_ITEM, item,
3189 "multiple L3 layers not supported");
3190 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3191 return rte_flow_error_set(error, EINVAL,
3192 RTE_FLOW_ERROR_TYPE_ITEM, item,
3193 "eCPRI cannot coexist with a TCP layer");
3194 /* In specification, eCPRI could be over UDP layer. */
3195 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3196 return rte_flow_error_set(error, EINVAL,
3197 RTE_FLOW_ERROR_TYPE_ITEM, item,
3198 "eCPRI over UDP layer is not yet supported right now");
3199 /* Mask for type field in common header could be zero. */
3201 mask = &rte_flow_item_ecpri_mask;
3202 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3203 /* Input mask is in big-endian format. */
3204 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3205 return rte_flow_error_set(error, EINVAL,
3206 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3207 "partial mask is not supported for protocol");
3208 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3209 return rte_flow_error_set(error, EINVAL,
3210 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3211 "message header mask must be after a type mask");
3212 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3213 acc_mask ? (const uint8_t *)acc_mask
3214 : (const uint8_t *)&nic_mask,
3215 sizeof(struct rte_flow_item_ecpri),
3216 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3220 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3221 const struct rte_flow_attr *attr __rte_unused,
3222 const struct rte_flow_item items[] __rte_unused,
3223 const struct rte_flow_action actions[] __rte_unused,
3224 bool external __rte_unused,
3225 int hairpin __rte_unused,
3226 struct rte_flow_error *error)
3228 return rte_flow_error_set(error, ENOTSUP,
3229 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3232 static struct mlx5_flow *
3233 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3234 const struct rte_flow_attr *attr __rte_unused,
3235 const struct rte_flow_item items[] __rte_unused,
3236 const struct rte_flow_action actions[] __rte_unused,
3237 struct rte_flow_error *error)
3239 rte_flow_error_set(error, ENOTSUP,
3240 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3245 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3246 struct mlx5_flow *dev_flow __rte_unused,
3247 const struct rte_flow_attr *attr __rte_unused,
3248 const struct rte_flow_item items[] __rte_unused,
3249 const struct rte_flow_action actions[] __rte_unused,
3250 struct rte_flow_error *error)
3252 return rte_flow_error_set(error, ENOTSUP,
3253 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3257 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3258 struct rte_flow *flow __rte_unused,
3259 struct rte_flow_error *error)
3261 return rte_flow_error_set(error, ENOTSUP,
3262 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3266 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3267 struct rte_flow *flow __rte_unused)
3272 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3273 struct rte_flow *flow __rte_unused)
3278 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3279 struct rte_flow *flow __rte_unused,
3280 const struct rte_flow_action *actions __rte_unused,
3281 void *data __rte_unused,
3282 struct rte_flow_error *error)
3284 return rte_flow_error_set(error, ENOTSUP,
3285 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3289 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3290 uint32_t domains __rte_unused,
3291 uint32_t flags __rte_unused)
3296 /* Void driver to protect from null pointer reference. */
3297 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3298 .validate = flow_null_validate,
3299 .prepare = flow_null_prepare,
3300 .translate = flow_null_translate,
3301 .apply = flow_null_apply,
3302 .remove = flow_null_remove,
3303 .destroy = flow_null_destroy,
3304 .query = flow_null_query,
3305 .sync_domain = flow_null_sync_domain,
3309 * Select flow driver type according to flow attributes and device
3313 * Pointer to the dev structure.
3315 * Pointer to the flow attributes.
3318 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3320 static enum mlx5_flow_drv_type
3321 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3323 struct mlx5_priv *priv = dev->data->dev_private;
3324 /* The OS can determine first a specific flow type (DV, VERBS) */
3325 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3327 if (type != MLX5_FLOW_TYPE_MAX)
3329 /* If no OS specific type - continue with DV/VERBS selection */
3330 if (attr->transfer && priv->config.dv_esw_en)
3331 type = MLX5_FLOW_TYPE_DV;
3332 if (!attr->transfer)
3333 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3334 MLX5_FLOW_TYPE_VERBS;
3338 #define flow_get_drv_ops(type) flow_drv_ops[type]
3341 * Flow driver validation API. This abstracts calling driver specific functions.
3342 * The type of flow driver is determined according to flow attributes.
3345 * Pointer to the dev structure.
3347 * Pointer to the flow attributes.
3349 * Pointer to the list of items.
3350 * @param[in] actions
3351 * Pointer to the list of actions.
3352 * @param[in] external
3353 * This flow rule is created by request external to PMD.
3354 * @param[in] hairpin
3355 * Number of hairpin TX actions, 0 means classic flow.
3357 * Pointer to the error structure.
3360 * 0 on success, a negative errno value otherwise and rte_errno is set.
3363 flow_drv_validate(struct rte_eth_dev *dev,
3364 const struct rte_flow_attr *attr,
3365 const struct rte_flow_item items[],
3366 const struct rte_flow_action actions[],
3367 bool external, int hairpin, struct rte_flow_error *error)
3369 const struct mlx5_flow_driver_ops *fops;
3370 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3372 fops = flow_get_drv_ops(type);
3373 return fops->validate(dev, attr, items, actions, external,
3378 * Flow driver preparation API. This abstracts calling driver specific
3379 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3380 * calculates the size of memory required for device flow, allocates the memory,
3381 * initializes the device flow and returns the pointer.
3384 * This function initializes device flow structure such as dv or verbs in
3385 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3386 * rest. For example, adding returning device flow to flow->dev_flow list and
3387 * setting backward reference to the flow should be done out of this function.
3388 * layers field is not filled either.
3391 * Pointer to the dev structure.
3393 * Pointer to the flow attributes.
3395 * Pointer to the list of items.
3396 * @param[in] actions
3397 * Pointer to the list of actions.
3398 * @param[in] flow_idx
3399 * This memory pool index to the flow.
3401 * Pointer to the error structure.
3404 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3406 static inline struct mlx5_flow *
3407 flow_drv_prepare(struct rte_eth_dev *dev,
3408 const struct rte_flow *flow,
3409 const struct rte_flow_attr *attr,
3410 const struct rte_flow_item items[],
3411 const struct rte_flow_action actions[],
3413 struct rte_flow_error *error)
3415 const struct mlx5_flow_driver_ops *fops;
3416 enum mlx5_flow_drv_type type = flow->drv_type;
3417 struct mlx5_flow *mlx5_flow = NULL;
3419 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3420 fops = flow_get_drv_ops(type);
3421 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3423 mlx5_flow->flow_idx = flow_idx;
3428 * Flow driver translation API. This abstracts calling driver specific
3429 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3430 * translates a generic flow into a driver flow. flow_drv_prepare() must
3434 * dev_flow->layers could be filled as a result of parsing during translation
3435 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3436 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3437 * flow->actions could be overwritten even though all the expanded dev_flows
3438 * have the same actions.
3441 * Pointer to the rte dev structure.
3442 * @param[in, out] dev_flow
3443 * Pointer to the mlx5 flow.
3445 * Pointer to the flow attributes.
3447 * Pointer to the list of items.
3448 * @param[in] actions
3449 * Pointer to the list of actions.
3451 * Pointer to the error structure.
3454 * 0 on success, a negative errno value otherwise and rte_errno is set.
3457 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3458 const struct rte_flow_attr *attr,
3459 const struct rte_flow_item items[],
3460 const struct rte_flow_action actions[],
3461 struct rte_flow_error *error)
3463 const struct mlx5_flow_driver_ops *fops;
3464 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3466 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3467 fops = flow_get_drv_ops(type);
3468 return fops->translate(dev, dev_flow, attr, items, actions, error);
3472 * Flow driver apply API. This abstracts calling driver specific functions.
3473 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3474 * translated driver flows on to device. flow_drv_translate() must precede.
3477 * Pointer to Ethernet device structure.
3478 * @param[in, out] flow
3479 * Pointer to flow structure.
3481 * Pointer to error structure.
3484 * 0 on success, a negative errno value otherwise and rte_errno is set.
3487 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3488 struct rte_flow_error *error)
3490 const struct mlx5_flow_driver_ops *fops;
3491 enum mlx5_flow_drv_type type = flow->drv_type;
3493 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3494 fops = flow_get_drv_ops(type);
3495 return fops->apply(dev, flow, error);
3499 * Flow driver destroy API. This abstracts calling driver specific functions.
3500 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3501 * on device and releases resources of the flow.
3504 * Pointer to Ethernet device.
3505 * @param[in, out] flow
3506 * Pointer to flow structure.
3509 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3511 const struct mlx5_flow_driver_ops *fops;
3512 enum mlx5_flow_drv_type type = flow->drv_type;
3514 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3515 fops = flow_get_drv_ops(type);
3516 fops->destroy(dev, flow);
3520 * Flow driver find RSS policy tbl API. This abstracts calling driver
3521 * specific functions. Parent flow (rte_flow) should have driver
3522 * type (drv_type). It will find the RSS policy table that has the rss_desc.
3525 * Pointer to Ethernet device.
3526 * @param[in, out] flow
3527 * Pointer to flow structure.
3529 * Pointer to meter policy table.
3530 * @param[in] rss_desc
3531 * Pointer to rss_desc
3533 static struct mlx5_flow_meter_sub_policy *
3534 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3535 struct rte_flow *flow,
3536 struct mlx5_flow_meter_policy *policy,
3537 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3539 const struct mlx5_flow_driver_ops *fops;
3540 enum mlx5_flow_drv_type type = flow->drv_type;
3542 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3543 fops = flow_get_drv_ops(type);
3544 return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3548 * Flow driver color tag rule API. This abstracts calling driver
3549 * specific functions. Parent flow (rte_flow) should have driver
3550 * type (drv_type). It will create the color tag rules in hierarchy meter.
3553 * Pointer to Ethernet device.
3554 * @param[in, out] flow
3555 * Pointer to flow structure.
3557 * Pointer to flow meter structure.
3558 * @param[in] src_port
3559 * The src port this extra rule should use.
3561 * The src port id match item.
3563 * Pointer to error structure.
3566 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3567 struct rte_flow *flow,
3568 struct mlx5_flow_meter_info *fm,
3570 const struct rte_flow_item *item,
3571 struct rte_flow_error *error)
3573 const struct mlx5_flow_driver_ops *fops;
3574 enum mlx5_flow_drv_type type = flow->drv_type;
3576 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3577 fops = flow_get_drv_ops(type);
3578 return fops->meter_hierarchy_rule_create(dev, fm,
3579 src_port, item, error);
3583 * Get RSS action from the action list.
3586 * Pointer to Ethernet device.
3587 * @param[in] actions
3588 * Pointer to the list of actions.
3590 * Parent flow structure pointer.
3593 * Pointer to the RSS action if exist, else return NULL.
3595 static const struct rte_flow_action_rss*
3596 flow_get_rss_action(struct rte_eth_dev *dev,
3597 const struct rte_flow_action actions[])
3599 struct mlx5_priv *priv = dev->data->dev_private;
3600 const struct rte_flow_action_rss *rss = NULL;
3601 struct mlx5_meter_policy_action_container *acg;
3602 struct mlx5_meter_policy_action_container *acy;
3604 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3605 switch (actions->type) {
3606 case RTE_FLOW_ACTION_TYPE_RSS:
3607 rss = actions->conf;
3609 case RTE_FLOW_ACTION_TYPE_SAMPLE:
3611 const struct rte_flow_action_sample *sample =
3613 const struct rte_flow_action *act = sample->actions;
3614 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3615 if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3619 case RTE_FLOW_ACTION_TYPE_METER:
3622 struct mlx5_flow_meter_info *fm;
3623 struct mlx5_flow_meter_policy *policy;
3624 const struct rte_flow_action_meter *mtr = actions->conf;
3626 fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3627 if (fm && !fm->def_policy) {
3628 policy = mlx5_flow_meter_policy_find(dev,
3629 fm->policy_id, NULL);
3630 MLX5_ASSERT(policy);
3631 if (policy->is_hierarchy) {
3633 mlx5_flow_meter_hierarchy_get_final_policy(dev,
3638 if (policy->is_rss) {
3640 &policy->act_cnt[RTE_COLOR_GREEN];
3642 &policy->act_cnt[RTE_COLOR_YELLOW];
3643 if (acg->fate_action ==
3644 MLX5_FLOW_FATE_SHARED_RSS)
3645 rss = acg->rss->conf;
3646 else if (acy->fate_action ==
3647 MLX5_FLOW_FATE_SHARED_RSS)
3648 rss = acy->rss->conf;
3661 * Get ASO age action by index.
3664 * Pointer to the Ethernet device structure.
3665 * @param[in] age_idx
3666 * Index to the ASO age action.
3669 * The specified ASO age action.
3671 struct mlx5_aso_age_action*
3672 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3674 uint16_t pool_idx = age_idx & UINT16_MAX;
3675 uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3676 struct mlx5_priv *priv = dev->data->dev_private;
3677 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3678 struct mlx5_aso_age_pool *pool;
3680 rte_rwlock_read_lock(&mng->resize_rwl);
3681 pool = mng->pools[pool_idx];
3682 rte_rwlock_read_unlock(&mng->resize_rwl);
3683 return &pool->actions[offset - 1];
3686 /* maps indirect action to translated direct in some actions array */
3687 struct mlx5_translated_action_handle {
3688 struct rte_flow_action_handle *action; /**< Indirect action handle. */
3689 int index; /**< Index in related array of rte_flow_action. */
3693 * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3694 * direct action if translation possible.
3695 * This functionality used to run same execution path for both direct and
3696 * indirect actions on flow create. All necessary preparations for indirect
3697 * action handling should be performed on *handle* actions list returned
3701 * Pointer to Ethernet device.
3702 * @param[in] actions
3703 * List of actions to translate.
3704 * @param[out] handle
3705 * List to store translated indirect action object handles.
3706 * @param[in, out] indir_n
3707 * Size of *handle* array. On return should be updated with number of
3708 * indirect actions retrieved from the *actions* list.
3709 * @param[out] translated_actions
3710 * List of actions where all indirect actions were translated to direct
3711 * if possible. NULL if no translation took place.
3713 * Pointer to the error structure.
3716 * 0 on success, a negative errno value otherwise and rte_errno is set.
3719 flow_action_handles_translate(struct rte_eth_dev *dev,
3720 const struct rte_flow_action actions[],
3721 struct mlx5_translated_action_handle *handle,
3723 struct rte_flow_action **translated_actions,
3724 struct rte_flow_error *error)
3726 struct mlx5_priv *priv = dev->data->dev_private;
3727 struct rte_flow_action *translated = NULL;
3728 size_t actions_size;
3731 struct mlx5_translated_action_handle *handle_end = NULL;
3733 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3734 if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3736 if (copied_n == *indir_n) {
3737 return rte_flow_error_set
3738 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3739 NULL, "too many shared actions");
3741 rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3742 sizeof(actions[n].conf));
3743 handle[copied_n].index = n;
3747 *indir_n = copied_n;
3750 actions_size = sizeof(struct rte_flow_action) * n;
3751 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3756 memcpy(translated, actions, actions_size);
3757 for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3758 struct mlx5_shared_action_rss *shared_rss;
3759 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3760 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3761 uint32_t idx = act_idx &
3762 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3765 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3766 shared_rss = mlx5_ipool_get
3767 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3768 translated[handle->index].type =
3769 RTE_FLOW_ACTION_TYPE_RSS;
3770 translated[handle->index].conf =
3771 &shared_rss->origin;
3773 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3774 translated[handle->index].type =
3775 (enum rte_flow_action_type)
3776 MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3777 translated[handle->index].conf = (void *)(uintptr_t)idx;
3779 case MLX5_INDIRECT_ACTION_TYPE_AGE:
3780 if (priv->sh->flow_hit_aso_en) {
3781 translated[handle->index].type =
3782 (enum rte_flow_action_type)
3783 MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3784 translated[handle->index].conf =
3785 (void *)(uintptr_t)idx;
3789 case MLX5_INDIRECT_ACTION_TYPE_CT:
3790 if (priv->sh->ct_aso_en) {
3791 translated[handle->index].type =
3792 RTE_FLOW_ACTION_TYPE_CONNTRACK;
3793 translated[handle->index].conf =
3794 (void *)(uintptr_t)idx;
3799 mlx5_free(translated);
3800 return rte_flow_error_set
3801 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3802 NULL, "invalid indirect action type");
3805 *translated_actions = translated;
3810 * Get Shared RSS action from the action list.
3813 * Pointer to Ethernet device.
3815 * Pointer to the list of actions.
3816 * @param[in] shared_n
3817 * Actions list length.
3820 * The MLX5 RSS action ID if exists, otherwise return 0.
3823 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3824 struct mlx5_translated_action_handle *handle,
3827 struct mlx5_translated_action_handle *handle_end;
3828 struct mlx5_priv *priv = dev->data->dev_private;
3829 struct mlx5_shared_action_rss *shared_rss;
3832 for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3833 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3834 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3835 uint32_t idx = act_idx &
3836 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3838 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3839 shared_rss = mlx5_ipool_get
3840 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3842 __atomic_add_fetch(&shared_rss->refcnt, 1,
3853 find_graph_root(uint32_t rss_level)
3855 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3856 MLX5_EXPANSION_ROOT_OUTER;
3860 * Get layer flags from the prefix flow.
3862 * Some flows may be split to several subflows, the prefix subflow gets the
3863 * match items and the suffix sub flow gets the actions.
3864 * Some actions need the user defined match item flags to get the detail for
3866 * This function helps the suffix flow to get the item layer flags from prefix
3869 * @param[in] dev_flow
3870 * Pointer the created preifx subflow.
3873 * The layers get from prefix subflow.
3875 static inline uint64_t
3876 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3878 uint64_t layers = 0;
3881 * Layers bits could be localization, but usually the compiler will
3882 * help to do the optimization work for source code.
3883 * If no decap actions, use the layers directly.
3885 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3886 return dev_flow->handle->layers;
3887 /* Convert L3 layers with decap action. */
3888 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3889 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3890 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3891 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3892 /* Convert L4 layers with decap action. */
3893 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3894 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3895 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3896 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3901 * Get metadata split action information.
3903 * @param[in] actions
3904 * Pointer to the list of actions.
3906 * Pointer to the return pointer.
3907 * @param[out] qrss_type
3908 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3909 * if no QUEUE/RSS is found.
3910 * @param[out] encap_idx
3911 * Pointer to the index of the encap action if exists, otherwise the last
3915 * Total number of actions.
3918 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3919 const struct rte_flow_action **qrss,
3922 const struct rte_flow_action_raw_encap *raw_encap;
3924 int raw_decap_idx = -1;
3927 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3928 switch (actions->type) {
3929 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3930 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3931 *encap_idx = actions_n;
3933 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3934 raw_decap_idx = actions_n;
3936 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3937 raw_encap = actions->conf;
3938 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3939 *encap_idx = raw_decap_idx != -1 ?
3940 raw_decap_idx : actions_n;
3942 case RTE_FLOW_ACTION_TYPE_QUEUE:
3943 case RTE_FLOW_ACTION_TYPE_RSS:
3951 if (*encap_idx == -1)
3952 *encap_idx = actions_n;
3953 /* Count RTE_FLOW_ACTION_TYPE_END. */
3954 return actions_n + 1;
3958 * Check if the action will change packet.
3961 * Pointer to Ethernet device.
3966 * true if action will change packet, false otherwise.
3968 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3969 enum rte_flow_action_type type)
3971 struct mlx5_priv *priv = dev->data->dev_private;
3974 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3975 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3976 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3977 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3978 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3979 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3980 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3981 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3982 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3983 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3984 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3985 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3986 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3987 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3988 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3989 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3990 case RTE_FLOW_ACTION_TYPE_SET_META:
3991 case RTE_FLOW_ACTION_TYPE_SET_TAG:
3992 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3993 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3994 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3995 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3996 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3997 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3998 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3999 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4000 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4001 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4002 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4004 case RTE_FLOW_ACTION_TYPE_FLAG:
4005 case RTE_FLOW_ACTION_TYPE_MARK:
4006 if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4016 * Check meter action from the action list.
4019 * Pointer to Ethernet device.
4020 * @param[in] actions
4021 * Pointer to the list of actions.
4022 * @param[out] has_mtr
4023 * Pointer to the meter exist flag.
4024 * @param[out] has_modify
4025 * Pointer to the flag showing there's packet change action.
4026 * @param[out] meter_id
4027 * Pointer to the meter id.
4030 * Total number of actions.
4033 flow_check_meter_action(struct rte_eth_dev *dev,
4034 const struct rte_flow_action actions[],
4035 bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4037 const struct rte_flow_action_meter *mtr = NULL;
4040 MLX5_ASSERT(has_mtr);
4042 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4043 switch (actions->type) {
4044 case RTE_FLOW_ACTION_TYPE_METER:
4045 mtr = actions->conf;
4046 *meter_id = mtr->mtr_id;
4053 *has_modify |= flow_check_modify_action_type(dev,
4057 /* Count RTE_FLOW_ACTION_TYPE_END. */
4058 return actions_n + 1;
4062 * Check if the flow should be split due to hairpin.
4063 * The reason for the split is that in current HW we can't
4064 * support encap and push-vlan on Rx, so if a flow contains
4065 * these actions we move it to Tx.
4068 * Pointer to Ethernet device.
4070 * Flow rule attributes.
4071 * @param[in] actions
4072 * Associated actions (list terminated by the END action).
4075 * > 0 the number of actions and the flow should be split,
4076 * 0 when no split required.
4079 flow_check_hairpin_split(struct rte_eth_dev *dev,
4080 const struct rte_flow_attr *attr,
4081 const struct rte_flow_action actions[])
4083 int queue_action = 0;
4086 const struct rte_flow_action_queue *queue;
4087 const struct rte_flow_action_rss *rss;
4088 const struct rte_flow_action_raw_encap *raw_encap;
4089 const struct rte_eth_hairpin_conf *conf;
4093 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4094 switch (actions->type) {
4095 case RTE_FLOW_ACTION_TYPE_QUEUE:
4096 queue = actions->conf;
4099 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4100 if (conf == NULL || conf->tx_explicit != 0)
4105 case RTE_FLOW_ACTION_TYPE_RSS:
4106 rss = actions->conf;
4107 if (rss == NULL || rss->queue_num == 0)
4109 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4110 if (conf == NULL || conf->tx_explicit != 0)
4115 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4116 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4117 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4118 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4119 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4123 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4124 raw_encap = actions->conf;
4125 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4134 if (split && queue_action)
4139 /* Declare flow create/destroy prototype in advance. */
4141 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4142 const struct rte_flow_attr *attr,
4143 const struct rte_flow_item items[],
4144 const struct rte_flow_action actions[],
4145 bool external, struct rte_flow_error *error);
4148 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4152 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4153 struct mlx5_list_entry *entry, void *cb_ctx)
4155 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4156 struct mlx5_flow_mreg_copy_resource *mcp_res =
4157 container_of(entry, typeof(*mcp_res), hlist_ent);
4159 return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4162 struct mlx5_list_entry *
4163 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4165 struct rte_eth_dev *dev = tool_ctx;
4166 struct mlx5_priv *priv = dev->data->dev_private;
4167 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4168 struct mlx5_flow_mreg_copy_resource *mcp_res;
4169 struct rte_flow_error *error = ctx->error;
4172 uint32_t mark_id = *(uint32_t *)(ctx->data);
4173 struct rte_flow_attr attr = {
4174 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4177 struct mlx5_rte_flow_item_tag tag_spec = {
4180 struct rte_flow_item items[] = {
4181 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4183 struct rte_flow_action_mark ftag = {
4186 struct mlx5_flow_action_copy_mreg cp_mreg = {
4190 struct rte_flow_action_jump jump = {
4191 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4193 struct rte_flow_action actions[] = {
4194 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4197 /* Fill the register fileds in the flow. */
4198 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4202 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4206 /* Provide the full width of FLAG specific value. */
4207 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4208 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4209 /* Build a new flow. */
4210 if (mark_id != MLX5_DEFAULT_COPY_ID) {
4211 items[0] = (struct rte_flow_item){
4212 .type = (enum rte_flow_item_type)
4213 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4216 items[1] = (struct rte_flow_item){
4217 .type = RTE_FLOW_ITEM_TYPE_END,
4219 actions[0] = (struct rte_flow_action){
4220 .type = (enum rte_flow_action_type)
4221 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4224 actions[1] = (struct rte_flow_action){
4225 .type = (enum rte_flow_action_type)
4226 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4229 actions[2] = (struct rte_flow_action){
4230 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4233 actions[3] = (struct rte_flow_action){
4234 .type = RTE_FLOW_ACTION_TYPE_END,
4237 /* Default rule, wildcard match. */
4238 attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4239 items[0] = (struct rte_flow_item){
4240 .type = RTE_FLOW_ITEM_TYPE_END,
4242 actions[0] = (struct rte_flow_action){
4243 .type = (enum rte_flow_action_type)
4244 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4247 actions[1] = (struct rte_flow_action){
4248 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4251 actions[2] = (struct rte_flow_action){
4252 .type = RTE_FLOW_ACTION_TYPE_END,
4255 /* Build a new entry. */
4256 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4262 mcp_res->mark_id = mark_id;
4264 * The copy Flows are not included in any list. There
4265 * ones are referenced from other Flows and can not
4266 * be applied, removed, deleted in ardbitrary order
4267 * by list traversing.
4269 mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4270 &attr, items, actions, false, error);
4271 if (!mcp_res->rix_flow) {
4272 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4275 return &mcp_res->hlist_ent;
4278 struct mlx5_list_entry *
4279 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4280 void *cb_ctx __rte_unused)
4282 struct rte_eth_dev *dev = tool_ctx;
4283 struct mlx5_priv *priv = dev->data->dev_private;
4284 struct mlx5_flow_mreg_copy_resource *mcp_res;
4287 mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4292 memcpy(mcp_res, oentry, sizeof(*mcp_res));
4294 return &mcp_res->hlist_ent;
4298 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4300 struct mlx5_flow_mreg_copy_resource *mcp_res =
4301 container_of(entry, typeof(*mcp_res), hlist_ent);
4302 struct rte_eth_dev *dev = tool_ctx;
4303 struct mlx5_priv *priv = dev->data->dev_private;
4305 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4309 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4311 * As mark_id is unique, if there's already a registered flow for the mark_id,
4312 * return by increasing the reference counter of the resource. Otherwise, create
4313 * the resource (mcp_res) and flow.
4316 * - If ingress port is ANY and reg_c[1] is mark_id,
4317 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4319 * For default flow (zero mark_id), flow is like,
4320 * - If ingress port is ANY,
4321 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
4324 * Pointer to Ethernet device.
4326 * ID of MARK action, zero means default flow for META.
4328 * Perform verbose error reporting if not NULL.
4331 * Associated resource on success, NULL otherwise and rte_errno is set.
4333 static struct mlx5_flow_mreg_copy_resource *
4334 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4335 struct rte_flow_error *error)
4337 struct mlx5_priv *priv = dev->data->dev_private;
4338 struct mlx5_list_entry *entry;
4339 struct mlx5_flow_cb_ctx ctx = {
4345 /* Check if already registered. */
4346 MLX5_ASSERT(priv->mreg_cp_tbl);
4347 entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4350 return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4355 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4357 struct mlx5_flow_mreg_copy_resource *mcp_res =
4358 container_of(entry, typeof(*mcp_res), hlist_ent);
4359 struct rte_eth_dev *dev = tool_ctx;
4360 struct mlx5_priv *priv = dev->data->dev_private;
4362 MLX5_ASSERT(mcp_res->rix_flow);
4363 flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4364 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4368 * Release flow in RX_CP_TBL.
4371 * Pointer to Ethernet device.
4373 * Parent flow for wich copying is provided.
4376 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4377 struct rte_flow *flow)
4379 struct mlx5_flow_mreg_copy_resource *mcp_res;
4380 struct mlx5_priv *priv = dev->data->dev_private;
4382 if (!flow->rix_mreg_copy)
4384 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4385 flow->rix_mreg_copy);
4386 if (!mcp_res || !priv->mreg_cp_tbl)
4388 MLX5_ASSERT(mcp_res->rix_flow);
4389 mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4390 flow->rix_mreg_copy = 0;
4394 * Remove the default copy action from RX_CP_TBL.
4396 * This functions is called in the mlx5_dev_start(). No thread safe
4400 * Pointer to Ethernet device.
4403 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4405 struct mlx5_list_entry *entry;
4406 struct mlx5_priv *priv = dev->data->dev_private;
4407 struct mlx5_flow_cb_ctx ctx;
4410 /* Check if default flow is registered. */
4411 if (!priv->mreg_cp_tbl)
4413 mark_id = MLX5_DEFAULT_COPY_ID;
4414 ctx.data = &mark_id;
4415 entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4418 mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4422 * Add the default copy action in in RX_CP_TBL.
4424 * This functions is called in the mlx5_dev_start(). No thread safe
4428 * Pointer to Ethernet device.
4430 * Perform verbose error reporting if not NULL.
4433 * 0 for success, negative value otherwise and rte_errno is set.
4436 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4437 struct rte_flow_error *error)
4439 struct mlx5_priv *priv = dev->data->dev_private;
4440 struct mlx5_flow_mreg_copy_resource *mcp_res;
4441 struct mlx5_flow_cb_ctx ctx;
4444 /* Check whether extensive metadata feature is engaged. */
4445 if (!priv->config.dv_flow_en ||
4446 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4447 !mlx5_flow_ext_mreg_supported(dev) ||
4448 !priv->sh->dv_regc0_mask)
4451 * Add default mreg copy flow may be called multiple time, but
4452 * only be called once in stop. Avoid register it twice.
4454 mark_id = MLX5_DEFAULT_COPY_ID;
4455 ctx.data = &mark_id;
4456 if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4458 mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4465 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4467 * All the flow having Q/RSS action should be split by
4468 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4469 * performs the following,
4470 * - CQE->flow_tag := reg_c[1] (MARK)
4471 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4472 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4473 * but there should be a flow per each MARK ID set by MARK action.
4475 * For the aforementioned reason, if there's a MARK action in flow's action
4476 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4477 * the MARK ID to CQE's flow_tag like,
4478 * - If reg_c[1] is mark_id,
4479 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4481 * For SET_META action which stores value in reg_c[0], as the destination is
4482 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4483 * MARK ID means the default flow. The default flow looks like,
4484 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4487 * Pointer to Ethernet device.
4489 * Pointer to flow structure.
4490 * @param[in] actions
4491 * Pointer to the list of actions.
4493 * Perform verbose error reporting if not NULL.
4496 * 0 on success, negative value otherwise and rte_errno is set.
4499 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4500 struct rte_flow *flow,
4501 const struct rte_flow_action *actions,
4502 struct rte_flow_error *error)
4504 struct mlx5_priv *priv = dev->data->dev_private;
4505 struct mlx5_dev_config *config = &priv->config;
4506 struct mlx5_flow_mreg_copy_resource *mcp_res;
4507 const struct rte_flow_action_mark *mark;
4509 /* Check whether extensive metadata feature is engaged. */
4510 if (!config->dv_flow_en ||
4511 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4512 !mlx5_flow_ext_mreg_supported(dev) ||
4513 !priv->sh->dv_regc0_mask)
4515 /* Find MARK action. */
4516 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4517 switch (actions->type) {
4518 case RTE_FLOW_ACTION_TYPE_FLAG:
4519 mcp_res = flow_mreg_add_copy_action
4520 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4523 flow->rix_mreg_copy = mcp_res->idx;
4525 case RTE_FLOW_ACTION_TYPE_MARK:
4526 mark = (const struct rte_flow_action_mark *)
4529 flow_mreg_add_copy_action(dev, mark->id, error);
4532 flow->rix_mreg_copy = mcp_res->idx;
4541 #define MLX5_MAX_SPLIT_ACTIONS 24
4542 #define MLX5_MAX_SPLIT_ITEMS 24
4545 * Split the hairpin flow.
4546 * Since HW can't support encap and push-vlan on Rx, we move these
4548 * If the count action is after the encap then we also
4549 * move the count action. in this case the count will also measure
4553 * Pointer to Ethernet device.
4554 * @param[in] actions
4555 * Associated actions (list terminated by the END action).
4556 * @param[out] actions_rx
4558 * @param[out] actions_tx
4560 * @param[out] pattern_tx
4561 * The pattern items for the Tx flow.
4562 * @param[out] flow_id
4563 * The flow ID connected to this flow.
4569 flow_hairpin_split(struct rte_eth_dev *dev,
4570 const struct rte_flow_action actions[],
4571 struct rte_flow_action actions_rx[],
4572 struct rte_flow_action actions_tx[],
4573 struct rte_flow_item pattern_tx[],
4576 const struct rte_flow_action_raw_encap *raw_encap;
4577 const struct rte_flow_action_raw_decap *raw_decap;
4578 struct mlx5_rte_flow_action_set_tag *set_tag;
4579 struct rte_flow_action *tag_action;
4580 struct mlx5_rte_flow_item_tag *tag_item;
4581 struct rte_flow_item *item;
4585 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4586 switch (actions->type) {
4587 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4588 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4589 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4590 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4591 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4592 rte_memcpy(actions_tx, actions,
4593 sizeof(struct rte_flow_action));
4596 case RTE_FLOW_ACTION_TYPE_COUNT:
4598 rte_memcpy(actions_tx, actions,
4599 sizeof(struct rte_flow_action));
4602 rte_memcpy(actions_rx, actions,
4603 sizeof(struct rte_flow_action));
4607 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4608 raw_encap = actions->conf;
4609 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4610 memcpy(actions_tx, actions,
4611 sizeof(struct rte_flow_action));
4615 rte_memcpy(actions_rx, actions,
4616 sizeof(struct rte_flow_action));
4620 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4621 raw_decap = actions->conf;
4622 if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4623 memcpy(actions_tx, actions,
4624 sizeof(struct rte_flow_action));
4627 rte_memcpy(actions_rx, actions,
4628 sizeof(struct rte_flow_action));
4633 rte_memcpy(actions_rx, actions,
4634 sizeof(struct rte_flow_action));
4639 /* Add set meta action and end action for the Rx flow. */
4640 tag_action = actions_rx;
4641 tag_action->type = (enum rte_flow_action_type)
4642 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4644 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4646 set_tag = (void *)actions_rx;
4647 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4648 .id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4651 MLX5_ASSERT(set_tag->id > REG_NON);
4652 tag_action->conf = set_tag;
4653 /* Create Tx item list. */
4654 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4655 addr = (void *)&pattern_tx[2];
4657 item->type = (enum rte_flow_item_type)
4658 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4659 tag_item = (void *)addr;
4660 tag_item->data = flow_id;
4661 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4662 MLX5_ASSERT(set_tag->id > REG_NON);
4663 item->spec = tag_item;
4664 addr += sizeof(struct mlx5_rte_flow_item_tag);
4665 tag_item = (void *)addr;
4666 tag_item->data = UINT32_MAX;
4667 tag_item->id = UINT16_MAX;
4668 item->mask = tag_item;
4671 item->type = RTE_FLOW_ITEM_TYPE_END;
4676 * The last stage of splitting chain, just creates the subflow
4677 * without any modification.
4680 * Pointer to Ethernet device.
4682 * Parent flow structure pointer.
4683 * @param[in, out] sub_flow
4684 * Pointer to return the created subflow, may be NULL.
4686 * Flow rule attributes.
4688 * Pattern specification (list terminated by the END pattern item).
4689 * @param[in] actions
4690 * Associated actions (list terminated by the END action).
4691 * @param[in] flow_split_info
4692 * Pointer to flow split info structure.
4694 * Perform verbose error reporting if not NULL.
4696 * 0 on success, negative value otherwise
4699 flow_create_split_inner(struct rte_eth_dev *dev,
4700 struct rte_flow *flow,
4701 struct mlx5_flow **sub_flow,
4702 const struct rte_flow_attr *attr,
4703 const struct rte_flow_item items[],
4704 const struct rte_flow_action actions[],
4705 struct mlx5_flow_split_info *flow_split_info,
4706 struct rte_flow_error *error)
4708 struct mlx5_flow *dev_flow;
4710 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4711 flow_split_info->flow_idx, error);
4714 dev_flow->flow = flow;
4715 dev_flow->external = flow_split_info->external;
4716 dev_flow->skip_scale = flow_split_info->skip_scale;
4717 /* Subflow object was created, we must include one in the list. */
4718 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4719 dev_flow->handle, next);
4721 * If dev_flow is as one of the suffix flow, some actions in suffix
4722 * flow may need some user defined item layer flags, and pass the
4723 * Metadate rxq mark flag to suffix flow as well.
4725 if (flow_split_info->prefix_layers)
4726 dev_flow->handle->layers = flow_split_info->prefix_layers;
4727 if (flow_split_info->prefix_mark)
4728 dev_flow->handle->mark = 1;
4730 *sub_flow = dev_flow;
4731 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4732 dev_flow->dv.table_id = flow_split_info->table_id;
4734 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4738 * Get the sub policy of a meter.
4741 * Pointer to Ethernet device.
4743 * Parent flow structure pointer.
4745 * Pointer to thread flow work space.
4747 * Flow rule attributes.
4749 * Pattern specification (list terminated by the END pattern item).
4751 * Perform verbose error reporting if not NULL.
4754 * Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4756 static struct mlx5_flow_meter_sub_policy *
4757 get_meter_sub_policy(struct rte_eth_dev *dev,
4758 struct rte_flow *flow,
4759 struct mlx5_flow_workspace *wks,
4760 const struct rte_flow_attr *attr,
4761 const struct rte_flow_item items[],
4762 struct rte_flow_error *error)
4764 struct mlx5_flow_meter_policy *policy;
4765 struct mlx5_flow_meter_policy *final_policy;
4766 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4768 policy = wks->policy;
4769 final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4770 if (final_policy->is_rss || final_policy->is_queue) {
4771 struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4772 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4776 * This is a tmp dev_flow,
4777 * no need to register any matcher for it in translate.
4779 wks->skip_matcher_reg = 1;
4780 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4781 struct mlx5_flow dev_flow = {0};
4782 struct mlx5_flow_handle dev_handle = { {0} };
4783 uint8_t fate = final_policy->act_cnt[i].fate_action;
4785 if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4786 const struct rte_flow_action_rss *rss_act =
4787 final_policy->act_cnt[i].rss->conf;
4788 struct rte_flow_action rss_actions[2] = {
4790 .type = RTE_FLOW_ACTION_TYPE_RSS,
4794 .type = RTE_FLOW_ACTION_TYPE_END,
4799 dev_flow.handle = &dev_handle;
4800 dev_flow.ingress = attr->ingress;
4801 dev_flow.flow = flow;
4802 dev_flow.external = 0;
4803 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4804 dev_flow.dv.transfer = attr->transfer;
4807 * Translate RSS action to get rss hash fields.
4809 if (flow_drv_translate(dev, &dev_flow, attr,
4810 items, rss_actions, error))
4812 rss_desc_v[i] = wks->rss_desc;
4813 rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4814 rss_desc_v[i].hash_fields =
4815 dev_flow.hash_fields;
4816 rss_desc_v[i].queue_num =
4817 rss_desc_v[i].hash_fields ?
4818 rss_desc_v[i].queue_num : 1;
4819 rss_desc_v[i].tunnel =
4820 !!(dev_flow.handle->layers &
4821 MLX5_FLOW_LAYER_TUNNEL);
4822 /* Use the RSS queues in the containers. */
4823 rss_desc_v[i].queue =
4824 (uint16_t *)(uintptr_t)rss_act->queue;
4825 rss_desc[i] = &rss_desc_v[i];
4826 } else if (fate == MLX5_FLOW_FATE_QUEUE) {
4827 /* This is queue action. */
4828 rss_desc_v[i] = wks->rss_desc;
4829 rss_desc_v[i].key_len = 0;
4830 rss_desc_v[i].hash_fields = 0;
4831 rss_desc_v[i].queue =
4832 &final_policy->act_cnt[i].queue;
4833 rss_desc_v[i].queue_num = 1;
4834 rss_desc[i] = &rss_desc_v[i];
4839 sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4840 flow, policy, rss_desc);
4842 enum mlx5_meter_domain mtr_domain =
4843 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4844 (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4845 MLX5_MTR_DOMAIN_INGRESS);
4846 sub_policy = policy->sub_policys[mtr_domain][0];
4849 rte_flow_error_set(error, EINVAL,
4850 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4851 "Failed to get meter sub-policy.");
4857 * Split the meter flow.
4859 * As meter flow will split to three sub flow, other than meter
4860 * action, the other actions make sense to only meter accepts
4861 * the packet. If it need to be dropped, no other additional
4862 * actions should be take.
4864 * One kind of special action which decapsulates the L3 tunnel
4865 * header will be in the prefix sub flow, as not to take the
4866 * L3 tunnel header into account.
4869 * Pointer to Ethernet device.
4871 * Parent flow structure pointer.
4873 * Pointer to thread flow work space.
4875 * Flow rule attributes.
4877 * Pattern specification (list terminated by the END pattern item).
4878 * @param[out] sfx_items
4879 * Suffix flow match items (list terminated by the END pattern item).
4880 * @param[in] actions
4881 * Associated actions (list terminated by the END action).
4882 * @param[out] actions_sfx
4883 * Suffix flow actions.
4884 * @param[out] actions_pre
4885 * Prefix flow actions.
4886 * @param[out] mtr_flow_id
4887 * Pointer to meter flow id.
4889 * Perform verbose error reporting if not NULL.
4892 * 0 on success, a negative errno value otherwise and rte_errno is set.
4895 flow_meter_split_prep(struct rte_eth_dev *dev,
4896 struct rte_flow *flow,
4897 struct mlx5_flow_workspace *wks,
4898 const struct rte_flow_attr *attr,
4899 const struct rte_flow_item items[],
4900 struct rte_flow_item sfx_items[],
4901 const struct rte_flow_action actions[],
4902 struct rte_flow_action actions_sfx[],
4903 struct rte_flow_action actions_pre[],
4904 uint32_t *mtr_flow_id,
4905 struct rte_flow_error *error)
4907 struct mlx5_priv *priv = dev->data->dev_private;
4908 struct mlx5_flow_meter_info *fm = wks->fm;
4909 struct rte_flow_action *tag_action = NULL;
4910 struct rte_flow_item *tag_item;
4911 struct mlx5_rte_flow_action_set_tag *set_tag;
4912 const struct rte_flow_action_raw_encap *raw_encap;
4913 const struct rte_flow_action_raw_decap *raw_decap;
4914 struct mlx5_rte_flow_item_tag *tag_item_spec;
4915 struct mlx5_rte_flow_item_tag *tag_item_mask;
4916 uint32_t tag_id = 0;
4917 struct rte_flow_item *vlan_item_dst = NULL;
4918 const struct rte_flow_item *vlan_item_src = NULL;
4919 struct rte_flow_action *hw_mtr_action;
4920 struct rte_flow_action *action_pre_head = NULL;
4921 int32_t flow_src_port = priv->representor_id;
4923 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4924 uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4925 MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4926 uint32_t flow_id = 0;
4927 uint32_t flow_id_reversed = 0;
4928 uint8_t flow_id_bits = 0;
4931 /* Prepare the suffix subflow items. */
4932 tag_item = sfx_items++;
4933 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4934 struct mlx5_priv *port_priv;
4935 const struct rte_flow_item_port_id *pid_v;
4936 int item_type = items->type;
4938 switch (item_type) {
4939 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4940 pid_v = items->spec;
4942 port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4944 return rte_flow_error_set(error,
4946 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
4948 "Failed to get port info.");
4949 flow_src_port = port_priv->representor_id;
4950 if (!fm->def_policy && wks->policy->is_hierarchy &&
4951 flow_src_port != priv->representor_id) {
4952 if (flow_drv_mtr_hierarchy_rule_create(dev,
4959 memcpy(sfx_items, items, sizeof(*sfx_items));
4962 case RTE_FLOW_ITEM_TYPE_VLAN:
4963 /* Determine if copy vlan item below. */
4964 vlan_item_src = items;
4965 vlan_item_dst = sfx_items++;
4966 vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
4972 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4974 mtr_first = priv->sh->meter_aso_en &&
4975 (attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
4976 /* For ASO meter, meter must be before tag in TX direction. */
4978 action_pre_head = actions_pre++;
4979 /* Leave space for tag action. */
4980 tag_action = actions_pre++;
4982 /* Prepare the actions for prefix and suffix flow. */
4983 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4984 struct rte_flow_action *action_cur = NULL;
4986 switch (actions->type) {
4987 case RTE_FLOW_ACTION_TYPE_METER:
4989 action_cur = action_pre_head;
4991 /* Leave space for tag action. */
4992 tag_action = actions_pre++;
4993 action_cur = actions_pre++;
4996 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4997 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4998 action_cur = actions_pre++;
5000 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5001 raw_encap = actions->conf;
5002 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5003 action_cur = actions_pre++;
5005 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5006 raw_decap = actions->conf;
5007 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5008 action_cur = actions_pre++;
5010 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5011 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5012 if (vlan_item_dst && vlan_item_src) {
5013 memcpy(vlan_item_dst, vlan_item_src,
5014 sizeof(*vlan_item_dst));
5016 * Convert to internal match item, it is used
5017 * for vlan push and set vid.
5019 vlan_item_dst->type = (enum rte_flow_item_type)
5020 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5027 action_cur = (fm->def_policy) ?
5028 actions_sfx++ : actions_pre++;
5029 memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5031 /* Add end action to the actions. */
5032 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5033 if (priv->sh->meter_aso_en) {
5035 * For ASO meter, need to add an extra jump action explicitly,
5036 * to jump from meter to policer table.
5038 struct mlx5_flow_meter_sub_policy *sub_policy;
5039 struct mlx5_flow_tbl_data_entry *tbl_data;
5041 if (!fm->def_policy) {
5042 sub_policy = get_meter_sub_policy(dev, flow, wks,
5043 attr, items, error);
5047 enum mlx5_meter_domain mtr_domain =
5048 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5049 (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5050 MLX5_MTR_DOMAIN_INGRESS);
5053 &priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5055 tbl_data = container_of(sub_policy->tbl_rsc,
5056 struct mlx5_flow_tbl_data_entry, tbl);
5057 hw_mtr_action = actions_pre++;
5058 hw_mtr_action->type = (enum rte_flow_action_type)
5059 MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5060 hw_mtr_action->conf = tbl_data->jump.action;
5062 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5065 return rte_flow_error_set(error, ENOMEM,
5066 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5067 NULL, "No tag action space.");
5069 tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5072 /* Only default-policy Meter creates mtr flow id. */
5073 if (fm->def_policy) {
5074 mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5076 return rte_flow_error_set(error, ENOMEM,
5077 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5078 "Failed to allocate meter flow id.");
5079 flow_id = tag_id - 1;
5080 flow_id_bits = (!flow_id) ? 1 :
5081 (MLX5_REG_BITS - __builtin_clz(flow_id));
5082 if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5084 mlx5_ipool_free(fm->flow_ipool, tag_id);
5085 return rte_flow_error_set(error, EINVAL,
5086 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5087 "Meter flow id exceeds max limit.");
5089 if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5090 priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5092 /* Build tag actions and items for meter_id/meter flow_id. */
5093 set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5094 tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5095 tag_item_mask = tag_item_spec + 1;
5096 /* Both flow_id and meter_id share the same register. */
5097 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5098 .id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5100 .offset = mtr_id_offset,
5101 .length = mtr_reg_bits,
5102 .data = flow->meter,
5105 * The color Reg bits used by flow_id are growing from
5106 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5108 for (shift = 0; shift < flow_id_bits; shift++)
5109 flow_id_reversed = (flow_id_reversed << 1) |
5110 ((flow_id >> shift) & 0x1);
5112 flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5113 tag_item_spec->id = set_tag->id;
5114 tag_item_spec->data = set_tag->data << mtr_id_offset;
5115 tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5116 tag_action->type = (enum rte_flow_action_type)
5117 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5118 tag_action->conf = set_tag;
5119 tag_item->type = (enum rte_flow_item_type)
5120 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5121 tag_item->spec = tag_item_spec;
5122 tag_item->last = NULL;
5123 tag_item->mask = tag_item_mask;
5126 *mtr_flow_id = tag_id;
5131 * Split action list having QUEUE/RSS for metadata register copy.
5133 * Once Q/RSS action is detected in user's action list, the flow action
5134 * should be split in order to copy metadata registers, which will happen in
5136 * - CQE->flow_tag := reg_c[1] (MARK)
5137 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5138 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5139 * This is because the last action of each flow must be a terminal action
5140 * (QUEUE, RSS or DROP).
5142 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5143 * stored and kept in the mlx5_flow structure per each sub_flow.
5145 * The Q/RSS action is replaced with,
5146 * - SET_TAG, setting the allocated flow ID to reg_c[2].
5147 * And the following JUMP action is added at the end,
5148 * - JUMP, to RX_CP_TBL.
5150 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5151 * flow_create_split_metadata() routine. The flow will look like,
5152 * - If flow ID matches (reg_c[2]), perform Q/RSS.
5155 * Pointer to Ethernet device.
5156 * @param[out] split_actions
5157 * Pointer to store split actions to jump to CP_TBL.
5158 * @param[in] actions
5159 * Pointer to the list of original flow actions.
5161 * Pointer to the Q/RSS action.
5162 * @param[in] actions_n
5163 * Number of original actions.
5165 * Perform verbose error reporting if not NULL.
5168 * non-zero unique flow_id on success, otherwise 0 and
5169 * error/rte_error are set.
5172 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5173 struct rte_flow_action *split_actions,
5174 const struct rte_flow_action *actions,
5175 const struct rte_flow_action *qrss,
5176 int actions_n, struct rte_flow_error *error)
5178 struct mlx5_priv *priv = dev->data->dev_private;
5179 struct mlx5_rte_flow_action_set_tag *set_tag;
5180 struct rte_flow_action_jump *jump;
5181 const int qrss_idx = qrss - actions;
5182 uint32_t flow_id = 0;
5186 * Given actions will be split
5187 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5188 * - Add jump to mreg CP_TBL.
5189 * As a result, there will be one more action.
5192 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5193 set_tag = (void *)(split_actions + actions_n);
5195 * If tag action is not set to void(it means we are not the meter
5196 * suffix flow), add the tag action. Since meter suffix flow already
5197 * has the tag added.
5199 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5201 * Allocate the new subflow ID. This one is unique within
5202 * device and not shared with representors. Otherwise,
5203 * we would have to resolve multi-thread access synch
5204 * issue. Each flow on the shared device is appended
5205 * with source vport identifier, so the resulting
5206 * flows will be unique in the shared (by master and
5207 * representors) domain even if they have coinciding
5210 mlx5_ipool_malloc(priv->sh->ipool
5211 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5213 return rte_flow_error_set(error, ENOMEM,
5214 RTE_FLOW_ERROR_TYPE_ACTION,
5215 NULL, "can't allocate id "
5216 "for split Q/RSS subflow");
5217 /* Internal SET_TAG action to set flow ID. */
5218 *set_tag = (struct mlx5_rte_flow_action_set_tag){
5221 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5225 /* Construct new actions array. */
5226 /* Replace QUEUE/RSS action. */
5227 split_actions[qrss_idx] = (struct rte_flow_action){
5228 .type = (enum rte_flow_action_type)
5229 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5233 /* JUMP action to jump to mreg copy table (CP_TBL). */
5234 jump = (void *)(set_tag + 1);
5235 *jump = (struct rte_flow_action_jump){
5236 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5238 split_actions[actions_n - 2] = (struct rte_flow_action){
5239 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5242 split_actions[actions_n - 1] = (struct rte_flow_action){
5243 .type = RTE_FLOW_ACTION_TYPE_END,
5249 * Extend the given action list for Tx metadata copy.
5251 * Copy the given action list to the ext_actions and add flow metadata register
5252 * copy action in order to copy reg_a set by WQE to reg_c[0].
5254 * @param[out] ext_actions
5255 * Pointer to the extended action list.
5256 * @param[in] actions
5257 * Pointer to the list of actions.
5258 * @param[in] actions_n
5259 * Number of actions in the list.
5261 * Perform verbose error reporting if not NULL.
5262 * @param[in] encap_idx
5263 * The encap action inndex.
5266 * 0 on success, negative value otherwise
5269 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5270 struct rte_flow_action *ext_actions,
5271 const struct rte_flow_action *actions,
5272 int actions_n, struct rte_flow_error *error,
5275 struct mlx5_flow_action_copy_mreg *cp_mreg =
5276 (struct mlx5_flow_action_copy_mreg *)
5277 (ext_actions + actions_n + 1);
5280 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5284 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5289 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5290 if (encap_idx == actions_n - 1) {
5291 ext_actions[actions_n - 1] = (struct rte_flow_action){
5292 .type = (enum rte_flow_action_type)
5293 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5296 ext_actions[actions_n] = (struct rte_flow_action){
5297 .type = RTE_FLOW_ACTION_TYPE_END,
5300 ext_actions[encap_idx] = (struct rte_flow_action){
5301 .type = (enum rte_flow_action_type)
5302 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5305 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5306 sizeof(*ext_actions) * (actions_n - encap_idx));
5312 * Check the match action from the action list.
5314 * @param[in] actions
5315 * Pointer to the list of actions.
5317 * Flow rule attributes.
5319 * The action to be check if exist.
5320 * @param[out] match_action_pos
5321 * Pointer to the position of the matched action if exists, otherwise is -1.
5322 * @param[out] qrss_action_pos
5323 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5324 * @param[out] modify_after_mirror
5325 * Pointer to the flag of modify action after FDB mirroring.
5328 * > 0 the total number of actions.
5329 * 0 if not found match action in action list.
5332 flow_check_match_action(const struct rte_flow_action actions[],
5333 const struct rte_flow_attr *attr,
5334 enum rte_flow_action_type action,
5335 int *match_action_pos, int *qrss_action_pos,
5336 int *modify_after_mirror)
5338 const struct rte_flow_action_sample *sample;
5339 const struct rte_flow_action_raw_decap *decap;
5346 *match_action_pos = -1;
5347 *qrss_action_pos = -1;
5348 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5349 if (actions->type == action) {
5351 *match_action_pos = actions_n;
5353 switch (actions->type) {
5354 case RTE_FLOW_ACTION_TYPE_QUEUE:
5355 case RTE_FLOW_ACTION_TYPE_RSS:
5356 *qrss_action_pos = actions_n;
5358 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5359 sample = actions->conf;
5360 ratio = sample->ratio;
5361 sub_type = ((const struct rte_flow_action *)
5362 (sample->actions))->type;
5363 if (ratio == 1 && attr->transfer)
5366 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5367 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5368 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5369 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5370 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5371 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5372 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5373 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5374 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5375 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5376 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5377 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5378 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5379 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5380 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5381 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5382 case RTE_FLOW_ACTION_TYPE_FLAG:
5383 case RTE_FLOW_ACTION_TYPE_MARK:
5384 case RTE_FLOW_ACTION_TYPE_SET_META:
5385 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5386 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5387 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5388 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5389 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5390 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5391 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5392 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5393 case RTE_FLOW_ACTION_TYPE_METER:
5395 *modify_after_mirror = 1;
5397 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5398 decap = actions->conf;
5399 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5402 if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5403 const struct rte_flow_action_raw_encap *encap =
5406 MLX5_ENCAPSULATION_DECISION_SIZE &&
5408 MLX5_ENCAPSULATION_DECISION_SIZE)
5413 *modify_after_mirror = 1;
5420 if (flag && fdb_mirror && !*modify_after_mirror) {
5421 /* FDB mirroring uses the destination array to implement
5422 * instead of FLOW_SAMPLER object.
5424 if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5427 /* Count RTE_FLOW_ACTION_TYPE_END. */
5428 return flag ? actions_n + 1 : 0;
5431 #define SAMPLE_SUFFIX_ITEM 2
5434 * Split the sample flow.
5436 * As sample flow will split to two sub flow, sample flow with
5437 * sample action, the other actions will move to new suffix flow.
5439 * Also add unique tag id with tag action in the sample flow,
5440 * the same tag id will be as match in the suffix flow.
5443 * Pointer to Ethernet device.
5444 * @param[in] add_tag
5445 * Add extra tag action flag.
5446 * @param[out] sfx_items
5447 * Suffix flow match items (list terminated by the END pattern item).
5448 * @param[in] actions
5449 * Associated actions (list terminated by the END action).
5450 * @param[out] actions_sfx
5451 * Suffix flow actions.
5452 * @param[out] actions_pre
5453 * Prefix flow actions.
5454 * @param[in] actions_n
5455 * The total number of actions.
5456 * @param[in] sample_action_pos
5457 * The sample action position.
5458 * @param[in] qrss_action_pos
5459 * The Queue/RSS action position.
5460 * @param[in] jump_table
5461 * Add extra jump action flag.
5463 * Perform verbose error reporting if not NULL.
5466 * 0 on success, or unique flow_id, a negative errno value
5467 * otherwise and rte_errno is set.
5470 flow_sample_split_prep(struct rte_eth_dev *dev,
5472 struct rte_flow_item sfx_items[],
5473 const struct rte_flow_action actions[],
5474 struct rte_flow_action actions_sfx[],
5475 struct rte_flow_action actions_pre[],
5477 int sample_action_pos,
5478 int qrss_action_pos,
5480 struct rte_flow_error *error)
5482 struct mlx5_priv *priv = dev->data->dev_private;
5483 struct mlx5_rte_flow_action_set_tag *set_tag;
5484 struct mlx5_rte_flow_item_tag *tag_spec;
5485 struct mlx5_rte_flow_item_tag *tag_mask;
5486 struct rte_flow_action_jump *jump_action;
5487 uint32_t tag_id = 0;
5489 int append_index = 0;
5492 if (sample_action_pos < 0)
5493 return rte_flow_error_set(error, EINVAL,
5494 RTE_FLOW_ERROR_TYPE_ACTION,
5495 NULL, "invalid position of sample "
5497 /* Prepare the actions for prefix and suffix flow. */
5498 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5499 index = qrss_action_pos;
5500 /* Put the preceding the Queue/RSS action into prefix flow. */
5502 memcpy(actions_pre, actions,
5503 sizeof(struct rte_flow_action) * index);
5504 /* Put others preceding the sample action into prefix flow. */
5505 if (sample_action_pos > index + 1)
5506 memcpy(actions_pre + index, actions + index + 1,
5507 sizeof(struct rte_flow_action) *
5508 (sample_action_pos - index - 1));
5509 index = sample_action_pos - 1;
5510 /* Put Queue/RSS action into Suffix flow. */
5511 memcpy(actions_sfx, actions + qrss_action_pos,
5512 sizeof(struct rte_flow_action));
5515 index = sample_action_pos;
5517 memcpy(actions_pre, actions,
5518 sizeof(struct rte_flow_action) * index);
5520 /* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5521 * For CX6DX and above, metadata registers Cx preserve their value,
5522 * add an extra tag action for NIC-RX and E-Switch Domain.
5525 /* Prepare the prefix tag action. */
5527 set_tag = (void *)(actions_pre + actions_n + append_index);
5528 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5531 mlx5_ipool_malloc(priv->sh->ipool
5532 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5533 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5537 /* Prepare the suffix subflow items. */
5538 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5539 tag_spec->data = tag_id;
5540 tag_spec->id = set_tag->id;
5541 tag_mask = tag_spec + 1;
5542 tag_mask->data = UINT32_MAX;
5543 sfx_items[0] = (struct rte_flow_item){
5544 .type = (enum rte_flow_item_type)
5545 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5550 sfx_items[1] = (struct rte_flow_item){
5551 .type = (enum rte_flow_item_type)
5552 RTE_FLOW_ITEM_TYPE_END,
5554 /* Prepare the tag action in prefix subflow. */
5555 actions_pre[index++] =
5556 (struct rte_flow_action){
5557 .type = (enum rte_flow_action_type)
5558 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5562 memcpy(actions_pre + index, actions + sample_action_pos,
5563 sizeof(struct rte_flow_action));
5565 /* For the modify action after the sample action in E-Switch mirroring,
5566 * Add the extra jump action in prefix subflow and jump into the next
5567 * table, then do the modify action in the new table.
5570 /* Prepare the prefix jump action. */
5572 jump_action = (void *)(actions_pre + actions_n + append_index);
5573 jump_action->group = jump_table;
5574 actions_pre[index++] =
5575 (struct rte_flow_action){
5576 .type = (enum rte_flow_action_type)
5577 RTE_FLOW_ACTION_TYPE_JUMP,
5578 .conf = jump_action,
5581 actions_pre[index] = (struct rte_flow_action){
5582 .type = (enum rte_flow_action_type)
5583 RTE_FLOW_ACTION_TYPE_END,
5585 /* Put the actions after sample into Suffix flow. */
5586 memcpy(actions_sfx, actions + sample_action_pos + 1,
5587 sizeof(struct rte_flow_action) *
5588 (actions_n - sample_action_pos - 1));
5593 * The splitting for metadata feature.
5595 * - Q/RSS action on NIC Rx should be split in order to pass by
5596 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5597 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5599 * - All the actions on NIC Tx should have a mreg copy action to
5600 * copy reg_a from WQE to reg_c[0].
5603 * Pointer to Ethernet device.
5605 * Parent flow structure pointer.
5607 * Flow rule attributes.
5609 * Pattern specification (list terminated by the END pattern item).
5610 * @param[in] actions
5611 * Associated actions (list terminated by the END action).
5612 * @param[in] flow_split_info
5613 * Pointer to flow split info structure.
5615 * Perform verbose error reporting if not NULL.
5617 * 0 on success, negative value otherwise
5620 flow_create_split_metadata(struct rte_eth_dev *dev,
5621 struct rte_flow *flow,
5622 const struct rte_flow_attr *attr,
5623 const struct rte_flow_item items[],
5624 const struct rte_flow_action actions[],
5625 struct mlx5_flow_split_info *flow_split_info,
5626 struct rte_flow_error *error)
5628 struct mlx5_priv *priv = dev->data->dev_private;
5629 struct mlx5_dev_config *config = &priv->config;
5630 const struct rte_flow_action *qrss = NULL;
5631 struct rte_flow_action *ext_actions = NULL;
5632 struct mlx5_flow *dev_flow = NULL;
5633 uint32_t qrss_id = 0;
5640 /* Check whether extensive metadata feature is engaged. */
5641 if (!config->dv_flow_en ||
5642 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5643 !mlx5_flow_ext_mreg_supported(dev))
5644 return flow_create_split_inner(dev, flow, NULL, attr, items,
5645 actions, flow_split_info, error);
5646 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5649 /* Exclude hairpin flows from splitting. */
5650 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5651 const struct rte_flow_action_queue *queue;
5654 if (mlx5_rxq_get_type(dev, queue->index) ==
5655 MLX5_RXQ_TYPE_HAIRPIN)
5657 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5658 const struct rte_flow_action_rss *rss;
5661 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5662 MLX5_RXQ_TYPE_HAIRPIN)
5667 /* Check if it is in meter suffix table. */
5668 mtr_sfx = attr->group == (attr->transfer ?
5669 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5670 MLX5_FLOW_TABLE_LEVEL_METER);
5672 * Q/RSS action on NIC Rx should be split in order to pass by
5673 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5674 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5676 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5677 sizeof(struct rte_flow_action_set_tag) +
5678 sizeof(struct rte_flow_action_jump);
5679 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5682 return rte_flow_error_set(error, ENOMEM,
5683 RTE_FLOW_ERROR_TYPE_ACTION,
5684 NULL, "no memory to split "
5687 * If we are the suffix flow of meter, tag already exist.
5688 * Set the tag action to void.
5691 ext_actions[qrss - actions].type =
5692 RTE_FLOW_ACTION_TYPE_VOID;
5694 ext_actions[qrss - actions].type =
5695 (enum rte_flow_action_type)
5696 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5698 * Create the new actions list with removed Q/RSS action
5699 * and appended set tag and jump to register copy table
5700 * (RX_CP_TBL). We should preallocate unique tag ID here
5701 * in advance, because it is needed for set tag action.
5703 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5704 qrss, actions_n, error);
5705 if (!mtr_sfx && !qrss_id) {
5709 } else if (attr->egress && !attr->transfer) {
5711 * All the actions on NIC Tx should have a metadata register
5712 * copy action to copy reg_a from WQE to reg_c[meta]
5714 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5715 sizeof(struct mlx5_flow_action_copy_mreg);
5716 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5719 return rte_flow_error_set(error, ENOMEM,
5720 RTE_FLOW_ERROR_TYPE_ACTION,
5721 NULL, "no memory to split "
5723 /* Create the action list appended with copy register. */
5724 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5725 actions_n, error, encap_idx);
5729 /* Add the unmodified original or prefix subflow. */
5730 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5731 items, ext_actions ? ext_actions :
5732 actions, flow_split_info, error);
5735 MLX5_ASSERT(dev_flow);
5737 const struct rte_flow_attr q_attr = {
5738 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5741 /* Internal PMD action to set register. */
5742 struct mlx5_rte_flow_item_tag q_tag_spec = {
5746 struct rte_flow_item q_items[] = {
5748 .type = (enum rte_flow_item_type)
5749 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5750 .spec = &q_tag_spec,
5755 .type = RTE_FLOW_ITEM_TYPE_END,
5758 struct rte_flow_action q_actions[] = {
5764 .type = RTE_FLOW_ACTION_TYPE_END,
5767 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5770 * Configure the tag item only if there is no meter subflow.
5771 * Since tag is already marked in the meter suffix subflow
5772 * we can just use the meter suffix items as is.
5775 /* Not meter subflow. */
5776 MLX5_ASSERT(!mtr_sfx);
5778 * Put unique id in prefix flow due to it is destroyed
5779 * after suffix flow and id will be freed after there
5780 * is no actual flows with this id and identifier
5781 * reallocation becomes possible (for example, for
5782 * other flows in other threads).
5784 dev_flow->handle->split_flow_id = qrss_id;
5785 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5789 q_tag_spec.id = ret;
5792 /* Add suffix subflow to execute Q/RSS. */
5793 flow_split_info->prefix_layers = layers;
5794 flow_split_info->prefix_mark = 0;
5795 ret = flow_create_split_inner(dev, flow, &dev_flow,
5796 &q_attr, mtr_sfx ? items :
5798 flow_split_info, error);
5801 /* qrss ID should be freed if failed. */
5803 MLX5_ASSERT(dev_flow);
5808 * We do not destroy the partially created sub_flows in case of error.
5809 * These ones are included into parent flow list and will be destroyed
5810 * by flow_drv_destroy.
5812 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5814 mlx5_free(ext_actions);
5819 * Create meter internal drop flow with the original pattern.
5822 * Pointer to Ethernet device.
5824 * Parent flow structure pointer.
5826 * Flow rule attributes.
5828 * Pattern specification (list terminated by the END pattern item).
5829 * @param[in] flow_split_info
5830 * Pointer to flow split info structure.
5832 * Pointer to flow meter structure.
5834 * Perform verbose error reporting if not NULL.
5836 * 0 on success, negative value otherwise
5839 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5840 struct rte_flow *flow,
5841 const struct rte_flow_attr *attr,
5842 const struct rte_flow_item items[],
5843 struct mlx5_flow_split_info *flow_split_info,
5844 struct mlx5_flow_meter_info *fm,
5845 struct rte_flow_error *error)
5847 struct mlx5_flow *dev_flow = NULL;
5848 struct rte_flow_attr drop_attr = *attr;
5849 struct rte_flow_action drop_actions[3];
5850 struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5852 MLX5_ASSERT(fm->drop_cnt);
5853 drop_actions[0].type =
5854 (enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5855 drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5856 drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5857 drop_actions[1].conf = NULL;
5858 drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5859 drop_actions[2].conf = NULL;
5860 drop_split_info.external = false;
5861 drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5862 drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5863 drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5864 return flow_create_split_inner(dev, flow, &dev_flow,
5865 &drop_attr, items, drop_actions,
5866 &drop_split_info, error);
5870 * The splitting for meter feature.
5872 * - The meter flow will be split to two flows as prefix and
5873 * suffix flow. The packets make sense only it pass the prefix
5876 * - Reg_C_5 is used for the packet to match betweend prefix and
5880 * Pointer to Ethernet device.
5882 * Parent flow structure pointer.
5884 * Flow rule attributes.
5886 * Pattern specification (list terminated by the END pattern item).
5887 * @param[in] actions
5888 * Associated actions (list terminated by the END action).
5889 * @param[in] flow_split_info
5890 * Pointer to flow split info structure.
5892 * Perform verbose error reporting if not NULL.
5894 * 0 on success, negative value otherwise
5897 flow_create_split_meter(struct rte_eth_dev *dev,
5898 struct rte_flow *flow,
5899 const struct rte_flow_attr *attr,
5900 const struct rte_flow_item items[],
5901 const struct rte_flow_action actions[],
5902 struct mlx5_flow_split_info *flow_split_info,
5903 struct rte_flow_error *error)
5905 struct mlx5_priv *priv = dev->data->dev_private;
5906 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5907 struct rte_flow_action *sfx_actions = NULL;
5908 struct rte_flow_action *pre_actions = NULL;
5909 struct rte_flow_item *sfx_items = NULL;
5910 struct mlx5_flow *dev_flow = NULL;
5911 struct rte_flow_attr sfx_attr = *attr;
5912 struct mlx5_flow_meter_info *fm = NULL;
5913 uint8_t skip_scale_restore;
5914 bool has_mtr = false;
5915 bool has_modify = false;
5916 bool set_mtr_reg = true;
5917 bool is_mtr_hierarchy = false;
5918 uint32_t meter_id = 0;
5919 uint32_t mtr_idx = 0;
5920 uint32_t mtr_flow_id = 0;
5927 actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5928 &has_modify, &meter_id);
5931 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5933 return rte_flow_error_set(error, EINVAL,
5934 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5935 NULL, "Meter not found.");
5937 fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5939 return rte_flow_error_set(error, EINVAL,
5940 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5941 NULL, "Meter not found.");
5942 ret = mlx5_flow_meter_attach(priv, fm,
5946 flow->meter = mtr_idx;
5950 if (!fm->def_policy) {
5951 wks->policy = mlx5_flow_meter_policy_find(dev,
5954 MLX5_ASSERT(wks->policy);
5955 if (wks->policy->is_hierarchy) {
5957 mlx5_flow_meter_hierarchy_get_final_policy(dev,
5959 if (!wks->final_policy)
5960 return rte_flow_error_set(error,
5962 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5963 "Failed to find terminal policy of hierarchy.");
5964 is_mtr_hierarchy = true;
5968 * If it isn't default-policy Meter, and
5969 * 1. There's no action in flow to change
5970 * packet (modify/encap/decap etc.), OR
5971 * 2. No drop count needed for this meter.
5972 * 3. It's not meter hierarchy.
5973 * Then no need to use regC to save meter id anymore.
5975 if (!fm->def_policy && !is_mtr_hierarchy &&
5976 (!has_modify || !fm->drop_cnt))
5977 set_mtr_reg = false;
5978 /* Prefix actions: meter, decap, encap, tag, jump, end. */
5979 act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5980 sizeof(struct mlx5_rte_flow_action_set_tag);
5981 /* Suffix items: tag, vlan, port id, end. */
5982 #define METER_SUFFIX_ITEM 4
5983 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5984 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5985 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5988 return rte_flow_error_set(error, ENOMEM,
5989 RTE_FLOW_ERROR_TYPE_ACTION,
5990 NULL, "no memory to split "
5992 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5994 /* There's no suffix flow for meter of non-default policy. */
5995 if (!fm->def_policy)
5996 pre_actions = sfx_actions + 1;
5998 pre_actions = sfx_actions + actions_n;
5999 ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6000 items, sfx_items, actions,
6001 sfx_actions, pre_actions,
6002 (set_mtr_reg ? &mtr_flow_id : NULL),
6008 /* Add the prefix subflow. */
6009 flow_split_info->prefix_mark = 0;
6010 skip_scale_restore = flow_split_info->skip_scale;
6011 flow_split_info->skip_scale |=
6012 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6013 ret = flow_create_split_inner(dev, flow, &dev_flow,
6014 attr, items, pre_actions,
6015 flow_split_info, error);
6016 flow_split_info->skip_scale = skip_scale_restore;
6019 mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6024 dev_flow->handle->split_flow_id = mtr_flow_id;
6025 dev_flow->handle->is_meter_flow_id = 1;
6027 if (!fm->def_policy) {
6028 if (!set_mtr_reg && fm->drop_cnt)
6030 flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6036 /* Setting the sfx group atrr. */
6037 sfx_attr.group = sfx_attr.transfer ?
6038 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6039 MLX5_FLOW_TABLE_LEVEL_METER;
6040 flow_split_info->prefix_layers =
6041 flow_get_prefix_layer_flags(dev_flow);
6042 flow_split_info->prefix_mark = dev_flow->handle->mark;
6043 flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6045 /* Add the prefix subflow. */
6046 ret = flow_create_split_metadata(dev, flow,
6047 &sfx_attr, sfx_items ?
6049 sfx_actions ? sfx_actions : actions,
6050 flow_split_info, error);
6053 mlx5_free(sfx_actions);
6058 * The splitting for sample feature.
6060 * Once Sample action is detected in the action list, the flow actions should
6061 * be split into prefix sub flow and suffix sub flow.
6063 * The original items remain in the prefix sub flow, all actions preceding the
6064 * sample action and the sample action itself will be copied to the prefix
6065 * sub flow, the actions following the sample action will be copied to the
6066 * suffix sub flow, Queue action always be located in the suffix sub flow.
6068 * In order to make the packet from prefix sub flow matches with suffix sub
6069 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6070 * flow uses tag item with the unique flow id.
6073 * Pointer to Ethernet device.
6075 * Parent flow structure pointer.
6077 * Flow rule attributes.
6079 * Pattern specification (list terminated by the END pattern item).
6080 * @param[in] actions
6081 * Associated actions (list terminated by the END action).
6082 * @param[in] flow_split_info
6083 * Pointer to flow split info structure.
6085 * Perform verbose error reporting if not NULL.
6087 * 0 on success, negative value otherwise
6090 flow_create_split_sample(struct rte_eth_dev *dev,
6091 struct rte_flow *flow,
6092 const struct rte_flow_attr *attr,
6093 const struct rte_flow_item items[],
6094 const struct rte_flow_action actions[],
6095 struct mlx5_flow_split_info *flow_split_info,
6096 struct rte_flow_error *error)
6098 struct mlx5_priv *priv = dev->data->dev_private;
6099 struct rte_flow_action *sfx_actions = NULL;
6100 struct rte_flow_action *pre_actions = NULL;
6101 struct rte_flow_item *sfx_items = NULL;
6102 struct mlx5_flow *dev_flow = NULL;
6103 struct rte_flow_attr sfx_attr = *attr;
6104 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6105 struct mlx5_flow_dv_sample_resource *sample_res;
6106 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6107 struct mlx5_flow_tbl_resource *sfx_tbl;
6111 uint32_t fdb_tx = 0;
6114 int sample_action_pos;
6115 int qrss_action_pos;
6117 int modify_after_mirror = 0;
6118 uint16_t jump_table = 0;
6119 const uint32_t next_ft_step = 1;
6122 if (priv->sampler_en)
6123 actions_n = flow_check_match_action(actions, attr,
6124 RTE_FLOW_ACTION_TYPE_SAMPLE,
6125 &sample_action_pos, &qrss_action_pos,
6126 &modify_after_mirror);
6128 /* The prefix actions must includes sample, tag, end. */
6129 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6130 + sizeof(struct mlx5_rte_flow_action_set_tag);
6131 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6132 sizeof(struct mlx5_rte_flow_item_tag) * 2;
6133 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6134 item_size), 0, SOCKET_ID_ANY);
6136 return rte_flow_error_set(error, ENOMEM,
6137 RTE_FLOW_ERROR_TYPE_ACTION,
6138 NULL, "no memory to split "
6140 /* The representor_id is UINT16_MAX for uplink. */
6141 fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6143 * When reg_c_preserve is set, metadata registers Cx preserve
6144 * their value even through packet duplication.
6146 add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6148 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6150 if (modify_after_mirror)
6151 jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6153 pre_actions = sfx_actions + actions_n;
6154 tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6155 actions, sfx_actions,
6156 pre_actions, actions_n,
6158 qrss_action_pos, jump_table,
6160 if (tag_id < 0 || (add_tag && !tag_id)) {
6164 if (modify_after_mirror)
6165 flow_split_info->skip_scale =
6166 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6167 /* Add the prefix subflow. */
6168 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6170 flow_split_info, error);
6175 dev_flow->handle->split_flow_id = tag_id;
6176 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6177 if (!modify_after_mirror) {
6178 /* Set the sfx group attr. */
6179 sample_res = (struct mlx5_flow_dv_sample_resource *)
6180 dev_flow->dv.sample_res;
6181 sfx_tbl = (struct mlx5_flow_tbl_resource *)
6182 sample_res->normal_path_tbl;
6183 sfx_tbl_data = container_of(sfx_tbl,
6184 struct mlx5_flow_tbl_data_entry,
6186 sfx_attr.group = sfx_attr.transfer ?
6187 (sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6189 MLX5_ASSERT(attr->transfer);
6190 sfx_attr.group = jump_table;
6192 flow_split_info->prefix_layers =
6193 flow_get_prefix_layer_flags(dev_flow);
6194 flow_split_info->prefix_mark = dev_flow->handle->mark;
6195 /* Suffix group level already be scaled with factor, set
6196 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6197 * again in translation.
6199 flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6202 /* Add the suffix subflow. */
6203 ret = flow_create_split_meter(dev, flow, &sfx_attr,
6204 sfx_items ? sfx_items : items,
6205 sfx_actions ? sfx_actions : actions,
6206 flow_split_info, error);
6209 mlx5_free(sfx_actions);
6214 * Split the flow to subflow set. The splitters might be linked
6215 * in the chain, like this:
6216 * flow_create_split_outer() calls:
6217 * flow_create_split_meter() calls:
6218 * flow_create_split_metadata(meter_subflow_0) calls:
6219 * flow_create_split_inner(metadata_subflow_0)
6220 * flow_create_split_inner(metadata_subflow_1)
6221 * flow_create_split_inner(metadata_subflow_2)
6222 * flow_create_split_metadata(meter_subflow_1) calls:
6223 * flow_create_split_inner(metadata_subflow_0)
6224 * flow_create_split_inner(metadata_subflow_1)
6225 * flow_create_split_inner(metadata_subflow_2)
6227 * This provide flexible way to add new levels of flow splitting.
6228 * The all of successfully created subflows are included to the
6229 * parent flow dev_flow list.
6232 * Pointer to Ethernet device.
6234 * Parent flow structure pointer.
6236 * Flow rule attributes.
6238 * Pattern specification (list terminated by the END pattern item).
6239 * @param[in] actions
6240 * Associated actions (list terminated by the END action).
6241 * @param[in] flow_split_info
6242 * Pointer to flow split info structure.
6244 * Perform verbose error reporting if not NULL.
6246 * 0 on success, negative value otherwise
6249 flow_create_split_outer(struct rte_eth_dev *dev,
6250 struct rte_flow *flow,
6251 const struct rte_flow_attr *attr,
6252 const struct rte_flow_item items[],
6253 const struct rte_flow_action actions[],
6254 struct mlx5_flow_split_info *flow_split_info,
6255 struct rte_flow_error *error)
6259 ret = flow_create_split_sample(dev, flow, attr, items,
6260 actions, flow_split_info, error);
6261 MLX5_ASSERT(ret <= 0);
6265 static inline struct mlx5_flow_tunnel *
6266 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6268 struct mlx5_flow_tunnel *tunnel;
6270 #pragma GCC diagnostic push
6271 #pragma GCC diagnostic ignored "-Wcast-qual"
6272 tunnel = (typeof(tunnel))flow->tunnel;
6273 #pragma GCC diagnostic pop
6279 * Adjust flow RSS workspace if needed.
6282 * Pointer to thread flow work space.
6284 * Pointer to RSS descriptor.
6285 * @param[in] nrssq_num
6286 * New RSS queue number.
6289 * 0 on success, -1 otherwise and rte_errno is set.
6292 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6293 struct mlx5_flow_rss_desc *rss_desc,
6296 if (likely(nrssq_num <= wks->rssq_num))
6298 rss_desc->queue = realloc(rss_desc->queue,
6299 sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6300 if (!rss_desc->queue) {
6304 wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6309 * Create a flow and add it to @p list.
6312 * Pointer to Ethernet device.
6314 * Pointer to a TAILQ flow list. If this parameter NULL,
6315 * no list insertion occurred, flow is just created,
6316 * this is caller's responsibility to track the
6319 * Flow rule attributes.
6321 * Pattern specification (list terminated by the END pattern item).
6322 * @param[in] actions
6323 * Associated actions (list terminated by the END action).
6324 * @param[in] external
6325 * This flow rule is created by request external to PMD.
6327 * Perform verbose error reporting if not NULL.
6330 * A flow index on success, 0 otherwise and rte_errno is set.
6333 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6334 const struct rte_flow_attr *attr,
6335 const struct rte_flow_item items[],
6336 const struct rte_flow_action original_actions[],
6337 bool external, struct rte_flow_error *error)
6339 struct mlx5_priv *priv = dev->data->dev_private;
6340 struct rte_flow *flow = NULL;
6341 struct mlx5_flow *dev_flow;
6342 const struct rte_flow_action_rss *rss = NULL;
6343 struct mlx5_translated_action_handle
6344 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6345 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6347 struct mlx5_flow_expand_rss buf;
6348 uint8_t buffer[4096];
6351 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6352 uint8_t buffer[2048];
6355 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6356 uint8_t buffer[2048];
6357 } actions_hairpin_tx;
6359 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6360 uint8_t buffer[2048];
6362 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6363 struct mlx5_flow_rss_desc *rss_desc;
6364 const struct rte_flow_action *p_actions_rx;
6368 struct rte_flow_attr attr_tx = { .priority = 0 };
6369 const struct rte_flow_action *actions;
6370 struct rte_flow_action *translated_actions = NULL;
6371 struct mlx5_flow_tunnel *tunnel;
6372 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6373 struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6374 struct mlx5_flow_split_info flow_split_info = {
6375 .external = !!external,
6385 rss_desc = &wks->rss_desc;
6386 ret = flow_action_handles_translate(dev, original_actions,
6389 &translated_actions, error);
6391 MLX5_ASSERT(translated_actions == NULL);
6394 actions = translated_actions ? translated_actions : original_actions;
6395 p_actions_rx = actions;
6396 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6397 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6398 external, hairpin_flow, error);
6400 goto error_before_hairpin_split;
6401 flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6404 goto error_before_hairpin_split;
6406 if (hairpin_flow > 0) {
6407 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6409 goto error_before_hairpin_split;
6411 flow_hairpin_split(dev, actions, actions_rx.actions,
6412 actions_hairpin_tx.actions, items_tx.items,
6414 p_actions_rx = actions_rx.actions;
6416 flow_split_info.flow_idx = idx;
6417 flow->drv_type = flow_get_drv_type(dev, attr);
6418 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6419 flow->drv_type < MLX5_FLOW_TYPE_MAX);
6420 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6421 /* RSS Action only works on NIC RX domain */
6422 if (attr->ingress && !attr->transfer)
6423 rss = flow_get_rss_action(dev, p_actions_rx);
6425 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6428 * The following information is required by
6429 * mlx5_flow_hashfields_adjust() in advance.
6431 rss_desc->level = rss->level;
6432 /* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6433 rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6435 flow->dev_handles = 0;
6436 if (rss && rss->types) {
6437 unsigned int graph_root;
6439 graph_root = find_graph_root(rss->level);
6440 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6442 mlx5_support_expansion, graph_root);
6443 MLX5_ASSERT(ret > 0 &&
6444 (unsigned int)ret < sizeof(expand_buffer.buffer));
6445 if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6446 for (i = 0; i < buf->entries; ++i)
6447 mlx5_dbg__print_pattern(buf->entry[i].pattern);
6451 buf->entry[0].pattern = (void *)(uintptr_t)items;
6453 rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6455 for (i = 0; i < buf->entries; ++i) {
6456 /* Initialize flow split data. */
6457 flow_split_info.prefix_layers = 0;
6458 flow_split_info.prefix_mark = 0;
6459 flow_split_info.skip_scale = 0;
6461 * The splitter may create multiple dev_flows,
6462 * depending on configuration. In the simplest
6463 * case it just creates unmodified original flow.
6465 ret = flow_create_split_outer(dev, flow, attr,
6466 buf->entry[i].pattern,
6467 p_actions_rx, &flow_split_info,
6471 if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6472 ret = flow_tunnel_add_default_miss(dev, flow, attr,
6475 wks->flows[0].tunnel,
6479 mlx5_free(default_miss_ctx.queue);
6484 /* Create the tx flow. */
6486 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6487 attr_tx.ingress = 0;
6489 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6490 actions_hairpin_tx.actions,
6494 dev_flow->flow = flow;
6495 dev_flow->external = 0;
6496 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6497 dev_flow->handle, next);
6498 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6500 actions_hairpin_tx.actions, error);
6505 * Update the metadata register copy table. If extensive
6506 * metadata feature is enabled and registers are supported
6507 * we might create the extra rte_flow for each unique
6508 * MARK/FLAG action ID.
6510 * The table is updated for ingress Flows only, because
6511 * the egress Flows belong to the different device and
6512 * copy table should be updated in peer NIC Rx domain.
6514 if (attr->ingress &&
6515 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6516 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6521 * If the flow is external (from application) OR device is started,
6522 * OR mreg discover, then apply immediately.
6524 if (external || dev->data->dev_started ||
6525 (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6526 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6527 ret = flow_drv_apply(dev, flow, error);
6532 flow_rxq_flags_set(dev, flow);
6533 rte_free(translated_actions);
6534 tunnel = flow_tunnel_from_rule(wks->flows);
6537 flow->tunnel_id = tunnel->tunnel_id;
6538 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6539 mlx5_free(default_miss_ctx.queue);
6541 mlx5_flow_pop_thread_workspace();
6545 ret = rte_errno; /* Save rte_errno before cleanup. */
6546 flow_mreg_del_copy_action(dev, flow);
6547 flow_drv_destroy(dev, flow);
6548 if (rss_desc->shared_rss)
6549 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6551 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6552 rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6553 mlx5_ipool_free(priv->flows[type], idx);
6554 rte_errno = ret; /* Restore rte_errno. */
6557 mlx5_flow_pop_thread_workspace();
6558 error_before_hairpin_split:
6559 rte_free(translated_actions);
6564 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6565 * incoming packets to table 1.
6567 * Other flow rules, requested for group n, will be created in
6568 * e-switch table n+1.
6569 * Jump action to e-switch group n will be created to group n+1.
6571 * Used when working in switchdev mode, to utilise advantages of table 1
6575 * Pointer to Ethernet device.
6578 * Pointer to flow on success, NULL otherwise and rte_errno is set.
6581 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6583 const struct rte_flow_attr attr = {
6590 const struct rte_flow_item pattern = {
6591 .type = RTE_FLOW_ITEM_TYPE_END,
6593 struct rte_flow_action_jump jump = {
6596 const struct rte_flow_action actions[] = {
6598 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6602 .type = RTE_FLOW_ACTION_TYPE_END,
6605 struct rte_flow_error error;
6607 return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6609 actions, false, &error);
6613 * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6614 * and sq number, directs all packets to peer vport.
6617 * Pointer to Ethernet device.
6622 * Flow ID on success, 0 otherwise and rte_errno is set.
6625 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6627 struct rte_flow_attr attr = {
6629 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6634 struct rte_flow_item_port_id port_spec = {
6635 .id = MLX5_PORT_ESW_MGR,
6637 struct mlx5_rte_flow_item_tx_queue txq_spec = {
6640 struct rte_flow_item pattern[] = {
6642 .type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6646 .type = (enum rte_flow_item_type)
6647 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6651 .type = RTE_FLOW_ITEM_TYPE_END,
6654 struct rte_flow_action_jump jump = {
6657 struct rte_flow_action_port_id port = {
6658 .id = dev->data->port_id,
6660 struct rte_flow_action actions[] = {
6662 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6666 .type = RTE_FLOW_ACTION_TYPE_END,
6669 struct rte_flow_error error;
6672 * Creates group 0, highest priority jump flow.
6673 * Matches txq to bypass kernel packets.
6675 if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6676 false, &error) == 0)
6678 /* Create group 1, lowest priority redirect flow for txq. */
6680 actions[0].conf = &port;
6681 actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6682 return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6683 actions, false, &error);
6687 * Validate a flow supported by the NIC.
6689 * @see rte_flow_validate()
6693 mlx5_flow_validate(struct rte_eth_dev *dev,
6694 const struct rte_flow_attr *attr,
6695 const struct rte_flow_item items[],
6696 const struct rte_flow_action original_actions[],
6697 struct rte_flow_error *error)
6700 struct mlx5_translated_action_handle
6701 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6702 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6703 const struct rte_flow_action *actions;
6704 struct rte_flow_action *translated_actions = NULL;
6705 int ret = flow_action_handles_translate(dev, original_actions,
6708 &translated_actions, error);
6712 actions = translated_actions ? translated_actions : original_actions;
6713 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6714 ret = flow_drv_validate(dev, attr, items, actions,
6715 true, hairpin_flow, error);
6716 rte_free(translated_actions);
6723 * @see rte_flow_create()
6727 mlx5_flow_create(struct rte_eth_dev *dev,
6728 const struct rte_flow_attr *attr,
6729 const struct rte_flow_item items[],
6730 const struct rte_flow_action actions[],
6731 struct rte_flow_error *error)
6734 * If the device is not started yet, it is not allowed to created a
6735 * flow from application. PMD default flows and traffic control flows
6738 if (unlikely(!dev->data->dev_started)) {
6739 DRV_LOG(DEBUG, "port %u is not started when "
6740 "inserting a flow", dev->data->port_id);
6741 rte_flow_error_set(error, ENODEV,
6742 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6744 "port not started");
6748 return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6749 attr, items, actions,
6754 * Destroy a flow in a list.
6757 * Pointer to Ethernet device.
6758 * @param[in] flow_idx
6759 * Index of flow to destroy.
6762 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6765 struct mlx5_priv *priv = dev->data->dev_private;
6766 struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6770 MLX5_ASSERT(flow->type == type);
6772 * Update RX queue flags only if port is started, otherwise it is
6775 if (dev->data->dev_started)
6776 flow_rxq_flags_trim(dev, flow);
6777 flow_drv_destroy(dev, flow);
6779 struct mlx5_flow_tunnel *tunnel;
6781 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6783 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6784 mlx5_flow_tunnel_free(dev, tunnel);
6786 flow_mreg_del_copy_action(dev, flow);
6787 mlx5_ipool_free(priv->flows[type], flow_idx);
6791 * Destroy all flows.
6794 * Pointer to Ethernet device.
6796 * Flow type to be flushed.
6798 * If flushing is called avtively.
6801 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6804 struct mlx5_priv *priv = dev->data->dev_private;
6805 uint32_t num_flushed = 0, fidx = 1;
6806 struct rte_flow *flow;
6808 MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6809 flow_list_destroy(dev, type, fidx);
6813 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6814 dev->data->port_id, num_flushed);
6819 * Stop all default actions for flows.
6822 * Pointer to Ethernet device.
6825 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6827 flow_mreg_del_default_copy_action(dev);
6828 flow_rxq_flags_clear(dev);
6832 * Start all default actions for flows.
6835 * Pointer to Ethernet device.
6837 * 0 on success, a negative errno value otherwise and rte_errno is set.
6840 mlx5_flow_start_default(struct rte_eth_dev *dev)
6842 struct rte_flow_error error;
6844 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6845 return flow_mreg_add_default_copy_action(dev, &error);
6849 * Release key of thread specific flow workspace data.
6852 flow_release_workspace(void *data)
6854 struct mlx5_flow_workspace *wks = data;
6855 struct mlx5_flow_workspace *next;
6859 free(wks->rss_desc.queue);
6866 * Get thread specific current flow workspace.
6868 * @return pointer to thread specific flow workspace data, NULL on error.
6870 struct mlx5_flow_workspace*
6871 mlx5_flow_get_thread_workspace(void)
6873 struct mlx5_flow_workspace *data;
6875 data = mlx5_flow_os_get_specific_workspace();
6876 MLX5_ASSERT(data && data->inuse);
6877 if (!data || !data->inuse)
6878 DRV_LOG(ERR, "flow workspace not initialized.");
6883 * Allocate and init new flow workspace.
6885 * @return pointer to flow workspace data, NULL on error.
6887 static struct mlx5_flow_workspace*
6888 flow_alloc_thread_workspace(void)
6890 struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6893 DRV_LOG(ERR, "Failed to allocate flow workspace "
6897 data->rss_desc.queue = calloc(1,
6898 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6899 if (!data->rss_desc.queue)
6901 data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6904 if (data->rss_desc.queue)
6905 free(data->rss_desc.queue);
6911 * Get new thread specific flow workspace.
6913 * If current workspace inuse, create new one and set as current.
6915 * @return pointer to thread specific flow workspace data, NULL on error.
6917 static struct mlx5_flow_workspace*
6918 mlx5_flow_push_thread_workspace(void)
6920 struct mlx5_flow_workspace *curr;
6921 struct mlx5_flow_workspace *data;
6923 curr = mlx5_flow_os_get_specific_workspace();
6925 data = flow_alloc_thread_workspace();
6928 } else if (!curr->inuse) {
6930 } else if (curr->next) {
6933 data = flow_alloc_thread_workspace();
6941 /* Set as current workspace */
6942 if (mlx5_flow_os_set_specific_workspace(data))
6943 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6948 * Close current thread specific flow workspace.
6950 * If previous workspace available, set it as current.
6952 * @return pointer to thread specific flow workspace data, NULL on error.
6955 mlx5_flow_pop_thread_workspace(void)
6957 struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6962 DRV_LOG(ERR, "Failed to close unused flow workspace.");
6968 if (mlx5_flow_os_set_specific_workspace(data->prev))
6969 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6973 * Verify the flow list is empty
6976 * Pointer to Ethernet device.
6978 * @return the number of flows not released.
6981 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
6983 struct mlx5_priv *priv = dev->data->dev_private;
6984 struct rte_flow *flow;
6988 for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
6989 MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
6990 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6991 dev->data->port_id, (void *)flow);
6999 * Enable default hairpin egress flow.
7002 * Pointer to Ethernet device.
7007 * 0 on success, a negative errno value otherwise and rte_errno is set.
7010 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7013 const struct rte_flow_attr attr = {
7017 struct mlx5_rte_flow_item_tx_queue queue_spec = {
7020 struct mlx5_rte_flow_item_tx_queue queue_mask = {
7021 .queue = UINT32_MAX,
7023 struct rte_flow_item items[] = {
7025 .type = (enum rte_flow_item_type)
7026 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7027 .spec = &queue_spec,
7029 .mask = &queue_mask,
7032 .type = RTE_FLOW_ITEM_TYPE_END,
7035 struct rte_flow_action_jump jump = {
7036 .group = MLX5_HAIRPIN_TX_TABLE,
7038 struct rte_flow_action actions[2];
7040 struct rte_flow_error error;
7042 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7043 actions[0].conf = &jump;
7044 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7045 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7046 &attr, items, actions, false, &error);
7049 "Failed to create ctrl flow: rte_errno(%d),"
7050 " type(%d), message(%s)",
7051 rte_errno, error.type,
7052 error.message ? error.message : " (no stated reason)");
7059 * Enable a control flow configured from the control plane.
7062 * Pointer to Ethernet device.
7064 * An Ethernet flow spec to apply.
7066 * An Ethernet flow mask to apply.
7068 * A VLAN flow spec to apply.
7070 * A VLAN flow mask to apply.
7073 * 0 on success, a negative errno value otherwise and rte_errno is set.
7076 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7077 struct rte_flow_item_eth *eth_spec,
7078 struct rte_flow_item_eth *eth_mask,
7079 struct rte_flow_item_vlan *vlan_spec,
7080 struct rte_flow_item_vlan *vlan_mask)
7082 struct mlx5_priv *priv = dev->data->dev_private;
7083 const struct rte_flow_attr attr = {
7085 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7087 struct rte_flow_item items[] = {
7089 .type = RTE_FLOW_ITEM_TYPE_ETH,
7095 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7096 RTE_FLOW_ITEM_TYPE_END,
7102 .type = RTE_FLOW_ITEM_TYPE_END,
7105 uint16_t queue[priv->reta_idx_n];
7106 struct rte_flow_action_rss action_rss = {
7107 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7109 .types = priv->rss_conf.rss_hf,
7110 .key_len = priv->rss_conf.rss_key_len,
7111 .queue_num = priv->reta_idx_n,
7112 .key = priv->rss_conf.rss_key,
7115 struct rte_flow_action actions[] = {
7117 .type = RTE_FLOW_ACTION_TYPE_RSS,
7118 .conf = &action_rss,
7121 .type = RTE_FLOW_ACTION_TYPE_END,
7125 struct rte_flow_error error;
7128 if (!priv->reta_idx_n || !priv->rxqs_n) {
7131 if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7132 action_rss.types = 0;
7133 for (i = 0; i != priv->reta_idx_n; ++i)
7134 queue[i] = (*priv->reta_idx)[i];
7135 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7136 &attr, items, actions, false, &error);
7143 * Enable a flow control configured from the control plane.
7146 * Pointer to Ethernet device.
7148 * An Ethernet flow spec to apply.
7150 * An Ethernet flow mask to apply.
7153 * 0 on success, a negative errno value otherwise and rte_errno is set.
7156 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7157 struct rte_flow_item_eth *eth_spec,
7158 struct rte_flow_item_eth *eth_mask)
7160 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7164 * Create default miss flow rule matching lacp traffic
7167 * Pointer to Ethernet device.
7169 * An Ethernet flow spec to apply.
7172 * 0 on success, a negative errno value otherwise and rte_errno is set.
7175 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7178 * The LACP matching is done by only using ether type since using
7179 * a multicast dst mac causes kernel to give low priority to this flow.
7181 static const struct rte_flow_item_eth lacp_spec = {
7182 .type = RTE_BE16(0x8809),
7184 static const struct rte_flow_item_eth lacp_mask = {
7187 const struct rte_flow_attr attr = {
7190 struct rte_flow_item items[] = {
7192 .type = RTE_FLOW_ITEM_TYPE_ETH,
7197 .type = RTE_FLOW_ITEM_TYPE_END,
7200 struct rte_flow_action actions[] = {
7202 .type = (enum rte_flow_action_type)
7203 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7206 .type = RTE_FLOW_ACTION_TYPE_END,
7209 struct rte_flow_error error;
7210 uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7211 &attr, items, actions,
7222 * @see rte_flow_destroy()
7226 mlx5_flow_destroy(struct rte_eth_dev *dev,
7227 struct rte_flow *flow,
7228 struct rte_flow_error *error __rte_unused)
7230 flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7231 (uintptr_t)(void *)flow);
7236 * Destroy all flows.
7238 * @see rte_flow_flush()
7242 mlx5_flow_flush(struct rte_eth_dev *dev,
7243 struct rte_flow_error *error __rte_unused)
7245 mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7252 * @see rte_flow_isolate()
7256 mlx5_flow_isolate(struct rte_eth_dev *dev,
7258 struct rte_flow_error *error)
7260 struct mlx5_priv *priv = dev->data->dev_private;
7262 if (dev->data->dev_started) {
7263 rte_flow_error_set(error, EBUSY,
7264 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7266 "port must be stopped first");
7269 priv->isolated = !!enable;
7271 dev->dev_ops = &mlx5_dev_ops_isolate;
7273 dev->dev_ops = &mlx5_dev_ops;
7275 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7276 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7284 * @see rte_flow_query()
7288 flow_drv_query(struct rte_eth_dev *dev,
7290 const struct rte_flow_action *actions,
7292 struct rte_flow_error *error)
7294 struct mlx5_priv *priv = dev->data->dev_private;
7295 const struct mlx5_flow_driver_ops *fops;
7296 struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7298 enum mlx5_flow_drv_type ftype;
7301 return rte_flow_error_set(error, ENOENT,
7302 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7304 "invalid flow handle");
7306 ftype = flow->drv_type;
7307 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7308 fops = flow_get_drv_ops(ftype);
7310 return fops->query(dev, flow, actions, data, error);
7316 * @see rte_flow_query()
7320 mlx5_flow_query(struct rte_eth_dev *dev,
7321 struct rte_flow *flow,
7322 const struct rte_flow_action *actions,
7324 struct rte_flow_error *error)
7328 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7336 * Get rte_flow callbacks.
7339 * Pointer to Ethernet device structure.
7341 * Pointer to operation-specific structure.
7346 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7347 const struct rte_flow_ops **ops)
7349 *ops = &mlx5_flow_ops;
7354 * Validate meter policy actions.
7355 * Dispatcher for action type specific validation.
7358 * Pointer to the Ethernet device structure.
7360 * The meter policy action object to validate.
7362 * Attributes of flow to determine steering domain.
7363 * @param[out] is_rss
7365 * @param[out] domain_bitmap
7367 * @param[out] is_def_policy
7368 * Is default policy or not.
7370 * Perform verbose error reporting if not NULL. Initialized in case of
7374 * 0 on success, otherwise negative errno value.
7377 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7378 const struct rte_flow_action *actions[RTE_COLORS],
7379 struct rte_flow_attr *attr,
7381 uint8_t *domain_bitmap,
7382 uint8_t *policy_mode,
7383 struct rte_mtr_error *error)
7385 const struct mlx5_flow_driver_ops *fops;
7387 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7388 return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7389 domain_bitmap, policy_mode, error);
7393 * Destroy the meter table set.
7396 * Pointer to Ethernet device.
7397 * @param[in] mtr_policy
7398 * Meter policy struct.
7401 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7402 struct mlx5_flow_meter_policy *mtr_policy)
7404 const struct mlx5_flow_driver_ops *fops;
7406 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7407 fops->destroy_mtr_acts(dev, mtr_policy);
7411 * Create policy action, lock free,
7412 * (mutex should be acquired by caller).
7413 * Dispatcher for action type specific call.
7416 * Pointer to the Ethernet device structure.
7417 * @param[in] mtr_policy
7418 * Meter policy struct.
7420 * Action specification used to create meter actions.
7422 * Perform verbose error reporting if not NULL. Initialized in case of
7426 * 0 on success, otherwise negative errno value.
7429 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7430 struct mlx5_flow_meter_policy *mtr_policy,
7431 const struct rte_flow_action *actions[RTE_COLORS],
7432 struct rte_mtr_error *error)
7434 const struct mlx5_flow_driver_ops *fops;
7436 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7437 return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7441 * Create policy rules, lock free,
7442 * (mutex should be acquired by caller).
7443 * Dispatcher for action type specific call.
7446 * Pointer to the Ethernet device structure.
7447 * @param[in] mtr_policy
7448 * Meter policy struct.
7451 * 0 on success, -1 otherwise.
7454 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7455 struct mlx5_flow_meter_policy *mtr_policy)
7457 const struct mlx5_flow_driver_ops *fops;
7459 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7460 return fops->create_policy_rules(dev, mtr_policy);
7464 * Destroy policy rules, lock free,
7465 * (mutex should be acquired by caller).
7466 * Dispatcher for action type specific call.
7469 * Pointer to the Ethernet device structure.
7470 * @param[in] mtr_policy
7471 * Meter policy struct.
7474 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7475 struct mlx5_flow_meter_policy *mtr_policy)
7477 const struct mlx5_flow_driver_ops *fops;
7479 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7480 fops->destroy_policy_rules(dev, mtr_policy);
7484 * Destroy the default policy table set.
7487 * Pointer to Ethernet device.
7490 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7492 const struct mlx5_flow_driver_ops *fops;
7494 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7495 fops->destroy_def_policy(dev);
7499 * Destroy the default policy table set.
7502 * Pointer to Ethernet device.
7505 * 0 on success, -1 otherwise.
7508 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7510 const struct mlx5_flow_driver_ops *fops;
7512 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7513 return fops->create_def_policy(dev);
7517 * Create the needed meter and suffix tables.
7520 * Pointer to Ethernet device.
7523 * 0 on success, -1 otherwise.
7526 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7527 struct mlx5_flow_meter_info *fm,
7529 uint8_t domain_bitmap)
7531 const struct mlx5_flow_driver_ops *fops;
7533 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7534 return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7538 * Destroy the meter table set.
7541 * Pointer to Ethernet device.
7543 * Pointer to the meter table set.
7546 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7547 struct mlx5_flow_meter_info *fm)
7549 const struct mlx5_flow_driver_ops *fops;
7551 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7552 fops->destroy_mtr_tbls(dev, fm);
7556 * Destroy the global meter drop table.
7559 * Pointer to Ethernet device.
7562 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7564 const struct mlx5_flow_driver_ops *fops;
7566 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7567 fops->destroy_mtr_drop_tbls(dev);
7571 * Destroy the sub policy table with RX queue.
7574 * Pointer to Ethernet device.
7575 * @param[in] mtr_policy
7576 * Pointer to meter policy table.
7579 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7580 struct mlx5_flow_meter_policy *mtr_policy)
7582 const struct mlx5_flow_driver_ops *fops;
7584 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7585 fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7589 * Allocate the needed aso flow meter id.
7592 * Pointer to Ethernet device.
7595 * Index to aso flow meter on success, NULL otherwise.
7598 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7600 const struct mlx5_flow_driver_ops *fops;
7602 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7603 return fops->create_meter(dev);
7607 * Free the aso flow meter id.
7610 * Pointer to Ethernet device.
7611 * @param[in] mtr_idx
7612 * Index to aso flow meter to be free.
7618 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7620 const struct mlx5_flow_driver_ops *fops;
7622 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7623 fops->free_meter(dev, mtr_idx);
7627 * Allocate a counter.
7630 * Pointer to Ethernet device structure.
7633 * Index to allocated counter on success, 0 otherwise.
7636 mlx5_counter_alloc(struct rte_eth_dev *dev)
7638 const struct mlx5_flow_driver_ops *fops;
7639 struct rte_flow_attr attr = { .transfer = 0 };
7641 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7642 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7643 return fops->counter_alloc(dev);
7646 "port %u counter allocate is not supported.",
7647 dev->data->port_id);
7655 * Pointer to Ethernet device structure.
7657 * Index to counter to be free.
7660 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7662 const struct mlx5_flow_driver_ops *fops;
7663 struct rte_flow_attr attr = { .transfer = 0 };
7665 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7666 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7667 fops->counter_free(dev, cnt);
7671 "port %u counter free is not supported.",
7672 dev->data->port_id);
7676 * Query counter statistics.
7679 * Pointer to Ethernet device structure.
7681 * Index to counter to query.
7683 * Set to clear counter statistics.
7685 * The counter hits packets number to save.
7687 * The counter hits bytes number to save.
7690 * 0 on success, a negative errno value otherwise.
7693 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7694 bool clear, uint64_t *pkts, uint64_t *bytes)
7696 const struct mlx5_flow_driver_ops *fops;
7697 struct rte_flow_attr attr = { .transfer = 0 };
7699 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7700 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7701 return fops->counter_query(dev, cnt, clear, pkts, bytes);
7704 "port %u counter query is not supported.",
7705 dev->data->port_id);
7710 * Allocate a new memory for the counter values wrapped by all the needed
7714 * Pointer to mlx5_dev_ctx_shared object.
7717 * 0 on success, a negative errno value otherwise.
7720 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7722 struct mlx5_devx_mkey_attr mkey_attr;
7723 struct mlx5_counter_stats_mem_mng *mem_mng;
7724 volatile struct flow_counter_stats *raw_data;
7725 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7726 int size = (sizeof(struct flow_counter_stats) *
7727 MLX5_COUNTERS_PER_POOL +
7728 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7729 sizeof(struct mlx5_counter_stats_mem_mng);
7730 size_t pgsize = rte_mem_page_size();
7734 if (pgsize == (size_t)-1) {
7735 DRV_LOG(ERR, "Failed to get mem page size");
7739 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7744 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7745 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7746 mem_mng->umem = mlx5_os_umem_reg(sh->cdev->ctx, mem, size,
7747 IBV_ACCESS_LOCAL_WRITE);
7748 if (!mem_mng->umem) {
7753 memset(&mkey_attr, 0, sizeof(mkey_attr));
7754 mkey_attr.addr = (uintptr_t)mem;
7755 mkey_attr.size = size;
7756 mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7757 mkey_attr.pd = sh->cdev->pdn;
7758 mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7759 mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7760 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->cdev->ctx, &mkey_attr);
7762 mlx5_os_umem_dereg(mem_mng->umem);
7767 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7768 raw_data = (volatile struct flow_counter_stats *)mem;
7769 for (i = 0; i < raws_n; ++i) {
7770 mem_mng->raws[i].mem_mng = mem_mng;
7771 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7773 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7774 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7775 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7777 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7778 sh->cmng.mem_mng = mem_mng;
7783 * Set the statistic memory to the new counter pool.
7786 * Pointer to mlx5_dev_ctx_shared object.
7788 * Pointer to the pool to set the statistic memory.
7791 * 0 on success, a negative errno value otherwise.
7794 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7795 struct mlx5_flow_counter_pool *pool)
7797 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7798 /* Resize statistic memory once used out. */
7799 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7800 mlx5_flow_create_counter_stat_mem_mng(sh)) {
7801 DRV_LOG(ERR, "Cannot resize counter stat mem.");
7804 rte_spinlock_lock(&pool->sl);
7805 pool->raw = cmng->mem_mng->raws + pool->index %
7806 MLX5_CNT_CONTAINER_RESIZE;
7807 rte_spinlock_unlock(&pool->sl);
7808 pool->raw_hw = NULL;
7812 #define MLX5_POOL_QUERY_FREQ_US 1000000
7815 * Set the periodic procedure for triggering asynchronous batch queries for all
7816 * the counter pools.
7819 * Pointer to mlx5_dev_ctx_shared object.
7822 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7824 uint32_t pools_n, us;
7826 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7827 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7828 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7829 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7830 sh->cmng.query_thread_on = 0;
7831 DRV_LOG(ERR, "Cannot reinitialize query alarm");
7833 sh->cmng.query_thread_on = 1;
7838 * The periodic procedure for triggering asynchronous batch queries for all the
7839 * counter pools. This function is probably called by the host thread.
7842 * The parameter for the alarm process.
7845 mlx5_flow_query_alarm(void *arg)
7847 struct mlx5_dev_ctx_shared *sh = arg;
7849 uint16_t pool_index = sh->cmng.pool_index;
7850 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7851 struct mlx5_flow_counter_pool *pool;
7854 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7856 rte_spinlock_lock(&cmng->pool_update_sl);
7857 pool = cmng->pools[pool_index];
7858 n_valid = cmng->n_valid;
7859 rte_spinlock_unlock(&cmng->pool_update_sl);
7860 /* Set the statistic memory to the new created pool. */
7861 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7864 /* There is a pool query in progress. */
7867 LIST_FIRST(&sh->cmng.free_stat_raws);
7869 /* No free counter statistics raw memory. */
7872 * Identify the counters released between query trigger and query
7873 * handle more efficiently. The counter released in this gap period
7874 * should wait for a new round of query as the new arrived packets
7875 * will not be taken into account.
7878 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7879 MLX5_COUNTERS_PER_POOL,
7881 pool->raw_hw->mem_mng->dm->id,
7885 (uint64_t)(uintptr_t)pool);
7887 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7888 " %d", pool->min_dcs->id);
7889 pool->raw_hw = NULL;
7892 LIST_REMOVE(pool->raw_hw, next);
7893 sh->cmng.pending_queries++;
7895 if (pool_index >= n_valid)
7898 sh->cmng.pool_index = pool_index;
7899 mlx5_set_query_alarm(sh);
7903 * Check and callback event for new aged flow in the counter pool
7906 * Pointer to mlx5_dev_ctx_shared object.
7908 * Pointer to Current counter pool.
7911 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7912 struct mlx5_flow_counter_pool *pool)
7914 struct mlx5_priv *priv;
7915 struct mlx5_flow_counter *cnt;
7916 struct mlx5_age_info *age_info;
7917 struct mlx5_age_param *age_param;
7918 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7919 struct mlx5_counter_stats_raw *prev = pool->raw;
7920 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7921 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7922 uint16_t expected = AGE_CANDIDATE;
7925 pool->time_of_last_age_check = curr_time;
7926 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7927 cnt = MLX5_POOL_GET_CNT(pool, i);
7928 age_param = MLX5_CNT_TO_AGE(cnt);
7929 if (__atomic_load_n(&age_param->state,
7930 __ATOMIC_RELAXED) != AGE_CANDIDATE)
7932 if (cur->data[i].hits != prev->data[i].hits) {
7933 __atomic_store_n(&age_param->sec_since_last_hit, 0,
7937 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7939 __ATOMIC_RELAXED) <= age_param->timeout)
7942 * Hold the lock first, or if between the
7943 * state AGE_TMOUT and tailq operation the
7944 * release happened, the release procedure
7945 * may delete a non-existent tailq node.
7947 priv = rte_eth_devices[age_param->port_id].data->dev_private;
7948 age_info = GET_PORT_AGE_INFO(priv);
7949 rte_spinlock_lock(&age_info->aged_sl);
7950 if (__atomic_compare_exchange_n(&age_param->state, &expected,
7953 __ATOMIC_RELAXED)) {
7954 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7955 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7957 rte_spinlock_unlock(&age_info->aged_sl);
7959 mlx5_age_event_prepare(sh);
7963 * Handler for the HW respond about ready values from an asynchronous batch
7964 * query. This function is probably called by the host thread.
7967 * The pointer to the shared device context.
7968 * @param[in] async_id
7969 * The Devx async ID.
7971 * The status of the completion.
7974 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7975 uint64_t async_id, int status)
7977 struct mlx5_flow_counter_pool *pool =
7978 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7979 struct mlx5_counter_stats_raw *raw_to_free;
7980 uint8_t query_gen = pool->query_gen ^ 1;
7981 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7982 enum mlx5_counter_type cnt_type =
7983 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7984 MLX5_COUNTER_TYPE_ORIGIN;
7986 if (unlikely(status)) {
7987 raw_to_free = pool->raw_hw;
7989 raw_to_free = pool->raw;
7991 mlx5_flow_aging_check(sh, pool);
7992 rte_spinlock_lock(&pool->sl);
7993 pool->raw = pool->raw_hw;
7994 rte_spinlock_unlock(&pool->sl);
7995 /* Be sure the new raw counters data is updated in memory. */
7997 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7998 rte_spinlock_lock(&cmng->csl[cnt_type]);
7999 TAILQ_CONCAT(&cmng->counters[cnt_type],
8000 &pool->counters[query_gen], next);
8001 rte_spinlock_unlock(&cmng->csl[cnt_type]);
8004 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8005 pool->raw_hw = NULL;
8006 sh->cmng.pending_queries--;
8010 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8011 const struct flow_grp_info *grp_info,
8012 struct rte_flow_error *error)
8014 if (grp_info->transfer && grp_info->external &&
8015 grp_info->fdb_def_rule) {
8016 if (group == UINT32_MAX)
8017 return rte_flow_error_set
8019 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8021 "group index not supported");
8026 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8031 * Translate the rte_flow group index to HW table value.
8033 * If tunnel offload is disabled, all group ids converted to flow table
8034 * id using the standard method.
8035 * If tunnel offload is enabled, group id can be converted using the
8036 * standard or tunnel conversion method. Group conversion method
8037 * selection depends on flags in `grp_info` parameter:
8038 * - Internal (grp_info.external == 0) groups conversion uses the
8040 * - Group ids in JUMP action converted with the tunnel conversion.
8041 * - Group id in rule attribute conversion depends on a rule type and
8043 * ** non zero group attributes converted with the tunnel method
8044 * ** zero group attribute in non-tunnel rule is converted using the
8045 * standard method - there's only one root table
8046 * ** zero group attribute in steer tunnel rule is converted with the
8047 * standard method - single root table
8048 * ** zero group attribute in match tunnel rule is a special OvS
8049 * case: that value is used for portability reasons. That group
8050 * id is converted with the tunnel conversion method.
8055 * PMD tunnel offload object
8057 * rte_flow group index value.
8060 * @param[in] grp_info
8061 * flags used for conversion
8063 * Pointer to error structure.
8066 * 0 on success, a negative errno value otherwise and rte_errno is set.
8069 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8070 const struct mlx5_flow_tunnel *tunnel,
8071 uint32_t group, uint32_t *table,
8072 const struct flow_grp_info *grp_info,
8073 struct rte_flow_error *error)
8076 bool standard_translation;
8078 if (!grp_info->skip_scale && grp_info->external &&
8079 group < MLX5_MAX_TABLES_EXTERNAL)
8080 group *= MLX5_FLOW_TABLE_FACTOR;
8081 if (is_tunnel_offload_active(dev)) {
8082 standard_translation = !grp_info->external ||
8083 grp_info->std_tbl_fix;
8085 standard_translation = true;
8088 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8089 dev->data->port_id, group, grp_info->transfer,
8090 grp_info->external, grp_info->fdb_def_rule,
8091 standard_translation ? "STANDARD" : "TUNNEL");
8092 if (standard_translation)
8093 ret = flow_group_to_table(dev->data->port_id, group, table,
8096 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8103 * Discover availability of metadata reg_c's.
8105 * Iteratively use test flows to check availability.
8108 * Pointer to the Ethernet device structure.
8111 * 0 on success, a negative errno value otherwise and rte_errno is set.
8114 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8116 struct mlx5_priv *priv = dev->data->dev_private;
8117 enum modify_reg idx;
8120 /* reg_c[0] and reg_c[1] are reserved. */
8121 priv->sh->flow_mreg_c[n++] = REG_C_0;
8122 priv->sh->flow_mreg_c[n++] = REG_C_1;
8123 /* Discover availability of other reg_c's. */
8124 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8125 struct rte_flow_attr attr = {
8126 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8127 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8130 struct rte_flow_item items[] = {
8132 .type = RTE_FLOW_ITEM_TYPE_END,
8135 struct rte_flow_action actions[] = {
8137 .type = (enum rte_flow_action_type)
8138 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8139 .conf = &(struct mlx5_flow_action_copy_mreg){
8145 .type = RTE_FLOW_ACTION_TYPE_JUMP,
8146 .conf = &(struct rte_flow_action_jump){
8147 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8151 .type = RTE_FLOW_ACTION_TYPE_END,
8155 struct rte_flow *flow;
8156 struct rte_flow_error error;
8158 if (!priv->config.dv_flow_en)
8160 /* Create internal flow, validation skips copy action. */
8161 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8162 items, actions, false, &error);
8163 flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8167 priv->sh->flow_mreg_c[n++] = idx;
8168 flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8170 for (; n < MLX5_MREG_C_NUM; ++n)
8171 priv->sh->flow_mreg_c[n] = REG_NON;
8172 priv->sh->metadata_regc_check_flag = 1;
8177 save_dump_file(const uint8_t *data, uint32_t size,
8178 uint32_t type, uint64_t id, void *arg, FILE *file)
8180 char line[BUF_SIZE];
8183 uint32_t actions_num;
8184 struct rte_flow_query_count *count;
8186 memset(line, 0, BUF_SIZE);
8188 case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8189 actions_num = *(uint32_t *)(arg);
8190 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8191 type, id, actions_num);
8193 case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8194 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8197 case DR_DUMP_REC_TYPE_PMD_COUNTER:
8198 count = (struct rte_flow_query_count *)arg;
8200 "%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8201 type, id, count->hits, count->bytes);
8207 for (k = 0; k < size; k++) {
8208 /* Make sure we do not overrun the line buffer length. */
8209 if (out >= BUF_SIZE - 4) {
8213 out += snprintf(line + out, BUF_SIZE - out, "%02x",
8216 fprintf(file, "%s\n", line);
8221 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8222 struct rte_flow_query_count *count, struct rte_flow_error *error)
8224 struct rte_flow_action action[2];
8225 enum mlx5_flow_drv_type ftype;
8226 const struct mlx5_flow_driver_ops *fops;
8229 return rte_flow_error_set(error, ENOENT,
8230 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8232 "invalid flow handle");
8234 action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8235 action[1].type = RTE_FLOW_ACTION_TYPE_END;
8236 if (flow->counter) {
8237 memset(count, 0, sizeof(struct rte_flow_query_count));
8238 ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8239 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8240 ftype < MLX5_FLOW_TYPE_MAX);
8241 fops = flow_get_drv_ops(ftype);
8242 return fops->query(dev, flow, action, count, error);
8247 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8249 * Dump flow ipool data to file
8252 * The pointer to Ethernet device.
8254 * A pointer to a file for output.
8256 * Perform verbose error reporting if not NULL. PMDs initialize this
8257 * structure in case of error only.
8259 * 0 on success, a negative value otherwise.
8262 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8263 struct rte_flow *flow, FILE *file,
8264 struct rte_flow_error *error)
8266 struct mlx5_priv *priv = dev->data->dev_private;
8267 struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
8268 struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8269 uint32_t handle_idx;
8270 struct mlx5_flow_handle *dh;
8271 struct rte_flow_query_count count;
8272 uint32_t actions_num;
8273 const uint8_t *data;
8277 void *action = NULL;
8280 return rte_flow_error_set(error, ENOENT,
8281 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8283 "invalid flow handle");
8285 handle_idx = flow->dev_handles;
8286 while (handle_idx) {
8287 dh = mlx5_ipool_get(priv->sh->ipool
8288 [MLX5_IPOOL_MLX5_FLOW], handle_idx);
8291 handle_idx = dh->next.next;
8294 type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8295 flow_dv_query_count_ptr(dev, flow->counter,
8298 id = (uint64_t)(uintptr_t)action;
8299 if (!mlx5_flow_query_counter(dev, flow, &count, error))
8300 save_dump_file(NULL, 0, type,
8301 id, (void *)&count, file);
8303 /* Get modify_hdr and encap_decap buf from ipools. */
8305 modify_hdr = dh->dvh.modify_hdr;
8307 if (dh->dvh.rix_encap_decap) {
8308 encap_decap = mlx5_ipool_get(priv->sh->ipool
8309 [MLX5_IPOOL_DECAP_ENCAP],
8310 dh->dvh.rix_encap_decap);
8313 data = (const uint8_t *)modify_hdr->actions;
8314 size = (size_t)(modify_hdr->actions_num) * 8;
8315 id = (uint64_t)(uintptr_t)modify_hdr->action;
8316 actions_num = modify_hdr->actions_num;
8317 type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8318 save_dump_file(data, size, type, id,
8319 (void *)(&actions_num), file);
8322 data = encap_decap->buf;
8323 size = encap_decap->size;
8324 id = (uint64_t)(uintptr_t)encap_decap->action;
8325 type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8326 save_dump_file(data, size, type,
8334 * Dump all flow's encap_decap/modify_hdr/counter data to file
8337 * The pointer to Ethernet device.
8339 * A pointer to a file for output.
8341 * Perform verbose error reporting if not NULL. PMDs initialize this
8342 * structure in case of error only.
8344 * 0 on success, a negative value otherwise.
8347 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8348 FILE *file, struct rte_flow_error *error)
8350 struct mlx5_priv *priv = dev->data->dev_private;
8351 struct mlx5_dev_ctx_shared *sh = priv->sh;
8352 struct mlx5_hlist *h;
8353 struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
8354 struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8355 struct rte_flow_query_count count;
8356 uint32_t actions_num;
8357 const uint8_t *data;
8363 struct mlx5_list_inconst *l_inconst;
8364 struct mlx5_list_entry *e;
8366 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8370 /* encap_decap hlist is lcore_share, get global core cache. */
8371 i = MLX5_LIST_GLOBAL;
8372 h = sh->encaps_decaps;
8374 for (j = 0; j <= h->mask; j++) {
8375 l_inconst = &h->buckets[j].l;
8376 if (!l_inconst || !l_inconst->cache[i])
8379 e = LIST_FIRST(&l_inconst->cache[i]->h);
8382 (struct mlx5_flow_dv_encap_decap_resource *)e;
8383 data = encap_decap->buf;
8384 size = encap_decap->size;
8385 id = (uint64_t)(uintptr_t)encap_decap->action;
8386 type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8387 save_dump_file(data, size, type,
8389 e = LIST_NEXT(e, next);
8394 /* get modify_hdr */
8395 h = sh->modify_cmds;
8397 lcore_index = rte_lcore_index(rte_lcore_id());
8398 if (unlikely(lcore_index == -1)) {
8399 lcore_index = MLX5_LIST_NLCORE;
8400 rte_spinlock_lock(&h->l_const.lcore_lock);
8404 for (j = 0; j <= h->mask; j++) {
8405 l_inconst = &h->buckets[j].l;
8406 if (!l_inconst || !l_inconst->cache[i])
8409 e = LIST_FIRST(&l_inconst->cache[i]->h);
8412 (struct mlx5_flow_dv_modify_hdr_resource *)e;
8413 data = (const uint8_t *)modify_hdr->actions;
8414 size = (size_t)(modify_hdr->actions_num) * 8;
8415 actions_num = modify_hdr->actions_num;
8416 id = (uint64_t)(uintptr_t)modify_hdr->action;
8417 type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8418 save_dump_file(data, size, type, id,
8419 (void *)(&actions_num), file);
8420 e = LIST_NEXT(e, next);
8424 if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8425 rte_spinlock_unlock(&h->l_const.lcore_lock);
8429 MLX5_ASSERT(cmng->n_valid <= cmng->n);
8430 max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8431 for (j = 1; j <= max; j++) {
8433 flow_dv_query_count_ptr(dev, j, &action, error);
8435 if (!flow_dv_query_count(dev, j, &count, error)) {
8436 type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8437 id = (uint64_t)(uintptr_t)action;
8438 save_dump_file(NULL, 0, type,
8439 id, (void *)&count, file);
8448 * Dump flow raw hw data to file
8451 * The pointer to Ethernet device.
8453 * A pointer to a file for output.
8455 * Perform verbose error reporting if not NULL. PMDs initialize this
8456 * structure in case of error only.
8458 * 0 on success, a nagative value otherwise.
8461 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8463 struct rte_flow_error *error __rte_unused)
8465 struct mlx5_priv *priv = dev->data->dev_private;
8466 struct mlx5_dev_ctx_shared *sh = priv->sh;
8467 uint32_t handle_idx;
8469 struct mlx5_flow_handle *dh;
8470 struct rte_flow *flow;
8472 if (!priv->config.dv_flow_en) {
8473 if (fputs("device dv flow disabled\n", file) <= 0)
8480 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8481 if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8484 return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8486 sh->tx_domain, file);
8489 flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8490 (uintptr_t)(void *)flow_idx);
8494 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8495 mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8497 handle_idx = flow->dev_handles;
8498 while (handle_idx) {
8499 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8504 ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8509 handle_idx = dh->next.next;
8515 * Get aged-out flows.
8518 * Pointer to the Ethernet device structure.
8519 * @param[in] context
8520 * The address of an array of pointers to the aged-out flows contexts.
8521 * @param[in] nb_countexts
8522 * The length of context array pointers.
8524 * Perform verbose error reporting if not NULL. Initialized in case of
8528 * how many contexts get in success, otherwise negative errno value.
8529 * if nb_contexts is 0, return the amount of all aged contexts.
8530 * if nb_contexts is not 0 , return the amount of aged flows reported
8531 * in the context array.
8534 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8535 uint32_t nb_contexts, struct rte_flow_error *error)
8537 const struct mlx5_flow_driver_ops *fops;
8538 struct rte_flow_attr attr = { .transfer = 0 };
8540 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8541 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8542 return fops->get_aged_flows(dev, contexts, nb_contexts,
8546 "port %u get aged flows is not supported.",
8547 dev->data->port_id);
8551 /* Wrapper for driver action_validate op callback */
8553 flow_drv_action_validate(struct rte_eth_dev *dev,
8554 const struct rte_flow_indir_action_conf *conf,
8555 const struct rte_flow_action *action,
8556 const struct mlx5_flow_driver_ops *fops,
8557 struct rte_flow_error *error)
8559 static const char err_msg[] = "indirect action validation unsupported";
8561 if (!fops->action_validate) {
8562 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8563 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8567 return fops->action_validate(dev, conf, action, error);
8571 * Destroys the shared action by handle.
8574 * Pointer to Ethernet device structure.
8576 * Handle for the indirect action object to be destroyed.
8578 * Perform verbose error reporting if not NULL. PMDs initialize this
8579 * structure in case of error only.
8582 * 0 on success, a negative errno value otherwise and rte_errno is set.
8584 * @note: wrapper for driver action_create op callback.
8587 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8588 struct rte_flow_action_handle *handle,
8589 struct rte_flow_error *error)
8591 static const char err_msg[] = "indirect action destruction unsupported";
8592 struct rte_flow_attr attr = { .transfer = 0 };
8593 const struct mlx5_flow_driver_ops *fops =
8594 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8596 if (!fops->action_destroy) {
8597 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8598 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8602 return fops->action_destroy(dev, handle, error);
8605 /* Wrapper for driver action_destroy op callback */
8607 flow_drv_action_update(struct rte_eth_dev *dev,
8608 struct rte_flow_action_handle *handle,
8610 const struct mlx5_flow_driver_ops *fops,
8611 struct rte_flow_error *error)
8613 static const char err_msg[] = "indirect action update unsupported";
8615 if (!fops->action_update) {
8616 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8617 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8621 return fops->action_update(dev, handle, update, error);
8624 /* Wrapper for driver action_destroy op callback */
8626 flow_drv_action_query(struct rte_eth_dev *dev,
8627 const struct rte_flow_action_handle *handle,
8629 const struct mlx5_flow_driver_ops *fops,
8630 struct rte_flow_error *error)
8632 static const char err_msg[] = "indirect action query unsupported";
8634 if (!fops->action_query) {
8635 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8636 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8640 return fops->action_query(dev, handle, data, error);
8644 * Create indirect action for reuse in multiple flow rules.
8647 * Pointer to Ethernet device structure.
8649 * Pointer to indirect action object configuration.
8651 * Action configuration for indirect action object creation.
8653 * Perform verbose error reporting if not NULL. PMDs initialize this
8654 * structure in case of error only.
8656 * A valid handle in case of success, NULL otherwise and rte_errno is set.
8658 static struct rte_flow_action_handle *
8659 mlx5_action_handle_create(struct rte_eth_dev *dev,
8660 const struct rte_flow_indir_action_conf *conf,
8661 const struct rte_flow_action *action,
8662 struct rte_flow_error *error)
8664 static const char err_msg[] = "indirect action creation unsupported";
8665 struct rte_flow_attr attr = { .transfer = 0 };
8666 const struct mlx5_flow_driver_ops *fops =
8667 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8669 if (flow_drv_action_validate(dev, conf, action, fops, error))
8671 if (!fops->action_create) {
8672 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8673 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8677 return fops->action_create(dev, conf, action, error);
8681 * Updates inplace the indirect action configuration pointed by *handle*
8682 * with the configuration provided as *update* argument.
8683 * The update of the indirect action configuration effects all flow rules
8684 * reusing the action via handle.
8687 * Pointer to Ethernet device structure.
8689 * Handle for the indirect action to be updated.
8691 * Action specification used to modify the action pointed by handle.
8692 * *update* could be of same type with the action pointed by the *handle*
8693 * handle argument, or some other structures like a wrapper, depending on
8694 * the indirect action type.
8696 * Perform verbose error reporting if not NULL. PMDs initialize this
8697 * structure in case of error only.
8700 * 0 on success, a negative errno value otherwise and rte_errno is set.
8703 mlx5_action_handle_update(struct rte_eth_dev *dev,
8704 struct rte_flow_action_handle *handle,
8706 struct rte_flow_error *error)
8708 struct rte_flow_attr attr = { .transfer = 0 };
8709 const struct mlx5_flow_driver_ops *fops =
8710 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8713 ret = flow_drv_action_validate(dev, NULL,
8714 (const struct rte_flow_action *)update, fops, error);
8717 return flow_drv_action_update(dev, handle, update, fops,
8722 * Query the indirect action by handle.
8724 * This function allows retrieving action-specific data such as counters.
8725 * Data is gathered by special action which may be present/referenced in
8726 * more than one flow rule definition.
8728 * see @RTE_FLOW_ACTION_TYPE_COUNT
8731 * Pointer to Ethernet device structure.
8733 * Handle for the indirect action to query.
8734 * @param[in, out] data
8735 * Pointer to storage for the associated query data type.
8737 * Perform verbose error reporting if not NULL. PMDs initialize this
8738 * structure in case of error only.
8741 * 0 on success, a negative errno value otherwise and rte_errno is set.
8744 mlx5_action_handle_query(struct rte_eth_dev *dev,
8745 const struct rte_flow_action_handle *handle,
8747 struct rte_flow_error *error)
8749 struct rte_flow_attr attr = { .transfer = 0 };
8750 const struct mlx5_flow_driver_ops *fops =
8751 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8753 return flow_drv_action_query(dev, handle, data, fops, error);
8757 * Destroy all indirect actions (shared RSS).
8760 * Pointer to Ethernet device.
8763 * 0 on success, a negative errno value otherwise and rte_errno is set.
8766 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8768 struct rte_flow_error error;
8769 struct mlx5_priv *priv = dev->data->dev_private;
8770 struct mlx5_shared_action_rss *shared_rss;
8774 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8775 priv->rss_shared_actions, idx, shared_rss, next) {
8776 ret |= mlx5_action_handle_destroy(dev,
8777 (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8782 #ifndef HAVE_MLX5DV_DR
8783 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8785 #define MLX5_DOMAIN_SYNC_FLOW \
8786 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8789 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8791 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8792 const struct mlx5_flow_driver_ops *fops;
8794 struct rte_flow_attr attr = { .transfer = 0 };
8796 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8797 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8803 const struct mlx5_flow_tunnel *
8804 mlx5_get_tof(const struct rte_flow_item *item,
8805 const struct rte_flow_action *action,
8806 enum mlx5_tof_rule_type *rule_type)
8808 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8809 if (item->type == (typeof(item->type))
8810 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8811 *rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8812 return flow_items_to_tunnel(item);
8815 for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8816 if (action->type == (typeof(action->type))
8817 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8818 *rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8819 return flow_actions_to_tunnel(action);
8826 * tunnel offload functionalilty is defined for DV environment only
8828 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8830 union tunnel_offload_mark {
8833 uint32_t app_reserve:8;
8834 uint32_t table_id:15;
8835 uint32_t transfer:1;
8836 uint32_t _unused_:8;
8841 mlx5_access_tunnel_offload_db
8842 (struct rte_eth_dev *dev,
8843 bool (*match)(struct rte_eth_dev *,
8844 struct mlx5_flow_tunnel *, const void *),
8845 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8846 void (*miss)(struct rte_eth_dev *, void *),
8847 void *ctx, bool lock_op);
8850 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8851 struct rte_flow *flow,
8852 const struct rte_flow_attr *attr,
8853 const struct rte_flow_action *app_actions,
8855 const struct mlx5_flow_tunnel *tunnel,
8856 struct tunnel_default_miss_ctx *ctx,
8857 struct rte_flow_error *error)
8859 struct mlx5_priv *priv = dev->data->dev_private;
8860 struct mlx5_flow *dev_flow;
8861 struct rte_flow_attr miss_attr = *attr;
8862 const struct rte_flow_item miss_items[2] = {
8864 .type = RTE_FLOW_ITEM_TYPE_ETH,
8870 .type = RTE_FLOW_ITEM_TYPE_END,
8876 union tunnel_offload_mark mark_id;
8877 struct rte_flow_action_mark miss_mark;
8878 struct rte_flow_action miss_actions[3] = {
8879 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8880 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
8882 const struct rte_flow_action_jump *jump_data;
8883 uint32_t i, flow_table = 0; /* prevent compilation warning */
8884 struct flow_grp_info grp_info = {
8886 .transfer = attr->transfer,
8887 .fdb_def_rule = !!priv->fdb_def_rule,
8892 if (!attr->transfer) {
8895 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8896 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8897 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8900 return rte_flow_error_set
8902 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8903 NULL, "invalid default miss RSS");
8904 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8905 ctx->action_rss.level = 0,
8906 ctx->action_rss.types = priv->rss_conf.rss_hf,
8907 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8908 ctx->action_rss.queue_num = priv->reta_idx_n,
8909 ctx->action_rss.key = priv->rss_conf.rss_key,
8910 ctx->action_rss.queue = ctx->queue;
8911 if (!priv->reta_idx_n || !priv->rxqs_n)
8912 return rte_flow_error_set
8914 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8915 NULL, "invalid port configuration");
8916 if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
8917 ctx->action_rss.types = 0;
8918 for (i = 0; i != priv->reta_idx_n; ++i)
8919 ctx->queue[i] = (*priv->reta_idx)[i];
8921 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8922 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8924 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8925 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8926 jump_data = app_actions->conf;
8927 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8928 miss_attr.group = jump_data->group;
8929 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8930 &flow_table, &grp_info, error);
8932 return rte_flow_error_set(error, EINVAL,
8933 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8934 NULL, "invalid tunnel id");
8935 mark_id.app_reserve = 0;
8936 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8937 mark_id.transfer = !!attr->transfer;
8938 mark_id._unused_ = 0;
8939 miss_mark.id = mark_id.val;
8940 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8941 miss_items, miss_actions, flow_idx, error);
8944 dev_flow->flow = flow;
8945 dev_flow->external = true;
8946 dev_flow->tunnel = tunnel;
8947 dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8948 /* Subflow object was created, we must include one in the list. */
8949 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8950 dev_flow->handle, next);
8952 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8953 dev->data->port_id, tunnel->app_tunnel.type,
8954 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8955 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8956 miss_actions, error);
8958 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8964 static const struct mlx5_flow_tbl_data_entry *
8965 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8967 struct mlx5_priv *priv = dev->data->dev_private;
8968 struct mlx5_dev_ctx_shared *sh = priv->sh;
8969 struct mlx5_list_entry *he;
8970 union tunnel_offload_mark mbits = { .val = mark };
8971 union mlx5_flow_tbl_key table_key = {
8973 .level = tunnel_id_to_flow_tbl(mbits.table_id),
8977 .is_fdb = !!mbits.transfer,
8981 struct mlx5_flow_cb_ctx ctx = {
8982 .data = &table_key.v64,
8985 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
8987 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8991 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
8992 struct mlx5_list_entry *entry)
8994 struct mlx5_dev_ctx_shared *sh = tool_ctx;
8995 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8997 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8998 tunnel_flow_tbl_to_id(tte->flow_table));
9003 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9004 struct mlx5_list_entry *entry, void *cb_ctx)
9006 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9007 union tunnel_tbl_key tbl = {
9008 .val = *(uint64_t *)(ctx->data),
9010 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9012 return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9015 static struct mlx5_list_entry *
9016 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9018 struct mlx5_dev_ctx_shared *sh = tool_ctx;
9019 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9020 struct tunnel_tbl_entry *tte;
9021 union tunnel_tbl_key tbl = {
9022 .val = *(uint64_t *)(ctx->data),
9025 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9030 mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9032 if (tte->flow_table >= MLX5_MAX_TABLES) {
9033 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9035 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9038 } else if (!tte->flow_table) {
9041 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9042 tte->tunnel_id = tbl.tunnel_id;
9043 tte->group = tbl.group;
9051 static struct mlx5_list_entry *
9052 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9053 struct mlx5_list_entry *oentry,
9054 void *cb_ctx __rte_unused)
9056 struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9061 memcpy(tte, oentry, sizeof(*tte));
9066 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9067 struct mlx5_list_entry *entry)
9069 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9075 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9076 const struct mlx5_flow_tunnel *tunnel,
9077 uint32_t group, uint32_t *table,
9078 struct rte_flow_error *error)
9080 struct mlx5_list_entry *he;
9081 struct tunnel_tbl_entry *tte;
9082 union tunnel_tbl_key key = {
9083 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9086 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9087 struct mlx5_hlist *group_hash;
9088 struct mlx5_flow_cb_ctx ctx = {
9092 group_hash = tunnel ? tunnel->groups : thub->groups;
9093 he = mlx5_hlist_register(group_hash, key.val, &ctx);
9095 return rte_flow_error_set(error, EINVAL,
9096 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9098 "tunnel group index not supported");
9099 tte = container_of(he, typeof(*tte), hash);
9100 *table = tte->flow_table;
9101 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9102 dev->data->port_id, key.tunnel_id, group, *table);
9107 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9108 struct mlx5_flow_tunnel *tunnel)
9110 struct mlx5_priv *priv = dev->data->dev_private;
9111 struct mlx5_indexed_pool *ipool;
9113 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9114 dev->data->port_id, tunnel->tunnel_id);
9115 LIST_REMOVE(tunnel, chain);
9116 mlx5_hlist_destroy(tunnel->groups);
9117 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9118 mlx5_ipool_free(ipool, tunnel->tunnel_id);
9122 mlx5_access_tunnel_offload_db
9123 (struct rte_eth_dev *dev,
9124 bool (*match)(struct rte_eth_dev *,
9125 struct mlx5_flow_tunnel *, const void *),
9126 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9127 void (*miss)(struct rte_eth_dev *, void *),
9128 void *ctx, bool lock_op)
9130 bool verdict = false;
9131 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9132 struct mlx5_flow_tunnel *tunnel;
9134 rte_spinlock_lock(&thub->sl);
9135 LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9136 verdict = match(dev, tunnel, (const void *)ctx);
9141 rte_spinlock_unlock(&thub->sl);
9143 hit(dev, tunnel, ctx);
9144 if (!verdict && miss)
9147 rte_spinlock_unlock(&thub->sl);
9152 struct tunnel_db_find_tunnel_id_ctx {
9154 struct mlx5_flow_tunnel *tunnel;
9158 find_tunnel_id_match(struct rte_eth_dev *dev,
9159 struct mlx5_flow_tunnel *tunnel, const void *x)
9161 const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9164 return tunnel->tunnel_id == ctx->tunnel_id;
9168 find_tunnel_id_hit(struct rte_eth_dev *dev,
9169 struct mlx5_flow_tunnel *tunnel, void *x)
9171 struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9173 ctx->tunnel = tunnel;
9176 static struct mlx5_flow_tunnel *
9177 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9179 struct tunnel_db_find_tunnel_id_ctx ctx = {
9183 mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9184 find_tunnel_id_hit, NULL, &ctx, true);
9189 static struct mlx5_flow_tunnel *
9190 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9191 const struct rte_flow_tunnel *app_tunnel)
9193 struct mlx5_priv *priv = dev->data->dev_private;
9194 struct mlx5_indexed_pool *ipool;
9195 struct mlx5_flow_tunnel *tunnel;
9198 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9199 tunnel = mlx5_ipool_zmalloc(ipool, &id);
9202 if (id >= MLX5_MAX_TUNNELS) {
9203 mlx5_ipool_free(ipool, id);
9204 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9207 tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9209 mlx5_flow_tunnel_grp2tbl_create_cb,
9210 mlx5_flow_tunnel_grp2tbl_match_cb,
9211 mlx5_flow_tunnel_grp2tbl_remove_cb,
9212 mlx5_flow_tunnel_grp2tbl_clone_cb,
9213 mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9214 if (!tunnel->groups) {
9215 mlx5_ipool_free(ipool, id);
9218 /* initiate new PMD tunnel */
9219 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9220 tunnel->tunnel_id = id;
9221 tunnel->action.type = (typeof(tunnel->action.type))
9222 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9223 tunnel->action.conf = tunnel;
9224 tunnel->item.type = (typeof(tunnel->item.type))
9225 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9226 tunnel->item.spec = tunnel;
9227 tunnel->item.last = NULL;
9228 tunnel->item.mask = NULL;
9230 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9231 dev->data->port_id, tunnel->tunnel_id);
9236 struct tunnel_db_get_tunnel_ctx {
9237 const struct rte_flow_tunnel *app_tunnel;
9238 struct mlx5_flow_tunnel *tunnel;
9241 static bool get_tunnel_match(struct rte_eth_dev *dev,
9242 struct mlx5_flow_tunnel *tunnel, const void *x)
9244 const struct tunnel_db_get_tunnel_ctx *ctx = x;
9247 return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9248 sizeof(*ctx->app_tunnel));
9251 static void get_tunnel_hit(struct rte_eth_dev *dev,
9252 struct mlx5_flow_tunnel *tunnel, void *x)
9254 /* called under tunnel spinlock protection */
9255 struct tunnel_db_get_tunnel_ctx *ctx = x;
9259 ctx->tunnel = tunnel;
9262 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9264 /* called under tunnel spinlock protection */
9265 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9266 struct tunnel_db_get_tunnel_ctx *ctx = x;
9268 rte_spinlock_unlock(&thub->sl);
9269 ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9270 rte_spinlock_lock(&thub->sl);
9272 ctx->tunnel->refctn = 1;
9273 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9279 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9280 const struct rte_flow_tunnel *app_tunnel,
9281 struct mlx5_flow_tunnel **tunnel)
9283 struct tunnel_db_get_tunnel_ctx ctx = {
9284 .app_tunnel = app_tunnel,
9287 mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9288 get_tunnel_miss, &ctx, true);
9289 *tunnel = ctx.tunnel;
9290 return ctx.tunnel ? 0 : -ENOMEM;
9293 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9295 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9299 if (!LIST_EMPTY(&thub->tunnels))
9300 DRV_LOG(WARNING, "port %u tunnels present", port_id);
9301 mlx5_hlist_destroy(thub->groups);
9305 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9308 struct mlx5_flow_tunnel_hub *thub;
9310 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9314 LIST_INIT(&thub->tunnels);
9315 rte_spinlock_init(&thub->sl);
9316 thub->groups = mlx5_hlist_create("flow groups", 64,
9318 mlx5_flow_tunnel_grp2tbl_create_cb,
9319 mlx5_flow_tunnel_grp2tbl_match_cb,
9320 mlx5_flow_tunnel_grp2tbl_remove_cb,
9321 mlx5_flow_tunnel_grp2tbl_clone_cb,
9322 mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9323 if (!thub->groups) {
9327 sh->tunnel_hub = thub;
9333 mlx5_hlist_destroy(thub->groups);
9340 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9341 struct rte_flow_tunnel *tunnel,
9342 const char *err_msg)
9345 if (!is_tunnel_offload_active(dev)) {
9346 err_msg = "tunnel offload was not activated";
9348 } else if (!tunnel) {
9349 err_msg = "no application tunnel";
9353 switch (tunnel->type) {
9355 err_msg = "unsupported tunnel type";
9357 case RTE_FLOW_ITEM_TYPE_VXLAN:
9358 case RTE_FLOW_ITEM_TYPE_GRE:
9359 case RTE_FLOW_ITEM_TYPE_NVGRE:
9360 case RTE_FLOW_ITEM_TYPE_GENEVE:
9369 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9370 struct rte_flow_tunnel *app_tunnel,
9371 struct rte_flow_action **actions,
9372 uint32_t *num_of_actions,
9373 struct rte_flow_error *error)
9376 struct mlx5_flow_tunnel *tunnel;
9377 const char *err_msg = NULL;
9378 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9381 return rte_flow_error_set(error, EINVAL,
9382 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9384 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9386 return rte_flow_error_set(error, ret,
9387 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9388 "failed to initialize pmd tunnel");
9390 *actions = &tunnel->action;
9391 *num_of_actions = 1;
9396 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9397 struct rte_flow_tunnel *app_tunnel,
9398 struct rte_flow_item **items,
9399 uint32_t *num_of_items,
9400 struct rte_flow_error *error)
9403 struct mlx5_flow_tunnel *tunnel;
9404 const char *err_msg = NULL;
9405 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9408 return rte_flow_error_set(error, EINVAL,
9409 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9411 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9413 return rte_flow_error_set(error, ret,
9414 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9415 "failed to initialize pmd tunnel");
9417 *items = &tunnel->item;
9422 struct tunnel_db_element_release_ctx {
9423 struct rte_flow_item *items;
9424 struct rte_flow_action *actions;
9425 uint32_t num_elements;
9426 struct rte_flow_error *error;
9431 tunnel_element_release_match(struct rte_eth_dev *dev,
9432 struct mlx5_flow_tunnel *tunnel, const void *x)
9434 const struct tunnel_db_element_release_ctx *ctx = x;
9437 if (ctx->num_elements != 1)
9439 else if (ctx->items)
9440 return ctx->items == &tunnel->item;
9441 else if (ctx->actions)
9442 return ctx->actions == &tunnel->action;
9448 tunnel_element_release_hit(struct rte_eth_dev *dev,
9449 struct mlx5_flow_tunnel *tunnel, void *x)
9451 struct tunnel_db_element_release_ctx *ctx = x;
9453 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9454 mlx5_flow_tunnel_free(dev, tunnel);
9458 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9460 struct tunnel_db_element_release_ctx *ctx = x;
9462 ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9463 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9464 "invalid argument");
9468 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9469 struct rte_flow_item *pmd_items,
9470 uint32_t num_items, struct rte_flow_error *err)
9472 struct tunnel_db_element_release_ctx ctx = {
9475 .num_elements = num_items,
9479 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9480 tunnel_element_release_hit,
9481 tunnel_element_release_miss, &ctx, false);
9487 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9488 struct rte_flow_action *pmd_actions,
9489 uint32_t num_actions, struct rte_flow_error *err)
9491 struct tunnel_db_element_release_ctx ctx = {
9493 .actions = pmd_actions,
9494 .num_elements = num_actions,
9498 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9499 tunnel_element_release_hit,
9500 tunnel_element_release_miss, &ctx, false);
9506 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9508 struct rte_flow_restore_info *info,
9509 struct rte_flow_error *err)
9511 uint64_t ol_flags = m->ol_flags;
9512 const struct mlx5_flow_tbl_data_entry *tble;
9513 const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9515 if (!is_tunnel_offload_active(dev)) {
9520 if ((ol_flags & mask) != mask)
9522 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9524 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9525 dev->data->port_id, m->hash.fdir.hi);
9528 MLX5_ASSERT(tble->tunnel);
9529 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9530 info->group_id = tble->group_id;
9531 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9532 RTE_FLOW_RESTORE_INFO_GROUP_ID |
9533 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9538 return rte_flow_error_set(err, EINVAL,
9539 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9540 "failed to get restore info");
9543 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9545 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9546 __rte_unused struct rte_flow_tunnel *app_tunnel,
9547 __rte_unused struct rte_flow_action **actions,
9548 __rte_unused uint32_t *num_of_actions,
9549 __rte_unused struct rte_flow_error *error)
9555 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9556 __rte_unused struct rte_flow_tunnel *app_tunnel,
9557 __rte_unused struct rte_flow_item **items,
9558 __rte_unused uint32_t *num_of_items,
9559 __rte_unused struct rte_flow_error *error)
9565 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9566 __rte_unused struct rte_flow_item *pmd_items,
9567 __rte_unused uint32_t num_items,
9568 __rte_unused struct rte_flow_error *err)
9574 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9575 __rte_unused struct rte_flow_action *pmd_action,
9576 __rte_unused uint32_t num_actions,
9577 __rte_unused struct rte_flow_error *err)
9583 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9584 __rte_unused struct rte_mbuf *m,
9585 __rte_unused struct rte_flow_restore_info *i,
9586 __rte_unused struct rte_flow_error *err)
9592 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9593 __rte_unused struct rte_flow *flow,
9594 __rte_unused const struct rte_flow_attr *attr,
9595 __rte_unused const struct rte_flow_action *actions,
9596 __rte_unused uint32_t flow_idx,
9597 __rte_unused const struct mlx5_flow_tunnel *tunnel,
9598 __rte_unused struct tunnel_default_miss_ctx *ctx,
9599 __rte_unused struct rte_flow_error *error)
9604 static struct mlx5_flow_tunnel *
9605 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9606 __rte_unused uint32_t id)
9612 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9613 __rte_unused struct mlx5_flow_tunnel *tunnel)
9618 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9619 __rte_unused const struct mlx5_flow_tunnel *t,
9620 __rte_unused uint32_t group,
9621 __rte_unused uint32_t *table,
9622 struct rte_flow_error *error)
9624 return rte_flow_error_set(error, ENOTSUP,
9625 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9626 "tunnel offload requires DV support");
9630 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9631 __rte_unused uint16_t port_id)
9634 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9637 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9640 struct rte_flow_error error;
9642 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9644 ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9646 (void *)(uintptr_t)item->type, &error);
9648 printf("%s ", item_name);
9650 printf("%d\n", (int)item->type);
9656 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9658 const struct rte_flow_item_udp *spec = udp_item->spec;
9659 const struct rte_flow_item_udp *mask = udp_item->mask;
9660 uint16_t udp_dport = 0;
9664 mask = &rte_flow_item_udp_mask;
9665 udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9666 mask->hdr.dst_port);
9668 return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9671 static const struct mlx5_flow_expand_node *
9672 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9673 unsigned int item_idx,
9674 const struct mlx5_flow_expand_node graph[],
9675 const struct mlx5_flow_expand_node *node)
9677 const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9678 switch (item->type) {
9679 case RTE_FLOW_ITEM_TYPE_VXLAN:
9680 MLX5_ASSERT(item_idx > 0);
9681 prev_item = pattern + item_idx - 1;
9682 MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9683 if (mlx5_flow_is_std_vxlan_port(prev_item))
9684 return &graph[MLX5_EXPANSION_STD_VXLAN];
9686 return &graph[MLX5_EXPANSION_L3_VXLAN];