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,
79 [MLX5_FLOW_TYPE_HW] = &mlx5_flow_hw_drv_ops,
81 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
82 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
85 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
86 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
91 /** Node object of input graph for mlx5_flow_expand_rss(). */
92 struct mlx5_flow_expand_node {
93 const int *const next;
95 * List of next node indexes. Index 0 is interpreted as a terminator.
97 const enum rte_flow_item_type type;
98 /**< Pattern item type of current node. */
101 * RSS types bit-field associated with this node
102 * (see RTE_ETH_RSS_* definitions).
106 * Bit-fields that define how the node is used in the expansion.
107 * (see MLX5_EXPANSION_NODE_* definitions).
111 /* Optional expand field. The expansion alg will not go deeper. */
112 #define MLX5_EXPANSION_NODE_OPTIONAL (UINT64_C(1) << 0)
114 /* The node is not added implicitly as expansion to the flow pattern.
115 * If the node type does not match the flow pattern item type, the
116 * expansion alg will go deeper to its next items.
117 * In the current implementation, the list of next nodes indexes can
118 * have up to one node with this flag set and it has to be the last
119 * node index (before the list terminator).
121 #define MLX5_EXPANSION_NODE_EXPLICIT (UINT64_C(1) << 1)
123 /** Object returned by mlx5_flow_expand_rss(). */
124 struct mlx5_flow_expand_rss {
126 /**< Number of entries @p patterns and @p priorities. */
128 struct rte_flow_item *pattern; /**< Expanded pattern array. */
129 uint32_t priority; /**< Priority offset for each expansion. */
134 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
136 static const struct mlx5_flow_expand_node *
137 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
138 unsigned int item_idx,
139 const struct mlx5_flow_expand_node graph[],
140 const struct mlx5_flow_expand_node *node);
143 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
145 switch (item->type) {
146 case RTE_FLOW_ITEM_TYPE_ETH:
147 case RTE_FLOW_ITEM_TYPE_VLAN:
148 case RTE_FLOW_ITEM_TYPE_IPV4:
149 case RTE_FLOW_ITEM_TYPE_IPV6:
150 case RTE_FLOW_ITEM_TYPE_UDP:
151 case RTE_FLOW_ITEM_TYPE_TCP:
152 case RTE_FLOW_ITEM_TYPE_VXLAN:
153 case RTE_FLOW_ITEM_TYPE_NVGRE:
154 case RTE_FLOW_ITEM_TYPE_GRE:
155 case RTE_FLOW_ITEM_TYPE_GENEVE:
156 case RTE_FLOW_ITEM_TYPE_MPLS:
157 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
158 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
159 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
160 case RTE_FLOW_ITEM_TYPE_GTP:
169 * Network Service Header (NSH) and its next protocol values
170 * are described in RFC-8393.
172 static enum rte_flow_item_type
173 mlx5_nsh_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
175 enum rte_flow_item_type type;
177 switch (proto_mask & proto_spec) {
178 case RTE_VXLAN_GPE_TYPE_IPV4:
179 type = RTE_FLOW_ITEM_TYPE_IPV4;
181 case RTE_VXLAN_GPE_TYPE_IPV6:
182 type = RTE_VXLAN_GPE_TYPE_IPV6;
184 case RTE_VXLAN_GPE_TYPE_ETH:
185 type = RTE_FLOW_ITEM_TYPE_ETH;
188 type = RTE_FLOW_ITEM_TYPE_END;
193 static enum rte_flow_item_type
194 mlx5_inet_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
196 enum rte_flow_item_type type;
198 switch (proto_mask & proto_spec) {
200 type = RTE_FLOW_ITEM_TYPE_UDP;
203 type = RTE_FLOW_ITEM_TYPE_TCP;
206 type = RTE_FLOW_ITEM_TYPE_IPV4;
209 type = RTE_FLOW_ITEM_TYPE_IPV6;
212 type = RTE_FLOW_ITEM_TYPE_END;
217 static enum rte_flow_item_type
218 mlx5_ethertype_to_item_type(rte_be16_t type_spec,
219 rte_be16_t type_mask, bool is_tunnel)
221 enum rte_flow_item_type type;
223 switch (rte_be_to_cpu_16(type_spec & type_mask)) {
224 case RTE_ETHER_TYPE_TEB:
226 RTE_FLOW_ITEM_TYPE_ETH : RTE_FLOW_ITEM_TYPE_END;
228 case RTE_ETHER_TYPE_VLAN:
230 RTE_FLOW_ITEM_TYPE_VLAN : RTE_FLOW_ITEM_TYPE_END;
232 case RTE_ETHER_TYPE_IPV4:
233 type = RTE_FLOW_ITEM_TYPE_IPV4;
235 case RTE_ETHER_TYPE_IPV6:
236 type = RTE_FLOW_ITEM_TYPE_IPV6;
239 type = RTE_FLOW_ITEM_TYPE_END;
244 static enum rte_flow_item_type
245 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
247 #define MLX5_XSET_ITEM_MASK_SPEC(type, fld) \
249 const void *m = item->mask; \
250 const void *s = item->spec; \
252 ((const struct rte_flow_item_##type *)m)->fld : \
253 rte_flow_item_##type##_mask.fld; \
254 spec = ((const struct rte_flow_item_##type *)s)->fld; \
257 enum rte_flow_item_type ret;
260 if (item == NULL || item->spec == NULL)
261 return RTE_FLOW_ITEM_TYPE_VOID;
262 switch (item->type) {
263 case RTE_FLOW_ITEM_TYPE_ETH:
264 MLX5_XSET_ITEM_MASK_SPEC(eth, type);
266 return RTE_FLOW_ITEM_TYPE_VOID;
267 ret = mlx5_ethertype_to_item_type(spec, mask, false);
269 case RTE_FLOW_ITEM_TYPE_VLAN:
270 MLX5_XSET_ITEM_MASK_SPEC(vlan, inner_type);
272 return RTE_FLOW_ITEM_TYPE_VOID;
273 ret = mlx5_ethertype_to_item_type(spec, mask, false);
275 case RTE_FLOW_ITEM_TYPE_IPV4:
276 MLX5_XSET_ITEM_MASK_SPEC(ipv4, hdr.next_proto_id);
278 return RTE_FLOW_ITEM_TYPE_VOID;
279 ret = mlx5_inet_proto_to_item_type(spec, mask);
281 case RTE_FLOW_ITEM_TYPE_IPV6:
282 MLX5_XSET_ITEM_MASK_SPEC(ipv6, hdr.proto);
284 return RTE_FLOW_ITEM_TYPE_VOID;
285 ret = mlx5_inet_proto_to_item_type(spec, mask);
287 case RTE_FLOW_ITEM_TYPE_GENEVE:
288 MLX5_XSET_ITEM_MASK_SPEC(geneve, protocol);
289 ret = mlx5_ethertype_to_item_type(spec, mask, true);
291 case RTE_FLOW_ITEM_TYPE_GRE:
292 MLX5_XSET_ITEM_MASK_SPEC(gre, protocol);
293 ret = mlx5_ethertype_to_item_type(spec, mask, true);
295 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
296 MLX5_XSET_ITEM_MASK_SPEC(vxlan_gpe, protocol);
297 ret = mlx5_nsh_proto_to_item_type(spec, mask);
300 ret = RTE_FLOW_ITEM_TYPE_VOID;
304 #undef MLX5_XSET_ITEM_MASK_SPEC
308 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
309 const int *next_node)
311 const struct mlx5_flow_expand_node *node = NULL;
312 const int *next = next_node;
314 while (next && *next) {
316 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
317 * flag set, because they were not found in the flow pattern.
319 node = &graph[*next];
320 if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
327 #define MLX5_RSS_EXP_ELT_N 16
330 * Expand RSS flows into several possible flows according to the RSS hash
331 * fields requested and the driver capabilities.
334 * Buffer to store the result expansion.
336 * Buffer size in bytes. If 0, @p buf can be NULL.
340 * RSS types to expand (see RTE_ETH_RSS_* definitions).
342 * Input graph to expand @p pattern according to @p types.
343 * @param[in] graph_root_index
344 * Index of root node in @p graph, typically 0.
347 * A positive value representing the size of @p buf in bytes regardless of
348 * @p size on success, a negative errno value otherwise and rte_errno is
349 * set, the following errors are defined:
351 * -E2BIG: graph-depth @p graph is too deep.
352 * -EINVAL: @p size has not enough space for expanded pattern.
355 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
356 const struct rte_flow_item *pattern, uint64_t types,
357 const struct mlx5_flow_expand_node graph[],
358 int graph_root_index)
360 const struct rte_flow_item *item;
361 const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
362 const int *next_node;
363 const int *stack[MLX5_RSS_EXP_ELT_N];
365 struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
366 unsigned int i, item_idx, last_expand_item_idx = 0;
368 size_t user_pattern_size = 0;
370 const struct mlx5_flow_expand_node *next = NULL;
371 struct rte_flow_item missed_item;
374 const struct rte_flow_item *last_expand_item = NULL;
376 memset(&missed_item, 0, sizeof(missed_item));
377 lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
378 MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
381 buf->entry[0].priority = 0;
382 buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
384 addr = buf->entry[0].pattern;
385 for (item = pattern, item_idx = 0;
386 item->type != RTE_FLOW_ITEM_TYPE_END;
387 item++, item_idx++) {
388 if (!mlx5_flow_is_rss_expandable_item(item)) {
389 user_pattern_size += sizeof(*item);
392 last_expand_item = item;
393 last_expand_item_idx = item_idx;
395 while (node->next && node->next[i]) {
396 next = &graph[node->next[i]];
397 if (next->type == item->type)
399 if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
408 user_pattern_size += sizeof(*item);
410 user_pattern_size += sizeof(*item); /* Handle END item. */
411 lsize += user_pattern_size;
414 /* Copy the user pattern in the first entry of the buffer. */
415 rte_memcpy(addr, pattern, user_pattern_size);
416 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
418 /* Start expanding. */
419 memset(flow_items, 0, sizeof(flow_items));
420 user_pattern_size -= sizeof(*item);
422 * Check if the last valid item has spec set, need complete pattern,
423 * and the pattern can be used for expansion.
425 missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
426 if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
427 /* Item type END indicates expansion is not required. */
430 if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
434 while (node->next && node->next[i]) {
435 next = &graph[node->next[i]];
436 if (next->type == missed_item.type) {
437 flow_items[0].type = missed_item.type;
438 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
441 if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
450 if (next && missed) {
451 elt = 2; /* missed item + item end. */
453 lsize += elt * sizeof(*item) + user_pattern_size;
456 if (node->rss_types & types) {
457 buf->entry[buf->entries].priority = 1;
458 buf->entry[buf->entries].pattern = addr;
460 rte_memcpy(addr, buf->entry[0].pattern,
462 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
463 rte_memcpy(addr, flow_items, elt * sizeof(*item));
464 addr = (void *)(((uintptr_t)addr) +
465 elt * sizeof(*item));
467 } else if (last_expand_item != NULL) {
468 node = mlx5_flow_expand_rss_adjust_node(pattern,
469 last_expand_item_idx, graph, node);
471 memset(flow_items, 0, sizeof(flow_items));
472 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
474 stack[stack_pos] = next_node;
475 node = next_node ? &graph[*next_node] : NULL;
477 flow_items[stack_pos].type = node->type;
478 if (node->rss_types & types) {
481 * compute the number of items to copy from the
482 * expansion and copy it.
483 * When the stack_pos is 0, there are 1 element in it,
484 * plus the addition END item.
487 flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
488 lsize += elt * sizeof(*item) + user_pattern_size;
491 n = elt * sizeof(*item);
492 buf->entry[buf->entries].priority =
493 stack_pos + 1 + missed;
494 buf->entry[buf->entries].pattern = addr;
496 rte_memcpy(addr, buf->entry[0].pattern,
498 addr = (void *)(((uintptr_t)addr) +
500 rte_memcpy(addr, &missed_item,
501 missed * sizeof(*item));
502 addr = (void *)(((uintptr_t)addr) +
503 missed * sizeof(*item));
504 rte_memcpy(addr, flow_items, n);
505 addr = (void *)(((uintptr_t)addr) + n);
508 if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
510 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
512 if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
516 stack[stack_pos] = next_node;
517 } else if (*(next_node + 1)) {
518 /* Follow up with the next possibility. */
519 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
521 } else if (!stack_pos) {
523 * Completing the traverse over the different paths.
524 * The next_node is advanced to the terminator.
528 /* Move to the next path. */
530 next_node = stack[--stack_pos];
535 next_node = mlx5_flow_expand_rss_skip_explicit(graph,
537 stack[stack_pos] = next_node;
539 node = next_node && *next_node ? &graph[*next_node] : NULL;
544 enum mlx5_expansion {
546 MLX5_EXPANSION_ROOT_OUTER,
547 MLX5_EXPANSION_OUTER_ETH,
548 MLX5_EXPANSION_OUTER_VLAN,
549 MLX5_EXPANSION_OUTER_IPV4,
550 MLX5_EXPANSION_OUTER_IPV4_UDP,
551 MLX5_EXPANSION_OUTER_IPV4_TCP,
552 MLX5_EXPANSION_OUTER_IPV6,
553 MLX5_EXPANSION_OUTER_IPV6_UDP,
554 MLX5_EXPANSION_OUTER_IPV6_TCP,
555 MLX5_EXPANSION_VXLAN,
556 MLX5_EXPANSION_STD_VXLAN,
557 MLX5_EXPANSION_L3_VXLAN,
558 MLX5_EXPANSION_VXLAN_GPE,
560 MLX5_EXPANSION_NVGRE,
561 MLX5_EXPANSION_GRE_KEY,
566 MLX5_EXPANSION_IPV4_UDP,
567 MLX5_EXPANSION_IPV4_TCP,
569 MLX5_EXPANSION_IPV6_UDP,
570 MLX5_EXPANSION_IPV6_TCP,
571 MLX5_EXPANSION_IPV6_FRAG_EXT,
573 MLX5_EXPANSION_GENEVE,
576 /** Supported expansion of items. */
577 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
578 [MLX5_EXPANSION_ROOT] = {
579 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
581 MLX5_EXPANSION_IPV6),
582 .type = RTE_FLOW_ITEM_TYPE_END,
584 [MLX5_EXPANSION_ROOT_OUTER] = {
585 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
586 MLX5_EXPANSION_OUTER_IPV4,
587 MLX5_EXPANSION_OUTER_IPV6),
588 .type = RTE_FLOW_ITEM_TYPE_END,
590 [MLX5_EXPANSION_OUTER_ETH] = {
591 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
592 .type = RTE_FLOW_ITEM_TYPE_ETH,
595 [MLX5_EXPANSION_OUTER_VLAN] = {
596 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
597 MLX5_EXPANSION_OUTER_IPV6),
598 .type = RTE_FLOW_ITEM_TYPE_VLAN,
599 .node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
601 [MLX5_EXPANSION_OUTER_IPV4] = {
602 .next = MLX5_FLOW_EXPAND_RSS_NEXT
603 (MLX5_EXPANSION_OUTER_IPV4_UDP,
604 MLX5_EXPANSION_OUTER_IPV4_TCP,
606 MLX5_EXPANSION_NVGRE,
608 MLX5_EXPANSION_IPV6),
609 .type = RTE_FLOW_ITEM_TYPE_IPV4,
610 .rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
611 RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
613 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
614 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
615 MLX5_EXPANSION_VXLAN_GPE,
617 MLX5_EXPANSION_GENEVE,
619 .type = RTE_FLOW_ITEM_TYPE_UDP,
620 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
622 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
623 .type = RTE_FLOW_ITEM_TYPE_TCP,
624 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
626 [MLX5_EXPANSION_OUTER_IPV6] = {
627 .next = MLX5_FLOW_EXPAND_RSS_NEXT
628 (MLX5_EXPANSION_OUTER_IPV6_UDP,
629 MLX5_EXPANSION_OUTER_IPV6_TCP,
633 MLX5_EXPANSION_NVGRE),
634 .type = RTE_FLOW_ITEM_TYPE_IPV6,
635 .rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
636 RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
638 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
639 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
640 MLX5_EXPANSION_VXLAN_GPE,
642 MLX5_EXPANSION_GENEVE,
644 .type = RTE_FLOW_ITEM_TYPE_UDP,
645 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
647 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
648 .type = RTE_FLOW_ITEM_TYPE_TCP,
649 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
651 [MLX5_EXPANSION_VXLAN] = {
652 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
654 MLX5_EXPANSION_IPV6),
655 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
657 [MLX5_EXPANSION_STD_VXLAN] = {
658 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
659 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
661 [MLX5_EXPANSION_L3_VXLAN] = {
662 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
663 MLX5_EXPANSION_IPV6),
664 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
666 [MLX5_EXPANSION_VXLAN_GPE] = {
667 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
669 MLX5_EXPANSION_IPV6),
670 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
672 [MLX5_EXPANSION_GRE] = {
673 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
676 MLX5_EXPANSION_GRE_KEY,
677 MLX5_EXPANSION_MPLS),
678 .type = RTE_FLOW_ITEM_TYPE_GRE,
680 [MLX5_EXPANSION_GRE_KEY] = {
681 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
683 MLX5_EXPANSION_MPLS),
684 .type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
685 .node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
687 [MLX5_EXPANSION_NVGRE] = {
688 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
689 .type = RTE_FLOW_ITEM_TYPE_NVGRE,
691 [MLX5_EXPANSION_MPLS] = {
692 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
695 .type = RTE_FLOW_ITEM_TYPE_MPLS,
696 .node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
698 [MLX5_EXPANSION_ETH] = {
699 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
700 .type = RTE_FLOW_ITEM_TYPE_ETH,
702 [MLX5_EXPANSION_VLAN] = {
703 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
704 MLX5_EXPANSION_IPV6),
705 .type = RTE_FLOW_ITEM_TYPE_VLAN,
706 .node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
708 [MLX5_EXPANSION_IPV4] = {
709 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
710 MLX5_EXPANSION_IPV4_TCP),
711 .type = RTE_FLOW_ITEM_TYPE_IPV4,
712 .rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
713 RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
715 [MLX5_EXPANSION_IPV4_UDP] = {
716 .type = RTE_FLOW_ITEM_TYPE_UDP,
717 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
719 [MLX5_EXPANSION_IPV4_TCP] = {
720 .type = RTE_FLOW_ITEM_TYPE_TCP,
721 .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
723 [MLX5_EXPANSION_IPV6] = {
724 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
725 MLX5_EXPANSION_IPV6_TCP,
726 MLX5_EXPANSION_IPV6_FRAG_EXT),
727 .type = RTE_FLOW_ITEM_TYPE_IPV6,
728 .rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
729 RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
731 [MLX5_EXPANSION_IPV6_UDP] = {
732 .type = RTE_FLOW_ITEM_TYPE_UDP,
733 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
735 [MLX5_EXPANSION_IPV6_TCP] = {
736 .type = RTE_FLOW_ITEM_TYPE_TCP,
737 .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
739 [MLX5_EXPANSION_IPV6_FRAG_EXT] = {
740 .type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
742 [MLX5_EXPANSION_GTP] = {
743 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
744 MLX5_EXPANSION_IPV6),
745 .type = RTE_FLOW_ITEM_TYPE_GTP,
747 [MLX5_EXPANSION_GENEVE] = {
748 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
750 MLX5_EXPANSION_IPV6),
751 .type = RTE_FLOW_ITEM_TYPE_GENEVE,
755 static struct rte_flow_action_handle *
756 mlx5_action_handle_create(struct rte_eth_dev *dev,
757 const struct rte_flow_indir_action_conf *conf,
758 const struct rte_flow_action *action,
759 struct rte_flow_error *error);
760 static int mlx5_action_handle_destroy
761 (struct rte_eth_dev *dev,
762 struct rte_flow_action_handle *handle,
763 struct rte_flow_error *error);
764 static int mlx5_action_handle_update
765 (struct rte_eth_dev *dev,
766 struct rte_flow_action_handle *handle,
768 struct rte_flow_error *error);
769 static int mlx5_action_handle_query
770 (struct rte_eth_dev *dev,
771 const struct rte_flow_action_handle *handle,
773 struct rte_flow_error *error);
775 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
776 struct rte_flow_tunnel *app_tunnel,
777 struct rte_flow_action **actions,
778 uint32_t *num_of_actions,
779 struct rte_flow_error *error);
781 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
782 struct rte_flow_tunnel *app_tunnel,
783 struct rte_flow_item **items,
784 uint32_t *num_of_items,
785 struct rte_flow_error *error);
787 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
788 struct rte_flow_item *pmd_items,
789 uint32_t num_items, struct rte_flow_error *err);
791 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
792 struct rte_flow_action *pmd_actions,
793 uint32_t num_actions,
794 struct rte_flow_error *err);
796 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
798 struct rte_flow_restore_info *info,
799 struct rte_flow_error *err);
800 static struct rte_flow_item_flex_handle *
801 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
802 const struct rte_flow_item_flex_conf *conf,
803 struct rte_flow_error *error);
805 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
806 const struct rte_flow_item_flex_handle *handle,
807 struct rte_flow_error *error);
809 static const struct rte_flow_ops mlx5_flow_ops = {
810 .validate = mlx5_flow_validate,
811 .create = mlx5_flow_create,
812 .destroy = mlx5_flow_destroy,
813 .flush = mlx5_flow_flush,
814 .isolate = mlx5_flow_isolate,
815 .query = mlx5_flow_query,
816 .dev_dump = mlx5_flow_dev_dump,
817 .get_aged_flows = mlx5_flow_get_aged_flows,
818 .action_handle_create = mlx5_action_handle_create,
819 .action_handle_destroy = mlx5_action_handle_destroy,
820 .action_handle_update = mlx5_action_handle_update,
821 .action_handle_query = mlx5_action_handle_query,
822 .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
823 .tunnel_match = mlx5_flow_tunnel_match,
824 .tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
825 .tunnel_item_release = mlx5_flow_tunnel_item_release,
826 .get_restore_info = mlx5_flow_tunnel_get_restore_info,
827 .flex_item_create = mlx5_flow_flex_item_create,
828 .flex_item_release = mlx5_flow_flex_item_release,
831 /* Tunnel information. */
832 struct mlx5_flow_tunnel_info {
833 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
834 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
837 static struct mlx5_flow_tunnel_info tunnels_info[] = {
839 .tunnel = MLX5_FLOW_LAYER_VXLAN,
840 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
843 .tunnel = MLX5_FLOW_LAYER_GENEVE,
844 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
847 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
848 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
851 .tunnel = MLX5_FLOW_LAYER_GRE,
852 .ptype = RTE_PTYPE_TUNNEL_GRE,
855 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
856 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
859 .tunnel = MLX5_FLOW_LAYER_MPLS,
860 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
863 .tunnel = MLX5_FLOW_LAYER_NVGRE,
864 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
867 .tunnel = MLX5_FLOW_LAYER_IPIP,
868 .ptype = RTE_PTYPE_TUNNEL_IP,
871 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
872 .ptype = RTE_PTYPE_TUNNEL_IP,
875 .tunnel = MLX5_FLOW_LAYER_GTP,
876 .ptype = RTE_PTYPE_TUNNEL_GTPU,
883 * Translate tag ID to register.
886 * Pointer to the Ethernet device structure.
888 * The feature that request the register.
890 * The request register ID.
892 * Error description in case of any.
895 * The request register on success, a negative errno
896 * value otherwise and rte_errno is set.
899 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
900 enum mlx5_feature_name feature,
902 struct rte_flow_error *error)
904 struct mlx5_priv *priv = dev->data->dev_private;
905 struct mlx5_sh_config *config = &priv->sh->config;
906 enum modify_reg start_reg;
907 bool skip_mtr_reg = false;
910 case MLX5_HAIRPIN_RX:
912 case MLX5_HAIRPIN_TX:
914 case MLX5_METADATA_RX:
915 switch (config->dv_xmeta_en) {
916 case MLX5_XMETA_MODE_LEGACY:
918 case MLX5_XMETA_MODE_META16:
920 case MLX5_XMETA_MODE_META32:
924 case MLX5_METADATA_TX:
926 case MLX5_METADATA_FDB:
927 switch (config->dv_xmeta_en) {
928 case MLX5_XMETA_MODE_LEGACY:
930 case MLX5_XMETA_MODE_META16:
932 case MLX5_XMETA_MODE_META32:
937 switch (config->dv_xmeta_en) {
938 case MLX5_XMETA_MODE_LEGACY:
940 case MLX5_XMETA_MODE_META16:
942 case MLX5_XMETA_MODE_META32:
948 * If meter color and meter id share one register, flow match
949 * should use the meter color register for match.
951 if (priv->mtr_reg_share)
952 return priv->mtr_color_reg;
954 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
957 case MLX5_ASO_FLOW_HIT:
958 case MLX5_ASO_CONNTRACK:
960 /* All features use the same REG_C. */
961 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
962 return priv->mtr_color_reg;
965 * Metadata COPY_MARK register using is in meter suffix sub
966 * flow while with meter. It's safe to share the same register.
968 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
971 * If meter is enable, it will engage the register for color
972 * match and flow match. If meter color match is not using the
973 * REG_C_2, need to skip the REG_C_x be used by meter color
975 * If meter is disable, free to use all available registers.
977 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
978 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
979 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
980 if (id > (uint32_t)(REG_C_7 - start_reg))
981 return rte_flow_error_set(error, EINVAL,
982 RTE_FLOW_ERROR_TYPE_ITEM,
983 NULL, "invalid tag id");
984 if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
985 return rte_flow_error_set(error, ENOTSUP,
986 RTE_FLOW_ERROR_TYPE_ITEM,
987 NULL, "unsupported tag id");
989 * This case means meter is using the REG_C_x great than 2.
990 * Take care not to conflict with meter color REG_C_x.
991 * If the available index REG_C_y >= REG_C_x, skip the
994 if (skip_mtr_reg && priv->sh->flow_mreg_c
995 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
996 if (id >= (uint32_t)(REG_C_7 - start_reg))
997 return rte_flow_error_set(error, EINVAL,
998 RTE_FLOW_ERROR_TYPE_ITEM,
999 NULL, "invalid tag id");
1000 if (priv->sh->flow_mreg_c
1001 [id + 1 + start_reg - REG_C_0] != REG_NON)
1002 return priv->sh->flow_mreg_c
1003 [id + 1 + start_reg - REG_C_0];
1004 return rte_flow_error_set(error, ENOTSUP,
1005 RTE_FLOW_ERROR_TYPE_ITEM,
1006 NULL, "unsupported tag id");
1008 return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
1011 return rte_flow_error_set(error, EINVAL,
1012 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1013 NULL, "invalid feature name");
1017 * Check extensive flow metadata register support.
1020 * Pointer to rte_eth_dev structure.
1023 * True if device supports extensive flow metadata register, otherwise false.
1026 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
1028 struct mlx5_priv *priv = dev->data->dev_private;
1031 * Having available reg_c can be regarded inclusively as supporting
1032 * extensive flow metadata register, which could mean,
1033 * - metadata register copy action by modify header.
1034 * - 16 modify header actions is supported.
1035 * - reg_c's are preserved across different domain (FDB and NIC) on
1036 * packet loopback by flow lookup miss.
1038 return priv->sh->flow_mreg_c[2] != REG_NON;
1042 * Get the lowest priority.
1045 * Pointer to the Ethernet device structure.
1046 * @param[in] attributes
1047 * Pointer to device flow rule attributes.
1050 * The value of lowest priority of flow.
1053 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
1054 const struct rte_flow_attr *attr)
1056 struct mlx5_priv *priv = dev->data->dev_private;
1058 if (!attr->group && !attr->transfer)
1059 return priv->sh->flow_max_priority - 2;
1060 return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1064 * Calculate matcher priority of the flow.
1067 * Pointer to the Ethernet device structure.
1069 * Pointer to device flow rule attributes.
1070 * @param[in] subpriority
1071 * The priority based on the items.
1072 * @param[in] external
1073 * Flow is user flow.
1075 * The matcher priority of the flow.
1078 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1079 const struct rte_flow_attr *attr,
1080 uint32_t subpriority, bool external)
1082 uint16_t priority = (uint16_t)attr->priority;
1083 struct mlx5_priv *priv = dev->data->dev_private;
1085 if (!attr->group && !attr->transfer) {
1086 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1087 priority = priv->sh->flow_max_priority - 1;
1088 return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1089 } else if (!external && attr->transfer && attr->group == 0 &&
1090 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1091 return (priv->sh->flow_max_priority - 1) * 3;
1093 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1094 priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1095 return priority * 3 + subpriority;
1099 * Verify the @p item specifications (spec, last, mask) are compatible with the
1103 * Item specification.
1105 * @p item->mask or flow default bit-masks.
1106 * @param[in] nic_mask
1107 * Bit-masks covering supported fields by the NIC to compare with user mask.
1109 * Bit-masks size in bytes.
1110 * @param[in] range_accepted
1111 * True if range of values is accepted for specific fields, false otherwise.
1113 * Pointer to error structure.
1116 * 0 on success, a negative errno value otherwise and rte_errno is set.
1119 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1120 const uint8_t *mask,
1121 const uint8_t *nic_mask,
1123 bool range_accepted,
1124 struct rte_flow_error *error)
1128 MLX5_ASSERT(nic_mask);
1129 for (i = 0; i < size; ++i)
1130 if ((nic_mask[i] | mask[i]) != nic_mask[i])
1131 return rte_flow_error_set(error, ENOTSUP,
1132 RTE_FLOW_ERROR_TYPE_ITEM,
1134 "mask enables non supported"
1136 if (!item->spec && (item->mask || item->last))
1137 return rte_flow_error_set(error, EINVAL,
1138 RTE_FLOW_ERROR_TYPE_ITEM, item,
1139 "mask/last without a spec is not"
1141 if (item->spec && item->last && !range_accepted) {
1147 for (i = 0; i < size; ++i) {
1148 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1149 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1151 ret = memcmp(spec, last, size);
1153 return rte_flow_error_set(error, EINVAL,
1154 RTE_FLOW_ERROR_TYPE_ITEM,
1156 "range is not valid");
1162 * Adjust the hash fields according to the @p flow information.
1164 * @param[in] dev_flow.
1165 * Pointer to the mlx5_flow.
1167 * 1 when the hash field is for a tunnel item.
1168 * @param[in] layer_types
1169 * RTE_ETH_RSS_* types.
1170 * @param[in] hash_fields
1174 * The hash fields that should be used.
1177 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1178 int tunnel __rte_unused, uint64_t layer_types,
1179 uint64_t hash_fields)
1181 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1182 int rss_request_inner = rss_desc->level >= 2;
1184 /* Check RSS hash level for tunnel. */
1185 if (tunnel && rss_request_inner)
1186 hash_fields |= IBV_RX_HASH_INNER;
1187 else if (tunnel || rss_request_inner)
1190 /* Check if requested layer matches RSS hash fields. */
1191 if (!(rss_desc->types & layer_types))
1197 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
1198 * if several tunnel rules are used on this queue, the tunnel ptype will be
1202 * Rx queue to update.
1205 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1208 uint32_t tunnel_ptype = 0;
1210 /* Look up for the ptype to use. */
1211 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1212 if (!rxq_ctrl->flow_tunnels_n[i])
1214 if (!tunnel_ptype) {
1215 tunnel_ptype = tunnels_info[i].ptype;
1221 rxq_ctrl->rxq.tunnel = tunnel_ptype;
1225 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the device
1229 * Pointer to the Ethernet device structure.
1230 * @param[in] dev_handle
1231 * Pointer to device flow handle structure.
1234 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1235 struct mlx5_flow_handle *dev_handle)
1237 struct mlx5_priv *priv = dev->data->dev_private;
1238 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1239 struct mlx5_ind_table_obj *ind_tbl = NULL;
1242 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1243 struct mlx5_hrxq *hrxq;
1245 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1246 dev_handle->rix_hrxq);
1248 ind_tbl = hrxq->ind_table;
1249 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1250 struct mlx5_shared_action_rss *shared_rss;
1252 shared_rss = mlx5_ipool_get
1253 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1254 dev_handle->rix_srss);
1256 ind_tbl = shared_rss->ind_tbl;
1260 for (i = 0; i != ind_tbl->queues_n; ++i) {
1261 int idx = ind_tbl->queues[i];
1262 struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1264 MLX5_ASSERT(rxq_ctrl != NULL);
1265 if (rxq_ctrl == NULL)
1268 * To support metadata register copy on Tx loopback,
1269 * this must be always enabled (metadata may arive
1270 * from other port - not from local flows only.
1275 /* Increase the counter matching the flow. */
1276 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1277 if ((tunnels_info[j].tunnel &
1278 dev_handle->layers) ==
1279 tunnels_info[j].tunnel) {
1280 rxq_ctrl->flow_tunnels_n[j]++;
1284 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1290 flow_rxq_mark_flag_set(struct rte_eth_dev *dev)
1292 struct mlx5_priv *priv = dev->data->dev_private;
1293 struct mlx5_rxq_ctrl *rxq_ctrl;
1295 if (priv->mark_enabled)
1297 LIST_FOREACH(rxq_ctrl, &priv->rxqsctrl, next) {
1298 rxq_ctrl->rxq.mark = 1;
1300 priv->mark_enabled = 1;
1304 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1307 * Pointer to the Ethernet device structure.
1309 * Pointer to flow structure.
1312 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1314 struct mlx5_priv *priv = dev->data->dev_private;
1315 uint32_t handle_idx;
1316 struct mlx5_flow_handle *dev_handle;
1317 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
1321 flow_rxq_mark_flag_set(dev);
1322 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1323 handle_idx, dev_handle, next)
1324 flow_drv_rxq_flags_set(dev, dev_handle);
1328 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1329 * device flow if no other flow uses it with the same kind of request.
1332 * Pointer to Ethernet device.
1333 * @param[in] dev_handle
1334 * Pointer to the device flow handle structure.
1337 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1338 struct mlx5_flow_handle *dev_handle)
1340 struct mlx5_priv *priv = dev->data->dev_private;
1341 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1342 struct mlx5_ind_table_obj *ind_tbl = NULL;
1345 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1346 struct mlx5_hrxq *hrxq;
1348 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1349 dev_handle->rix_hrxq);
1351 ind_tbl = hrxq->ind_table;
1352 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1353 struct mlx5_shared_action_rss *shared_rss;
1355 shared_rss = mlx5_ipool_get
1356 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1357 dev_handle->rix_srss);
1359 ind_tbl = shared_rss->ind_tbl;
1363 MLX5_ASSERT(dev->data->dev_started);
1364 for (i = 0; i != ind_tbl->queues_n; ++i) {
1365 int idx = ind_tbl->queues[i];
1366 struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1368 MLX5_ASSERT(rxq_ctrl != NULL);
1369 if (rxq_ctrl == NULL)
1374 /* Decrease the counter matching the flow. */
1375 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1376 if ((tunnels_info[j].tunnel &
1377 dev_handle->layers) ==
1378 tunnels_info[j].tunnel) {
1379 rxq_ctrl->flow_tunnels_n[j]--;
1383 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1389 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1390 * @p flow if no other flow uses it with the same kind of request.
1393 * Pointer to Ethernet device.
1395 * Pointer to the flow.
1398 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1400 struct mlx5_priv *priv = dev->data->dev_private;
1401 uint32_t handle_idx;
1402 struct mlx5_flow_handle *dev_handle;
1404 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1405 handle_idx, dev_handle, next)
1406 flow_drv_rxq_flags_trim(dev, dev_handle);
1410 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1413 * Pointer to Ethernet device.
1416 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1418 struct mlx5_priv *priv = dev->data->dev_private;
1421 for (i = 0; i != priv->rxqs_n; ++i) {
1422 struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1425 if (rxq == NULL || rxq->ctrl == NULL)
1427 rxq->ctrl->rxq.mark = 0;
1428 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1429 rxq->ctrl->flow_tunnels_n[j] = 0;
1430 rxq->ctrl->rxq.tunnel = 0;
1432 priv->mark_enabled = 0;
1436 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1439 * Pointer to the Ethernet device structure.
1442 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1444 struct mlx5_priv *priv = dev->data->dev_private;
1447 for (i = 0; i != priv->rxqs_n; ++i) {
1448 struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1449 struct mlx5_rxq_data *data;
1451 if (rxq == NULL || rxq->ctrl == NULL)
1453 data = &rxq->ctrl->rxq;
1454 if (!rte_flow_dynf_metadata_avail()) {
1455 data->dynf_meta = 0;
1456 data->flow_meta_mask = 0;
1457 data->flow_meta_offset = -1;
1458 data->flow_meta_port_mask = 0;
1460 data->dynf_meta = 1;
1461 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1462 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1463 data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1469 * return a pointer to the desired action in the list of actions.
1471 * @param[in] actions
1472 * The list of actions to search the action in.
1474 * The action to find.
1477 * Pointer to the action in the list, if found. NULL otherwise.
1479 const struct rte_flow_action *
1480 mlx5_flow_find_action(const struct rte_flow_action *actions,
1481 enum rte_flow_action_type action)
1483 if (actions == NULL)
1485 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1486 if (actions->type == action)
1492 * Validate the flag action.
1494 * @param[in] action_flags
1495 * Bit-fields that holds the actions detected until now.
1497 * Attributes of flow that includes this action.
1499 * Pointer to error structure.
1502 * 0 on success, a negative errno value otherwise and rte_errno is set.
1505 mlx5_flow_validate_action_flag(uint64_t action_flags,
1506 const struct rte_flow_attr *attr,
1507 struct rte_flow_error *error)
1509 if (action_flags & MLX5_FLOW_ACTION_MARK)
1510 return rte_flow_error_set(error, EINVAL,
1511 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1512 "can't mark and flag in same flow");
1513 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1514 return rte_flow_error_set(error, EINVAL,
1515 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1517 " actions in same flow");
1519 return rte_flow_error_set(error, ENOTSUP,
1520 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1521 "flag action not supported for "
1527 * Validate the mark action.
1530 * Pointer to the queue action.
1531 * @param[in] action_flags
1532 * Bit-fields that holds the actions detected until now.
1534 * Attributes of flow that includes this action.
1536 * Pointer to error structure.
1539 * 0 on success, a negative errno value otherwise and rte_errno is set.
1542 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1543 uint64_t action_flags,
1544 const struct rte_flow_attr *attr,
1545 struct rte_flow_error *error)
1547 const struct rte_flow_action_mark *mark = action->conf;
1550 return rte_flow_error_set(error, EINVAL,
1551 RTE_FLOW_ERROR_TYPE_ACTION,
1553 "configuration cannot be null");
1554 if (mark->id >= MLX5_FLOW_MARK_MAX)
1555 return rte_flow_error_set(error, EINVAL,
1556 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1558 "mark id must in 0 <= id < "
1559 RTE_STR(MLX5_FLOW_MARK_MAX));
1560 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1561 return rte_flow_error_set(error, EINVAL,
1562 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1563 "can't flag and mark in same flow");
1564 if (action_flags & MLX5_FLOW_ACTION_MARK)
1565 return rte_flow_error_set(error, EINVAL,
1566 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1567 "can't have 2 mark actions in same"
1570 return rte_flow_error_set(error, ENOTSUP,
1571 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1572 "mark action not supported for "
1578 * Validate the drop action.
1580 * @param[in] action_flags
1581 * Bit-fields that holds the actions detected until now.
1583 * Attributes of flow that includes this action.
1585 * Pointer to error structure.
1588 * 0 on success, a negative errno value otherwise and rte_errno is set.
1591 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1592 const struct rte_flow_attr *attr,
1593 struct rte_flow_error *error)
1596 return rte_flow_error_set(error, ENOTSUP,
1597 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1598 "drop action not supported for "
1604 * Validate the queue action.
1607 * Pointer to the queue action.
1608 * @param[in] action_flags
1609 * Bit-fields that holds the actions detected until now.
1611 * Pointer to the Ethernet device structure.
1613 * Attributes of flow that includes this action.
1615 * Pointer to error structure.
1618 * 0 on success, a negative errno value otherwise and rte_errno is set.
1621 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1622 uint64_t action_flags,
1623 struct rte_eth_dev *dev,
1624 const struct rte_flow_attr *attr,
1625 struct rte_flow_error *error)
1627 struct mlx5_priv *priv = dev->data->dev_private;
1628 const struct rte_flow_action_queue *queue = action->conf;
1630 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1631 return rte_flow_error_set(error, EINVAL,
1632 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1633 "can't have 2 fate actions in"
1636 return rte_flow_error_set(error, EINVAL,
1637 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1638 NULL, "No Rx queues configured");
1639 if (queue->index >= priv->rxqs_n)
1640 return rte_flow_error_set(error, EINVAL,
1641 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1643 "queue index out of range");
1644 if (mlx5_rxq_get(dev, queue->index) == NULL)
1645 return rte_flow_error_set(error, EINVAL,
1646 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1648 "queue is not configured");
1650 return rte_flow_error_set(error, ENOTSUP,
1651 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1652 "queue action not supported for "
1658 * Validate queue numbers for device RSS.
1661 * Configured device.
1663 * Array of queue numbers.
1664 * @param[in] queues_n
1665 * Size of the @p queues array.
1667 * On error, filled with a textual error description.
1669 * On error, filled with an offending queue index in @p queues array.
1672 * 0 on success, a negative errno code on error.
1675 mlx5_validate_rss_queues(struct rte_eth_dev *dev,
1676 const uint16_t *queues, uint32_t queues_n,
1677 const char **error, uint32_t *queue_idx)
1679 const struct mlx5_priv *priv = dev->data->dev_private;
1680 enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1683 for (i = 0; i != queues_n; ++i) {
1684 struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev,
1687 if (queues[i] >= priv->rxqs_n) {
1688 *error = "queue index out of range";
1692 if (rxq_ctrl == NULL) {
1693 *error = "queue is not configured";
1698 rxq_type = rxq_ctrl->type;
1699 if (rxq_type != rxq_ctrl->type) {
1700 *error = "combining hairpin and regular RSS queues is not supported";
1709 * Validate the rss action.
1712 * Pointer to the Ethernet device structure.
1714 * Pointer to the queue action.
1716 * Pointer to error structure.
1719 * 0 on success, a negative errno value otherwise and rte_errno is set.
1722 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1723 const struct rte_flow_action *action,
1724 struct rte_flow_error *error)
1726 struct mlx5_priv *priv = dev->data->dev_private;
1727 const struct rte_flow_action_rss *rss = action->conf;
1729 const char *message;
1732 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1733 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1734 return rte_flow_error_set(error, ENOTSUP,
1735 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1737 "RSS hash function not supported");
1738 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1743 return rte_flow_error_set(error, ENOTSUP,
1744 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1746 "tunnel RSS is not supported");
1747 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1748 if (rss->key_len == 0 && rss->key != NULL)
1749 return rte_flow_error_set(error, ENOTSUP,
1750 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1752 "RSS hash key length 0");
1753 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1754 return rte_flow_error_set(error, ENOTSUP,
1755 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1757 "RSS hash key too small");
1758 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1759 return rte_flow_error_set(error, ENOTSUP,
1760 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1762 "RSS hash key too large");
1763 if (rss->queue_num > priv->sh->dev_cap.ind_table_max_size)
1764 return rte_flow_error_set(error, ENOTSUP,
1765 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1767 "number of queues too large");
1768 if (rss->types & MLX5_RSS_HF_MASK)
1769 return rte_flow_error_set(error, ENOTSUP,
1770 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1772 "some RSS protocols are not"
1774 if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1775 !(rss->types & RTE_ETH_RSS_IP))
1776 return rte_flow_error_set(error, EINVAL,
1777 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1778 "L3 partial RSS requested but L3 RSS"
1779 " type not specified");
1780 if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1781 !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1782 return rte_flow_error_set(error, EINVAL,
1783 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1784 "L4 partial RSS requested but L4 RSS"
1785 " type not specified");
1787 return rte_flow_error_set(error, EINVAL,
1788 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1789 NULL, "No Rx queues configured");
1790 if (!rss->queue_num)
1791 return rte_flow_error_set(error, EINVAL,
1792 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1793 NULL, "No queues configured");
1794 ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1795 &message, &queue_idx);
1797 return rte_flow_error_set(error, -ret,
1798 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1799 &rss->queue[queue_idx], message);
1805 * Validate the rss action.
1808 * Pointer to the queue action.
1809 * @param[in] action_flags
1810 * Bit-fields that holds the actions detected until now.
1812 * Pointer to the Ethernet device structure.
1814 * Attributes of flow that includes this action.
1815 * @param[in] item_flags
1816 * Items that were detected.
1818 * Pointer to error structure.
1821 * 0 on success, a negative errno value otherwise and rte_errno is set.
1824 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1825 uint64_t action_flags,
1826 struct rte_eth_dev *dev,
1827 const struct rte_flow_attr *attr,
1828 uint64_t item_flags,
1829 struct rte_flow_error *error)
1831 const struct rte_flow_action_rss *rss = action->conf;
1832 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1835 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1836 return rte_flow_error_set(error, EINVAL,
1837 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1838 "can't have 2 fate actions"
1840 ret = mlx5_validate_action_rss(dev, action, error);
1844 return rte_flow_error_set(error, ENOTSUP,
1845 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1846 "rss action not supported for "
1848 if (rss->level > 1 && !tunnel)
1849 return rte_flow_error_set(error, EINVAL,
1850 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1851 "inner RSS is not supported for "
1852 "non-tunnel flows");
1853 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1854 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1855 return rte_flow_error_set(error, EINVAL,
1856 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1857 "RSS on eCPRI is not supported now");
1859 if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1861 (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1863 return rte_flow_error_set(error, EINVAL,
1864 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1865 "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1870 * Validate the default miss action.
1872 * @param[in] action_flags
1873 * Bit-fields that holds the actions detected until now.
1875 * Pointer to error structure.
1878 * 0 on success, a negative errno value otherwise and rte_errno is set.
1881 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1882 const struct rte_flow_attr *attr,
1883 struct rte_flow_error *error)
1885 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1886 return rte_flow_error_set(error, EINVAL,
1887 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1888 "can't have 2 fate actions in"
1891 return rte_flow_error_set(error, ENOTSUP,
1892 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1893 "default miss action not supported "
1896 return rte_flow_error_set(error, ENOTSUP,
1897 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1898 "only group 0 is supported");
1900 return rte_flow_error_set(error, ENOTSUP,
1901 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1902 NULL, "transfer is not supported");
1907 * Validate the count action.
1910 * Pointer to the Ethernet device structure.
1912 * Attributes of flow that includes this action.
1914 * Pointer to error structure.
1917 * 0 on success, a negative errno value otherwise and rte_errno is set.
1920 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1921 const struct rte_flow_attr *attr,
1922 struct rte_flow_error *error)
1925 return rte_flow_error_set(error, ENOTSUP,
1926 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1927 "count action not supported for "
1933 * Validate the ASO CT action.
1936 * Pointer to the Ethernet device structure.
1937 * @param[in] conntrack
1938 * Pointer to the CT action profile.
1940 * Pointer to error structure.
1943 * 0 on success, a negative errno value otherwise and rte_errno is set.
1946 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1947 const struct rte_flow_action_conntrack *conntrack,
1948 struct rte_flow_error *error)
1952 if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1953 return rte_flow_error_set(error, EINVAL,
1954 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1955 "Invalid CT state");
1956 if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1957 return rte_flow_error_set(error, EINVAL,
1958 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1959 "Invalid last TCP packet flag");
1964 * Verify the @p attributes will be correctly understood by the NIC and store
1965 * them in the @p flow if everything is correct.
1968 * Pointer to the Ethernet device structure.
1969 * @param[in] attributes
1970 * Pointer to flow attributes
1972 * Pointer to error structure.
1975 * 0 on success, a negative errno value otherwise and rte_errno is set.
1978 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1979 const struct rte_flow_attr *attributes,
1980 struct rte_flow_error *error)
1982 struct mlx5_priv *priv = dev->data->dev_private;
1983 uint32_t priority_max = priv->sh->flow_max_priority - 1;
1985 if (attributes->group)
1986 return rte_flow_error_set(error, ENOTSUP,
1987 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1988 NULL, "groups is not supported");
1989 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1990 attributes->priority >= priority_max)
1991 return rte_flow_error_set(error, ENOTSUP,
1992 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1993 NULL, "priority out of range");
1994 if (attributes->egress)
1995 return rte_flow_error_set(error, ENOTSUP,
1996 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1997 "egress is not supported");
1998 if (attributes->transfer && !priv->sh->config.dv_esw_en)
1999 return rte_flow_error_set(error, ENOTSUP,
2000 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
2001 NULL, "transfer is not supported");
2002 if (!attributes->ingress)
2003 return rte_flow_error_set(error, EINVAL,
2004 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
2006 "ingress attribute is mandatory");
2011 * Validate ICMP6 item.
2014 * Item specification.
2015 * @param[in] item_flags
2016 * Bit-fields that holds the items detected until now.
2017 * @param[in] ext_vlan_sup
2018 * Whether extended VLAN features are supported or not.
2020 * Pointer to error structure.
2023 * 0 on success, a negative errno value otherwise and rte_errno is set.
2026 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
2027 uint64_t item_flags,
2028 uint8_t target_protocol,
2029 struct rte_flow_error *error)
2031 const struct rte_flow_item_icmp6 *mask = item->mask;
2032 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2033 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2034 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2035 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2036 MLX5_FLOW_LAYER_OUTER_L4;
2039 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
2040 return rte_flow_error_set(error, EINVAL,
2041 RTE_FLOW_ERROR_TYPE_ITEM, item,
2042 "protocol filtering not compatible"
2043 " with ICMP6 layer");
2044 if (!(item_flags & l3m))
2045 return rte_flow_error_set(error, EINVAL,
2046 RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 "IPv6 is mandatory to filter on"
2049 if (item_flags & l4m)
2050 return rte_flow_error_set(error, EINVAL,
2051 RTE_FLOW_ERROR_TYPE_ITEM, item,
2052 "multiple L4 layers not supported");
2054 mask = &rte_flow_item_icmp6_mask;
2055 ret = mlx5_flow_item_acceptable
2056 (item, (const uint8_t *)mask,
2057 (const uint8_t *)&rte_flow_item_icmp6_mask,
2058 sizeof(struct rte_flow_item_icmp6),
2059 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2066 * Validate ICMP item.
2069 * Item specification.
2070 * @param[in] item_flags
2071 * Bit-fields that holds the items detected until now.
2073 * Pointer to error structure.
2076 * 0 on success, a negative errno value otherwise and rte_errno is set.
2079 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2080 uint64_t item_flags,
2081 uint8_t target_protocol,
2082 struct rte_flow_error *error)
2084 const struct rte_flow_item_icmp *mask = item->mask;
2085 const struct rte_flow_item_icmp nic_mask = {
2086 .hdr.icmp_type = 0xff,
2087 .hdr.icmp_code = 0xff,
2088 .hdr.icmp_ident = RTE_BE16(0xffff),
2089 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
2091 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2092 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2093 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2094 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2095 MLX5_FLOW_LAYER_OUTER_L4;
2098 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2099 return rte_flow_error_set(error, EINVAL,
2100 RTE_FLOW_ERROR_TYPE_ITEM, item,
2101 "protocol filtering not compatible"
2102 " with ICMP layer");
2103 if (!(item_flags & l3m))
2104 return rte_flow_error_set(error, EINVAL,
2105 RTE_FLOW_ERROR_TYPE_ITEM, item,
2106 "IPv4 is mandatory to filter"
2108 if (item_flags & l4m)
2109 return rte_flow_error_set(error, EINVAL,
2110 RTE_FLOW_ERROR_TYPE_ITEM, item,
2111 "multiple L4 layers not supported");
2114 ret = mlx5_flow_item_acceptable
2115 (item, (const uint8_t *)mask,
2116 (const uint8_t *)&nic_mask,
2117 sizeof(struct rte_flow_item_icmp),
2118 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2125 * Validate Ethernet item.
2128 * Item specification.
2129 * @param[in] item_flags
2130 * Bit-fields that holds the items detected until now.
2132 * Pointer to error structure.
2135 * 0 on success, a negative errno value otherwise and rte_errno is set.
2138 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2139 uint64_t item_flags, bool ext_vlan_sup,
2140 struct rte_flow_error *error)
2142 const struct rte_flow_item_eth *mask = item->mask;
2143 const struct rte_flow_item_eth nic_mask = {
2144 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2145 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2146 .type = RTE_BE16(0xffff),
2147 .has_vlan = ext_vlan_sup ? 1 : 0,
2150 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2151 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
2152 MLX5_FLOW_LAYER_OUTER_L2;
2154 if (item_flags & ethm)
2155 return rte_flow_error_set(error, ENOTSUP,
2156 RTE_FLOW_ERROR_TYPE_ITEM, item,
2157 "multiple L2 layers not supported");
2158 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2159 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2160 return rte_flow_error_set(error, EINVAL,
2161 RTE_FLOW_ERROR_TYPE_ITEM, item,
2162 "L2 layer should not follow "
2164 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2165 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2166 return rte_flow_error_set(error, EINVAL,
2167 RTE_FLOW_ERROR_TYPE_ITEM, item,
2168 "L2 layer should not follow VLAN");
2169 if (item_flags & MLX5_FLOW_LAYER_GTP)
2170 return rte_flow_error_set(error, EINVAL,
2171 RTE_FLOW_ERROR_TYPE_ITEM, item,
2172 "L2 layer should not follow GTP");
2174 mask = &rte_flow_item_eth_mask;
2175 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2176 (const uint8_t *)&nic_mask,
2177 sizeof(struct rte_flow_item_eth),
2178 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2183 * Validate VLAN item.
2186 * Item specification.
2187 * @param[in] item_flags
2188 * Bit-fields that holds the items detected until now.
2190 * Ethernet device flow is being created on.
2192 * Pointer to error structure.
2195 * 0 on success, a negative errno value otherwise and rte_errno is set.
2198 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2199 uint64_t item_flags,
2200 struct rte_eth_dev *dev,
2201 struct rte_flow_error *error)
2203 const struct rte_flow_item_vlan *spec = item->spec;
2204 const struct rte_flow_item_vlan *mask = item->mask;
2205 const struct rte_flow_item_vlan nic_mask = {
2206 .tci = RTE_BE16(UINT16_MAX),
2207 .inner_type = RTE_BE16(UINT16_MAX),
2209 uint16_t vlan_tag = 0;
2210 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2212 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2213 MLX5_FLOW_LAYER_INNER_L4) :
2214 (MLX5_FLOW_LAYER_OUTER_L3 |
2215 MLX5_FLOW_LAYER_OUTER_L4);
2216 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2217 MLX5_FLOW_LAYER_OUTER_VLAN;
2219 if (item_flags & vlanm)
2220 return rte_flow_error_set(error, EINVAL,
2221 RTE_FLOW_ERROR_TYPE_ITEM, item,
2222 "multiple VLAN layers not supported");
2223 else if ((item_flags & l34m) != 0)
2224 return rte_flow_error_set(error, EINVAL,
2225 RTE_FLOW_ERROR_TYPE_ITEM, item,
2226 "VLAN cannot follow L3/L4 layer");
2228 mask = &rte_flow_item_vlan_mask;
2229 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2230 (const uint8_t *)&nic_mask,
2231 sizeof(struct rte_flow_item_vlan),
2232 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2235 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2236 struct mlx5_priv *priv = dev->data->dev_private;
2238 if (priv->vmwa_context) {
2240 * Non-NULL context means we have a virtual machine
2241 * and SR-IOV enabled, we have to create VLAN interface
2242 * to make hypervisor to setup E-Switch vport
2243 * context correctly. We avoid creating the multiple
2244 * VLAN interfaces, so we cannot support VLAN tag mask.
2246 return rte_flow_error_set(error, EINVAL,
2247 RTE_FLOW_ERROR_TYPE_ITEM,
2249 "VLAN tag mask is not"
2250 " supported in virtual"
2255 vlan_tag = spec->tci;
2256 vlan_tag &= mask->tci;
2259 * From verbs perspective an empty VLAN is equivalent
2260 * to a packet without VLAN layer.
2263 return rte_flow_error_set(error, EINVAL,
2264 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2266 "VLAN cannot be empty");
2271 * Validate IPV4 item.
2274 * Item specification.
2275 * @param[in] item_flags
2276 * Bit-fields that holds the items detected until now.
2277 * @param[in] last_item
2278 * Previous validated item in the pattern items.
2279 * @param[in] ether_type
2280 * Type in the ethernet layer header (including dot1q).
2281 * @param[in] acc_mask
2282 * Acceptable mask, if NULL default internal default mask
2283 * will be used to check whether item fields are supported.
2284 * @param[in] range_accepted
2285 * True if range of values is accepted for specific fields, false otherwise.
2287 * Pointer to error structure.
2290 * 0 on success, a negative errno value otherwise and rte_errno is set.
2293 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2294 uint64_t item_flags,
2296 uint16_t ether_type,
2297 const struct rte_flow_item_ipv4 *acc_mask,
2298 bool range_accepted,
2299 struct rte_flow_error *error)
2301 const struct rte_flow_item_ipv4 *mask = item->mask;
2302 const struct rte_flow_item_ipv4 *spec = item->spec;
2303 const struct rte_flow_item_ipv4 nic_mask = {
2305 .src_addr = RTE_BE32(0xffffffff),
2306 .dst_addr = RTE_BE32(0xffffffff),
2307 .type_of_service = 0xff,
2308 .next_proto_id = 0xff,
2311 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2312 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2313 MLX5_FLOW_LAYER_OUTER_L3;
2314 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2315 MLX5_FLOW_LAYER_OUTER_L4;
2317 uint8_t next_proto = 0xFF;
2318 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2319 MLX5_FLOW_LAYER_OUTER_VLAN |
2320 MLX5_FLOW_LAYER_INNER_VLAN);
2322 if ((last_item & l2_vlan) && ether_type &&
2323 ether_type != RTE_ETHER_TYPE_IPV4)
2324 return rte_flow_error_set(error, EINVAL,
2325 RTE_FLOW_ERROR_TYPE_ITEM, item,
2326 "IPv4 cannot follow L2/VLAN layer "
2327 "which ether type is not IPv4");
2328 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2330 next_proto = mask->hdr.next_proto_id &
2331 spec->hdr.next_proto_id;
2332 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2333 return rte_flow_error_set(error, EINVAL,
2334 RTE_FLOW_ERROR_TYPE_ITEM,
2339 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2340 return rte_flow_error_set(error, EINVAL,
2341 RTE_FLOW_ERROR_TYPE_ITEM, item,
2342 "wrong tunnel type - IPv6 specified "
2343 "but IPv4 item provided");
2344 if (item_flags & l3m)
2345 return rte_flow_error_set(error, ENOTSUP,
2346 RTE_FLOW_ERROR_TYPE_ITEM, item,
2347 "multiple L3 layers not supported");
2348 else if (item_flags & l4m)
2349 return rte_flow_error_set(error, EINVAL,
2350 RTE_FLOW_ERROR_TYPE_ITEM, item,
2351 "L3 cannot follow an L4 layer.");
2352 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2353 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2354 return rte_flow_error_set(error, EINVAL,
2355 RTE_FLOW_ERROR_TYPE_ITEM, item,
2356 "L3 cannot follow an NVGRE layer.");
2358 mask = &rte_flow_item_ipv4_mask;
2359 else if (mask->hdr.next_proto_id != 0 &&
2360 mask->hdr.next_proto_id != 0xff)
2361 return rte_flow_error_set(error, EINVAL,
2362 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2363 "partial mask is not supported"
2365 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2366 acc_mask ? (const uint8_t *)acc_mask
2367 : (const uint8_t *)&nic_mask,
2368 sizeof(struct rte_flow_item_ipv4),
2369 range_accepted, error);
2376 * Validate IPV6 item.
2379 * Item specification.
2380 * @param[in] item_flags
2381 * Bit-fields that holds the items detected until now.
2382 * @param[in] last_item
2383 * Previous validated item in the pattern items.
2384 * @param[in] ether_type
2385 * Type in the ethernet layer header (including dot1q).
2386 * @param[in] acc_mask
2387 * Acceptable mask, if NULL default internal default mask
2388 * will be used to check whether item fields are supported.
2390 * Pointer to error structure.
2393 * 0 on success, a negative errno value otherwise and rte_errno is set.
2396 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2397 uint64_t item_flags,
2399 uint16_t ether_type,
2400 const struct rte_flow_item_ipv6 *acc_mask,
2401 struct rte_flow_error *error)
2403 const struct rte_flow_item_ipv6 *mask = item->mask;
2404 const struct rte_flow_item_ipv6 *spec = item->spec;
2405 const struct rte_flow_item_ipv6 nic_mask = {
2408 "\xff\xff\xff\xff\xff\xff\xff\xff"
2409 "\xff\xff\xff\xff\xff\xff\xff\xff",
2411 "\xff\xff\xff\xff\xff\xff\xff\xff"
2412 "\xff\xff\xff\xff\xff\xff\xff\xff",
2413 .vtc_flow = RTE_BE32(0xffffffff),
2417 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2418 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2419 MLX5_FLOW_LAYER_OUTER_L3;
2420 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2421 MLX5_FLOW_LAYER_OUTER_L4;
2423 uint8_t next_proto = 0xFF;
2424 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2425 MLX5_FLOW_LAYER_OUTER_VLAN |
2426 MLX5_FLOW_LAYER_INNER_VLAN);
2428 if ((last_item & l2_vlan) && ether_type &&
2429 ether_type != RTE_ETHER_TYPE_IPV6)
2430 return rte_flow_error_set(error, EINVAL,
2431 RTE_FLOW_ERROR_TYPE_ITEM, item,
2432 "IPv6 cannot follow L2/VLAN layer "
2433 "which ether type is not IPv6");
2434 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2435 next_proto = spec->hdr.proto;
2436 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2437 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2438 return rte_flow_error_set(error, EINVAL,
2439 RTE_FLOW_ERROR_TYPE_ITEM,
2444 if (next_proto == IPPROTO_HOPOPTS ||
2445 next_proto == IPPROTO_ROUTING ||
2446 next_proto == IPPROTO_FRAGMENT ||
2447 next_proto == IPPROTO_ESP ||
2448 next_proto == IPPROTO_AH ||
2449 next_proto == IPPROTO_DSTOPTS)
2450 return rte_flow_error_set(error, EINVAL,
2451 RTE_FLOW_ERROR_TYPE_ITEM, item,
2452 "IPv6 proto (next header) should "
2453 "not be set as extension header");
2454 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2455 return rte_flow_error_set(error, EINVAL,
2456 RTE_FLOW_ERROR_TYPE_ITEM, item,
2457 "wrong tunnel type - IPv4 specified "
2458 "but IPv6 item provided");
2459 if (item_flags & l3m)
2460 return rte_flow_error_set(error, ENOTSUP,
2461 RTE_FLOW_ERROR_TYPE_ITEM, item,
2462 "multiple L3 layers not supported");
2463 else if (item_flags & l4m)
2464 return rte_flow_error_set(error, EINVAL,
2465 RTE_FLOW_ERROR_TYPE_ITEM, item,
2466 "L3 cannot follow an L4 layer.");
2467 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2468 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2469 return rte_flow_error_set(error, EINVAL,
2470 RTE_FLOW_ERROR_TYPE_ITEM, item,
2471 "L3 cannot follow an NVGRE layer.");
2473 mask = &rte_flow_item_ipv6_mask;
2474 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2475 acc_mask ? (const uint8_t *)acc_mask
2476 : (const uint8_t *)&nic_mask,
2477 sizeof(struct rte_flow_item_ipv6),
2478 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2485 * Validate UDP item.
2488 * Item specification.
2489 * @param[in] item_flags
2490 * Bit-fields that holds the items detected until now.
2491 * @param[in] target_protocol
2492 * The next protocol in the previous item.
2493 * @param[in] flow_mask
2494 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2496 * Pointer to error structure.
2499 * 0 on success, a negative errno value otherwise and rte_errno is set.
2502 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2503 uint64_t item_flags,
2504 uint8_t target_protocol,
2505 struct rte_flow_error *error)
2507 const struct rte_flow_item_udp *mask = item->mask;
2508 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2509 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2510 MLX5_FLOW_LAYER_OUTER_L3;
2511 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2512 MLX5_FLOW_LAYER_OUTER_L4;
2515 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2516 return rte_flow_error_set(error, EINVAL,
2517 RTE_FLOW_ERROR_TYPE_ITEM, item,
2518 "protocol filtering not compatible"
2520 if (!(item_flags & l3m))
2521 return rte_flow_error_set(error, EINVAL,
2522 RTE_FLOW_ERROR_TYPE_ITEM, item,
2523 "L3 is mandatory to filter on L4");
2524 if (item_flags & l4m)
2525 return rte_flow_error_set(error, EINVAL,
2526 RTE_FLOW_ERROR_TYPE_ITEM, item,
2527 "multiple L4 layers not supported");
2529 mask = &rte_flow_item_udp_mask;
2530 ret = mlx5_flow_item_acceptable
2531 (item, (const uint8_t *)mask,
2532 (const uint8_t *)&rte_flow_item_udp_mask,
2533 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2541 * Validate TCP item.
2544 * Item specification.
2545 * @param[in] item_flags
2546 * Bit-fields that holds the items detected until now.
2547 * @param[in] target_protocol
2548 * The next protocol in the previous item.
2550 * Pointer to error structure.
2553 * 0 on success, a negative errno value otherwise and rte_errno is set.
2556 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2557 uint64_t item_flags,
2558 uint8_t target_protocol,
2559 const struct rte_flow_item_tcp *flow_mask,
2560 struct rte_flow_error *error)
2562 const struct rte_flow_item_tcp *mask = item->mask;
2563 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2564 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2565 MLX5_FLOW_LAYER_OUTER_L3;
2566 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2567 MLX5_FLOW_LAYER_OUTER_L4;
2570 MLX5_ASSERT(flow_mask);
2571 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2572 return rte_flow_error_set(error, EINVAL,
2573 RTE_FLOW_ERROR_TYPE_ITEM, item,
2574 "protocol filtering not compatible"
2576 if (!(item_flags & l3m))
2577 return rte_flow_error_set(error, EINVAL,
2578 RTE_FLOW_ERROR_TYPE_ITEM, item,
2579 "L3 is mandatory to filter on L4");
2580 if (item_flags & l4m)
2581 return rte_flow_error_set(error, EINVAL,
2582 RTE_FLOW_ERROR_TYPE_ITEM, item,
2583 "multiple L4 layers not supported");
2585 mask = &rte_flow_item_tcp_mask;
2586 ret = mlx5_flow_item_acceptable
2587 (item, (const uint8_t *)mask,
2588 (const uint8_t *)flow_mask,
2589 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2597 * Validate VXLAN item.
2600 * Pointer to the Ethernet device structure.
2601 * @param[in] udp_dport
2602 * UDP destination port
2604 * Item specification.
2605 * @param[in] item_flags
2606 * Bit-fields that holds the items detected until now.
2608 * Flow rule attributes.
2610 * Pointer to error structure.
2613 * 0 on success, a negative errno value otherwise and rte_errno is set.
2616 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2618 const struct rte_flow_item *item,
2619 uint64_t item_flags,
2620 const struct rte_flow_attr *attr,
2621 struct rte_flow_error *error)
2623 const struct rte_flow_item_vxlan *spec = item->spec;
2624 const struct rte_flow_item_vxlan *mask = item->mask;
2626 struct mlx5_priv *priv = dev->data->dev_private;
2630 } id = { .vlan_id = 0, };
2631 const struct rte_flow_item_vxlan nic_mask = {
2632 .vni = "\xff\xff\xff",
2635 const struct rte_flow_item_vxlan *valid_mask;
2637 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2638 return rte_flow_error_set(error, ENOTSUP,
2639 RTE_FLOW_ERROR_TYPE_ITEM, item,
2640 "multiple tunnel layers not"
2642 valid_mask = &rte_flow_item_vxlan_mask;
2644 * Verify only UDPv4 is present as defined in
2645 * https://tools.ietf.org/html/rfc7348
2647 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2648 return rte_flow_error_set(error, EINVAL,
2649 RTE_FLOW_ERROR_TYPE_ITEM, item,
2650 "no outer UDP layer found");
2652 mask = &rte_flow_item_vxlan_mask;
2654 if (priv->sh->steering_format_version !=
2655 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2656 !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2657 /* FDB domain & NIC domain non-zero group */
2658 if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2659 valid_mask = &nic_mask;
2660 /* Group zero in NIC domain */
2661 if (!attr->group && !attr->transfer &&
2662 priv->sh->tunnel_header_0_1)
2663 valid_mask = &nic_mask;
2665 ret = mlx5_flow_item_acceptable
2666 (item, (const uint8_t *)mask,
2667 (const uint8_t *)valid_mask,
2668 sizeof(struct rte_flow_item_vxlan),
2669 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2673 memcpy(&id.vni[1], spec->vni, 3);
2674 memcpy(&id.vni[1], mask->vni, 3);
2676 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2677 return rte_flow_error_set(error, ENOTSUP,
2678 RTE_FLOW_ERROR_TYPE_ITEM, item,
2679 "VXLAN tunnel must be fully defined");
2684 * Validate VXLAN_GPE item.
2687 * Item specification.
2688 * @param[in] item_flags
2689 * Bit-fields that holds the items detected until now.
2691 * Pointer to the private data structure.
2692 * @param[in] target_protocol
2693 * The next protocol in the previous item.
2695 * Pointer to error structure.
2698 * 0 on success, a negative errno value otherwise and rte_errno is set.
2701 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2702 uint64_t item_flags,
2703 struct rte_eth_dev *dev,
2704 struct rte_flow_error *error)
2706 struct mlx5_priv *priv = dev->data->dev_private;
2707 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2708 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2713 } id = { .vlan_id = 0, };
2715 if (!priv->sh->config.l3_vxlan_en)
2716 return rte_flow_error_set(error, ENOTSUP,
2717 RTE_FLOW_ERROR_TYPE_ITEM, item,
2718 "L3 VXLAN is not enabled by device"
2719 " parameter and/or not configured in"
2721 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2722 return rte_flow_error_set(error, ENOTSUP,
2723 RTE_FLOW_ERROR_TYPE_ITEM, item,
2724 "multiple tunnel layers not"
2727 * Verify only UDPv4 is present as defined in
2728 * https://tools.ietf.org/html/rfc7348
2730 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2731 return rte_flow_error_set(error, EINVAL,
2732 RTE_FLOW_ERROR_TYPE_ITEM, item,
2733 "no outer UDP layer found");
2735 mask = &rte_flow_item_vxlan_gpe_mask;
2736 ret = mlx5_flow_item_acceptable
2737 (item, (const uint8_t *)mask,
2738 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2739 sizeof(struct rte_flow_item_vxlan_gpe),
2740 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2745 return rte_flow_error_set(error, ENOTSUP,
2746 RTE_FLOW_ERROR_TYPE_ITEM,
2748 "VxLAN-GPE protocol"
2750 memcpy(&id.vni[1], spec->vni, 3);
2751 memcpy(&id.vni[1], mask->vni, 3);
2753 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2754 return rte_flow_error_set(error, ENOTSUP,
2755 RTE_FLOW_ERROR_TYPE_ITEM, item,
2756 "VXLAN-GPE tunnel must be fully"
2761 * Validate GRE Key item.
2764 * Item specification.
2765 * @param[in] item_flags
2766 * Bit flags to mark detected items.
2767 * @param[in] gre_item
2768 * Pointer to gre_item
2770 * Pointer to error structure.
2773 * 0 on success, a negative errno value otherwise and rte_errno is set.
2776 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2777 uint64_t item_flags,
2778 const struct rte_flow_item *gre_item,
2779 struct rte_flow_error *error)
2781 const rte_be32_t *mask = item->mask;
2783 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2784 const struct rte_flow_item_gre *gre_spec;
2785 const struct rte_flow_item_gre *gre_mask;
2787 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2788 return rte_flow_error_set(error, ENOTSUP,
2789 RTE_FLOW_ERROR_TYPE_ITEM, item,
2790 "Multiple GRE key not support");
2791 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2792 return rte_flow_error_set(error, ENOTSUP,
2793 RTE_FLOW_ERROR_TYPE_ITEM, item,
2794 "No preceding GRE header");
2795 if (item_flags & MLX5_FLOW_LAYER_INNER)
2796 return rte_flow_error_set(error, ENOTSUP,
2797 RTE_FLOW_ERROR_TYPE_ITEM, item,
2798 "GRE key following a wrong item");
2799 gre_mask = gre_item->mask;
2801 gre_mask = &rte_flow_item_gre_mask;
2802 gre_spec = gre_item->spec;
2803 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2804 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2805 return rte_flow_error_set(error, EINVAL,
2806 RTE_FLOW_ERROR_TYPE_ITEM, item,
2807 "Key bit must be on");
2810 mask = &gre_key_default_mask;
2811 ret = mlx5_flow_item_acceptable
2812 (item, (const uint8_t *)mask,
2813 (const uint8_t *)&gre_key_default_mask,
2814 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2819 * Validate GRE item.
2822 * Item specification.
2823 * @param[in] item_flags
2824 * Bit flags to mark detected items.
2825 * @param[in] target_protocol
2826 * The next protocol in the previous item.
2828 * Pointer to error structure.
2831 * 0 on success, a negative errno value otherwise and rte_errno is set.
2834 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2835 uint64_t item_flags,
2836 uint8_t target_protocol,
2837 struct rte_flow_error *error)
2839 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2840 const struct rte_flow_item_gre *mask = item->mask;
2842 const struct rte_flow_item_gre nic_mask = {
2843 .c_rsvd0_ver = RTE_BE16(0xB000),
2844 .protocol = RTE_BE16(UINT16_MAX),
2847 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2848 return rte_flow_error_set(error, EINVAL,
2849 RTE_FLOW_ERROR_TYPE_ITEM, item,
2850 "protocol filtering not compatible"
2851 " with this GRE layer");
2852 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2853 return rte_flow_error_set(error, ENOTSUP,
2854 RTE_FLOW_ERROR_TYPE_ITEM, item,
2855 "multiple tunnel layers not"
2857 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2858 return rte_flow_error_set(error, ENOTSUP,
2859 RTE_FLOW_ERROR_TYPE_ITEM, item,
2860 "L3 Layer is missing");
2862 mask = &rte_flow_item_gre_mask;
2863 ret = mlx5_flow_item_acceptable
2864 (item, (const uint8_t *)mask,
2865 (const uint8_t *)&nic_mask,
2866 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2870 #ifndef HAVE_MLX5DV_DR
2871 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2872 if (spec && (spec->protocol & mask->protocol))
2873 return rte_flow_error_set(error, ENOTSUP,
2874 RTE_FLOW_ERROR_TYPE_ITEM, item,
2875 "without MPLS support the"
2876 " specification cannot be used for"
2884 * Validate Geneve item.
2887 * Item specification.
2888 * @param[in] itemFlags
2889 * Bit-fields that holds the items detected until now.
2891 * Pointer to the private data structure.
2893 * Pointer to error structure.
2896 * 0 on success, a negative errno value otherwise and rte_errno is set.
2900 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2901 uint64_t item_flags,
2902 struct rte_eth_dev *dev,
2903 struct rte_flow_error *error)
2905 struct mlx5_priv *priv = dev->data->dev_private;
2906 const struct rte_flow_item_geneve *spec = item->spec;
2907 const struct rte_flow_item_geneve *mask = item->mask;
2910 uint8_t opt_len = priv->sh->cdev->config.hca_attr.geneve_max_opt_len ?
2911 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2912 const struct rte_flow_item_geneve nic_mask = {
2913 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2914 .vni = "\xff\xff\xff",
2915 .protocol = RTE_BE16(UINT16_MAX),
2918 if (!priv->sh->cdev->config.hca_attr.tunnel_stateless_geneve_rx)
2919 return rte_flow_error_set(error, ENOTSUP,
2920 RTE_FLOW_ERROR_TYPE_ITEM, item,
2921 "L3 Geneve is not enabled by device"
2922 " parameter and/or not configured in"
2924 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2925 return rte_flow_error_set(error, ENOTSUP,
2926 RTE_FLOW_ERROR_TYPE_ITEM, item,
2927 "multiple tunnel layers not"
2930 * Verify only UDPv4 is present as defined in
2931 * https://tools.ietf.org/html/rfc7348
2933 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2934 return rte_flow_error_set(error, EINVAL,
2935 RTE_FLOW_ERROR_TYPE_ITEM, item,
2936 "no outer UDP layer found");
2938 mask = &rte_flow_item_geneve_mask;
2939 ret = mlx5_flow_item_acceptable
2940 (item, (const uint8_t *)mask,
2941 (const uint8_t *)&nic_mask,
2942 sizeof(struct rte_flow_item_geneve),
2943 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2947 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2948 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2949 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2950 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2951 return rte_flow_error_set(error, ENOTSUP,
2952 RTE_FLOW_ERROR_TYPE_ITEM,
2954 "Geneve protocol unsupported"
2955 " fields are being used");
2956 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2957 return rte_flow_error_set
2959 RTE_FLOW_ERROR_TYPE_ITEM,
2961 "Unsupported Geneve options length");
2963 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2964 return rte_flow_error_set
2966 RTE_FLOW_ERROR_TYPE_ITEM, item,
2967 "Geneve tunnel must be fully defined");
2972 * Validate Geneve TLV option item.
2975 * Item specification.
2976 * @param[in] last_item
2977 * Previous validated item in the pattern items.
2978 * @param[in] geneve_item
2979 * Previous GENEVE item specification.
2981 * Pointer to the rte_eth_dev structure.
2983 * Pointer to error structure.
2986 * 0 on success, a negative errno value otherwise and rte_errno is set.
2989 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2991 const struct rte_flow_item *geneve_item,
2992 struct rte_eth_dev *dev,
2993 struct rte_flow_error *error)
2995 struct mlx5_priv *priv = dev->data->dev_private;
2996 struct mlx5_dev_ctx_shared *sh = priv->sh;
2997 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2998 struct mlx5_hca_attr *hca_attr = &sh->cdev->config.hca_attr;
2999 uint8_t data_max_supported =
3000 hca_attr->max_geneve_tlv_option_data_len * 4;
3001 const struct rte_flow_item_geneve *geneve_spec;
3002 const struct rte_flow_item_geneve *geneve_mask;
3003 const struct rte_flow_item_geneve_opt *spec = item->spec;
3004 const struct rte_flow_item_geneve_opt *mask = item->mask;
3006 unsigned int data_len;
3007 uint8_t tlv_option_len;
3008 uint16_t optlen_m, optlen_v;
3009 const struct rte_flow_item_geneve_opt full_mask = {
3010 .option_class = RTE_BE16(0xffff),
3011 .option_type = 0xff,
3016 mask = &rte_flow_item_geneve_opt_mask;
3018 return rte_flow_error_set
3019 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3020 "Geneve TLV opt class/type/length must be specified");
3021 if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
3022 return rte_flow_error_set
3023 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3024 "Geneve TLV opt length exceeds the limit (31)");
3025 /* Check if class type and length masks are full. */
3026 if (full_mask.option_class != mask->option_class ||
3027 full_mask.option_type != mask->option_type ||
3028 full_mask.option_len != (mask->option_len & full_mask.option_len))
3029 return rte_flow_error_set
3030 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3031 "Geneve TLV opt class/type/length masks must be full");
3032 /* Check if length is supported */
3033 if ((uint32_t)spec->option_len >
3034 hca_attr->max_geneve_tlv_option_data_len)
3035 return rte_flow_error_set
3036 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3037 "Geneve TLV opt length not supported");
3038 if (hca_attr->max_geneve_tlv_options > 1)
3040 "max_geneve_tlv_options supports more than 1 option");
3041 /* Check GENEVE item preceding. */
3042 if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
3043 return rte_flow_error_set
3044 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3045 "Geneve opt item must be preceded with Geneve item");
3046 geneve_spec = geneve_item->spec;
3047 geneve_mask = geneve_item->mask ? geneve_item->mask :
3048 &rte_flow_item_geneve_mask;
3049 /* Check if GENEVE TLV option size doesn't exceed option length */
3050 if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
3051 geneve_spec->ver_opt_len_o_c_rsvd0)) {
3052 tlv_option_len = spec->option_len & mask->option_len;
3053 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
3054 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
3055 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
3056 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
3057 if ((optlen_v & optlen_m) <= tlv_option_len)
3058 return rte_flow_error_set
3059 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3060 "GENEVE TLV option length exceeds optlen");
3062 /* Check if length is 0 or data is 0. */
3063 if (spec->data == NULL || spec->option_len == 0)
3064 return rte_flow_error_set
3065 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3066 "Geneve TLV opt with zero data/length not supported");
3067 /* Check not all data & mask are 0. */
3068 data_len = spec->option_len * 4;
3069 if (mask->data == NULL) {
3070 for (i = 0; i < data_len; i++)
3074 return rte_flow_error_set(error, ENOTSUP,
3075 RTE_FLOW_ERROR_TYPE_ITEM, item,
3076 "Can't match on Geneve option data 0");
3078 for (i = 0; i < data_len; i++)
3079 if (spec->data[i] & mask->data[i])
3082 return rte_flow_error_set(error, ENOTSUP,
3083 RTE_FLOW_ERROR_TYPE_ITEM, item,
3084 "Can't match on Geneve option data and mask 0");
3085 /* Check data mask supported. */
3086 for (i = data_max_supported; i < data_len ; i++)
3088 return rte_flow_error_set(error, ENOTSUP,
3089 RTE_FLOW_ERROR_TYPE_ITEM, item,
3090 "Data mask is of unsupported size");
3092 /* Check GENEVE option is supported in NIC. */
3093 if (!hca_attr->geneve_tlv_opt)
3094 return rte_flow_error_set
3095 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3096 "Geneve TLV opt not supported");
3097 /* Check if we already have geneve option with different type/class. */
3098 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3099 geneve_opt_resource = sh->geneve_tlv_option_resource;
3100 if (geneve_opt_resource != NULL)
3101 if (geneve_opt_resource->option_class != spec->option_class ||
3102 geneve_opt_resource->option_type != spec->option_type ||
3103 geneve_opt_resource->length != spec->option_len) {
3104 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3105 return rte_flow_error_set(error, ENOTSUP,
3106 RTE_FLOW_ERROR_TYPE_ITEM, item,
3107 "Only one Geneve TLV option supported");
3109 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3114 * Validate MPLS item.
3117 * Pointer to the rte_eth_dev structure.
3119 * Item specification.
3120 * @param[in] item_flags
3121 * Bit-fields that holds the items detected until now.
3122 * @param[in] prev_layer
3123 * The protocol layer indicated in previous item.
3125 * Pointer to error structure.
3128 * 0 on success, a negative errno value otherwise and rte_errno is set.
3131 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3132 const struct rte_flow_item *item __rte_unused,
3133 uint64_t item_flags __rte_unused,
3134 uint64_t prev_layer __rte_unused,
3135 struct rte_flow_error *error)
3137 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3138 const struct rte_flow_item_mpls *mask = item->mask;
3139 struct mlx5_priv *priv = dev->data->dev_private;
3142 if (!priv->sh->dev_cap.mpls_en)
3143 return rte_flow_error_set(error, ENOTSUP,
3144 RTE_FLOW_ERROR_TYPE_ITEM, item,
3145 "MPLS not supported or"
3146 " disabled in firmware"
3148 /* MPLS over UDP, GRE is allowed */
3149 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3150 MLX5_FLOW_LAYER_GRE |
3151 MLX5_FLOW_LAYER_GRE_KEY)))
3152 return rte_flow_error_set(error, EINVAL,
3153 RTE_FLOW_ERROR_TYPE_ITEM, item,
3154 "protocol filtering not compatible"
3155 " with MPLS layer");
3156 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3157 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3158 !(item_flags & MLX5_FLOW_LAYER_GRE))
3159 return rte_flow_error_set(error, ENOTSUP,
3160 RTE_FLOW_ERROR_TYPE_ITEM, item,
3161 "multiple tunnel layers not"
3164 mask = &rte_flow_item_mpls_mask;
3165 ret = mlx5_flow_item_acceptable
3166 (item, (const uint8_t *)mask,
3167 (const uint8_t *)&rte_flow_item_mpls_mask,
3168 sizeof(struct rte_flow_item_mpls),
3169 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3174 return rte_flow_error_set(error, ENOTSUP,
3175 RTE_FLOW_ERROR_TYPE_ITEM, item,
3176 "MPLS is not supported by Verbs, please"
3182 * Validate NVGRE item.
3185 * Item specification.
3186 * @param[in] item_flags
3187 * Bit flags to mark detected items.
3188 * @param[in] target_protocol
3189 * The next protocol in the previous item.
3191 * Pointer to error structure.
3194 * 0 on success, a negative errno value otherwise and rte_errno is set.
3197 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3198 uint64_t item_flags,
3199 uint8_t target_protocol,
3200 struct rte_flow_error *error)
3202 const struct rte_flow_item_nvgre *mask = item->mask;
3205 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3206 return rte_flow_error_set(error, EINVAL,
3207 RTE_FLOW_ERROR_TYPE_ITEM, item,
3208 "protocol filtering not compatible"
3209 " with this GRE layer");
3210 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3211 return rte_flow_error_set(error, ENOTSUP,
3212 RTE_FLOW_ERROR_TYPE_ITEM, item,
3213 "multiple tunnel layers not"
3215 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3216 return rte_flow_error_set(error, ENOTSUP,
3217 RTE_FLOW_ERROR_TYPE_ITEM, item,
3218 "L3 Layer is missing");
3220 mask = &rte_flow_item_nvgre_mask;
3221 ret = mlx5_flow_item_acceptable
3222 (item, (const uint8_t *)mask,
3223 (const uint8_t *)&rte_flow_item_nvgre_mask,
3224 sizeof(struct rte_flow_item_nvgre),
3225 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3232 * Validate eCPRI item.
3235 * Item specification.
3236 * @param[in] item_flags
3237 * Bit-fields that holds the items detected until now.
3238 * @param[in] last_item
3239 * Previous validated item in the pattern items.
3240 * @param[in] ether_type
3241 * Type in the ethernet layer header (including dot1q).
3242 * @param[in] acc_mask
3243 * Acceptable mask, if NULL default internal default mask
3244 * will be used to check whether item fields are supported.
3246 * Pointer to error structure.
3249 * 0 on success, a negative errno value otherwise and rte_errno is set.
3252 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3253 uint64_t item_flags,
3255 uint16_t ether_type,
3256 const struct rte_flow_item_ecpri *acc_mask,
3257 struct rte_flow_error *error)
3259 const struct rte_flow_item_ecpri *mask = item->mask;
3260 const struct rte_flow_item_ecpri nic_mask = {
3264 RTE_BE32(((const struct rte_ecpri_common_hdr) {
3268 .dummy[0] = 0xFFFFFFFF,
3271 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3272 MLX5_FLOW_LAYER_OUTER_VLAN);
3273 struct rte_flow_item_ecpri mask_lo;
3275 if (!(last_item & outer_l2_vlan) &&
3276 last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3277 return rte_flow_error_set(error, EINVAL,
3278 RTE_FLOW_ERROR_TYPE_ITEM, item,
3279 "eCPRI can only follow L2/VLAN layer or UDP layer");
3280 if ((last_item & outer_l2_vlan) && ether_type &&
3281 ether_type != RTE_ETHER_TYPE_ECPRI)
3282 return rte_flow_error_set(error, EINVAL,
3283 RTE_FLOW_ERROR_TYPE_ITEM, item,
3284 "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3285 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3286 return rte_flow_error_set(error, EINVAL,
3287 RTE_FLOW_ERROR_TYPE_ITEM, item,
3288 "eCPRI with tunnel is not supported right now");
3289 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3290 return rte_flow_error_set(error, ENOTSUP,
3291 RTE_FLOW_ERROR_TYPE_ITEM, item,
3292 "multiple L3 layers not supported");
3293 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3294 return rte_flow_error_set(error, EINVAL,
3295 RTE_FLOW_ERROR_TYPE_ITEM, item,
3296 "eCPRI cannot coexist with a TCP layer");
3297 /* In specification, eCPRI could be over UDP layer. */
3298 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3299 return rte_flow_error_set(error, EINVAL,
3300 RTE_FLOW_ERROR_TYPE_ITEM, item,
3301 "eCPRI over UDP layer is not yet supported right now");
3302 /* Mask for type field in common header could be zero. */
3304 mask = &rte_flow_item_ecpri_mask;
3305 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3306 /* Input mask is in big-endian format. */
3307 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3308 return rte_flow_error_set(error, EINVAL,
3309 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3310 "partial mask is not supported for protocol");
3311 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3312 return rte_flow_error_set(error, EINVAL,
3313 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3314 "message header mask must be after a type mask");
3315 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3316 acc_mask ? (const uint8_t *)acc_mask
3317 : (const uint8_t *)&nic_mask,
3318 sizeof(struct rte_flow_item_ecpri),
3319 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3323 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3324 const struct rte_flow_attr *attr __rte_unused,
3325 const struct rte_flow_item items[] __rte_unused,
3326 const struct rte_flow_action actions[] __rte_unused,
3327 bool external __rte_unused,
3328 int hairpin __rte_unused,
3329 struct rte_flow_error *error)
3331 return rte_flow_error_set(error, ENOTSUP,
3332 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3335 static struct mlx5_flow *
3336 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3337 const struct rte_flow_attr *attr __rte_unused,
3338 const struct rte_flow_item items[] __rte_unused,
3339 const struct rte_flow_action actions[] __rte_unused,
3340 struct rte_flow_error *error)
3342 rte_flow_error_set(error, ENOTSUP,
3343 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3348 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3349 struct mlx5_flow *dev_flow __rte_unused,
3350 const struct rte_flow_attr *attr __rte_unused,
3351 const struct rte_flow_item items[] __rte_unused,
3352 const struct rte_flow_action actions[] __rte_unused,
3353 struct rte_flow_error *error)
3355 return rte_flow_error_set(error, ENOTSUP,
3356 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3360 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3361 struct rte_flow *flow __rte_unused,
3362 struct rte_flow_error *error)
3364 return rte_flow_error_set(error, ENOTSUP,
3365 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3369 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3370 struct rte_flow *flow __rte_unused)
3375 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3376 struct rte_flow *flow __rte_unused)
3381 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3382 struct rte_flow *flow __rte_unused,
3383 const struct rte_flow_action *actions __rte_unused,
3384 void *data __rte_unused,
3385 struct rte_flow_error *error)
3387 return rte_flow_error_set(error, ENOTSUP,
3388 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3392 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3393 uint32_t domains __rte_unused,
3394 uint32_t flags __rte_unused)
3399 /* Void driver to protect from null pointer reference. */
3400 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3401 .validate = flow_null_validate,
3402 .prepare = flow_null_prepare,
3403 .translate = flow_null_translate,
3404 .apply = flow_null_apply,
3405 .remove = flow_null_remove,
3406 .destroy = flow_null_destroy,
3407 .query = flow_null_query,
3408 .sync_domain = flow_null_sync_domain,
3412 * Select flow driver type according to flow attributes and device
3416 * Pointer to the dev structure.
3418 * Pointer to the flow attributes.
3421 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3423 static enum mlx5_flow_drv_type
3424 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3426 struct mlx5_priv *priv = dev->data->dev_private;
3427 /* The OS can determine first a specific flow type (DV, VERBS) */
3428 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3430 if (type != MLX5_FLOW_TYPE_MAX)
3432 /* If no OS specific type - continue with DV/VERBS selection */
3433 if (attr->transfer && priv->sh->config.dv_esw_en)
3434 type = MLX5_FLOW_TYPE_DV;
3435 if (!attr->transfer)
3436 type = priv->sh->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3437 MLX5_FLOW_TYPE_VERBS;
3441 #define flow_get_drv_ops(type) flow_drv_ops[type]
3444 * Flow driver validation API. This abstracts calling driver specific functions.
3445 * The type of flow driver is determined according to flow attributes.
3448 * Pointer to the dev structure.
3450 * Pointer to the flow attributes.
3452 * Pointer to the list of items.
3453 * @param[in] actions
3454 * Pointer to the list of actions.
3455 * @param[in] external
3456 * This flow rule is created by request external to PMD.
3457 * @param[in] hairpin
3458 * Number of hairpin TX actions, 0 means classic flow.
3460 * Pointer to the error structure.
3463 * 0 on success, a negative errno value otherwise and rte_errno is set.
3466 flow_drv_validate(struct rte_eth_dev *dev,
3467 const struct rte_flow_attr *attr,
3468 const struct rte_flow_item items[],
3469 const struct rte_flow_action actions[],
3470 bool external, int hairpin, struct rte_flow_error *error)
3472 const struct mlx5_flow_driver_ops *fops;
3473 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3475 fops = flow_get_drv_ops(type);
3476 return fops->validate(dev, attr, items, actions, external,
3481 * Flow driver preparation API. This abstracts calling driver specific
3482 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3483 * calculates the size of memory required for device flow, allocates the memory,
3484 * initializes the device flow and returns the pointer.
3487 * This function initializes device flow structure such as dv or verbs in
3488 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3489 * rest. For example, adding returning device flow to flow->dev_flow list and
3490 * setting backward reference to the flow should be done out of this function.
3491 * layers field is not filled either.
3494 * Pointer to the dev structure.
3496 * Pointer to the flow attributes.
3498 * Pointer to the list of items.
3499 * @param[in] actions
3500 * Pointer to the list of actions.
3501 * @param[in] flow_idx
3502 * This memory pool index to the flow.
3504 * Pointer to the error structure.
3507 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3509 static inline struct mlx5_flow *
3510 flow_drv_prepare(struct rte_eth_dev *dev,
3511 const struct rte_flow *flow,
3512 const struct rte_flow_attr *attr,
3513 const struct rte_flow_item items[],
3514 const struct rte_flow_action actions[],
3516 struct rte_flow_error *error)
3518 const struct mlx5_flow_driver_ops *fops;
3519 enum mlx5_flow_drv_type type = flow->drv_type;
3520 struct mlx5_flow *mlx5_flow = NULL;
3522 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3523 fops = flow_get_drv_ops(type);
3524 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3526 mlx5_flow->flow_idx = flow_idx;
3531 * Flow driver translation API. This abstracts calling driver specific
3532 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3533 * translates a generic flow into a driver flow. flow_drv_prepare() must
3537 * dev_flow->layers could be filled as a result of parsing during translation
3538 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3539 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3540 * flow->actions could be overwritten even though all the expanded dev_flows
3541 * have the same actions.
3544 * Pointer to the rte dev structure.
3545 * @param[in, out] dev_flow
3546 * Pointer to the mlx5 flow.
3548 * Pointer to the flow attributes.
3550 * Pointer to the list of items.
3551 * @param[in] actions
3552 * Pointer to the list of actions.
3554 * Pointer to the error structure.
3557 * 0 on success, a negative errno value otherwise and rte_errno is set.
3560 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3561 const struct rte_flow_attr *attr,
3562 const struct rte_flow_item items[],
3563 const struct rte_flow_action actions[],
3564 struct rte_flow_error *error)
3566 const struct mlx5_flow_driver_ops *fops;
3567 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3569 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3570 fops = flow_get_drv_ops(type);
3571 return fops->translate(dev, dev_flow, attr, items, actions, error);
3575 * Flow driver apply API. This abstracts calling driver specific functions.
3576 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3577 * translated driver flows on to device. flow_drv_translate() must precede.
3580 * Pointer to Ethernet device structure.
3581 * @param[in, out] flow
3582 * Pointer to flow structure.
3584 * Pointer to error structure.
3587 * 0 on success, a negative errno value otherwise and rte_errno is set.
3590 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3591 struct rte_flow_error *error)
3593 const struct mlx5_flow_driver_ops *fops;
3594 enum mlx5_flow_drv_type type = flow->drv_type;
3596 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3597 fops = flow_get_drv_ops(type);
3598 return fops->apply(dev, flow, error);
3602 * Flow driver destroy API. This abstracts calling driver specific functions.
3603 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3604 * on device and releases resources of the flow.
3607 * Pointer to Ethernet device.
3608 * @param[in, out] flow
3609 * Pointer to flow structure.
3612 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3614 const struct mlx5_flow_driver_ops *fops;
3615 enum mlx5_flow_drv_type type = flow->drv_type;
3617 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3618 fops = flow_get_drv_ops(type);
3619 fops->destroy(dev, flow);
3623 * Flow driver find RSS policy tbl API. This abstracts calling driver
3624 * specific functions. Parent flow (rte_flow) should have driver
3625 * type (drv_type). It will find the RSS policy table that has the rss_desc.
3628 * Pointer to Ethernet device.
3629 * @param[in, out] flow
3630 * Pointer to flow structure.
3632 * Pointer to meter policy table.
3633 * @param[in] rss_desc
3634 * Pointer to rss_desc
3636 static struct mlx5_flow_meter_sub_policy *
3637 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3638 struct rte_flow *flow,
3639 struct mlx5_flow_meter_policy *policy,
3640 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3642 const struct mlx5_flow_driver_ops *fops;
3643 enum mlx5_flow_drv_type type = flow->drv_type;
3645 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3646 fops = flow_get_drv_ops(type);
3647 return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3651 * Flow driver color tag rule API. This abstracts calling driver
3652 * specific functions. Parent flow (rte_flow) should have driver
3653 * type (drv_type). It will create the color tag rules in hierarchy meter.
3656 * Pointer to Ethernet device.
3657 * @param[in, out] flow
3658 * Pointer to flow structure.
3660 * Pointer to flow meter structure.
3661 * @param[in] src_port
3662 * The src port this extra rule should use.
3664 * The src port id match item.
3666 * Pointer to error structure.
3669 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3670 struct rte_flow *flow,
3671 struct mlx5_flow_meter_info *fm,
3673 const struct rte_flow_item *item,
3674 struct rte_flow_error *error)
3676 const struct mlx5_flow_driver_ops *fops;
3677 enum mlx5_flow_drv_type type = flow->drv_type;
3679 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3680 fops = flow_get_drv_ops(type);
3681 return fops->meter_hierarchy_rule_create(dev, fm,
3682 src_port, item, error);
3686 * Get RSS action from the action list.
3689 * Pointer to Ethernet device.
3690 * @param[in] actions
3691 * Pointer to the list of actions.
3693 * Parent flow structure pointer.
3696 * Pointer to the RSS action if exist, else return NULL.
3698 static const struct rte_flow_action_rss*
3699 flow_get_rss_action(struct rte_eth_dev *dev,
3700 const struct rte_flow_action actions[])
3702 struct mlx5_priv *priv = dev->data->dev_private;
3703 const struct rte_flow_action_rss *rss = NULL;
3704 struct mlx5_meter_policy_action_container *acg;
3705 struct mlx5_meter_policy_action_container *acy;
3707 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3708 switch (actions->type) {
3709 case RTE_FLOW_ACTION_TYPE_RSS:
3710 rss = actions->conf;
3712 case RTE_FLOW_ACTION_TYPE_SAMPLE:
3714 const struct rte_flow_action_sample *sample =
3716 const struct rte_flow_action *act = sample->actions;
3717 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3718 if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3722 case RTE_FLOW_ACTION_TYPE_METER:
3725 struct mlx5_flow_meter_info *fm;
3726 struct mlx5_flow_meter_policy *policy;
3727 const struct rte_flow_action_meter *mtr = actions->conf;
3729 fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3730 if (fm && !fm->def_policy) {
3731 policy = mlx5_flow_meter_policy_find(dev,
3732 fm->policy_id, NULL);
3733 MLX5_ASSERT(policy);
3734 if (policy->is_hierarchy) {
3736 mlx5_flow_meter_hierarchy_get_final_policy(dev,
3741 if (policy->is_rss) {
3743 &policy->act_cnt[RTE_COLOR_GREEN];
3745 &policy->act_cnt[RTE_COLOR_YELLOW];
3746 if (acg->fate_action ==
3747 MLX5_FLOW_FATE_SHARED_RSS)
3748 rss = acg->rss->conf;
3749 else if (acy->fate_action ==
3750 MLX5_FLOW_FATE_SHARED_RSS)
3751 rss = acy->rss->conf;
3764 * Get ASO age action by index.
3767 * Pointer to the Ethernet device structure.
3768 * @param[in] age_idx
3769 * Index to the ASO age action.
3772 * The specified ASO age action.
3774 struct mlx5_aso_age_action*
3775 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3777 uint16_t pool_idx = age_idx & UINT16_MAX;
3778 uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3779 struct mlx5_priv *priv = dev->data->dev_private;
3780 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3781 struct mlx5_aso_age_pool *pool;
3783 rte_rwlock_read_lock(&mng->resize_rwl);
3784 pool = mng->pools[pool_idx];
3785 rte_rwlock_read_unlock(&mng->resize_rwl);
3786 return &pool->actions[offset - 1];
3789 /* maps indirect action to translated direct in some actions array */
3790 struct mlx5_translated_action_handle {
3791 struct rte_flow_action_handle *action; /**< Indirect action handle. */
3792 int index; /**< Index in related array of rte_flow_action. */
3796 * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3797 * direct action if translation possible.
3798 * This functionality used to run same execution path for both direct and
3799 * indirect actions on flow create. All necessary preparations for indirect
3800 * action handling should be performed on *handle* actions list returned
3804 * Pointer to Ethernet device.
3805 * @param[in] actions
3806 * List of actions to translate.
3807 * @param[out] handle
3808 * List to store translated indirect action object handles.
3809 * @param[in, out] indir_n
3810 * Size of *handle* array. On return should be updated with number of
3811 * indirect actions retrieved from the *actions* list.
3812 * @param[out] translated_actions
3813 * List of actions where all indirect actions were translated to direct
3814 * if possible. NULL if no translation took place.
3816 * Pointer to the error structure.
3819 * 0 on success, a negative errno value otherwise and rte_errno is set.
3822 flow_action_handles_translate(struct rte_eth_dev *dev,
3823 const struct rte_flow_action actions[],
3824 struct mlx5_translated_action_handle *handle,
3826 struct rte_flow_action **translated_actions,
3827 struct rte_flow_error *error)
3829 struct mlx5_priv *priv = dev->data->dev_private;
3830 struct rte_flow_action *translated = NULL;
3831 size_t actions_size;
3834 struct mlx5_translated_action_handle *handle_end = NULL;
3836 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3837 if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3839 if (copied_n == *indir_n) {
3840 return rte_flow_error_set
3841 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3842 NULL, "too many shared actions");
3844 rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3845 sizeof(actions[n].conf));
3846 handle[copied_n].index = n;
3850 *indir_n = copied_n;
3853 actions_size = sizeof(struct rte_flow_action) * n;
3854 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3859 memcpy(translated, actions, actions_size);
3860 for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3861 struct mlx5_shared_action_rss *shared_rss;
3862 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3863 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3864 uint32_t idx = act_idx &
3865 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3868 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3869 shared_rss = mlx5_ipool_get
3870 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3871 translated[handle->index].type =
3872 RTE_FLOW_ACTION_TYPE_RSS;
3873 translated[handle->index].conf =
3874 &shared_rss->origin;
3876 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3877 translated[handle->index].type =
3878 (enum rte_flow_action_type)
3879 MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3880 translated[handle->index].conf = (void *)(uintptr_t)idx;
3882 case MLX5_INDIRECT_ACTION_TYPE_AGE:
3883 if (priv->sh->flow_hit_aso_en) {
3884 translated[handle->index].type =
3885 (enum rte_flow_action_type)
3886 MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3887 translated[handle->index].conf =
3888 (void *)(uintptr_t)idx;
3892 case MLX5_INDIRECT_ACTION_TYPE_CT:
3893 if (priv->sh->ct_aso_en) {
3894 translated[handle->index].type =
3895 RTE_FLOW_ACTION_TYPE_CONNTRACK;
3896 translated[handle->index].conf =
3897 (void *)(uintptr_t)idx;
3902 mlx5_free(translated);
3903 return rte_flow_error_set
3904 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3905 NULL, "invalid indirect action type");
3908 *translated_actions = translated;
3913 * Get Shared RSS action from the action list.
3916 * Pointer to Ethernet device.
3918 * Pointer to the list of actions.
3919 * @param[in] shared_n
3920 * Actions list length.
3923 * The MLX5 RSS action ID if exists, otherwise return 0.
3926 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3927 struct mlx5_translated_action_handle *handle,
3930 struct mlx5_translated_action_handle *handle_end;
3931 struct mlx5_priv *priv = dev->data->dev_private;
3932 struct mlx5_shared_action_rss *shared_rss;
3935 for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3936 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3937 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3938 uint32_t idx = act_idx &
3939 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3941 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3942 shared_rss = mlx5_ipool_get
3943 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3945 __atomic_add_fetch(&shared_rss->refcnt, 1,
3956 find_graph_root(uint32_t rss_level)
3958 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3959 MLX5_EXPANSION_ROOT_OUTER;
3963 * Get layer flags from the prefix flow.
3965 * Some flows may be split to several subflows, the prefix subflow gets the
3966 * match items and the suffix sub flow gets the actions.
3967 * Some actions need the user defined match item flags to get the detail for
3969 * This function helps the suffix flow to get the item layer flags from prefix
3972 * @param[in] dev_flow
3973 * Pointer the created prefix subflow.
3976 * The layers get from prefix subflow.
3978 static inline uint64_t
3979 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3981 uint64_t layers = 0;
3984 * Layers bits could be localization, but usually the compiler will
3985 * help to do the optimization work for source code.
3986 * If no decap actions, use the layers directly.
3988 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3989 return dev_flow->handle->layers;
3990 /* Convert L3 layers with decap action. */
3991 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3992 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3993 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3994 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3995 /* Convert L4 layers with decap action. */
3996 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3997 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3998 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3999 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
4004 * Get metadata split action information.
4006 * @param[in] actions
4007 * Pointer to the list of actions.
4009 * Pointer to the return pointer.
4010 * @param[out] qrss_type
4011 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
4012 * if no QUEUE/RSS is found.
4013 * @param[out] encap_idx
4014 * Pointer to the index of the encap action if exists, otherwise the last
4018 * Total number of actions.
4021 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
4022 const struct rte_flow_action **qrss,
4025 const struct rte_flow_action_raw_encap *raw_encap;
4027 int raw_decap_idx = -1;
4030 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4031 switch (actions->type) {
4032 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4033 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4034 *encap_idx = actions_n;
4036 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4037 raw_decap_idx = actions_n;
4039 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4040 raw_encap = actions->conf;
4041 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4042 *encap_idx = raw_decap_idx != -1 ?
4043 raw_decap_idx : actions_n;
4045 case RTE_FLOW_ACTION_TYPE_QUEUE:
4046 case RTE_FLOW_ACTION_TYPE_RSS:
4054 if (*encap_idx == -1)
4055 *encap_idx = actions_n;
4056 /* Count RTE_FLOW_ACTION_TYPE_END. */
4057 return actions_n + 1;
4061 * Check if the action will change packet.
4064 * Pointer to Ethernet device.
4069 * true if action will change packet, false otherwise.
4071 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4072 enum rte_flow_action_type type)
4074 struct mlx5_priv *priv = dev->data->dev_private;
4077 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4078 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4079 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4080 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4081 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4082 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4083 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4084 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4085 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4086 case RTE_FLOW_ACTION_TYPE_SET_TTL:
4087 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4088 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4089 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4090 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4091 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4092 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4093 case RTE_FLOW_ACTION_TYPE_SET_META:
4094 case RTE_FLOW_ACTION_TYPE_SET_TAG:
4095 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4096 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4097 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4098 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4099 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4100 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4101 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4102 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4103 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4104 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4105 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4107 case RTE_FLOW_ACTION_TYPE_FLAG:
4108 case RTE_FLOW_ACTION_TYPE_MARK:
4109 if (priv->sh->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4119 * Check meter action from the action list.
4122 * Pointer to Ethernet device.
4123 * @param[in] actions
4124 * Pointer to the list of actions.
4125 * @param[out] has_mtr
4126 * Pointer to the meter exist flag.
4127 * @param[out] has_modify
4128 * Pointer to the flag showing there's packet change action.
4129 * @param[out] meter_id
4130 * Pointer to the meter id.
4133 * Total number of actions.
4136 flow_check_meter_action(struct rte_eth_dev *dev,
4137 const struct rte_flow_action actions[],
4138 bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4140 const struct rte_flow_action_meter *mtr = NULL;
4143 MLX5_ASSERT(has_mtr);
4145 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4146 switch (actions->type) {
4147 case RTE_FLOW_ACTION_TYPE_METER:
4148 mtr = actions->conf;
4149 *meter_id = mtr->mtr_id;
4156 *has_modify |= flow_check_modify_action_type(dev,
4160 /* Count RTE_FLOW_ACTION_TYPE_END. */
4161 return actions_n + 1;
4165 * Check if the flow should be split due to hairpin.
4166 * The reason for the split is that in current HW we can't
4167 * support encap and push-vlan on Rx, so if a flow contains
4168 * these actions we move it to Tx.
4171 * Pointer to Ethernet device.
4173 * Flow rule attributes.
4174 * @param[in] actions
4175 * Associated actions (list terminated by the END action).
4178 * > 0 the number of actions and the flow should be split,
4179 * 0 when no split required.
4182 flow_check_hairpin_split(struct rte_eth_dev *dev,
4183 const struct rte_flow_attr *attr,
4184 const struct rte_flow_action actions[])
4186 int queue_action = 0;
4189 const struct rte_flow_action_queue *queue;
4190 const struct rte_flow_action_rss *rss;
4191 const struct rte_flow_action_raw_encap *raw_encap;
4192 const struct rte_eth_hairpin_conf *conf;
4196 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4197 switch (actions->type) {
4198 case RTE_FLOW_ACTION_TYPE_QUEUE:
4199 queue = actions->conf;
4202 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4203 if (conf == NULL || conf->tx_explicit != 0)
4208 case RTE_FLOW_ACTION_TYPE_RSS:
4209 rss = actions->conf;
4210 if (rss == NULL || rss->queue_num == 0)
4212 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4213 if (conf == NULL || conf->tx_explicit != 0)
4218 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4219 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4220 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4221 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4222 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4226 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4227 raw_encap = actions->conf;
4228 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4237 if (split && queue_action)
4242 /* Declare flow create/destroy prototype in advance. */
4244 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4245 const struct rte_flow_attr *attr,
4246 const struct rte_flow_item items[],
4247 const struct rte_flow_action actions[],
4248 bool external, struct rte_flow_error *error);
4251 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4255 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4256 struct mlx5_list_entry *entry, void *cb_ctx)
4258 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4259 struct mlx5_flow_mreg_copy_resource *mcp_res =
4260 container_of(entry, typeof(*mcp_res), hlist_ent);
4262 return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4265 struct mlx5_list_entry *
4266 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4268 struct rte_eth_dev *dev = tool_ctx;
4269 struct mlx5_priv *priv = dev->data->dev_private;
4270 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4271 struct mlx5_flow_mreg_copy_resource *mcp_res;
4272 struct rte_flow_error *error = ctx->error;
4275 uint32_t mark_id = *(uint32_t *)(ctx->data);
4276 struct rte_flow_attr attr = {
4277 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4280 struct mlx5_rte_flow_item_tag tag_spec = {
4283 struct rte_flow_item items[] = {
4284 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4286 struct rte_flow_action_mark ftag = {
4289 struct mlx5_flow_action_copy_mreg cp_mreg = {
4293 struct rte_flow_action_jump jump = {
4294 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4296 struct rte_flow_action actions[] = {
4297 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4300 /* Fill the register fields in the flow. */
4301 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4305 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4309 /* Provide the full width of FLAG specific value. */
4310 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4311 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4312 /* Build a new flow. */
4313 if (mark_id != MLX5_DEFAULT_COPY_ID) {
4314 items[0] = (struct rte_flow_item){
4315 .type = (enum rte_flow_item_type)
4316 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4319 items[1] = (struct rte_flow_item){
4320 .type = RTE_FLOW_ITEM_TYPE_END,
4322 actions[0] = (struct rte_flow_action){
4323 .type = (enum rte_flow_action_type)
4324 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4327 actions[1] = (struct rte_flow_action){
4328 .type = (enum rte_flow_action_type)
4329 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4332 actions[2] = (struct rte_flow_action){
4333 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4336 actions[3] = (struct rte_flow_action){
4337 .type = RTE_FLOW_ACTION_TYPE_END,
4340 /* Default rule, wildcard match. */
4341 attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4342 items[0] = (struct rte_flow_item){
4343 .type = RTE_FLOW_ITEM_TYPE_END,
4345 actions[0] = (struct rte_flow_action){
4346 .type = (enum rte_flow_action_type)
4347 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4350 actions[1] = (struct rte_flow_action){
4351 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4354 actions[2] = (struct rte_flow_action){
4355 .type = RTE_FLOW_ACTION_TYPE_END,
4358 /* Build a new entry. */
4359 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4365 mcp_res->mark_id = mark_id;
4367 * The copy Flows are not included in any list. There
4368 * ones are referenced from other Flows and can not
4369 * be applied, removed, deleted in arbitrary order
4370 * by list traversing.
4372 mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4373 &attr, items, actions, false, error);
4374 if (!mcp_res->rix_flow) {
4375 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4378 return &mcp_res->hlist_ent;
4381 struct mlx5_list_entry *
4382 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4383 void *cb_ctx __rte_unused)
4385 struct rte_eth_dev *dev = tool_ctx;
4386 struct mlx5_priv *priv = dev->data->dev_private;
4387 struct mlx5_flow_mreg_copy_resource *mcp_res;
4390 mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4395 memcpy(mcp_res, oentry, sizeof(*mcp_res));
4397 return &mcp_res->hlist_ent;
4401 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4403 struct mlx5_flow_mreg_copy_resource *mcp_res =
4404 container_of(entry, typeof(*mcp_res), hlist_ent);
4405 struct rte_eth_dev *dev = tool_ctx;
4406 struct mlx5_priv *priv = dev->data->dev_private;
4408 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4412 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4414 * As mark_id is unique, if there's already a registered flow for the mark_id,
4415 * return by increasing the reference counter of the resource. Otherwise, create
4416 * the resource (mcp_res) and flow.
4419 * - If ingress port is ANY and reg_c[1] is mark_id,
4420 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4422 * For default flow (zero mark_id), flow is like,
4423 * - If ingress port is ANY,
4424 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
4427 * Pointer to Ethernet device.
4429 * ID of MARK action, zero means default flow for META.
4431 * Perform verbose error reporting if not NULL.
4434 * Associated resource on success, NULL otherwise and rte_errno is set.
4436 static struct mlx5_flow_mreg_copy_resource *
4437 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4438 struct rte_flow_error *error)
4440 struct mlx5_priv *priv = dev->data->dev_private;
4441 struct mlx5_list_entry *entry;
4442 struct mlx5_flow_cb_ctx ctx = {
4448 /* Check if already registered. */
4449 MLX5_ASSERT(priv->mreg_cp_tbl);
4450 entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4453 return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4458 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4460 struct mlx5_flow_mreg_copy_resource *mcp_res =
4461 container_of(entry, typeof(*mcp_res), hlist_ent);
4462 struct rte_eth_dev *dev = tool_ctx;
4463 struct mlx5_priv *priv = dev->data->dev_private;
4465 MLX5_ASSERT(mcp_res->rix_flow);
4466 flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4467 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4471 * Release flow in RX_CP_TBL.
4474 * Pointer to Ethernet device.
4476 * Parent flow for wich copying is provided.
4479 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4480 struct rte_flow *flow)
4482 struct mlx5_flow_mreg_copy_resource *mcp_res;
4483 struct mlx5_priv *priv = dev->data->dev_private;
4485 if (!flow->rix_mreg_copy)
4487 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4488 flow->rix_mreg_copy);
4489 if (!mcp_res || !priv->mreg_cp_tbl)
4491 MLX5_ASSERT(mcp_res->rix_flow);
4492 mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4493 flow->rix_mreg_copy = 0;
4497 * Remove the default copy action from RX_CP_TBL.
4499 * This functions is called in the mlx5_dev_start(). No thread safe
4503 * Pointer to Ethernet device.
4506 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4508 struct mlx5_list_entry *entry;
4509 struct mlx5_priv *priv = dev->data->dev_private;
4510 struct mlx5_flow_cb_ctx ctx;
4513 /* Check if default flow is registered. */
4514 if (!priv->mreg_cp_tbl)
4516 mark_id = MLX5_DEFAULT_COPY_ID;
4517 ctx.data = &mark_id;
4518 entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4521 mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4525 * Add the default copy action in in RX_CP_TBL.
4527 * This functions is called in the mlx5_dev_start(). No thread safe
4531 * Pointer to Ethernet device.
4533 * Perform verbose error reporting if not NULL.
4536 * 0 for success, negative value otherwise and rte_errno is set.
4539 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4540 struct rte_flow_error *error)
4542 struct mlx5_priv *priv = dev->data->dev_private;
4543 struct mlx5_flow_mreg_copy_resource *mcp_res;
4544 struct mlx5_flow_cb_ctx ctx;
4547 /* Check whether extensive metadata feature is engaged. */
4548 if (!priv->sh->config.dv_flow_en ||
4549 priv->sh->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4550 !mlx5_flow_ext_mreg_supported(dev) ||
4551 !priv->sh->dv_regc0_mask)
4554 * Add default mreg copy flow may be called multiple time, but
4555 * only be called once in stop. Avoid register it twice.
4557 mark_id = MLX5_DEFAULT_COPY_ID;
4558 ctx.data = &mark_id;
4559 if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4561 mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4568 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4570 * All the flow having Q/RSS action should be split by
4571 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4572 * performs the following,
4573 * - CQE->flow_tag := reg_c[1] (MARK)
4574 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4575 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4576 * but there should be a flow per each MARK ID set by MARK action.
4578 * For the aforementioned reason, if there's a MARK action in flow's action
4579 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4580 * the MARK ID to CQE's flow_tag like,
4581 * - If reg_c[1] is mark_id,
4582 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4584 * For SET_META action which stores value in reg_c[0], as the destination is
4585 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4586 * MARK ID means the default flow. The default flow looks like,
4587 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4590 * Pointer to Ethernet device.
4592 * Pointer to flow structure.
4593 * @param[in] actions
4594 * Pointer to the list of actions.
4596 * Perform verbose error reporting if not NULL.
4599 * 0 on success, negative value otherwise and rte_errno is set.
4602 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4603 struct rte_flow *flow,
4604 const struct rte_flow_action *actions,
4605 struct rte_flow_error *error)
4607 struct mlx5_priv *priv = dev->data->dev_private;
4608 struct mlx5_sh_config *config = &priv->sh->config;
4609 struct mlx5_flow_mreg_copy_resource *mcp_res;
4610 const struct rte_flow_action_mark *mark;
4612 /* Check whether extensive metadata feature is engaged. */
4613 if (!config->dv_flow_en ||
4614 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4615 !mlx5_flow_ext_mreg_supported(dev) ||
4616 !priv->sh->dv_regc0_mask)
4618 /* Find MARK action. */
4619 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4620 switch (actions->type) {
4621 case RTE_FLOW_ACTION_TYPE_FLAG:
4622 mcp_res = flow_mreg_add_copy_action
4623 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4626 flow->rix_mreg_copy = mcp_res->idx;
4628 case RTE_FLOW_ACTION_TYPE_MARK:
4629 mark = (const struct rte_flow_action_mark *)
4632 flow_mreg_add_copy_action(dev, mark->id, error);
4635 flow->rix_mreg_copy = mcp_res->idx;
4644 #define MLX5_MAX_SPLIT_ACTIONS 24
4645 #define MLX5_MAX_SPLIT_ITEMS 24
4648 * Split the hairpin flow.
4649 * Since HW can't support encap and push-vlan on Rx, we move these
4651 * If the count action is after the encap then we also
4652 * move the count action. in this case the count will also measure
4656 * Pointer to Ethernet device.
4657 * @param[in] actions
4658 * Associated actions (list terminated by the END action).
4659 * @param[out] actions_rx
4661 * @param[out] actions_tx
4663 * @param[out] pattern_tx
4664 * The pattern items for the Tx flow.
4665 * @param[out] flow_id
4666 * The flow ID connected to this flow.
4672 flow_hairpin_split(struct rte_eth_dev *dev,
4673 const struct rte_flow_action actions[],
4674 struct rte_flow_action actions_rx[],
4675 struct rte_flow_action actions_tx[],
4676 struct rte_flow_item pattern_tx[],
4679 const struct rte_flow_action_raw_encap *raw_encap;
4680 const struct rte_flow_action_raw_decap *raw_decap;
4681 struct mlx5_rte_flow_action_set_tag *set_tag;
4682 struct rte_flow_action *tag_action;
4683 struct mlx5_rte_flow_item_tag *tag_item;
4684 struct rte_flow_item *item;
4688 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4689 switch (actions->type) {
4690 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4691 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4692 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4693 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4694 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4695 rte_memcpy(actions_tx, actions,
4696 sizeof(struct rte_flow_action));
4699 case RTE_FLOW_ACTION_TYPE_COUNT:
4701 rte_memcpy(actions_tx, actions,
4702 sizeof(struct rte_flow_action));
4705 rte_memcpy(actions_rx, actions,
4706 sizeof(struct rte_flow_action));
4710 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4711 raw_encap = actions->conf;
4712 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4713 memcpy(actions_tx, actions,
4714 sizeof(struct rte_flow_action));
4718 rte_memcpy(actions_rx, actions,
4719 sizeof(struct rte_flow_action));
4723 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4724 raw_decap = actions->conf;
4725 if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4726 memcpy(actions_tx, actions,
4727 sizeof(struct rte_flow_action));
4730 rte_memcpy(actions_rx, actions,
4731 sizeof(struct rte_flow_action));
4736 rte_memcpy(actions_rx, actions,
4737 sizeof(struct rte_flow_action));
4742 /* Add set meta action and end action for the Rx flow. */
4743 tag_action = actions_rx;
4744 tag_action->type = (enum rte_flow_action_type)
4745 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4747 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4749 set_tag = (void *)actions_rx;
4750 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4751 .id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4754 MLX5_ASSERT(set_tag->id > REG_NON);
4755 tag_action->conf = set_tag;
4756 /* Create Tx item list. */
4757 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4758 addr = (void *)&pattern_tx[2];
4760 item->type = (enum rte_flow_item_type)
4761 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4762 tag_item = (void *)addr;
4763 tag_item->data = flow_id;
4764 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4765 MLX5_ASSERT(set_tag->id > REG_NON);
4766 item->spec = tag_item;
4767 addr += sizeof(struct mlx5_rte_flow_item_tag);
4768 tag_item = (void *)addr;
4769 tag_item->data = UINT32_MAX;
4770 tag_item->id = UINT16_MAX;
4771 item->mask = tag_item;
4774 item->type = RTE_FLOW_ITEM_TYPE_END;
4779 * The last stage of splitting chain, just creates the subflow
4780 * without any modification.
4783 * Pointer to Ethernet device.
4785 * Parent flow structure pointer.
4786 * @param[in, out] sub_flow
4787 * Pointer to return the created subflow, may be NULL.
4789 * Flow rule attributes.
4791 * Pattern specification (list terminated by the END pattern item).
4792 * @param[in] actions
4793 * Associated actions (list terminated by the END action).
4794 * @param[in] flow_split_info
4795 * Pointer to flow split info structure.
4797 * Perform verbose error reporting if not NULL.
4799 * 0 on success, negative value otherwise
4802 flow_create_split_inner(struct rte_eth_dev *dev,
4803 struct rte_flow *flow,
4804 struct mlx5_flow **sub_flow,
4805 const struct rte_flow_attr *attr,
4806 const struct rte_flow_item items[],
4807 const struct rte_flow_action actions[],
4808 struct mlx5_flow_split_info *flow_split_info,
4809 struct rte_flow_error *error)
4811 struct mlx5_flow *dev_flow;
4812 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
4814 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4815 flow_split_info->flow_idx, error);
4818 dev_flow->flow = flow;
4819 dev_flow->external = flow_split_info->external;
4820 dev_flow->skip_scale = flow_split_info->skip_scale;
4821 /* Subflow object was created, we must include one in the list. */
4822 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4823 dev_flow->handle, next);
4825 * If dev_flow is as one of the suffix flow, some actions in suffix
4826 * flow may need some user defined item layer flags, and pass the
4827 * Metadata rxq mark flag to suffix flow as well.
4829 if (flow_split_info->prefix_layers)
4830 dev_flow->handle->layers = flow_split_info->prefix_layers;
4831 if (flow_split_info->prefix_mark) {
4836 *sub_flow = dev_flow;
4837 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4838 dev_flow->dv.table_id = flow_split_info->table_id;
4840 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4844 * Get the sub policy of a meter.
4847 * Pointer to Ethernet device.
4849 * Parent flow structure pointer.
4851 * Pointer to thread flow work space.
4853 * Flow rule attributes.
4855 * Pattern specification (list terminated by the END pattern item).
4857 * Perform verbose error reporting if not NULL.
4860 * Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4862 static struct mlx5_flow_meter_sub_policy *
4863 get_meter_sub_policy(struct rte_eth_dev *dev,
4864 struct rte_flow *flow,
4865 struct mlx5_flow_workspace *wks,
4866 const struct rte_flow_attr *attr,
4867 const struct rte_flow_item items[],
4868 struct rte_flow_error *error)
4870 struct mlx5_flow_meter_policy *policy;
4871 struct mlx5_flow_meter_policy *final_policy;
4872 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4874 policy = wks->policy;
4875 final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4876 if (final_policy->is_rss || final_policy->is_queue) {
4877 struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4878 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4882 * This is a tmp dev_flow,
4883 * no need to register any matcher for it in translate.
4885 wks->skip_matcher_reg = 1;
4886 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4887 struct mlx5_flow dev_flow = {0};
4888 struct mlx5_flow_handle dev_handle = { {0} };
4889 uint8_t fate = final_policy->act_cnt[i].fate_action;
4891 if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4892 const struct rte_flow_action_rss *rss_act =
4893 final_policy->act_cnt[i].rss->conf;
4894 struct rte_flow_action rss_actions[2] = {
4896 .type = RTE_FLOW_ACTION_TYPE_RSS,
4900 .type = RTE_FLOW_ACTION_TYPE_END,
4905 dev_flow.handle = &dev_handle;
4906 dev_flow.ingress = attr->ingress;
4907 dev_flow.flow = flow;
4908 dev_flow.external = 0;
4909 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4910 dev_flow.dv.transfer = attr->transfer;
4913 * Translate RSS action to get rss hash fields.
4915 if (flow_drv_translate(dev, &dev_flow, attr,
4916 items, rss_actions, error))
4918 rss_desc_v[i] = wks->rss_desc;
4919 rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4920 rss_desc_v[i].hash_fields =
4921 dev_flow.hash_fields;
4922 rss_desc_v[i].queue_num =
4923 rss_desc_v[i].hash_fields ?
4924 rss_desc_v[i].queue_num : 1;
4925 rss_desc_v[i].tunnel =
4926 !!(dev_flow.handle->layers &
4927 MLX5_FLOW_LAYER_TUNNEL);
4928 /* Use the RSS queues in the containers. */
4929 rss_desc_v[i].queue =
4930 (uint16_t *)(uintptr_t)rss_act->queue;
4931 rss_desc[i] = &rss_desc_v[i];
4932 } else if (fate == MLX5_FLOW_FATE_QUEUE) {
4933 /* This is queue action. */
4934 rss_desc_v[i] = wks->rss_desc;
4935 rss_desc_v[i].key_len = 0;
4936 rss_desc_v[i].hash_fields = 0;
4937 rss_desc_v[i].queue =
4938 &final_policy->act_cnt[i].queue;
4939 rss_desc_v[i].queue_num = 1;
4940 rss_desc[i] = &rss_desc_v[i];
4945 sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4946 flow, policy, rss_desc);
4948 enum mlx5_meter_domain mtr_domain =
4949 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4950 (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4951 MLX5_MTR_DOMAIN_INGRESS);
4952 sub_policy = policy->sub_policys[mtr_domain][0];
4955 rte_flow_error_set(error, EINVAL,
4956 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4957 "Failed to get meter sub-policy.");
4963 * Split the meter flow.
4965 * As meter flow will split to three sub flow, other than meter
4966 * action, the other actions make sense to only meter accepts
4967 * the packet. If it need to be dropped, no other additional
4968 * actions should be take.
4970 * One kind of special action which decapsulates the L3 tunnel
4971 * header will be in the prefix sub flow, as not to take the
4972 * L3 tunnel header into account.
4975 * Pointer to Ethernet device.
4977 * Parent flow structure pointer.
4979 * Pointer to thread flow work space.
4981 * Flow rule attributes.
4983 * Pattern specification (list terminated by the END pattern item).
4984 * @param[out] sfx_items
4985 * Suffix flow match items (list terminated by the END pattern item).
4986 * @param[in] actions
4987 * Associated actions (list terminated by the END action).
4988 * @param[out] actions_sfx
4989 * Suffix flow actions.
4990 * @param[out] actions_pre
4991 * Prefix flow actions.
4992 * @param[out] mtr_flow_id
4993 * Pointer to meter flow id.
4995 * Perform verbose error reporting if not NULL.
4998 * 0 on success, a negative errno value otherwise and rte_errno is set.
5001 flow_meter_split_prep(struct rte_eth_dev *dev,
5002 struct rte_flow *flow,
5003 struct mlx5_flow_workspace *wks,
5004 const struct rte_flow_attr *attr,
5005 const struct rte_flow_item items[],
5006 struct rte_flow_item sfx_items[],
5007 const struct rte_flow_action actions[],
5008 struct rte_flow_action actions_sfx[],
5009 struct rte_flow_action actions_pre[],
5010 uint32_t *mtr_flow_id,
5011 struct rte_flow_error *error)
5013 struct mlx5_priv *priv = dev->data->dev_private;
5014 struct mlx5_flow_meter_info *fm = wks->fm;
5015 struct rte_flow_action *tag_action = NULL;
5016 struct rte_flow_item *tag_item;
5017 struct mlx5_rte_flow_action_set_tag *set_tag;
5018 const struct rte_flow_action_raw_encap *raw_encap;
5019 const struct rte_flow_action_raw_decap *raw_decap;
5020 struct mlx5_rte_flow_item_tag *tag_item_spec;
5021 struct mlx5_rte_flow_item_tag *tag_item_mask;
5022 uint32_t tag_id = 0;
5023 struct rte_flow_item *vlan_item_dst = NULL;
5024 const struct rte_flow_item *vlan_item_src = NULL;
5025 const struct rte_flow_item *orig_items = items;
5026 struct rte_flow_action *hw_mtr_action;
5027 struct rte_flow_action *action_pre_head = NULL;
5028 int32_t flow_src_port = priv->representor_id;
5030 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
5031 uint8_t mtr_reg_bits = priv->mtr_reg_share ?
5032 MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
5033 uint32_t flow_id = 0;
5034 uint32_t flow_id_reversed = 0;
5035 uint8_t flow_id_bits = 0;
5038 /* Prepare the suffix subflow items. */
5039 tag_item = sfx_items++;
5040 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5041 struct mlx5_priv *port_priv;
5042 const struct rte_flow_item_port_id *pid_v;
5043 int item_type = items->type;
5045 switch (item_type) {
5046 case RTE_FLOW_ITEM_TYPE_PORT_ID:
5047 pid_v = items->spec;
5049 port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
5051 return rte_flow_error_set(error,
5053 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5055 "Failed to get port info.");
5056 flow_src_port = port_priv->representor_id;
5057 if (!fm->def_policy && wks->policy->is_hierarchy &&
5058 flow_src_port != priv->representor_id) {
5059 if (flow_drv_mtr_hierarchy_rule_create(dev,
5066 memcpy(sfx_items, items, sizeof(*sfx_items));
5069 case RTE_FLOW_ITEM_TYPE_VLAN:
5070 /* Determine if copy vlan item below. */
5071 vlan_item_src = items;
5072 vlan_item_dst = sfx_items++;
5073 vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5079 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5081 mtr_first = priv->sh->meter_aso_en &&
5082 (attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5083 /* For ASO meter, meter must be before tag in TX direction. */
5085 action_pre_head = actions_pre++;
5086 /* Leave space for tag action. */
5087 tag_action = actions_pre++;
5089 /* Prepare the actions for prefix and suffix flow. */
5090 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5091 struct rte_flow_action *action_cur = NULL;
5093 switch (actions->type) {
5094 case RTE_FLOW_ACTION_TYPE_METER:
5096 action_cur = action_pre_head;
5098 /* Leave space for tag action. */
5099 tag_action = actions_pre++;
5100 action_cur = actions_pre++;
5103 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5104 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5105 action_cur = actions_pre++;
5107 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5108 raw_encap = actions->conf;
5109 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5110 action_cur = actions_pre++;
5112 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5113 raw_decap = actions->conf;
5114 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5115 action_cur = actions_pre++;
5117 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5118 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5119 if (vlan_item_dst && vlan_item_src) {
5120 memcpy(vlan_item_dst, vlan_item_src,
5121 sizeof(*vlan_item_dst));
5123 * Convert to internal match item, it is used
5124 * for vlan push and set vid.
5126 vlan_item_dst->type = (enum rte_flow_item_type)
5127 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5134 action_cur = (fm->def_policy) ?
5135 actions_sfx++ : actions_pre++;
5136 memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5138 /* Add end action to the actions. */
5139 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5140 if (priv->sh->meter_aso_en) {
5142 * For ASO meter, need to add an extra jump action explicitly,
5143 * to jump from meter to policer table.
5145 struct mlx5_flow_meter_sub_policy *sub_policy;
5146 struct mlx5_flow_tbl_data_entry *tbl_data;
5148 if (!fm->def_policy) {
5149 sub_policy = get_meter_sub_policy(dev, flow, wks,
5155 enum mlx5_meter_domain mtr_domain =
5156 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5157 (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5158 MLX5_MTR_DOMAIN_INGRESS);
5161 &priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5163 tbl_data = container_of(sub_policy->tbl_rsc,
5164 struct mlx5_flow_tbl_data_entry, tbl);
5165 hw_mtr_action = actions_pre++;
5166 hw_mtr_action->type = (enum rte_flow_action_type)
5167 MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5168 hw_mtr_action->conf = tbl_data->jump.action;
5170 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5173 return rte_flow_error_set(error, ENOMEM,
5174 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5175 NULL, "No tag action space.");
5177 tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5180 /* Only default-policy Meter creates mtr flow id. */
5181 if (fm->def_policy) {
5182 mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5184 return rte_flow_error_set(error, ENOMEM,
5185 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5186 "Failed to allocate meter flow id.");
5187 flow_id = tag_id - 1;
5188 flow_id_bits = (!flow_id) ? 1 :
5189 (MLX5_REG_BITS - __builtin_clz(flow_id));
5190 if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5192 mlx5_ipool_free(fm->flow_ipool, tag_id);
5193 return rte_flow_error_set(error, EINVAL,
5194 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5195 "Meter flow id exceeds max limit.");
5197 if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5198 priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5200 /* Build tag actions and items for meter_id/meter flow_id. */
5201 set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5202 tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5203 tag_item_mask = tag_item_spec + 1;
5204 /* Both flow_id and meter_id share the same register. */
5205 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5206 .id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5208 .offset = mtr_id_offset,
5209 .length = mtr_reg_bits,
5210 .data = flow->meter,
5213 * The color Reg bits used by flow_id are growing from
5214 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5216 for (shift = 0; shift < flow_id_bits; shift++)
5217 flow_id_reversed = (flow_id_reversed << 1) |
5218 ((flow_id >> shift) & 0x1);
5220 flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5221 tag_item_spec->id = set_tag->id;
5222 tag_item_spec->data = set_tag->data << mtr_id_offset;
5223 tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5224 tag_action->type = (enum rte_flow_action_type)
5225 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5226 tag_action->conf = set_tag;
5227 tag_item->type = (enum rte_flow_item_type)
5228 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5229 tag_item->spec = tag_item_spec;
5230 tag_item->last = NULL;
5231 tag_item->mask = tag_item_mask;
5234 *mtr_flow_id = tag_id;
5239 * Split action list having QUEUE/RSS for metadata register copy.
5241 * Once Q/RSS action is detected in user's action list, the flow action
5242 * should be split in order to copy metadata registers, which will happen in
5244 * - CQE->flow_tag := reg_c[1] (MARK)
5245 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5246 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5247 * This is because the last action of each flow must be a terminal action
5248 * (QUEUE, RSS or DROP).
5250 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5251 * stored and kept in the mlx5_flow structure per each sub_flow.
5253 * The Q/RSS action is replaced with,
5254 * - SET_TAG, setting the allocated flow ID to reg_c[2].
5255 * And the following JUMP action is added at the end,
5256 * - JUMP, to RX_CP_TBL.
5258 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5259 * flow_create_split_metadata() routine. The flow will look like,
5260 * - If flow ID matches (reg_c[2]), perform Q/RSS.
5263 * Pointer to Ethernet device.
5264 * @param[out] split_actions
5265 * Pointer to store split actions to jump to CP_TBL.
5266 * @param[in] actions
5267 * Pointer to the list of original flow actions.
5269 * Pointer to the Q/RSS action.
5270 * @param[in] actions_n
5271 * Number of original actions.
5272 * @param[in] mtr_sfx
5273 * Check if it is in meter suffix table.
5275 * Perform verbose error reporting if not NULL.
5278 * non-zero unique flow_id on success, otherwise 0 and
5279 * error/rte_error are set.
5282 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5283 struct rte_flow_action *split_actions,
5284 const struct rte_flow_action *actions,
5285 const struct rte_flow_action *qrss,
5286 int actions_n, int mtr_sfx,
5287 struct rte_flow_error *error)
5289 struct mlx5_priv *priv = dev->data->dev_private;
5290 struct mlx5_rte_flow_action_set_tag *set_tag;
5291 struct rte_flow_action_jump *jump;
5292 const int qrss_idx = qrss - actions;
5293 uint32_t flow_id = 0;
5297 * Given actions will be split
5298 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5299 * - Add jump to mreg CP_TBL.
5300 * As a result, there will be one more action.
5302 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5303 /* Count MLX5_RTE_FLOW_ACTION_TYPE_TAG. */
5305 set_tag = (void *)(split_actions + actions_n);
5307 * If we are not the meter suffix flow, add the tag action.
5308 * Since meter suffix flow already has the tag added.
5312 * Allocate the new subflow ID. This one is unique within
5313 * device and not shared with representors. Otherwise,
5314 * we would have to resolve multi-thread access synch
5315 * issue. Each flow on the shared device is appended
5316 * with source vport identifier, so the resulting
5317 * flows will be unique in the shared (by master and
5318 * representors) domain even if they have coinciding
5321 mlx5_ipool_malloc(priv->sh->ipool
5322 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5324 return rte_flow_error_set(error, ENOMEM,
5325 RTE_FLOW_ERROR_TYPE_ACTION,
5326 NULL, "can't allocate id "
5327 "for split Q/RSS subflow");
5328 /* Internal SET_TAG action to set flow ID. */
5329 *set_tag = (struct mlx5_rte_flow_action_set_tag){
5332 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5336 /* Construct new actions array. */
5337 /* Replace QUEUE/RSS action. */
5338 split_actions[qrss_idx] = (struct rte_flow_action){
5339 .type = (enum rte_flow_action_type)
5340 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5345 * If we are the suffix flow of meter, tag already exist.
5346 * Set the QUEUE/RSS action to void.
5348 split_actions[qrss_idx].type = RTE_FLOW_ACTION_TYPE_VOID;
5350 /* JUMP action to jump to mreg copy table (CP_TBL). */
5351 jump = (void *)(set_tag + 1);
5352 *jump = (struct rte_flow_action_jump){
5353 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5355 split_actions[actions_n - 2] = (struct rte_flow_action){
5356 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5359 split_actions[actions_n - 1] = (struct rte_flow_action){
5360 .type = RTE_FLOW_ACTION_TYPE_END,
5366 * Extend the given action list for Tx metadata copy.
5368 * Copy the given action list to the ext_actions and add flow metadata register
5369 * copy action in order to copy reg_a set by WQE to reg_c[0].
5371 * @param[out] ext_actions
5372 * Pointer to the extended action list.
5373 * @param[in] actions
5374 * Pointer to the list of actions.
5375 * @param[in] actions_n
5376 * Number of actions in the list.
5378 * Perform verbose error reporting if not NULL.
5379 * @param[in] encap_idx
5380 * The encap action index.
5383 * 0 on success, negative value otherwise
5386 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5387 struct rte_flow_action *ext_actions,
5388 const struct rte_flow_action *actions,
5389 int actions_n, struct rte_flow_error *error,
5392 struct mlx5_flow_action_copy_mreg *cp_mreg =
5393 (struct mlx5_flow_action_copy_mreg *)
5394 (ext_actions + actions_n + 1);
5397 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5401 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5406 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5407 if (encap_idx == actions_n - 1) {
5408 ext_actions[actions_n - 1] = (struct rte_flow_action){
5409 .type = (enum rte_flow_action_type)
5410 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5413 ext_actions[actions_n] = (struct rte_flow_action){
5414 .type = RTE_FLOW_ACTION_TYPE_END,
5417 ext_actions[encap_idx] = (struct rte_flow_action){
5418 .type = (enum rte_flow_action_type)
5419 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5422 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5423 sizeof(*ext_actions) * (actions_n - encap_idx));
5429 * Check the match action from the action list.
5431 * @param[in] actions
5432 * Pointer to the list of actions.
5434 * Flow rule attributes.
5436 * The action to be check if exist.
5437 * @param[out] match_action_pos
5438 * Pointer to the position of the matched action if exists, otherwise is -1.
5439 * @param[out] qrss_action_pos
5440 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5441 * @param[out] modify_after_mirror
5442 * Pointer to the flag of modify action after FDB mirroring.
5445 * > 0 the total number of actions.
5446 * 0 if not found match action in action list.
5449 flow_check_match_action(const struct rte_flow_action actions[],
5450 const struct rte_flow_attr *attr,
5451 enum rte_flow_action_type action,
5452 int *match_action_pos, int *qrss_action_pos,
5453 int *modify_after_mirror)
5455 const struct rte_flow_action_sample *sample;
5456 const struct rte_flow_action_raw_decap *decap;
5463 *match_action_pos = -1;
5464 *qrss_action_pos = -1;
5465 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5466 if (actions->type == action) {
5468 *match_action_pos = actions_n;
5470 switch (actions->type) {
5471 case RTE_FLOW_ACTION_TYPE_QUEUE:
5472 case RTE_FLOW_ACTION_TYPE_RSS:
5473 *qrss_action_pos = actions_n;
5475 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5476 sample = actions->conf;
5477 ratio = sample->ratio;
5478 sub_type = ((const struct rte_flow_action *)
5479 (sample->actions))->type;
5480 if (ratio == 1 && attr->transfer)
5483 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5484 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5485 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5486 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5487 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5488 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5489 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5490 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5491 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5492 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5493 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5494 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5495 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5496 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5497 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5498 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5499 case RTE_FLOW_ACTION_TYPE_FLAG:
5500 case RTE_FLOW_ACTION_TYPE_MARK:
5501 case RTE_FLOW_ACTION_TYPE_SET_META:
5502 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5503 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5504 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5505 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5506 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5507 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5508 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5509 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5510 case RTE_FLOW_ACTION_TYPE_METER:
5512 *modify_after_mirror = 1;
5514 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5515 decap = actions->conf;
5516 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5519 if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5520 const struct rte_flow_action_raw_encap *encap =
5523 MLX5_ENCAPSULATION_DECISION_SIZE &&
5525 MLX5_ENCAPSULATION_DECISION_SIZE)
5530 *modify_after_mirror = 1;
5537 if (flag && fdb_mirror && !*modify_after_mirror) {
5538 /* FDB mirroring uses the destination array to implement
5539 * instead of FLOW_SAMPLER object.
5541 if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5544 /* Count RTE_FLOW_ACTION_TYPE_END. */
5545 return flag ? actions_n + 1 : 0;
5548 #define SAMPLE_SUFFIX_ITEM 2
5551 * Split the sample flow.
5553 * As sample flow will split to two sub flow, sample flow with
5554 * sample action, the other actions will move to new suffix flow.
5556 * Also add unique tag id with tag action in the sample flow,
5557 * the same tag id will be as match in the suffix flow.
5560 * Pointer to Ethernet device.
5561 * @param[in] add_tag
5562 * Add extra tag action flag.
5563 * @param[out] sfx_items
5564 * Suffix flow match items (list terminated by the END pattern item).
5565 * @param[in] actions
5566 * Associated actions (list terminated by the END action).
5567 * @param[out] actions_sfx
5568 * Suffix flow actions.
5569 * @param[out] actions_pre
5570 * Prefix flow actions.
5571 * @param[in] actions_n
5572 * The total number of actions.
5573 * @param[in] sample_action_pos
5574 * The sample action position.
5575 * @param[in] qrss_action_pos
5576 * The Queue/RSS action position.
5577 * @param[in] jump_table
5578 * Add extra jump action flag.
5580 * Perform verbose error reporting if not NULL.
5583 * 0 on success, or unique flow_id, a negative errno value
5584 * otherwise and rte_errno is set.
5587 flow_sample_split_prep(struct rte_eth_dev *dev,
5589 struct rte_flow_item sfx_items[],
5590 const struct rte_flow_action actions[],
5591 struct rte_flow_action actions_sfx[],
5592 struct rte_flow_action actions_pre[],
5594 int sample_action_pos,
5595 int qrss_action_pos,
5597 struct rte_flow_error *error)
5599 struct mlx5_priv *priv = dev->data->dev_private;
5600 struct mlx5_rte_flow_action_set_tag *set_tag;
5601 struct mlx5_rte_flow_item_tag *tag_spec;
5602 struct mlx5_rte_flow_item_tag *tag_mask;
5603 struct rte_flow_action_jump *jump_action;
5604 uint32_t tag_id = 0;
5606 int append_index = 0;
5609 if (sample_action_pos < 0)
5610 return rte_flow_error_set(error, EINVAL,
5611 RTE_FLOW_ERROR_TYPE_ACTION,
5612 NULL, "invalid position of sample "
5614 /* Prepare the actions for prefix and suffix flow. */
5615 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5616 index = qrss_action_pos;
5617 /* Put the preceding the Queue/RSS action into prefix flow. */
5619 memcpy(actions_pre, actions,
5620 sizeof(struct rte_flow_action) * index);
5621 /* Put others preceding the sample action into prefix flow. */
5622 if (sample_action_pos > index + 1)
5623 memcpy(actions_pre + index, actions + index + 1,
5624 sizeof(struct rte_flow_action) *
5625 (sample_action_pos - index - 1));
5626 index = sample_action_pos - 1;
5627 /* Put Queue/RSS action into Suffix flow. */
5628 memcpy(actions_sfx, actions + qrss_action_pos,
5629 sizeof(struct rte_flow_action));
5632 index = sample_action_pos;
5634 memcpy(actions_pre, actions,
5635 sizeof(struct rte_flow_action) * index);
5637 /* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5638 * For CX6DX and above, metadata registers Cx preserve their value,
5639 * add an extra tag action for NIC-RX and E-Switch Domain.
5642 /* Prepare the prefix tag action. */
5644 set_tag = (void *)(actions_pre + actions_n + append_index);
5645 ret = mlx5_flow_get_reg_id(dev, MLX5_SAMPLE_ID, 0, error);
5648 mlx5_ipool_malloc(priv->sh->ipool
5649 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5650 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5654 /* Prepare the suffix subflow items. */
5655 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5656 tag_spec->data = tag_id;
5657 tag_spec->id = set_tag->id;
5658 tag_mask = tag_spec + 1;
5659 tag_mask->data = UINT32_MAX;
5660 sfx_items[0] = (struct rte_flow_item){
5661 .type = (enum rte_flow_item_type)
5662 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5667 sfx_items[1] = (struct rte_flow_item){
5668 .type = (enum rte_flow_item_type)
5669 RTE_FLOW_ITEM_TYPE_END,
5671 /* Prepare the tag action in prefix subflow. */
5672 actions_pre[index++] =
5673 (struct rte_flow_action){
5674 .type = (enum rte_flow_action_type)
5675 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5679 memcpy(actions_pre + index, actions + sample_action_pos,
5680 sizeof(struct rte_flow_action));
5682 /* For the modify action after the sample action in E-Switch mirroring,
5683 * Add the extra jump action in prefix subflow and jump into the next
5684 * table, then do the modify action in the new table.
5687 /* Prepare the prefix jump action. */
5689 jump_action = (void *)(actions_pre + actions_n + append_index);
5690 jump_action->group = jump_table;
5691 actions_pre[index++] =
5692 (struct rte_flow_action){
5693 .type = (enum rte_flow_action_type)
5694 RTE_FLOW_ACTION_TYPE_JUMP,
5695 .conf = jump_action,
5698 actions_pre[index] = (struct rte_flow_action){
5699 .type = (enum rte_flow_action_type)
5700 RTE_FLOW_ACTION_TYPE_END,
5702 /* Put the actions after sample into Suffix flow. */
5703 memcpy(actions_sfx, actions + sample_action_pos + 1,
5704 sizeof(struct rte_flow_action) *
5705 (actions_n - sample_action_pos - 1));
5710 * The splitting for metadata feature.
5712 * - Q/RSS action on NIC Rx should be split in order to pass by
5713 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5714 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5716 * - All the actions on NIC Tx should have a mreg copy action to
5717 * copy reg_a from WQE to reg_c[0].
5720 * Pointer to Ethernet device.
5722 * Parent flow structure pointer.
5724 * Flow rule attributes.
5726 * Pattern specification (list terminated by the END pattern item).
5727 * @param[in] actions
5728 * Associated actions (list terminated by the END action).
5729 * @param[in] flow_split_info
5730 * Pointer to flow split info structure.
5732 * Perform verbose error reporting if not NULL.
5734 * 0 on success, negative value otherwise
5737 flow_create_split_metadata(struct rte_eth_dev *dev,
5738 struct rte_flow *flow,
5739 const struct rte_flow_attr *attr,
5740 const struct rte_flow_item items[],
5741 const struct rte_flow_action actions[],
5742 struct mlx5_flow_split_info *flow_split_info,
5743 struct rte_flow_error *error)
5745 struct mlx5_priv *priv = dev->data->dev_private;
5746 struct mlx5_sh_config *config = &priv->sh->config;
5747 const struct rte_flow_action *qrss = NULL;
5748 struct rte_flow_action *ext_actions = NULL;
5749 struct mlx5_flow *dev_flow = NULL;
5750 uint32_t qrss_id = 0;
5757 /* Check whether extensive metadata feature is engaged. */
5758 if (!config->dv_flow_en ||
5759 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5760 !mlx5_flow_ext_mreg_supported(dev))
5761 return flow_create_split_inner(dev, flow, NULL, attr, items,
5762 actions, flow_split_info, error);
5763 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5766 /* Exclude hairpin flows from splitting. */
5767 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5768 const struct rte_flow_action_queue *queue;
5771 if (mlx5_rxq_get_type(dev, queue->index) ==
5772 MLX5_RXQ_TYPE_HAIRPIN)
5774 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5775 const struct rte_flow_action_rss *rss;
5778 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5779 MLX5_RXQ_TYPE_HAIRPIN)
5784 /* Check if it is in meter suffix table. */
5785 mtr_sfx = attr->group == (attr->transfer ?
5786 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5787 MLX5_FLOW_TABLE_LEVEL_METER);
5789 * Q/RSS action on NIC Rx should be split in order to pass by
5790 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5791 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5793 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5794 sizeof(struct rte_flow_action_set_tag) +
5795 sizeof(struct rte_flow_action_jump);
5796 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5799 return rte_flow_error_set(error, ENOMEM,
5800 RTE_FLOW_ERROR_TYPE_ACTION,
5801 NULL, "no memory to split "
5804 * Create the new actions list with removed Q/RSS action
5805 * and appended set tag and jump to register copy table
5806 * (RX_CP_TBL). We should preallocate unique tag ID here
5807 * in advance, because it is needed for set tag action.
5809 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5812 if (!mtr_sfx && !qrss_id) {
5816 } else if (attr->egress && !attr->transfer) {
5818 * All the actions on NIC Tx should have a metadata register
5819 * copy action to copy reg_a from WQE to reg_c[meta]
5821 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5822 sizeof(struct mlx5_flow_action_copy_mreg);
5823 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5826 return rte_flow_error_set(error, ENOMEM,
5827 RTE_FLOW_ERROR_TYPE_ACTION,
5828 NULL, "no memory to split "
5830 /* Create the action list appended with copy register. */
5831 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5832 actions_n, error, encap_idx);
5836 /* Add the unmodified original or prefix subflow. */
5837 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5838 items, ext_actions ? ext_actions :
5839 actions, flow_split_info, error);
5842 MLX5_ASSERT(dev_flow);
5844 const struct rte_flow_attr q_attr = {
5845 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5848 /* Internal PMD action to set register. */
5849 struct mlx5_rte_flow_item_tag q_tag_spec = {
5853 struct rte_flow_item q_items[] = {
5855 .type = (enum rte_flow_item_type)
5856 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5857 .spec = &q_tag_spec,
5862 .type = RTE_FLOW_ITEM_TYPE_END,
5865 struct rte_flow_action q_actions[] = {
5871 .type = RTE_FLOW_ACTION_TYPE_END,
5874 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5877 * Configure the tag item only if there is no meter subflow.
5878 * Since tag is already marked in the meter suffix subflow
5879 * we can just use the meter suffix items as is.
5882 /* Not meter subflow. */
5883 MLX5_ASSERT(!mtr_sfx);
5885 * Put unique id in prefix flow due to it is destroyed
5886 * after suffix flow and id will be freed after there
5887 * is no actual flows with this id and identifier
5888 * reallocation becomes possible (for example, for
5889 * other flows in other threads).
5891 dev_flow->handle->split_flow_id = qrss_id;
5892 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5896 q_tag_spec.id = ret;
5899 /* Add suffix subflow to execute Q/RSS. */
5900 flow_split_info->prefix_layers = layers;
5901 flow_split_info->prefix_mark = 0;
5902 flow_split_info->table_id = 0;
5903 ret = flow_create_split_inner(dev, flow, &dev_flow,
5904 &q_attr, mtr_sfx ? items :
5906 flow_split_info, error);
5909 /* qrss ID should be freed if failed. */
5911 MLX5_ASSERT(dev_flow);
5916 * We do not destroy the partially created sub_flows in case of error.
5917 * These ones are included into parent flow list and will be destroyed
5918 * by flow_drv_destroy.
5920 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5922 mlx5_free(ext_actions);
5927 * Create meter internal drop flow with the original pattern.
5930 * Pointer to Ethernet device.
5932 * Parent flow structure pointer.
5934 * Flow rule attributes.
5936 * Pattern specification (list terminated by the END pattern item).
5937 * @param[in] flow_split_info
5938 * Pointer to flow split info structure.
5940 * Pointer to flow meter structure.
5942 * Perform verbose error reporting if not NULL.
5944 * 0 on success, negative value otherwise
5947 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5948 struct rte_flow *flow,
5949 const struct rte_flow_attr *attr,
5950 const struct rte_flow_item items[],
5951 struct mlx5_flow_split_info *flow_split_info,
5952 struct mlx5_flow_meter_info *fm,
5953 struct rte_flow_error *error)
5955 struct mlx5_flow *dev_flow = NULL;
5956 struct rte_flow_attr drop_attr = *attr;
5957 struct rte_flow_action drop_actions[3];
5958 struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5960 MLX5_ASSERT(fm->drop_cnt);
5961 drop_actions[0].type =
5962 (enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5963 drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5964 drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5965 drop_actions[1].conf = NULL;
5966 drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5967 drop_actions[2].conf = NULL;
5968 drop_split_info.external = false;
5969 drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5970 drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5971 drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5972 return flow_create_split_inner(dev, flow, &dev_flow,
5973 &drop_attr, items, drop_actions,
5974 &drop_split_info, error);
5978 * The splitting for meter feature.
5980 * - The meter flow will be split to two flows as prefix and
5981 * suffix flow. The packets make sense only it pass the prefix
5984 * - Reg_C_5 is used for the packet to match betweend prefix and
5988 * Pointer to Ethernet device.
5990 * Parent flow structure pointer.
5992 * Flow rule attributes.
5994 * Pattern specification (list terminated by the END pattern item).
5995 * @param[in] actions
5996 * Associated actions (list terminated by the END action).
5997 * @param[in] flow_split_info
5998 * Pointer to flow split info structure.
6000 * Perform verbose error reporting if not NULL.
6002 * 0 on success, negative value otherwise
6005 flow_create_split_meter(struct rte_eth_dev *dev,
6006 struct rte_flow *flow,
6007 const struct rte_flow_attr *attr,
6008 const struct rte_flow_item items[],
6009 const struct rte_flow_action actions[],
6010 struct mlx5_flow_split_info *flow_split_info,
6011 struct rte_flow_error *error)
6013 struct mlx5_priv *priv = dev->data->dev_private;
6014 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6015 struct rte_flow_action *sfx_actions = NULL;
6016 struct rte_flow_action *pre_actions = NULL;
6017 struct rte_flow_item *sfx_items = NULL;
6018 struct mlx5_flow *dev_flow = NULL;
6019 struct rte_flow_attr sfx_attr = *attr;
6020 struct mlx5_flow_meter_info *fm = NULL;
6021 uint8_t skip_scale_restore;
6022 bool has_mtr = false;
6023 bool has_modify = false;
6024 bool set_mtr_reg = true;
6025 bool is_mtr_hierarchy = false;
6026 uint32_t meter_id = 0;
6027 uint32_t mtr_idx = 0;
6028 uint32_t mtr_flow_id = 0;
6035 actions_n = flow_check_meter_action(dev, actions, &has_mtr,
6036 &has_modify, &meter_id);
6039 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
6041 return rte_flow_error_set(error, EINVAL,
6042 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6043 NULL, "Meter not found.");
6045 fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
6047 return rte_flow_error_set(error, EINVAL,
6048 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6049 NULL, "Meter not found.");
6050 ret = mlx5_flow_meter_attach(priv, fm,
6054 flow->meter = mtr_idx;
6058 if (!fm->def_policy) {
6059 wks->policy = mlx5_flow_meter_policy_find(dev,
6062 MLX5_ASSERT(wks->policy);
6063 if (wks->policy->is_hierarchy) {
6065 mlx5_flow_meter_hierarchy_get_final_policy(dev,
6067 if (!wks->final_policy)
6068 return rte_flow_error_set(error,
6070 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6071 "Failed to find terminal policy of hierarchy.");
6072 is_mtr_hierarchy = true;
6076 * If it isn't default-policy Meter, and
6077 * 1. There's no action in flow to change
6078 * packet (modify/encap/decap etc.), OR
6079 * 2. No drop count needed for this meter.
6080 * 3. It's not meter hierarchy.
6081 * Then no need to use regC to save meter id anymore.
6083 if (!fm->def_policy && !is_mtr_hierarchy &&
6084 (!has_modify || !fm->drop_cnt))
6085 set_mtr_reg = false;
6086 /* Prefix actions: meter, decap, encap, tag, jump, end. */
6087 act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
6088 sizeof(struct mlx5_rte_flow_action_set_tag);
6089 /* Suffix items: tag, vlan, port id, end. */
6090 #define METER_SUFFIX_ITEM 4
6091 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6092 sizeof(struct mlx5_rte_flow_item_tag) * 2;
6093 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6096 return rte_flow_error_set(error, ENOMEM,
6097 RTE_FLOW_ERROR_TYPE_ACTION,
6098 NULL, "no memory to split "
6100 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6102 /* There's no suffix flow for meter of non-default policy. */
6103 if (!fm->def_policy)
6104 pre_actions = sfx_actions + 1;
6106 pre_actions = sfx_actions + actions_n;
6107 ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6108 items, sfx_items, actions,
6109 sfx_actions, pre_actions,
6110 (set_mtr_reg ? &mtr_flow_id : NULL),
6116 /* Add the prefix subflow. */
6117 skip_scale_restore = flow_split_info->skip_scale;
6118 flow_split_info->skip_scale |=
6119 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6120 ret = flow_create_split_inner(dev, flow, &dev_flow,
6121 attr, items, pre_actions,
6122 flow_split_info, error);
6123 flow_split_info->skip_scale = skip_scale_restore;
6126 mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6131 dev_flow->handle->split_flow_id = mtr_flow_id;
6132 dev_flow->handle->is_meter_flow_id = 1;
6134 if (!fm->def_policy) {
6135 if (!set_mtr_reg && fm->drop_cnt)
6137 flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6143 /* Setting the sfx group atrr. */
6144 sfx_attr.group = sfx_attr.transfer ?
6145 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6146 MLX5_FLOW_TABLE_LEVEL_METER;
6147 flow_split_info->prefix_layers =
6148 flow_get_prefix_layer_flags(dev_flow);
6149 flow_split_info->prefix_mark |= wks->mark;
6150 flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6152 /* Add the prefix subflow. */
6153 ret = flow_create_split_metadata(dev, flow,
6154 &sfx_attr, sfx_items ?
6156 sfx_actions ? sfx_actions : actions,
6157 flow_split_info, error);
6160 mlx5_free(sfx_actions);
6165 * The splitting for sample feature.
6167 * Once Sample action is detected in the action list, the flow actions should
6168 * be split into prefix sub flow and suffix sub flow.
6170 * The original items remain in the prefix sub flow, all actions preceding the
6171 * sample action and the sample action itself will be copied to the prefix
6172 * sub flow, the actions following the sample action will be copied to the
6173 * suffix sub flow, Queue action always be located in the suffix sub flow.
6175 * In order to make the packet from prefix sub flow matches with suffix sub
6176 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6177 * flow uses tag item with the unique flow id.
6180 * Pointer to Ethernet device.
6182 * Parent flow structure pointer.
6184 * Flow rule attributes.
6186 * Pattern specification (list terminated by the END pattern item).
6187 * @param[in] actions
6188 * Associated actions (list terminated by the END action).
6189 * @param[in] flow_split_info
6190 * Pointer to flow split info structure.
6192 * Perform verbose error reporting if not NULL.
6194 * 0 on success, negative value otherwise
6197 flow_create_split_sample(struct rte_eth_dev *dev,
6198 struct rte_flow *flow,
6199 const struct rte_flow_attr *attr,
6200 const struct rte_flow_item items[],
6201 const struct rte_flow_action actions[],
6202 struct mlx5_flow_split_info *flow_split_info,
6203 struct rte_flow_error *error)
6205 struct mlx5_priv *priv = dev->data->dev_private;
6206 struct rte_flow_action *sfx_actions = NULL;
6207 struct rte_flow_action *pre_actions = NULL;
6208 struct rte_flow_item *sfx_items = NULL;
6209 struct mlx5_flow *dev_flow = NULL;
6210 struct rte_flow_attr sfx_attr = *attr;
6211 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6212 struct mlx5_flow_dv_sample_resource *sample_res;
6213 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6214 struct mlx5_flow_tbl_resource *sfx_tbl;
6215 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6219 uint32_t fdb_tx = 0;
6222 int sample_action_pos;
6223 int qrss_action_pos;
6225 int modify_after_mirror = 0;
6226 uint16_t jump_table = 0;
6227 const uint32_t next_ft_step = 1;
6230 if (priv->sampler_en)
6231 actions_n = flow_check_match_action(actions, attr,
6232 RTE_FLOW_ACTION_TYPE_SAMPLE,
6233 &sample_action_pos, &qrss_action_pos,
6234 &modify_after_mirror);
6236 /* The prefix actions must includes sample, tag, end. */
6237 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6238 + sizeof(struct mlx5_rte_flow_action_set_tag);
6239 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6240 sizeof(struct mlx5_rte_flow_item_tag) * 2;
6241 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6242 item_size), 0, SOCKET_ID_ANY);
6244 return rte_flow_error_set(error, ENOMEM,
6245 RTE_FLOW_ERROR_TYPE_ACTION,
6246 NULL, "no memory to split "
6248 /* The representor_id is UINT16_MAX for uplink. */
6249 fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6251 * When reg_c_preserve is set, metadata registers Cx preserve
6252 * their value even through packet duplication.
6254 add_tag = (!fdb_tx ||
6255 priv->sh->cdev->config.hca_attr.reg_c_preserve);
6257 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6259 if (modify_after_mirror)
6260 jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6262 pre_actions = sfx_actions + actions_n;
6263 tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6264 actions, sfx_actions,
6265 pre_actions, actions_n,
6267 qrss_action_pos, jump_table,
6269 if (tag_id < 0 || (add_tag && !tag_id)) {
6273 if (modify_after_mirror)
6274 flow_split_info->skip_scale =
6275 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6276 /* Add the prefix subflow. */
6277 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6279 flow_split_info, error);
6284 dev_flow->handle->split_flow_id = tag_id;
6285 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6286 if (!modify_after_mirror) {
6287 /* Set the sfx group attr. */
6288 sample_res = (struct mlx5_flow_dv_sample_resource *)
6289 dev_flow->dv.sample_res;
6290 sfx_tbl = (struct mlx5_flow_tbl_resource *)
6291 sample_res->normal_path_tbl;
6292 sfx_tbl_data = container_of(sfx_tbl,
6293 struct mlx5_flow_tbl_data_entry,
6295 sfx_attr.group = sfx_attr.transfer ?
6296 (sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6298 MLX5_ASSERT(attr->transfer);
6299 sfx_attr.group = jump_table;
6301 flow_split_info->prefix_layers =
6302 flow_get_prefix_layer_flags(dev_flow);
6304 flow_split_info->prefix_mark |= wks->mark;
6305 /* Suffix group level already be scaled with factor, set
6306 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6307 * again in translation.
6309 flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6312 /* Add the suffix subflow. */
6313 ret = flow_create_split_meter(dev, flow, &sfx_attr,
6314 sfx_items ? sfx_items : items,
6315 sfx_actions ? sfx_actions : actions,
6316 flow_split_info, error);
6319 mlx5_free(sfx_actions);
6324 * Split the flow to subflow set. The splitters might be linked
6325 * in the chain, like this:
6326 * flow_create_split_outer() calls:
6327 * flow_create_split_meter() calls:
6328 * flow_create_split_metadata(meter_subflow_0) calls:
6329 * flow_create_split_inner(metadata_subflow_0)
6330 * flow_create_split_inner(metadata_subflow_1)
6331 * flow_create_split_inner(metadata_subflow_2)
6332 * flow_create_split_metadata(meter_subflow_1) calls:
6333 * flow_create_split_inner(metadata_subflow_0)
6334 * flow_create_split_inner(metadata_subflow_1)
6335 * flow_create_split_inner(metadata_subflow_2)
6337 * This provide flexible way to add new levels of flow splitting.
6338 * The all of successfully created subflows are included to the
6339 * parent flow dev_flow list.
6342 * Pointer to Ethernet device.
6344 * Parent flow structure pointer.
6346 * Flow rule attributes.
6348 * Pattern specification (list terminated by the END pattern item).
6349 * @param[in] actions
6350 * Associated actions (list terminated by the END action).
6351 * @param[in] flow_split_info
6352 * Pointer to flow split info structure.
6354 * Perform verbose error reporting if not NULL.
6356 * 0 on success, negative value otherwise
6359 flow_create_split_outer(struct rte_eth_dev *dev,
6360 struct rte_flow *flow,
6361 const struct rte_flow_attr *attr,
6362 const struct rte_flow_item items[],
6363 const struct rte_flow_action actions[],
6364 struct mlx5_flow_split_info *flow_split_info,
6365 struct rte_flow_error *error)
6369 ret = flow_create_split_sample(dev, flow, attr, items,
6370 actions, flow_split_info, error);
6371 MLX5_ASSERT(ret <= 0);
6375 static inline struct mlx5_flow_tunnel *
6376 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6378 struct mlx5_flow_tunnel *tunnel;
6380 #pragma GCC diagnostic push
6381 #pragma GCC diagnostic ignored "-Wcast-qual"
6382 tunnel = (typeof(tunnel))flow->tunnel;
6383 #pragma GCC diagnostic pop
6389 * Adjust flow RSS workspace if needed.
6392 * Pointer to thread flow work space.
6394 * Pointer to RSS descriptor.
6395 * @param[in] nrssq_num
6396 * New RSS queue number.
6399 * 0 on success, -1 otherwise and rte_errno is set.
6402 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6403 struct mlx5_flow_rss_desc *rss_desc,
6406 if (likely(nrssq_num <= wks->rssq_num))
6408 rss_desc->queue = realloc(rss_desc->queue,
6409 sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6410 if (!rss_desc->queue) {
6414 wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6419 * Create a flow and add it to @p list.
6422 * Pointer to Ethernet device.
6424 * Pointer to a TAILQ flow list. If this parameter NULL,
6425 * no list insertion occurred, flow is just created,
6426 * this is caller's responsibility to track the
6429 * Flow rule attributes.
6431 * Pattern specification (list terminated by the END pattern item).
6432 * @param[in] actions
6433 * Associated actions (list terminated by the END action).
6434 * @param[in] external
6435 * This flow rule is created by request external to PMD.
6437 * Perform verbose error reporting if not NULL.
6440 * A flow index on success, 0 otherwise and rte_errno is set.
6443 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6444 const struct rte_flow_attr *attr,
6445 const struct rte_flow_item items[],
6446 const struct rte_flow_action original_actions[],
6447 bool external, struct rte_flow_error *error)
6449 struct mlx5_priv *priv = dev->data->dev_private;
6450 struct rte_flow *flow = NULL;
6451 struct mlx5_flow *dev_flow;
6452 const struct rte_flow_action_rss *rss = NULL;
6453 struct mlx5_translated_action_handle
6454 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6455 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6457 struct mlx5_flow_expand_rss buf;
6458 uint8_t buffer[4096];
6461 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6462 uint8_t buffer[2048];
6465 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6466 uint8_t buffer[2048];
6467 } actions_hairpin_tx;
6469 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6470 uint8_t buffer[2048];
6472 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6473 struct mlx5_flow_rss_desc *rss_desc;
6474 const struct rte_flow_action *p_actions_rx;
6478 struct rte_flow_attr attr_tx = { .priority = 0 };
6479 const struct rte_flow_action *actions;
6480 struct rte_flow_action *translated_actions = NULL;
6481 struct mlx5_flow_tunnel *tunnel;
6482 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6483 struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6484 struct mlx5_flow_split_info flow_split_info = {
6485 .external = !!external,
6495 rss_desc = &wks->rss_desc;
6496 ret = flow_action_handles_translate(dev, original_actions,
6499 &translated_actions, error);
6501 MLX5_ASSERT(translated_actions == NULL);
6504 actions = translated_actions ? translated_actions : original_actions;
6505 p_actions_rx = actions;
6506 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6507 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6508 external, hairpin_flow, error);
6510 goto error_before_hairpin_split;
6511 flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6514 goto error_before_hairpin_split;
6516 if (hairpin_flow > 0) {
6517 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6519 goto error_before_hairpin_split;
6521 flow_hairpin_split(dev, actions, actions_rx.actions,
6522 actions_hairpin_tx.actions, items_tx.items,
6524 p_actions_rx = actions_rx.actions;
6526 flow_split_info.flow_idx = idx;
6527 flow->drv_type = flow_get_drv_type(dev, attr);
6528 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6529 flow->drv_type < MLX5_FLOW_TYPE_MAX);
6530 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6531 /* RSS Action only works on NIC RX domain */
6532 if (attr->ingress && !attr->transfer)
6533 rss = flow_get_rss_action(dev, p_actions_rx);
6535 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6538 * The following information is required by
6539 * mlx5_flow_hashfields_adjust() in advance.
6541 rss_desc->level = rss->level;
6542 /* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6543 rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6545 flow->dev_handles = 0;
6546 if (rss && rss->types) {
6547 unsigned int graph_root;
6549 graph_root = find_graph_root(rss->level);
6550 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6552 mlx5_support_expansion, graph_root);
6553 MLX5_ASSERT(ret > 0 &&
6554 (unsigned int)ret < sizeof(expand_buffer.buffer));
6555 if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6556 for (i = 0; i < buf->entries; ++i)
6557 mlx5_dbg__print_pattern(buf->entry[i].pattern);
6561 buf->entry[0].pattern = (void *)(uintptr_t)items;
6563 rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6565 for (i = 0; i < buf->entries; ++i) {
6566 /* Initialize flow split data. */
6567 flow_split_info.prefix_layers = 0;
6568 flow_split_info.prefix_mark = 0;
6569 flow_split_info.skip_scale = 0;
6571 * The splitter may create multiple dev_flows,
6572 * depending on configuration. In the simplest
6573 * case it just creates unmodified original flow.
6575 ret = flow_create_split_outer(dev, flow, attr,
6576 buf->entry[i].pattern,
6577 p_actions_rx, &flow_split_info,
6581 if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6582 ret = flow_tunnel_add_default_miss(dev, flow, attr,
6585 wks->flows[0].tunnel,
6589 mlx5_free(default_miss_ctx.queue);
6594 /* Create the tx flow. */
6596 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6597 attr_tx.ingress = 0;
6599 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6600 actions_hairpin_tx.actions,
6604 dev_flow->flow = flow;
6605 dev_flow->external = 0;
6606 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6607 dev_flow->handle, next);
6608 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6610 actions_hairpin_tx.actions, error);
6615 * Update the metadata register copy table. If extensive
6616 * metadata feature is enabled and registers are supported
6617 * we might create the extra rte_flow for each unique
6618 * MARK/FLAG action ID.
6620 * The table is updated for ingress Flows only, because
6621 * the egress Flows belong to the different device and
6622 * copy table should be updated in peer NIC Rx domain.
6624 if (attr->ingress &&
6625 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6626 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6631 * If the flow is external (from application) OR device is started,
6632 * OR mreg discover, then apply immediately.
6634 if (external || dev->data->dev_started ||
6635 (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6636 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6637 ret = flow_drv_apply(dev, flow, error);
6642 flow_rxq_flags_set(dev, flow);
6643 rte_free(translated_actions);
6644 tunnel = flow_tunnel_from_rule(wks->flows);
6647 flow->tunnel_id = tunnel->tunnel_id;
6648 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6649 mlx5_free(default_miss_ctx.queue);
6651 mlx5_flow_pop_thread_workspace();
6655 ret = rte_errno; /* Save rte_errno before cleanup. */
6656 flow_mreg_del_copy_action(dev, flow);
6657 flow_drv_destroy(dev, flow);
6658 if (rss_desc->shared_rss)
6659 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6661 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6662 rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6663 mlx5_ipool_free(priv->flows[type], idx);
6664 rte_errno = ret; /* Restore rte_errno. */
6667 mlx5_flow_pop_thread_workspace();
6668 error_before_hairpin_split:
6669 rte_free(translated_actions);
6674 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6675 * incoming packets to table 1.
6677 * Other flow rules, requested for group n, will be created in
6678 * e-switch table n+1.
6679 * Jump action to e-switch group n will be created to group n+1.
6681 * Used when working in switchdev mode, to utilise advantages of table 1
6685 * Pointer to Ethernet device.
6688 * Pointer to flow on success, NULL otherwise and rte_errno is set.
6691 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6693 const struct rte_flow_attr attr = {
6700 const struct rte_flow_item pattern = {
6701 .type = RTE_FLOW_ITEM_TYPE_END,
6703 struct rte_flow_action_jump jump = {
6706 const struct rte_flow_action actions[] = {
6708 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6712 .type = RTE_FLOW_ACTION_TYPE_END,
6715 struct rte_flow_error error;
6717 return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6719 actions, false, &error);
6723 * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6724 * and sq number, directs all packets to peer vport.
6727 * Pointer to Ethernet device.
6732 * Flow ID on success, 0 otherwise and rte_errno is set.
6735 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6737 struct rte_flow_attr attr = {
6739 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6744 struct rte_flow_item_port_id port_spec = {
6745 .id = MLX5_PORT_ESW_MGR,
6747 struct mlx5_rte_flow_item_tx_queue txq_spec = {
6750 struct rte_flow_item pattern[] = {
6752 .type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6756 .type = (enum rte_flow_item_type)
6757 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6761 .type = RTE_FLOW_ITEM_TYPE_END,
6764 struct rte_flow_action_jump jump = {
6767 struct rte_flow_action_port_id port = {
6768 .id = dev->data->port_id,
6770 struct rte_flow_action actions[] = {
6772 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6776 .type = RTE_FLOW_ACTION_TYPE_END,
6779 struct rte_flow_error error;
6782 * Creates group 0, highest priority jump flow.
6783 * Matches txq to bypass kernel packets.
6785 if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6786 false, &error) == 0)
6788 /* Create group 1, lowest priority redirect flow for txq. */
6790 actions[0].conf = &port;
6791 actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6792 return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6793 actions, false, &error);
6797 * Validate a flow supported by the NIC.
6799 * @see rte_flow_validate()
6803 mlx5_flow_validate(struct rte_eth_dev *dev,
6804 const struct rte_flow_attr *attr,
6805 const struct rte_flow_item items[],
6806 const struct rte_flow_action original_actions[],
6807 struct rte_flow_error *error)
6810 struct mlx5_translated_action_handle
6811 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6812 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6813 const struct rte_flow_action *actions;
6814 struct rte_flow_action *translated_actions = NULL;
6815 int ret = flow_action_handles_translate(dev, original_actions,
6818 &translated_actions, error);
6822 actions = translated_actions ? translated_actions : original_actions;
6823 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6824 ret = flow_drv_validate(dev, attr, items, actions,
6825 true, hairpin_flow, error);
6826 rte_free(translated_actions);
6833 * @see rte_flow_create()
6837 mlx5_flow_create(struct rte_eth_dev *dev,
6838 const struct rte_flow_attr *attr,
6839 const struct rte_flow_item items[],
6840 const struct rte_flow_action actions[],
6841 struct rte_flow_error *error)
6844 * If the device is not started yet, it is not allowed to created a
6845 * flow from application. PMD default flows and traffic control flows
6848 if (unlikely(!dev->data->dev_started)) {
6849 DRV_LOG(DEBUG, "port %u is not started when "
6850 "inserting a flow", dev->data->port_id);
6851 rte_flow_error_set(error, ENODEV,
6852 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6854 "port not started");
6858 return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6859 attr, items, actions,
6864 * Destroy a flow in a list.
6867 * Pointer to Ethernet device.
6868 * @param[in] flow_idx
6869 * Index of flow to destroy.
6872 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6875 struct mlx5_priv *priv = dev->data->dev_private;
6876 struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6880 MLX5_ASSERT(flow->type == type);
6882 * Update RX queue flags only if port is started, otherwise it is
6885 if (dev->data->dev_started)
6886 flow_rxq_flags_trim(dev, flow);
6887 flow_drv_destroy(dev, flow);
6889 struct mlx5_flow_tunnel *tunnel;
6891 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6893 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6894 mlx5_flow_tunnel_free(dev, tunnel);
6896 flow_mreg_del_copy_action(dev, flow);
6897 mlx5_ipool_free(priv->flows[type], flow_idx);
6901 * Destroy all flows.
6904 * Pointer to Ethernet device.
6906 * Flow type to be flushed.
6908 * If flushing is called actively.
6911 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6914 struct mlx5_priv *priv = dev->data->dev_private;
6915 uint32_t num_flushed = 0, fidx = 1;
6916 struct rte_flow *flow;
6918 MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6919 flow_list_destroy(dev, type, fidx);
6923 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6924 dev->data->port_id, num_flushed);
6929 * Stop all default actions for flows.
6932 * Pointer to Ethernet device.
6935 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6937 flow_mreg_del_default_copy_action(dev);
6938 flow_rxq_flags_clear(dev);
6942 * Start all default actions for flows.
6945 * Pointer to Ethernet device.
6947 * 0 on success, a negative errno value otherwise and rte_errno is set.
6950 mlx5_flow_start_default(struct rte_eth_dev *dev)
6952 struct rte_flow_error error;
6954 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6955 return flow_mreg_add_default_copy_action(dev, &error);
6959 * Release key of thread specific flow workspace data.
6962 flow_release_workspace(void *data)
6964 struct mlx5_flow_workspace *wks = data;
6965 struct mlx5_flow_workspace *next;
6969 free(wks->rss_desc.queue);
6976 * Get thread specific current flow workspace.
6978 * @return pointer to thread specific flow workspace data, NULL on error.
6980 struct mlx5_flow_workspace*
6981 mlx5_flow_get_thread_workspace(void)
6983 struct mlx5_flow_workspace *data;
6985 data = mlx5_flow_os_get_specific_workspace();
6986 MLX5_ASSERT(data && data->inuse);
6987 if (!data || !data->inuse)
6988 DRV_LOG(ERR, "flow workspace not initialized.");
6993 * Allocate and init new flow workspace.
6995 * @return pointer to flow workspace data, NULL on error.
6997 static struct mlx5_flow_workspace*
6998 flow_alloc_thread_workspace(void)
7000 struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
7003 DRV_LOG(ERR, "Failed to allocate flow workspace "
7007 data->rss_desc.queue = calloc(1,
7008 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
7009 if (!data->rss_desc.queue)
7011 data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
7014 free(data->rss_desc.queue);
7020 * Get new thread specific flow workspace.
7022 * If current workspace inuse, create new one and set as current.
7024 * @return pointer to thread specific flow workspace data, NULL on error.
7026 static struct mlx5_flow_workspace*
7027 mlx5_flow_push_thread_workspace(void)
7029 struct mlx5_flow_workspace *curr;
7030 struct mlx5_flow_workspace *data;
7032 curr = mlx5_flow_os_get_specific_workspace();
7034 data = flow_alloc_thread_workspace();
7037 } else if (!curr->inuse) {
7039 } else if (curr->next) {
7042 data = flow_alloc_thread_workspace();
7050 /* Set as current workspace */
7051 if (mlx5_flow_os_set_specific_workspace(data))
7052 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7057 * Close current thread specific flow workspace.
7059 * If previous workspace available, set it as current.
7061 * @return pointer to thread specific flow workspace data, NULL on error.
7064 mlx5_flow_pop_thread_workspace(void)
7066 struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7071 DRV_LOG(ERR, "Failed to close unused flow workspace.");
7077 if (mlx5_flow_os_set_specific_workspace(data->prev))
7078 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7082 * Verify the flow list is empty
7085 * Pointer to Ethernet device.
7087 * @return the number of flows not released.
7090 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7092 struct mlx5_priv *priv = dev->data->dev_private;
7093 struct rte_flow *flow;
7097 for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7098 MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7099 DRV_LOG(DEBUG, "port %u flow %p still referenced",
7100 dev->data->port_id, (void *)flow);
7108 * Enable default hairpin egress flow.
7111 * Pointer to Ethernet device.
7116 * 0 on success, a negative errno value otherwise and rte_errno is set.
7119 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7122 const struct rte_flow_attr attr = {
7126 struct mlx5_rte_flow_item_tx_queue queue_spec = {
7129 struct mlx5_rte_flow_item_tx_queue queue_mask = {
7130 .queue = UINT32_MAX,
7132 struct rte_flow_item items[] = {
7134 .type = (enum rte_flow_item_type)
7135 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7136 .spec = &queue_spec,
7138 .mask = &queue_mask,
7141 .type = RTE_FLOW_ITEM_TYPE_END,
7144 struct rte_flow_action_jump jump = {
7145 .group = MLX5_HAIRPIN_TX_TABLE,
7147 struct rte_flow_action actions[2];
7149 struct rte_flow_error error;
7151 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7152 actions[0].conf = &jump;
7153 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7154 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7155 &attr, items, actions, false, &error);
7158 "Failed to create ctrl flow: rte_errno(%d),"
7159 " type(%d), message(%s)",
7160 rte_errno, error.type,
7161 error.message ? error.message : " (no stated reason)");
7168 * Enable a control flow configured from the control plane.
7171 * Pointer to Ethernet device.
7173 * An Ethernet flow spec to apply.
7175 * An Ethernet flow mask to apply.
7177 * A VLAN flow spec to apply.
7179 * A VLAN flow mask to apply.
7182 * 0 on success, a negative errno value otherwise and rte_errno is set.
7185 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7186 struct rte_flow_item_eth *eth_spec,
7187 struct rte_flow_item_eth *eth_mask,
7188 struct rte_flow_item_vlan *vlan_spec,
7189 struct rte_flow_item_vlan *vlan_mask)
7191 struct mlx5_priv *priv = dev->data->dev_private;
7192 const struct rte_flow_attr attr = {
7194 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7196 struct rte_flow_item items[] = {
7198 .type = RTE_FLOW_ITEM_TYPE_ETH,
7204 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7205 RTE_FLOW_ITEM_TYPE_END,
7211 .type = RTE_FLOW_ITEM_TYPE_END,
7214 uint16_t queue[priv->reta_idx_n];
7215 struct rte_flow_action_rss action_rss = {
7216 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7218 .types = priv->rss_conf.rss_hf,
7219 .key_len = priv->rss_conf.rss_key_len,
7220 .queue_num = priv->reta_idx_n,
7221 .key = priv->rss_conf.rss_key,
7224 struct rte_flow_action actions[] = {
7226 .type = RTE_FLOW_ACTION_TYPE_RSS,
7227 .conf = &action_rss,
7230 .type = RTE_FLOW_ACTION_TYPE_END,
7234 struct rte_flow_error error;
7237 if (!priv->reta_idx_n || !priv->rxqs_n) {
7240 if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7241 action_rss.types = 0;
7242 for (i = 0; i != priv->reta_idx_n; ++i)
7243 queue[i] = (*priv->reta_idx)[i];
7244 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7245 &attr, items, actions, false, &error);
7252 * Enable a flow control configured from the control plane.
7255 * Pointer to Ethernet device.
7257 * An Ethernet flow spec to apply.
7259 * An Ethernet flow mask to apply.
7262 * 0 on success, a negative errno value otherwise and rte_errno is set.
7265 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7266 struct rte_flow_item_eth *eth_spec,
7267 struct rte_flow_item_eth *eth_mask)
7269 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7273 * Create default miss flow rule matching lacp traffic
7276 * Pointer to Ethernet device.
7278 * An Ethernet flow spec to apply.
7281 * 0 on success, a negative errno value otherwise and rte_errno is set.
7284 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7287 * The LACP matching is done by only using ether type since using
7288 * a multicast dst mac causes kernel to give low priority to this flow.
7290 static const struct rte_flow_item_eth lacp_spec = {
7291 .type = RTE_BE16(0x8809),
7293 static const struct rte_flow_item_eth lacp_mask = {
7296 const struct rte_flow_attr attr = {
7299 struct rte_flow_item items[] = {
7301 .type = RTE_FLOW_ITEM_TYPE_ETH,
7306 .type = RTE_FLOW_ITEM_TYPE_END,
7309 struct rte_flow_action actions[] = {
7311 .type = (enum rte_flow_action_type)
7312 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7315 .type = RTE_FLOW_ACTION_TYPE_END,
7318 struct rte_flow_error error;
7319 uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7320 &attr, items, actions,
7331 * @see rte_flow_destroy()
7335 mlx5_flow_destroy(struct rte_eth_dev *dev,
7336 struct rte_flow *flow,
7337 struct rte_flow_error *error __rte_unused)
7339 flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7340 (uintptr_t)(void *)flow);
7345 * Destroy all flows.
7347 * @see rte_flow_flush()
7351 mlx5_flow_flush(struct rte_eth_dev *dev,
7352 struct rte_flow_error *error __rte_unused)
7354 mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7361 * @see rte_flow_isolate()
7365 mlx5_flow_isolate(struct rte_eth_dev *dev,
7367 struct rte_flow_error *error)
7369 struct mlx5_priv *priv = dev->data->dev_private;
7371 if (dev->data->dev_started) {
7372 rte_flow_error_set(error, EBUSY,
7373 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7375 "port must be stopped first");
7378 priv->isolated = !!enable;
7380 dev->dev_ops = &mlx5_dev_ops_isolate;
7382 dev->dev_ops = &mlx5_dev_ops;
7384 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7385 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7393 * @see rte_flow_query()
7397 flow_drv_query(struct rte_eth_dev *dev,
7399 const struct rte_flow_action *actions,
7401 struct rte_flow_error *error)
7403 struct mlx5_priv *priv = dev->data->dev_private;
7404 const struct mlx5_flow_driver_ops *fops;
7405 struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7407 enum mlx5_flow_drv_type ftype;
7410 return rte_flow_error_set(error, ENOENT,
7411 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7413 "invalid flow handle");
7415 ftype = flow->drv_type;
7416 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7417 fops = flow_get_drv_ops(ftype);
7419 return fops->query(dev, flow, actions, data, error);
7425 * @see rte_flow_query()
7429 mlx5_flow_query(struct rte_eth_dev *dev,
7430 struct rte_flow *flow,
7431 const struct rte_flow_action *actions,
7433 struct rte_flow_error *error)
7437 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7445 * Get rte_flow callbacks.
7448 * Pointer to Ethernet device structure.
7450 * Pointer to operation-specific structure.
7455 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7456 const struct rte_flow_ops **ops)
7458 *ops = &mlx5_flow_ops;
7463 * Validate meter policy actions.
7464 * Dispatcher for action type specific validation.
7467 * Pointer to the Ethernet device structure.
7469 * The meter policy action object to validate.
7471 * Attributes of flow to determine steering domain.
7472 * @param[out] is_rss
7474 * @param[out] domain_bitmap
7476 * @param[out] is_def_policy
7477 * Is default policy or not.
7479 * Perform verbose error reporting if not NULL. Initialized in case of
7483 * 0 on success, otherwise negative errno value.
7486 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7487 const struct rte_flow_action *actions[RTE_COLORS],
7488 struct rte_flow_attr *attr,
7490 uint8_t *domain_bitmap,
7491 uint8_t *policy_mode,
7492 struct rte_mtr_error *error)
7494 const struct mlx5_flow_driver_ops *fops;
7496 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7497 return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7498 domain_bitmap, policy_mode, error);
7502 * Destroy the meter table set.
7505 * Pointer to Ethernet device.
7506 * @param[in] mtr_policy
7507 * Meter policy struct.
7510 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7511 struct mlx5_flow_meter_policy *mtr_policy)
7513 const struct mlx5_flow_driver_ops *fops;
7515 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7516 fops->destroy_mtr_acts(dev, mtr_policy);
7520 * Create policy action, lock free,
7521 * (mutex should be acquired by caller).
7522 * Dispatcher for action type specific call.
7525 * Pointer to the Ethernet device structure.
7526 * @param[in] mtr_policy
7527 * Meter policy struct.
7529 * Action specification used to create meter actions.
7531 * Perform verbose error reporting if not NULL. Initialized in case of
7535 * 0 on success, otherwise negative errno value.
7538 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7539 struct mlx5_flow_meter_policy *mtr_policy,
7540 const struct rte_flow_action *actions[RTE_COLORS],
7541 struct rte_mtr_error *error)
7543 const struct mlx5_flow_driver_ops *fops;
7545 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7546 return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7550 * Create policy rules, lock free,
7551 * (mutex should be acquired by caller).
7552 * Dispatcher for action type specific call.
7555 * Pointer to the Ethernet device structure.
7556 * @param[in] mtr_policy
7557 * Meter policy struct.
7560 * 0 on success, -1 otherwise.
7563 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7564 struct mlx5_flow_meter_policy *mtr_policy)
7566 const struct mlx5_flow_driver_ops *fops;
7568 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7569 return fops->create_policy_rules(dev, mtr_policy);
7573 * Destroy policy rules, lock free,
7574 * (mutex should be acquired by caller).
7575 * Dispatcher for action type specific call.
7578 * Pointer to the Ethernet device structure.
7579 * @param[in] mtr_policy
7580 * Meter policy struct.
7583 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7584 struct mlx5_flow_meter_policy *mtr_policy)
7586 const struct mlx5_flow_driver_ops *fops;
7588 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7589 fops->destroy_policy_rules(dev, mtr_policy);
7593 * Destroy the default policy table set.
7596 * Pointer to Ethernet device.
7599 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7601 const struct mlx5_flow_driver_ops *fops;
7603 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7604 fops->destroy_def_policy(dev);
7608 * Destroy the default policy table set.
7611 * Pointer to Ethernet device.
7614 * 0 on success, -1 otherwise.
7617 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7619 const struct mlx5_flow_driver_ops *fops;
7621 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7622 return fops->create_def_policy(dev);
7626 * Create the needed meter and suffix tables.
7629 * Pointer to Ethernet device.
7632 * 0 on success, -1 otherwise.
7635 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7636 struct mlx5_flow_meter_info *fm,
7638 uint8_t domain_bitmap)
7640 const struct mlx5_flow_driver_ops *fops;
7642 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7643 return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7647 * Destroy the meter table set.
7650 * Pointer to Ethernet device.
7652 * Pointer to the meter table set.
7655 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7656 struct mlx5_flow_meter_info *fm)
7658 const struct mlx5_flow_driver_ops *fops;
7660 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7661 fops->destroy_mtr_tbls(dev, fm);
7665 * Destroy the global meter drop table.
7668 * Pointer to Ethernet device.
7671 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7673 const struct mlx5_flow_driver_ops *fops;
7675 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7676 fops->destroy_mtr_drop_tbls(dev);
7680 * Destroy the sub policy table with RX queue.
7683 * Pointer to Ethernet device.
7684 * @param[in] mtr_policy
7685 * Pointer to meter policy table.
7688 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7689 struct mlx5_flow_meter_policy *mtr_policy)
7691 const struct mlx5_flow_driver_ops *fops;
7693 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7694 fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7698 * Allocate the needed aso flow meter id.
7701 * Pointer to Ethernet device.
7704 * Index to aso flow meter on success, NULL otherwise.
7707 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7709 const struct mlx5_flow_driver_ops *fops;
7711 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7712 return fops->create_meter(dev);
7716 * Free the aso flow meter id.
7719 * Pointer to Ethernet device.
7720 * @param[in] mtr_idx
7721 * Index to aso flow meter to be free.
7727 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7729 const struct mlx5_flow_driver_ops *fops;
7731 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7732 fops->free_meter(dev, mtr_idx);
7736 * Allocate a counter.
7739 * Pointer to Ethernet device structure.
7742 * Index to allocated counter on success, 0 otherwise.
7745 mlx5_counter_alloc(struct rte_eth_dev *dev)
7747 const struct mlx5_flow_driver_ops *fops;
7748 struct rte_flow_attr attr = { .transfer = 0 };
7750 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7751 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7752 return fops->counter_alloc(dev);
7755 "port %u counter allocate is not supported.",
7756 dev->data->port_id);
7764 * Pointer to Ethernet device structure.
7766 * Index to counter to be free.
7769 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7771 const struct mlx5_flow_driver_ops *fops;
7772 struct rte_flow_attr attr = { .transfer = 0 };
7774 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7775 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7776 fops->counter_free(dev, cnt);
7780 "port %u counter free is not supported.",
7781 dev->data->port_id);
7785 * Query counter statistics.
7788 * Pointer to Ethernet device structure.
7790 * Index to counter to query.
7792 * Set to clear counter statistics.
7794 * The counter hits packets number to save.
7796 * The counter hits bytes number to save.
7799 * 0 on success, a negative errno value otherwise.
7802 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7803 bool clear, uint64_t *pkts, uint64_t *bytes, void **action)
7805 const struct mlx5_flow_driver_ops *fops;
7806 struct rte_flow_attr attr = { .transfer = 0 };
7808 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7809 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7810 return fops->counter_query(dev, cnt, clear, pkts,
7814 "port %u counter query is not supported.",
7815 dev->data->port_id);
7820 * Allocate a new memory for the counter values wrapped by all the needed
7824 * Pointer to mlx5_dev_ctx_shared object.
7827 * 0 on success, a negative errno value otherwise.
7830 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7832 struct mlx5_counter_stats_mem_mng *mem_mng;
7833 volatile struct flow_counter_stats *raw_data;
7834 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7835 int size = (sizeof(struct flow_counter_stats) *
7836 MLX5_COUNTERS_PER_POOL +
7837 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7838 sizeof(struct mlx5_counter_stats_mem_mng);
7839 size_t pgsize = rte_mem_page_size();
7844 if (pgsize == (size_t)-1) {
7845 DRV_LOG(ERR, "Failed to get mem page size");
7849 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7854 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7855 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7856 ret = mlx5_os_wrapped_mkey_create(sh->cdev->ctx, sh->cdev->pd,
7857 sh->cdev->pdn, mem, size,
7864 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7865 raw_data = (volatile struct flow_counter_stats *)mem;
7866 for (i = 0; i < raws_n; ++i) {
7867 mem_mng->raws[i].mem_mng = mem_mng;
7868 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7870 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7871 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7872 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7874 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7875 sh->cmng.mem_mng = mem_mng;
7880 * Set the statistic memory to the new counter pool.
7883 * Pointer to mlx5_dev_ctx_shared object.
7885 * Pointer to the pool to set the statistic memory.
7888 * 0 on success, a negative errno value otherwise.
7891 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7892 struct mlx5_flow_counter_pool *pool)
7894 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7895 /* Resize statistic memory once used out. */
7896 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7897 mlx5_flow_create_counter_stat_mem_mng(sh)) {
7898 DRV_LOG(ERR, "Cannot resize counter stat mem.");
7901 rte_spinlock_lock(&pool->sl);
7902 pool->raw = cmng->mem_mng->raws + pool->index %
7903 MLX5_CNT_CONTAINER_RESIZE;
7904 rte_spinlock_unlock(&pool->sl);
7905 pool->raw_hw = NULL;
7909 #define MLX5_POOL_QUERY_FREQ_US 1000000
7912 * Set the periodic procedure for triggering asynchronous batch queries for all
7913 * the counter pools.
7916 * Pointer to mlx5_dev_ctx_shared object.
7919 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7921 uint32_t pools_n, us;
7923 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7924 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7925 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7926 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7927 sh->cmng.query_thread_on = 0;
7928 DRV_LOG(ERR, "Cannot reinitialize query alarm");
7930 sh->cmng.query_thread_on = 1;
7935 * The periodic procedure for triggering asynchronous batch queries for all the
7936 * counter pools. This function is probably called by the host thread.
7939 * The parameter for the alarm process.
7942 mlx5_flow_query_alarm(void *arg)
7944 struct mlx5_dev_ctx_shared *sh = arg;
7946 uint16_t pool_index = sh->cmng.pool_index;
7947 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7948 struct mlx5_flow_counter_pool *pool;
7951 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7953 rte_spinlock_lock(&cmng->pool_update_sl);
7954 pool = cmng->pools[pool_index];
7955 n_valid = cmng->n_valid;
7956 rte_spinlock_unlock(&cmng->pool_update_sl);
7957 /* Set the statistic memory to the new created pool. */
7958 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7961 /* There is a pool query in progress. */
7964 LIST_FIRST(&sh->cmng.free_stat_raws);
7966 /* No free counter statistics raw memory. */
7969 * Identify the counters released between query trigger and query
7970 * handle more efficiently. The counter released in this gap period
7971 * should wait for a new round of query as the new arrived packets
7972 * will not be taken into account.
7975 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7976 MLX5_COUNTERS_PER_POOL,
7978 pool->raw_hw->mem_mng->wm.lkey,
7982 (uint64_t)(uintptr_t)pool);
7984 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7985 " %d", pool->min_dcs->id);
7986 pool->raw_hw = NULL;
7989 LIST_REMOVE(pool->raw_hw, next);
7990 sh->cmng.pending_queries++;
7992 if (pool_index >= n_valid)
7995 sh->cmng.pool_index = pool_index;
7996 mlx5_set_query_alarm(sh);
8000 * Check and callback event for new aged flow in the counter pool
8003 * Pointer to mlx5_dev_ctx_shared object.
8005 * Pointer to Current counter pool.
8008 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
8009 struct mlx5_flow_counter_pool *pool)
8011 struct mlx5_priv *priv;
8012 struct mlx5_flow_counter *cnt;
8013 struct mlx5_age_info *age_info;
8014 struct mlx5_age_param *age_param;
8015 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
8016 struct mlx5_counter_stats_raw *prev = pool->raw;
8017 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
8018 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
8019 uint16_t expected = AGE_CANDIDATE;
8022 pool->time_of_last_age_check = curr_time;
8023 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
8024 cnt = MLX5_POOL_GET_CNT(pool, i);
8025 age_param = MLX5_CNT_TO_AGE(cnt);
8026 if (__atomic_load_n(&age_param->state,
8027 __ATOMIC_RELAXED) != AGE_CANDIDATE)
8029 if (cur->data[i].hits != prev->data[i].hits) {
8030 __atomic_store_n(&age_param->sec_since_last_hit, 0,
8034 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
8036 __ATOMIC_RELAXED) <= age_param->timeout)
8039 * Hold the lock first, or if between the
8040 * state AGE_TMOUT and tailq operation the
8041 * release happened, the release procedure
8042 * may delete a non-existent tailq node.
8044 priv = rte_eth_devices[age_param->port_id].data->dev_private;
8045 age_info = GET_PORT_AGE_INFO(priv);
8046 rte_spinlock_lock(&age_info->aged_sl);
8047 if (__atomic_compare_exchange_n(&age_param->state, &expected,
8050 __ATOMIC_RELAXED)) {
8051 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
8052 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8054 rte_spinlock_unlock(&age_info->aged_sl);
8056 mlx5_age_event_prepare(sh);
8060 * Handler for the HW respond about ready values from an asynchronous batch
8061 * query. This function is probably called by the host thread.
8064 * The pointer to the shared device context.
8065 * @param[in] async_id
8066 * The Devx async ID.
8068 * The status of the completion.
8071 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8072 uint64_t async_id, int status)
8074 struct mlx5_flow_counter_pool *pool =
8075 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8076 struct mlx5_counter_stats_raw *raw_to_free;
8077 uint8_t query_gen = pool->query_gen ^ 1;
8078 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8079 enum mlx5_counter_type cnt_type =
8080 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8081 MLX5_COUNTER_TYPE_ORIGIN;
8083 if (unlikely(status)) {
8084 raw_to_free = pool->raw_hw;
8086 raw_to_free = pool->raw;
8088 mlx5_flow_aging_check(sh, pool);
8089 rte_spinlock_lock(&pool->sl);
8090 pool->raw = pool->raw_hw;
8091 rte_spinlock_unlock(&pool->sl);
8092 /* Be sure the new raw counters data is updated in memory. */
8094 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8095 rte_spinlock_lock(&cmng->csl[cnt_type]);
8096 TAILQ_CONCAT(&cmng->counters[cnt_type],
8097 &pool->counters[query_gen], next);
8098 rte_spinlock_unlock(&cmng->csl[cnt_type]);
8101 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8102 pool->raw_hw = NULL;
8103 sh->cmng.pending_queries--;
8107 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8108 const struct flow_grp_info *grp_info,
8109 struct rte_flow_error *error)
8111 if (grp_info->transfer && grp_info->external &&
8112 grp_info->fdb_def_rule) {
8113 if (group == UINT32_MAX)
8114 return rte_flow_error_set
8116 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8118 "group index not supported");
8123 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8128 * Translate the rte_flow group index to HW table value.
8130 * If tunnel offload is disabled, all group ids converted to flow table
8131 * id using the standard method.
8132 * If tunnel offload is enabled, group id can be converted using the
8133 * standard or tunnel conversion method. Group conversion method
8134 * selection depends on flags in `grp_info` parameter:
8135 * - Internal (grp_info.external == 0) groups conversion uses the
8137 * - Group ids in JUMP action converted with the tunnel conversion.
8138 * - Group id in rule attribute conversion depends on a rule type and
8140 * ** non zero group attributes converted with the tunnel method
8141 * ** zero group attribute in non-tunnel rule is converted using the
8142 * standard method - there's only one root table
8143 * ** zero group attribute in steer tunnel rule is converted with the
8144 * standard method - single root table
8145 * ** zero group attribute in match tunnel rule is a special OvS
8146 * case: that value is used for portability reasons. That group
8147 * id is converted with the tunnel conversion method.
8152 * PMD tunnel offload object
8154 * rte_flow group index value.
8157 * @param[in] grp_info
8158 * flags used for conversion
8160 * Pointer to error structure.
8163 * 0 on success, a negative errno value otherwise and rte_errno is set.
8166 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8167 const struct mlx5_flow_tunnel *tunnel,
8168 uint32_t group, uint32_t *table,
8169 const struct flow_grp_info *grp_info,
8170 struct rte_flow_error *error)
8173 bool standard_translation;
8175 if (!grp_info->skip_scale && grp_info->external &&
8176 group < MLX5_MAX_TABLES_EXTERNAL)
8177 group *= MLX5_FLOW_TABLE_FACTOR;
8178 if (is_tunnel_offload_active(dev)) {
8179 standard_translation = !grp_info->external ||
8180 grp_info->std_tbl_fix;
8182 standard_translation = true;
8185 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8186 dev->data->port_id, group, grp_info->transfer,
8187 grp_info->external, grp_info->fdb_def_rule,
8188 standard_translation ? "STANDARD" : "TUNNEL");
8189 if (standard_translation)
8190 ret = flow_group_to_table(dev->data->port_id, group, table,
8193 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8200 * Discover availability of metadata reg_c's.
8202 * Iteratively use test flows to check availability.
8205 * Pointer to the Ethernet device structure.
8208 * 0 on success, a negative errno value otherwise and rte_errno is set.
8211 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8213 struct mlx5_priv *priv = dev->data->dev_private;
8214 enum modify_reg idx;
8217 /* reg_c[0] and reg_c[1] are reserved. */
8218 priv->sh->flow_mreg_c[n++] = REG_C_0;
8219 priv->sh->flow_mreg_c[n++] = REG_C_1;
8220 /* Discover availability of other reg_c's. */
8221 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8222 struct rte_flow_attr attr = {
8223 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8224 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8227 struct rte_flow_item items[] = {
8229 .type = RTE_FLOW_ITEM_TYPE_END,
8232 struct rte_flow_action actions[] = {
8234 .type = (enum rte_flow_action_type)
8235 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8236 .conf = &(struct mlx5_flow_action_copy_mreg){
8242 .type = RTE_FLOW_ACTION_TYPE_JUMP,
8243 .conf = &(struct rte_flow_action_jump){
8244 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8248 .type = RTE_FLOW_ACTION_TYPE_END,
8252 struct rte_flow *flow;
8253 struct rte_flow_error error;
8255 if (!priv->sh->config.dv_flow_en)
8257 /* Create internal flow, validation skips copy action. */
8258 flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8259 items, actions, false, &error);
8260 flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8264 priv->sh->flow_mreg_c[n++] = idx;
8265 flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8267 for (; n < MLX5_MREG_C_NUM; ++n)
8268 priv->sh->flow_mreg_c[n] = REG_NON;
8269 priv->sh->metadata_regc_check_flag = 1;
8274 save_dump_file(const uint8_t *data, uint32_t size,
8275 uint32_t type, uint64_t id, void *arg, FILE *file)
8277 char line[BUF_SIZE];
8280 uint32_t actions_num;
8281 struct rte_flow_query_count *count;
8283 memset(line, 0, BUF_SIZE);
8285 case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8286 actions_num = *(uint32_t *)(arg);
8287 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8288 type, id, actions_num);
8290 case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8291 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8294 case DR_DUMP_REC_TYPE_PMD_COUNTER:
8295 count = (struct rte_flow_query_count *)arg;
8297 "%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8298 type, id, count->hits, count->bytes);
8304 for (k = 0; k < size; k++) {
8305 /* Make sure we do not overrun the line buffer length. */
8306 if (out >= BUF_SIZE - 4) {
8310 out += snprintf(line + out, BUF_SIZE - out, "%02x",
8313 fprintf(file, "%s\n", line);
8318 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8319 struct rte_flow_query_count *count, struct rte_flow_error *error)
8321 struct rte_flow_action action[2];
8322 enum mlx5_flow_drv_type ftype;
8323 const struct mlx5_flow_driver_ops *fops;
8326 return rte_flow_error_set(error, ENOENT,
8327 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8329 "invalid flow handle");
8331 action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8332 action[1].type = RTE_FLOW_ACTION_TYPE_END;
8333 if (flow->counter) {
8334 memset(count, 0, sizeof(struct rte_flow_query_count));
8335 ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8336 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8337 ftype < MLX5_FLOW_TYPE_MAX);
8338 fops = flow_get_drv_ops(ftype);
8339 return fops->query(dev, flow, action, count, error);
8344 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8346 * Dump flow ipool data to file
8349 * The pointer to Ethernet device.
8351 * A pointer to a file for output.
8353 * Perform verbose error reporting if not NULL. PMDs initialize this
8354 * structure in case of error only.
8356 * 0 on success, a negative value otherwise.
8359 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8360 struct rte_flow *flow, FILE *file,
8361 struct rte_flow_error *error)
8363 struct mlx5_priv *priv = dev->data->dev_private;
8364 struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
8365 struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8366 uint32_t handle_idx;
8367 struct mlx5_flow_handle *dh;
8368 struct rte_flow_query_count count;
8369 uint32_t actions_num;
8370 const uint8_t *data;
8374 void *action = NULL;
8377 return rte_flow_error_set(error, ENOENT,
8378 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8380 "invalid flow handle");
8382 handle_idx = flow->dev_handles;
8384 if (flow->counter &&
8385 (!mlx5_counter_query(dev, flow->counter, false,
8386 &count.hits, &count.bytes, &action)) && action) {
8387 id = (uint64_t)(uintptr_t)action;
8388 type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8389 save_dump_file(NULL, 0, type,
8390 id, (void *)&count, file);
8393 while (handle_idx) {
8394 dh = mlx5_ipool_get(priv->sh->ipool
8395 [MLX5_IPOOL_MLX5_FLOW], handle_idx);
8398 handle_idx = dh->next.next;
8400 /* Get modify_hdr and encap_decap buf from ipools. */
8402 modify_hdr = dh->dvh.modify_hdr;
8404 if (dh->dvh.rix_encap_decap) {
8405 encap_decap = mlx5_ipool_get(priv->sh->ipool
8406 [MLX5_IPOOL_DECAP_ENCAP],
8407 dh->dvh.rix_encap_decap);
8410 data = (const uint8_t *)modify_hdr->actions;
8411 size = (size_t)(modify_hdr->actions_num) * 8;
8412 id = (uint64_t)(uintptr_t)modify_hdr->action;
8413 actions_num = modify_hdr->actions_num;
8414 type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8415 save_dump_file(data, size, type, id,
8416 (void *)(&actions_num), file);
8419 data = encap_decap->buf;
8420 size = encap_decap->size;
8421 id = (uint64_t)(uintptr_t)encap_decap->action;
8422 type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8423 save_dump_file(data, size, type,
8431 * Dump all flow's encap_decap/modify_hdr/counter data to file
8434 * The pointer to Ethernet device.
8436 * A pointer to a file for output.
8438 * Perform verbose error reporting if not NULL. PMDs initialize this
8439 * structure in case of error only.
8441 * 0 on success, a negative value otherwise.
8444 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8445 FILE *file, struct rte_flow_error *error __rte_unused)
8447 struct mlx5_priv *priv = dev->data->dev_private;
8448 struct mlx5_dev_ctx_shared *sh = priv->sh;
8449 struct mlx5_hlist *h;
8450 struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
8451 struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8452 struct rte_flow_query_count count;
8453 uint32_t actions_num;
8454 const uint8_t *data;
8460 struct mlx5_list_inconst *l_inconst;
8461 struct mlx5_list_entry *e;
8463 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8467 /* encap_decap hlist is lcore_share, get global core cache. */
8468 i = MLX5_LIST_GLOBAL;
8469 h = sh->encaps_decaps;
8471 for (j = 0; j <= h->mask; j++) {
8472 l_inconst = &h->buckets[j].l;
8473 if (!l_inconst || !l_inconst->cache[i])
8476 e = LIST_FIRST(&l_inconst->cache[i]->h);
8479 (struct mlx5_flow_dv_encap_decap_resource *)e;
8480 data = encap_decap->buf;
8481 size = encap_decap->size;
8482 id = (uint64_t)(uintptr_t)encap_decap->action;
8483 type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8484 save_dump_file(data, size, type,
8486 e = LIST_NEXT(e, next);
8491 /* get modify_hdr */
8492 h = sh->modify_cmds;
8494 lcore_index = rte_lcore_index(rte_lcore_id());
8495 if (unlikely(lcore_index == -1)) {
8496 lcore_index = MLX5_LIST_NLCORE;
8497 rte_spinlock_lock(&h->l_const.lcore_lock);
8501 for (j = 0; j <= h->mask; j++) {
8502 l_inconst = &h->buckets[j].l;
8503 if (!l_inconst || !l_inconst->cache[i])
8506 e = LIST_FIRST(&l_inconst->cache[i]->h);
8509 (struct mlx5_flow_dv_modify_hdr_resource *)e;
8510 data = (const uint8_t *)modify_hdr->actions;
8511 size = (size_t)(modify_hdr->actions_num) * 8;
8512 actions_num = modify_hdr->actions_num;
8513 id = (uint64_t)(uintptr_t)modify_hdr->action;
8514 type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8515 save_dump_file(data, size, type, id,
8516 (void *)(&actions_num), file);
8517 e = LIST_NEXT(e, next);
8521 if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8522 rte_spinlock_unlock(&h->l_const.lcore_lock);
8526 MLX5_ASSERT(cmng->n_valid <= cmng->n);
8527 max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8528 for (j = 1; j <= max; j++) {
8530 if ((!mlx5_counter_query(dev, j, false, &count.hits,
8531 &count.bytes, &action)) && action) {
8532 id = (uint64_t)(uintptr_t)action;
8533 type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8534 save_dump_file(NULL, 0, type,
8535 id, (void *)&count, file);
8543 * Dump flow raw hw data to file
8546 * The pointer to Ethernet device.
8548 * A pointer to a file for output.
8550 * Perform verbose error reporting if not NULL. PMDs initialize this
8551 * structure in case of error only.
8553 * 0 on success, a negative value otherwise.
8556 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8558 struct rte_flow_error *error __rte_unused)
8560 struct mlx5_priv *priv = dev->data->dev_private;
8561 struct mlx5_dev_ctx_shared *sh = priv->sh;
8562 uint32_t handle_idx;
8564 struct mlx5_flow_handle *dh;
8565 struct rte_flow *flow;
8567 if (!sh->config.dv_flow_en) {
8568 if (fputs("device dv flow disabled\n", file) <= 0)
8575 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8576 if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8579 return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8581 sh->tx_domain, file);
8584 flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8585 (uintptr_t)(void *)flow_idx);
8589 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8590 mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8592 handle_idx = flow->dev_handles;
8593 while (handle_idx) {
8594 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8599 ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8604 handle_idx = dh->next.next;
8610 * Get aged-out flows.
8613 * Pointer to the Ethernet device structure.
8614 * @param[in] context
8615 * The address of an array of pointers to the aged-out flows contexts.
8616 * @param[in] nb_countexts
8617 * The length of context array pointers.
8619 * Perform verbose error reporting if not NULL. Initialized in case of
8623 * how many contexts get in success, otherwise negative errno value.
8624 * if nb_contexts is 0, return the amount of all aged contexts.
8625 * if nb_contexts is not 0 , return the amount of aged flows reported
8626 * in the context array.
8629 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8630 uint32_t nb_contexts, struct rte_flow_error *error)
8632 const struct mlx5_flow_driver_ops *fops;
8633 struct rte_flow_attr attr = { .transfer = 0 };
8635 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8636 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8637 return fops->get_aged_flows(dev, contexts, nb_contexts,
8641 "port %u get aged flows is not supported.",
8642 dev->data->port_id);
8646 /* Wrapper for driver action_validate op callback */
8648 flow_drv_action_validate(struct rte_eth_dev *dev,
8649 const struct rte_flow_indir_action_conf *conf,
8650 const struct rte_flow_action *action,
8651 const struct mlx5_flow_driver_ops *fops,
8652 struct rte_flow_error *error)
8654 static const char err_msg[] = "indirect action validation unsupported";
8656 if (!fops->action_validate) {
8657 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8658 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8662 return fops->action_validate(dev, conf, action, error);
8666 * Destroys the shared action by handle.
8669 * Pointer to Ethernet device structure.
8671 * Handle for the indirect action object to be destroyed.
8673 * Perform verbose error reporting if not NULL. PMDs initialize this
8674 * structure in case of error only.
8677 * 0 on success, a negative errno value otherwise and rte_errno is set.
8679 * @note: wrapper for driver action_create op callback.
8682 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8683 struct rte_flow_action_handle *handle,
8684 struct rte_flow_error *error)
8686 static const char err_msg[] = "indirect action destruction unsupported";
8687 struct rte_flow_attr attr = { .transfer = 0 };
8688 const struct mlx5_flow_driver_ops *fops =
8689 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8691 if (!fops->action_destroy) {
8692 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8693 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8697 return fops->action_destroy(dev, handle, error);
8700 /* Wrapper for driver action_destroy op callback */
8702 flow_drv_action_update(struct rte_eth_dev *dev,
8703 struct rte_flow_action_handle *handle,
8705 const struct mlx5_flow_driver_ops *fops,
8706 struct rte_flow_error *error)
8708 static const char err_msg[] = "indirect action update unsupported";
8710 if (!fops->action_update) {
8711 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8712 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8716 return fops->action_update(dev, handle, update, error);
8719 /* Wrapper for driver action_destroy op callback */
8721 flow_drv_action_query(struct rte_eth_dev *dev,
8722 const struct rte_flow_action_handle *handle,
8724 const struct mlx5_flow_driver_ops *fops,
8725 struct rte_flow_error *error)
8727 static const char err_msg[] = "indirect action query unsupported";
8729 if (!fops->action_query) {
8730 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8731 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8735 return fops->action_query(dev, handle, data, error);
8739 * Create indirect action for reuse in multiple flow rules.
8742 * Pointer to Ethernet device structure.
8744 * Pointer to indirect action object configuration.
8746 * Action configuration for indirect action object creation.
8748 * Perform verbose error reporting if not NULL. PMDs initialize this
8749 * structure in case of error only.
8751 * A valid handle in case of success, NULL otherwise and rte_errno is set.
8753 static struct rte_flow_action_handle *
8754 mlx5_action_handle_create(struct rte_eth_dev *dev,
8755 const struct rte_flow_indir_action_conf *conf,
8756 const struct rte_flow_action *action,
8757 struct rte_flow_error *error)
8759 static const char err_msg[] = "indirect action creation unsupported";
8760 struct rte_flow_attr attr = { .transfer = 0 };
8761 const struct mlx5_flow_driver_ops *fops =
8762 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8764 if (flow_drv_action_validate(dev, conf, action, fops, error))
8766 if (!fops->action_create) {
8767 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8768 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8772 return fops->action_create(dev, conf, action, error);
8776 * Updates inplace the indirect action configuration pointed by *handle*
8777 * with the configuration provided as *update* argument.
8778 * The update of the indirect action configuration effects all flow rules
8779 * reusing the action via handle.
8782 * Pointer to Ethernet device structure.
8784 * Handle for the indirect action to be updated.
8786 * Action specification used to modify the action pointed by handle.
8787 * *update* could be of same type with the action pointed by the *handle*
8788 * handle argument, or some other structures like a wrapper, depending on
8789 * the indirect action type.
8791 * Perform verbose error reporting if not NULL. PMDs initialize this
8792 * structure in case of error only.
8795 * 0 on success, a negative errno value otherwise and rte_errno is set.
8798 mlx5_action_handle_update(struct rte_eth_dev *dev,
8799 struct rte_flow_action_handle *handle,
8801 struct rte_flow_error *error)
8803 struct rte_flow_attr attr = { .transfer = 0 };
8804 const struct mlx5_flow_driver_ops *fops =
8805 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8808 ret = flow_drv_action_validate(dev, NULL,
8809 (const struct rte_flow_action *)update, fops, error);
8812 return flow_drv_action_update(dev, handle, update, fops,
8817 * Query the indirect action by handle.
8819 * This function allows retrieving action-specific data such as counters.
8820 * Data is gathered by special action which may be present/referenced in
8821 * more than one flow rule definition.
8823 * see @RTE_FLOW_ACTION_TYPE_COUNT
8826 * Pointer to Ethernet device structure.
8828 * Handle for the indirect action to query.
8829 * @param[in, out] data
8830 * Pointer to storage for the associated query data type.
8832 * Perform verbose error reporting if not NULL. PMDs initialize this
8833 * structure in case of error only.
8836 * 0 on success, a negative errno value otherwise and rte_errno is set.
8839 mlx5_action_handle_query(struct rte_eth_dev *dev,
8840 const struct rte_flow_action_handle *handle,
8842 struct rte_flow_error *error)
8844 struct rte_flow_attr attr = { .transfer = 0 };
8845 const struct mlx5_flow_driver_ops *fops =
8846 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8848 return flow_drv_action_query(dev, handle, data, fops, error);
8852 * Destroy all indirect actions (shared RSS).
8855 * Pointer to Ethernet device.
8858 * 0 on success, a negative errno value otherwise and rte_errno is set.
8861 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8863 struct rte_flow_error error;
8864 struct mlx5_priv *priv = dev->data->dev_private;
8865 struct mlx5_shared_action_rss *shared_rss;
8869 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8870 priv->rss_shared_actions, idx, shared_rss, next) {
8871 ret |= mlx5_action_handle_destroy(dev,
8872 (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8878 * Validate existing indirect actions against current device configuration
8879 * and attach them to device resources.
8882 * Pointer to Ethernet device.
8885 * 0 on success, a negative errno value otherwise and rte_errno is set.
8888 mlx5_action_handle_attach(struct rte_eth_dev *dev)
8890 struct mlx5_priv *priv = dev->data->dev_private;
8891 struct mlx5_indexed_pool *ipool =
8892 priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8893 struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8897 ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8898 struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8899 const char *message;
8902 ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
8904 &message, &queue_idx);
8906 DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
8907 dev->data->port_id, ind_tbl->queues[queue_idx],
8914 ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8915 struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8917 ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
8919 DRV_LOG(ERR, "Port %u could not attach "
8920 "indirection table obj %p",
8921 dev->data->port_id, (void *)ind_tbl);
8927 shared_rss_last = shared_rss;
8928 ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8929 struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8931 if (shared_rss == shared_rss_last)
8933 if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
8934 DRV_LOG(CRIT, "Port %u could not detach "
8935 "indirection table obj %p on rollback",
8936 dev->data->port_id, (void *)ind_tbl);
8942 * Detach indirect actions of the device from its resources.
8945 * Pointer to Ethernet device.
8948 * 0 on success, a negative errno value otherwise and rte_errno is set.
8951 mlx5_action_handle_detach(struct rte_eth_dev *dev)
8953 struct mlx5_priv *priv = dev->data->dev_private;
8954 struct mlx5_indexed_pool *ipool =
8955 priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8956 struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8960 ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8961 struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8963 ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
8965 DRV_LOG(ERR, "Port %u could not detach "
8966 "indirection table obj %p",
8967 dev->data->port_id, (void *)ind_tbl);
8973 shared_rss_last = shared_rss;
8974 ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8975 struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8977 if (shared_rss == shared_rss_last)
8979 if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
8980 DRV_LOG(CRIT, "Port %u could not attach "
8981 "indirection table obj %p on rollback",
8982 dev->data->port_id, (void *)ind_tbl);
8987 #ifndef HAVE_MLX5DV_DR
8988 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8990 #define MLX5_DOMAIN_SYNC_FLOW \
8991 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8994 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8996 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8997 const struct mlx5_flow_driver_ops *fops;
8999 struct rte_flow_attr attr = { .transfer = 0 };
9001 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9002 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
9008 const struct mlx5_flow_tunnel *
9009 mlx5_get_tof(const struct rte_flow_item *item,
9010 const struct rte_flow_action *action,
9011 enum mlx5_tof_rule_type *rule_type)
9013 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9014 if (item->type == (typeof(item->type))
9015 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
9016 *rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
9017 return flow_items_to_tunnel(item);
9020 for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
9021 if (action->type == (typeof(action->type))
9022 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
9023 *rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
9024 return flow_actions_to_tunnel(action);
9031 * tunnel offload functionality is defined for DV environment only
9033 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9035 union tunnel_offload_mark {
9038 uint32_t app_reserve:8;
9039 uint32_t table_id:15;
9040 uint32_t transfer:1;
9041 uint32_t _unused_:8;
9046 mlx5_access_tunnel_offload_db
9047 (struct rte_eth_dev *dev,
9048 bool (*match)(struct rte_eth_dev *,
9049 struct mlx5_flow_tunnel *, const void *),
9050 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9051 void (*miss)(struct rte_eth_dev *, void *),
9052 void *ctx, bool lock_op);
9055 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9056 struct rte_flow *flow,
9057 const struct rte_flow_attr *attr,
9058 const struct rte_flow_action *app_actions,
9060 const struct mlx5_flow_tunnel *tunnel,
9061 struct tunnel_default_miss_ctx *ctx,
9062 struct rte_flow_error *error)
9064 struct mlx5_priv *priv = dev->data->dev_private;
9065 struct mlx5_flow *dev_flow;
9066 struct rte_flow_attr miss_attr = *attr;
9067 const struct rte_flow_item miss_items[2] = {
9069 .type = RTE_FLOW_ITEM_TYPE_ETH,
9075 .type = RTE_FLOW_ITEM_TYPE_END,
9081 union tunnel_offload_mark mark_id;
9082 struct rte_flow_action_mark miss_mark;
9083 struct rte_flow_action miss_actions[3] = {
9084 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9085 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
9087 const struct rte_flow_action_jump *jump_data;
9088 uint32_t i, flow_table = 0; /* prevent compilation warning */
9089 struct flow_grp_info grp_info = {
9091 .transfer = attr->transfer,
9092 .fdb_def_rule = !!priv->fdb_def_rule,
9097 if (!attr->transfer) {
9100 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9101 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9102 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9105 return rte_flow_error_set
9107 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9108 NULL, "invalid default miss RSS");
9109 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9110 ctx->action_rss.level = 0,
9111 ctx->action_rss.types = priv->rss_conf.rss_hf,
9112 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9113 ctx->action_rss.queue_num = priv->reta_idx_n,
9114 ctx->action_rss.key = priv->rss_conf.rss_key,
9115 ctx->action_rss.queue = ctx->queue;
9116 if (!priv->reta_idx_n || !priv->rxqs_n)
9117 return rte_flow_error_set
9119 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9120 NULL, "invalid port configuration");
9121 if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9122 ctx->action_rss.types = 0;
9123 for (i = 0; i != priv->reta_idx_n; ++i)
9124 ctx->queue[i] = (*priv->reta_idx)[i];
9126 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9127 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9129 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
9130 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
9131 jump_data = app_actions->conf;
9132 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
9133 miss_attr.group = jump_data->group;
9134 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
9135 &flow_table, &grp_info, error);
9137 return rte_flow_error_set(error, EINVAL,
9138 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9139 NULL, "invalid tunnel id");
9140 mark_id.app_reserve = 0;
9141 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
9142 mark_id.transfer = !!attr->transfer;
9143 mark_id._unused_ = 0;
9144 miss_mark.id = mark_id.val;
9145 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
9146 miss_items, miss_actions, flow_idx, error);
9149 dev_flow->flow = flow;
9150 dev_flow->external = true;
9151 dev_flow->tunnel = tunnel;
9152 dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
9153 /* Subflow object was created, we must include one in the list. */
9154 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
9155 dev_flow->handle, next);
9157 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
9158 dev->data->port_id, tunnel->app_tunnel.type,
9159 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
9160 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
9161 miss_actions, error);
9163 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
9169 static const struct mlx5_flow_tbl_data_entry *
9170 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
9172 struct mlx5_priv *priv = dev->data->dev_private;
9173 struct mlx5_dev_ctx_shared *sh = priv->sh;
9174 struct mlx5_list_entry *he;
9175 union tunnel_offload_mark mbits = { .val = mark };
9176 union mlx5_flow_tbl_key table_key = {
9178 .level = tunnel_id_to_flow_tbl(mbits.table_id),
9182 .is_fdb = !!mbits.transfer,
9186 struct mlx5_flow_cb_ctx ctx = {
9187 .data = &table_key.v64,
9190 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
9192 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
9196 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
9197 struct mlx5_list_entry *entry)
9199 struct mlx5_dev_ctx_shared *sh = tool_ctx;
9200 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9202 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9203 tunnel_flow_tbl_to_id(tte->flow_table));
9208 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9209 struct mlx5_list_entry *entry, void *cb_ctx)
9211 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9212 union tunnel_tbl_key tbl = {
9213 .val = *(uint64_t *)(ctx->data),
9215 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9217 return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9220 static struct mlx5_list_entry *
9221 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9223 struct mlx5_dev_ctx_shared *sh = tool_ctx;
9224 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9225 struct tunnel_tbl_entry *tte;
9226 union tunnel_tbl_key tbl = {
9227 .val = *(uint64_t *)(ctx->data),
9230 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9235 mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9237 if (tte->flow_table >= MLX5_MAX_TABLES) {
9238 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9240 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9243 } else if (!tte->flow_table) {
9246 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9247 tte->tunnel_id = tbl.tunnel_id;
9248 tte->group = tbl.group;
9256 static struct mlx5_list_entry *
9257 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9258 struct mlx5_list_entry *oentry,
9259 void *cb_ctx __rte_unused)
9261 struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9266 memcpy(tte, oentry, sizeof(*tte));
9271 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9272 struct mlx5_list_entry *entry)
9274 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9280 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9281 const struct mlx5_flow_tunnel *tunnel,
9282 uint32_t group, uint32_t *table,
9283 struct rte_flow_error *error)
9285 struct mlx5_list_entry *he;
9286 struct tunnel_tbl_entry *tte;
9287 union tunnel_tbl_key key = {
9288 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9291 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9292 struct mlx5_hlist *group_hash;
9293 struct mlx5_flow_cb_ctx ctx = {
9297 group_hash = tunnel ? tunnel->groups : thub->groups;
9298 he = mlx5_hlist_register(group_hash, key.val, &ctx);
9300 return rte_flow_error_set(error, EINVAL,
9301 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9303 "tunnel group index not supported");
9304 tte = container_of(he, typeof(*tte), hash);
9305 *table = tte->flow_table;
9306 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9307 dev->data->port_id, key.tunnel_id, group, *table);
9312 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9313 struct mlx5_flow_tunnel *tunnel)
9315 struct mlx5_priv *priv = dev->data->dev_private;
9316 struct mlx5_indexed_pool *ipool;
9318 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9319 dev->data->port_id, tunnel->tunnel_id);
9320 LIST_REMOVE(tunnel, chain);
9321 mlx5_hlist_destroy(tunnel->groups);
9322 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9323 mlx5_ipool_free(ipool, tunnel->tunnel_id);
9327 mlx5_access_tunnel_offload_db
9328 (struct rte_eth_dev *dev,
9329 bool (*match)(struct rte_eth_dev *,
9330 struct mlx5_flow_tunnel *, const void *),
9331 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9332 void (*miss)(struct rte_eth_dev *, void *),
9333 void *ctx, bool lock_op)
9335 bool verdict = false;
9336 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9337 struct mlx5_flow_tunnel *tunnel;
9339 rte_spinlock_lock(&thub->sl);
9340 LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9341 verdict = match(dev, tunnel, (const void *)ctx);
9346 rte_spinlock_unlock(&thub->sl);
9348 hit(dev, tunnel, ctx);
9349 if (!verdict && miss)
9352 rte_spinlock_unlock(&thub->sl);
9357 struct tunnel_db_find_tunnel_id_ctx {
9359 struct mlx5_flow_tunnel *tunnel;
9363 find_tunnel_id_match(struct rte_eth_dev *dev,
9364 struct mlx5_flow_tunnel *tunnel, const void *x)
9366 const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9369 return tunnel->tunnel_id == ctx->tunnel_id;
9373 find_tunnel_id_hit(struct rte_eth_dev *dev,
9374 struct mlx5_flow_tunnel *tunnel, void *x)
9376 struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9378 ctx->tunnel = tunnel;
9381 static struct mlx5_flow_tunnel *
9382 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9384 struct tunnel_db_find_tunnel_id_ctx ctx = {
9388 mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9389 find_tunnel_id_hit, NULL, &ctx, true);
9394 static struct mlx5_flow_tunnel *
9395 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9396 const struct rte_flow_tunnel *app_tunnel)
9398 struct mlx5_priv *priv = dev->data->dev_private;
9399 struct mlx5_indexed_pool *ipool;
9400 struct mlx5_flow_tunnel *tunnel;
9403 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9404 tunnel = mlx5_ipool_zmalloc(ipool, &id);
9407 if (id >= MLX5_MAX_TUNNELS) {
9408 mlx5_ipool_free(ipool, id);
9409 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9412 tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9414 mlx5_flow_tunnel_grp2tbl_create_cb,
9415 mlx5_flow_tunnel_grp2tbl_match_cb,
9416 mlx5_flow_tunnel_grp2tbl_remove_cb,
9417 mlx5_flow_tunnel_grp2tbl_clone_cb,
9418 mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9419 if (!tunnel->groups) {
9420 mlx5_ipool_free(ipool, id);
9423 /* initiate new PMD tunnel */
9424 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9425 tunnel->tunnel_id = id;
9426 tunnel->action.type = (typeof(tunnel->action.type))
9427 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9428 tunnel->action.conf = tunnel;
9429 tunnel->item.type = (typeof(tunnel->item.type))
9430 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9431 tunnel->item.spec = tunnel;
9432 tunnel->item.last = NULL;
9433 tunnel->item.mask = NULL;
9435 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9436 dev->data->port_id, tunnel->tunnel_id);
9441 struct tunnel_db_get_tunnel_ctx {
9442 const struct rte_flow_tunnel *app_tunnel;
9443 struct mlx5_flow_tunnel *tunnel;
9446 static bool get_tunnel_match(struct rte_eth_dev *dev,
9447 struct mlx5_flow_tunnel *tunnel, const void *x)
9449 const struct tunnel_db_get_tunnel_ctx *ctx = x;
9452 return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9453 sizeof(*ctx->app_tunnel));
9456 static void get_tunnel_hit(struct rte_eth_dev *dev,
9457 struct mlx5_flow_tunnel *tunnel, void *x)
9459 /* called under tunnel spinlock protection */
9460 struct tunnel_db_get_tunnel_ctx *ctx = x;
9464 ctx->tunnel = tunnel;
9467 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9469 /* called under tunnel spinlock protection */
9470 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9471 struct tunnel_db_get_tunnel_ctx *ctx = x;
9473 rte_spinlock_unlock(&thub->sl);
9474 ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9475 rte_spinlock_lock(&thub->sl);
9477 ctx->tunnel->refctn = 1;
9478 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9484 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9485 const struct rte_flow_tunnel *app_tunnel,
9486 struct mlx5_flow_tunnel **tunnel)
9488 struct tunnel_db_get_tunnel_ctx ctx = {
9489 .app_tunnel = app_tunnel,
9492 mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9493 get_tunnel_miss, &ctx, true);
9494 *tunnel = ctx.tunnel;
9495 return ctx.tunnel ? 0 : -ENOMEM;
9498 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9500 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9504 if (!LIST_EMPTY(&thub->tunnels))
9505 DRV_LOG(WARNING, "port %u tunnels present", port_id);
9506 mlx5_hlist_destroy(thub->groups);
9510 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9513 struct mlx5_flow_tunnel_hub *thub;
9515 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9519 LIST_INIT(&thub->tunnels);
9520 rte_spinlock_init(&thub->sl);
9521 thub->groups = mlx5_hlist_create("flow groups", 64,
9523 mlx5_flow_tunnel_grp2tbl_create_cb,
9524 mlx5_flow_tunnel_grp2tbl_match_cb,
9525 mlx5_flow_tunnel_grp2tbl_remove_cb,
9526 mlx5_flow_tunnel_grp2tbl_clone_cb,
9527 mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9528 if (!thub->groups) {
9532 sh->tunnel_hub = thub;
9538 mlx5_hlist_destroy(thub->groups);
9545 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9546 struct rte_flow_tunnel *tunnel,
9547 struct rte_flow_error *error)
9549 struct mlx5_priv *priv = dev->data->dev_private;
9551 if (!priv->sh->config.dv_flow_en)
9552 return rte_flow_error_set(error, ENOTSUP,
9553 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9554 "flow DV interface is off");
9555 if (!is_tunnel_offload_active(dev))
9556 return rte_flow_error_set(error, ENOTSUP,
9557 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9558 "tunnel offload was not activated");
9560 return rte_flow_error_set(error, EINVAL,
9561 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9562 "no application tunnel");
9563 switch (tunnel->type) {
9565 return rte_flow_error_set(error, EINVAL,
9566 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9567 "unsupported tunnel type");
9568 case RTE_FLOW_ITEM_TYPE_VXLAN:
9569 case RTE_FLOW_ITEM_TYPE_GRE:
9570 case RTE_FLOW_ITEM_TYPE_NVGRE:
9571 case RTE_FLOW_ITEM_TYPE_GENEVE:
9578 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9579 struct rte_flow_tunnel *app_tunnel,
9580 struct rte_flow_action **actions,
9581 uint32_t *num_of_actions,
9582 struct rte_flow_error *error)
9584 struct mlx5_flow_tunnel *tunnel;
9585 int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9589 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9591 return rte_flow_error_set(error, ret,
9592 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9593 "failed to initialize pmd tunnel");
9595 *actions = &tunnel->action;
9596 *num_of_actions = 1;
9601 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9602 struct rte_flow_tunnel *app_tunnel,
9603 struct rte_flow_item **items,
9604 uint32_t *num_of_items,
9605 struct rte_flow_error *error)
9607 struct mlx5_flow_tunnel *tunnel;
9608 int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9612 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9614 return rte_flow_error_set(error, ret,
9615 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9616 "failed to initialize pmd tunnel");
9618 *items = &tunnel->item;
9623 struct tunnel_db_element_release_ctx {
9624 struct rte_flow_item *items;
9625 struct rte_flow_action *actions;
9626 uint32_t num_elements;
9627 struct rte_flow_error *error;
9632 tunnel_element_release_match(struct rte_eth_dev *dev,
9633 struct mlx5_flow_tunnel *tunnel, const void *x)
9635 const struct tunnel_db_element_release_ctx *ctx = x;
9638 if (ctx->num_elements != 1)
9640 else if (ctx->items)
9641 return ctx->items == &tunnel->item;
9642 else if (ctx->actions)
9643 return ctx->actions == &tunnel->action;
9649 tunnel_element_release_hit(struct rte_eth_dev *dev,
9650 struct mlx5_flow_tunnel *tunnel, void *x)
9652 struct tunnel_db_element_release_ctx *ctx = x;
9654 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9655 mlx5_flow_tunnel_free(dev, tunnel);
9659 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9661 struct tunnel_db_element_release_ctx *ctx = x;
9663 ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9664 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9665 "invalid argument");
9669 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9670 struct rte_flow_item *pmd_items,
9671 uint32_t num_items, struct rte_flow_error *err)
9673 struct tunnel_db_element_release_ctx ctx = {
9676 .num_elements = num_items,
9680 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9681 tunnel_element_release_hit,
9682 tunnel_element_release_miss, &ctx, false);
9688 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9689 struct rte_flow_action *pmd_actions,
9690 uint32_t num_actions, struct rte_flow_error *err)
9692 struct tunnel_db_element_release_ctx ctx = {
9694 .actions = pmd_actions,
9695 .num_elements = num_actions,
9699 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9700 tunnel_element_release_hit,
9701 tunnel_element_release_miss, &ctx, false);
9707 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9709 struct rte_flow_restore_info *info,
9710 struct rte_flow_error *err)
9712 uint64_t ol_flags = m->ol_flags;
9713 const struct mlx5_flow_tbl_data_entry *tble;
9714 const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9716 if (!is_tunnel_offload_active(dev)) {
9721 if ((ol_flags & mask) != mask)
9723 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9725 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9726 dev->data->port_id, m->hash.fdir.hi);
9729 MLX5_ASSERT(tble->tunnel);
9730 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9731 info->group_id = tble->group_id;
9732 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9733 RTE_FLOW_RESTORE_INFO_GROUP_ID |
9734 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9739 return rte_flow_error_set(err, EINVAL,
9740 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9741 "failed to get restore info");
9744 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9746 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9747 __rte_unused struct rte_flow_tunnel *app_tunnel,
9748 __rte_unused struct rte_flow_action **actions,
9749 __rte_unused uint32_t *num_of_actions,
9750 __rte_unused struct rte_flow_error *error)
9756 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9757 __rte_unused struct rte_flow_tunnel *app_tunnel,
9758 __rte_unused struct rte_flow_item **items,
9759 __rte_unused uint32_t *num_of_items,
9760 __rte_unused struct rte_flow_error *error)
9766 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9767 __rte_unused struct rte_flow_item *pmd_items,
9768 __rte_unused uint32_t num_items,
9769 __rte_unused struct rte_flow_error *err)
9775 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9776 __rte_unused struct rte_flow_action *pmd_action,
9777 __rte_unused uint32_t num_actions,
9778 __rte_unused struct rte_flow_error *err)
9784 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9785 __rte_unused struct rte_mbuf *m,
9786 __rte_unused struct rte_flow_restore_info *i,
9787 __rte_unused struct rte_flow_error *err)
9793 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9794 __rte_unused struct rte_flow *flow,
9795 __rte_unused const struct rte_flow_attr *attr,
9796 __rte_unused const struct rte_flow_action *actions,
9797 __rte_unused uint32_t flow_idx,
9798 __rte_unused const struct mlx5_flow_tunnel *tunnel,
9799 __rte_unused struct tunnel_default_miss_ctx *ctx,
9800 __rte_unused struct rte_flow_error *error)
9805 static struct mlx5_flow_tunnel *
9806 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9807 __rte_unused uint32_t id)
9813 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9814 __rte_unused struct mlx5_flow_tunnel *tunnel)
9819 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9820 __rte_unused const struct mlx5_flow_tunnel *t,
9821 __rte_unused uint32_t group,
9822 __rte_unused uint32_t *table,
9823 struct rte_flow_error *error)
9825 return rte_flow_error_set(error, ENOTSUP,
9826 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9827 "tunnel offload requires DV support");
9831 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9832 __rte_unused uint16_t port_id)
9835 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9837 /* Flex flow item API */
9838 static struct rte_flow_item_flex_handle *
9839 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
9840 const struct rte_flow_item_flex_conf *conf,
9841 struct rte_flow_error *error)
9843 static const char err_msg[] = "flex item creation unsupported";
9844 struct rte_flow_attr attr = { .transfer = 0 };
9845 const struct mlx5_flow_driver_ops *fops =
9846 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9848 if (!fops->item_create) {
9849 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9850 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9854 return fops->item_create(dev, conf, error);
9858 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
9859 const struct rte_flow_item_flex_handle *handle,
9860 struct rte_flow_error *error)
9862 static const char err_msg[] = "flex item release unsupported";
9863 struct rte_flow_attr attr = { .transfer = 0 };
9864 const struct mlx5_flow_driver_ops *fops =
9865 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9867 if (!fops->item_release) {
9868 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9869 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9873 return fops->item_release(dev, handle, error);
9877 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9880 struct rte_flow_error error;
9882 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9884 ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9886 (void *)(uintptr_t)item->type, &error);
9888 printf("%s ", item_name);
9890 printf("%d\n", (int)item->type);
9896 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9898 const struct rte_flow_item_udp *spec = udp_item->spec;
9899 const struct rte_flow_item_udp *mask = udp_item->mask;
9900 uint16_t udp_dport = 0;
9904 mask = &rte_flow_item_udp_mask;
9905 udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9906 mask->hdr.dst_port);
9908 return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9911 static const struct mlx5_flow_expand_node *
9912 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9913 unsigned int item_idx,
9914 const struct mlx5_flow_expand_node graph[],
9915 const struct mlx5_flow_expand_node *node)
9917 const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9919 if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
9921 node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
9923 * The expansion node is VXLAN and it is also the last
9924 * expandable item in the pattern, so need to continue
9925 * expansion of the inner tunnel.
9927 MLX5_ASSERT(item_idx > 0);
9928 prev_item = pattern + item_idx - 1;
9929 MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9930 if (mlx5_flow_is_std_vxlan_port(prev_item))
9931 return &graph[MLX5_EXPANSION_STD_VXLAN];
9932 return &graph[MLX5_EXPANSION_L3_VXLAN];
9937 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
9938 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
9939 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
9942 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
9943 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
9944 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
9945 { 9, 10, 11 }, { 12, 13, 14 },
9949 * Discover the number of available flow priorities.
9955 * On success, number of available flow priorities.
9956 * On failure, a negative errno-style code and rte_errno is set.
9959 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
9961 static const uint16_t vprio[] = {8, 16};
9962 const struct mlx5_priv *priv = dev->data->dev_private;
9963 const struct mlx5_flow_driver_ops *fops;
9964 enum mlx5_flow_drv_type type;
9967 type = mlx5_flow_os_get_type();
9968 if (type == MLX5_FLOW_TYPE_MAX) {
9969 type = MLX5_FLOW_TYPE_VERBS;
9970 if (priv->sh->cdev->config.devx && priv->sh->config.dv_flow_en)
9971 type = MLX5_FLOW_TYPE_DV;
9973 fops = flow_get_drv_ops(type);
9974 if (fops->discover_priorities == NULL) {
9975 DRV_LOG(ERR, "Priority discovery not supported");
9976 rte_errno = ENOTSUP;
9979 ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
9984 ret = RTE_DIM(priority_map_3);
9987 ret = RTE_DIM(priority_map_5);
9990 rte_errno = ENOTSUP;
9992 "port %u maximum priority: %d expected 8/16",
9993 dev->data->port_id, ret);
9996 DRV_LOG(INFO, "port %u supported flow priorities:"
9997 " 0-%d for ingress or egress root table,"
9998 " 0-%d for non-root table or transfer root table.",
9999 dev->data->port_id, ret - 2,
10000 MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
10005 * Adjust flow priority based on the highest layer and the request priority.
10008 * Pointer to the Ethernet device structure.
10009 * @param[in] priority
10010 * The rule base priority.
10011 * @param[in] subpriority
10012 * The priority based on the items.
10015 * The new priority.
10018 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
10019 uint32_t subpriority)
10022 struct mlx5_priv *priv = dev->data->dev_private;
10024 switch (priv->sh->flow_max_priority) {
10025 case RTE_DIM(priority_map_3):
10026 res = priority_map_3[priority][subpriority];
10028 case RTE_DIM(priority_map_5):
10029 res = priority_map_5[priority][subpriority];