1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
10 #include <sys/queue.h>
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
25 #include <mlx5_malloc.h>
27 #include "mlx5_defs.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
36 struct tunnel_default_miss_ctx {
40 struct rte_flow_action_rss action_rss;
41 struct rte_flow_action_queue miss_queue;
42 struct rte_flow_action_jump miss_jump;
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 struct rte_flow *flow,
50 const struct rte_flow_attr *attr,
51 const struct rte_flow_action *app_actions,
53 const struct mlx5_flow_tunnel *tunnel,
54 struct tunnel_default_miss_ctx *ctx,
55 struct rte_flow_error *error);
56 static struct mlx5_flow_tunnel *
57 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
59 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
61 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
62 const struct mlx5_flow_tunnel *tunnel,
63 uint32_t group, uint32_t *table,
64 struct rte_flow_error *error);
66 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
67 static void mlx5_flow_pop_thread_workspace(void);
70 /** Device flow drivers. */
71 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
73 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
75 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
76 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
77 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
78 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
80 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
81 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
84 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
85 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
90 /** Node object of input graph for mlx5_flow_expand_rss(). */
91 struct mlx5_flow_expand_node {
92 const int *const next;
94 * List of next node indexes. Index 0 is interpreted as a terminator.
96 const enum rte_flow_item_type type;
97 /**< Pattern item type of current node. */
100 * RSS types bit-field associated with this node
101 * (see ETH_RSS_* definitions).
104 /**< optional expand field. Default 0 to expand, 1 not go deeper. */
107 /** Object returned by mlx5_flow_expand_rss(). */
108 struct mlx5_flow_expand_rss {
110 /**< Number of entries @p patterns and @p priorities. */
112 struct rte_flow_item *pattern; /**< Expanded pattern array. */
113 uint32_t priority; /**< Priority offset for each expansion. */
118 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
121 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
123 switch (item->type) {
124 case RTE_FLOW_ITEM_TYPE_ETH:
125 case RTE_FLOW_ITEM_TYPE_VLAN:
126 case RTE_FLOW_ITEM_TYPE_IPV4:
127 case RTE_FLOW_ITEM_TYPE_IPV6:
128 case RTE_FLOW_ITEM_TYPE_UDP:
129 case RTE_FLOW_ITEM_TYPE_TCP:
130 case RTE_FLOW_ITEM_TYPE_VXLAN:
131 case RTE_FLOW_ITEM_TYPE_NVGRE:
132 case RTE_FLOW_ITEM_TYPE_GRE:
133 case RTE_FLOW_ITEM_TYPE_GENEVE:
134 case RTE_FLOW_ITEM_TYPE_MPLS:
142 static enum rte_flow_item_type
143 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
145 enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
146 uint16_t ether_type = 0;
147 uint16_t ether_type_m;
148 uint8_t ip_next_proto = 0;
149 uint8_t ip_next_proto_m;
151 if (item == NULL || item->spec == NULL)
153 switch (item->type) {
154 case RTE_FLOW_ITEM_TYPE_ETH:
156 ether_type_m = ((const struct rte_flow_item_eth *)
159 ether_type_m = rte_flow_item_eth_mask.type;
160 if (ether_type_m != RTE_BE16(0xFFFF))
162 ether_type = ((const struct rte_flow_item_eth *)
164 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
165 ret = RTE_FLOW_ITEM_TYPE_IPV4;
166 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
167 ret = RTE_FLOW_ITEM_TYPE_IPV6;
168 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
169 ret = RTE_FLOW_ITEM_TYPE_VLAN;
171 ret = RTE_FLOW_ITEM_TYPE_END;
173 case RTE_FLOW_ITEM_TYPE_VLAN:
175 ether_type_m = ((const struct rte_flow_item_vlan *)
176 (item->mask))->inner_type;
178 ether_type_m = rte_flow_item_vlan_mask.inner_type;
179 if (ether_type_m != RTE_BE16(0xFFFF))
181 ether_type = ((const struct rte_flow_item_vlan *)
182 (item->spec))->inner_type;
183 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
184 ret = RTE_FLOW_ITEM_TYPE_IPV4;
185 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
186 ret = RTE_FLOW_ITEM_TYPE_IPV6;
187 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
188 ret = RTE_FLOW_ITEM_TYPE_VLAN;
190 ret = RTE_FLOW_ITEM_TYPE_END;
192 case RTE_FLOW_ITEM_TYPE_IPV4:
194 ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
195 (item->mask))->hdr.next_proto_id;
198 rte_flow_item_ipv4_mask.hdr.next_proto_id;
199 if (ip_next_proto_m != 0xFF)
201 ip_next_proto = ((const struct rte_flow_item_ipv4 *)
202 (item->spec))->hdr.next_proto_id;
203 if (ip_next_proto == IPPROTO_UDP)
204 ret = RTE_FLOW_ITEM_TYPE_UDP;
205 else if (ip_next_proto == IPPROTO_TCP)
206 ret = RTE_FLOW_ITEM_TYPE_TCP;
207 else if (ip_next_proto == IPPROTO_IP)
208 ret = RTE_FLOW_ITEM_TYPE_IPV4;
209 else if (ip_next_proto == IPPROTO_IPV6)
210 ret = RTE_FLOW_ITEM_TYPE_IPV6;
212 ret = RTE_FLOW_ITEM_TYPE_END;
214 case RTE_FLOW_ITEM_TYPE_IPV6:
216 ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
217 (item->mask))->hdr.proto;
220 rte_flow_item_ipv6_mask.hdr.proto;
221 if (ip_next_proto_m != 0xFF)
223 ip_next_proto = ((const struct rte_flow_item_ipv6 *)
224 (item->spec))->hdr.proto;
225 if (ip_next_proto == IPPROTO_UDP)
226 ret = RTE_FLOW_ITEM_TYPE_UDP;
227 else if (ip_next_proto == IPPROTO_TCP)
228 ret = RTE_FLOW_ITEM_TYPE_TCP;
229 else if (ip_next_proto == IPPROTO_IP)
230 ret = RTE_FLOW_ITEM_TYPE_IPV4;
231 else if (ip_next_proto == IPPROTO_IPV6)
232 ret = RTE_FLOW_ITEM_TYPE_IPV6;
234 ret = RTE_FLOW_ITEM_TYPE_END;
237 ret = RTE_FLOW_ITEM_TYPE_VOID;
243 #define MLX5_RSS_EXP_ELT_N 16
246 * Expand RSS flows into several possible flows according to the RSS hash
247 * fields requested and the driver capabilities.
250 * Buffer to store the result expansion.
252 * Buffer size in bytes. If 0, @p buf can be NULL.
256 * RSS types to expand (see ETH_RSS_* definitions).
258 * Input graph to expand @p pattern according to @p types.
259 * @param[in] graph_root_index
260 * Index of root node in @p graph, typically 0.
263 * A positive value representing the size of @p buf in bytes regardless of
264 * @p size on success, a negative errno value otherwise and rte_errno is
265 * set, the following errors are defined:
267 * -E2BIG: graph-depth @p graph is too deep.
268 * -EINVAL: @p size has not enough space for expanded pattern.
271 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
272 const struct rte_flow_item *pattern, uint64_t types,
273 const struct mlx5_flow_expand_node graph[],
274 int graph_root_index)
276 const struct rte_flow_item *item;
277 const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
278 const int *next_node;
279 const int *stack[MLX5_RSS_EXP_ELT_N];
281 struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
284 size_t user_pattern_size = 0;
286 const struct mlx5_flow_expand_node *next = NULL;
287 struct rte_flow_item missed_item;
290 const struct rte_flow_item *last_item = NULL;
292 memset(&missed_item, 0, sizeof(missed_item));
293 lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
294 MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
297 buf->entry[0].priority = 0;
298 buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
300 addr = buf->entry[0].pattern;
301 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
302 if (!mlx5_flow_is_rss_expandable_item(item)) {
303 user_pattern_size += sizeof(*item);
307 for (i = 0; node->next && node->next[i]; ++i) {
308 next = &graph[node->next[i]];
309 if (next->type == item->type)
314 user_pattern_size += sizeof(*item);
316 user_pattern_size += sizeof(*item); /* Handle END item. */
317 lsize += user_pattern_size;
320 /* Copy the user pattern in the first entry of the buffer. */
321 rte_memcpy(addr, pattern, user_pattern_size);
322 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
324 /* Start expanding. */
325 memset(flow_items, 0, sizeof(flow_items));
326 user_pattern_size -= sizeof(*item);
328 * Check if the last valid item has spec set, need complete pattern,
329 * and the pattern can be used for expansion.
331 missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
332 if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
333 /* Item type END indicates expansion is not required. */
336 if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
339 for (i = 0; node->next && node->next[i]; ++i) {
340 next = &graph[node->next[i]];
341 if (next->type == missed_item.type) {
342 flow_items[0].type = missed_item.type;
343 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
349 if (next && missed) {
350 elt = 2; /* missed item + item end. */
352 lsize += elt * sizeof(*item) + user_pattern_size;
355 if (node->rss_types & types) {
356 buf->entry[buf->entries].priority = 1;
357 buf->entry[buf->entries].pattern = addr;
359 rte_memcpy(addr, buf->entry[0].pattern,
361 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
362 rte_memcpy(addr, flow_items, elt * sizeof(*item));
363 addr = (void *)(((uintptr_t)addr) +
364 elt * sizeof(*item));
367 memset(flow_items, 0, sizeof(flow_items));
368 next_node = node->next;
369 stack[stack_pos] = next_node;
370 node = next_node ? &graph[*next_node] : NULL;
372 flow_items[stack_pos].type = node->type;
373 if (node->rss_types & types) {
376 * compute the number of items to copy from the
377 * expansion and copy it.
378 * When the stack_pos is 0, there are 1 element in it,
379 * plus the addition END item.
382 flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
383 lsize += elt * sizeof(*item) + user_pattern_size;
386 n = elt * sizeof(*item);
387 buf->entry[buf->entries].priority =
388 stack_pos + 1 + missed;
389 buf->entry[buf->entries].pattern = addr;
391 rte_memcpy(addr, buf->entry[0].pattern,
393 addr = (void *)(((uintptr_t)addr) +
395 rte_memcpy(addr, &missed_item,
396 missed * sizeof(*item));
397 addr = (void *)(((uintptr_t)addr) +
398 missed * sizeof(*item));
399 rte_memcpy(addr, flow_items, n);
400 addr = (void *)(((uintptr_t)addr) + n);
403 if (!node->optional && node->next) {
404 next_node = node->next;
405 if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
409 stack[stack_pos] = next_node;
410 } else if (*(next_node + 1)) {
411 /* Follow up with the next possibility. */
414 /* Move to the next path. */
416 next_node = stack[--stack_pos];
418 stack[stack_pos] = next_node;
420 node = *next_node ? &graph[*next_node] : NULL;
425 enum mlx5_expansion {
427 MLX5_EXPANSION_ROOT_OUTER,
428 MLX5_EXPANSION_ROOT_ETH_VLAN,
429 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
430 MLX5_EXPANSION_OUTER_ETH,
431 MLX5_EXPANSION_OUTER_ETH_VLAN,
432 MLX5_EXPANSION_OUTER_VLAN,
433 MLX5_EXPANSION_OUTER_IPV4,
434 MLX5_EXPANSION_OUTER_IPV4_UDP,
435 MLX5_EXPANSION_OUTER_IPV4_TCP,
436 MLX5_EXPANSION_OUTER_IPV6,
437 MLX5_EXPANSION_OUTER_IPV6_UDP,
438 MLX5_EXPANSION_OUTER_IPV6_TCP,
439 MLX5_EXPANSION_VXLAN,
440 MLX5_EXPANSION_VXLAN_GPE,
442 MLX5_EXPANSION_NVGRE,
443 MLX5_EXPANSION_GRE_KEY,
446 MLX5_EXPANSION_ETH_VLAN,
449 MLX5_EXPANSION_IPV4_UDP,
450 MLX5_EXPANSION_IPV4_TCP,
452 MLX5_EXPANSION_IPV6_UDP,
453 MLX5_EXPANSION_IPV6_TCP,
456 /** Supported expansion of items. */
457 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
458 [MLX5_EXPANSION_ROOT] = {
459 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
461 MLX5_EXPANSION_IPV6),
462 .type = RTE_FLOW_ITEM_TYPE_END,
464 [MLX5_EXPANSION_ROOT_OUTER] = {
465 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
466 MLX5_EXPANSION_OUTER_IPV4,
467 MLX5_EXPANSION_OUTER_IPV6),
468 .type = RTE_FLOW_ITEM_TYPE_END,
470 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
471 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
472 .type = RTE_FLOW_ITEM_TYPE_END,
474 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
475 .next = MLX5_FLOW_EXPAND_RSS_NEXT
476 (MLX5_EXPANSION_OUTER_ETH_VLAN),
477 .type = RTE_FLOW_ITEM_TYPE_END,
479 [MLX5_EXPANSION_OUTER_ETH] = {
480 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
481 MLX5_EXPANSION_OUTER_IPV6),
482 .type = RTE_FLOW_ITEM_TYPE_ETH,
485 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
486 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
487 .type = RTE_FLOW_ITEM_TYPE_ETH,
490 [MLX5_EXPANSION_OUTER_VLAN] = {
491 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
492 MLX5_EXPANSION_OUTER_IPV6),
493 .type = RTE_FLOW_ITEM_TYPE_VLAN,
495 [MLX5_EXPANSION_OUTER_IPV4] = {
496 .next = MLX5_FLOW_EXPAND_RSS_NEXT
497 (MLX5_EXPANSION_OUTER_IPV4_UDP,
498 MLX5_EXPANSION_OUTER_IPV4_TCP,
500 MLX5_EXPANSION_NVGRE,
502 MLX5_EXPANSION_IPV6),
503 .type = RTE_FLOW_ITEM_TYPE_IPV4,
504 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
505 ETH_RSS_NONFRAG_IPV4_OTHER,
507 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
508 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
509 MLX5_EXPANSION_VXLAN_GPE,
510 MLX5_EXPANSION_MPLS),
511 .type = RTE_FLOW_ITEM_TYPE_UDP,
512 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
514 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
515 .type = RTE_FLOW_ITEM_TYPE_TCP,
516 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
518 [MLX5_EXPANSION_OUTER_IPV6] = {
519 .next = MLX5_FLOW_EXPAND_RSS_NEXT
520 (MLX5_EXPANSION_OUTER_IPV6_UDP,
521 MLX5_EXPANSION_OUTER_IPV6_TCP,
525 MLX5_EXPANSION_NVGRE),
526 .type = RTE_FLOW_ITEM_TYPE_IPV6,
527 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
528 ETH_RSS_NONFRAG_IPV6_OTHER,
530 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
531 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
532 MLX5_EXPANSION_VXLAN_GPE,
533 MLX5_EXPANSION_MPLS),
534 .type = RTE_FLOW_ITEM_TYPE_UDP,
535 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
537 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
538 .type = RTE_FLOW_ITEM_TYPE_TCP,
539 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
541 [MLX5_EXPANSION_VXLAN] = {
542 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
544 MLX5_EXPANSION_IPV6),
545 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
547 [MLX5_EXPANSION_VXLAN_GPE] = {
548 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
550 MLX5_EXPANSION_IPV6),
551 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
553 [MLX5_EXPANSION_GRE] = {
554 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
556 MLX5_EXPANSION_GRE_KEY,
557 MLX5_EXPANSION_MPLS),
558 .type = RTE_FLOW_ITEM_TYPE_GRE,
560 [MLX5_EXPANSION_GRE_KEY] = {
561 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
563 MLX5_EXPANSION_MPLS),
564 .type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
567 [MLX5_EXPANSION_NVGRE] = {
568 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
569 .type = RTE_FLOW_ITEM_TYPE_NVGRE,
571 [MLX5_EXPANSION_MPLS] = {
572 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
575 .type = RTE_FLOW_ITEM_TYPE_MPLS,
577 [MLX5_EXPANSION_ETH] = {
578 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
579 MLX5_EXPANSION_IPV6),
580 .type = RTE_FLOW_ITEM_TYPE_ETH,
582 [MLX5_EXPANSION_ETH_VLAN] = {
583 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
584 .type = RTE_FLOW_ITEM_TYPE_ETH,
586 [MLX5_EXPANSION_VLAN] = {
587 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
588 MLX5_EXPANSION_IPV6),
589 .type = RTE_FLOW_ITEM_TYPE_VLAN,
591 [MLX5_EXPANSION_IPV4] = {
592 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
593 MLX5_EXPANSION_IPV4_TCP),
594 .type = RTE_FLOW_ITEM_TYPE_IPV4,
595 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
596 ETH_RSS_NONFRAG_IPV4_OTHER,
598 [MLX5_EXPANSION_IPV4_UDP] = {
599 .type = RTE_FLOW_ITEM_TYPE_UDP,
600 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
602 [MLX5_EXPANSION_IPV4_TCP] = {
603 .type = RTE_FLOW_ITEM_TYPE_TCP,
604 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
606 [MLX5_EXPANSION_IPV6] = {
607 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
608 MLX5_EXPANSION_IPV6_TCP),
609 .type = RTE_FLOW_ITEM_TYPE_IPV6,
610 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
611 ETH_RSS_NONFRAG_IPV6_OTHER,
613 [MLX5_EXPANSION_IPV6_UDP] = {
614 .type = RTE_FLOW_ITEM_TYPE_UDP,
615 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
617 [MLX5_EXPANSION_IPV6_TCP] = {
618 .type = RTE_FLOW_ITEM_TYPE_TCP,
619 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
623 static struct rte_flow_action_handle *
624 mlx5_action_handle_create(struct rte_eth_dev *dev,
625 const struct rte_flow_indir_action_conf *conf,
626 const struct rte_flow_action *action,
627 struct rte_flow_error *error);
628 static int mlx5_action_handle_destroy
629 (struct rte_eth_dev *dev,
630 struct rte_flow_action_handle *handle,
631 struct rte_flow_error *error);
632 static int mlx5_action_handle_update
633 (struct rte_eth_dev *dev,
634 struct rte_flow_action_handle *handle,
636 struct rte_flow_error *error);
637 static int mlx5_action_handle_query
638 (struct rte_eth_dev *dev,
639 const struct rte_flow_action_handle *handle,
641 struct rte_flow_error *error);
643 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
644 struct rte_flow_tunnel *app_tunnel,
645 struct rte_flow_action **actions,
646 uint32_t *num_of_actions,
647 struct rte_flow_error *error);
649 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
650 struct rte_flow_tunnel *app_tunnel,
651 struct rte_flow_item **items,
652 uint32_t *num_of_items,
653 struct rte_flow_error *error);
655 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
656 struct rte_flow_item *pmd_items,
657 uint32_t num_items, struct rte_flow_error *err);
659 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
660 struct rte_flow_action *pmd_actions,
661 uint32_t num_actions,
662 struct rte_flow_error *err);
664 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
666 struct rte_flow_restore_info *info,
667 struct rte_flow_error *err);
669 static const struct rte_flow_ops mlx5_flow_ops = {
670 .validate = mlx5_flow_validate,
671 .create = mlx5_flow_create,
672 .destroy = mlx5_flow_destroy,
673 .flush = mlx5_flow_flush,
674 .isolate = mlx5_flow_isolate,
675 .query = mlx5_flow_query,
676 .dev_dump = mlx5_flow_dev_dump,
677 .get_aged_flows = mlx5_flow_get_aged_flows,
678 .action_handle_create = mlx5_action_handle_create,
679 .action_handle_destroy = mlx5_action_handle_destroy,
680 .action_handle_update = mlx5_action_handle_update,
681 .action_handle_query = mlx5_action_handle_query,
682 .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
683 .tunnel_match = mlx5_flow_tunnel_match,
684 .tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
685 .tunnel_item_release = mlx5_flow_tunnel_item_release,
686 .get_restore_info = mlx5_flow_tunnel_get_restore_info,
689 /* Tunnel information. */
690 struct mlx5_flow_tunnel_info {
691 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
692 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
695 static struct mlx5_flow_tunnel_info tunnels_info[] = {
697 .tunnel = MLX5_FLOW_LAYER_VXLAN,
698 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
701 .tunnel = MLX5_FLOW_LAYER_GENEVE,
702 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
705 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
706 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
709 .tunnel = MLX5_FLOW_LAYER_GRE,
710 .ptype = RTE_PTYPE_TUNNEL_GRE,
713 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
714 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
717 .tunnel = MLX5_FLOW_LAYER_MPLS,
718 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
721 .tunnel = MLX5_FLOW_LAYER_NVGRE,
722 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
725 .tunnel = MLX5_FLOW_LAYER_IPIP,
726 .ptype = RTE_PTYPE_TUNNEL_IP,
729 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
730 .ptype = RTE_PTYPE_TUNNEL_IP,
733 .tunnel = MLX5_FLOW_LAYER_GTP,
734 .ptype = RTE_PTYPE_TUNNEL_GTPU,
741 * Translate tag ID to register.
744 * Pointer to the Ethernet device structure.
746 * The feature that request the register.
748 * The request register ID.
750 * Error description in case of any.
753 * The request register on success, a negative errno
754 * value otherwise and rte_errno is set.
757 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
758 enum mlx5_feature_name feature,
760 struct rte_flow_error *error)
762 struct mlx5_priv *priv = dev->data->dev_private;
763 struct mlx5_dev_config *config = &priv->config;
764 enum modify_reg start_reg;
765 bool skip_mtr_reg = false;
768 case MLX5_HAIRPIN_RX:
770 case MLX5_HAIRPIN_TX:
772 case MLX5_METADATA_RX:
773 switch (config->dv_xmeta_en) {
774 case MLX5_XMETA_MODE_LEGACY:
776 case MLX5_XMETA_MODE_META16:
778 case MLX5_XMETA_MODE_META32:
782 case MLX5_METADATA_TX:
784 case MLX5_METADATA_FDB:
785 switch (config->dv_xmeta_en) {
786 case MLX5_XMETA_MODE_LEGACY:
788 case MLX5_XMETA_MODE_META16:
790 case MLX5_XMETA_MODE_META32:
795 switch (config->dv_xmeta_en) {
796 case MLX5_XMETA_MODE_LEGACY:
798 case MLX5_XMETA_MODE_META16:
800 case MLX5_XMETA_MODE_META32:
806 * If meter color and meter id share one register, flow match
807 * should use the meter color register for match.
809 if (priv->mtr_reg_share)
810 return priv->mtr_color_reg;
812 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
815 case MLX5_ASO_FLOW_HIT:
816 case MLX5_ASO_CONNTRACK:
817 /* All features use the same REG_C. */
818 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
819 return priv->mtr_color_reg;
822 * Metadata COPY_MARK register using is in meter suffix sub
823 * flow while with meter. It's safe to share the same register.
825 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
828 * If meter is enable, it will engage the register for color
829 * match and flow match. If meter color match is not using the
830 * REG_C_2, need to skip the REG_C_x be used by meter color
832 * If meter is disable, free to use all available registers.
834 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
835 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
836 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
837 if (id > (uint32_t)(REG_C_7 - start_reg))
838 return rte_flow_error_set(error, EINVAL,
839 RTE_FLOW_ERROR_TYPE_ITEM,
840 NULL, "invalid tag id");
841 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
842 return rte_flow_error_set(error, ENOTSUP,
843 RTE_FLOW_ERROR_TYPE_ITEM,
844 NULL, "unsupported tag id");
846 * This case means meter is using the REG_C_x great than 2.
847 * Take care not to conflict with meter color REG_C_x.
848 * If the available index REG_C_y >= REG_C_x, skip the
851 if (skip_mtr_reg && config->flow_mreg_c
852 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
853 if (id >= (uint32_t)(REG_C_7 - start_reg))
854 return rte_flow_error_set(error, EINVAL,
855 RTE_FLOW_ERROR_TYPE_ITEM,
856 NULL, "invalid tag id");
857 if (config->flow_mreg_c
858 [id + 1 + start_reg - REG_C_0] != REG_NON)
859 return config->flow_mreg_c
860 [id + 1 + start_reg - REG_C_0];
861 return rte_flow_error_set(error, ENOTSUP,
862 RTE_FLOW_ERROR_TYPE_ITEM,
863 NULL, "unsupported tag id");
865 return config->flow_mreg_c[id + start_reg - REG_C_0];
868 return rte_flow_error_set(error, EINVAL,
869 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
870 NULL, "invalid feature name");
874 * Check extensive flow metadata register support.
877 * Pointer to rte_eth_dev structure.
880 * True if device supports extensive flow metadata register, otherwise false.
883 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
885 struct mlx5_priv *priv = dev->data->dev_private;
886 struct mlx5_dev_config *config = &priv->config;
889 * Having available reg_c can be regarded inclusively as supporting
890 * extensive flow metadata register, which could mean,
891 * - metadata register copy action by modify header.
892 * - 16 modify header actions is supported.
893 * - reg_c's are preserved across different domain (FDB and NIC) on
894 * packet loopback by flow lookup miss.
896 return config->flow_mreg_c[2] != REG_NON;
900 * Get the lowest priority.
903 * Pointer to the Ethernet device structure.
904 * @param[in] attributes
905 * Pointer to device flow rule attributes.
908 * The value of lowest priority of flow.
911 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
912 const struct rte_flow_attr *attr)
914 struct mlx5_priv *priv = dev->data->dev_private;
916 if (!attr->group && !attr->transfer)
917 return priv->config.flow_prio - 2;
918 return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
922 * Calculate matcher priority of the flow.
925 * Pointer to the Ethernet device structure.
927 * Pointer to device flow rule attributes.
928 * @param[in] subpriority
929 * The priority based on the items.
931 * The matcher priority of the flow.
934 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
935 const struct rte_flow_attr *attr,
936 uint32_t subpriority)
938 uint16_t priority = (uint16_t)attr->priority;
939 struct mlx5_priv *priv = dev->data->dev_private;
941 if (!attr->group && !attr->transfer) {
942 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
943 priority = priv->config.flow_prio - 1;
944 return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
946 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
947 priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
948 return priority * 3 + subpriority;
952 * Verify the @p item specifications (spec, last, mask) are compatible with the
956 * Item specification.
958 * @p item->mask or flow default bit-masks.
959 * @param[in] nic_mask
960 * Bit-masks covering supported fields by the NIC to compare with user mask.
962 * Bit-masks size in bytes.
963 * @param[in] range_accepted
964 * True if range of values is accepted for specific fields, false otherwise.
966 * Pointer to error structure.
969 * 0 on success, a negative errno value otherwise and rte_errno is set.
972 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
974 const uint8_t *nic_mask,
977 struct rte_flow_error *error)
981 MLX5_ASSERT(nic_mask);
982 for (i = 0; i < size; ++i)
983 if ((nic_mask[i] | mask[i]) != nic_mask[i])
984 return rte_flow_error_set(error, ENOTSUP,
985 RTE_FLOW_ERROR_TYPE_ITEM,
987 "mask enables non supported"
989 if (!item->spec && (item->mask || item->last))
990 return rte_flow_error_set(error, EINVAL,
991 RTE_FLOW_ERROR_TYPE_ITEM, item,
992 "mask/last without a spec is not"
994 if (item->spec && item->last && !range_accepted) {
1000 for (i = 0; i < size; ++i) {
1001 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1002 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1004 ret = memcmp(spec, last, size);
1006 return rte_flow_error_set(error, EINVAL,
1007 RTE_FLOW_ERROR_TYPE_ITEM,
1009 "range is not valid");
1015 * Adjust the hash fields according to the @p flow information.
1017 * @param[in] dev_flow.
1018 * Pointer to the mlx5_flow.
1020 * 1 when the hash field is for a tunnel item.
1021 * @param[in] layer_types
1023 * @param[in] hash_fields
1027 * The hash fields that should be used.
1030 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1031 int tunnel __rte_unused, uint64_t layer_types,
1032 uint64_t hash_fields)
1034 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1035 int rss_request_inner = rss_desc->level >= 2;
1037 /* Check RSS hash level for tunnel. */
1038 if (tunnel && rss_request_inner)
1039 hash_fields |= IBV_RX_HASH_INNER;
1040 else if (tunnel || rss_request_inner)
1043 /* Check if requested layer matches RSS hash fields. */
1044 if (!(rss_desc->types & layer_types))
1050 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
1051 * if several tunnel rules are used on this queue, the tunnel ptype will be
1055 * Rx queue to update.
1058 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1061 uint32_t tunnel_ptype = 0;
1063 /* Look up for the ptype to use. */
1064 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1065 if (!rxq_ctrl->flow_tunnels_n[i])
1067 if (!tunnel_ptype) {
1068 tunnel_ptype = tunnels_info[i].ptype;
1074 rxq_ctrl->rxq.tunnel = tunnel_ptype;
1078 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1082 * Pointer to the Ethernet device structure.
1083 * @param[in] dev_handle
1084 * Pointer to device flow handle structure.
1087 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1088 struct mlx5_flow_handle *dev_handle)
1090 struct mlx5_priv *priv = dev->data->dev_private;
1091 const int mark = dev_handle->mark;
1092 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1093 struct mlx5_ind_table_obj *ind_tbl = NULL;
1096 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1097 struct mlx5_hrxq *hrxq;
1099 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1100 dev_handle->rix_hrxq);
1102 ind_tbl = hrxq->ind_table;
1103 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1104 struct mlx5_shared_action_rss *shared_rss;
1106 shared_rss = mlx5_ipool_get
1107 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1108 dev_handle->rix_srss);
1110 ind_tbl = shared_rss->ind_tbl;
1114 for (i = 0; i != ind_tbl->queues_n; ++i) {
1115 int idx = ind_tbl->queues[i];
1116 struct mlx5_rxq_ctrl *rxq_ctrl =
1117 container_of((*priv->rxqs)[idx],
1118 struct mlx5_rxq_ctrl, rxq);
1121 * To support metadata register copy on Tx loopback,
1122 * this must be always enabled (metadata may arive
1123 * from other port - not from local flows only.
1125 if (priv->config.dv_flow_en &&
1126 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1127 mlx5_flow_ext_mreg_supported(dev)) {
1128 rxq_ctrl->rxq.mark = 1;
1129 rxq_ctrl->flow_mark_n = 1;
1131 rxq_ctrl->rxq.mark = 1;
1132 rxq_ctrl->flow_mark_n++;
1137 /* Increase the counter matching the flow. */
1138 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1139 if ((tunnels_info[j].tunnel &
1140 dev_handle->layers) ==
1141 tunnels_info[j].tunnel) {
1142 rxq_ctrl->flow_tunnels_n[j]++;
1146 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1152 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1155 * Pointer to the Ethernet device structure.
1157 * Pointer to flow structure.
1160 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1162 struct mlx5_priv *priv = dev->data->dev_private;
1163 uint32_t handle_idx;
1164 struct mlx5_flow_handle *dev_handle;
1166 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1167 handle_idx, dev_handle, next)
1168 flow_drv_rxq_flags_set(dev, dev_handle);
1172 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1173 * device flow if no other flow uses it with the same kind of request.
1176 * Pointer to Ethernet device.
1177 * @param[in] dev_handle
1178 * Pointer to the device flow handle structure.
1181 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1182 struct mlx5_flow_handle *dev_handle)
1184 struct mlx5_priv *priv = dev->data->dev_private;
1185 const int mark = dev_handle->mark;
1186 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1187 struct mlx5_ind_table_obj *ind_tbl = NULL;
1190 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1191 struct mlx5_hrxq *hrxq;
1193 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1194 dev_handle->rix_hrxq);
1196 ind_tbl = hrxq->ind_table;
1197 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1198 struct mlx5_shared_action_rss *shared_rss;
1200 shared_rss = mlx5_ipool_get
1201 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1202 dev_handle->rix_srss);
1204 ind_tbl = shared_rss->ind_tbl;
1208 MLX5_ASSERT(dev->data->dev_started);
1209 for (i = 0; i != ind_tbl->queues_n; ++i) {
1210 int idx = ind_tbl->queues[i];
1211 struct mlx5_rxq_ctrl *rxq_ctrl =
1212 container_of((*priv->rxqs)[idx],
1213 struct mlx5_rxq_ctrl, rxq);
1215 if (priv->config.dv_flow_en &&
1216 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1217 mlx5_flow_ext_mreg_supported(dev)) {
1218 rxq_ctrl->rxq.mark = 1;
1219 rxq_ctrl->flow_mark_n = 1;
1221 rxq_ctrl->flow_mark_n--;
1222 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1227 /* Decrease the counter matching the flow. */
1228 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1229 if ((tunnels_info[j].tunnel &
1230 dev_handle->layers) ==
1231 tunnels_info[j].tunnel) {
1232 rxq_ctrl->flow_tunnels_n[j]--;
1236 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1242 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1243 * @p flow if no other flow uses it with the same kind of request.
1246 * Pointer to Ethernet device.
1248 * Pointer to the flow.
1251 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1253 struct mlx5_priv *priv = dev->data->dev_private;
1254 uint32_t handle_idx;
1255 struct mlx5_flow_handle *dev_handle;
1257 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1258 handle_idx, dev_handle, next)
1259 flow_drv_rxq_flags_trim(dev, dev_handle);
1263 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1266 * Pointer to Ethernet device.
1269 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1271 struct mlx5_priv *priv = dev->data->dev_private;
1274 for (i = 0; i != priv->rxqs_n; ++i) {
1275 struct mlx5_rxq_ctrl *rxq_ctrl;
1278 if (!(*priv->rxqs)[i])
1280 rxq_ctrl = container_of((*priv->rxqs)[i],
1281 struct mlx5_rxq_ctrl, rxq);
1282 rxq_ctrl->flow_mark_n = 0;
1283 rxq_ctrl->rxq.mark = 0;
1284 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1285 rxq_ctrl->flow_tunnels_n[j] = 0;
1286 rxq_ctrl->rxq.tunnel = 0;
1291 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1294 * Pointer to the Ethernet device structure.
1297 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1299 struct mlx5_priv *priv = dev->data->dev_private;
1300 struct mlx5_rxq_data *data;
1303 for (i = 0; i != priv->rxqs_n; ++i) {
1304 if (!(*priv->rxqs)[i])
1306 data = (*priv->rxqs)[i];
1307 if (!rte_flow_dynf_metadata_avail()) {
1308 data->dynf_meta = 0;
1309 data->flow_meta_mask = 0;
1310 data->flow_meta_offset = -1;
1311 data->flow_meta_port_mask = 0;
1313 data->dynf_meta = 1;
1314 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1315 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1316 data->flow_meta_port_mask = (uint32_t)~0;
1317 if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1318 data->flow_meta_port_mask >>= 16;
1324 * return a pointer to the desired action in the list of actions.
1326 * @param[in] actions
1327 * The list of actions to search the action in.
1329 * The action to find.
1332 * Pointer to the action in the list, if found. NULL otherwise.
1334 const struct rte_flow_action *
1335 mlx5_flow_find_action(const struct rte_flow_action *actions,
1336 enum rte_flow_action_type action)
1338 if (actions == NULL)
1340 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1341 if (actions->type == action)
1347 * Validate the flag action.
1349 * @param[in] action_flags
1350 * Bit-fields that holds the actions detected until now.
1352 * Attributes of flow that includes this action.
1354 * Pointer to error structure.
1357 * 0 on success, a negative errno value otherwise and rte_errno is set.
1360 mlx5_flow_validate_action_flag(uint64_t action_flags,
1361 const struct rte_flow_attr *attr,
1362 struct rte_flow_error *error)
1364 if (action_flags & MLX5_FLOW_ACTION_MARK)
1365 return rte_flow_error_set(error, EINVAL,
1366 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1367 "can't mark and flag in same flow");
1368 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1369 return rte_flow_error_set(error, EINVAL,
1370 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1372 " actions in same flow");
1374 return rte_flow_error_set(error, ENOTSUP,
1375 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1376 "flag action not supported for "
1382 * Validate the mark action.
1385 * Pointer to the queue action.
1386 * @param[in] action_flags
1387 * Bit-fields that holds the actions detected until now.
1389 * Attributes of flow that includes this action.
1391 * Pointer to error structure.
1394 * 0 on success, a negative errno value otherwise and rte_errno is set.
1397 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1398 uint64_t action_flags,
1399 const struct rte_flow_attr *attr,
1400 struct rte_flow_error *error)
1402 const struct rte_flow_action_mark *mark = action->conf;
1405 return rte_flow_error_set(error, EINVAL,
1406 RTE_FLOW_ERROR_TYPE_ACTION,
1408 "configuration cannot be null");
1409 if (mark->id >= MLX5_FLOW_MARK_MAX)
1410 return rte_flow_error_set(error, EINVAL,
1411 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1413 "mark id must in 0 <= id < "
1414 RTE_STR(MLX5_FLOW_MARK_MAX));
1415 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1416 return rte_flow_error_set(error, EINVAL,
1417 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1418 "can't flag and mark in same flow");
1419 if (action_flags & MLX5_FLOW_ACTION_MARK)
1420 return rte_flow_error_set(error, EINVAL,
1421 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1422 "can't have 2 mark actions in same"
1425 return rte_flow_error_set(error, ENOTSUP,
1426 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1427 "mark action not supported for "
1433 * Validate the drop action.
1435 * @param[in] action_flags
1436 * Bit-fields that holds the actions detected until now.
1438 * Attributes of flow that includes this action.
1440 * Pointer to error structure.
1443 * 0 on success, a negative errno value otherwise and rte_errno is set.
1446 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1447 const struct rte_flow_attr *attr,
1448 struct rte_flow_error *error)
1451 return rte_flow_error_set(error, ENOTSUP,
1452 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1453 "drop action not supported for "
1459 * Validate the queue action.
1462 * Pointer to the queue action.
1463 * @param[in] action_flags
1464 * Bit-fields that holds the actions detected until now.
1466 * Pointer to the Ethernet device structure.
1468 * Attributes of flow that includes this action.
1470 * Pointer to error structure.
1473 * 0 on success, a negative errno value otherwise and rte_errno is set.
1476 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1477 uint64_t action_flags,
1478 struct rte_eth_dev *dev,
1479 const struct rte_flow_attr *attr,
1480 struct rte_flow_error *error)
1482 struct mlx5_priv *priv = dev->data->dev_private;
1483 const struct rte_flow_action_queue *queue = action->conf;
1485 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1486 return rte_flow_error_set(error, EINVAL,
1487 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1488 "can't have 2 fate actions in"
1491 return rte_flow_error_set(error, EINVAL,
1492 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1493 NULL, "No Rx queues configured");
1494 if (queue->index >= priv->rxqs_n)
1495 return rte_flow_error_set(error, EINVAL,
1496 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1498 "queue index out of range");
1499 if (!(*priv->rxqs)[queue->index])
1500 return rte_flow_error_set(error, EINVAL,
1501 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1503 "queue is not configured");
1505 return rte_flow_error_set(error, ENOTSUP,
1506 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1507 "queue action not supported for "
1513 * Validate the rss action.
1516 * Pointer to the Ethernet device structure.
1518 * Pointer to the queue action.
1520 * Pointer to error structure.
1523 * 0 on success, a negative errno value otherwise and rte_errno is set.
1526 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1527 const struct rte_flow_action *action,
1528 struct rte_flow_error *error)
1530 struct mlx5_priv *priv = dev->data->dev_private;
1531 const struct rte_flow_action_rss *rss = action->conf;
1532 enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1535 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1536 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1537 return rte_flow_error_set(error, ENOTSUP,
1538 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1540 "RSS hash function not supported");
1541 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1546 return rte_flow_error_set(error, ENOTSUP,
1547 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1549 "tunnel RSS is not supported");
1550 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1551 if (rss->key_len == 0 && rss->key != NULL)
1552 return rte_flow_error_set(error, ENOTSUP,
1553 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1555 "RSS hash key length 0");
1556 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1557 return rte_flow_error_set(error, ENOTSUP,
1558 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1560 "RSS hash key too small");
1561 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1562 return rte_flow_error_set(error, ENOTSUP,
1563 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1565 "RSS hash key too large");
1566 if (rss->queue_num > priv->config.ind_table_max_size)
1567 return rte_flow_error_set(error, ENOTSUP,
1568 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1570 "number of queues too large");
1571 if (rss->types & MLX5_RSS_HF_MASK)
1572 return rte_flow_error_set(error, ENOTSUP,
1573 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1575 "some RSS protocols are not"
1577 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1578 !(rss->types & ETH_RSS_IP))
1579 return rte_flow_error_set(error, EINVAL,
1580 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1581 "L3 partial RSS requested but L3 RSS"
1582 " type not specified");
1583 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1584 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1585 return rte_flow_error_set(error, EINVAL,
1586 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1587 "L4 partial RSS requested but L4 RSS"
1588 " type not specified");
1590 return rte_flow_error_set(error, EINVAL,
1591 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1592 NULL, "No Rx queues configured");
1593 if (!rss->queue_num)
1594 return rte_flow_error_set(error, EINVAL,
1595 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1596 NULL, "No queues configured");
1597 for (i = 0; i != rss->queue_num; ++i) {
1598 struct mlx5_rxq_ctrl *rxq_ctrl;
1600 if (rss->queue[i] >= priv->rxqs_n)
1601 return rte_flow_error_set
1603 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1604 &rss->queue[i], "queue index out of range");
1605 if (!(*priv->rxqs)[rss->queue[i]])
1606 return rte_flow_error_set
1607 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1608 &rss->queue[i], "queue is not configured");
1609 rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1610 struct mlx5_rxq_ctrl, rxq);
1612 rxq_type = rxq_ctrl->type;
1613 if (rxq_type != rxq_ctrl->type)
1614 return rte_flow_error_set
1615 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1617 "combining hairpin and regular RSS queues is not supported");
1623 * Validate the rss action.
1626 * Pointer to the queue action.
1627 * @param[in] action_flags
1628 * Bit-fields that holds the actions detected until now.
1630 * Pointer to the Ethernet device structure.
1632 * Attributes of flow that includes this action.
1633 * @param[in] item_flags
1634 * Items that were detected.
1636 * Pointer to error structure.
1639 * 0 on success, a negative errno value otherwise and rte_errno is set.
1642 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1643 uint64_t action_flags,
1644 struct rte_eth_dev *dev,
1645 const struct rte_flow_attr *attr,
1646 uint64_t item_flags,
1647 struct rte_flow_error *error)
1649 const struct rte_flow_action_rss *rss = action->conf;
1650 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1653 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1654 return rte_flow_error_set(error, EINVAL,
1655 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1656 "can't have 2 fate actions"
1658 ret = mlx5_validate_action_rss(dev, action, error);
1662 return rte_flow_error_set(error, ENOTSUP,
1663 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1664 "rss action not supported for "
1666 if (rss->level > 1 && !tunnel)
1667 return rte_flow_error_set(error, EINVAL,
1668 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1669 "inner RSS is not supported for "
1670 "non-tunnel flows");
1671 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1672 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1673 return rte_flow_error_set(error, EINVAL,
1674 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1675 "RSS on eCPRI is not supported now");
1677 if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1679 (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1681 return rte_flow_error_set(error, EINVAL,
1682 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1683 "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1688 * Validate the default miss action.
1690 * @param[in] action_flags
1691 * Bit-fields that holds the actions detected until now.
1693 * Pointer to error structure.
1696 * 0 on success, a negative errno value otherwise and rte_errno is set.
1699 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1700 const struct rte_flow_attr *attr,
1701 struct rte_flow_error *error)
1703 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1704 return rte_flow_error_set(error, EINVAL,
1705 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1706 "can't have 2 fate actions in"
1709 return rte_flow_error_set(error, ENOTSUP,
1710 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1711 "default miss action not supported "
1714 return rte_flow_error_set(error, ENOTSUP,
1715 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1716 "only group 0 is supported");
1718 return rte_flow_error_set(error, ENOTSUP,
1719 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1720 NULL, "transfer is not supported");
1725 * Validate the count action.
1728 * Pointer to the Ethernet device structure.
1730 * Attributes of flow that includes this action.
1732 * Pointer to error structure.
1735 * 0 on success, a negative errno value otherwise and rte_errno is set.
1738 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1739 const struct rte_flow_attr *attr,
1740 struct rte_flow_error *error)
1743 return rte_flow_error_set(error, ENOTSUP,
1744 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1745 "count action not supported for "
1751 * Validate the ASO CT action.
1754 * Pointer to the Ethernet device structure.
1755 * @param[in] conntrack
1756 * Pointer to the CT action profile.
1758 * Pointer to error structure.
1761 * 0 on success, a negative errno value otherwise and rte_errno is set.
1764 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1765 const struct rte_flow_action_conntrack *conntrack,
1766 struct rte_flow_error *error)
1770 if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1771 return rte_flow_error_set(error, EINVAL,
1772 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1773 "Invalid CT state");
1774 if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1775 return rte_flow_error_set(error, EINVAL,
1776 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1777 "Invalid last TCP packet flag");
1782 * Verify the @p attributes will be correctly understood by the NIC and store
1783 * them in the @p flow if everything is correct.
1786 * Pointer to the Ethernet device structure.
1787 * @param[in] attributes
1788 * Pointer to flow attributes
1790 * Pointer to error structure.
1793 * 0 on success, a negative errno value otherwise and rte_errno is set.
1796 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1797 const struct rte_flow_attr *attributes,
1798 struct rte_flow_error *error)
1800 struct mlx5_priv *priv = dev->data->dev_private;
1801 uint32_t priority_max = priv->config.flow_prio - 1;
1803 if (attributes->group)
1804 return rte_flow_error_set(error, ENOTSUP,
1805 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1806 NULL, "groups is not supported");
1807 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1808 attributes->priority >= priority_max)
1809 return rte_flow_error_set(error, ENOTSUP,
1810 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1811 NULL, "priority out of range");
1812 if (attributes->egress)
1813 return rte_flow_error_set(error, ENOTSUP,
1814 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1815 "egress is not supported");
1816 if (attributes->transfer && !priv->config.dv_esw_en)
1817 return rte_flow_error_set(error, ENOTSUP,
1818 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1819 NULL, "transfer is not supported");
1820 if (!attributes->ingress)
1821 return rte_flow_error_set(error, EINVAL,
1822 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1824 "ingress attribute is mandatory");
1829 * Validate ICMP6 item.
1832 * Item specification.
1833 * @param[in] item_flags
1834 * Bit-fields that holds the items detected until now.
1835 * @param[in] ext_vlan_sup
1836 * Whether extended VLAN features are supported or not.
1838 * Pointer to error structure.
1841 * 0 on success, a negative errno value otherwise and rte_errno is set.
1844 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1845 uint64_t item_flags,
1846 uint8_t target_protocol,
1847 struct rte_flow_error *error)
1849 const struct rte_flow_item_icmp6 *mask = item->mask;
1850 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1851 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1852 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1853 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1854 MLX5_FLOW_LAYER_OUTER_L4;
1857 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1858 return rte_flow_error_set(error, EINVAL,
1859 RTE_FLOW_ERROR_TYPE_ITEM, item,
1860 "protocol filtering not compatible"
1861 " with ICMP6 layer");
1862 if (!(item_flags & l3m))
1863 return rte_flow_error_set(error, EINVAL,
1864 RTE_FLOW_ERROR_TYPE_ITEM, item,
1865 "IPv6 is mandatory to filter on"
1867 if (item_flags & l4m)
1868 return rte_flow_error_set(error, EINVAL,
1869 RTE_FLOW_ERROR_TYPE_ITEM, item,
1870 "multiple L4 layers not supported");
1872 mask = &rte_flow_item_icmp6_mask;
1873 ret = mlx5_flow_item_acceptable
1874 (item, (const uint8_t *)mask,
1875 (const uint8_t *)&rte_flow_item_icmp6_mask,
1876 sizeof(struct rte_flow_item_icmp6),
1877 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1884 * Validate ICMP item.
1887 * Item specification.
1888 * @param[in] item_flags
1889 * Bit-fields that holds the items detected until now.
1891 * Pointer to error structure.
1894 * 0 on success, a negative errno value otherwise and rte_errno is set.
1897 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1898 uint64_t item_flags,
1899 uint8_t target_protocol,
1900 struct rte_flow_error *error)
1902 const struct rte_flow_item_icmp *mask = item->mask;
1903 const struct rte_flow_item_icmp nic_mask = {
1904 .hdr.icmp_type = 0xff,
1905 .hdr.icmp_code = 0xff,
1906 .hdr.icmp_ident = RTE_BE16(0xffff),
1907 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1909 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1910 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1911 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1912 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1913 MLX5_FLOW_LAYER_OUTER_L4;
1916 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1917 return rte_flow_error_set(error, EINVAL,
1918 RTE_FLOW_ERROR_TYPE_ITEM, item,
1919 "protocol filtering not compatible"
1920 " with ICMP layer");
1921 if (!(item_flags & l3m))
1922 return rte_flow_error_set(error, EINVAL,
1923 RTE_FLOW_ERROR_TYPE_ITEM, item,
1924 "IPv4 is mandatory to filter"
1926 if (item_flags & l4m)
1927 return rte_flow_error_set(error, EINVAL,
1928 RTE_FLOW_ERROR_TYPE_ITEM, item,
1929 "multiple L4 layers not supported");
1932 ret = mlx5_flow_item_acceptable
1933 (item, (const uint8_t *)mask,
1934 (const uint8_t *)&nic_mask,
1935 sizeof(struct rte_flow_item_icmp),
1936 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1943 * Validate Ethernet item.
1946 * Item specification.
1947 * @param[in] item_flags
1948 * Bit-fields that holds the items detected until now.
1950 * Pointer to error structure.
1953 * 0 on success, a negative errno value otherwise and rte_errno is set.
1956 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1957 uint64_t item_flags, bool ext_vlan_sup,
1958 struct rte_flow_error *error)
1960 const struct rte_flow_item_eth *mask = item->mask;
1961 const struct rte_flow_item_eth nic_mask = {
1962 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1963 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1964 .type = RTE_BE16(0xffff),
1965 .has_vlan = ext_vlan_sup ? 1 : 0,
1968 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1969 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1970 MLX5_FLOW_LAYER_OUTER_L2;
1972 if (item_flags & ethm)
1973 return rte_flow_error_set(error, ENOTSUP,
1974 RTE_FLOW_ERROR_TYPE_ITEM, item,
1975 "multiple L2 layers not supported");
1976 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1977 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1978 return rte_flow_error_set(error, EINVAL,
1979 RTE_FLOW_ERROR_TYPE_ITEM, item,
1980 "L2 layer should not follow "
1982 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1983 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1984 return rte_flow_error_set(error, EINVAL,
1985 RTE_FLOW_ERROR_TYPE_ITEM, item,
1986 "L2 layer should not follow VLAN");
1988 mask = &rte_flow_item_eth_mask;
1989 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1990 (const uint8_t *)&nic_mask,
1991 sizeof(struct rte_flow_item_eth),
1992 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1997 * Validate VLAN item.
2000 * Item specification.
2001 * @param[in] item_flags
2002 * Bit-fields that holds the items detected until now.
2004 * Ethernet device flow is being created on.
2006 * Pointer to error structure.
2009 * 0 on success, a negative errno value otherwise and rte_errno is set.
2012 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2013 uint64_t item_flags,
2014 struct rte_eth_dev *dev,
2015 struct rte_flow_error *error)
2017 const struct rte_flow_item_vlan *spec = item->spec;
2018 const struct rte_flow_item_vlan *mask = item->mask;
2019 const struct rte_flow_item_vlan nic_mask = {
2020 .tci = RTE_BE16(UINT16_MAX),
2021 .inner_type = RTE_BE16(UINT16_MAX),
2023 uint16_t vlan_tag = 0;
2024 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2026 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2027 MLX5_FLOW_LAYER_INNER_L4) :
2028 (MLX5_FLOW_LAYER_OUTER_L3 |
2029 MLX5_FLOW_LAYER_OUTER_L4);
2030 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2031 MLX5_FLOW_LAYER_OUTER_VLAN;
2033 if (item_flags & vlanm)
2034 return rte_flow_error_set(error, EINVAL,
2035 RTE_FLOW_ERROR_TYPE_ITEM, item,
2036 "multiple VLAN layers not supported");
2037 else if ((item_flags & l34m) != 0)
2038 return rte_flow_error_set(error, EINVAL,
2039 RTE_FLOW_ERROR_TYPE_ITEM, item,
2040 "VLAN cannot follow L3/L4 layer");
2042 mask = &rte_flow_item_vlan_mask;
2043 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2044 (const uint8_t *)&nic_mask,
2045 sizeof(struct rte_flow_item_vlan),
2046 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2049 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2050 struct mlx5_priv *priv = dev->data->dev_private;
2052 if (priv->vmwa_context) {
2054 * Non-NULL context means we have a virtual machine
2055 * and SR-IOV enabled, we have to create VLAN interface
2056 * to make hypervisor to setup E-Switch vport
2057 * context correctly. We avoid creating the multiple
2058 * VLAN interfaces, so we cannot support VLAN tag mask.
2060 return rte_flow_error_set(error, EINVAL,
2061 RTE_FLOW_ERROR_TYPE_ITEM,
2063 "VLAN tag mask is not"
2064 " supported in virtual"
2069 vlan_tag = spec->tci;
2070 vlan_tag &= mask->tci;
2073 * From verbs perspective an empty VLAN is equivalent
2074 * to a packet without VLAN layer.
2077 return rte_flow_error_set(error, EINVAL,
2078 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2080 "VLAN cannot be empty");
2085 * Validate IPV4 item.
2088 * Item specification.
2089 * @param[in] item_flags
2090 * Bit-fields that holds the items detected until now.
2091 * @param[in] last_item
2092 * Previous validated item in the pattern items.
2093 * @param[in] ether_type
2094 * Type in the ethernet layer header (including dot1q).
2095 * @param[in] acc_mask
2096 * Acceptable mask, if NULL default internal default mask
2097 * will be used to check whether item fields are supported.
2098 * @param[in] range_accepted
2099 * True if range of values is accepted for specific fields, false otherwise.
2101 * Pointer to error structure.
2104 * 0 on success, a negative errno value otherwise and rte_errno is set.
2107 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2108 uint64_t item_flags,
2110 uint16_t ether_type,
2111 const struct rte_flow_item_ipv4 *acc_mask,
2112 bool range_accepted,
2113 struct rte_flow_error *error)
2115 const struct rte_flow_item_ipv4 *mask = item->mask;
2116 const struct rte_flow_item_ipv4 *spec = item->spec;
2117 const struct rte_flow_item_ipv4 nic_mask = {
2119 .src_addr = RTE_BE32(0xffffffff),
2120 .dst_addr = RTE_BE32(0xffffffff),
2121 .type_of_service = 0xff,
2122 .next_proto_id = 0xff,
2125 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2126 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2127 MLX5_FLOW_LAYER_OUTER_L3;
2128 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2129 MLX5_FLOW_LAYER_OUTER_L4;
2131 uint8_t next_proto = 0xFF;
2132 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2133 MLX5_FLOW_LAYER_OUTER_VLAN |
2134 MLX5_FLOW_LAYER_INNER_VLAN);
2136 if ((last_item & l2_vlan) && ether_type &&
2137 ether_type != RTE_ETHER_TYPE_IPV4)
2138 return rte_flow_error_set(error, EINVAL,
2139 RTE_FLOW_ERROR_TYPE_ITEM, item,
2140 "IPv4 cannot follow L2/VLAN layer "
2141 "which ether type is not IPv4");
2142 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2144 next_proto = mask->hdr.next_proto_id &
2145 spec->hdr.next_proto_id;
2146 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2147 return rte_flow_error_set(error, EINVAL,
2148 RTE_FLOW_ERROR_TYPE_ITEM,
2153 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2154 return rte_flow_error_set(error, EINVAL,
2155 RTE_FLOW_ERROR_TYPE_ITEM, item,
2156 "wrong tunnel type - IPv6 specified "
2157 "but IPv4 item provided");
2158 if (item_flags & l3m)
2159 return rte_flow_error_set(error, ENOTSUP,
2160 RTE_FLOW_ERROR_TYPE_ITEM, item,
2161 "multiple L3 layers not supported");
2162 else if (item_flags & l4m)
2163 return rte_flow_error_set(error, EINVAL,
2164 RTE_FLOW_ERROR_TYPE_ITEM, item,
2165 "L3 cannot follow an L4 layer.");
2166 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2167 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2168 return rte_flow_error_set(error, EINVAL,
2169 RTE_FLOW_ERROR_TYPE_ITEM, item,
2170 "L3 cannot follow an NVGRE layer.");
2172 mask = &rte_flow_item_ipv4_mask;
2173 else if (mask->hdr.next_proto_id != 0 &&
2174 mask->hdr.next_proto_id != 0xff)
2175 return rte_flow_error_set(error, EINVAL,
2176 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2177 "partial mask is not supported"
2179 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2180 acc_mask ? (const uint8_t *)acc_mask
2181 : (const uint8_t *)&nic_mask,
2182 sizeof(struct rte_flow_item_ipv4),
2183 range_accepted, error);
2190 * Validate IPV6 item.
2193 * Item specification.
2194 * @param[in] item_flags
2195 * Bit-fields that holds the items detected until now.
2196 * @param[in] last_item
2197 * Previous validated item in the pattern items.
2198 * @param[in] ether_type
2199 * Type in the ethernet layer header (including dot1q).
2200 * @param[in] acc_mask
2201 * Acceptable mask, if NULL default internal default mask
2202 * will be used to check whether item fields are supported.
2204 * Pointer to error structure.
2207 * 0 on success, a negative errno value otherwise and rte_errno is set.
2210 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2211 uint64_t item_flags,
2213 uint16_t ether_type,
2214 const struct rte_flow_item_ipv6 *acc_mask,
2215 struct rte_flow_error *error)
2217 const struct rte_flow_item_ipv6 *mask = item->mask;
2218 const struct rte_flow_item_ipv6 *spec = item->spec;
2219 const struct rte_flow_item_ipv6 nic_mask = {
2222 "\xff\xff\xff\xff\xff\xff\xff\xff"
2223 "\xff\xff\xff\xff\xff\xff\xff\xff",
2225 "\xff\xff\xff\xff\xff\xff\xff\xff"
2226 "\xff\xff\xff\xff\xff\xff\xff\xff",
2227 .vtc_flow = RTE_BE32(0xffffffff),
2231 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2232 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2233 MLX5_FLOW_LAYER_OUTER_L3;
2234 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2235 MLX5_FLOW_LAYER_OUTER_L4;
2237 uint8_t next_proto = 0xFF;
2238 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2239 MLX5_FLOW_LAYER_OUTER_VLAN |
2240 MLX5_FLOW_LAYER_INNER_VLAN);
2242 if ((last_item & l2_vlan) && ether_type &&
2243 ether_type != RTE_ETHER_TYPE_IPV6)
2244 return rte_flow_error_set(error, EINVAL,
2245 RTE_FLOW_ERROR_TYPE_ITEM, item,
2246 "IPv6 cannot follow L2/VLAN layer "
2247 "which ether type is not IPv6");
2248 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2249 next_proto = spec->hdr.proto;
2250 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2251 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2252 return rte_flow_error_set(error, EINVAL,
2253 RTE_FLOW_ERROR_TYPE_ITEM,
2258 if (next_proto == IPPROTO_HOPOPTS ||
2259 next_proto == IPPROTO_ROUTING ||
2260 next_proto == IPPROTO_FRAGMENT ||
2261 next_proto == IPPROTO_ESP ||
2262 next_proto == IPPROTO_AH ||
2263 next_proto == IPPROTO_DSTOPTS)
2264 return rte_flow_error_set(error, EINVAL,
2265 RTE_FLOW_ERROR_TYPE_ITEM, item,
2266 "IPv6 proto (next header) should "
2267 "not be set as extension header");
2268 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2269 return rte_flow_error_set(error, EINVAL,
2270 RTE_FLOW_ERROR_TYPE_ITEM, item,
2271 "wrong tunnel type - IPv4 specified "
2272 "but IPv6 item provided");
2273 if (item_flags & l3m)
2274 return rte_flow_error_set(error, ENOTSUP,
2275 RTE_FLOW_ERROR_TYPE_ITEM, item,
2276 "multiple L3 layers not supported");
2277 else if (item_flags & l4m)
2278 return rte_flow_error_set(error, EINVAL,
2279 RTE_FLOW_ERROR_TYPE_ITEM, item,
2280 "L3 cannot follow an L4 layer.");
2281 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2282 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2283 return rte_flow_error_set(error, EINVAL,
2284 RTE_FLOW_ERROR_TYPE_ITEM, item,
2285 "L3 cannot follow an NVGRE layer.");
2287 mask = &rte_flow_item_ipv6_mask;
2288 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2289 acc_mask ? (const uint8_t *)acc_mask
2290 : (const uint8_t *)&nic_mask,
2291 sizeof(struct rte_flow_item_ipv6),
2292 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2299 * Validate UDP item.
2302 * Item specification.
2303 * @param[in] item_flags
2304 * Bit-fields that holds the items detected until now.
2305 * @param[in] target_protocol
2306 * The next protocol in the previous item.
2307 * @param[in] flow_mask
2308 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2310 * Pointer to error structure.
2313 * 0 on success, a negative errno value otherwise and rte_errno is set.
2316 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2317 uint64_t item_flags,
2318 uint8_t target_protocol,
2319 struct rte_flow_error *error)
2321 const struct rte_flow_item_udp *mask = item->mask;
2322 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2323 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2324 MLX5_FLOW_LAYER_OUTER_L3;
2325 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2326 MLX5_FLOW_LAYER_OUTER_L4;
2329 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2330 return rte_flow_error_set(error, EINVAL,
2331 RTE_FLOW_ERROR_TYPE_ITEM, item,
2332 "protocol filtering not compatible"
2334 if (!(item_flags & l3m))
2335 return rte_flow_error_set(error, EINVAL,
2336 RTE_FLOW_ERROR_TYPE_ITEM, item,
2337 "L3 is mandatory to filter on L4");
2338 if (item_flags & l4m)
2339 return rte_flow_error_set(error, EINVAL,
2340 RTE_FLOW_ERROR_TYPE_ITEM, item,
2341 "multiple L4 layers not supported");
2343 mask = &rte_flow_item_udp_mask;
2344 ret = mlx5_flow_item_acceptable
2345 (item, (const uint8_t *)mask,
2346 (const uint8_t *)&rte_flow_item_udp_mask,
2347 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2355 * Validate TCP item.
2358 * Item specification.
2359 * @param[in] item_flags
2360 * Bit-fields that holds the items detected until now.
2361 * @param[in] target_protocol
2362 * The next protocol in the previous item.
2364 * Pointer to error structure.
2367 * 0 on success, a negative errno value otherwise and rte_errno is set.
2370 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2371 uint64_t item_flags,
2372 uint8_t target_protocol,
2373 const struct rte_flow_item_tcp *flow_mask,
2374 struct rte_flow_error *error)
2376 const struct rte_flow_item_tcp *mask = item->mask;
2377 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2378 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2379 MLX5_FLOW_LAYER_OUTER_L3;
2380 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2381 MLX5_FLOW_LAYER_OUTER_L4;
2384 MLX5_ASSERT(flow_mask);
2385 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2386 return rte_flow_error_set(error, EINVAL,
2387 RTE_FLOW_ERROR_TYPE_ITEM, item,
2388 "protocol filtering not compatible"
2390 if (!(item_flags & l3m))
2391 return rte_flow_error_set(error, EINVAL,
2392 RTE_FLOW_ERROR_TYPE_ITEM, item,
2393 "L3 is mandatory to filter on L4");
2394 if (item_flags & l4m)
2395 return rte_flow_error_set(error, EINVAL,
2396 RTE_FLOW_ERROR_TYPE_ITEM, item,
2397 "multiple L4 layers not supported");
2399 mask = &rte_flow_item_tcp_mask;
2400 ret = mlx5_flow_item_acceptable
2401 (item, (const uint8_t *)mask,
2402 (const uint8_t *)flow_mask,
2403 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2411 * Validate VXLAN item.
2414 * Pointer to the Ethernet device structure.
2416 * Item specification.
2417 * @param[in] item_flags
2418 * Bit-fields that holds the items detected until now.
2420 * Flow rule attributes.
2422 * Pointer to error structure.
2425 * 0 on success, a negative errno value otherwise and rte_errno is set.
2428 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2429 const struct rte_flow_item *item,
2430 uint64_t item_flags,
2431 const struct rte_flow_attr *attr,
2432 struct rte_flow_error *error)
2434 const struct rte_flow_item_vxlan *spec = item->spec;
2435 const struct rte_flow_item_vxlan *mask = item->mask;
2437 struct mlx5_priv *priv = dev->data->dev_private;
2441 } id = { .vlan_id = 0, };
2442 const struct rte_flow_item_vxlan nic_mask = {
2443 .vni = "\xff\xff\xff",
2446 const struct rte_flow_item_vxlan *valid_mask;
2448 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2449 return rte_flow_error_set(error, ENOTSUP,
2450 RTE_FLOW_ERROR_TYPE_ITEM, item,
2451 "multiple tunnel layers not"
2453 valid_mask = &rte_flow_item_vxlan_mask;
2455 * Verify only UDPv4 is present as defined in
2456 * https://tools.ietf.org/html/rfc7348
2458 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2459 return rte_flow_error_set(error, EINVAL,
2460 RTE_FLOW_ERROR_TYPE_ITEM, item,
2461 "no outer UDP layer found");
2463 mask = &rte_flow_item_vxlan_mask;
2464 /* FDB domain & NIC domain non-zero group */
2465 if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2466 valid_mask = &nic_mask;
2467 /* Group zero in NIC domain */
2468 if (!attr->group && !attr->transfer && priv->sh->tunnel_header_0_1)
2469 valid_mask = &nic_mask;
2470 ret = mlx5_flow_item_acceptable
2471 (item, (const uint8_t *)mask,
2472 (const uint8_t *)valid_mask,
2473 sizeof(struct rte_flow_item_vxlan),
2474 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2478 memcpy(&id.vni[1], spec->vni, 3);
2479 memcpy(&id.vni[1], mask->vni, 3);
2481 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2482 return rte_flow_error_set(error, ENOTSUP,
2483 RTE_FLOW_ERROR_TYPE_ITEM, item,
2484 "VXLAN tunnel must be fully defined");
2489 * Validate VXLAN_GPE item.
2492 * Item specification.
2493 * @param[in] item_flags
2494 * Bit-fields that holds the items detected until now.
2496 * Pointer to the private data structure.
2497 * @param[in] target_protocol
2498 * The next protocol in the previous item.
2500 * Pointer to error structure.
2503 * 0 on success, a negative errno value otherwise and rte_errno is set.
2506 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2507 uint64_t item_flags,
2508 struct rte_eth_dev *dev,
2509 struct rte_flow_error *error)
2511 struct mlx5_priv *priv = dev->data->dev_private;
2512 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2513 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2518 } id = { .vlan_id = 0, };
2520 if (!priv->config.l3_vxlan_en)
2521 return rte_flow_error_set(error, ENOTSUP,
2522 RTE_FLOW_ERROR_TYPE_ITEM, item,
2523 "L3 VXLAN is not enabled by device"
2524 " parameter and/or not configured in"
2526 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2527 return rte_flow_error_set(error, ENOTSUP,
2528 RTE_FLOW_ERROR_TYPE_ITEM, item,
2529 "multiple tunnel layers not"
2532 * Verify only UDPv4 is present as defined in
2533 * https://tools.ietf.org/html/rfc7348
2535 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2536 return rte_flow_error_set(error, EINVAL,
2537 RTE_FLOW_ERROR_TYPE_ITEM, item,
2538 "no outer UDP layer found");
2540 mask = &rte_flow_item_vxlan_gpe_mask;
2541 ret = mlx5_flow_item_acceptable
2542 (item, (const uint8_t *)mask,
2543 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2544 sizeof(struct rte_flow_item_vxlan_gpe),
2545 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2550 return rte_flow_error_set(error, ENOTSUP,
2551 RTE_FLOW_ERROR_TYPE_ITEM,
2553 "VxLAN-GPE protocol"
2555 memcpy(&id.vni[1], spec->vni, 3);
2556 memcpy(&id.vni[1], mask->vni, 3);
2558 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2559 return rte_flow_error_set(error, ENOTSUP,
2560 RTE_FLOW_ERROR_TYPE_ITEM, item,
2561 "VXLAN-GPE tunnel must be fully"
2566 * Validate GRE Key item.
2569 * Item specification.
2570 * @param[in] item_flags
2571 * Bit flags to mark detected items.
2572 * @param[in] gre_item
2573 * Pointer to gre_item
2575 * Pointer to error structure.
2578 * 0 on success, a negative errno value otherwise and rte_errno is set.
2581 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2582 uint64_t item_flags,
2583 const struct rte_flow_item *gre_item,
2584 struct rte_flow_error *error)
2586 const rte_be32_t *mask = item->mask;
2588 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2589 const struct rte_flow_item_gre *gre_spec;
2590 const struct rte_flow_item_gre *gre_mask;
2592 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2593 return rte_flow_error_set(error, ENOTSUP,
2594 RTE_FLOW_ERROR_TYPE_ITEM, item,
2595 "Multiple GRE key not support");
2596 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2597 return rte_flow_error_set(error, ENOTSUP,
2598 RTE_FLOW_ERROR_TYPE_ITEM, item,
2599 "No preceding GRE header");
2600 if (item_flags & MLX5_FLOW_LAYER_INNER)
2601 return rte_flow_error_set(error, ENOTSUP,
2602 RTE_FLOW_ERROR_TYPE_ITEM, item,
2603 "GRE key following a wrong item");
2604 gre_mask = gre_item->mask;
2606 gre_mask = &rte_flow_item_gre_mask;
2607 gre_spec = gre_item->spec;
2608 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2609 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2610 return rte_flow_error_set(error, EINVAL,
2611 RTE_FLOW_ERROR_TYPE_ITEM, item,
2612 "Key bit must be on");
2615 mask = &gre_key_default_mask;
2616 ret = mlx5_flow_item_acceptable
2617 (item, (const uint8_t *)mask,
2618 (const uint8_t *)&gre_key_default_mask,
2619 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2624 * Validate GRE item.
2627 * Item specification.
2628 * @param[in] item_flags
2629 * Bit flags to mark detected items.
2630 * @param[in] target_protocol
2631 * The next protocol in the previous item.
2633 * Pointer to error structure.
2636 * 0 on success, a negative errno value otherwise and rte_errno is set.
2639 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2640 uint64_t item_flags,
2641 uint8_t target_protocol,
2642 struct rte_flow_error *error)
2644 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2645 const struct rte_flow_item_gre *mask = item->mask;
2647 const struct rte_flow_item_gre nic_mask = {
2648 .c_rsvd0_ver = RTE_BE16(0xB000),
2649 .protocol = RTE_BE16(UINT16_MAX),
2652 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2653 return rte_flow_error_set(error, EINVAL,
2654 RTE_FLOW_ERROR_TYPE_ITEM, item,
2655 "protocol filtering not compatible"
2656 " with this GRE layer");
2657 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2658 return rte_flow_error_set(error, ENOTSUP,
2659 RTE_FLOW_ERROR_TYPE_ITEM, item,
2660 "multiple tunnel layers not"
2662 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2663 return rte_flow_error_set(error, ENOTSUP,
2664 RTE_FLOW_ERROR_TYPE_ITEM, item,
2665 "L3 Layer is missing");
2667 mask = &rte_flow_item_gre_mask;
2668 ret = mlx5_flow_item_acceptable
2669 (item, (const uint8_t *)mask,
2670 (const uint8_t *)&nic_mask,
2671 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2675 #ifndef HAVE_MLX5DV_DR
2676 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2677 if (spec && (spec->protocol & mask->protocol))
2678 return rte_flow_error_set(error, ENOTSUP,
2679 RTE_FLOW_ERROR_TYPE_ITEM, item,
2680 "without MPLS support the"
2681 " specification cannot be used for"
2689 * Validate Geneve item.
2692 * Item specification.
2693 * @param[in] itemFlags
2694 * Bit-fields that holds the items detected until now.
2696 * Pointer to the private data structure.
2698 * Pointer to error structure.
2701 * 0 on success, a negative errno value otherwise and rte_errno is set.
2705 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2706 uint64_t item_flags,
2707 struct rte_eth_dev *dev,
2708 struct rte_flow_error *error)
2710 struct mlx5_priv *priv = dev->data->dev_private;
2711 const struct rte_flow_item_geneve *spec = item->spec;
2712 const struct rte_flow_item_geneve *mask = item->mask;
2715 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2716 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2717 const struct rte_flow_item_geneve nic_mask = {
2718 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2719 .vni = "\xff\xff\xff",
2720 .protocol = RTE_BE16(UINT16_MAX),
2723 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2724 return rte_flow_error_set(error, ENOTSUP,
2725 RTE_FLOW_ERROR_TYPE_ITEM, item,
2726 "L3 Geneve is not enabled by device"
2727 " parameter and/or not configured in"
2729 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2730 return rte_flow_error_set(error, ENOTSUP,
2731 RTE_FLOW_ERROR_TYPE_ITEM, item,
2732 "multiple tunnel layers not"
2735 * Verify only UDPv4 is present as defined in
2736 * https://tools.ietf.org/html/rfc7348
2738 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2739 return rte_flow_error_set(error, EINVAL,
2740 RTE_FLOW_ERROR_TYPE_ITEM, item,
2741 "no outer UDP layer found");
2743 mask = &rte_flow_item_geneve_mask;
2744 ret = mlx5_flow_item_acceptable
2745 (item, (const uint8_t *)mask,
2746 (const uint8_t *)&nic_mask,
2747 sizeof(struct rte_flow_item_geneve),
2748 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2752 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2753 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2754 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2755 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2756 return rte_flow_error_set(error, ENOTSUP,
2757 RTE_FLOW_ERROR_TYPE_ITEM,
2759 "Geneve protocol unsupported"
2760 " fields are being used");
2761 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2762 return rte_flow_error_set
2764 RTE_FLOW_ERROR_TYPE_ITEM,
2766 "Unsupported Geneve options length");
2768 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2769 return rte_flow_error_set
2771 RTE_FLOW_ERROR_TYPE_ITEM, item,
2772 "Geneve tunnel must be fully defined");
2777 * Validate Geneve TLV option item.
2780 * Item specification.
2781 * @param[in] last_item
2782 * Previous validated item in the pattern items.
2783 * @param[in] geneve_item
2784 * Previous GENEVE item specification.
2786 * Pointer to the rte_eth_dev structure.
2788 * Pointer to error structure.
2791 * 0 on success, a negative errno value otherwise and rte_errno is set.
2794 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2796 const struct rte_flow_item *geneve_item,
2797 struct rte_eth_dev *dev,
2798 struct rte_flow_error *error)
2800 struct mlx5_priv *priv = dev->data->dev_private;
2801 struct mlx5_dev_ctx_shared *sh = priv->sh;
2802 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2803 struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2804 uint8_t data_max_supported =
2805 hca_attr->max_geneve_tlv_option_data_len * 4;
2806 struct mlx5_dev_config *config = &priv->config;
2807 const struct rte_flow_item_geneve *geneve_spec;
2808 const struct rte_flow_item_geneve *geneve_mask;
2809 const struct rte_flow_item_geneve_opt *spec = item->spec;
2810 const struct rte_flow_item_geneve_opt *mask = item->mask;
2812 unsigned int data_len;
2813 uint8_t tlv_option_len;
2814 uint16_t optlen_m, optlen_v;
2815 const struct rte_flow_item_geneve_opt full_mask = {
2816 .option_class = RTE_BE16(0xffff),
2817 .option_type = 0xff,
2822 mask = &rte_flow_item_geneve_opt_mask;
2824 return rte_flow_error_set
2825 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2826 "Geneve TLV opt class/type/length must be specified");
2827 if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2828 return rte_flow_error_set
2829 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2830 "Geneve TLV opt length exceeeds the limit (31)");
2831 /* Check if class type and length masks are full. */
2832 if (full_mask.option_class != mask->option_class ||
2833 full_mask.option_type != mask->option_type ||
2834 full_mask.option_len != (mask->option_len & full_mask.option_len))
2835 return rte_flow_error_set
2836 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2837 "Geneve TLV opt class/type/length masks must be full");
2838 /* Check if length is supported */
2839 if ((uint32_t)spec->option_len >
2840 config->hca_attr.max_geneve_tlv_option_data_len)
2841 return rte_flow_error_set
2842 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2843 "Geneve TLV opt length not supported");
2844 if (config->hca_attr.max_geneve_tlv_options > 1)
2846 "max_geneve_tlv_options supports more than 1 option");
2847 /* Check GENEVE item preceding. */
2848 if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2849 return rte_flow_error_set
2850 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2851 "Geneve opt item must be preceded with Geneve item");
2852 geneve_spec = geneve_item->spec;
2853 geneve_mask = geneve_item->mask ? geneve_item->mask :
2854 &rte_flow_item_geneve_mask;
2855 /* Check if GENEVE TLV option size doesn't exceed option length */
2856 if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2857 geneve_spec->ver_opt_len_o_c_rsvd0)) {
2858 tlv_option_len = spec->option_len & mask->option_len;
2859 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2860 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2861 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2862 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2863 if ((optlen_v & optlen_m) <= tlv_option_len)
2864 return rte_flow_error_set
2865 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2866 "GENEVE TLV option length exceeds optlen");
2868 /* Check if length is 0 or data is 0. */
2869 if (spec->data == NULL || spec->option_len == 0)
2870 return rte_flow_error_set
2871 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2872 "Geneve TLV opt with zero data/length not supported");
2873 /* Check not all data & mask are 0. */
2874 data_len = spec->option_len * 4;
2875 if (mask->data == NULL) {
2876 for (i = 0; i < data_len; i++)
2880 return rte_flow_error_set(error, ENOTSUP,
2881 RTE_FLOW_ERROR_TYPE_ITEM, item,
2882 "Can't match on Geneve option data 0");
2884 for (i = 0; i < data_len; i++)
2885 if (spec->data[i] & mask->data[i])
2888 return rte_flow_error_set(error, ENOTSUP,
2889 RTE_FLOW_ERROR_TYPE_ITEM, item,
2890 "Can't match on Geneve option data and mask 0");
2891 /* Check data mask supported. */
2892 for (i = data_max_supported; i < data_len ; i++)
2894 return rte_flow_error_set(error, ENOTSUP,
2895 RTE_FLOW_ERROR_TYPE_ITEM, item,
2896 "Data mask is of unsupported size");
2898 /* Check GENEVE option is supported in NIC. */
2899 if (!config->hca_attr.geneve_tlv_opt)
2900 return rte_flow_error_set
2901 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2902 "Geneve TLV opt not supported");
2903 /* Check if we already have geneve option with different type/class. */
2904 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2905 geneve_opt_resource = sh->geneve_tlv_option_resource;
2906 if (geneve_opt_resource != NULL)
2907 if (geneve_opt_resource->option_class != spec->option_class ||
2908 geneve_opt_resource->option_type != spec->option_type ||
2909 geneve_opt_resource->length != spec->option_len) {
2910 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2911 return rte_flow_error_set(error, ENOTSUP,
2912 RTE_FLOW_ERROR_TYPE_ITEM, item,
2913 "Only one Geneve TLV option supported");
2915 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2920 * Validate MPLS item.
2923 * Pointer to the rte_eth_dev structure.
2925 * Item specification.
2926 * @param[in] item_flags
2927 * Bit-fields that holds the items detected until now.
2928 * @param[in] prev_layer
2929 * The protocol layer indicated in previous item.
2931 * Pointer to error structure.
2934 * 0 on success, a negative errno value otherwise and rte_errno is set.
2937 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2938 const struct rte_flow_item *item __rte_unused,
2939 uint64_t item_flags __rte_unused,
2940 uint64_t prev_layer __rte_unused,
2941 struct rte_flow_error *error)
2943 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2944 const struct rte_flow_item_mpls *mask = item->mask;
2945 struct mlx5_priv *priv = dev->data->dev_private;
2948 if (!priv->config.mpls_en)
2949 return rte_flow_error_set(error, ENOTSUP,
2950 RTE_FLOW_ERROR_TYPE_ITEM, item,
2951 "MPLS not supported or"
2952 " disabled in firmware"
2954 /* MPLS over UDP, GRE is allowed */
2955 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
2956 MLX5_FLOW_LAYER_GRE |
2957 MLX5_FLOW_LAYER_GRE_KEY)))
2958 return rte_flow_error_set(error, EINVAL,
2959 RTE_FLOW_ERROR_TYPE_ITEM, item,
2960 "protocol filtering not compatible"
2961 " with MPLS layer");
2962 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2963 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2964 !(item_flags & MLX5_FLOW_LAYER_GRE))
2965 return rte_flow_error_set(error, ENOTSUP,
2966 RTE_FLOW_ERROR_TYPE_ITEM, item,
2967 "multiple tunnel layers not"
2970 mask = &rte_flow_item_mpls_mask;
2971 ret = mlx5_flow_item_acceptable
2972 (item, (const uint8_t *)mask,
2973 (const uint8_t *)&rte_flow_item_mpls_mask,
2974 sizeof(struct rte_flow_item_mpls),
2975 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2980 return rte_flow_error_set(error, ENOTSUP,
2981 RTE_FLOW_ERROR_TYPE_ITEM, item,
2982 "MPLS is not supported by Verbs, please"
2988 * Validate NVGRE item.
2991 * Item specification.
2992 * @param[in] item_flags
2993 * Bit flags to mark detected items.
2994 * @param[in] target_protocol
2995 * The next protocol in the previous item.
2997 * Pointer to error structure.
3000 * 0 on success, a negative errno value otherwise and rte_errno is set.
3003 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3004 uint64_t item_flags,
3005 uint8_t target_protocol,
3006 struct rte_flow_error *error)
3008 const struct rte_flow_item_nvgre *mask = item->mask;
3011 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3012 return rte_flow_error_set(error, EINVAL,
3013 RTE_FLOW_ERROR_TYPE_ITEM, item,
3014 "protocol filtering not compatible"
3015 " with this GRE layer");
3016 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3017 return rte_flow_error_set(error, ENOTSUP,
3018 RTE_FLOW_ERROR_TYPE_ITEM, item,
3019 "multiple tunnel layers not"
3021 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3022 return rte_flow_error_set(error, ENOTSUP,
3023 RTE_FLOW_ERROR_TYPE_ITEM, item,
3024 "L3 Layer is missing");
3026 mask = &rte_flow_item_nvgre_mask;
3027 ret = mlx5_flow_item_acceptable
3028 (item, (const uint8_t *)mask,
3029 (const uint8_t *)&rte_flow_item_nvgre_mask,
3030 sizeof(struct rte_flow_item_nvgre),
3031 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3038 * Validate eCPRI item.
3041 * Item specification.
3042 * @param[in] item_flags
3043 * Bit-fields that holds the items detected until now.
3044 * @param[in] last_item
3045 * Previous validated item in the pattern items.
3046 * @param[in] ether_type
3047 * Type in the ethernet layer header (including dot1q).
3048 * @param[in] acc_mask
3049 * Acceptable mask, if NULL default internal default mask
3050 * will be used to check whether item fields are supported.
3052 * Pointer to error structure.
3055 * 0 on success, a negative errno value otherwise and rte_errno is set.
3058 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3059 uint64_t item_flags,
3061 uint16_t ether_type,
3062 const struct rte_flow_item_ecpri *acc_mask,
3063 struct rte_flow_error *error)
3065 const struct rte_flow_item_ecpri *mask = item->mask;
3066 const struct rte_flow_item_ecpri nic_mask = {
3070 RTE_BE32(((const struct rte_ecpri_common_hdr) {
3074 .dummy[0] = 0xFFFFFFFF,
3077 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3078 MLX5_FLOW_LAYER_OUTER_VLAN);
3079 struct rte_flow_item_ecpri mask_lo;
3081 if (!(last_item & outer_l2_vlan) &&
3082 last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3083 return rte_flow_error_set(error, EINVAL,
3084 RTE_FLOW_ERROR_TYPE_ITEM, item,
3085 "eCPRI can only follow L2/VLAN layer or UDP layer");
3086 if ((last_item & outer_l2_vlan) && ether_type &&
3087 ether_type != RTE_ETHER_TYPE_ECPRI)
3088 return rte_flow_error_set(error, EINVAL,
3089 RTE_FLOW_ERROR_TYPE_ITEM, item,
3090 "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3091 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3092 return rte_flow_error_set(error, EINVAL,
3093 RTE_FLOW_ERROR_TYPE_ITEM, item,
3094 "eCPRI with tunnel is not supported right now");
3095 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3096 return rte_flow_error_set(error, ENOTSUP,
3097 RTE_FLOW_ERROR_TYPE_ITEM, item,
3098 "multiple L3 layers not supported");
3099 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3100 return rte_flow_error_set(error, EINVAL,
3101 RTE_FLOW_ERROR_TYPE_ITEM, item,
3102 "eCPRI cannot coexist with a TCP layer");
3103 /* In specification, eCPRI could be over UDP layer. */
3104 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3105 return rte_flow_error_set(error, EINVAL,
3106 RTE_FLOW_ERROR_TYPE_ITEM, item,
3107 "eCPRI over UDP layer is not yet supported right now");
3108 /* Mask for type field in common header could be zero. */
3110 mask = &rte_flow_item_ecpri_mask;
3111 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3112 /* Input mask is in big-endian format. */
3113 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3114 return rte_flow_error_set(error, EINVAL,
3115 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3116 "partial mask is not supported for protocol");
3117 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3118 return rte_flow_error_set(error, EINVAL,
3119 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3120 "message header mask must be after a type mask");
3121 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3122 acc_mask ? (const uint8_t *)acc_mask
3123 : (const uint8_t *)&nic_mask,
3124 sizeof(struct rte_flow_item_ecpri),
3125 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3129 * Release resource related QUEUE/RSS action split.
3132 * Pointer to Ethernet device.
3134 * Flow to release id's from.
3137 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3138 struct rte_flow *flow)
3140 struct mlx5_priv *priv = dev->data->dev_private;
3141 uint32_t handle_idx;
3142 struct mlx5_flow_handle *dev_handle;
3144 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3145 handle_idx, dev_handle, next)
3146 if (dev_handle->split_flow_id &&
3147 !dev_handle->is_meter_flow_id)
3148 mlx5_ipool_free(priv->sh->ipool
3149 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3150 dev_handle->split_flow_id);
3154 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3155 const struct rte_flow_attr *attr __rte_unused,
3156 const struct rte_flow_item items[] __rte_unused,
3157 const struct rte_flow_action actions[] __rte_unused,
3158 bool external __rte_unused,
3159 int hairpin __rte_unused,
3160 struct rte_flow_error *error)
3162 return rte_flow_error_set(error, ENOTSUP,
3163 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3166 static struct mlx5_flow *
3167 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3168 const struct rte_flow_attr *attr __rte_unused,
3169 const struct rte_flow_item items[] __rte_unused,
3170 const struct rte_flow_action actions[] __rte_unused,
3171 struct rte_flow_error *error)
3173 rte_flow_error_set(error, ENOTSUP,
3174 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3179 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3180 struct mlx5_flow *dev_flow __rte_unused,
3181 const struct rte_flow_attr *attr __rte_unused,
3182 const struct rte_flow_item items[] __rte_unused,
3183 const struct rte_flow_action actions[] __rte_unused,
3184 struct rte_flow_error *error)
3186 return rte_flow_error_set(error, ENOTSUP,
3187 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3191 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3192 struct rte_flow *flow __rte_unused,
3193 struct rte_flow_error *error)
3195 return rte_flow_error_set(error, ENOTSUP,
3196 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3200 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3201 struct rte_flow *flow __rte_unused)
3206 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3207 struct rte_flow *flow __rte_unused)
3212 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3213 struct rte_flow *flow __rte_unused,
3214 const struct rte_flow_action *actions __rte_unused,
3215 void *data __rte_unused,
3216 struct rte_flow_error *error)
3218 return rte_flow_error_set(error, ENOTSUP,
3219 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3223 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3224 uint32_t domains __rte_unused,
3225 uint32_t flags __rte_unused)
3230 /* Void driver to protect from null pointer reference. */
3231 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3232 .validate = flow_null_validate,
3233 .prepare = flow_null_prepare,
3234 .translate = flow_null_translate,
3235 .apply = flow_null_apply,
3236 .remove = flow_null_remove,
3237 .destroy = flow_null_destroy,
3238 .query = flow_null_query,
3239 .sync_domain = flow_null_sync_domain,
3243 * Select flow driver type according to flow attributes and device
3247 * Pointer to the dev structure.
3249 * Pointer to the flow attributes.
3252 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3254 static enum mlx5_flow_drv_type
3255 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3257 struct mlx5_priv *priv = dev->data->dev_private;
3258 /* The OS can determine first a specific flow type (DV, VERBS) */
3259 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3261 if (type != MLX5_FLOW_TYPE_MAX)
3263 /* If no OS specific type - continue with DV/VERBS selection */
3264 if (attr->transfer && priv->config.dv_esw_en)
3265 type = MLX5_FLOW_TYPE_DV;
3266 if (!attr->transfer)
3267 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3268 MLX5_FLOW_TYPE_VERBS;
3272 #define flow_get_drv_ops(type) flow_drv_ops[type]
3275 * Flow driver validation API. This abstracts calling driver specific functions.
3276 * The type of flow driver is determined according to flow attributes.
3279 * Pointer to the dev structure.
3281 * Pointer to the flow attributes.
3283 * Pointer to the list of items.
3284 * @param[in] actions
3285 * Pointer to the list of actions.
3286 * @param[in] external
3287 * This flow rule is created by request external to PMD.
3288 * @param[in] hairpin
3289 * Number of hairpin TX actions, 0 means classic flow.
3291 * Pointer to the error structure.
3294 * 0 on success, a negative errno value otherwise and rte_errno is set.
3297 flow_drv_validate(struct rte_eth_dev *dev,
3298 const struct rte_flow_attr *attr,
3299 const struct rte_flow_item items[],
3300 const struct rte_flow_action actions[],
3301 bool external, int hairpin, struct rte_flow_error *error)
3303 const struct mlx5_flow_driver_ops *fops;
3304 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3306 fops = flow_get_drv_ops(type);
3307 return fops->validate(dev, attr, items, actions, external,
3312 * Flow driver preparation API. This abstracts calling driver specific
3313 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3314 * calculates the size of memory required for device flow, allocates the memory,
3315 * initializes the device flow and returns the pointer.
3318 * This function initializes device flow structure such as dv or verbs in
3319 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3320 * rest. For example, adding returning device flow to flow->dev_flow list and
3321 * setting backward reference to the flow should be done out of this function.
3322 * layers field is not filled either.
3325 * Pointer to the dev structure.
3327 * Pointer to the flow attributes.
3329 * Pointer to the list of items.
3330 * @param[in] actions
3331 * Pointer to the list of actions.
3332 * @param[in] flow_idx
3333 * This memory pool index to the flow.
3335 * Pointer to the error structure.
3338 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3340 static inline struct mlx5_flow *
3341 flow_drv_prepare(struct rte_eth_dev *dev,
3342 const struct rte_flow *flow,
3343 const struct rte_flow_attr *attr,
3344 const struct rte_flow_item items[],
3345 const struct rte_flow_action actions[],
3347 struct rte_flow_error *error)
3349 const struct mlx5_flow_driver_ops *fops;
3350 enum mlx5_flow_drv_type type = flow->drv_type;
3351 struct mlx5_flow *mlx5_flow = NULL;
3353 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3354 fops = flow_get_drv_ops(type);
3355 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3357 mlx5_flow->flow_idx = flow_idx;
3362 * Flow driver translation API. This abstracts calling driver specific
3363 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3364 * translates a generic flow into a driver flow. flow_drv_prepare() must
3368 * dev_flow->layers could be filled as a result of parsing during translation
3369 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3370 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3371 * flow->actions could be overwritten even though all the expanded dev_flows
3372 * have the same actions.
3375 * Pointer to the rte dev structure.
3376 * @param[in, out] dev_flow
3377 * Pointer to the mlx5 flow.
3379 * Pointer to the flow attributes.
3381 * Pointer to the list of items.
3382 * @param[in] actions
3383 * Pointer to the list of actions.
3385 * Pointer to the error structure.
3388 * 0 on success, a negative errno value otherwise and rte_errno is set.
3391 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3392 const struct rte_flow_attr *attr,
3393 const struct rte_flow_item items[],
3394 const struct rte_flow_action actions[],
3395 struct rte_flow_error *error)
3397 const struct mlx5_flow_driver_ops *fops;
3398 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3400 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3401 fops = flow_get_drv_ops(type);
3402 return fops->translate(dev, dev_flow, attr, items, actions, error);
3406 * Flow driver apply API. This abstracts calling driver specific functions.
3407 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3408 * translated driver flows on to device. flow_drv_translate() must precede.
3411 * Pointer to Ethernet device structure.
3412 * @param[in, out] flow
3413 * Pointer to flow structure.
3415 * Pointer to error structure.
3418 * 0 on success, a negative errno value otherwise and rte_errno is set.
3421 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3422 struct rte_flow_error *error)
3424 const struct mlx5_flow_driver_ops *fops;
3425 enum mlx5_flow_drv_type type = flow->drv_type;
3427 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3428 fops = flow_get_drv_ops(type);
3429 return fops->apply(dev, flow, error);
3433 * Flow driver destroy API. This abstracts calling driver specific functions.
3434 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3435 * on device and releases resources of the flow.
3438 * Pointer to Ethernet device.
3439 * @param[in, out] flow
3440 * Pointer to flow structure.
3443 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3445 const struct mlx5_flow_driver_ops *fops;
3446 enum mlx5_flow_drv_type type = flow->drv_type;
3448 flow_mreg_split_qrss_release(dev, flow);
3449 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3450 fops = flow_get_drv_ops(type);
3451 fops->destroy(dev, flow);
3455 * Flow driver find RSS policy tbl API. This abstracts calling driver
3456 * specific functions. Parent flow (rte_flow) should have driver
3457 * type (drv_type). It will find the RSS policy table that has the rss_desc.
3460 * Pointer to Ethernet device.
3461 * @param[in, out] flow
3462 * Pointer to flow structure.
3464 * Pointer to meter policy table.
3465 * @param[in] rss_desc
3466 * Pointer to rss_desc
3468 static struct mlx5_flow_meter_sub_policy *
3469 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3470 struct rte_flow *flow,
3471 struct mlx5_flow_meter_policy *policy,
3472 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3474 const struct mlx5_flow_driver_ops *fops;
3475 enum mlx5_flow_drv_type type = flow->drv_type;
3477 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3478 fops = flow_get_drv_ops(type);
3479 return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3483 * Flow driver color tag rule API. This abstracts calling driver
3484 * specific functions. Parent flow (rte_flow) should have driver
3485 * type (drv_type). It will create the color tag rules in hierarchy meter.
3488 * Pointer to Ethernet device.
3489 * @param[in, out] flow
3490 * Pointer to flow structure.
3492 * Pointer to flow meter structure.
3493 * @param[in] src_port
3494 * The src port this extra rule should use.
3496 * The src port id match item.
3498 * Pointer to error structure.
3501 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3502 struct rte_flow *flow,
3503 struct mlx5_flow_meter_info *fm,
3505 const struct rte_flow_item *item,
3506 struct rte_flow_error *error)
3508 const struct mlx5_flow_driver_ops *fops;
3509 enum mlx5_flow_drv_type type = flow->drv_type;
3511 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3512 fops = flow_get_drv_ops(type);
3513 return fops->meter_hierarchy_rule_create(dev, fm,
3514 src_port, item, error);
3518 * Get RSS action from the action list.
3521 * Pointer to Ethernet device.
3522 * @param[in] actions
3523 * Pointer to the list of actions.
3525 * Parent flow structure pointer.
3528 * Pointer to the RSS action if exist, else return NULL.
3530 static const struct rte_flow_action_rss*
3531 flow_get_rss_action(struct rte_eth_dev *dev,
3532 const struct rte_flow_action actions[])
3534 struct mlx5_priv *priv = dev->data->dev_private;
3535 const struct rte_flow_action_rss *rss = NULL;
3537 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3538 switch (actions->type) {
3539 case RTE_FLOW_ACTION_TYPE_RSS:
3540 rss = actions->conf;
3542 case RTE_FLOW_ACTION_TYPE_SAMPLE:
3544 const struct rte_flow_action_sample *sample =
3546 const struct rte_flow_action *act = sample->actions;
3547 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3548 if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3552 case RTE_FLOW_ACTION_TYPE_METER:
3555 struct mlx5_flow_meter_info *fm;
3556 struct mlx5_flow_meter_policy *policy;
3557 const struct rte_flow_action_meter *mtr = actions->conf;
3559 fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3560 if (fm && !fm->def_policy) {
3561 policy = mlx5_flow_meter_policy_find(dev,
3562 fm->policy_id, NULL);
3563 MLX5_ASSERT(policy);
3564 if (policy->is_hierarchy) {
3566 mlx5_flow_meter_hierarchy_get_final_policy(dev,
3573 policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3585 * Get ASO age action by index.
3588 * Pointer to the Ethernet device structure.
3589 * @param[in] age_idx
3590 * Index to the ASO age action.
3593 * The specified ASO age action.
3595 struct mlx5_aso_age_action*
3596 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3598 uint16_t pool_idx = age_idx & UINT16_MAX;
3599 uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3600 struct mlx5_priv *priv = dev->data->dev_private;
3601 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3602 struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3604 return &pool->actions[offset - 1];
3607 /* maps indirect action to translated direct in some actions array */
3608 struct mlx5_translated_action_handle {
3609 struct rte_flow_action_handle *action; /**< Indirect action handle. */
3610 int index; /**< Index in related array of rte_flow_action. */
3614 * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3615 * direct action if translation possible.
3616 * This functionality used to run same execution path for both direct and
3617 * indirect actions on flow create. All necessary preparations for indirect
3618 * action handling should be performed on *handle* actions list returned
3622 * Pointer to Ethernet device.
3623 * @param[in] actions
3624 * List of actions to translate.
3625 * @param[out] handle
3626 * List to store translated indirect action object handles.
3627 * @param[in, out] indir_n
3628 * Size of *handle* array. On return should be updated with number of
3629 * indirect actions retrieved from the *actions* list.
3630 * @param[out] translated_actions
3631 * List of actions where all indirect actions were translated to direct
3632 * if possible. NULL if no translation took place.
3634 * Pointer to the error structure.
3637 * 0 on success, a negative errno value otherwise and rte_errno is set.
3640 flow_action_handles_translate(struct rte_eth_dev *dev,
3641 const struct rte_flow_action actions[],
3642 struct mlx5_translated_action_handle *handle,
3644 struct rte_flow_action **translated_actions,
3645 struct rte_flow_error *error)
3647 struct mlx5_priv *priv = dev->data->dev_private;
3648 struct rte_flow_action *translated = NULL;
3649 size_t actions_size;
3652 struct mlx5_translated_action_handle *handle_end = NULL;
3654 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3655 if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3657 if (copied_n == *indir_n) {
3658 return rte_flow_error_set
3659 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3660 NULL, "too many shared actions");
3662 rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3663 sizeof(actions[n].conf));
3664 handle[copied_n].index = n;
3668 *indir_n = copied_n;
3671 actions_size = sizeof(struct rte_flow_action) * n;
3672 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3677 memcpy(translated, actions, actions_size);
3678 for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3679 struct mlx5_shared_action_rss *shared_rss;
3680 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3681 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3682 uint32_t idx = act_idx &
3683 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3686 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3687 shared_rss = mlx5_ipool_get
3688 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3689 translated[handle->index].type =
3690 RTE_FLOW_ACTION_TYPE_RSS;
3691 translated[handle->index].conf =
3692 &shared_rss->origin;
3694 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3695 translated[handle->index].type =
3696 (enum rte_flow_action_type)
3697 MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3698 translated[handle->index].conf = (void *)(uintptr_t)idx;
3700 case MLX5_INDIRECT_ACTION_TYPE_AGE:
3701 if (priv->sh->flow_hit_aso_en) {
3702 translated[handle->index].type =
3703 (enum rte_flow_action_type)
3704 MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3705 translated[handle->index].conf =
3706 (void *)(uintptr_t)idx;
3710 case MLX5_INDIRECT_ACTION_TYPE_CT:
3711 if (priv->sh->ct_aso_en) {
3712 translated[handle->index].type =
3713 RTE_FLOW_ACTION_TYPE_CONNTRACK;
3714 translated[handle->index].conf =
3715 (void *)(uintptr_t)idx;
3720 mlx5_free(translated);
3721 return rte_flow_error_set
3722 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3723 NULL, "invalid indirect action type");
3726 *translated_actions = translated;
3731 * Get Shared RSS action from the action list.
3734 * Pointer to Ethernet device.
3736 * Pointer to the list of actions.
3737 * @param[in] shared_n
3738 * Actions list length.
3741 * The MLX5 RSS action ID if exists, otherwise return 0.
3744 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3745 struct mlx5_translated_action_handle *handle,
3748 struct mlx5_translated_action_handle *handle_end;
3749 struct mlx5_priv *priv = dev->data->dev_private;
3750 struct mlx5_shared_action_rss *shared_rss;
3753 for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3754 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3755 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3756 uint32_t idx = act_idx &
3757 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3759 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3760 shared_rss = mlx5_ipool_get
3761 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3763 __atomic_add_fetch(&shared_rss->refcnt, 1,
3774 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3776 const struct rte_flow_item *item;
3777 unsigned int has_vlan = 0;
3779 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3780 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3786 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3787 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3788 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3789 MLX5_EXPANSION_ROOT_OUTER;
3793 * Get layer flags from the prefix flow.
3795 * Some flows may be split to several subflows, the prefix subflow gets the
3796 * match items and the suffix sub flow gets the actions.
3797 * Some actions need the user defined match item flags to get the detail for
3799 * This function helps the suffix flow to get the item layer flags from prefix
3802 * @param[in] dev_flow
3803 * Pointer the created preifx subflow.
3806 * The layers get from prefix subflow.
3808 static inline uint64_t
3809 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3811 uint64_t layers = 0;
3814 * Layers bits could be localization, but usually the compiler will
3815 * help to do the optimization work for source code.
3816 * If no decap actions, use the layers directly.
3818 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3819 return dev_flow->handle->layers;
3820 /* Convert L3 layers with decap action. */
3821 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3822 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3823 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3824 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3825 /* Convert L4 layers with decap action. */
3826 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3827 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3828 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3829 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3834 * Get metadata split action information.
3836 * @param[in] actions
3837 * Pointer to the list of actions.
3839 * Pointer to the return pointer.
3840 * @param[out] qrss_type
3841 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3842 * if no QUEUE/RSS is found.
3843 * @param[out] encap_idx
3844 * Pointer to the index of the encap action if exists, otherwise the last
3848 * Total number of actions.
3851 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3852 const struct rte_flow_action **qrss,
3855 const struct rte_flow_action_raw_encap *raw_encap;
3857 int raw_decap_idx = -1;
3860 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3861 switch (actions->type) {
3862 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3863 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3864 *encap_idx = actions_n;
3866 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3867 raw_decap_idx = actions_n;
3869 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3870 raw_encap = actions->conf;
3871 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3872 *encap_idx = raw_decap_idx != -1 ?
3873 raw_decap_idx : actions_n;
3875 case RTE_FLOW_ACTION_TYPE_QUEUE:
3876 case RTE_FLOW_ACTION_TYPE_RSS:
3884 if (*encap_idx == -1)
3885 *encap_idx = actions_n;
3886 /* Count RTE_FLOW_ACTION_TYPE_END. */
3887 return actions_n + 1;
3891 * Check if the action will change packet.
3894 * Pointer to Ethernet device.
3899 * true if action will change packet, false otherwise.
3901 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3902 enum rte_flow_action_type type)
3904 struct mlx5_priv *priv = dev->data->dev_private;
3907 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3908 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3909 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3910 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3911 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3912 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3913 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3914 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3915 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3916 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3917 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3918 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3919 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3920 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3921 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3922 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3923 case RTE_FLOW_ACTION_TYPE_SET_META:
3924 case RTE_FLOW_ACTION_TYPE_SET_TAG:
3925 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3926 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3927 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3928 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3929 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3930 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3931 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3932 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3933 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3934 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3935 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3937 case RTE_FLOW_ACTION_TYPE_FLAG:
3938 case RTE_FLOW_ACTION_TYPE_MARK:
3939 if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3949 * Check meter action from the action list.
3952 * Pointer to Ethernet device.
3953 * @param[in] actions
3954 * Pointer to the list of actions.
3955 * @param[out] has_mtr
3956 * Pointer to the meter exist flag.
3957 * @param[out] has_modify
3958 * Pointer to the flag showing there's packet change action.
3959 * @param[out] meter_id
3960 * Pointer to the meter id.
3963 * Total number of actions.
3966 flow_check_meter_action(struct rte_eth_dev *dev,
3967 const struct rte_flow_action actions[],
3968 bool *has_mtr, bool *has_modify, uint32_t *meter_id)
3970 const struct rte_flow_action_meter *mtr = NULL;
3973 MLX5_ASSERT(has_mtr);
3975 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3976 switch (actions->type) {
3977 case RTE_FLOW_ACTION_TYPE_METER:
3978 mtr = actions->conf;
3979 *meter_id = mtr->mtr_id;
3986 *has_modify |= flow_check_modify_action_type(dev,
3990 /* Count RTE_FLOW_ACTION_TYPE_END. */
3991 return actions_n + 1;
3995 * Check if the flow should be split due to hairpin.
3996 * The reason for the split is that in current HW we can't
3997 * support encap and push-vlan on Rx, so if a flow contains
3998 * these actions we move it to Tx.
4001 * Pointer to Ethernet device.
4003 * Flow rule attributes.
4004 * @param[in] actions
4005 * Associated actions (list terminated by the END action).
4008 * > 0 the number of actions and the flow should be split,
4009 * 0 when no split required.
4012 flow_check_hairpin_split(struct rte_eth_dev *dev,
4013 const struct rte_flow_attr *attr,
4014 const struct rte_flow_action actions[])
4016 int queue_action = 0;
4019 const struct rte_flow_action_queue *queue;
4020 const struct rte_flow_action_rss *rss;
4021 const struct rte_flow_action_raw_encap *raw_encap;
4022 const struct rte_eth_hairpin_conf *conf;
4026 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4027 switch (actions->type) {
4028 case RTE_FLOW_ACTION_TYPE_QUEUE:
4029 queue = actions->conf;
4032 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4033 if (conf == NULL || conf->tx_explicit != 0)
4038 case RTE_FLOW_ACTION_TYPE_RSS:
4039 rss = actions->conf;
4040 if (rss == NULL || rss->queue_num == 0)
4042 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4043 if (conf == NULL || conf->tx_explicit != 0)
4048 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4049 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4050 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4051 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4052 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4056 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4057 raw_encap = actions->conf;
4058 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4067 if (split && queue_action)
4072 /* Declare flow create/destroy prototype in advance. */
4074 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4075 const struct rte_flow_attr *attr,
4076 const struct rte_flow_item items[],
4077 const struct rte_flow_action actions[],
4078 bool external, struct rte_flow_error *error);
4081 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4085 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
4086 struct mlx5_hlist_entry *entry,
4087 uint64_t key, void *cb_ctx __rte_unused)
4089 struct mlx5_flow_mreg_copy_resource *mcp_res =
4090 container_of(entry, typeof(*mcp_res), hlist_ent);
4092 return mcp_res->mark_id != key;
4095 struct mlx5_hlist_entry *
4096 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
4099 struct rte_eth_dev *dev = list->ctx;
4100 struct mlx5_priv *priv = dev->data->dev_private;
4101 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4102 struct mlx5_flow_mreg_copy_resource *mcp_res;
4103 struct rte_flow_error *error = ctx->error;
4106 uint32_t mark_id = key;
4107 struct rte_flow_attr attr = {
4108 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4111 struct mlx5_rte_flow_item_tag tag_spec = {
4114 struct rte_flow_item items[] = {
4115 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4117 struct rte_flow_action_mark ftag = {
4120 struct mlx5_flow_action_copy_mreg cp_mreg = {
4124 struct rte_flow_action_jump jump = {
4125 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4127 struct rte_flow_action actions[] = {
4128 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4131 /* Fill the register fileds in the flow. */
4132 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4136 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4140 /* Provide the full width of FLAG specific value. */
4141 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4142 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4143 /* Build a new flow. */
4144 if (mark_id != MLX5_DEFAULT_COPY_ID) {
4145 items[0] = (struct rte_flow_item){
4146 .type = (enum rte_flow_item_type)
4147 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4150 items[1] = (struct rte_flow_item){
4151 .type = RTE_FLOW_ITEM_TYPE_END,
4153 actions[0] = (struct rte_flow_action){
4154 .type = (enum rte_flow_action_type)
4155 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4158 actions[1] = (struct rte_flow_action){
4159 .type = (enum rte_flow_action_type)
4160 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4163 actions[2] = (struct rte_flow_action){
4164 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4167 actions[3] = (struct rte_flow_action){
4168 .type = RTE_FLOW_ACTION_TYPE_END,
4171 /* Default rule, wildcard match. */
4172 attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4173 items[0] = (struct rte_flow_item){
4174 .type = RTE_FLOW_ITEM_TYPE_END,
4176 actions[0] = (struct rte_flow_action){
4177 .type = (enum rte_flow_action_type)
4178 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4181 actions[1] = (struct rte_flow_action){
4182 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4185 actions[2] = (struct rte_flow_action){
4186 .type = RTE_FLOW_ACTION_TYPE_END,
4189 /* Build a new entry. */
4190 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4196 mcp_res->mark_id = mark_id;
4198 * The copy Flows are not included in any list. There
4199 * ones are referenced from other Flows and can not
4200 * be applied, removed, deleted in ardbitrary order
4201 * by list traversing.
4203 mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
4204 actions, false, error);
4205 if (!mcp_res->rix_flow) {
4206 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4209 return &mcp_res->hlist_ent;
4213 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4215 * As mark_id is unique, if there's already a registered flow for the mark_id,
4216 * return by increasing the reference counter of the resource. Otherwise, create
4217 * the resource (mcp_res) and flow.
4220 * - If ingress port is ANY and reg_c[1] is mark_id,
4221 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4223 * For default flow (zero mark_id), flow is like,
4224 * - If ingress port is ANY,
4225 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
4228 * Pointer to Ethernet device.
4230 * ID of MARK action, zero means default flow for META.
4232 * Perform verbose error reporting if not NULL.
4235 * Associated resource on success, NULL otherwise and rte_errno is set.
4237 static struct mlx5_flow_mreg_copy_resource *
4238 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4239 struct rte_flow_error *error)
4241 struct mlx5_priv *priv = dev->data->dev_private;
4242 struct mlx5_hlist_entry *entry;
4243 struct mlx5_flow_cb_ctx ctx = {
4248 /* Check if already registered. */
4249 MLX5_ASSERT(priv->mreg_cp_tbl);
4250 entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4253 return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4258 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
4260 struct mlx5_flow_mreg_copy_resource *mcp_res =
4261 container_of(entry, typeof(*mcp_res), hlist_ent);
4262 struct rte_eth_dev *dev = list->ctx;
4263 struct mlx5_priv *priv = dev->data->dev_private;
4265 MLX5_ASSERT(mcp_res->rix_flow);
4266 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
4267 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4271 * Release flow in RX_CP_TBL.
4274 * Pointer to Ethernet device.
4276 * Parent flow for wich copying is provided.
4279 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4280 struct rte_flow *flow)
4282 struct mlx5_flow_mreg_copy_resource *mcp_res;
4283 struct mlx5_priv *priv = dev->data->dev_private;
4285 if (!flow->rix_mreg_copy)
4287 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4288 flow->rix_mreg_copy);
4289 if (!mcp_res || !priv->mreg_cp_tbl)
4291 MLX5_ASSERT(mcp_res->rix_flow);
4292 mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4293 flow->rix_mreg_copy = 0;
4297 * Remove the default copy action from RX_CP_TBL.
4299 * This functions is called in the mlx5_dev_start(). No thread safe
4303 * Pointer to Ethernet device.
4306 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4308 struct mlx5_hlist_entry *entry;
4309 struct mlx5_priv *priv = dev->data->dev_private;
4311 /* Check if default flow is registered. */
4312 if (!priv->mreg_cp_tbl)
4314 entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4315 MLX5_DEFAULT_COPY_ID, NULL);
4318 mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4322 * Add the default copy action in in RX_CP_TBL.
4324 * This functions is called in the mlx5_dev_start(). No thread safe
4328 * Pointer to Ethernet device.
4330 * Perform verbose error reporting if not NULL.
4333 * 0 for success, negative value otherwise and rte_errno is set.
4336 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4337 struct rte_flow_error *error)
4339 struct mlx5_priv *priv = dev->data->dev_private;
4340 struct mlx5_flow_mreg_copy_resource *mcp_res;
4342 /* Check whether extensive metadata feature is engaged. */
4343 if (!priv->config.dv_flow_en ||
4344 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4345 !mlx5_flow_ext_mreg_supported(dev) ||
4346 !priv->sh->dv_regc0_mask)
4349 * Add default mreg copy flow may be called multiple time, but
4350 * only be called once in stop. Avoid register it twice.
4352 if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4354 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4361 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4363 * All the flow having Q/RSS action should be split by
4364 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4365 * performs the following,
4366 * - CQE->flow_tag := reg_c[1] (MARK)
4367 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4368 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4369 * but there should be a flow per each MARK ID set by MARK action.
4371 * For the aforementioned reason, if there's a MARK action in flow's action
4372 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4373 * the MARK ID to CQE's flow_tag like,
4374 * - If reg_c[1] is mark_id,
4375 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4377 * For SET_META action which stores value in reg_c[0], as the destination is
4378 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4379 * MARK ID means the default flow. The default flow looks like,
4380 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4383 * Pointer to Ethernet device.
4385 * Pointer to flow structure.
4386 * @param[in] actions
4387 * Pointer to the list of actions.
4389 * Perform verbose error reporting if not NULL.
4392 * 0 on success, negative value otherwise and rte_errno is set.
4395 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4396 struct rte_flow *flow,
4397 const struct rte_flow_action *actions,
4398 struct rte_flow_error *error)
4400 struct mlx5_priv *priv = dev->data->dev_private;
4401 struct mlx5_dev_config *config = &priv->config;
4402 struct mlx5_flow_mreg_copy_resource *mcp_res;
4403 const struct rte_flow_action_mark *mark;
4405 /* Check whether extensive metadata feature is engaged. */
4406 if (!config->dv_flow_en ||
4407 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4408 !mlx5_flow_ext_mreg_supported(dev) ||
4409 !priv->sh->dv_regc0_mask)
4411 /* Find MARK action. */
4412 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4413 switch (actions->type) {
4414 case RTE_FLOW_ACTION_TYPE_FLAG:
4415 mcp_res = flow_mreg_add_copy_action
4416 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4419 flow->rix_mreg_copy = mcp_res->idx;
4421 case RTE_FLOW_ACTION_TYPE_MARK:
4422 mark = (const struct rte_flow_action_mark *)
4425 flow_mreg_add_copy_action(dev, mark->id, error);
4428 flow->rix_mreg_copy = mcp_res->idx;
4437 #define MLX5_MAX_SPLIT_ACTIONS 24
4438 #define MLX5_MAX_SPLIT_ITEMS 24
4441 * Split the hairpin flow.
4442 * Since HW can't support encap and push-vlan on Rx, we move these
4444 * If the count action is after the encap then we also
4445 * move the count action. in this case the count will also measure
4449 * Pointer to Ethernet device.
4450 * @param[in] actions
4451 * Associated actions (list terminated by the END action).
4452 * @param[out] actions_rx
4454 * @param[out] actions_tx
4456 * @param[out] pattern_tx
4457 * The pattern items for the Tx flow.
4458 * @param[out] flow_id
4459 * The flow ID connected to this flow.
4465 flow_hairpin_split(struct rte_eth_dev *dev,
4466 const struct rte_flow_action actions[],
4467 struct rte_flow_action actions_rx[],
4468 struct rte_flow_action actions_tx[],
4469 struct rte_flow_item pattern_tx[],
4472 const struct rte_flow_action_raw_encap *raw_encap;
4473 const struct rte_flow_action_raw_decap *raw_decap;
4474 struct mlx5_rte_flow_action_set_tag *set_tag;
4475 struct rte_flow_action *tag_action;
4476 struct mlx5_rte_flow_item_tag *tag_item;
4477 struct rte_flow_item *item;
4481 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4482 switch (actions->type) {
4483 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4484 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4485 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4486 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4487 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4488 rte_memcpy(actions_tx, actions,
4489 sizeof(struct rte_flow_action));
4492 case RTE_FLOW_ACTION_TYPE_COUNT:
4494 rte_memcpy(actions_tx, actions,
4495 sizeof(struct rte_flow_action));
4498 rte_memcpy(actions_rx, actions,
4499 sizeof(struct rte_flow_action));
4503 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4504 raw_encap = actions->conf;
4505 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4506 memcpy(actions_tx, actions,
4507 sizeof(struct rte_flow_action));
4511 rte_memcpy(actions_rx, actions,
4512 sizeof(struct rte_flow_action));
4516 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4517 raw_decap = actions->conf;
4518 if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4519 memcpy(actions_tx, actions,
4520 sizeof(struct rte_flow_action));
4523 rte_memcpy(actions_rx, actions,
4524 sizeof(struct rte_flow_action));
4529 rte_memcpy(actions_rx, actions,
4530 sizeof(struct rte_flow_action));
4535 /* Add set meta action and end action for the Rx flow. */
4536 tag_action = actions_rx;
4537 tag_action->type = (enum rte_flow_action_type)
4538 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4540 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4542 set_tag = (void *)actions_rx;
4543 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4544 .id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4547 MLX5_ASSERT(set_tag->id > REG_NON);
4548 tag_action->conf = set_tag;
4549 /* Create Tx item list. */
4550 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4551 addr = (void *)&pattern_tx[2];
4553 item->type = (enum rte_flow_item_type)
4554 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4555 tag_item = (void *)addr;
4556 tag_item->data = flow_id;
4557 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4558 MLX5_ASSERT(set_tag->id > REG_NON);
4559 item->spec = tag_item;
4560 addr += sizeof(struct mlx5_rte_flow_item_tag);
4561 tag_item = (void *)addr;
4562 tag_item->data = UINT32_MAX;
4563 tag_item->id = UINT16_MAX;
4564 item->mask = tag_item;
4567 item->type = RTE_FLOW_ITEM_TYPE_END;
4572 * The last stage of splitting chain, just creates the subflow
4573 * without any modification.
4576 * Pointer to Ethernet device.
4578 * Parent flow structure pointer.
4579 * @param[in, out] sub_flow
4580 * Pointer to return the created subflow, may be NULL.
4582 * Flow rule attributes.
4584 * Pattern specification (list terminated by the END pattern item).
4585 * @param[in] actions
4586 * Associated actions (list terminated by the END action).
4587 * @param[in] flow_split_info
4588 * Pointer to flow split info structure.
4590 * Perform verbose error reporting if not NULL.
4592 * 0 on success, negative value otherwise
4595 flow_create_split_inner(struct rte_eth_dev *dev,
4596 struct rte_flow *flow,
4597 struct mlx5_flow **sub_flow,
4598 const struct rte_flow_attr *attr,
4599 const struct rte_flow_item items[],
4600 const struct rte_flow_action actions[],
4601 struct mlx5_flow_split_info *flow_split_info,
4602 struct rte_flow_error *error)
4604 struct mlx5_flow *dev_flow;
4606 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4607 flow_split_info->flow_idx, error);
4610 dev_flow->flow = flow;
4611 dev_flow->external = flow_split_info->external;
4612 dev_flow->skip_scale = flow_split_info->skip_scale;
4613 /* Subflow object was created, we must include one in the list. */
4614 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4615 dev_flow->handle, next);
4617 * If dev_flow is as one of the suffix flow, some actions in suffix
4618 * flow may need some user defined item layer flags, and pass the
4619 * Metadate rxq mark flag to suffix flow as well.
4621 if (flow_split_info->prefix_layers)
4622 dev_flow->handle->layers = flow_split_info->prefix_layers;
4623 if (flow_split_info->prefix_mark)
4624 dev_flow->handle->mark = 1;
4626 *sub_flow = dev_flow;
4627 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4628 dev_flow->dv.table_id = flow_split_info->table_id;
4630 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4634 * Get the sub policy of a meter.
4637 * Pointer to Ethernet device.
4639 * Parent flow structure pointer.
4641 * Pointer to thread flow work space.
4643 * Flow rule attributes.
4645 * Pattern specification (list terminated by the END pattern item).
4647 * Perform verbose error reporting if not NULL.
4650 * Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4652 static struct mlx5_flow_meter_sub_policy *
4653 get_meter_sub_policy(struct rte_eth_dev *dev,
4654 struct rte_flow *flow,
4655 struct mlx5_flow_workspace *wks,
4656 const struct rte_flow_attr *attr,
4657 const struct rte_flow_item items[],
4658 struct rte_flow_error *error)
4660 struct mlx5_flow_meter_policy *policy;
4661 struct mlx5_flow_meter_policy *final_policy;
4662 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4664 policy = wks->policy;
4665 final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4666 if (final_policy->is_rss || final_policy->is_queue) {
4667 struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4668 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4672 * This is a tmp dev_flow,
4673 * no need to register any matcher for it in translate.
4675 wks->skip_matcher_reg = 1;
4676 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4677 struct mlx5_flow dev_flow = {0};
4678 struct mlx5_flow_handle dev_handle = { {0} };
4680 if (final_policy->is_rss) {
4681 const void *rss_act =
4682 final_policy->act_cnt[i].rss->conf;
4683 struct rte_flow_action rss_actions[2] = {
4685 .type = RTE_FLOW_ACTION_TYPE_RSS,
4689 .type = RTE_FLOW_ACTION_TYPE_END,
4694 dev_flow.handle = &dev_handle;
4695 dev_flow.ingress = attr->ingress;
4696 dev_flow.flow = flow;
4697 dev_flow.external = 0;
4698 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4699 dev_flow.dv.transfer = attr->transfer;
4702 * Translate RSS action to get rss hash fields.
4704 if (flow_drv_translate(dev, &dev_flow, attr,
4705 items, rss_actions, error))
4707 rss_desc_v[i] = wks->rss_desc;
4708 rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4709 rss_desc_v[i].hash_fields =
4710 dev_flow.hash_fields;
4711 rss_desc_v[i].queue_num =
4712 rss_desc_v[i].hash_fields ?
4713 rss_desc_v[i].queue_num : 1;
4714 rss_desc_v[i].tunnel =
4715 !!(dev_flow.handle->layers &
4716 MLX5_FLOW_LAYER_TUNNEL);
4718 /* This is queue action. */
4719 rss_desc_v[i] = wks->rss_desc;
4720 rss_desc_v[i].key_len = 0;
4721 rss_desc_v[i].hash_fields = 0;
4722 rss_desc_v[i].queue =
4723 &final_policy->act_cnt[i].queue;
4724 rss_desc_v[i].queue_num = 1;
4726 rss_desc[i] = &rss_desc_v[i];
4728 sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4729 flow, policy, rss_desc);
4731 enum mlx5_meter_domain mtr_domain =
4732 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4733 attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4734 MLX5_MTR_DOMAIN_INGRESS;
4735 sub_policy = policy->sub_policys[mtr_domain][0];
4738 rte_flow_error_set(error, EINVAL,
4739 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4740 "Failed to get meter sub-policy.");
4748 * Split the meter flow.
4750 * As meter flow will split to three sub flow, other than meter
4751 * action, the other actions make sense to only meter accepts
4752 * the packet. If it need to be dropped, no other additional
4753 * actions should be take.
4755 * One kind of special action which decapsulates the L3 tunnel
4756 * header will be in the prefix sub flow, as not to take the
4757 * L3 tunnel header into account.
4760 * Pointer to Ethernet device.
4762 * Parent flow structure pointer.
4764 * Pointer to thread flow work space.
4766 * Flow rule attributes.
4768 * Pattern specification (list terminated by the END pattern item).
4769 * @param[out] sfx_items
4770 * Suffix flow match items (list terminated by the END pattern item).
4771 * @param[in] actions
4772 * Associated actions (list terminated by the END action).
4773 * @param[out] actions_sfx
4774 * Suffix flow actions.
4775 * @param[out] actions_pre
4776 * Prefix flow actions.
4777 * @param[out] mtr_flow_id
4778 * Pointer to meter flow id.
4780 * Perform verbose error reporting if not NULL.
4783 * 0 on success, a negative errno value otherwise and rte_errno is set.
4786 flow_meter_split_prep(struct rte_eth_dev *dev,
4787 struct rte_flow *flow,
4788 struct mlx5_flow_workspace *wks,
4789 const struct rte_flow_attr *attr,
4790 const struct rte_flow_item items[],
4791 struct rte_flow_item sfx_items[],
4792 const struct rte_flow_action actions[],
4793 struct rte_flow_action actions_sfx[],
4794 struct rte_flow_action actions_pre[],
4795 uint32_t *mtr_flow_id,
4796 struct rte_flow_error *error)
4798 struct mlx5_priv *priv = dev->data->dev_private;
4799 struct mlx5_flow_meter_info *fm = wks->fm;
4800 struct rte_flow_action *tag_action = NULL;
4801 struct rte_flow_item *tag_item;
4802 struct mlx5_rte_flow_action_set_tag *set_tag;
4803 const struct rte_flow_action_raw_encap *raw_encap;
4804 const struct rte_flow_action_raw_decap *raw_decap;
4805 struct mlx5_rte_flow_item_tag *tag_item_spec;
4806 struct mlx5_rte_flow_item_tag *tag_item_mask;
4807 uint32_t tag_id = 0;
4808 struct rte_flow_item *vlan_item_dst = NULL;
4809 const struct rte_flow_item *vlan_item_src = NULL;
4810 struct rte_flow_action *hw_mtr_action;
4811 struct rte_flow_action *action_pre_head = NULL;
4812 int32_t flow_src_port = priv->representor_id;
4814 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4815 uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4816 MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4817 uint32_t flow_id = 0;
4818 uint32_t flow_id_reversed = 0;
4819 uint8_t flow_id_bits = 0;
4822 /* Prepare the suffix subflow items. */
4823 tag_item = sfx_items++;
4824 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4825 struct mlx5_priv *port_priv;
4826 const struct rte_flow_item_port_id *pid_v;
4827 int item_type = items->type;
4829 switch (item_type) {
4830 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4831 pid_v = items->spec;
4833 port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4835 return rte_flow_error_set(error,
4837 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
4839 "Failed to get port info.");
4840 flow_src_port = port_priv->representor_id;
4841 if (!fm->def_policy && wks->policy->is_hierarchy &&
4842 flow_src_port != priv->representor_id) {
4843 if (flow_drv_mtr_hierarchy_rule_create(dev,
4850 memcpy(sfx_items, items, sizeof(*sfx_items));
4853 case RTE_FLOW_ITEM_TYPE_VLAN:
4854 /* Determine if copy vlan item below. */
4855 vlan_item_src = items;
4856 vlan_item_dst = sfx_items++;
4857 vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
4863 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4865 mtr_first = priv->sh->meter_aso_en &&
4866 (attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
4867 /* For ASO meter, meter must be before tag in TX direction. */
4869 action_pre_head = actions_pre++;
4870 /* Leave space for tag action. */
4871 tag_action = actions_pre++;
4873 /* Prepare the actions for prefix and suffix flow. */
4874 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4875 struct rte_flow_action *action_cur = NULL;
4877 switch (actions->type) {
4878 case RTE_FLOW_ACTION_TYPE_METER:
4880 action_cur = action_pre_head;
4882 /* Leave space for tag action. */
4883 tag_action = actions_pre++;
4884 action_cur = actions_pre++;
4887 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4888 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4889 action_cur = actions_pre++;
4891 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4892 raw_encap = actions->conf;
4893 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4894 action_cur = actions_pre++;
4896 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4897 raw_decap = actions->conf;
4898 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4899 action_cur = actions_pre++;
4901 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4902 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4903 if (vlan_item_dst && vlan_item_src) {
4904 memcpy(vlan_item_dst, vlan_item_src,
4905 sizeof(*vlan_item_dst));
4907 * Convert to internal match item, it is used
4908 * for vlan push and set vid.
4910 vlan_item_dst->type = (enum rte_flow_item_type)
4911 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4918 action_cur = (fm->def_policy) ?
4919 actions_sfx++ : actions_pre++;
4920 memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4922 /* Add end action to the actions. */
4923 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4924 if (priv->sh->meter_aso_en) {
4926 * For ASO meter, need to add an extra jump action explicitly,
4927 * to jump from meter to policer table.
4929 struct mlx5_flow_meter_sub_policy *sub_policy;
4930 struct mlx5_flow_tbl_data_entry *tbl_data;
4932 if (!fm->def_policy) {
4933 sub_policy = get_meter_sub_policy(dev, flow, wks,
4934 attr, items, error);
4938 enum mlx5_meter_domain mtr_domain =
4939 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4940 attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4941 MLX5_MTR_DOMAIN_INGRESS;
4944 &priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
4946 tbl_data = container_of(sub_policy->tbl_rsc,
4947 struct mlx5_flow_tbl_data_entry, tbl);
4948 hw_mtr_action = actions_pre++;
4949 hw_mtr_action->type = (enum rte_flow_action_type)
4950 MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
4951 hw_mtr_action->conf = tbl_data->jump.action;
4953 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4956 return rte_flow_error_set(error, ENOMEM,
4957 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4958 "No tag action space.");
4960 tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
4963 /* Only default-policy Meter creates mtr flow id. */
4964 if (fm->def_policy) {
4965 mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4967 return rte_flow_error_set(error, ENOMEM,
4968 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4969 "Failed to allocate meter flow id.");
4970 flow_id = tag_id - 1;
4971 flow_id_bits = (!flow_id) ? 1 :
4972 (MLX5_REG_BITS - __builtin_clz(flow_id));
4973 if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
4975 mlx5_ipool_free(fm->flow_ipool, tag_id);
4976 return rte_flow_error_set(error, EINVAL,
4977 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4978 "Meter flow id exceeds max limit.");
4980 if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
4981 priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
4983 /* Build tag actions and items for meter_id/meter flow_id. */
4984 set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
4985 tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4986 tag_item_mask = tag_item_spec + 1;
4987 /* Both flow_id and meter_id share the same register. */
4988 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4989 .id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
4991 .offset = mtr_id_offset,
4992 .length = mtr_reg_bits,
4993 .data = flow->meter,
4996 * The color Reg bits used by flow_id are growing from
4997 * msb to lsb, so must do bit reverse for flow_id val in RegC.
4999 for (shift = 0; shift < flow_id_bits; shift++)
5000 flow_id_reversed = (flow_id_reversed << 1) |
5001 ((flow_id >> shift) & 0x1);
5003 flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5004 tag_item_spec->id = set_tag->id;
5005 tag_item_spec->data = set_tag->data << mtr_id_offset;
5006 tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5007 tag_action->type = (enum rte_flow_action_type)
5008 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5009 tag_action->conf = set_tag;
5010 tag_item->type = (enum rte_flow_item_type)
5011 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5012 tag_item->spec = tag_item_spec;
5013 tag_item->last = NULL;
5014 tag_item->mask = tag_item_mask;
5017 *mtr_flow_id = tag_id;
5022 * Split action list having QUEUE/RSS for metadata register copy.
5024 * Once Q/RSS action is detected in user's action list, the flow action
5025 * should be split in order to copy metadata registers, which will happen in
5027 * - CQE->flow_tag := reg_c[1] (MARK)
5028 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5029 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5030 * This is because the last action of each flow must be a terminal action
5031 * (QUEUE, RSS or DROP).
5033 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5034 * stored and kept in the mlx5_flow structure per each sub_flow.
5036 * The Q/RSS action is replaced with,
5037 * - SET_TAG, setting the allocated flow ID to reg_c[2].
5038 * And the following JUMP action is added at the end,
5039 * - JUMP, to RX_CP_TBL.
5041 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5042 * flow_create_split_metadata() routine. The flow will look like,
5043 * - If flow ID matches (reg_c[2]), perform Q/RSS.
5046 * Pointer to Ethernet device.
5047 * @param[out] split_actions
5048 * Pointer to store split actions to jump to CP_TBL.
5049 * @param[in] actions
5050 * Pointer to the list of original flow actions.
5052 * Pointer to the Q/RSS action.
5053 * @param[in] actions_n
5054 * Number of original actions.
5056 * Perform verbose error reporting if not NULL.
5059 * non-zero unique flow_id on success, otherwise 0 and
5060 * error/rte_error are set.
5063 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5064 struct rte_flow_action *split_actions,
5065 const struct rte_flow_action *actions,
5066 const struct rte_flow_action *qrss,
5067 int actions_n, struct rte_flow_error *error)
5069 struct mlx5_priv *priv = dev->data->dev_private;
5070 struct mlx5_rte_flow_action_set_tag *set_tag;
5071 struct rte_flow_action_jump *jump;
5072 const int qrss_idx = qrss - actions;
5073 uint32_t flow_id = 0;
5077 * Given actions will be split
5078 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5079 * - Add jump to mreg CP_TBL.
5080 * As a result, there will be one more action.
5083 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5084 set_tag = (void *)(split_actions + actions_n);
5086 * If tag action is not set to void(it means we are not the meter
5087 * suffix flow), add the tag action. Since meter suffix flow already
5088 * has the tag added.
5090 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5092 * Allocate the new subflow ID. This one is unique within
5093 * device and not shared with representors. Otherwise,
5094 * we would have to resolve multi-thread access synch
5095 * issue. Each flow on the shared device is appended
5096 * with source vport identifier, so the resulting
5097 * flows will be unique in the shared (by master and
5098 * representors) domain even if they have coinciding
5101 mlx5_ipool_malloc(priv->sh->ipool
5102 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5104 return rte_flow_error_set(error, ENOMEM,
5105 RTE_FLOW_ERROR_TYPE_ACTION,
5106 NULL, "can't allocate id "
5107 "for split Q/RSS subflow");
5108 /* Internal SET_TAG action to set flow ID. */
5109 *set_tag = (struct mlx5_rte_flow_action_set_tag){
5112 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5116 /* Construct new actions array. */
5117 /* Replace QUEUE/RSS action. */
5118 split_actions[qrss_idx] = (struct rte_flow_action){
5119 .type = (enum rte_flow_action_type)
5120 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5124 /* JUMP action to jump to mreg copy table (CP_TBL). */
5125 jump = (void *)(set_tag + 1);
5126 *jump = (struct rte_flow_action_jump){
5127 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5129 split_actions[actions_n - 2] = (struct rte_flow_action){
5130 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5133 split_actions[actions_n - 1] = (struct rte_flow_action){
5134 .type = RTE_FLOW_ACTION_TYPE_END,
5140 * Extend the given action list for Tx metadata copy.
5142 * Copy the given action list to the ext_actions and add flow metadata register
5143 * copy action in order to copy reg_a set by WQE to reg_c[0].
5145 * @param[out] ext_actions
5146 * Pointer to the extended action list.
5147 * @param[in] actions
5148 * Pointer to the list of actions.
5149 * @param[in] actions_n
5150 * Number of actions in the list.
5152 * Perform verbose error reporting if not NULL.
5153 * @param[in] encap_idx
5154 * The encap action inndex.
5157 * 0 on success, negative value otherwise
5160 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5161 struct rte_flow_action *ext_actions,
5162 const struct rte_flow_action *actions,
5163 int actions_n, struct rte_flow_error *error,
5166 struct mlx5_flow_action_copy_mreg *cp_mreg =
5167 (struct mlx5_flow_action_copy_mreg *)
5168 (ext_actions + actions_n + 1);
5171 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5175 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5180 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5181 if (encap_idx == actions_n - 1) {
5182 ext_actions[actions_n - 1] = (struct rte_flow_action){
5183 .type = (enum rte_flow_action_type)
5184 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5187 ext_actions[actions_n] = (struct rte_flow_action){
5188 .type = RTE_FLOW_ACTION_TYPE_END,
5191 ext_actions[encap_idx] = (struct rte_flow_action){
5192 .type = (enum rte_flow_action_type)
5193 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5196 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5197 sizeof(*ext_actions) * (actions_n - encap_idx));
5203 * Check the match action from the action list.
5205 * @param[in] actions
5206 * Pointer to the list of actions.
5208 * Flow rule attributes.
5210 * The action to be check if exist.
5211 * @param[out] match_action_pos
5212 * Pointer to the position of the matched action if exists, otherwise is -1.
5213 * @param[out] qrss_action_pos
5214 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5215 * @param[out] modify_after_mirror
5216 * Pointer to the flag of modify action after FDB mirroring.
5219 * > 0 the total number of actions.
5220 * 0 if not found match action in action list.
5223 flow_check_match_action(const struct rte_flow_action actions[],
5224 const struct rte_flow_attr *attr,
5225 enum rte_flow_action_type action,
5226 int *match_action_pos, int *qrss_action_pos,
5227 int *modify_after_mirror)
5229 const struct rte_flow_action_sample *sample;
5236 *match_action_pos = -1;
5237 *qrss_action_pos = -1;
5238 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5239 if (actions->type == action) {
5241 *match_action_pos = actions_n;
5243 switch (actions->type) {
5244 case RTE_FLOW_ACTION_TYPE_QUEUE:
5245 case RTE_FLOW_ACTION_TYPE_RSS:
5246 *qrss_action_pos = actions_n;
5248 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5249 sample = actions->conf;
5250 ratio = sample->ratio;
5251 sub_type = ((const struct rte_flow_action *)
5252 (sample->actions))->type;
5253 if (ratio == 1 && attr->transfer)
5256 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5257 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5258 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5259 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5260 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5261 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5262 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5263 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5264 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5265 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5266 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5267 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5268 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5269 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5270 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5271 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5272 case RTE_FLOW_ACTION_TYPE_FLAG:
5273 case RTE_FLOW_ACTION_TYPE_MARK:
5274 case RTE_FLOW_ACTION_TYPE_SET_META:
5275 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5276 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5277 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5278 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5279 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5280 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5281 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5282 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5283 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5284 case RTE_FLOW_ACTION_TYPE_METER:
5286 *modify_after_mirror = 1;
5293 if (flag && fdb_mirror && !*modify_after_mirror) {
5294 /* FDB mirroring uses the destination array to implement
5295 * instead of FLOW_SAMPLER object.
5297 if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5300 /* Count RTE_FLOW_ACTION_TYPE_END. */
5301 return flag ? actions_n + 1 : 0;
5304 #define SAMPLE_SUFFIX_ITEM 2
5307 * Split the sample flow.
5309 * As sample flow will split to two sub flow, sample flow with
5310 * sample action, the other actions will move to new suffix flow.
5312 * Also add unique tag id with tag action in the sample flow,
5313 * the same tag id will be as match in the suffix flow.
5316 * Pointer to Ethernet device.
5317 * @param[in] add_tag
5318 * Add extra tag action flag.
5319 * @param[out] sfx_items
5320 * Suffix flow match items (list terminated by the END pattern item).
5321 * @param[in] actions
5322 * Associated actions (list terminated by the END action).
5323 * @param[out] actions_sfx
5324 * Suffix flow actions.
5325 * @param[out] actions_pre
5326 * Prefix flow actions.
5327 * @param[in] actions_n
5328 * The total number of actions.
5329 * @param[in] sample_action_pos
5330 * The sample action position.
5331 * @param[in] qrss_action_pos
5332 * The Queue/RSS action position.
5333 * @param[in] jump_table
5334 * Add extra jump action flag.
5336 * Perform verbose error reporting if not NULL.
5339 * 0 on success, or unique flow_id, a negative errno value
5340 * otherwise and rte_errno is set.
5343 flow_sample_split_prep(struct rte_eth_dev *dev,
5345 struct rte_flow_item sfx_items[],
5346 const struct rte_flow_action actions[],
5347 struct rte_flow_action actions_sfx[],
5348 struct rte_flow_action actions_pre[],
5350 int sample_action_pos,
5351 int qrss_action_pos,
5353 struct rte_flow_error *error)
5355 struct mlx5_priv *priv = dev->data->dev_private;
5356 struct mlx5_rte_flow_action_set_tag *set_tag;
5357 struct mlx5_rte_flow_item_tag *tag_spec;
5358 struct mlx5_rte_flow_item_tag *tag_mask;
5359 struct rte_flow_action_jump *jump_action;
5360 uint32_t tag_id = 0;
5362 int append_index = 0;
5365 if (sample_action_pos < 0)
5366 return rte_flow_error_set(error, EINVAL,
5367 RTE_FLOW_ERROR_TYPE_ACTION,
5368 NULL, "invalid position of sample "
5370 /* Prepare the actions for prefix and suffix flow. */
5371 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5372 index = qrss_action_pos;
5373 /* Put the preceding the Queue/RSS action into prefix flow. */
5375 memcpy(actions_pre, actions,
5376 sizeof(struct rte_flow_action) * index);
5377 /* Put others preceding the sample action into prefix flow. */
5378 if (sample_action_pos > index + 1)
5379 memcpy(actions_pre + index, actions + index + 1,
5380 sizeof(struct rte_flow_action) *
5381 (sample_action_pos - index - 1));
5382 index = sample_action_pos - 1;
5383 /* Put Queue/RSS action into Suffix flow. */
5384 memcpy(actions_sfx, actions + qrss_action_pos,
5385 sizeof(struct rte_flow_action));
5388 index = sample_action_pos;
5390 memcpy(actions_pre, actions,
5391 sizeof(struct rte_flow_action) * index);
5393 /* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5394 * For CX6DX and above, metadata registers Cx preserve their value,
5395 * add an extra tag action for NIC-RX and E-Switch Domain.
5398 /* Prepare the prefix tag action. */
5400 set_tag = (void *)(actions_pre + actions_n + append_index);
5401 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5404 mlx5_ipool_malloc(priv->sh->ipool
5405 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5406 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5410 /* Prepare the suffix subflow items. */
5411 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5412 tag_spec->data = tag_id;
5413 tag_spec->id = set_tag->id;
5414 tag_mask = tag_spec + 1;
5415 tag_mask->data = UINT32_MAX;
5416 sfx_items[0] = (struct rte_flow_item){
5417 .type = (enum rte_flow_item_type)
5418 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5423 sfx_items[1] = (struct rte_flow_item){
5424 .type = (enum rte_flow_item_type)
5425 RTE_FLOW_ITEM_TYPE_END,
5427 /* Prepare the tag action in prefix subflow. */
5428 actions_pre[index++] =
5429 (struct rte_flow_action){
5430 .type = (enum rte_flow_action_type)
5431 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5435 memcpy(actions_pre + index, actions + sample_action_pos,
5436 sizeof(struct rte_flow_action));
5438 /* For the modify action after the sample action in E-Switch mirroring,
5439 * Add the extra jump action in prefix subflow and jump into the next
5440 * table, then do the modify action in the new table.
5443 /* Prepare the prefix jump action. */
5445 jump_action = (void *)(actions_pre + actions_n + append_index);
5446 jump_action->group = jump_table;
5447 actions_pre[index++] =
5448 (struct rte_flow_action){
5449 .type = (enum rte_flow_action_type)
5450 RTE_FLOW_ACTION_TYPE_JUMP,
5451 .conf = jump_action,
5454 actions_pre[index] = (struct rte_flow_action){
5455 .type = (enum rte_flow_action_type)
5456 RTE_FLOW_ACTION_TYPE_END,
5458 /* Put the actions after sample into Suffix flow. */
5459 memcpy(actions_sfx, actions + sample_action_pos + 1,
5460 sizeof(struct rte_flow_action) *
5461 (actions_n - sample_action_pos - 1));
5466 * The splitting for metadata feature.
5468 * - Q/RSS action on NIC Rx should be split in order to pass by
5469 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5470 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5472 * - All the actions on NIC Tx should have a mreg copy action to
5473 * copy reg_a from WQE to reg_c[0].
5476 * Pointer to Ethernet device.
5478 * Parent flow structure pointer.
5480 * Flow rule attributes.
5482 * Pattern specification (list terminated by the END pattern item).
5483 * @param[in] actions
5484 * Associated actions (list terminated by the END action).
5485 * @param[in] flow_split_info
5486 * Pointer to flow split info structure.
5488 * Perform verbose error reporting if not NULL.
5490 * 0 on success, negative value otherwise
5493 flow_create_split_metadata(struct rte_eth_dev *dev,
5494 struct rte_flow *flow,
5495 const struct rte_flow_attr *attr,
5496 const struct rte_flow_item items[],
5497 const struct rte_flow_action actions[],
5498 struct mlx5_flow_split_info *flow_split_info,
5499 struct rte_flow_error *error)
5501 struct mlx5_priv *priv = dev->data->dev_private;
5502 struct mlx5_dev_config *config = &priv->config;
5503 const struct rte_flow_action *qrss = NULL;
5504 struct rte_flow_action *ext_actions = NULL;
5505 struct mlx5_flow *dev_flow = NULL;
5506 uint32_t qrss_id = 0;
5513 /* Check whether extensive metadata feature is engaged. */
5514 if (!config->dv_flow_en ||
5515 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5516 !mlx5_flow_ext_mreg_supported(dev))
5517 return flow_create_split_inner(dev, flow, NULL, attr, items,
5518 actions, flow_split_info, error);
5519 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5522 /* Exclude hairpin flows from splitting. */
5523 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5524 const struct rte_flow_action_queue *queue;
5527 if (mlx5_rxq_get_type(dev, queue->index) ==
5528 MLX5_RXQ_TYPE_HAIRPIN)
5530 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5531 const struct rte_flow_action_rss *rss;
5534 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5535 MLX5_RXQ_TYPE_HAIRPIN)
5540 /* Check if it is in meter suffix table. */
5541 mtr_sfx = attr->group == (attr->transfer ?
5542 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5543 MLX5_FLOW_TABLE_LEVEL_METER);
5545 * Q/RSS action on NIC Rx should be split in order to pass by
5546 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5547 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5549 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5550 sizeof(struct rte_flow_action_set_tag) +
5551 sizeof(struct rte_flow_action_jump);
5552 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5555 return rte_flow_error_set(error, ENOMEM,
5556 RTE_FLOW_ERROR_TYPE_ACTION,
5557 NULL, "no memory to split "
5560 * If we are the suffix flow of meter, tag already exist.
5561 * Set the tag action to void.
5564 ext_actions[qrss - actions].type =
5565 RTE_FLOW_ACTION_TYPE_VOID;
5567 ext_actions[qrss - actions].type =
5568 (enum rte_flow_action_type)
5569 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5571 * Create the new actions list with removed Q/RSS action
5572 * and appended set tag and jump to register copy table
5573 * (RX_CP_TBL). We should preallocate unique tag ID here
5574 * in advance, because it is needed for set tag action.
5576 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5577 qrss, actions_n, error);
5578 if (!mtr_sfx && !qrss_id) {
5582 } else if (attr->egress && !attr->transfer) {
5584 * All the actions on NIC Tx should have a metadata register
5585 * copy action to copy reg_a from WQE to reg_c[meta]
5587 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5588 sizeof(struct mlx5_flow_action_copy_mreg);
5589 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5592 return rte_flow_error_set(error, ENOMEM,
5593 RTE_FLOW_ERROR_TYPE_ACTION,
5594 NULL, "no memory to split "
5596 /* Create the action list appended with copy register. */
5597 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5598 actions_n, error, encap_idx);
5602 /* Add the unmodified original or prefix subflow. */
5603 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5604 items, ext_actions ? ext_actions :
5605 actions, flow_split_info, error);
5608 MLX5_ASSERT(dev_flow);
5610 const struct rte_flow_attr q_attr = {
5611 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5614 /* Internal PMD action to set register. */
5615 struct mlx5_rte_flow_item_tag q_tag_spec = {
5619 struct rte_flow_item q_items[] = {
5621 .type = (enum rte_flow_item_type)
5622 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5623 .spec = &q_tag_spec,
5628 .type = RTE_FLOW_ITEM_TYPE_END,
5631 struct rte_flow_action q_actions[] = {
5637 .type = RTE_FLOW_ACTION_TYPE_END,
5640 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5643 * Configure the tag item only if there is no meter subflow.
5644 * Since tag is already marked in the meter suffix subflow
5645 * we can just use the meter suffix items as is.
5648 /* Not meter subflow. */
5649 MLX5_ASSERT(!mtr_sfx);
5651 * Put unique id in prefix flow due to it is destroyed
5652 * after suffix flow and id will be freed after there
5653 * is no actual flows with this id and identifier
5654 * reallocation becomes possible (for example, for
5655 * other flows in other threads).
5657 dev_flow->handle->split_flow_id = qrss_id;
5658 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5662 q_tag_spec.id = ret;
5665 /* Add suffix subflow to execute Q/RSS. */
5666 flow_split_info->prefix_layers = layers;
5667 flow_split_info->prefix_mark = 0;
5668 ret = flow_create_split_inner(dev, flow, &dev_flow,
5669 &q_attr, mtr_sfx ? items :
5671 flow_split_info, error);
5674 /* qrss ID should be freed if failed. */
5676 MLX5_ASSERT(dev_flow);
5681 * We do not destroy the partially created sub_flows in case of error.
5682 * These ones are included into parent flow list and will be destroyed
5683 * by flow_drv_destroy.
5685 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5687 mlx5_free(ext_actions);
5692 * Create meter internal drop flow with the original pattern.
5695 * Pointer to Ethernet device.
5697 * Parent flow structure pointer.
5699 * Flow rule attributes.
5701 * Pattern specification (list terminated by the END pattern item).
5702 * @param[in] flow_split_info
5703 * Pointer to flow split info structure.
5705 * Pointer to flow meter structure.
5707 * Perform verbose error reporting if not NULL.
5709 * 0 on success, negative value otherwise
5712 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5713 struct rte_flow *flow,
5714 const struct rte_flow_attr *attr,
5715 const struct rte_flow_item items[],
5716 struct mlx5_flow_split_info *flow_split_info,
5717 struct mlx5_flow_meter_info *fm,
5718 struct rte_flow_error *error)
5720 struct mlx5_flow *dev_flow = NULL;
5721 struct rte_flow_attr drop_attr = *attr;
5722 struct rte_flow_action drop_actions[3];
5723 struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5725 MLX5_ASSERT(fm->drop_cnt);
5726 drop_actions[0].type =
5727 (enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5728 drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5729 drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5730 drop_actions[1].conf = NULL;
5731 drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5732 drop_actions[2].conf = NULL;
5733 drop_split_info.external = false;
5734 drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5735 drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5736 drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5737 return flow_create_split_inner(dev, flow, &dev_flow,
5738 &drop_attr, items, drop_actions,
5739 &drop_split_info, error);
5743 * The splitting for meter feature.
5745 * - The meter flow will be split to two flows as prefix and
5746 * suffix flow. The packets make sense only it pass the prefix
5749 * - Reg_C_5 is used for the packet to match betweend prefix and
5753 * Pointer to Ethernet device.
5755 * Parent flow structure pointer.
5757 * Flow rule attributes.
5759 * Pattern specification (list terminated by the END pattern item).
5760 * @param[in] actions
5761 * Associated actions (list terminated by the END action).
5762 * @param[in] flow_split_info
5763 * Pointer to flow split info structure.
5765 * Perform verbose error reporting if not NULL.
5767 * 0 on success, negative value otherwise
5770 flow_create_split_meter(struct rte_eth_dev *dev,
5771 struct rte_flow *flow,
5772 const struct rte_flow_attr *attr,
5773 const struct rte_flow_item items[],
5774 const struct rte_flow_action actions[],
5775 struct mlx5_flow_split_info *flow_split_info,
5776 struct rte_flow_error *error)
5778 struct mlx5_priv *priv = dev->data->dev_private;
5779 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5780 struct rte_flow_action *sfx_actions = NULL;
5781 struct rte_flow_action *pre_actions = NULL;
5782 struct rte_flow_item *sfx_items = NULL;
5783 struct mlx5_flow *dev_flow = NULL;
5784 struct rte_flow_attr sfx_attr = *attr;
5785 struct mlx5_flow_meter_info *fm = NULL;
5786 uint8_t skip_scale_restore;
5787 bool has_mtr = false;
5788 bool has_modify = false;
5789 bool set_mtr_reg = true;
5790 bool is_mtr_hierarchy = false;
5791 uint32_t meter_id = 0;
5792 uint32_t mtr_idx = 0;
5793 uint32_t mtr_flow_id = 0;
5800 actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5801 &has_modify, &meter_id);
5804 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5806 return rte_flow_error_set(error, EINVAL,
5807 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5808 NULL, "Meter not found.");
5810 fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5812 return rte_flow_error_set(error, EINVAL,
5813 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5814 NULL, "Meter not found.");
5815 ret = mlx5_flow_meter_attach(priv, fm,
5819 flow->meter = mtr_idx;
5823 if (!fm->def_policy) {
5824 wks->policy = mlx5_flow_meter_policy_find(dev,
5827 MLX5_ASSERT(wks->policy);
5828 if (wks->policy->is_hierarchy) {
5830 mlx5_flow_meter_hierarchy_get_final_policy(dev,
5832 if (!wks->final_policy)
5833 return rte_flow_error_set(error,
5835 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5836 "Failed to find terminal policy of hierarchy.");
5837 is_mtr_hierarchy = true;
5841 * If it isn't default-policy Meter, and
5842 * 1. There's no action in flow to change
5843 * packet (modify/encap/decap etc.), OR
5844 * 2. No drop count needed for this meter.
5845 * 3. It's not meter hierarchy.
5846 * Then no need to use regC to save meter id anymore.
5848 if (!fm->def_policy && !is_mtr_hierarchy &&
5849 (!has_modify || !fm->drop_cnt))
5850 set_mtr_reg = false;
5851 /* Prefix actions: meter, decap, encap, tag, jump, end. */
5852 act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5853 sizeof(struct mlx5_rte_flow_action_set_tag);
5854 /* Suffix items: tag, vlan, port id, end. */
5855 #define METER_SUFFIX_ITEM 4
5856 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5857 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5858 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5861 return rte_flow_error_set(error, ENOMEM,
5862 RTE_FLOW_ERROR_TYPE_ACTION,
5863 NULL, "no memory to split "
5865 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5867 /* There's no suffix flow for meter of non-default policy. */
5868 if (!fm->def_policy)
5869 pre_actions = sfx_actions + 1;
5871 pre_actions = sfx_actions + actions_n;
5872 ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
5873 items, sfx_items, actions,
5874 sfx_actions, pre_actions,
5875 (set_mtr_reg ? &mtr_flow_id : NULL),
5881 /* Add the prefix subflow. */
5882 flow_split_info->prefix_mark = 0;
5883 skip_scale_restore = flow_split_info->skip_scale;
5884 flow_split_info->skip_scale |=
5885 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5886 ret = flow_create_split_inner(dev, flow, &dev_flow,
5887 attr, items, pre_actions,
5888 flow_split_info, error);
5889 flow_split_info->skip_scale = skip_scale_restore;
5892 mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
5897 dev_flow->handle->split_flow_id = mtr_flow_id;
5898 dev_flow->handle->is_meter_flow_id = 1;
5900 if (!fm->def_policy) {
5901 if (!set_mtr_reg && fm->drop_cnt)
5903 flow_meter_create_drop_flow_with_org_pattern(dev, flow,
5909 /* Setting the sfx group atrr. */
5910 sfx_attr.group = sfx_attr.transfer ?
5911 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5912 MLX5_FLOW_TABLE_LEVEL_METER;
5913 flow_split_info->prefix_layers =
5914 flow_get_prefix_layer_flags(dev_flow);
5915 flow_split_info->prefix_mark = dev_flow->handle->mark;
5916 flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
5918 /* Add the prefix subflow. */
5919 ret = flow_create_split_metadata(dev, flow,
5920 &sfx_attr, sfx_items ?
5922 sfx_actions ? sfx_actions : actions,
5923 flow_split_info, error);
5926 mlx5_free(sfx_actions);
5931 * The splitting for sample feature.
5933 * Once Sample action is detected in the action list, the flow actions should
5934 * be split into prefix sub flow and suffix sub flow.
5936 * The original items remain in the prefix sub flow, all actions preceding the
5937 * sample action and the sample action itself will be copied to the prefix
5938 * sub flow, the actions following the sample action will be copied to the
5939 * suffix sub flow, Queue action always be located in the suffix sub flow.
5941 * In order to make the packet from prefix sub flow matches with suffix sub
5942 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5943 * flow uses tag item with the unique flow id.
5946 * Pointer to Ethernet device.
5948 * Parent flow structure pointer.
5950 * Flow rule attributes.
5952 * Pattern specification (list terminated by the END pattern item).
5953 * @param[in] actions
5954 * Associated actions (list terminated by the END action).
5955 * @param[in] flow_split_info
5956 * Pointer to flow split info structure.
5958 * Perform verbose error reporting if not NULL.
5960 * 0 on success, negative value otherwise
5963 flow_create_split_sample(struct rte_eth_dev *dev,
5964 struct rte_flow *flow,
5965 const struct rte_flow_attr *attr,
5966 const struct rte_flow_item items[],
5967 const struct rte_flow_action actions[],
5968 struct mlx5_flow_split_info *flow_split_info,
5969 struct rte_flow_error *error)
5971 struct mlx5_priv *priv = dev->data->dev_private;
5972 struct rte_flow_action *sfx_actions = NULL;
5973 struct rte_flow_action *pre_actions = NULL;
5974 struct rte_flow_item *sfx_items = NULL;
5975 struct mlx5_flow *dev_flow = NULL;
5976 struct rte_flow_attr sfx_attr = *attr;
5977 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5978 struct mlx5_flow_dv_sample_resource *sample_res;
5979 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5980 struct mlx5_flow_tbl_resource *sfx_tbl;
5984 uint32_t fdb_tx = 0;
5987 int sample_action_pos;
5988 int qrss_action_pos;
5990 int modify_after_mirror = 0;
5991 uint16_t jump_table = 0;
5992 const uint32_t next_ft_step = 1;
5995 if (priv->sampler_en)
5996 actions_n = flow_check_match_action(actions, attr,
5997 RTE_FLOW_ACTION_TYPE_SAMPLE,
5998 &sample_action_pos, &qrss_action_pos,
5999 &modify_after_mirror);
6001 /* The prefix actions must includes sample, tag, end. */
6002 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6003 + sizeof(struct mlx5_rte_flow_action_set_tag);
6004 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6005 sizeof(struct mlx5_rte_flow_item_tag) * 2;
6006 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6007 item_size), 0, SOCKET_ID_ANY);
6009 return rte_flow_error_set(error, ENOMEM,
6010 RTE_FLOW_ERROR_TYPE_ACTION,
6011 NULL, "no memory to split "
6013 /* The representor_id is UINT16_MAX for uplink. */
6014 fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6016 * When reg_c_preserve is set, metadata registers Cx preserve
6017 * their value even through packet duplication.
6019 add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6021 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6023 if (modify_after_mirror)
6024 jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6026 pre_actions = sfx_actions + actions_n;
6027 tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6028 actions, sfx_actions,
6029 pre_actions, actions_n,
6031 qrss_action_pos, jump_table,
6033 if (tag_id < 0 || (add_tag && !tag_id)) {
6037 if (modify_after_mirror)
6038 flow_split_info->skip_scale =
6039 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6040 /* Add the prefix subflow. */
6041 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6043 flow_split_info, error);
6048 dev_flow->handle->split_flow_id = tag_id;
6049 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6050 if (!modify_after_mirror) {
6051 /* Set the sfx group attr. */
6052 sample_res = (struct mlx5_flow_dv_sample_resource *)
6053 dev_flow->dv.sample_res;
6054 sfx_tbl = (struct mlx5_flow_tbl_resource *)
6055 sample_res->normal_path_tbl;
6056 sfx_tbl_data = container_of(sfx_tbl,
6057 struct mlx5_flow_tbl_data_entry,
6059 sfx_attr.group = sfx_attr.transfer ?
6060 (sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6062 MLX5_ASSERT(attr->transfer);
6063 sfx_attr.group = jump_table;
6065 flow_split_info->prefix_layers =
6066 flow_get_prefix_layer_flags(dev_flow);
6067 flow_split_info->prefix_mark = dev_flow->handle->mark;
6068 /* Suffix group level already be scaled with factor, set
6069 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6070 * again in translation.
6072 flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6075 /* Add the suffix subflow. */
6076 ret = flow_create_split_meter(dev, flow, &sfx_attr,
6077 sfx_items ? sfx_items : items,
6078 sfx_actions ? sfx_actions : actions,
6079 flow_split_info, error);
6082 mlx5_free(sfx_actions);
6087 * Split the flow to subflow set. The splitters might be linked
6088 * in the chain, like this:
6089 * flow_create_split_outer() calls:
6090 * flow_create_split_meter() calls:
6091 * flow_create_split_metadata(meter_subflow_0) calls:
6092 * flow_create_split_inner(metadata_subflow_0)
6093 * flow_create_split_inner(metadata_subflow_1)
6094 * flow_create_split_inner(metadata_subflow_2)
6095 * flow_create_split_metadata(meter_subflow_1) calls:
6096 * flow_create_split_inner(metadata_subflow_0)
6097 * flow_create_split_inner(metadata_subflow_1)
6098 * flow_create_split_inner(metadata_subflow_2)
6100 * This provide flexible way to add new levels of flow splitting.
6101 * The all of successfully created subflows are included to the
6102 * parent flow dev_flow list.
6105 * Pointer to Ethernet device.
6107 * Parent flow structure pointer.
6109 * Flow rule attributes.
6111 * Pattern specification (list terminated by the END pattern item).
6112 * @param[in] actions
6113 * Associated actions (list terminated by the END action).
6114 * @param[in] flow_split_info
6115 * Pointer to flow split info structure.
6117 * Perform verbose error reporting if not NULL.
6119 * 0 on success, negative value otherwise
6122 flow_create_split_outer(struct rte_eth_dev *dev,
6123 struct rte_flow *flow,
6124 const struct rte_flow_attr *attr,
6125 const struct rte_flow_item items[],
6126 const struct rte_flow_action actions[],
6127 struct mlx5_flow_split_info *flow_split_info,
6128 struct rte_flow_error *error)
6132 ret = flow_create_split_sample(dev, flow, attr, items,
6133 actions, flow_split_info, error);
6134 MLX5_ASSERT(ret <= 0);
6138 static inline struct mlx5_flow_tunnel *
6139 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6141 struct mlx5_flow_tunnel *tunnel;
6143 #pragma GCC diagnostic push
6144 #pragma GCC diagnostic ignored "-Wcast-qual"
6145 tunnel = (typeof(tunnel))flow->tunnel;
6146 #pragma GCC diagnostic pop
6152 * Adjust flow RSS workspace if needed.
6155 * Pointer to thread flow work space.
6157 * Pointer to RSS descriptor.
6158 * @param[in] nrssq_num
6159 * New RSS queue number.
6162 * 0 on success, -1 otherwise and rte_errno is set.
6165 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6166 struct mlx5_flow_rss_desc *rss_desc,
6169 if (likely(nrssq_num <= wks->rssq_num))
6171 rss_desc->queue = realloc(rss_desc->queue,
6172 sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6173 if (!rss_desc->queue) {
6177 wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6182 * Create a flow and add it to @p list.
6185 * Pointer to Ethernet device.
6187 * Pointer to a TAILQ flow list. If this parameter NULL,
6188 * no list insertion occurred, flow is just created,
6189 * this is caller's responsibility to track the
6192 * Flow rule attributes.
6194 * Pattern specification (list terminated by the END pattern item).
6195 * @param[in] actions
6196 * Associated actions (list terminated by the END action).
6197 * @param[in] external
6198 * This flow rule is created by request external to PMD.
6200 * Perform verbose error reporting if not NULL.
6203 * A flow index on success, 0 otherwise and rte_errno is set.
6206 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
6207 const struct rte_flow_attr *attr,
6208 const struct rte_flow_item items[],
6209 const struct rte_flow_action original_actions[],
6210 bool external, struct rte_flow_error *error)
6212 struct mlx5_priv *priv = dev->data->dev_private;
6213 struct rte_flow *flow = NULL;
6214 struct mlx5_flow *dev_flow;
6215 const struct rte_flow_action_rss *rss = NULL;
6216 struct mlx5_translated_action_handle
6217 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6218 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6220 struct mlx5_flow_expand_rss buf;
6221 uint8_t buffer[2048];
6224 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6225 uint8_t buffer[2048];
6228 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6229 uint8_t buffer[2048];
6230 } actions_hairpin_tx;
6232 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6233 uint8_t buffer[2048];
6235 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6236 struct mlx5_flow_rss_desc *rss_desc;
6237 const struct rte_flow_action *p_actions_rx;
6241 struct rte_flow_attr attr_tx = { .priority = 0 };
6242 const struct rte_flow_action *actions;
6243 struct rte_flow_action *translated_actions = NULL;
6244 struct mlx5_flow_tunnel *tunnel;
6245 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6246 struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6247 struct mlx5_flow_split_info flow_split_info = {
6248 .external = !!external,
6258 rss_desc = &wks->rss_desc;
6259 ret = flow_action_handles_translate(dev, original_actions,
6262 &translated_actions, error);
6264 MLX5_ASSERT(translated_actions == NULL);
6267 actions = translated_actions ? translated_actions : original_actions;
6268 p_actions_rx = actions;
6269 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6270 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6271 external, hairpin_flow, error);
6273 goto error_before_hairpin_split;
6274 flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
6277 goto error_before_hairpin_split;
6279 if (hairpin_flow > 0) {
6280 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6282 goto error_before_hairpin_split;
6284 flow_hairpin_split(dev, actions, actions_rx.actions,
6285 actions_hairpin_tx.actions, items_tx.items,
6287 p_actions_rx = actions_rx.actions;
6289 flow_split_info.flow_idx = idx;
6290 flow->drv_type = flow_get_drv_type(dev, attr);
6291 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6292 flow->drv_type < MLX5_FLOW_TYPE_MAX);
6293 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6294 /* RSS Action only works on NIC RX domain */
6295 if (attr->ingress && !attr->transfer)
6296 rss = flow_get_rss_action(dev, p_actions_rx);
6298 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6301 * The following information is required by
6302 * mlx5_flow_hashfields_adjust() in advance.
6304 rss_desc->level = rss->level;
6305 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6306 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6308 flow->dev_handles = 0;
6309 if (rss && rss->types) {
6310 unsigned int graph_root;
6312 graph_root = find_graph_root(items, rss->level);
6313 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6315 mlx5_support_expansion, graph_root);
6316 MLX5_ASSERT(ret > 0 &&
6317 (unsigned int)ret < sizeof(expand_buffer.buffer));
6318 if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6319 for (i = 0; i < buf->entries; ++i)
6320 mlx5_dbg__print_pattern(buf->entry[i].pattern);
6324 buf->entry[0].pattern = (void *)(uintptr_t)items;
6326 rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6328 for (i = 0; i < buf->entries; ++i) {
6329 /* Initialize flow split data. */
6330 flow_split_info.prefix_layers = 0;
6331 flow_split_info.prefix_mark = 0;
6332 flow_split_info.skip_scale = 0;
6334 * The splitter may create multiple dev_flows,
6335 * depending on configuration. In the simplest
6336 * case it just creates unmodified original flow.
6338 ret = flow_create_split_outer(dev, flow, attr,
6339 buf->entry[i].pattern,
6340 p_actions_rx, &flow_split_info,
6344 if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6345 ret = flow_tunnel_add_default_miss(dev, flow, attr,
6348 wks->flows[0].tunnel,
6352 mlx5_free(default_miss_ctx.queue);
6357 /* Create the tx flow. */
6359 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6360 attr_tx.ingress = 0;
6362 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6363 actions_hairpin_tx.actions,
6367 dev_flow->flow = flow;
6368 dev_flow->external = 0;
6369 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6370 dev_flow->handle, next);
6371 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6373 actions_hairpin_tx.actions, error);
6378 * Update the metadata register copy table. If extensive
6379 * metadata feature is enabled and registers are supported
6380 * we might create the extra rte_flow for each unique
6381 * MARK/FLAG action ID.
6383 * The table is updated for ingress Flows only, because
6384 * the egress Flows belong to the different device and
6385 * copy table should be updated in peer NIC Rx domain.
6387 if (attr->ingress &&
6388 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6389 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6394 * If the flow is external (from application) OR device is started,
6395 * OR mreg discover, then apply immediately.
6397 if (external || dev->data->dev_started ||
6398 (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6399 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6400 ret = flow_drv_apply(dev, flow, error);
6405 rte_spinlock_lock(&priv->flow_list_lock);
6406 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
6408 rte_spinlock_unlock(&priv->flow_list_lock);
6410 flow_rxq_flags_set(dev, flow);
6411 rte_free(translated_actions);
6412 tunnel = flow_tunnel_from_rule(wks->flows);
6415 flow->tunnel_id = tunnel->tunnel_id;
6416 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6417 mlx5_free(default_miss_ctx.queue);
6419 mlx5_flow_pop_thread_workspace();
6423 ret = rte_errno; /* Save rte_errno before cleanup. */
6424 flow_mreg_del_copy_action(dev, flow);
6425 flow_drv_destroy(dev, flow);
6426 if (rss_desc->shared_rss)
6427 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6429 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6430 rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6431 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
6432 rte_errno = ret; /* Restore rte_errno. */
6435 mlx5_flow_pop_thread_workspace();
6436 error_before_hairpin_split:
6437 rte_free(translated_actions);
6442 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6443 * incoming packets to table 1.
6445 * Other flow rules, requested for group n, will be created in
6446 * e-switch table n+1.
6447 * Jump action to e-switch group n will be created to group n+1.
6449 * Used when working in switchdev mode, to utilise advantages of table 1
6453 * Pointer to Ethernet device.
6456 * Pointer to flow on success, NULL otherwise and rte_errno is set.
6459 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6461 const struct rte_flow_attr attr = {
6468 const struct rte_flow_item pattern = {
6469 .type = RTE_FLOW_ITEM_TYPE_END,
6471 struct rte_flow_action_jump jump = {
6474 const struct rte_flow_action actions[] = {
6476 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6480 .type = RTE_FLOW_ACTION_TYPE_END,
6483 struct mlx5_priv *priv = dev->data->dev_private;
6484 struct rte_flow_error error;
6486 return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
6488 actions, false, &error);
6492 * Validate a flow supported by the NIC.
6494 * @see rte_flow_validate()
6498 mlx5_flow_validate(struct rte_eth_dev *dev,
6499 const struct rte_flow_attr *attr,
6500 const struct rte_flow_item items[],
6501 const struct rte_flow_action original_actions[],
6502 struct rte_flow_error *error)
6505 struct mlx5_translated_action_handle
6506 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6507 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6508 const struct rte_flow_action *actions;
6509 struct rte_flow_action *translated_actions = NULL;
6510 int ret = flow_action_handles_translate(dev, original_actions,
6513 &translated_actions, error);
6517 actions = translated_actions ? translated_actions : original_actions;
6518 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6519 ret = flow_drv_validate(dev, attr, items, actions,
6520 true, hairpin_flow, error);
6521 rte_free(translated_actions);
6528 * @see rte_flow_create()
6532 mlx5_flow_create(struct rte_eth_dev *dev,
6533 const struct rte_flow_attr *attr,
6534 const struct rte_flow_item items[],
6535 const struct rte_flow_action actions[],
6536 struct rte_flow_error *error)
6538 struct mlx5_priv *priv = dev->data->dev_private;
6541 * If the device is not started yet, it is not allowed to created a
6542 * flow from application. PMD default flows and traffic control flows
6545 if (unlikely(!dev->data->dev_started)) {
6546 DRV_LOG(DEBUG, "port %u is not started when "
6547 "inserting a flow", dev->data->port_id);
6548 rte_flow_error_set(error, ENODEV,
6549 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6551 "port not started");
6555 return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
6556 attr, items, actions, true, error);
6560 * Destroy a flow in a list.
6563 * Pointer to Ethernet device.
6565 * Pointer to the Indexed flow list. If this parameter NULL,
6566 * there is no flow removal from the list. Be noted that as
6567 * flow is add to the indexed list, memory of the indexed
6568 * list points to maybe changed as flow destroyed.
6569 * @param[in] flow_idx
6570 * Index of flow to destroy.
6573 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
6576 struct mlx5_priv *priv = dev->data->dev_private;
6577 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6578 [MLX5_IPOOL_RTE_FLOW], flow_idx);
6583 * Update RX queue flags only if port is started, otherwise it is
6586 if (dev->data->dev_started)
6587 flow_rxq_flags_trim(dev, flow);
6588 flow_drv_destroy(dev, flow);
6590 rte_spinlock_lock(&priv->flow_list_lock);
6591 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
6592 flow_idx, flow, next);
6593 rte_spinlock_unlock(&priv->flow_list_lock);
6596 struct mlx5_flow_tunnel *tunnel;
6598 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6600 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6601 mlx5_flow_tunnel_free(dev, tunnel);
6603 flow_mreg_del_copy_action(dev, flow);
6604 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6608 * Destroy all flows.
6611 * Pointer to Ethernet device.
6613 * Pointer to the Indexed flow list.
6615 * If flushing is called avtively.
6618 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
6620 uint32_t num_flushed = 0;
6623 flow_list_destroy(dev, list, *list);
6627 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6628 dev->data->port_id, num_flushed);
6633 * Stop all default actions for flows.
6636 * Pointer to Ethernet device.
6639 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6641 flow_mreg_del_default_copy_action(dev);
6642 flow_rxq_flags_clear(dev);
6646 * Start all default actions for flows.
6649 * Pointer to Ethernet device.
6651 * 0 on success, a negative errno value otherwise and rte_errno is set.
6654 mlx5_flow_start_default(struct rte_eth_dev *dev)
6656 struct rte_flow_error error;
6658 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6659 return flow_mreg_add_default_copy_action(dev, &error);
6663 * Release key of thread specific flow workspace data.
6666 flow_release_workspace(void *data)
6668 struct mlx5_flow_workspace *wks = data;
6669 struct mlx5_flow_workspace *next;
6673 free(wks->rss_desc.queue);
6680 * Get thread specific current flow workspace.
6682 * @return pointer to thread specific flow workspace data, NULL on error.
6684 struct mlx5_flow_workspace*
6685 mlx5_flow_get_thread_workspace(void)
6687 struct mlx5_flow_workspace *data;
6689 data = mlx5_flow_os_get_specific_workspace();
6690 MLX5_ASSERT(data && data->inuse);
6691 if (!data || !data->inuse)
6692 DRV_LOG(ERR, "flow workspace not initialized.");
6697 * Allocate and init new flow workspace.
6699 * @return pointer to flow workspace data, NULL on error.
6701 static struct mlx5_flow_workspace*
6702 flow_alloc_thread_workspace(void)
6704 struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6707 DRV_LOG(ERR, "Failed to allocate flow workspace "
6711 data->rss_desc.queue = calloc(1,
6712 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6713 if (!data->rss_desc.queue)
6715 data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6718 if (data->rss_desc.queue)
6719 free(data->rss_desc.queue);
6725 * Get new thread specific flow workspace.
6727 * If current workspace inuse, create new one and set as current.
6729 * @return pointer to thread specific flow workspace data, NULL on error.
6731 static struct mlx5_flow_workspace*
6732 mlx5_flow_push_thread_workspace(void)
6734 struct mlx5_flow_workspace *curr;
6735 struct mlx5_flow_workspace *data;
6737 curr = mlx5_flow_os_get_specific_workspace();
6739 data = flow_alloc_thread_workspace();
6742 } else if (!curr->inuse) {
6744 } else if (curr->next) {
6747 data = flow_alloc_thread_workspace();
6755 /* Set as current workspace */
6756 if (mlx5_flow_os_set_specific_workspace(data))
6757 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6762 * Close current thread specific flow workspace.
6764 * If previous workspace available, set it as current.
6766 * @return pointer to thread specific flow workspace data, NULL on error.
6769 mlx5_flow_pop_thread_workspace(void)
6771 struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6776 DRV_LOG(ERR, "Failed to close unused flow workspace.");
6782 if (mlx5_flow_os_set_specific_workspace(data->prev))
6783 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6787 * Verify the flow list is empty
6790 * Pointer to Ethernet device.
6792 * @return the number of flows not released.
6795 mlx5_flow_verify(struct rte_eth_dev *dev)
6797 struct mlx5_priv *priv = dev->data->dev_private;
6798 struct rte_flow *flow;
6802 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6804 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6805 dev->data->port_id, (void *)flow);
6812 * Enable default hairpin egress flow.
6815 * Pointer to Ethernet device.
6820 * 0 on success, a negative errno value otherwise and rte_errno is set.
6823 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6826 struct mlx5_priv *priv = dev->data->dev_private;
6827 const struct rte_flow_attr attr = {
6831 struct mlx5_rte_flow_item_tx_queue queue_spec = {
6834 struct mlx5_rte_flow_item_tx_queue queue_mask = {
6835 .queue = UINT32_MAX,
6837 struct rte_flow_item items[] = {
6839 .type = (enum rte_flow_item_type)
6840 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6841 .spec = &queue_spec,
6843 .mask = &queue_mask,
6846 .type = RTE_FLOW_ITEM_TYPE_END,
6849 struct rte_flow_action_jump jump = {
6850 .group = MLX5_HAIRPIN_TX_TABLE,
6852 struct rte_flow_action actions[2];
6854 struct rte_flow_error error;
6856 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6857 actions[0].conf = &jump;
6858 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6859 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6860 &attr, items, actions, false, &error);
6863 "Failed to create ctrl flow: rte_errno(%d),"
6864 " type(%d), message(%s)",
6865 rte_errno, error.type,
6866 error.message ? error.message : " (no stated reason)");
6873 * Enable a control flow configured from the control plane.
6876 * Pointer to Ethernet device.
6878 * An Ethernet flow spec to apply.
6880 * An Ethernet flow mask to apply.
6882 * A VLAN flow spec to apply.
6884 * A VLAN flow mask to apply.
6887 * 0 on success, a negative errno value otherwise and rte_errno is set.
6890 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6891 struct rte_flow_item_eth *eth_spec,
6892 struct rte_flow_item_eth *eth_mask,
6893 struct rte_flow_item_vlan *vlan_spec,
6894 struct rte_flow_item_vlan *vlan_mask)
6896 struct mlx5_priv *priv = dev->data->dev_private;
6897 const struct rte_flow_attr attr = {
6899 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6901 struct rte_flow_item items[] = {
6903 .type = RTE_FLOW_ITEM_TYPE_ETH,
6909 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6910 RTE_FLOW_ITEM_TYPE_END,
6916 .type = RTE_FLOW_ITEM_TYPE_END,
6919 uint16_t queue[priv->reta_idx_n];
6920 struct rte_flow_action_rss action_rss = {
6921 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6923 .types = priv->rss_conf.rss_hf,
6924 .key_len = priv->rss_conf.rss_key_len,
6925 .queue_num = priv->reta_idx_n,
6926 .key = priv->rss_conf.rss_key,
6929 struct rte_flow_action actions[] = {
6931 .type = RTE_FLOW_ACTION_TYPE_RSS,
6932 .conf = &action_rss,
6935 .type = RTE_FLOW_ACTION_TYPE_END,
6939 struct rte_flow_error error;
6942 if (!priv->reta_idx_n || !priv->rxqs_n) {
6945 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6946 action_rss.types = 0;
6947 for (i = 0; i != priv->reta_idx_n; ++i)
6948 queue[i] = (*priv->reta_idx)[i];
6949 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6950 &attr, items, actions, false, &error);
6957 * Enable a flow control configured from the control plane.
6960 * Pointer to Ethernet device.
6962 * An Ethernet flow spec to apply.
6964 * An Ethernet flow mask to apply.
6967 * 0 on success, a negative errno value otherwise and rte_errno is set.
6970 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6971 struct rte_flow_item_eth *eth_spec,
6972 struct rte_flow_item_eth *eth_mask)
6974 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6978 * Create default miss flow rule matching lacp traffic
6981 * Pointer to Ethernet device.
6983 * An Ethernet flow spec to apply.
6986 * 0 on success, a negative errno value otherwise and rte_errno is set.
6989 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6991 struct mlx5_priv *priv = dev->data->dev_private;
6993 * The LACP matching is done by only using ether type since using
6994 * a multicast dst mac causes kernel to give low priority to this flow.
6996 static const struct rte_flow_item_eth lacp_spec = {
6997 .type = RTE_BE16(0x8809),
6999 static const struct rte_flow_item_eth lacp_mask = {
7002 const struct rte_flow_attr attr = {
7005 struct rte_flow_item items[] = {
7007 .type = RTE_FLOW_ITEM_TYPE_ETH,
7012 .type = RTE_FLOW_ITEM_TYPE_END,
7015 struct rte_flow_action actions[] = {
7017 .type = (enum rte_flow_action_type)
7018 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7021 .type = RTE_FLOW_ACTION_TYPE_END,
7024 struct rte_flow_error error;
7025 uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
7026 &attr, items, actions, false, &error);
7036 * @see rte_flow_destroy()
7040 mlx5_flow_destroy(struct rte_eth_dev *dev,
7041 struct rte_flow *flow,
7042 struct rte_flow_error *error __rte_unused)
7044 struct mlx5_priv *priv = dev->data->dev_private;
7046 flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
7051 * Destroy all flows.
7053 * @see rte_flow_flush()
7057 mlx5_flow_flush(struct rte_eth_dev *dev,
7058 struct rte_flow_error *error __rte_unused)
7060 struct mlx5_priv *priv = dev->data->dev_private;
7062 mlx5_flow_list_flush(dev, &priv->flows, false);
7069 * @see rte_flow_isolate()
7073 mlx5_flow_isolate(struct rte_eth_dev *dev,
7075 struct rte_flow_error *error)
7077 struct mlx5_priv *priv = dev->data->dev_private;
7079 if (dev->data->dev_started) {
7080 rte_flow_error_set(error, EBUSY,
7081 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7083 "port must be stopped first");
7086 priv->isolated = !!enable;
7088 dev->dev_ops = &mlx5_dev_ops_isolate;
7090 dev->dev_ops = &mlx5_dev_ops;
7092 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7093 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7101 * @see rte_flow_query()
7105 flow_drv_query(struct rte_eth_dev *dev,
7107 const struct rte_flow_action *actions,
7109 struct rte_flow_error *error)
7111 struct mlx5_priv *priv = dev->data->dev_private;
7112 const struct mlx5_flow_driver_ops *fops;
7113 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
7114 [MLX5_IPOOL_RTE_FLOW],
7116 enum mlx5_flow_drv_type ftype;
7119 return rte_flow_error_set(error, ENOENT,
7120 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7122 "invalid flow handle");
7124 ftype = flow->drv_type;
7125 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7126 fops = flow_get_drv_ops(ftype);
7128 return fops->query(dev, flow, actions, data, error);
7134 * @see rte_flow_query()
7138 mlx5_flow_query(struct rte_eth_dev *dev,
7139 struct rte_flow *flow,
7140 const struct rte_flow_action *actions,
7142 struct rte_flow_error *error)
7146 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7154 * Get rte_flow callbacks.
7157 * Pointer to Ethernet device structure.
7159 * Pointer to operation-specific structure.
7164 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7165 const struct rte_flow_ops **ops)
7167 *ops = &mlx5_flow_ops;
7172 * Validate meter policy actions.
7173 * Dispatcher for action type specific validation.
7176 * Pointer to the Ethernet device structure.
7178 * The meter policy action object to validate.
7180 * Attributes of flow to determine steering domain.
7181 * @param[out] is_rss
7183 * @param[out] domain_bitmap
7185 * @param[out] is_def_policy
7186 * Is default policy or not.
7188 * Perform verbose error reporting if not NULL. Initialized in case of
7192 * 0 on success, otherwise negative errno value.
7195 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7196 const struct rte_flow_action *actions[RTE_COLORS],
7197 struct rte_flow_attr *attr,
7199 uint8_t *domain_bitmap,
7200 bool *is_def_policy,
7201 struct rte_mtr_error *error)
7203 const struct mlx5_flow_driver_ops *fops;
7205 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7206 return fops->validate_mtr_acts(dev, actions, attr,
7207 is_rss, domain_bitmap, is_def_policy, error);
7211 * Destroy the meter table set.
7214 * Pointer to Ethernet device.
7215 * @param[in] mtr_policy
7216 * Meter policy struct.
7219 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7220 struct mlx5_flow_meter_policy *mtr_policy)
7222 const struct mlx5_flow_driver_ops *fops;
7224 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7225 fops->destroy_mtr_acts(dev, mtr_policy);
7229 * Create policy action, lock free,
7230 * (mutex should be acquired by caller).
7231 * Dispatcher for action type specific call.
7234 * Pointer to the Ethernet device structure.
7235 * @param[in] mtr_policy
7236 * Meter policy struct.
7238 * Action specification used to create meter actions.
7240 * Perform verbose error reporting if not NULL. Initialized in case of
7244 * 0 on success, otherwise negative errno value.
7247 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7248 struct mlx5_flow_meter_policy *mtr_policy,
7249 const struct rte_flow_action *actions[RTE_COLORS],
7250 struct rte_mtr_error *error)
7252 const struct mlx5_flow_driver_ops *fops;
7254 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7255 return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7259 * Create policy rules, lock free,
7260 * (mutex should be acquired by caller).
7261 * Dispatcher for action type specific call.
7264 * Pointer to the Ethernet device structure.
7265 * @param[in] mtr_policy
7266 * Meter policy struct.
7269 * 0 on success, -1 otherwise.
7272 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7273 struct mlx5_flow_meter_policy *mtr_policy)
7275 const struct mlx5_flow_driver_ops *fops;
7277 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7278 return fops->create_policy_rules(dev, mtr_policy);
7282 * Destroy policy rules, lock free,
7283 * (mutex should be acquired by caller).
7284 * Dispatcher for action type specific call.
7287 * Pointer to the Ethernet device structure.
7288 * @param[in] mtr_policy
7289 * Meter policy struct.
7292 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7293 struct mlx5_flow_meter_policy *mtr_policy)
7295 const struct mlx5_flow_driver_ops *fops;
7297 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7298 fops->destroy_policy_rules(dev, mtr_policy);
7302 * Destroy the default policy table set.
7305 * Pointer to Ethernet device.
7308 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7310 const struct mlx5_flow_driver_ops *fops;
7312 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7313 fops->destroy_def_policy(dev);
7317 * Destroy the default policy table set.
7320 * Pointer to Ethernet device.
7323 * 0 on success, -1 otherwise.
7326 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7328 const struct mlx5_flow_driver_ops *fops;
7330 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7331 return fops->create_def_policy(dev);
7335 * Create the needed meter and suffix tables.
7338 * Pointer to Ethernet device.
7341 * 0 on success, -1 otherwise.
7344 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7345 struct mlx5_flow_meter_info *fm,
7347 uint8_t domain_bitmap)
7349 const struct mlx5_flow_driver_ops *fops;
7351 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7352 return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7356 * Destroy the meter table set.
7359 * Pointer to Ethernet device.
7361 * Pointer to the meter table set.
7364 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7365 struct mlx5_flow_meter_info *fm)
7367 const struct mlx5_flow_driver_ops *fops;
7369 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7370 fops->destroy_mtr_tbls(dev, fm);
7374 * Destroy the global meter drop table.
7377 * Pointer to Ethernet device.
7380 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7382 const struct mlx5_flow_driver_ops *fops;
7384 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7385 fops->destroy_mtr_drop_tbls(dev);
7389 * Destroy the sub policy table with RX queue.
7392 * Pointer to Ethernet device.
7393 * @param[in] mtr_policy
7394 * Pointer to meter policy table.
7397 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7398 struct mlx5_flow_meter_policy *mtr_policy)
7400 const struct mlx5_flow_driver_ops *fops;
7402 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7403 fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7407 * Allocate the needed aso flow meter id.
7410 * Pointer to Ethernet device.
7413 * Index to aso flow meter on success, NULL otherwise.
7416 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7418 const struct mlx5_flow_driver_ops *fops;
7420 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7421 return fops->create_meter(dev);
7425 * Free the aso flow meter id.
7428 * Pointer to Ethernet device.
7429 * @param[in] mtr_idx
7430 * Index to aso flow meter to be free.
7436 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7438 const struct mlx5_flow_driver_ops *fops;
7440 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7441 fops->free_meter(dev, mtr_idx);
7445 * Allocate a counter.
7448 * Pointer to Ethernet device structure.
7451 * Index to allocated counter on success, 0 otherwise.
7454 mlx5_counter_alloc(struct rte_eth_dev *dev)
7456 const struct mlx5_flow_driver_ops *fops;
7457 struct rte_flow_attr attr = { .transfer = 0 };
7459 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7460 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7461 return fops->counter_alloc(dev);
7464 "port %u counter allocate is not supported.",
7465 dev->data->port_id);
7473 * Pointer to Ethernet device structure.
7475 * Index to counter to be free.
7478 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7480 const struct mlx5_flow_driver_ops *fops;
7481 struct rte_flow_attr attr = { .transfer = 0 };
7483 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7484 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7485 fops->counter_free(dev, cnt);
7489 "port %u counter free is not supported.",
7490 dev->data->port_id);
7494 * Query counter statistics.
7497 * Pointer to Ethernet device structure.
7499 * Index to counter to query.
7501 * Set to clear counter statistics.
7503 * The counter hits packets number to save.
7505 * The counter hits bytes number to save.
7508 * 0 on success, a negative errno value otherwise.
7511 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7512 bool clear, uint64_t *pkts, uint64_t *bytes)
7514 const struct mlx5_flow_driver_ops *fops;
7515 struct rte_flow_attr attr = { .transfer = 0 };
7517 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7518 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7519 return fops->counter_query(dev, cnt, clear, pkts, bytes);
7522 "port %u counter query is not supported.",
7523 dev->data->port_id);
7528 * Allocate a new memory for the counter values wrapped by all the needed
7532 * Pointer to mlx5_dev_ctx_shared object.
7535 * 0 on success, a negative errno value otherwise.
7538 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7540 struct mlx5_devx_mkey_attr mkey_attr;
7541 struct mlx5_counter_stats_mem_mng *mem_mng;
7542 volatile struct flow_counter_stats *raw_data;
7543 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7544 int size = (sizeof(struct flow_counter_stats) *
7545 MLX5_COUNTERS_PER_POOL +
7546 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7547 sizeof(struct mlx5_counter_stats_mem_mng);
7548 size_t pgsize = rte_mem_page_size();
7552 if (pgsize == (size_t)-1) {
7553 DRV_LOG(ERR, "Failed to get mem page size");
7557 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7562 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7563 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7564 mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7565 IBV_ACCESS_LOCAL_WRITE);
7566 if (!mem_mng->umem) {
7571 memset(&mkey_attr, 0, sizeof(mkey_attr));
7572 mkey_attr.addr = (uintptr_t)mem;
7573 mkey_attr.size = size;
7574 mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7575 mkey_attr.pd = sh->pdn;
7576 mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7577 mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7578 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7580 mlx5_os_umem_dereg(mem_mng->umem);
7585 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7586 raw_data = (volatile struct flow_counter_stats *)mem;
7587 for (i = 0; i < raws_n; ++i) {
7588 mem_mng->raws[i].mem_mng = mem_mng;
7589 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7591 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7592 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7593 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7595 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7596 sh->cmng.mem_mng = mem_mng;
7601 * Set the statistic memory to the new counter pool.
7604 * Pointer to mlx5_dev_ctx_shared object.
7606 * Pointer to the pool to set the statistic memory.
7609 * 0 on success, a negative errno value otherwise.
7612 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7613 struct mlx5_flow_counter_pool *pool)
7615 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7616 /* Resize statistic memory once used out. */
7617 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7618 mlx5_flow_create_counter_stat_mem_mng(sh)) {
7619 DRV_LOG(ERR, "Cannot resize counter stat mem.");
7622 rte_spinlock_lock(&pool->sl);
7623 pool->raw = cmng->mem_mng->raws + pool->index %
7624 MLX5_CNT_CONTAINER_RESIZE;
7625 rte_spinlock_unlock(&pool->sl);
7626 pool->raw_hw = NULL;
7630 #define MLX5_POOL_QUERY_FREQ_US 1000000
7633 * Set the periodic procedure for triggering asynchronous batch queries for all
7634 * the counter pools.
7637 * Pointer to mlx5_dev_ctx_shared object.
7640 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7642 uint32_t pools_n, us;
7644 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7645 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7646 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7647 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7648 sh->cmng.query_thread_on = 0;
7649 DRV_LOG(ERR, "Cannot reinitialize query alarm");
7651 sh->cmng.query_thread_on = 1;
7656 * The periodic procedure for triggering asynchronous batch queries for all the
7657 * counter pools. This function is probably called by the host thread.
7660 * The parameter for the alarm process.
7663 mlx5_flow_query_alarm(void *arg)
7665 struct mlx5_dev_ctx_shared *sh = arg;
7667 uint16_t pool_index = sh->cmng.pool_index;
7668 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7669 struct mlx5_flow_counter_pool *pool;
7672 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7674 rte_spinlock_lock(&cmng->pool_update_sl);
7675 pool = cmng->pools[pool_index];
7676 n_valid = cmng->n_valid;
7677 rte_spinlock_unlock(&cmng->pool_update_sl);
7678 /* Set the statistic memory to the new created pool. */
7679 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7682 /* There is a pool query in progress. */
7685 LIST_FIRST(&sh->cmng.free_stat_raws);
7687 /* No free counter statistics raw memory. */
7690 * Identify the counters released between query trigger and query
7691 * handle more efficiently. The counter released in this gap period
7692 * should wait for a new round of query as the new arrived packets
7693 * will not be taken into account.
7696 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7697 MLX5_COUNTERS_PER_POOL,
7699 pool->raw_hw->mem_mng->dm->id,
7703 (uint64_t)(uintptr_t)pool);
7705 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7706 " %d", pool->min_dcs->id);
7707 pool->raw_hw = NULL;
7710 LIST_REMOVE(pool->raw_hw, next);
7711 sh->cmng.pending_queries++;
7713 if (pool_index >= n_valid)
7716 sh->cmng.pool_index = pool_index;
7717 mlx5_set_query_alarm(sh);
7721 * Check and callback event for new aged flow in the counter pool
7724 * Pointer to mlx5_dev_ctx_shared object.
7726 * Pointer to Current counter pool.
7729 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7730 struct mlx5_flow_counter_pool *pool)
7732 struct mlx5_priv *priv;
7733 struct mlx5_flow_counter *cnt;
7734 struct mlx5_age_info *age_info;
7735 struct mlx5_age_param *age_param;
7736 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7737 struct mlx5_counter_stats_raw *prev = pool->raw;
7738 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7739 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7740 uint16_t expected = AGE_CANDIDATE;
7743 pool->time_of_last_age_check = curr_time;
7744 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7745 cnt = MLX5_POOL_GET_CNT(pool, i);
7746 age_param = MLX5_CNT_TO_AGE(cnt);
7747 if (__atomic_load_n(&age_param->state,
7748 __ATOMIC_RELAXED) != AGE_CANDIDATE)
7750 if (cur->data[i].hits != prev->data[i].hits) {
7751 __atomic_store_n(&age_param->sec_since_last_hit, 0,
7755 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7757 __ATOMIC_RELAXED) <= age_param->timeout)
7760 * Hold the lock first, or if between the
7761 * state AGE_TMOUT and tailq operation the
7762 * release happened, the release procedure
7763 * may delete a non-existent tailq node.
7765 priv = rte_eth_devices[age_param->port_id].data->dev_private;
7766 age_info = GET_PORT_AGE_INFO(priv);
7767 rte_spinlock_lock(&age_info->aged_sl);
7768 if (__atomic_compare_exchange_n(&age_param->state, &expected,
7771 __ATOMIC_RELAXED)) {
7772 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7773 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7775 rte_spinlock_unlock(&age_info->aged_sl);
7777 mlx5_age_event_prepare(sh);
7781 * Handler for the HW respond about ready values from an asynchronous batch
7782 * query. This function is probably called by the host thread.
7785 * The pointer to the shared device context.
7786 * @param[in] async_id
7787 * The Devx async ID.
7789 * The status of the completion.
7792 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7793 uint64_t async_id, int status)
7795 struct mlx5_flow_counter_pool *pool =
7796 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7797 struct mlx5_counter_stats_raw *raw_to_free;
7798 uint8_t query_gen = pool->query_gen ^ 1;
7799 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7800 enum mlx5_counter_type cnt_type =
7801 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7802 MLX5_COUNTER_TYPE_ORIGIN;
7804 if (unlikely(status)) {
7805 raw_to_free = pool->raw_hw;
7807 raw_to_free = pool->raw;
7809 mlx5_flow_aging_check(sh, pool);
7810 rte_spinlock_lock(&pool->sl);
7811 pool->raw = pool->raw_hw;
7812 rte_spinlock_unlock(&pool->sl);
7813 /* Be sure the new raw counters data is updated in memory. */
7815 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7816 rte_spinlock_lock(&cmng->csl[cnt_type]);
7817 TAILQ_CONCAT(&cmng->counters[cnt_type],
7818 &pool->counters[query_gen], next);
7819 rte_spinlock_unlock(&cmng->csl[cnt_type]);
7822 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7823 pool->raw_hw = NULL;
7824 sh->cmng.pending_queries--;
7828 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7829 const struct flow_grp_info *grp_info,
7830 struct rte_flow_error *error)
7832 if (grp_info->transfer && grp_info->external &&
7833 grp_info->fdb_def_rule) {
7834 if (group == UINT32_MAX)
7835 return rte_flow_error_set
7837 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7839 "group index not supported");
7844 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7849 * Translate the rte_flow group index to HW table value.
7851 * If tunnel offload is disabled, all group ids converted to flow table
7852 * id using the standard method.
7853 * If tunnel offload is enabled, group id can be converted using the
7854 * standard or tunnel conversion method. Group conversion method
7855 * selection depends on flags in `grp_info` parameter:
7856 * - Internal (grp_info.external == 0) groups conversion uses the
7858 * - Group ids in JUMP action converted with the tunnel conversion.
7859 * - Group id in rule attribute conversion depends on a rule type and
7861 * ** non zero group attributes converted with the tunnel method
7862 * ** zero group attribute in non-tunnel rule is converted using the
7863 * standard method - there's only one root table
7864 * ** zero group attribute in steer tunnel rule is converted with the
7865 * standard method - single root table
7866 * ** zero group attribute in match tunnel rule is a special OvS
7867 * case: that value is used for portability reasons. That group
7868 * id is converted with the tunnel conversion method.
7873 * PMD tunnel offload object
7875 * rte_flow group index value.
7878 * @param[in] grp_info
7879 * flags used for conversion
7881 * Pointer to error structure.
7884 * 0 on success, a negative errno value otherwise and rte_errno is set.
7887 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7888 const struct mlx5_flow_tunnel *tunnel,
7889 uint32_t group, uint32_t *table,
7890 const struct flow_grp_info *grp_info,
7891 struct rte_flow_error *error)
7894 bool standard_translation;
7896 if (!grp_info->skip_scale && grp_info->external &&
7897 group < MLX5_MAX_TABLES_EXTERNAL)
7898 group *= MLX5_FLOW_TABLE_FACTOR;
7899 if (is_tunnel_offload_active(dev)) {
7900 standard_translation = !grp_info->external ||
7901 grp_info->std_tbl_fix;
7903 standard_translation = true;
7906 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7907 dev->data->port_id, group, grp_info->transfer,
7908 grp_info->external, grp_info->fdb_def_rule,
7909 standard_translation ? "STANDARD" : "TUNNEL");
7910 if (standard_translation)
7911 ret = flow_group_to_table(dev->data->port_id, group, table,
7914 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7921 * Discover availability of metadata reg_c's.
7923 * Iteratively use test flows to check availability.
7926 * Pointer to the Ethernet device structure.
7929 * 0 on success, a negative errno value otherwise and rte_errno is set.
7932 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7934 struct mlx5_priv *priv = dev->data->dev_private;
7935 struct mlx5_dev_config *config = &priv->config;
7936 enum modify_reg idx;
7939 /* reg_c[0] and reg_c[1] are reserved. */
7940 config->flow_mreg_c[n++] = REG_C_0;
7941 config->flow_mreg_c[n++] = REG_C_1;
7942 /* Discover availability of other reg_c's. */
7943 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7944 struct rte_flow_attr attr = {
7945 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7946 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7949 struct rte_flow_item items[] = {
7951 .type = RTE_FLOW_ITEM_TYPE_END,
7954 struct rte_flow_action actions[] = {
7956 .type = (enum rte_flow_action_type)
7957 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7958 .conf = &(struct mlx5_flow_action_copy_mreg){
7964 .type = RTE_FLOW_ACTION_TYPE_JUMP,
7965 .conf = &(struct rte_flow_action_jump){
7966 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7970 .type = RTE_FLOW_ACTION_TYPE_END,
7974 struct rte_flow *flow;
7975 struct rte_flow_error error;
7977 if (!config->dv_flow_en)
7979 /* Create internal flow, validation skips copy action. */
7980 flow_idx = flow_list_create(dev, NULL, &attr, items,
7981 actions, false, &error);
7982 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7986 config->flow_mreg_c[n++] = idx;
7987 flow_list_destroy(dev, NULL, flow_idx);
7989 for (; n < MLX5_MREG_C_NUM; ++n)
7990 config->flow_mreg_c[n] = REG_NON;
7995 save_dump_file(const uint8_t *data, uint32_t size,
7996 uint32_t type, uint32_t id, void *arg, FILE *file)
7998 char line[BUF_SIZE];
8001 uint32_t actions_num;
8002 struct rte_flow_query_count *count;
8004 memset(line, 0, BUF_SIZE);
8006 case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8007 actions_num = *(uint32_t *)(arg);
8008 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,%d,",
8009 type, id, actions_num);
8011 case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8012 out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,",
8015 case DR_DUMP_REC_TYPE_PMD_COUNTER:
8016 count = (struct rte_flow_query_count *)arg;
8017 fprintf(file, "%d,0x%x,%" PRIu64 ",%" PRIu64 "\n", type,
8018 id, count->hits, count->bytes);
8024 for (k = 0; k < size; k++) {
8025 /* Make sure we do not overrun the line buffer length. */
8026 if (out >= BUF_SIZE - 4) {
8030 out += snprintf(line + out, BUF_SIZE - out, "%02x",
8033 fprintf(file, "%s\n", line);
8038 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8039 struct rte_flow_query_count *count, struct rte_flow_error *error)
8041 struct rte_flow_action action[2];
8042 enum mlx5_flow_drv_type ftype;
8043 const struct mlx5_flow_driver_ops *fops;
8046 return rte_flow_error_set(error, ENOENT,
8047 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8049 "invalid flow handle");
8051 action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8052 action[1].type = RTE_FLOW_ACTION_TYPE_END;
8053 if (flow->counter) {
8054 memset(count, 0, sizeof(struct rte_flow_query_count));
8055 ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8056 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8057 ftype < MLX5_FLOW_TYPE_MAX);
8058 fops = flow_get_drv_ops(ftype);
8059 return fops->query(dev, flow, action, count, error);
8064 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8066 * Dump flow ipool data to file
8069 * The pointer to Ethernet device.
8071 * A pointer to a file for output.
8073 * Perform verbose error reporting if not NULL. PMDs initialize this
8074 * structure in case of error only.
8076 * 0 on success, a negative value otherwise.
8079 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8080 struct rte_flow *flow, FILE *file,
8081 struct rte_flow_error *error)
8083 struct mlx5_priv *priv = dev->data->dev_private;
8084 struct mlx5_flow_dv_modify_hdr_resource *modify_hdr;
8085 struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8086 uint32_t handle_idx;
8087 struct mlx5_flow_handle *dh;
8088 struct rte_flow_query_count count;
8089 uint32_t actions_num;
8090 const uint8_t *data;
8096 return rte_flow_error_set(error, ENOENT,
8097 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8099 "invalid flow handle");
8101 handle_idx = flow->dev_handles;
8102 while (handle_idx) {
8103 dh = mlx5_ipool_get(priv->sh->ipool
8104 [MLX5_IPOOL_MLX5_FLOW], handle_idx);
8107 handle_idx = dh->next.next;
8108 id = (uint32_t)(uintptr_t)dh->drv_flow;
8111 type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8112 if (!mlx5_flow_query_counter(dev, flow, &count, error))
8113 save_dump_file(NULL, 0, type,
8114 id, (void *)&count, file);
8116 /* Get modify_hdr and encap_decap buf from ipools. */
8118 modify_hdr = dh->dvh.modify_hdr;
8120 if (dh->dvh.rix_encap_decap) {
8121 encap_decap = mlx5_ipool_get(priv->sh->ipool
8122 [MLX5_IPOOL_DECAP_ENCAP],
8123 dh->dvh.rix_encap_decap);
8126 data = (const uint8_t *)modify_hdr->actions;
8127 size = (size_t)(modify_hdr->actions_num) * 8;
8128 actions_num = modify_hdr->actions_num;
8129 type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8130 save_dump_file(data, size, type, id,
8131 (void *)(&actions_num), file);
8134 data = encap_decap->buf;
8135 size = encap_decap->size;
8136 type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8137 save_dump_file(data, size, type,
8146 * Dump flow raw hw data to file
8149 * The pointer to Ethernet device.
8151 * A pointer to a file for output.
8153 * Perform verbose error reporting if not NULL. PMDs initialize this
8154 * structure in case of error only.
8156 * 0 on success, a nagative value otherwise.
8159 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8161 struct rte_flow_error *error __rte_unused)
8163 struct mlx5_priv *priv = dev->data->dev_private;
8164 struct mlx5_dev_ctx_shared *sh = priv->sh;
8165 uint32_t handle_idx;
8167 struct mlx5_flow_handle *dh;
8168 struct rte_flow *flow;
8169 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8173 if (!priv->config.dv_flow_en) {
8174 if (fputs("device dv flow disabled\n", file) <= 0)
8181 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8182 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
8183 priv->flows, idx, flow, next)
8184 mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8186 return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8188 sh->tx_domain, file);
8191 flow = mlx5_ipool_get(priv->sh->ipool
8192 [MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
8196 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8197 mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8199 handle_idx = flow->dev_handles;
8200 while (handle_idx) {
8201 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8206 ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8211 handle_idx = dh->next.next;
8217 * Get aged-out flows.
8220 * Pointer to the Ethernet device structure.
8221 * @param[in] context
8222 * The address of an array of pointers to the aged-out flows contexts.
8223 * @param[in] nb_countexts
8224 * The length of context array pointers.
8226 * Perform verbose error reporting if not NULL. Initialized in case of
8230 * how many contexts get in success, otherwise negative errno value.
8231 * if nb_contexts is 0, return the amount of all aged contexts.
8232 * if nb_contexts is not 0 , return the amount of aged flows reported
8233 * in the context array.
8236 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8237 uint32_t nb_contexts, struct rte_flow_error *error)
8239 const struct mlx5_flow_driver_ops *fops;
8240 struct rte_flow_attr attr = { .transfer = 0 };
8242 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8243 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8244 return fops->get_aged_flows(dev, contexts, nb_contexts,
8248 "port %u get aged flows is not supported.",
8249 dev->data->port_id);
8253 /* Wrapper for driver action_validate op callback */
8255 flow_drv_action_validate(struct rte_eth_dev *dev,
8256 const struct rte_flow_indir_action_conf *conf,
8257 const struct rte_flow_action *action,
8258 const struct mlx5_flow_driver_ops *fops,
8259 struct rte_flow_error *error)
8261 static const char err_msg[] = "indirect action validation unsupported";
8263 if (!fops->action_validate) {
8264 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8265 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8269 return fops->action_validate(dev, conf, action, error);
8273 * Destroys the shared action by handle.
8276 * Pointer to Ethernet device structure.
8278 * Handle for the indirect action object to be destroyed.
8280 * Perform verbose error reporting if not NULL. PMDs initialize this
8281 * structure in case of error only.
8284 * 0 on success, a negative errno value otherwise and rte_errno is set.
8286 * @note: wrapper for driver action_create op callback.
8289 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8290 struct rte_flow_action_handle *handle,
8291 struct rte_flow_error *error)
8293 static const char err_msg[] = "indirect action destruction unsupported";
8294 struct rte_flow_attr attr = { .transfer = 0 };
8295 const struct mlx5_flow_driver_ops *fops =
8296 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8298 if (!fops->action_destroy) {
8299 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8300 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8304 return fops->action_destroy(dev, handle, error);
8307 /* Wrapper for driver action_destroy op callback */
8309 flow_drv_action_update(struct rte_eth_dev *dev,
8310 struct rte_flow_action_handle *handle,
8312 const struct mlx5_flow_driver_ops *fops,
8313 struct rte_flow_error *error)
8315 static const char err_msg[] = "indirect action update unsupported";
8317 if (!fops->action_update) {
8318 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8319 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8323 return fops->action_update(dev, handle, update, error);
8326 /* Wrapper for driver action_destroy op callback */
8328 flow_drv_action_query(struct rte_eth_dev *dev,
8329 const struct rte_flow_action_handle *handle,
8331 const struct mlx5_flow_driver_ops *fops,
8332 struct rte_flow_error *error)
8334 static const char err_msg[] = "indirect action query unsupported";
8336 if (!fops->action_query) {
8337 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8338 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8342 return fops->action_query(dev, handle, data, error);
8346 * Create indirect action for reuse in multiple flow rules.
8349 * Pointer to Ethernet device structure.
8351 * Pointer to indirect action object configuration.
8353 * Action configuration for indirect action object creation.
8355 * Perform verbose error reporting if not NULL. PMDs initialize this
8356 * structure in case of error only.
8358 * A valid handle in case of success, NULL otherwise and rte_errno is set.
8360 static struct rte_flow_action_handle *
8361 mlx5_action_handle_create(struct rte_eth_dev *dev,
8362 const struct rte_flow_indir_action_conf *conf,
8363 const struct rte_flow_action *action,
8364 struct rte_flow_error *error)
8366 static const char err_msg[] = "indirect action creation unsupported";
8367 struct rte_flow_attr attr = { .transfer = 0 };
8368 const struct mlx5_flow_driver_ops *fops =
8369 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8371 if (flow_drv_action_validate(dev, conf, action, fops, error))
8373 if (!fops->action_create) {
8374 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8375 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8379 return fops->action_create(dev, conf, action, error);
8383 * Updates inplace the indirect action configuration pointed by *handle*
8384 * with the configuration provided as *update* argument.
8385 * The update of the indirect action configuration effects all flow rules
8386 * reusing the action via handle.
8389 * Pointer to Ethernet device structure.
8391 * Handle for the indirect action to be updated.
8393 * Action specification used to modify the action pointed by handle.
8394 * *update* could be of same type with the action pointed by the *handle*
8395 * handle argument, or some other structures like a wrapper, depending on
8396 * the indirect action type.
8398 * Perform verbose error reporting if not NULL. PMDs initialize this
8399 * structure in case of error only.
8402 * 0 on success, a negative errno value otherwise and rte_errno is set.
8405 mlx5_action_handle_update(struct rte_eth_dev *dev,
8406 struct rte_flow_action_handle *handle,
8408 struct rte_flow_error *error)
8410 struct rte_flow_attr attr = { .transfer = 0 };
8411 const struct mlx5_flow_driver_ops *fops =
8412 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8415 ret = flow_drv_action_validate(dev, NULL,
8416 (const struct rte_flow_action *)update, fops, error);
8419 return flow_drv_action_update(dev, handle, update, fops,
8424 * Query the indirect action by handle.
8426 * This function allows retrieving action-specific data such as counters.
8427 * Data is gathered by special action which may be present/referenced in
8428 * more than one flow rule definition.
8430 * see @RTE_FLOW_ACTION_TYPE_COUNT
8433 * Pointer to Ethernet device structure.
8435 * Handle for the indirect action to query.
8436 * @param[in, out] data
8437 * Pointer to storage for the associated query data type.
8439 * Perform verbose error reporting if not NULL. PMDs initialize this
8440 * structure in case of error only.
8443 * 0 on success, a negative errno value otherwise and rte_errno is set.
8446 mlx5_action_handle_query(struct rte_eth_dev *dev,
8447 const struct rte_flow_action_handle *handle,
8449 struct rte_flow_error *error)
8451 struct rte_flow_attr attr = { .transfer = 0 };
8452 const struct mlx5_flow_driver_ops *fops =
8453 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8455 return flow_drv_action_query(dev, handle, data, fops, error);
8459 * Destroy all indirect actions (shared RSS).
8462 * Pointer to Ethernet device.
8465 * 0 on success, a negative errno value otherwise and rte_errno is set.
8468 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8470 struct rte_flow_error error;
8471 struct mlx5_priv *priv = dev->data->dev_private;
8472 struct mlx5_shared_action_rss *shared_rss;
8476 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8477 priv->rss_shared_actions, idx, shared_rss, next) {
8478 ret |= mlx5_action_handle_destroy(dev,
8479 (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8484 #ifndef HAVE_MLX5DV_DR
8485 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8487 #define MLX5_DOMAIN_SYNC_FLOW \
8488 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8491 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8493 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8494 const struct mlx5_flow_driver_ops *fops;
8496 struct rte_flow_attr attr = { .transfer = 0 };
8498 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8499 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8505 const struct mlx5_flow_tunnel *
8506 mlx5_get_tof(const struct rte_flow_item *item,
8507 const struct rte_flow_action *action,
8508 enum mlx5_tof_rule_type *rule_type)
8510 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8511 if (item->type == (typeof(item->type))
8512 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8513 *rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8514 return flow_items_to_tunnel(item);
8517 for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8518 if (action->type == (typeof(action->type))
8519 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8520 *rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8521 return flow_actions_to_tunnel(action);
8528 * tunnel offload functionalilty is defined for DV environment only
8530 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8532 union tunnel_offload_mark {
8535 uint32_t app_reserve:8;
8536 uint32_t table_id:15;
8537 uint32_t transfer:1;
8538 uint32_t _unused_:8;
8543 mlx5_access_tunnel_offload_db
8544 (struct rte_eth_dev *dev,
8545 bool (*match)(struct rte_eth_dev *,
8546 struct mlx5_flow_tunnel *, const void *),
8547 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8548 void (*miss)(struct rte_eth_dev *, void *),
8549 void *ctx, bool lock_op);
8552 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8553 struct rte_flow *flow,
8554 const struct rte_flow_attr *attr,
8555 const struct rte_flow_action *app_actions,
8557 const struct mlx5_flow_tunnel *tunnel,
8558 struct tunnel_default_miss_ctx *ctx,
8559 struct rte_flow_error *error)
8561 struct mlx5_priv *priv = dev->data->dev_private;
8562 struct mlx5_flow *dev_flow;
8563 struct rte_flow_attr miss_attr = *attr;
8564 const struct rte_flow_item miss_items[2] = {
8566 .type = RTE_FLOW_ITEM_TYPE_ETH,
8572 .type = RTE_FLOW_ITEM_TYPE_END,
8578 union tunnel_offload_mark mark_id;
8579 struct rte_flow_action_mark miss_mark;
8580 struct rte_flow_action miss_actions[3] = {
8581 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8582 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
8584 const struct rte_flow_action_jump *jump_data;
8585 uint32_t i, flow_table = 0; /* prevent compilation warning */
8586 struct flow_grp_info grp_info = {
8588 .transfer = attr->transfer,
8589 .fdb_def_rule = !!priv->fdb_def_rule,
8594 if (!attr->transfer) {
8597 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8598 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8599 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8602 return rte_flow_error_set
8604 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8605 NULL, "invalid default miss RSS");
8606 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8607 ctx->action_rss.level = 0,
8608 ctx->action_rss.types = priv->rss_conf.rss_hf,
8609 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8610 ctx->action_rss.queue_num = priv->reta_idx_n,
8611 ctx->action_rss.key = priv->rss_conf.rss_key,
8612 ctx->action_rss.queue = ctx->queue;
8613 if (!priv->reta_idx_n || !priv->rxqs_n)
8614 return rte_flow_error_set
8616 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8617 NULL, "invalid port configuration");
8618 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8619 ctx->action_rss.types = 0;
8620 for (i = 0; i != priv->reta_idx_n; ++i)
8621 ctx->queue[i] = (*priv->reta_idx)[i];
8623 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8624 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8626 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8627 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8628 jump_data = app_actions->conf;
8629 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8630 miss_attr.group = jump_data->group;
8631 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8632 &flow_table, &grp_info, error);
8634 return rte_flow_error_set(error, EINVAL,
8635 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8636 NULL, "invalid tunnel id");
8637 mark_id.app_reserve = 0;
8638 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8639 mark_id.transfer = !!attr->transfer;
8640 mark_id._unused_ = 0;
8641 miss_mark.id = mark_id.val;
8642 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8643 miss_items, miss_actions, flow_idx, error);
8646 dev_flow->flow = flow;
8647 dev_flow->external = true;
8648 dev_flow->tunnel = tunnel;
8649 dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8650 /* Subflow object was created, we must include one in the list. */
8651 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8652 dev_flow->handle, next);
8654 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8655 dev->data->port_id, tunnel->app_tunnel.type,
8656 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8657 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8658 miss_actions, error);
8660 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8666 static const struct mlx5_flow_tbl_data_entry *
8667 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8669 struct mlx5_priv *priv = dev->data->dev_private;
8670 struct mlx5_dev_ctx_shared *sh = priv->sh;
8671 struct mlx5_hlist_entry *he;
8672 union tunnel_offload_mark mbits = { .val = mark };
8673 union mlx5_flow_tbl_key table_key = {
8675 .level = tunnel_id_to_flow_tbl(mbits.table_id),
8679 .is_fdb = !!mbits.transfer,
8683 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
8685 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8689 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
8690 struct mlx5_hlist_entry *entry)
8692 struct mlx5_dev_ctx_shared *sh = list->ctx;
8693 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8695 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8696 tunnel_flow_tbl_to_id(tte->flow_table));
8701 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
8702 struct mlx5_hlist_entry *entry,
8703 uint64_t key, void *cb_ctx __rte_unused)
8705 union tunnel_tbl_key tbl = {
8708 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8710 return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8713 static struct mlx5_hlist_entry *
8714 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
8715 void *ctx __rte_unused)
8717 struct mlx5_dev_ctx_shared *sh = list->ctx;
8718 struct tunnel_tbl_entry *tte;
8719 union tunnel_tbl_key tbl = {
8723 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8728 mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8730 if (tte->flow_table >= MLX5_MAX_TABLES) {
8731 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8733 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8736 } else if (!tte->flow_table) {
8739 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8740 tte->tunnel_id = tbl.tunnel_id;
8741 tte->group = tbl.group;
8750 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8751 const struct mlx5_flow_tunnel *tunnel,
8752 uint32_t group, uint32_t *table,
8753 struct rte_flow_error *error)
8755 struct mlx5_hlist_entry *he;
8756 struct tunnel_tbl_entry *tte;
8757 union tunnel_tbl_key key = {
8758 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8761 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8762 struct mlx5_hlist *group_hash;
8764 group_hash = tunnel ? tunnel->groups : thub->groups;
8765 he = mlx5_hlist_register(group_hash, key.val, NULL);
8767 return rte_flow_error_set(error, EINVAL,
8768 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8770 "tunnel group index not supported");
8771 tte = container_of(he, typeof(*tte), hash);
8772 *table = tte->flow_table;
8773 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8774 dev->data->port_id, key.tunnel_id, group, *table);
8779 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8780 struct mlx5_flow_tunnel *tunnel)
8782 struct mlx5_priv *priv = dev->data->dev_private;
8783 struct mlx5_indexed_pool *ipool;
8785 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8786 dev->data->port_id, tunnel->tunnel_id);
8787 LIST_REMOVE(tunnel, chain);
8788 mlx5_hlist_destroy(tunnel->groups);
8789 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8790 mlx5_ipool_free(ipool, tunnel->tunnel_id);
8794 mlx5_access_tunnel_offload_db
8795 (struct rte_eth_dev *dev,
8796 bool (*match)(struct rte_eth_dev *,
8797 struct mlx5_flow_tunnel *, const void *),
8798 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8799 void (*miss)(struct rte_eth_dev *, void *),
8800 void *ctx, bool lock_op)
8802 bool verdict = false;
8803 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8804 struct mlx5_flow_tunnel *tunnel;
8806 rte_spinlock_lock(&thub->sl);
8807 LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8808 verdict = match(dev, tunnel, (const void *)ctx);
8813 rte_spinlock_unlock(&thub->sl);
8815 hit(dev, tunnel, ctx);
8816 if (!verdict && miss)
8819 rte_spinlock_unlock(&thub->sl);
8824 struct tunnel_db_find_tunnel_id_ctx {
8826 struct mlx5_flow_tunnel *tunnel;
8830 find_tunnel_id_match(struct rte_eth_dev *dev,
8831 struct mlx5_flow_tunnel *tunnel, const void *x)
8833 const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8836 return tunnel->tunnel_id == ctx->tunnel_id;
8840 find_tunnel_id_hit(struct rte_eth_dev *dev,
8841 struct mlx5_flow_tunnel *tunnel, void *x)
8843 struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8845 ctx->tunnel = tunnel;
8848 static struct mlx5_flow_tunnel *
8849 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8851 struct tunnel_db_find_tunnel_id_ctx ctx = {
8855 mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8856 find_tunnel_id_hit, NULL, &ctx, true);
8861 static struct mlx5_flow_tunnel *
8862 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8863 const struct rte_flow_tunnel *app_tunnel)
8865 struct mlx5_priv *priv = dev->data->dev_private;
8866 struct mlx5_indexed_pool *ipool;
8867 struct mlx5_flow_tunnel *tunnel;
8870 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8871 tunnel = mlx5_ipool_zmalloc(ipool, &id);
8874 if (id >= MLX5_MAX_TUNNELS) {
8875 mlx5_ipool_free(ipool, id);
8876 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8879 tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
8880 mlx5_flow_tunnel_grp2tbl_create_cb,
8881 mlx5_flow_tunnel_grp2tbl_match_cb,
8882 mlx5_flow_tunnel_grp2tbl_remove_cb);
8883 if (!tunnel->groups) {
8884 mlx5_ipool_free(ipool, id);
8887 tunnel->groups->ctx = priv->sh;
8888 /* initiate new PMD tunnel */
8889 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8890 tunnel->tunnel_id = id;
8891 tunnel->action.type = (typeof(tunnel->action.type))
8892 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8893 tunnel->action.conf = tunnel;
8894 tunnel->item.type = (typeof(tunnel->item.type))
8895 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8896 tunnel->item.spec = tunnel;
8897 tunnel->item.last = NULL;
8898 tunnel->item.mask = NULL;
8900 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8901 dev->data->port_id, tunnel->tunnel_id);
8906 struct tunnel_db_get_tunnel_ctx {
8907 const struct rte_flow_tunnel *app_tunnel;
8908 struct mlx5_flow_tunnel *tunnel;
8911 static bool get_tunnel_match(struct rte_eth_dev *dev,
8912 struct mlx5_flow_tunnel *tunnel, const void *x)
8914 const struct tunnel_db_get_tunnel_ctx *ctx = x;
8917 return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
8918 sizeof(*ctx->app_tunnel));
8921 static void get_tunnel_hit(struct rte_eth_dev *dev,
8922 struct mlx5_flow_tunnel *tunnel, void *x)
8924 /* called under tunnel spinlock protection */
8925 struct tunnel_db_get_tunnel_ctx *ctx = x;
8929 ctx->tunnel = tunnel;
8932 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8934 /* called under tunnel spinlock protection */
8935 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8936 struct tunnel_db_get_tunnel_ctx *ctx = x;
8938 rte_spinlock_unlock(&thub->sl);
8939 ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8940 rte_spinlock_lock(&thub->sl);
8942 ctx->tunnel->refctn = 1;
8943 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8949 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8950 const struct rte_flow_tunnel *app_tunnel,
8951 struct mlx5_flow_tunnel **tunnel)
8953 struct tunnel_db_get_tunnel_ctx ctx = {
8954 .app_tunnel = app_tunnel,
8957 mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8958 get_tunnel_miss, &ctx, true);
8959 *tunnel = ctx.tunnel;
8960 return ctx.tunnel ? 0 : -ENOMEM;
8963 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8965 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8969 if (!LIST_EMPTY(&thub->tunnels))
8970 DRV_LOG(WARNING, "port %u tunnels present", port_id);
8971 mlx5_hlist_destroy(thub->groups);
8975 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8978 struct mlx5_flow_tunnel_hub *thub;
8980 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8984 LIST_INIT(&thub->tunnels);
8985 rte_spinlock_init(&thub->sl);
8986 thub->groups = mlx5_hlist_create("flow groups",
8987 rte_align32pow2(MLX5_MAX_TABLES), 0,
8988 0, mlx5_flow_tunnel_grp2tbl_create_cb,
8989 mlx5_flow_tunnel_grp2tbl_match_cb,
8990 mlx5_flow_tunnel_grp2tbl_remove_cb);
8991 if (!thub->groups) {
8995 thub->groups->ctx = sh;
8996 sh->tunnel_hub = thub;
9002 mlx5_hlist_destroy(thub->groups);
9009 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9010 struct rte_flow_tunnel *tunnel,
9011 const char *err_msg)
9014 if (!is_tunnel_offload_active(dev)) {
9015 err_msg = "tunnel offload was not activated";
9017 } else if (!tunnel) {
9018 err_msg = "no application tunnel";
9022 switch (tunnel->type) {
9024 err_msg = "unsupported tunnel type";
9026 case RTE_FLOW_ITEM_TYPE_VXLAN:
9035 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9036 struct rte_flow_tunnel *app_tunnel,
9037 struct rte_flow_action **actions,
9038 uint32_t *num_of_actions,
9039 struct rte_flow_error *error)
9042 struct mlx5_flow_tunnel *tunnel;
9043 const char *err_msg = NULL;
9044 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9047 return rte_flow_error_set(error, EINVAL,
9048 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9050 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9052 return rte_flow_error_set(error, ret,
9053 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9054 "failed to initialize pmd tunnel");
9056 *actions = &tunnel->action;
9057 *num_of_actions = 1;
9062 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9063 struct rte_flow_tunnel *app_tunnel,
9064 struct rte_flow_item **items,
9065 uint32_t *num_of_items,
9066 struct rte_flow_error *error)
9069 struct mlx5_flow_tunnel *tunnel;
9070 const char *err_msg = NULL;
9071 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9074 return rte_flow_error_set(error, EINVAL,
9075 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9077 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9079 return rte_flow_error_set(error, ret,
9080 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9081 "failed to initialize pmd tunnel");
9083 *items = &tunnel->item;
9088 struct tunnel_db_element_release_ctx {
9089 struct rte_flow_item *items;
9090 struct rte_flow_action *actions;
9091 uint32_t num_elements;
9092 struct rte_flow_error *error;
9097 tunnel_element_release_match(struct rte_eth_dev *dev,
9098 struct mlx5_flow_tunnel *tunnel, const void *x)
9100 const struct tunnel_db_element_release_ctx *ctx = x;
9103 if (ctx->num_elements != 1)
9105 else if (ctx->items)
9106 return ctx->items == &tunnel->item;
9107 else if (ctx->actions)
9108 return ctx->actions == &tunnel->action;
9114 tunnel_element_release_hit(struct rte_eth_dev *dev,
9115 struct mlx5_flow_tunnel *tunnel, void *x)
9117 struct tunnel_db_element_release_ctx *ctx = x;
9119 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9120 mlx5_flow_tunnel_free(dev, tunnel);
9124 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9126 struct tunnel_db_element_release_ctx *ctx = x;
9128 ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9129 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9130 "invalid argument");
9134 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9135 struct rte_flow_item *pmd_items,
9136 uint32_t num_items, struct rte_flow_error *err)
9138 struct tunnel_db_element_release_ctx ctx = {
9141 .num_elements = num_items,
9145 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9146 tunnel_element_release_hit,
9147 tunnel_element_release_miss, &ctx, false);
9153 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9154 struct rte_flow_action *pmd_actions,
9155 uint32_t num_actions, struct rte_flow_error *err)
9157 struct tunnel_db_element_release_ctx ctx = {
9159 .actions = pmd_actions,
9160 .num_elements = num_actions,
9164 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9165 tunnel_element_release_hit,
9166 tunnel_element_release_miss, &ctx, false);
9172 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9174 struct rte_flow_restore_info *info,
9175 struct rte_flow_error *err)
9177 uint64_t ol_flags = m->ol_flags;
9178 const struct mlx5_flow_tbl_data_entry *tble;
9179 const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
9181 if (!is_tunnel_offload_active(dev)) {
9186 if ((ol_flags & mask) != mask)
9188 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9190 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9191 dev->data->port_id, m->hash.fdir.hi);
9194 MLX5_ASSERT(tble->tunnel);
9195 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9196 info->group_id = tble->group_id;
9197 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9198 RTE_FLOW_RESTORE_INFO_GROUP_ID |
9199 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9204 return rte_flow_error_set(err, EINVAL,
9205 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9206 "failed to get restore info");
9209 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9211 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9212 __rte_unused struct rte_flow_tunnel *app_tunnel,
9213 __rte_unused struct rte_flow_action **actions,
9214 __rte_unused uint32_t *num_of_actions,
9215 __rte_unused struct rte_flow_error *error)
9221 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9222 __rte_unused struct rte_flow_tunnel *app_tunnel,
9223 __rte_unused struct rte_flow_item **items,
9224 __rte_unused uint32_t *num_of_items,
9225 __rte_unused struct rte_flow_error *error)
9231 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9232 __rte_unused struct rte_flow_item *pmd_items,
9233 __rte_unused uint32_t num_items,
9234 __rte_unused struct rte_flow_error *err)
9240 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9241 __rte_unused struct rte_flow_action *pmd_action,
9242 __rte_unused uint32_t num_actions,
9243 __rte_unused struct rte_flow_error *err)
9249 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9250 __rte_unused struct rte_mbuf *m,
9251 __rte_unused struct rte_flow_restore_info *i,
9252 __rte_unused struct rte_flow_error *err)
9258 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9259 __rte_unused struct rte_flow *flow,
9260 __rte_unused const struct rte_flow_attr *attr,
9261 __rte_unused const struct rte_flow_action *actions,
9262 __rte_unused uint32_t flow_idx,
9263 __rte_unused const struct mlx5_flow_tunnel *tunnel,
9264 __rte_unused struct tunnel_default_miss_ctx *ctx,
9265 __rte_unused struct rte_flow_error *error)
9270 static struct mlx5_flow_tunnel *
9271 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9272 __rte_unused uint32_t id)
9278 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9279 __rte_unused struct mlx5_flow_tunnel *tunnel)
9284 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9285 __rte_unused const struct mlx5_flow_tunnel *t,
9286 __rte_unused uint32_t group,
9287 __rte_unused uint32_t *table,
9288 struct rte_flow_error *error)
9290 return rte_flow_error_set(error, ENOTSUP,
9291 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9292 "tunnel offload requires DV support");
9296 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9297 __rte_unused uint16_t port_id)
9300 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9303 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9306 struct rte_flow_error error;
9308 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9310 ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9312 (void *)(uintptr_t)item->type, &error);
9314 printf("%s ", item_name);
9316 printf("%d\n", (int)item->type);