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 struct tunnel_default_miss_ctx *ctx,
54 struct rte_flow_error *error);
55 static struct mlx5_flow_tunnel *
56 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
58 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
60 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
61 const struct mlx5_flow_tunnel *tunnel,
62 uint32_t group, uint32_t *table,
63 struct rte_flow_error *error);
65 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
66 static void mlx5_flow_pop_thread_workspace(void);
69 /** Device flow drivers. */
70 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
72 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
74 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
75 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
76 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
77 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
79 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
80 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
83 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
84 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
89 /** Node object of input graph for mlx5_flow_expand_rss(). */
90 struct mlx5_flow_expand_node {
91 const int *const next;
93 * List of next node indexes. Index 0 is interpreted as a terminator.
95 const enum rte_flow_item_type type;
96 /**< Pattern item type of current node. */
99 * RSS types bit-field associated with this node
100 * (see ETH_RSS_* definitions).
104 /** Object returned by mlx5_flow_expand_rss(). */
105 struct mlx5_flow_expand_rss {
107 /**< Number of entries @p patterns and @p priorities. */
109 struct rte_flow_item *pattern; /**< Expanded pattern array. */
110 uint32_t priority; /**< Priority offset for each expansion. */
114 static enum rte_flow_item_type
115 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
117 enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
118 uint16_t ether_type = 0;
119 uint16_t ether_type_m;
120 uint8_t ip_next_proto = 0;
121 uint8_t ip_next_proto_m;
123 if (item == NULL || item->spec == NULL)
125 switch (item->type) {
126 case RTE_FLOW_ITEM_TYPE_ETH:
128 ether_type_m = ((const struct rte_flow_item_eth *)
131 ether_type_m = rte_flow_item_eth_mask.type;
132 if (ether_type_m != RTE_BE16(0xFFFF))
134 ether_type = ((const struct rte_flow_item_eth *)
136 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
137 ret = RTE_FLOW_ITEM_TYPE_IPV4;
138 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
139 ret = RTE_FLOW_ITEM_TYPE_IPV6;
140 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
141 ret = RTE_FLOW_ITEM_TYPE_VLAN;
143 ret = RTE_FLOW_ITEM_TYPE_END;
145 case RTE_FLOW_ITEM_TYPE_VLAN:
147 ether_type_m = ((const struct rte_flow_item_vlan *)
148 (item->mask))->inner_type;
150 ether_type_m = rte_flow_item_vlan_mask.inner_type;
151 if (ether_type_m != RTE_BE16(0xFFFF))
153 ether_type = ((const struct rte_flow_item_vlan *)
154 (item->spec))->inner_type;
155 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
156 ret = RTE_FLOW_ITEM_TYPE_IPV4;
157 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
158 ret = RTE_FLOW_ITEM_TYPE_IPV6;
159 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
160 ret = RTE_FLOW_ITEM_TYPE_VLAN;
162 ret = RTE_FLOW_ITEM_TYPE_END;
164 case RTE_FLOW_ITEM_TYPE_IPV4:
166 ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
167 (item->mask))->hdr.next_proto_id;
170 rte_flow_item_ipv4_mask.hdr.next_proto_id;
171 if (ip_next_proto_m != 0xFF)
173 ip_next_proto = ((const struct rte_flow_item_ipv4 *)
174 (item->spec))->hdr.next_proto_id;
175 if (ip_next_proto == IPPROTO_UDP)
176 ret = RTE_FLOW_ITEM_TYPE_UDP;
177 else if (ip_next_proto == IPPROTO_TCP)
178 ret = RTE_FLOW_ITEM_TYPE_TCP;
179 else if (ip_next_proto == IPPROTO_IP)
180 ret = RTE_FLOW_ITEM_TYPE_IPV4;
181 else if (ip_next_proto == IPPROTO_IPV6)
182 ret = RTE_FLOW_ITEM_TYPE_IPV6;
184 ret = RTE_FLOW_ITEM_TYPE_END;
186 case RTE_FLOW_ITEM_TYPE_IPV6:
188 ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
189 (item->mask))->hdr.proto;
192 rte_flow_item_ipv6_mask.hdr.proto;
193 if (ip_next_proto_m != 0xFF)
195 ip_next_proto = ((const struct rte_flow_item_ipv6 *)
196 (item->spec))->hdr.proto;
197 if (ip_next_proto == IPPROTO_UDP)
198 ret = RTE_FLOW_ITEM_TYPE_UDP;
199 else if (ip_next_proto == IPPROTO_TCP)
200 ret = RTE_FLOW_ITEM_TYPE_TCP;
201 else if (ip_next_proto == IPPROTO_IP)
202 ret = RTE_FLOW_ITEM_TYPE_IPV4;
203 else if (ip_next_proto == IPPROTO_IPV6)
204 ret = RTE_FLOW_ITEM_TYPE_IPV6;
206 ret = RTE_FLOW_ITEM_TYPE_END;
209 ret = RTE_FLOW_ITEM_TYPE_VOID;
215 #define MLX5_RSS_EXP_ELT_N 8
218 * Expand RSS flows into several possible flows according to the RSS hash
219 * fields requested and the driver capabilities.
222 * Buffer to store the result expansion.
224 * Buffer size in bytes. If 0, @p buf can be NULL.
228 * RSS types to expand (see ETH_RSS_* definitions).
230 * Input graph to expand @p pattern according to @p types.
231 * @param[in] graph_root_index
232 * Index of root node in @p graph, typically 0.
235 * A positive value representing the size of @p buf in bytes regardless of
236 * @p size on success, a negative errno value otherwise and rte_errno is
237 * set, the following errors are defined:
239 * -E2BIG: graph-depth @p graph is too deep.
242 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
243 const struct rte_flow_item *pattern, uint64_t types,
244 const struct mlx5_flow_expand_node graph[],
245 int graph_root_index)
247 const struct rte_flow_item *item;
248 const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
249 const int *next_node;
250 const int *stack[MLX5_RSS_EXP_ELT_N];
252 struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
255 size_t user_pattern_size = 0;
257 const struct mlx5_flow_expand_node *next = NULL;
258 struct rte_flow_item missed_item;
261 const struct rte_flow_item *last_item = NULL;
263 memset(&missed_item, 0, sizeof(missed_item));
264 lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
265 MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
267 buf->entry[0].priority = 0;
268 buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
270 addr = buf->entry[0].pattern;
272 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
273 if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
275 for (i = 0; node->next && node->next[i]; ++i) {
276 next = &graph[node->next[i]];
277 if (next->type == item->type)
282 user_pattern_size += sizeof(*item);
284 user_pattern_size += sizeof(*item); /* Handle END item. */
285 lsize += user_pattern_size;
286 /* Copy the user pattern in the first entry of the buffer. */
288 rte_memcpy(addr, pattern, user_pattern_size);
289 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
292 /* Start expanding. */
293 memset(flow_items, 0, sizeof(flow_items));
294 user_pattern_size -= sizeof(*item);
296 * Check if the last valid item has spec set, need complete pattern,
297 * and the pattern can be used for expansion.
299 missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
300 if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
301 /* Item type END indicates expansion is not required. */
304 if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
307 for (i = 0; node->next && node->next[i]; ++i) {
308 next = &graph[node->next[i]];
309 if (next->type == missed_item.type) {
310 flow_items[0].type = missed_item.type;
311 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
317 if (next && missed) {
318 elt = 2; /* missed item + item end. */
320 lsize += elt * sizeof(*item) + user_pattern_size;
321 if ((node->rss_types & types) && lsize <= size) {
322 buf->entry[buf->entries].priority = 1;
323 buf->entry[buf->entries].pattern = addr;
325 rte_memcpy(addr, buf->entry[0].pattern,
327 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
328 rte_memcpy(addr, flow_items, elt * sizeof(*item));
329 addr = (void *)(((uintptr_t)addr) +
330 elt * sizeof(*item));
333 memset(flow_items, 0, sizeof(flow_items));
334 next_node = node->next;
335 stack[stack_pos] = next_node;
336 node = next_node ? &graph[*next_node] : NULL;
338 flow_items[stack_pos].type = node->type;
339 if (node->rss_types & types) {
341 * compute the number of items to copy from the
342 * expansion and copy it.
343 * When the stack_pos is 0, there are 1 element in it,
344 * plus the addition END item.
347 flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
348 lsize += elt * sizeof(*item) + user_pattern_size;
350 size_t n = elt * sizeof(*item);
352 buf->entry[buf->entries].priority =
353 stack_pos + 1 + missed;
354 buf->entry[buf->entries].pattern = addr;
356 rte_memcpy(addr, buf->entry[0].pattern,
358 addr = (void *)(((uintptr_t)addr) +
360 rte_memcpy(addr, &missed_item,
361 missed * sizeof(*item));
362 addr = (void *)(((uintptr_t)addr) +
363 missed * sizeof(*item));
364 rte_memcpy(addr, flow_items, n);
365 addr = (void *)(((uintptr_t)addr) + n);
370 next_node = node->next;
371 if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
375 stack[stack_pos] = next_node;
376 } else if (*(next_node + 1)) {
377 /* Follow up with the next possibility. */
380 /* Move to the next path. */
382 next_node = stack[--stack_pos];
384 stack[stack_pos] = next_node;
386 node = *next_node ? &graph[*next_node] : NULL;
391 enum mlx5_expansion {
393 MLX5_EXPANSION_ROOT_OUTER,
394 MLX5_EXPANSION_ROOT_ETH_VLAN,
395 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
396 MLX5_EXPANSION_OUTER_ETH,
397 MLX5_EXPANSION_OUTER_ETH_VLAN,
398 MLX5_EXPANSION_OUTER_VLAN,
399 MLX5_EXPANSION_OUTER_IPV4,
400 MLX5_EXPANSION_OUTER_IPV4_UDP,
401 MLX5_EXPANSION_OUTER_IPV4_TCP,
402 MLX5_EXPANSION_OUTER_IPV6,
403 MLX5_EXPANSION_OUTER_IPV6_UDP,
404 MLX5_EXPANSION_OUTER_IPV6_TCP,
405 MLX5_EXPANSION_VXLAN,
406 MLX5_EXPANSION_VXLAN_GPE,
410 MLX5_EXPANSION_ETH_VLAN,
413 MLX5_EXPANSION_IPV4_UDP,
414 MLX5_EXPANSION_IPV4_TCP,
416 MLX5_EXPANSION_IPV6_UDP,
417 MLX5_EXPANSION_IPV6_TCP,
420 /** Supported expansion of items. */
421 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
422 [MLX5_EXPANSION_ROOT] = {
423 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
425 MLX5_EXPANSION_IPV6),
426 .type = RTE_FLOW_ITEM_TYPE_END,
428 [MLX5_EXPANSION_ROOT_OUTER] = {
429 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
430 MLX5_EXPANSION_OUTER_IPV4,
431 MLX5_EXPANSION_OUTER_IPV6),
432 .type = RTE_FLOW_ITEM_TYPE_END,
434 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
435 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
436 .type = RTE_FLOW_ITEM_TYPE_END,
438 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
439 .next = MLX5_FLOW_EXPAND_RSS_NEXT
440 (MLX5_EXPANSION_OUTER_ETH_VLAN),
441 .type = RTE_FLOW_ITEM_TYPE_END,
443 [MLX5_EXPANSION_OUTER_ETH] = {
444 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
445 MLX5_EXPANSION_OUTER_IPV6,
446 MLX5_EXPANSION_MPLS),
447 .type = RTE_FLOW_ITEM_TYPE_ETH,
450 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
451 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
452 .type = RTE_FLOW_ITEM_TYPE_ETH,
455 [MLX5_EXPANSION_OUTER_VLAN] = {
456 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
457 MLX5_EXPANSION_OUTER_IPV6),
458 .type = RTE_FLOW_ITEM_TYPE_VLAN,
460 [MLX5_EXPANSION_OUTER_IPV4] = {
461 .next = MLX5_FLOW_EXPAND_RSS_NEXT
462 (MLX5_EXPANSION_OUTER_IPV4_UDP,
463 MLX5_EXPANSION_OUTER_IPV4_TCP,
466 MLX5_EXPANSION_IPV6),
467 .type = RTE_FLOW_ITEM_TYPE_IPV4,
468 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
469 ETH_RSS_NONFRAG_IPV4_OTHER,
471 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
472 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
473 MLX5_EXPANSION_VXLAN_GPE),
474 .type = RTE_FLOW_ITEM_TYPE_UDP,
475 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
477 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
478 .type = RTE_FLOW_ITEM_TYPE_TCP,
479 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
481 [MLX5_EXPANSION_OUTER_IPV6] = {
482 .next = MLX5_FLOW_EXPAND_RSS_NEXT
483 (MLX5_EXPANSION_OUTER_IPV6_UDP,
484 MLX5_EXPANSION_OUTER_IPV6_TCP,
488 .type = RTE_FLOW_ITEM_TYPE_IPV6,
489 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
490 ETH_RSS_NONFRAG_IPV6_OTHER,
492 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
493 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
494 MLX5_EXPANSION_VXLAN_GPE),
495 .type = RTE_FLOW_ITEM_TYPE_UDP,
496 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
498 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
499 .type = RTE_FLOW_ITEM_TYPE_TCP,
500 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
502 [MLX5_EXPANSION_VXLAN] = {
503 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
505 MLX5_EXPANSION_IPV6),
506 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
508 [MLX5_EXPANSION_VXLAN_GPE] = {
509 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
511 MLX5_EXPANSION_IPV6),
512 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
514 [MLX5_EXPANSION_GRE] = {
515 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
516 MLX5_EXPANSION_IPV6),
517 .type = RTE_FLOW_ITEM_TYPE_GRE,
519 [MLX5_EXPANSION_MPLS] = {
520 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
521 MLX5_EXPANSION_IPV6),
522 .type = RTE_FLOW_ITEM_TYPE_MPLS,
524 [MLX5_EXPANSION_ETH] = {
525 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
526 MLX5_EXPANSION_IPV6),
527 .type = RTE_FLOW_ITEM_TYPE_ETH,
529 [MLX5_EXPANSION_ETH_VLAN] = {
530 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
531 .type = RTE_FLOW_ITEM_TYPE_ETH,
533 [MLX5_EXPANSION_VLAN] = {
534 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
535 MLX5_EXPANSION_IPV6),
536 .type = RTE_FLOW_ITEM_TYPE_VLAN,
538 [MLX5_EXPANSION_IPV4] = {
539 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
540 MLX5_EXPANSION_IPV4_TCP),
541 .type = RTE_FLOW_ITEM_TYPE_IPV4,
542 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
543 ETH_RSS_NONFRAG_IPV4_OTHER,
545 [MLX5_EXPANSION_IPV4_UDP] = {
546 .type = RTE_FLOW_ITEM_TYPE_UDP,
547 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
549 [MLX5_EXPANSION_IPV4_TCP] = {
550 .type = RTE_FLOW_ITEM_TYPE_TCP,
551 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
553 [MLX5_EXPANSION_IPV6] = {
554 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
555 MLX5_EXPANSION_IPV6_TCP),
556 .type = RTE_FLOW_ITEM_TYPE_IPV6,
557 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
558 ETH_RSS_NONFRAG_IPV6_OTHER,
560 [MLX5_EXPANSION_IPV6_UDP] = {
561 .type = RTE_FLOW_ITEM_TYPE_UDP,
562 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
564 [MLX5_EXPANSION_IPV6_TCP] = {
565 .type = RTE_FLOW_ITEM_TYPE_TCP,
566 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
570 static struct rte_flow_action_handle *
571 mlx5_action_handle_create(struct rte_eth_dev *dev,
572 const struct rte_flow_indir_action_conf *conf,
573 const struct rte_flow_action *action,
574 struct rte_flow_error *error);
575 static int mlx5_action_handle_destroy
576 (struct rte_eth_dev *dev,
577 struct rte_flow_action_handle *handle,
578 struct rte_flow_error *error);
579 static int mlx5_action_handle_update
580 (struct rte_eth_dev *dev,
581 struct rte_flow_action_handle *handle,
583 struct rte_flow_error *error);
584 static int mlx5_action_handle_query
585 (struct rte_eth_dev *dev,
586 const struct rte_flow_action_handle *handle,
588 struct rte_flow_error *error);
590 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
591 struct rte_flow_tunnel *app_tunnel,
592 struct rte_flow_action **actions,
593 uint32_t *num_of_actions,
594 struct rte_flow_error *error);
596 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
597 struct rte_flow_tunnel *app_tunnel,
598 struct rte_flow_item **items,
599 uint32_t *num_of_items,
600 struct rte_flow_error *error);
602 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
603 struct rte_flow_item *pmd_items,
604 uint32_t num_items, struct rte_flow_error *err);
606 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
607 struct rte_flow_action *pmd_actions,
608 uint32_t num_actions,
609 struct rte_flow_error *err);
611 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
613 struct rte_flow_restore_info *info,
614 struct rte_flow_error *err);
616 static const struct rte_flow_ops mlx5_flow_ops = {
617 .validate = mlx5_flow_validate,
618 .create = mlx5_flow_create,
619 .destroy = mlx5_flow_destroy,
620 .flush = mlx5_flow_flush,
621 .isolate = mlx5_flow_isolate,
622 .query = mlx5_flow_query,
623 .dev_dump = mlx5_flow_dev_dump,
624 .get_aged_flows = mlx5_flow_get_aged_flows,
625 .action_handle_create = mlx5_action_handle_create,
626 .action_handle_destroy = mlx5_action_handle_destroy,
627 .action_handle_update = mlx5_action_handle_update,
628 .action_handle_query = mlx5_action_handle_query,
629 .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
630 .tunnel_match = mlx5_flow_tunnel_match,
631 .tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
632 .tunnel_item_release = mlx5_flow_tunnel_item_release,
633 .get_restore_info = mlx5_flow_tunnel_get_restore_info,
636 /* Tunnel information. */
637 struct mlx5_flow_tunnel_info {
638 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
639 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
642 static struct mlx5_flow_tunnel_info tunnels_info[] = {
644 .tunnel = MLX5_FLOW_LAYER_VXLAN,
645 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
648 .tunnel = MLX5_FLOW_LAYER_GENEVE,
649 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
652 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
653 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
656 .tunnel = MLX5_FLOW_LAYER_GRE,
657 .ptype = RTE_PTYPE_TUNNEL_GRE,
660 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
661 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
664 .tunnel = MLX5_FLOW_LAYER_MPLS,
665 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
668 .tunnel = MLX5_FLOW_LAYER_NVGRE,
669 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
672 .tunnel = MLX5_FLOW_LAYER_IPIP,
673 .ptype = RTE_PTYPE_TUNNEL_IP,
676 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
677 .ptype = RTE_PTYPE_TUNNEL_IP,
680 .tunnel = MLX5_FLOW_LAYER_GTP,
681 .ptype = RTE_PTYPE_TUNNEL_GTPU,
688 * Translate tag ID to register.
691 * Pointer to the Ethernet device structure.
693 * The feature that request the register.
695 * The request register ID.
697 * Error description in case of any.
700 * The request register on success, a negative errno
701 * value otherwise and rte_errno is set.
704 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
705 enum mlx5_feature_name feature,
707 struct rte_flow_error *error)
709 struct mlx5_priv *priv = dev->data->dev_private;
710 struct mlx5_dev_config *config = &priv->config;
711 enum modify_reg start_reg;
712 bool skip_mtr_reg = false;
715 case MLX5_HAIRPIN_RX:
717 case MLX5_HAIRPIN_TX:
719 case MLX5_METADATA_RX:
720 switch (config->dv_xmeta_en) {
721 case MLX5_XMETA_MODE_LEGACY:
723 case MLX5_XMETA_MODE_META16:
725 case MLX5_XMETA_MODE_META32:
729 case MLX5_METADATA_TX:
731 case MLX5_METADATA_FDB:
732 switch (config->dv_xmeta_en) {
733 case MLX5_XMETA_MODE_LEGACY:
735 case MLX5_XMETA_MODE_META16:
737 case MLX5_XMETA_MODE_META32:
742 switch (config->dv_xmeta_en) {
743 case MLX5_XMETA_MODE_LEGACY:
745 case MLX5_XMETA_MODE_META16:
747 case MLX5_XMETA_MODE_META32:
753 * If meter color and meter id share one register, flow match
754 * should use the meter color register for match.
756 if (priv->mtr_reg_share)
757 return priv->mtr_color_reg;
759 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
762 case MLX5_ASO_FLOW_HIT:
763 case MLX5_ASO_CONNTRACK:
764 /* All features use the same REG_C. */
765 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
766 return priv->mtr_color_reg;
769 * Metadata COPY_MARK register using is in meter suffix sub
770 * flow while with meter. It's safe to share the same register.
772 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
775 * If meter is enable, it will engage the register for color
776 * match and flow match. If meter color match is not using the
777 * REG_C_2, need to skip the REG_C_x be used by meter color
779 * If meter is disable, free to use all available registers.
781 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
782 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
783 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
784 if (id > (uint32_t)(REG_C_7 - start_reg))
785 return rte_flow_error_set(error, EINVAL,
786 RTE_FLOW_ERROR_TYPE_ITEM,
787 NULL, "invalid tag id");
788 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
789 return rte_flow_error_set(error, ENOTSUP,
790 RTE_FLOW_ERROR_TYPE_ITEM,
791 NULL, "unsupported tag id");
793 * This case means meter is using the REG_C_x great than 2.
794 * Take care not to conflict with meter color REG_C_x.
795 * If the available index REG_C_y >= REG_C_x, skip the
798 if (skip_mtr_reg && config->flow_mreg_c
799 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
800 if (id >= (uint32_t)(REG_C_7 - start_reg))
801 return rte_flow_error_set(error, EINVAL,
802 RTE_FLOW_ERROR_TYPE_ITEM,
803 NULL, "invalid tag id");
804 if (config->flow_mreg_c
805 [id + 1 + start_reg - REG_C_0] != REG_NON)
806 return config->flow_mreg_c
807 [id + 1 + start_reg - REG_C_0];
808 return rte_flow_error_set(error, ENOTSUP,
809 RTE_FLOW_ERROR_TYPE_ITEM,
810 NULL, "unsupported tag id");
812 return config->flow_mreg_c[id + start_reg - REG_C_0];
815 return rte_flow_error_set(error, EINVAL,
816 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
817 NULL, "invalid feature name");
821 * Check extensive flow metadata register support.
824 * Pointer to rte_eth_dev structure.
827 * True if device supports extensive flow metadata register, otherwise false.
830 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
832 struct mlx5_priv *priv = dev->data->dev_private;
833 struct mlx5_dev_config *config = &priv->config;
836 * Having available reg_c can be regarded inclusively as supporting
837 * extensive flow metadata register, which could mean,
838 * - metadata register copy action by modify header.
839 * - 16 modify header actions is supported.
840 * - reg_c's are preserved across different domain (FDB and NIC) on
841 * packet loopback by flow lookup miss.
843 return config->flow_mreg_c[2] != REG_NON;
847 * Get the lowest priority.
850 * Pointer to the Ethernet device structure.
851 * @param[in] attributes
852 * Pointer to device flow rule attributes.
855 * The value of lowest priority of flow.
858 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
859 const struct rte_flow_attr *attr)
861 struct mlx5_priv *priv = dev->data->dev_private;
863 if (!attr->group && !attr->transfer)
864 return priv->config.flow_prio - 2;
865 return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
869 * Calculate matcher priority of the flow.
872 * Pointer to the Ethernet device structure.
874 * Pointer to device flow rule attributes.
875 * @param[in] subpriority
876 * The priority based on the items.
878 * The matcher priority of the flow.
881 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
882 const struct rte_flow_attr *attr,
883 uint32_t subpriority)
885 uint16_t priority = (uint16_t)attr->priority;
886 struct mlx5_priv *priv = dev->data->dev_private;
888 if (!attr->group && !attr->transfer) {
889 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
890 priority = priv->config.flow_prio - 1;
891 return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
893 if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
894 priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
895 return priority * 3 + subpriority;
899 * Verify the @p item specifications (spec, last, mask) are compatible with the
903 * Item specification.
905 * @p item->mask or flow default bit-masks.
906 * @param[in] nic_mask
907 * Bit-masks covering supported fields by the NIC to compare with user mask.
909 * Bit-masks size in bytes.
910 * @param[in] range_accepted
911 * True if range of values is accepted for specific fields, false otherwise.
913 * Pointer to error structure.
916 * 0 on success, a negative errno value otherwise and rte_errno is set.
919 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
921 const uint8_t *nic_mask,
924 struct rte_flow_error *error)
928 MLX5_ASSERT(nic_mask);
929 for (i = 0; i < size; ++i)
930 if ((nic_mask[i] | mask[i]) != nic_mask[i])
931 return rte_flow_error_set(error, ENOTSUP,
932 RTE_FLOW_ERROR_TYPE_ITEM,
934 "mask enables non supported"
936 if (!item->spec && (item->mask || item->last))
937 return rte_flow_error_set(error, EINVAL,
938 RTE_FLOW_ERROR_TYPE_ITEM, item,
939 "mask/last without a spec is not"
941 if (item->spec && item->last && !range_accepted) {
947 for (i = 0; i < size; ++i) {
948 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
949 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
951 ret = memcmp(spec, last, size);
953 return rte_flow_error_set(error, EINVAL,
954 RTE_FLOW_ERROR_TYPE_ITEM,
956 "range is not valid");
962 * Adjust the hash fields according to the @p flow information.
964 * @param[in] dev_flow.
965 * Pointer to the mlx5_flow.
967 * 1 when the hash field is for a tunnel item.
968 * @param[in] layer_types
970 * @param[in] hash_fields
974 * The hash fields that should be used.
977 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
978 int tunnel __rte_unused, uint64_t layer_types,
979 uint64_t hash_fields)
981 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
982 int rss_request_inner = rss_desc->level >= 2;
984 /* Check RSS hash level for tunnel. */
985 if (tunnel && rss_request_inner)
986 hash_fields |= IBV_RX_HASH_INNER;
987 else if (tunnel || rss_request_inner)
990 /* Check if requested layer matches RSS hash fields. */
991 if (!(rss_desc->types & layer_types))
997 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
998 * if several tunnel rules are used on this queue, the tunnel ptype will be
1002 * Rx queue to update.
1005 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1008 uint32_t tunnel_ptype = 0;
1010 /* Look up for the ptype to use. */
1011 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1012 if (!rxq_ctrl->flow_tunnels_n[i])
1014 if (!tunnel_ptype) {
1015 tunnel_ptype = tunnels_info[i].ptype;
1021 rxq_ctrl->rxq.tunnel = tunnel_ptype;
1025 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1029 * Pointer to the Ethernet device structure.
1030 * @param[in] dev_handle
1031 * Pointer to device flow handle structure.
1034 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1035 struct mlx5_flow_handle *dev_handle)
1037 struct mlx5_priv *priv = dev->data->dev_private;
1038 const int mark = dev_handle->mark;
1039 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1040 struct mlx5_ind_table_obj *ind_tbl = NULL;
1043 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1044 struct mlx5_hrxq *hrxq;
1046 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1047 dev_handle->rix_hrxq);
1049 ind_tbl = hrxq->ind_table;
1050 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1051 struct mlx5_shared_action_rss *shared_rss;
1053 shared_rss = mlx5_ipool_get
1054 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1055 dev_handle->rix_srss);
1057 ind_tbl = shared_rss->ind_tbl;
1061 for (i = 0; i != ind_tbl->queues_n; ++i) {
1062 int idx = ind_tbl->queues[i];
1063 struct mlx5_rxq_ctrl *rxq_ctrl =
1064 container_of((*priv->rxqs)[idx],
1065 struct mlx5_rxq_ctrl, rxq);
1068 * To support metadata register copy on Tx loopback,
1069 * this must be always enabled (metadata may arive
1070 * from other port - not from local flows only.
1072 if (priv->config.dv_flow_en &&
1073 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1074 mlx5_flow_ext_mreg_supported(dev)) {
1075 rxq_ctrl->rxq.mark = 1;
1076 rxq_ctrl->flow_mark_n = 1;
1078 rxq_ctrl->rxq.mark = 1;
1079 rxq_ctrl->flow_mark_n++;
1084 /* Increase the counter matching the flow. */
1085 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1086 if ((tunnels_info[j].tunnel &
1087 dev_handle->layers) ==
1088 tunnels_info[j].tunnel) {
1089 rxq_ctrl->flow_tunnels_n[j]++;
1093 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1099 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1102 * Pointer to the Ethernet device structure.
1104 * Pointer to flow structure.
1107 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1109 struct mlx5_priv *priv = dev->data->dev_private;
1110 uint32_t handle_idx;
1111 struct mlx5_flow_handle *dev_handle;
1113 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1114 handle_idx, dev_handle, next)
1115 flow_drv_rxq_flags_set(dev, dev_handle);
1119 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1120 * device flow if no other flow uses it with the same kind of request.
1123 * Pointer to Ethernet device.
1124 * @param[in] dev_handle
1125 * Pointer to the device flow handle structure.
1128 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1129 struct mlx5_flow_handle *dev_handle)
1131 struct mlx5_priv *priv = dev->data->dev_private;
1132 const int mark = dev_handle->mark;
1133 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1134 struct mlx5_ind_table_obj *ind_tbl = NULL;
1137 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1138 struct mlx5_hrxq *hrxq;
1140 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1141 dev_handle->rix_hrxq);
1143 ind_tbl = hrxq->ind_table;
1144 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1145 struct mlx5_shared_action_rss *shared_rss;
1147 shared_rss = mlx5_ipool_get
1148 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1149 dev_handle->rix_srss);
1151 ind_tbl = shared_rss->ind_tbl;
1155 MLX5_ASSERT(dev->data->dev_started);
1156 for (i = 0; i != ind_tbl->queues_n; ++i) {
1157 int idx = ind_tbl->queues[i];
1158 struct mlx5_rxq_ctrl *rxq_ctrl =
1159 container_of((*priv->rxqs)[idx],
1160 struct mlx5_rxq_ctrl, rxq);
1162 if (priv->config.dv_flow_en &&
1163 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1164 mlx5_flow_ext_mreg_supported(dev)) {
1165 rxq_ctrl->rxq.mark = 1;
1166 rxq_ctrl->flow_mark_n = 1;
1168 rxq_ctrl->flow_mark_n--;
1169 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1174 /* Decrease the counter matching the flow. */
1175 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1176 if ((tunnels_info[j].tunnel &
1177 dev_handle->layers) ==
1178 tunnels_info[j].tunnel) {
1179 rxq_ctrl->flow_tunnels_n[j]--;
1183 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1189 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1190 * @p flow if no other flow uses it with the same kind of request.
1193 * Pointer to Ethernet device.
1195 * Pointer to the flow.
1198 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1200 struct mlx5_priv *priv = dev->data->dev_private;
1201 uint32_t handle_idx;
1202 struct mlx5_flow_handle *dev_handle;
1204 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1205 handle_idx, dev_handle, next)
1206 flow_drv_rxq_flags_trim(dev, dev_handle);
1210 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1213 * Pointer to Ethernet device.
1216 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1218 struct mlx5_priv *priv = dev->data->dev_private;
1221 for (i = 0; i != priv->rxqs_n; ++i) {
1222 struct mlx5_rxq_ctrl *rxq_ctrl;
1225 if (!(*priv->rxqs)[i])
1227 rxq_ctrl = container_of((*priv->rxqs)[i],
1228 struct mlx5_rxq_ctrl, rxq);
1229 rxq_ctrl->flow_mark_n = 0;
1230 rxq_ctrl->rxq.mark = 0;
1231 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1232 rxq_ctrl->flow_tunnels_n[j] = 0;
1233 rxq_ctrl->rxq.tunnel = 0;
1238 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1241 * Pointer to the Ethernet device structure.
1244 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1246 struct mlx5_priv *priv = dev->data->dev_private;
1247 struct mlx5_rxq_data *data;
1250 for (i = 0; i != priv->rxqs_n; ++i) {
1251 if (!(*priv->rxqs)[i])
1253 data = (*priv->rxqs)[i];
1254 if (!rte_flow_dynf_metadata_avail()) {
1255 data->dynf_meta = 0;
1256 data->flow_meta_mask = 0;
1257 data->flow_meta_offset = -1;
1258 data->flow_meta_port_mask = 0;
1260 data->dynf_meta = 1;
1261 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1262 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1263 data->flow_meta_port_mask = (uint32_t)~0;
1264 if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1265 data->flow_meta_port_mask >>= 16;
1271 * return a pointer to the desired action in the list of actions.
1273 * @param[in] actions
1274 * The list of actions to search the action in.
1276 * The action to find.
1279 * Pointer to the action in the list, if found. NULL otherwise.
1281 const struct rte_flow_action *
1282 mlx5_flow_find_action(const struct rte_flow_action *actions,
1283 enum rte_flow_action_type action)
1285 if (actions == NULL)
1287 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1288 if (actions->type == action)
1294 * Validate the flag action.
1296 * @param[in] action_flags
1297 * Bit-fields that holds the actions detected until now.
1299 * Attributes of flow that includes this action.
1301 * Pointer to error structure.
1304 * 0 on success, a negative errno value otherwise and rte_errno is set.
1307 mlx5_flow_validate_action_flag(uint64_t action_flags,
1308 const struct rte_flow_attr *attr,
1309 struct rte_flow_error *error)
1311 if (action_flags & MLX5_FLOW_ACTION_MARK)
1312 return rte_flow_error_set(error, EINVAL,
1313 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1314 "can't mark and flag in same flow");
1315 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1316 return rte_flow_error_set(error, EINVAL,
1317 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1319 " actions in same flow");
1321 return rte_flow_error_set(error, ENOTSUP,
1322 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1323 "flag action not supported for "
1329 * Validate the mark action.
1332 * Pointer to the queue action.
1333 * @param[in] action_flags
1334 * Bit-fields that holds the actions detected until now.
1336 * Attributes of flow that includes this action.
1338 * Pointer to error structure.
1341 * 0 on success, a negative errno value otherwise and rte_errno is set.
1344 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1345 uint64_t action_flags,
1346 const struct rte_flow_attr *attr,
1347 struct rte_flow_error *error)
1349 const struct rte_flow_action_mark *mark = action->conf;
1352 return rte_flow_error_set(error, EINVAL,
1353 RTE_FLOW_ERROR_TYPE_ACTION,
1355 "configuration cannot be null");
1356 if (mark->id >= MLX5_FLOW_MARK_MAX)
1357 return rte_flow_error_set(error, EINVAL,
1358 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1360 "mark id must in 0 <= id < "
1361 RTE_STR(MLX5_FLOW_MARK_MAX));
1362 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1363 return rte_flow_error_set(error, EINVAL,
1364 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1365 "can't flag and mark in same flow");
1366 if (action_flags & MLX5_FLOW_ACTION_MARK)
1367 return rte_flow_error_set(error, EINVAL,
1368 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1369 "can't have 2 mark actions in same"
1372 return rte_flow_error_set(error, ENOTSUP,
1373 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1374 "mark action not supported for "
1380 * Validate the drop action.
1382 * @param[in] action_flags
1383 * Bit-fields that holds the actions detected until now.
1385 * Attributes of flow that includes this action.
1387 * Pointer to error structure.
1390 * 0 on success, a negative errno value otherwise and rte_errno is set.
1393 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1394 const struct rte_flow_attr *attr,
1395 struct rte_flow_error *error)
1398 return rte_flow_error_set(error, ENOTSUP,
1399 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1400 "drop action not supported for "
1406 * Validate the queue action.
1409 * Pointer to the queue action.
1410 * @param[in] action_flags
1411 * Bit-fields that holds the actions detected until now.
1413 * Pointer to the Ethernet device structure.
1415 * Attributes of flow that includes this action.
1417 * Pointer to error structure.
1420 * 0 on success, a negative errno value otherwise and rte_errno is set.
1423 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1424 uint64_t action_flags,
1425 struct rte_eth_dev *dev,
1426 const struct rte_flow_attr *attr,
1427 struct rte_flow_error *error)
1429 struct mlx5_priv *priv = dev->data->dev_private;
1430 const struct rte_flow_action_queue *queue = action->conf;
1432 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1433 return rte_flow_error_set(error, EINVAL,
1434 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1435 "can't have 2 fate actions in"
1438 return rte_flow_error_set(error, EINVAL,
1439 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1440 NULL, "No Rx queues configured");
1441 if (queue->index >= priv->rxqs_n)
1442 return rte_flow_error_set(error, EINVAL,
1443 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1445 "queue index out of range");
1446 if (!(*priv->rxqs)[queue->index])
1447 return rte_flow_error_set(error, EINVAL,
1448 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1450 "queue is not configured");
1452 return rte_flow_error_set(error, ENOTSUP,
1453 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1454 "queue action not supported for "
1460 * Validate the rss action.
1463 * Pointer to the Ethernet device structure.
1465 * Pointer to the queue action.
1467 * Pointer to error structure.
1470 * 0 on success, a negative errno value otherwise and rte_errno is set.
1473 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1474 const struct rte_flow_action *action,
1475 struct rte_flow_error *error)
1477 struct mlx5_priv *priv = dev->data->dev_private;
1478 const struct rte_flow_action_rss *rss = action->conf;
1479 enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1482 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1483 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1484 return rte_flow_error_set(error, ENOTSUP,
1485 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1487 "RSS hash function not supported");
1488 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1493 return rte_flow_error_set(error, ENOTSUP,
1494 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1496 "tunnel RSS is not supported");
1497 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1498 if (rss->key_len == 0 && rss->key != NULL)
1499 return rte_flow_error_set(error, ENOTSUP,
1500 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1502 "RSS hash key length 0");
1503 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1504 return rte_flow_error_set(error, ENOTSUP,
1505 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1507 "RSS hash key too small");
1508 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1509 return rte_flow_error_set(error, ENOTSUP,
1510 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1512 "RSS hash key too large");
1513 if (rss->queue_num > priv->config.ind_table_max_size)
1514 return rte_flow_error_set(error, ENOTSUP,
1515 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1517 "number of queues too large");
1518 if (rss->types & MLX5_RSS_HF_MASK)
1519 return rte_flow_error_set(error, ENOTSUP,
1520 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1522 "some RSS protocols are not"
1524 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1525 !(rss->types & ETH_RSS_IP))
1526 return rte_flow_error_set(error, EINVAL,
1527 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1528 "L3 partial RSS requested but L3 RSS"
1529 " type not specified");
1530 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1531 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1532 return rte_flow_error_set(error, EINVAL,
1533 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1534 "L4 partial RSS requested but L4 RSS"
1535 " type not specified");
1537 return rte_flow_error_set(error, EINVAL,
1538 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1539 NULL, "No Rx queues configured");
1540 if (!rss->queue_num)
1541 return rte_flow_error_set(error, EINVAL,
1542 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1543 NULL, "No queues configured");
1544 for (i = 0; i != rss->queue_num; ++i) {
1545 struct mlx5_rxq_ctrl *rxq_ctrl;
1547 if (rss->queue[i] >= priv->rxqs_n)
1548 return rte_flow_error_set
1550 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1551 &rss->queue[i], "queue index out of range");
1552 if (!(*priv->rxqs)[rss->queue[i]])
1553 return rte_flow_error_set
1554 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1555 &rss->queue[i], "queue is not configured");
1556 rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1557 struct mlx5_rxq_ctrl, rxq);
1559 rxq_type = rxq_ctrl->type;
1560 if (rxq_type != rxq_ctrl->type)
1561 return rte_flow_error_set
1562 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1564 "combining hairpin and regular RSS queues is not supported");
1570 * Validate the rss action.
1573 * Pointer to the queue action.
1574 * @param[in] action_flags
1575 * Bit-fields that holds the actions detected until now.
1577 * Pointer to the Ethernet device structure.
1579 * Attributes of flow that includes this action.
1580 * @param[in] item_flags
1581 * Items that were detected.
1583 * Pointer to error structure.
1586 * 0 on success, a negative errno value otherwise and rte_errno is set.
1589 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1590 uint64_t action_flags,
1591 struct rte_eth_dev *dev,
1592 const struct rte_flow_attr *attr,
1593 uint64_t item_flags,
1594 struct rte_flow_error *error)
1596 const struct rte_flow_action_rss *rss = action->conf;
1597 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1600 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1601 return rte_flow_error_set(error, EINVAL,
1602 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1603 "can't have 2 fate actions"
1605 ret = mlx5_validate_action_rss(dev, action, error);
1609 return rte_flow_error_set(error, ENOTSUP,
1610 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1611 "rss action not supported for "
1613 if (rss->level > 1 && !tunnel)
1614 return rte_flow_error_set(error, EINVAL,
1615 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1616 "inner RSS is not supported for "
1617 "non-tunnel flows");
1618 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1619 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1620 return rte_flow_error_set(error, EINVAL,
1621 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1622 "RSS on eCPRI is not supported now");
1628 * Validate the default miss action.
1630 * @param[in] action_flags
1631 * Bit-fields that holds the actions detected until now.
1633 * Pointer to error structure.
1636 * 0 on success, a negative errno value otherwise and rte_errno is set.
1639 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1640 const struct rte_flow_attr *attr,
1641 struct rte_flow_error *error)
1643 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1644 return rte_flow_error_set(error, EINVAL,
1645 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1646 "can't have 2 fate actions in"
1649 return rte_flow_error_set(error, ENOTSUP,
1650 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1651 "default miss action not supported "
1654 return rte_flow_error_set(error, ENOTSUP,
1655 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1656 "only group 0 is supported");
1658 return rte_flow_error_set(error, ENOTSUP,
1659 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1660 NULL, "transfer is not supported");
1665 * Validate the count action.
1668 * Pointer to the Ethernet device structure.
1670 * Attributes of flow that includes this action.
1672 * Pointer to error structure.
1675 * 0 on success, a negative errno value otherwise and rte_errno is set.
1678 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1679 const struct rte_flow_attr *attr,
1680 struct rte_flow_error *error)
1683 return rte_flow_error_set(error, ENOTSUP,
1684 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1685 "count action not supported for "
1691 * Verify the @p attributes will be correctly understood by the NIC and store
1692 * them in the @p flow if everything is correct.
1695 * Pointer to the Ethernet device structure.
1696 * @param[in] attributes
1697 * Pointer to flow attributes
1699 * Pointer to error structure.
1702 * 0 on success, a negative errno value otherwise and rte_errno is set.
1705 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1706 const struct rte_flow_attr *attributes,
1707 struct rte_flow_error *error)
1709 struct mlx5_priv *priv = dev->data->dev_private;
1710 uint32_t priority_max = priv->config.flow_prio - 1;
1712 if (attributes->group)
1713 return rte_flow_error_set(error, ENOTSUP,
1714 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1715 NULL, "groups is not supported");
1716 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1717 attributes->priority >= priority_max)
1718 return rte_flow_error_set(error, ENOTSUP,
1719 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1720 NULL, "priority out of range");
1721 if (attributes->egress)
1722 return rte_flow_error_set(error, ENOTSUP,
1723 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1724 "egress is not supported");
1725 if (attributes->transfer && !priv->config.dv_esw_en)
1726 return rte_flow_error_set(error, ENOTSUP,
1727 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1728 NULL, "transfer is not supported");
1729 if (!attributes->ingress)
1730 return rte_flow_error_set(error, EINVAL,
1731 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1733 "ingress attribute is mandatory");
1738 * Validate ICMP6 item.
1741 * Item specification.
1742 * @param[in] item_flags
1743 * Bit-fields that holds the items detected until now.
1744 * @param[in] ext_vlan_sup
1745 * Whether extended VLAN features are supported or not.
1747 * Pointer to error structure.
1750 * 0 on success, a negative errno value otherwise and rte_errno is set.
1753 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1754 uint64_t item_flags,
1755 uint8_t target_protocol,
1756 struct rte_flow_error *error)
1758 const struct rte_flow_item_icmp6 *mask = item->mask;
1759 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1760 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1761 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1762 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1763 MLX5_FLOW_LAYER_OUTER_L4;
1766 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1767 return rte_flow_error_set(error, EINVAL,
1768 RTE_FLOW_ERROR_TYPE_ITEM, item,
1769 "protocol filtering not compatible"
1770 " with ICMP6 layer");
1771 if (!(item_flags & l3m))
1772 return rte_flow_error_set(error, EINVAL,
1773 RTE_FLOW_ERROR_TYPE_ITEM, item,
1774 "IPv6 is mandatory to filter on"
1776 if (item_flags & l4m)
1777 return rte_flow_error_set(error, EINVAL,
1778 RTE_FLOW_ERROR_TYPE_ITEM, item,
1779 "multiple L4 layers not supported");
1781 mask = &rte_flow_item_icmp6_mask;
1782 ret = mlx5_flow_item_acceptable
1783 (item, (const uint8_t *)mask,
1784 (const uint8_t *)&rte_flow_item_icmp6_mask,
1785 sizeof(struct rte_flow_item_icmp6),
1786 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1793 * Validate ICMP item.
1796 * Item specification.
1797 * @param[in] item_flags
1798 * Bit-fields that holds the items detected until now.
1800 * Pointer to error structure.
1803 * 0 on success, a negative errno value otherwise and rte_errno is set.
1806 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1807 uint64_t item_flags,
1808 uint8_t target_protocol,
1809 struct rte_flow_error *error)
1811 const struct rte_flow_item_icmp *mask = item->mask;
1812 const struct rte_flow_item_icmp nic_mask = {
1813 .hdr.icmp_type = 0xff,
1814 .hdr.icmp_code = 0xff,
1815 .hdr.icmp_ident = RTE_BE16(0xffff),
1816 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1818 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1819 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1820 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1821 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1822 MLX5_FLOW_LAYER_OUTER_L4;
1825 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1826 return rte_flow_error_set(error, EINVAL,
1827 RTE_FLOW_ERROR_TYPE_ITEM, item,
1828 "protocol filtering not compatible"
1829 " with ICMP layer");
1830 if (!(item_flags & l3m))
1831 return rte_flow_error_set(error, EINVAL,
1832 RTE_FLOW_ERROR_TYPE_ITEM, item,
1833 "IPv4 is mandatory to filter"
1835 if (item_flags & l4m)
1836 return rte_flow_error_set(error, EINVAL,
1837 RTE_FLOW_ERROR_TYPE_ITEM, item,
1838 "multiple L4 layers not supported");
1841 ret = mlx5_flow_item_acceptable
1842 (item, (const uint8_t *)mask,
1843 (const uint8_t *)&nic_mask,
1844 sizeof(struct rte_flow_item_icmp),
1845 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1852 * Validate Ethernet item.
1855 * Item specification.
1856 * @param[in] item_flags
1857 * Bit-fields that holds the items detected until now.
1859 * Pointer to error structure.
1862 * 0 on success, a negative errno value otherwise and rte_errno is set.
1865 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1866 uint64_t item_flags, bool ext_vlan_sup,
1867 struct rte_flow_error *error)
1869 const struct rte_flow_item_eth *mask = item->mask;
1870 const struct rte_flow_item_eth nic_mask = {
1871 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1872 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1873 .type = RTE_BE16(0xffff),
1874 .has_vlan = ext_vlan_sup ? 1 : 0,
1877 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1878 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1879 MLX5_FLOW_LAYER_OUTER_L2;
1881 if (item_flags & ethm)
1882 return rte_flow_error_set(error, ENOTSUP,
1883 RTE_FLOW_ERROR_TYPE_ITEM, item,
1884 "multiple L2 layers not supported");
1885 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1886 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1887 return rte_flow_error_set(error, EINVAL,
1888 RTE_FLOW_ERROR_TYPE_ITEM, item,
1889 "L2 layer should not follow "
1891 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1892 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1893 return rte_flow_error_set(error, EINVAL,
1894 RTE_FLOW_ERROR_TYPE_ITEM, item,
1895 "L2 layer should not follow VLAN");
1897 mask = &rte_flow_item_eth_mask;
1898 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1899 (const uint8_t *)&nic_mask,
1900 sizeof(struct rte_flow_item_eth),
1901 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1906 * Validate VLAN item.
1909 * Item specification.
1910 * @param[in] item_flags
1911 * Bit-fields that holds the items detected until now.
1913 * Ethernet device flow is being created on.
1915 * Pointer to error structure.
1918 * 0 on success, a negative errno value otherwise and rte_errno is set.
1921 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1922 uint64_t item_flags,
1923 struct rte_eth_dev *dev,
1924 struct rte_flow_error *error)
1926 const struct rte_flow_item_vlan *spec = item->spec;
1927 const struct rte_flow_item_vlan *mask = item->mask;
1928 const struct rte_flow_item_vlan nic_mask = {
1929 .tci = RTE_BE16(UINT16_MAX),
1930 .inner_type = RTE_BE16(UINT16_MAX),
1932 uint16_t vlan_tag = 0;
1933 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1935 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1936 MLX5_FLOW_LAYER_INNER_L4) :
1937 (MLX5_FLOW_LAYER_OUTER_L3 |
1938 MLX5_FLOW_LAYER_OUTER_L4);
1939 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1940 MLX5_FLOW_LAYER_OUTER_VLAN;
1942 if (item_flags & vlanm)
1943 return rte_flow_error_set(error, EINVAL,
1944 RTE_FLOW_ERROR_TYPE_ITEM, item,
1945 "multiple VLAN layers not supported");
1946 else if ((item_flags & l34m) != 0)
1947 return rte_flow_error_set(error, EINVAL,
1948 RTE_FLOW_ERROR_TYPE_ITEM, item,
1949 "VLAN cannot follow L3/L4 layer");
1951 mask = &rte_flow_item_vlan_mask;
1952 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1953 (const uint8_t *)&nic_mask,
1954 sizeof(struct rte_flow_item_vlan),
1955 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1958 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1959 struct mlx5_priv *priv = dev->data->dev_private;
1961 if (priv->vmwa_context) {
1963 * Non-NULL context means we have a virtual machine
1964 * and SR-IOV enabled, we have to create VLAN interface
1965 * to make hypervisor to setup E-Switch vport
1966 * context correctly. We avoid creating the multiple
1967 * VLAN interfaces, so we cannot support VLAN tag mask.
1969 return rte_flow_error_set(error, EINVAL,
1970 RTE_FLOW_ERROR_TYPE_ITEM,
1972 "VLAN tag mask is not"
1973 " supported in virtual"
1978 vlan_tag = spec->tci;
1979 vlan_tag &= mask->tci;
1982 * From verbs perspective an empty VLAN is equivalent
1983 * to a packet without VLAN layer.
1986 return rte_flow_error_set(error, EINVAL,
1987 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1989 "VLAN cannot be empty");
1994 * Validate IPV4 item.
1997 * Item specification.
1998 * @param[in] item_flags
1999 * Bit-fields that holds the items detected until now.
2000 * @param[in] last_item
2001 * Previous validated item in the pattern items.
2002 * @param[in] ether_type
2003 * Type in the ethernet layer header (including dot1q).
2004 * @param[in] acc_mask
2005 * Acceptable mask, if NULL default internal default mask
2006 * will be used to check whether item fields are supported.
2007 * @param[in] range_accepted
2008 * True if range of values is accepted for specific fields, false otherwise.
2010 * Pointer to error structure.
2013 * 0 on success, a negative errno value otherwise and rte_errno is set.
2016 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2017 uint64_t item_flags,
2019 uint16_t ether_type,
2020 const struct rte_flow_item_ipv4 *acc_mask,
2021 bool range_accepted,
2022 struct rte_flow_error *error)
2024 const struct rte_flow_item_ipv4 *mask = item->mask;
2025 const struct rte_flow_item_ipv4 *spec = item->spec;
2026 const struct rte_flow_item_ipv4 nic_mask = {
2028 .src_addr = RTE_BE32(0xffffffff),
2029 .dst_addr = RTE_BE32(0xffffffff),
2030 .type_of_service = 0xff,
2031 .next_proto_id = 0xff,
2034 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2035 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2036 MLX5_FLOW_LAYER_OUTER_L3;
2037 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2038 MLX5_FLOW_LAYER_OUTER_L4;
2040 uint8_t next_proto = 0xFF;
2041 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2042 MLX5_FLOW_LAYER_OUTER_VLAN |
2043 MLX5_FLOW_LAYER_INNER_VLAN);
2045 if ((last_item & l2_vlan) && ether_type &&
2046 ether_type != RTE_ETHER_TYPE_IPV4)
2047 return rte_flow_error_set(error, EINVAL,
2048 RTE_FLOW_ERROR_TYPE_ITEM, item,
2049 "IPv4 cannot follow L2/VLAN layer "
2050 "which ether type is not IPv4");
2051 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2053 next_proto = mask->hdr.next_proto_id &
2054 spec->hdr.next_proto_id;
2055 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2056 return rte_flow_error_set(error, EINVAL,
2057 RTE_FLOW_ERROR_TYPE_ITEM,
2062 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2063 return rte_flow_error_set(error, EINVAL,
2064 RTE_FLOW_ERROR_TYPE_ITEM, item,
2065 "wrong tunnel type - IPv6 specified "
2066 "but IPv4 item provided");
2067 if (item_flags & l3m)
2068 return rte_flow_error_set(error, ENOTSUP,
2069 RTE_FLOW_ERROR_TYPE_ITEM, item,
2070 "multiple L3 layers not supported");
2071 else if (item_flags & l4m)
2072 return rte_flow_error_set(error, EINVAL,
2073 RTE_FLOW_ERROR_TYPE_ITEM, item,
2074 "L3 cannot follow an L4 layer.");
2075 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2076 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2077 return rte_flow_error_set(error, EINVAL,
2078 RTE_FLOW_ERROR_TYPE_ITEM, item,
2079 "L3 cannot follow an NVGRE layer.");
2081 mask = &rte_flow_item_ipv4_mask;
2082 else if (mask->hdr.next_proto_id != 0 &&
2083 mask->hdr.next_proto_id != 0xff)
2084 return rte_flow_error_set(error, EINVAL,
2085 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2086 "partial mask is not supported"
2088 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2089 acc_mask ? (const uint8_t *)acc_mask
2090 : (const uint8_t *)&nic_mask,
2091 sizeof(struct rte_flow_item_ipv4),
2092 range_accepted, error);
2099 * Validate IPV6 item.
2102 * Item specification.
2103 * @param[in] item_flags
2104 * Bit-fields that holds the items detected until now.
2105 * @param[in] last_item
2106 * Previous validated item in the pattern items.
2107 * @param[in] ether_type
2108 * Type in the ethernet layer header (including dot1q).
2109 * @param[in] acc_mask
2110 * Acceptable mask, if NULL default internal default mask
2111 * will be used to check whether item fields are supported.
2113 * Pointer to error structure.
2116 * 0 on success, a negative errno value otherwise and rte_errno is set.
2119 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2120 uint64_t item_flags,
2122 uint16_t ether_type,
2123 const struct rte_flow_item_ipv6 *acc_mask,
2124 struct rte_flow_error *error)
2126 const struct rte_flow_item_ipv6 *mask = item->mask;
2127 const struct rte_flow_item_ipv6 *spec = item->spec;
2128 const struct rte_flow_item_ipv6 nic_mask = {
2131 "\xff\xff\xff\xff\xff\xff\xff\xff"
2132 "\xff\xff\xff\xff\xff\xff\xff\xff",
2134 "\xff\xff\xff\xff\xff\xff\xff\xff"
2135 "\xff\xff\xff\xff\xff\xff\xff\xff",
2136 .vtc_flow = RTE_BE32(0xffffffff),
2140 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2141 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2142 MLX5_FLOW_LAYER_OUTER_L3;
2143 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2144 MLX5_FLOW_LAYER_OUTER_L4;
2146 uint8_t next_proto = 0xFF;
2147 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2148 MLX5_FLOW_LAYER_OUTER_VLAN |
2149 MLX5_FLOW_LAYER_INNER_VLAN);
2151 if ((last_item & l2_vlan) && ether_type &&
2152 ether_type != RTE_ETHER_TYPE_IPV6)
2153 return rte_flow_error_set(error, EINVAL,
2154 RTE_FLOW_ERROR_TYPE_ITEM, item,
2155 "IPv6 cannot follow L2/VLAN layer "
2156 "which ether type is not IPv6");
2157 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2158 next_proto = spec->hdr.proto;
2159 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2160 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2161 return rte_flow_error_set(error, EINVAL,
2162 RTE_FLOW_ERROR_TYPE_ITEM,
2167 if (next_proto == IPPROTO_HOPOPTS ||
2168 next_proto == IPPROTO_ROUTING ||
2169 next_proto == IPPROTO_FRAGMENT ||
2170 next_proto == IPPROTO_ESP ||
2171 next_proto == IPPROTO_AH ||
2172 next_proto == IPPROTO_DSTOPTS)
2173 return rte_flow_error_set(error, EINVAL,
2174 RTE_FLOW_ERROR_TYPE_ITEM, item,
2175 "IPv6 proto (next header) should "
2176 "not be set as extension header");
2177 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2178 return rte_flow_error_set(error, EINVAL,
2179 RTE_FLOW_ERROR_TYPE_ITEM, item,
2180 "wrong tunnel type - IPv4 specified "
2181 "but IPv6 item provided");
2182 if (item_flags & l3m)
2183 return rte_flow_error_set(error, ENOTSUP,
2184 RTE_FLOW_ERROR_TYPE_ITEM, item,
2185 "multiple L3 layers not supported");
2186 else if (item_flags & l4m)
2187 return rte_flow_error_set(error, EINVAL,
2188 RTE_FLOW_ERROR_TYPE_ITEM, item,
2189 "L3 cannot follow an L4 layer.");
2190 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2191 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2192 return rte_flow_error_set(error, EINVAL,
2193 RTE_FLOW_ERROR_TYPE_ITEM, item,
2194 "L3 cannot follow an NVGRE layer.");
2196 mask = &rte_flow_item_ipv6_mask;
2197 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2198 acc_mask ? (const uint8_t *)acc_mask
2199 : (const uint8_t *)&nic_mask,
2200 sizeof(struct rte_flow_item_ipv6),
2201 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2208 * Validate UDP item.
2211 * Item specification.
2212 * @param[in] item_flags
2213 * Bit-fields that holds the items detected until now.
2214 * @param[in] target_protocol
2215 * The next protocol in the previous item.
2216 * @param[in] flow_mask
2217 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2219 * Pointer to error structure.
2222 * 0 on success, a negative errno value otherwise and rte_errno is set.
2225 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2226 uint64_t item_flags,
2227 uint8_t target_protocol,
2228 struct rte_flow_error *error)
2230 const struct rte_flow_item_udp *mask = item->mask;
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;
2238 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2239 return rte_flow_error_set(error, EINVAL,
2240 RTE_FLOW_ERROR_TYPE_ITEM, item,
2241 "protocol filtering not compatible"
2243 if (!(item_flags & l3m))
2244 return rte_flow_error_set(error, EINVAL,
2245 RTE_FLOW_ERROR_TYPE_ITEM, item,
2246 "L3 is mandatory to filter on L4");
2247 if (item_flags & l4m)
2248 return rte_flow_error_set(error, EINVAL,
2249 RTE_FLOW_ERROR_TYPE_ITEM, item,
2250 "multiple L4 layers not supported");
2252 mask = &rte_flow_item_udp_mask;
2253 ret = mlx5_flow_item_acceptable
2254 (item, (const uint8_t *)mask,
2255 (const uint8_t *)&rte_flow_item_udp_mask,
2256 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2264 * Validate TCP item.
2267 * Item specification.
2268 * @param[in] item_flags
2269 * Bit-fields that holds the items detected until now.
2270 * @param[in] target_protocol
2271 * The next protocol in the previous item.
2273 * Pointer to error structure.
2276 * 0 on success, a negative errno value otherwise and rte_errno is set.
2279 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2280 uint64_t item_flags,
2281 uint8_t target_protocol,
2282 const struct rte_flow_item_tcp *flow_mask,
2283 struct rte_flow_error *error)
2285 const struct rte_flow_item_tcp *mask = item->mask;
2286 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2287 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2288 MLX5_FLOW_LAYER_OUTER_L3;
2289 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2290 MLX5_FLOW_LAYER_OUTER_L4;
2293 MLX5_ASSERT(flow_mask);
2294 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2295 return rte_flow_error_set(error, EINVAL,
2296 RTE_FLOW_ERROR_TYPE_ITEM, item,
2297 "protocol filtering not compatible"
2299 if (!(item_flags & l3m))
2300 return rte_flow_error_set(error, EINVAL,
2301 RTE_FLOW_ERROR_TYPE_ITEM, item,
2302 "L3 is mandatory to filter on L4");
2303 if (item_flags & l4m)
2304 return rte_flow_error_set(error, EINVAL,
2305 RTE_FLOW_ERROR_TYPE_ITEM, item,
2306 "multiple L4 layers not supported");
2308 mask = &rte_flow_item_tcp_mask;
2309 ret = mlx5_flow_item_acceptable
2310 (item, (const uint8_t *)mask,
2311 (const uint8_t *)flow_mask,
2312 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2320 * Validate VXLAN item.
2323 * Item specification.
2324 * @param[in] item_flags
2325 * Bit-fields that holds the items detected until now.
2326 * @param[in] target_protocol
2327 * The next protocol in the previous item.
2329 * Pointer to error structure.
2332 * 0 on success, a negative errno value otherwise and rte_errno is set.
2335 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2336 uint64_t item_flags,
2337 struct rte_flow_error *error)
2339 const struct rte_flow_item_vxlan *spec = item->spec;
2340 const struct rte_flow_item_vxlan *mask = item->mask;
2345 } id = { .vlan_id = 0, };
2348 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2349 return rte_flow_error_set(error, ENOTSUP,
2350 RTE_FLOW_ERROR_TYPE_ITEM, item,
2351 "multiple tunnel layers not"
2354 * Verify only UDPv4 is present as defined in
2355 * https://tools.ietf.org/html/rfc7348
2357 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2358 return rte_flow_error_set(error, EINVAL,
2359 RTE_FLOW_ERROR_TYPE_ITEM, item,
2360 "no outer UDP layer found");
2362 mask = &rte_flow_item_vxlan_mask;
2363 ret = mlx5_flow_item_acceptable
2364 (item, (const uint8_t *)mask,
2365 (const uint8_t *)&rte_flow_item_vxlan_mask,
2366 sizeof(struct rte_flow_item_vxlan),
2367 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2371 memcpy(&id.vni[1], spec->vni, 3);
2372 memcpy(&id.vni[1], mask->vni, 3);
2374 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2375 return rte_flow_error_set(error, ENOTSUP,
2376 RTE_FLOW_ERROR_TYPE_ITEM, item,
2377 "VXLAN tunnel must be fully defined");
2382 * Validate VXLAN_GPE item.
2385 * Item specification.
2386 * @param[in] item_flags
2387 * Bit-fields that holds the items detected until now.
2389 * Pointer to the private data structure.
2390 * @param[in] target_protocol
2391 * The next protocol in the previous item.
2393 * Pointer to error structure.
2396 * 0 on success, a negative errno value otherwise and rte_errno is set.
2399 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2400 uint64_t item_flags,
2401 struct rte_eth_dev *dev,
2402 struct rte_flow_error *error)
2404 struct mlx5_priv *priv = dev->data->dev_private;
2405 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2406 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2411 } id = { .vlan_id = 0, };
2413 if (!priv->config.l3_vxlan_en)
2414 return rte_flow_error_set(error, ENOTSUP,
2415 RTE_FLOW_ERROR_TYPE_ITEM, item,
2416 "L3 VXLAN is not enabled by device"
2417 " parameter and/or not configured in"
2419 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2420 return rte_flow_error_set(error, ENOTSUP,
2421 RTE_FLOW_ERROR_TYPE_ITEM, item,
2422 "multiple tunnel layers not"
2425 * Verify only UDPv4 is present as defined in
2426 * https://tools.ietf.org/html/rfc7348
2428 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2429 return rte_flow_error_set(error, EINVAL,
2430 RTE_FLOW_ERROR_TYPE_ITEM, item,
2431 "no outer UDP layer found");
2433 mask = &rte_flow_item_vxlan_gpe_mask;
2434 ret = mlx5_flow_item_acceptable
2435 (item, (const uint8_t *)mask,
2436 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2437 sizeof(struct rte_flow_item_vxlan_gpe),
2438 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2443 return rte_flow_error_set(error, ENOTSUP,
2444 RTE_FLOW_ERROR_TYPE_ITEM,
2446 "VxLAN-GPE protocol"
2448 memcpy(&id.vni[1], spec->vni, 3);
2449 memcpy(&id.vni[1], mask->vni, 3);
2451 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2452 return rte_flow_error_set(error, ENOTSUP,
2453 RTE_FLOW_ERROR_TYPE_ITEM, item,
2454 "VXLAN-GPE tunnel must be fully"
2459 * Validate GRE Key item.
2462 * Item specification.
2463 * @param[in] item_flags
2464 * Bit flags to mark detected items.
2465 * @param[in] gre_item
2466 * Pointer to gre_item
2468 * Pointer to error structure.
2471 * 0 on success, a negative errno value otherwise and rte_errno is set.
2474 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2475 uint64_t item_flags,
2476 const struct rte_flow_item *gre_item,
2477 struct rte_flow_error *error)
2479 const rte_be32_t *mask = item->mask;
2481 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2482 const struct rte_flow_item_gre *gre_spec;
2483 const struct rte_flow_item_gre *gre_mask;
2485 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2486 return rte_flow_error_set(error, ENOTSUP,
2487 RTE_FLOW_ERROR_TYPE_ITEM, item,
2488 "Multiple GRE key not support");
2489 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2490 return rte_flow_error_set(error, ENOTSUP,
2491 RTE_FLOW_ERROR_TYPE_ITEM, item,
2492 "No preceding GRE header");
2493 if (item_flags & MLX5_FLOW_LAYER_INNER)
2494 return rte_flow_error_set(error, ENOTSUP,
2495 RTE_FLOW_ERROR_TYPE_ITEM, item,
2496 "GRE key following a wrong item");
2497 gre_mask = gre_item->mask;
2499 gre_mask = &rte_flow_item_gre_mask;
2500 gre_spec = gre_item->spec;
2501 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2502 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2503 return rte_flow_error_set(error, EINVAL,
2504 RTE_FLOW_ERROR_TYPE_ITEM, item,
2505 "Key bit must be on");
2508 mask = &gre_key_default_mask;
2509 ret = mlx5_flow_item_acceptable
2510 (item, (const uint8_t *)mask,
2511 (const uint8_t *)&gre_key_default_mask,
2512 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2517 * Validate GRE item.
2520 * Item specification.
2521 * @param[in] item_flags
2522 * Bit flags to mark detected items.
2523 * @param[in] target_protocol
2524 * The next protocol in the previous item.
2526 * Pointer to error structure.
2529 * 0 on success, a negative errno value otherwise and rte_errno is set.
2532 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2533 uint64_t item_flags,
2534 uint8_t target_protocol,
2535 struct rte_flow_error *error)
2537 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2538 const struct rte_flow_item_gre *mask = item->mask;
2540 const struct rte_flow_item_gre nic_mask = {
2541 .c_rsvd0_ver = RTE_BE16(0xB000),
2542 .protocol = RTE_BE16(UINT16_MAX),
2545 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2546 return rte_flow_error_set(error, EINVAL,
2547 RTE_FLOW_ERROR_TYPE_ITEM, item,
2548 "protocol filtering not compatible"
2549 " with this GRE layer");
2550 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2551 return rte_flow_error_set(error, ENOTSUP,
2552 RTE_FLOW_ERROR_TYPE_ITEM, item,
2553 "multiple tunnel layers not"
2555 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2556 return rte_flow_error_set(error, ENOTSUP,
2557 RTE_FLOW_ERROR_TYPE_ITEM, item,
2558 "L3 Layer is missing");
2560 mask = &rte_flow_item_gre_mask;
2561 ret = mlx5_flow_item_acceptable
2562 (item, (const uint8_t *)mask,
2563 (const uint8_t *)&nic_mask,
2564 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2568 #ifndef HAVE_MLX5DV_DR
2569 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2570 if (spec && (spec->protocol & mask->protocol))
2571 return rte_flow_error_set(error, ENOTSUP,
2572 RTE_FLOW_ERROR_TYPE_ITEM, item,
2573 "without MPLS support the"
2574 " specification cannot be used for"
2582 * Validate Geneve item.
2585 * Item specification.
2586 * @param[in] itemFlags
2587 * Bit-fields that holds the items detected until now.
2589 * Pointer to the private data structure.
2591 * Pointer to error structure.
2594 * 0 on success, a negative errno value otherwise and rte_errno is set.
2598 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2599 uint64_t item_flags,
2600 struct rte_eth_dev *dev,
2601 struct rte_flow_error *error)
2603 struct mlx5_priv *priv = dev->data->dev_private;
2604 const struct rte_flow_item_geneve *spec = item->spec;
2605 const struct rte_flow_item_geneve *mask = item->mask;
2608 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2609 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2610 const struct rte_flow_item_geneve nic_mask = {
2611 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2612 .vni = "\xff\xff\xff",
2613 .protocol = RTE_BE16(UINT16_MAX),
2616 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2617 return rte_flow_error_set(error, ENOTSUP,
2618 RTE_FLOW_ERROR_TYPE_ITEM, item,
2619 "L3 Geneve is not enabled by device"
2620 " parameter and/or not configured in"
2622 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2623 return rte_flow_error_set(error, ENOTSUP,
2624 RTE_FLOW_ERROR_TYPE_ITEM, item,
2625 "multiple tunnel layers not"
2628 * Verify only UDPv4 is present as defined in
2629 * https://tools.ietf.org/html/rfc7348
2631 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2632 return rte_flow_error_set(error, EINVAL,
2633 RTE_FLOW_ERROR_TYPE_ITEM, item,
2634 "no outer UDP layer found");
2636 mask = &rte_flow_item_geneve_mask;
2637 ret = mlx5_flow_item_acceptable
2638 (item, (const uint8_t *)mask,
2639 (const uint8_t *)&nic_mask,
2640 sizeof(struct rte_flow_item_geneve),
2641 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2645 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2646 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2647 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2648 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2649 return rte_flow_error_set(error, ENOTSUP,
2650 RTE_FLOW_ERROR_TYPE_ITEM,
2652 "Geneve protocol unsupported"
2653 " fields are being used");
2654 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2655 return rte_flow_error_set
2657 RTE_FLOW_ERROR_TYPE_ITEM,
2659 "Unsupported Geneve options length");
2661 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2662 return rte_flow_error_set
2664 RTE_FLOW_ERROR_TYPE_ITEM, item,
2665 "Geneve tunnel must be fully defined");
2670 * Validate Geneve TLV option item.
2673 * Item specification.
2674 * @param[in] last_item
2675 * Previous validated item in the pattern items.
2676 * @param[in] geneve_item
2677 * Previous GENEVE item specification.
2679 * Pointer to the rte_eth_dev structure.
2681 * Pointer to error structure.
2684 * 0 on success, a negative errno value otherwise and rte_errno is set.
2687 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2689 const struct rte_flow_item *geneve_item,
2690 struct rte_eth_dev *dev,
2691 struct rte_flow_error *error)
2693 struct mlx5_priv *priv = dev->data->dev_private;
2694 struct mlx5_dev_ctx_shared *sh = priv->sh;
2695 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2696 struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2697 uint8_t data_max_supported =
2698 hca_attr->max_geneve_tlv_option_data_len * 4;
2699 struct mlx5_dev_config *config = &priv->config;
2700 const struct rte_flow_item_geneve *geneve_spec;
2701 const struct rte_flow_item_geneve *geneve_mask;
2702 const struct rte_flow_item_geneve_opt *spec = item->spec;
2703 const struct rte_flow_item_geneve_opt *mask = item->mask;
2705 unsigned int data_len;
2706 uint8_t tlv_option_len;
2707 uint16_t optlen_m, optlen_v;
2708 const struct rte_flow_item_geneve_opt full_mask = {
2709 .option_class = RTE_BE16(0xffff),
2710 .option_type = 0xff,
2715 mask = &rte_flow_item_geneve_opt_mask;
2717 return rte_flow_error_set
2718 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2719 "Geneve TLV opt class/type/length must be specified");
2720 if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2721 return rte_flow_error_set
2722 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2723 "Geneve TLV opt length exceeeds the limit (31)");
2724 /* Check if class type and length masks are full. */
2725 if (full_mask.option_class != mask->option_class ||
2726 full_mask.option_type != mask->option_type ||
2727 full_mask.option_len != (mask->option_len & full_mask.option_len))
2728 return rte_flow_error_set
2729 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2730 "Geneve TLV opt class/type/length masks must be full");
2731 /* Check if length is supported */
2732 if ((uint32_t)spec->option_len >
2733 config->hca_attr.max_geneve_tlv_option_data_len)
2734 return rte_flow_error_set
2735 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2736 "Geneve TLV opt length not supported");
2737 if (config->hca_attr.max_geneve_tlv_options > 1)
2739 "max_geneve_tlv_options supports more than 1 option");
2740 /* Check GENEVE item preceding. */
2741 if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2742 return rte_flow_error_set
2743 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2744 "Geneve opt item must be preceded with Geneve item");
2745 geneve_spec = geneve_item->spec;
2746 geneve_mask = geneve_item->mask ? geneve_item->mask :
2747 &rte_flow_item_geneve_mask;
2748 /* Check if GENEVE TLV option size doesn't exceed option length */
2749 if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2750 geneve_spec->ver_opt_len_o_c_rsvd0)) {
2751 tlv_option_len = spec->option_len & mask->option_len;
2752 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2753 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2754 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2755 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2756 if ((optlen_v & optlen_m) <= tlv_option_len)
2757 return rte_flow_error_set
2758 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2759 "GENEVE TLV option length exceeds optlen");
2761 /* Check if length is 0 or data is 0. */
2762 if (spec->data == NULL || spec->option_len == 0)
2763 return rte_flow_error_set
2764 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2765 "Geneve TLV opt with zero data/length not supported");
2766 /* Check not all data & mask are 0. */
2767 data_len = spec->option_len * 4;
2768 if (mask->data == NULL) {
2769 for (i = 0; i < data_len; i++)
2773 return rte_flow_error_set(error, ENOTSUP,
2774 RTE_FLOW_ERROR_TYPE_ITEM, item,
2775 "Can't match on Geneve option data 0");
2777 for (i = 0; i < data_len; i++)
2778 if (spec->data[i] & mask->data[i])
2781 return rte_flow_error_set(error, ENOTSUP,
2782 RTE_FLOW_ERROR_TYPE_ITEM, item,
2783 "Can't match on Geneve option data and mask 0");
2784 /* Check data mask supported. */
2785 for (i = data_max_supported; i < data_len ; i++)
2787 return rte_flow_error_set(error, ENOTSUP,
2788 RTE_FLOW_ERROR_TYPE_ITEM, item,
2789 "Data mask is of unsupported size");
2791 /* Check GENEVE option is supported in NIC. */
2792 if (!config->hca_attr.geneve_tlv_opt)
2793 return rte_flow_error_set
2794 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2795 "Geneve TLV opt not supported");
2796 /* Check if we already have geneve option with different type/class. */
2797 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2798 geneve_opt_resource = sh->geneve_tlv_option_resource;
2799 if (geneve_opt_resource != NULL)
2800 if (geneve_opt_resource->option_class != spec->option_class ||
2801 geneve_opt_resource->option_type != spec->option_type ||
2802 geneve_opt_resource->length != spec->option_len) {
2803 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2804 return rte_flow_error_set(error, ENOTSUP,
2805 RTE_FLOW_ERROR_TYPE_ITEM, item,
2806 "Only one Geneve TLV option supported");
2808 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2813 * Validate MPLS item.
2816 * Pointer to the rte_eth_dev structure.
2818 * Item specification.
2819 * @param[in] item_flags
2820 * Bit-fields that holds the items detected until now.
2821 * @param[in] prev_layer
2822 * The protocol layer indicated in previous item.
2824 * Pointer to error structure.
2827 * 0 on success, a negative errno value otherwise and rte_errno is set.
2830 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2831 const struct rte_flow_item *item __rte_unused,
2832 uint64_t item_flags __rte_unused,
2833 uint64_t prev_layer __rte_unused,
2834 struct rte_flow_error *error)
2836 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2837 const struct rte_flow_item_mpls *mask = item->mask;
2838 struct mlx5_priv *priv = dev->data->dev_private;
2841 if (!priv->config.mpls_en)
2842 return rte_flow_error_set(error, ENOTSUP,
2843 RTE_FLOW_ERROR_TYPE_ITEM, item,
2844 "MPLS not supported or"
2845 " disabled in firmware"
2847 /* MPLS over IP, UDP, GRE is allowed */
2848 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2849 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2850 MLX5_FLOW_LAYER_GRE |
2851 MLX5_FLOW_LAYER_GRE_KEY)))
2852 return rte_flow_error_set(error, EINVAL,
2853 RTE_FLOW_ERROR_TYPE_ITEM, item,
2854 "protocol filtering not compatible"
2855 " with MPLS layer");
2856 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2857 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2858 !(item_flags & MLX5_FLOW_LAYER_GRE))
2859 return rte_flow_error_set(error, ENOTSUP,
2860 RTE_FLOW_ERROR_TYPE_ITEM, item,
2861 "multiple tunnel layers not"
2864 mask = &rte_flow_item_mpls_mask;
2865 ret = mlx5_flow_item_acceptable
2866 (item, (const uint8_t *)mask,
2867 (const uint8_t *)&rte_flow_item_mpls_mask,
2868 sizeof(struct rte_flow_item_mpls),
2869 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2874 return rte_flow_error_set(error, ENOTSUP,
2875 RTE_FLOW_ERROR_TYPE_ITEM, item,
2876 "MPLS is not supported by Verbs, please"
2882 * Validate NVGRE item.
2885 * Item specification.
2886 * @param[in] item_flags
2887 * Bit flags to mark detected items.
2888 * @param[in] target_protocol
2889 * The next protocol in the previous item.
2891 * Pointer to error structure.
2894 * 0 on success, a negative errno value otherwise and rte_errno is set.
2897 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2898 uint64_t item_flags,
2899 uint8_t target_protocol,
2900 struct rte_flow_error *error)
2902 const struct rte_flow_item_nvgre *mask = item->mask;
2905 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2906 return rte_flow_error_set(error, EINVAL,
2907 RTE_FLOW_ERROR_TYPE_ITEM, item,
2908 "protocol filtering not compatible"
2909 " with this GRE layer");
2910 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2911 return rte_flow_error_set(error, ENOTSUP,
2912 RTE_FLOW_ERROR_TYPE_ITEM, item,
2913 "multiple tunnel layers not"
2915 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2916 return rte_flow_error_set(error, ENOTSUP,
2917 RTE_FLOW_ERROR_TYPE_ITEM, item,
2918 "L3 Layer is missing");
2920 mask = &rte_flow_item_nvgre_mask;
2921 ret = mlx5_flow_item_acceptable
2922 (item, (const uint8_t *)mask,
2923 (const uint8_t *)&rte_flow_item_nvgre_mask,
2924 sizeof(struct rte_flow_item_nvgre),
2925 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2932 * Validate eCPRI item.
2935 * Item specification.
2936 * @param[in] item_flags
2937 * Bit-fields that holds the items detected until now.
2938 * @param[in] last_item
2939 * Previous validated item in the pattern items.
2940 * @param[in] ether_type
2941 * Type in the ethernet layer header (including dot1q).
2942 * @param[in] acc_mask
2943 * Acceptable mask, if NULL default internal default mask
2944 * will be used to check whether item fields are supported.
2946 * Pointer to error structure.
2949 * 0 on success, a negative errno value otherwise and rte_errno is set.
2952 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2953 uint64_t item_flags,
2955 uint16_t ether_type,
2956 const struct rte_flow_item_ecpri *acc_mask,
2957 struct rte_flow_error *error)
2959 const struct rte_flow_item_ecpri *mask = item->mask;
2960 const struct rte_flow_item_ecpri nic_mask = {
2964 RTE_BE32(((const struct rte_ecpri_common_hdr) {
2968 .dummy[0] = 0xFFFFFFFF,
2971 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2972 MLX5_FLOW_LAYER_OUTER_VLAN);
2973 struct rte_flow_item_ecpri mask_lo;
2975 if (!(last_item & outer_l2_vlan) &&
2976 last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2977 return rte_flow_error_set(error, EINVAL,
2978 RTE_FLOW_ERROR_TYPE_ITEM, item,
2979 "eCPRI can only follow L2/VLAN layer or UDP layer");
2980 if ((last_item & outer_l2_vlan) && ether_type &&
2981 ether_type != RTE_ETHER_TYPE_ECPRI)
2982 return rte_flow_error_set(error, EINVAL,
2983 RTE_FLOW_ERROR_TYPE_ITEM, item,
2984 "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2985 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2986 return rte_flow_error_set(error, EINVAL,
2987 RTE_FLOW_ERROR_TYPE_ITEM, item,
2988 "eCPRI with tunnel is not supported right now");
2989 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2990 return rte_flow_error_set(error, ENOTSUP,
2991 RTE_FLOW_ERROR_TYPE_ITEM, item,
2992 "multiple L3 layers not supported");
2993 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2994 return rte_flow_error_set(error, EINVAL,
2995 RTE_FLOW_ERROR_TYPE_ITEM, item,
2996 "eCPRI cannot coexist with a TCP layer");
2997 /* In specification, eCPRI could be over UDP layer. */
2998 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2999 return rte_flow_error_set(error, EINVAL,
3000 RTE_FLOW_ERROR_TYPE_ITEM, item,
3001 "eCPRI over UDP layer is not yet supported right now");
3002 /* Mask for type field in common header could be zero. */
3004 mask = &rte_flow_item_ecpri_mask;
3005 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3006 /* Input mask is in big-endian format. */
3007 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3008 return rte_flow_error_set(error, EINVAL,
3009 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3010 "partial mask is not supported for protocol");
3011 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3012 return rte_flow_error_set(error, EINVAL,
3013 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3014 "message header mask must be after a type mask");
3015 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3016 acc_mask ? (const uint8_t *)acc_mask
3017 : (const uint8_t *)&nic_mask,
3018 sizeof(struct rte_flow_item_ecpri),
3019 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3023 * Release resource related QUEUE/RSS action split.
3026 * Pointer to Ethernet device.
3028 * Flow to release id's from.
3031 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3032 struct rte_flow *flow)
3034 struct mlx5_priv *priv = dev->data->dev_private;
3035 uint32_t handle_idx;
3036 struct mlx5_flow_handle *dev_handle;
3038 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3039 handle_idx, dev_handle, next)
3040 if (dev_handle->split_flow_id &&
3041 !dev_handle->is_meter_flow_id)
3042 mlx5_ipool_free(priv->sh->ipool
3043 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3044 dev_handle->split_flow_id);
3048 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3049 const struct rte_flow_attr *attr __rte_unused,
3050 const struct rte_flow_item items[] __rte_unused,
3051 const struct rte_flow_action actions[] __rte_unused,
3052 bool external __rte_unused,
3053 int hairpin __rte_unused,
3054 struct rte_flow_error *error)
3056 return rte_flow_error_set(error, ENOTSUP,
3057 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3060 static struct mlx5_flow *
3061 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3062 const struct rte_flow_attr *attr __rte_unused,
3063 const struct rte_flow_item items[] __rte_unused,
3064 const struct rte_flow_action actions[] __rte_unused,
3065 struct rte_flow_error *error)
3067 rte_flow_error_set(error, ENOTSUP,
3068 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3073 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3074 struct mlx5_flow *dev_flow __rte_unused,
3075 const struct rte_flow_attr *attr __rte_unused,
3076 const struct rte_flow_item items[] __rte_unused,
3077 const struct rte_flow_action actions[] __rte_unused,
3078 struct rte_flow_error *error)
3080 return rte_flow_error_set(error, ENOTSUP,
3081 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3085 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3086 struct rte_flow *flow __rte_unused,
3087 struct rte_flow_error *error)
3089 return rte_flow_error_set(error, ENOTSUP,
3090 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3094 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3095 struct rte_flow *flow __rte_unused)
3100 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3101 struct rte_flow *flow __rte_unused)
3106 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3107 struct rte_flow *flow __rte_unused,
3108 const struct rte_flow_action *actions __rte_unused,
3109 void *data __rte_unused,
3110 struct rte_flow_error *error)
3112 return rte_flow_error_set(error, ENOTSUP,
3113 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3117 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3118 uint32_t domains __rte_unused,
3119 uint32_t flags __rte_unused)
3124 /* Void driver to protect from null pointer reference. */
3125 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3126 .validate = flow_null_validate,
3127 .prepare = flow_null_prepare,
3128 .translate = flow_null_translate,
3129 .apply = flow_null_apply,
3130 .remove = flow_null_remove,
3131 .destroy = flow_null_destroy,
3132 .query = flow_null_query,
3133 .sync_domain = flow_null_sync_domain,
3137 * Select flow driver type according to flow attributes and device
3141 * Pointer to the dev structure.
3143 * Pointer to the flow attributes.
3146 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3148 static enum mlx5_flow_drv_type
3149 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3151 struct mlx5_priv *priv = dev->data->dev_private;
3152 /* The OS can determine first a specific flow type (DV, VERBS) */
3153 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3155 if (type != MLX5_FLOW_TYPE_MAX)
3157 /* If no OS specific type - continue with DV/VERBS selection */
3158 if (attr->transfer && priv->config.dv_esw_en)
3159 type = MLX5_FLOW_TYPE_DV;
3160 if (!attr->transfer)
3161 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3162 MLX5_FLOW_TYPE_VERBS;
3166 #define flow_get_drv_ops(type) flow_drv_ops[type]
3169 * Flow driver validation API. This abstracts calling driver specific functions.
3170 * The type of flow driver is determined according to flow attributes.
3173 * Pointer to the dev structure.
3175 * Pointer to the flow attributes.
3177 * Pointer to the list of items.
3178 * @param[in] actions
3179 * Pointer to the list of actions.
3180 * @param[in] external
3181 * This flow rule is created by request external to PMD.
3182 * @param[in] hairpin
3183 * Number of hairpin TX actions, 0 means classic flow.
3185 * Pointer to the error structure.
3188 * 0 on success, a negative errno value otherwise and rte_errno is set.
3191 flow_drv_validate(struct rte_eth_dev *dev,
3192 const struct rte_flow_attr *attr,
3193 const struct rte_flow_item items[],
3194 const struct rte_flow_action actions[],
3195 bool external, int hairpin, struct rte_flow_error *error)
3197 const struct mlx5_flow_driver_ops *fops;
3198 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3200 fops = flow_get_drv_ops(type);
3201 return fops->validate(dev, attr, items, actions, external,
3206 * Flow driver preparation API. This abstracts calling driver specific
3207 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3208 * calculates the size of memory required for device flow, allocates the memory,
3209 * initializes the device flow and returns the pointer.
3212 * This function initializes device flow structure such as dv or verbs in
3213 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3214 * rest. For example, adding returning device flow to flow->dev_flow list and
3215 * setting backward reference to the flow should be done out of this function.
3216 * layers field is not filled either.
3219 * Pointer to the dev structure.
3221 * Pointer to the flow attributes.
3223 * Pointer to the list of items.
3224 * @param[in] actions
3225 * Pointer to the list of actions.
3226 * @param[in] flow_idx
3227 * This memory pool index to the flow.
3229 * Pointer to the error structure.
3232 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3234 static inline struct mlx5_flow *
3235 flow_drv_prepare(struct rte_eth_dev *dev,
3236 const struct rte_flow *flow,
3237 const struct rte_flow_attr *attr,
3238 const struct rte_flow_item items[],
3239 const struct rte_flow_action actions[],
3241 struct rte_flow_error *error)
3243 const struct mlx5_flow_driver_ops *fops;
3244 enum mlx5_flow_drv_type type = flow->drv_type;
3245 struct mlx5_flow *mlx5_flow = NULL;
3247 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3248 fops = flow_get_drv_ops(type);
3249 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3251 mlx5_flow->flow_idx = flow_idx;
3256 * Flow driver translation API. This abstracts calling driver specific
3257 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3258 * translates a generic flow into a driver flow. flow_drv_prepare() must
3262 * dev_flow->layers could be filled as a result of parsing during translation
3263 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3264 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3265 * flow->actions could be overwritten even though all the expanded dev_flows
3266 * have the same actions.
3269 * Pointer to the rte dev structure.
3270 * @param[in, out] dev_flow
3271 * Pointer to the mlx5 flow.
3273 * Pointer to the flow attributes.
3275 * Pointer to the list of items.
3276 * @param[in] actions
3277 * Pointer to the list of actions.
3279 * Pointer to the error structure.
3282 * 0 on success, a negative errno value otherwise and rte_errno is set.
3285 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3286 const struct rte_flow_attr *attr,
3287 const struct rte_flow_item items[],
3288 const struct rte_flow_action actions[],
3289 struct rte_flow_error *error)
3291 const struct mlx5_flow_driver_ops *fops;
3292 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3294 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3295 fops = flow_get_drv_ops(type);
3296 return fops->translate(dev, dev_flow, attr, items, actions, error);
3300 * Flow driver apply API. This abstracts calling driver specific functions.
3301 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3302 * translated driver flows on to device. flow_drv_translate() must precede.
3305 * Pointer to Ethernet device structure.
3306 * @param[in, out] flow
3307 * Pointer to flow structure.
3309 * Pointer to error structure.
3312 * 0 on success, a negative errno value otherwise and rte_errno is set.
3315 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3316 struct rte_flow_error *error)
3318 const struct mlx5_flow_driver_ops *fops;
3319 enum mlx5_flow_drv_type type = flow->drv_type;
3321 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3322 fops = flow_get_drv_ops(type);
3323 return fops->apply(dev, flow, error);
3327 * Flow driver destroy API. This abstracts calling driver specific functions.
3328 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3329 * on device and releases resources of the flow.
3332 * Pointer to Ethernet device.
3333 * @param[in, out] flow
3334 * Pointer to flow structure.
3337 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3339 const struct mlx5_flow_driver_ops *fops;
3340 enum mlx5_flow_drv_type type = flow->drv_type;
3342 flow_mreg_split_qrss_release(dev, flow);
3343 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3344 fops = flow_get_drv_ops(type);
3345 fops->destroy(dev, flow);
3349 * Flow driver find RSS policy tbl API. This abstracts calling driver
3350 * specific functions. Parent flow (rte_flow) should have driver
3351 * type (drv_type). It will find the RSS policy table that has the rss_desc.
3354 * Pointer to Ethernet device.
3355 * @param[in, out] flow
3356 * Pointer to flow structure.
3358 * Pointer to meter policy table.
3359 * @param[in] rss_desc
3360 * Pointer to rss_desc
3362 static struct mlx5_flow_meter_sub_policy *
3363 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3364 struct rte_flow *flow,
3365 struct mlx5_flow_meter_policy *policy,
3366 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3368 const struct mlx5_flow_driver_ops *fops;
3369 enum mlx5_flow_drv_type type = flow->drv_type;
3371 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3372 fops = flow_get_drv_ops(type);
3373 return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3377 * Get RSS action from the action list.
3380 * Pointer to Ethernet device.
3381 * @param[in] actions
3382 * Pointer to the list of actions.
3384 * Parent flow structure pointer.
3387 * Pointer to the RSS action if exist, else return NULL.
3389 static const struct rte_flow_action_rss*
3390 flow_get_rss_action(struct rte_eth_dev *dev,
3391 const struct rte_flow_action actions[])
3393 struct mlx5_priv *priv = dev->data->dev_private;
3394 const struct rte_flow_action_rss *rss = NULL;
3396 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3397 switch (actions->type) {
3398 case RTE_FLOW_ACTION_TYPE_RSS:
3399 rss = actions->conf;
3401 case RTE_FLOW_ACTION_TYPE_SAMPLE:
3403 const struct rte_flow_action_sample *sample =
3405 const struct rte_flow_action *act = sample->actions;
3406 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3407 if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3411 case RTE_FLOW_ACTION_TYPE_METER:
3414 struct mlx5_flow_meter_info *fm;
3415 struct mlx5_flow_meter_policy *policy;
3416 const struct rte_flow_action_meter *mtr = actions->conf;
3418 fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3420 policy = mlx5_flow_meter_policy_find(dev,
3421 fm->policy_id, NULL);
3422 if (policy && policy->is_rss)
3424 policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3436 * Get ASO age action by index.
3439 * Pointer to the Ethernet device structure.
3440 * @param[in] age_idx
3441 * Index to the ASO age action.
3444 * The specified ASO age action.
3446 struct mlx5_aso_age_action*
3447 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3449 uint16_t pool_idx = age_idx & UINT16_MAX;
3450 uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3451 struct mlx5_priv *priv = dev->data->dev_private;
3452 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3453 struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3455 return &pool->actions[offset - 1];
3458 /* maps indirect action to translated direct in some actions array */
3459 struct mlx5_translated_action_handle {
3460 struct rte_flow_action_handle *action; /**< Indirect action handle. */
3461 int index; /**< Index in related array of rte_flow_action. */
3465 * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3466 * direct action if translation possible.
3467 * This functionality used to run same execution path for both direct and
3468 * indirect actions on flow create. All necessary preparations for indirect
3469 * action handling should be performed on *handle* actions list returned
3473 * Pointer to Ethernet device.
3474 * @param[in] actions
3475 * List of actions to translate.
3476 * @param[out] handle
3477 * List to store translated indirect action object handles.
3478 * @param[in, out] indir_n
3479 * Size of *handle* array. On return should be updated with number of
3480 * indirect actions retrieved from the *actions* list.
3481 * @param[out] translated_actions
3482 * List of actions where all indirect actions were translated to direct
3483 * if possible. NULL if no translation took place.
3485 * Pointer to the error structure.
3488 * 0 on success, a negative errno value otherwise and rte_errno is set.
3491 flow_action_handles_translate(struct rte_eth_dev *dev,
3492 const struct rte_flow_action actions[],
3493 struct mlx5_translated_action_handle *handle,
3495 struct rte_flow_action **translated_actions,
3496 struct rte_flow_error *error)
3498 struct mlx5_priv *priv = dev->data->dev_private;
3499 struct rte_flow_action *translated = NULL;
3500 size_t actions_size;
3503 struct mlx5_translated_action_handle *handle_end = NULL;
3505 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3506 if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3508 if (copied_n == *indir_n) {
3509 return rte_flow_error_set
3510 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3511 NULL, "too many shared actions");
3513 rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3514 sizeof(actions[n].conf));
3515 handle[copied_n].index = n;
3519 *indir_n = copied_n;
3522 actions_size = sizeof(struct rte_flow_action) * n;
3523 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3528 memcpy(translated, actions, actions_size);
3529 for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3530 struct mlx5_shared_action_rss *shared_rss;
3531 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3532 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3533 uint32_t idx = act_idx &
3534 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3537 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3538 shared_rss = mlx5_ipool_get
3539 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3540 translated[handle->index].type =
3541 RTE_FLOW_ACTION_TYPE_RSS;
3542 translated[handle->index].conf =
3543 &shared_rss->origin;
3545 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3546 translated[handle->index].type =
3547 (enum rte_flow_action_type)
3548 MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3549 translated[handle->index].conf = (void *)(uintptr_t)idx;
3551 case MLX5_INDIRECT_ACTION_TYPE_AGE:
3552 if (priv->sh->flow_hit_aso_en) {
3553 translated[handle->index].type =
3554 (enum rte_flow_action_type)
3555 MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3556 translated[handle->index].conf =
3557 (void *)(uintptr_t)idx;
3561 case MLX5_INDIRECT_ACTION_TYPE_CT:
3562 if (priv->sh->ct_aso_en) {
3563 translated[handle->index].type =
3564 RTE_FLOW_ACTION_TYPE_CONNTRACK;
3565 translated[handle->index].conf =
3566 (void *)(uintptr_t)idx;
3571 mlx5_free(translated);
3572 return rte_flow_error_set
3573 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3574 NULL, "invalid indirect action type");
3577 *translated_actions = translated;
3582 * Get Shared RSS action from the action list.
3585 * Pointer to Ethernet device.
3587 * Pointer to the list of actions.
3588 * @param[in] shared_n
3589 * Actions list length.
3592 * The MLX5 RSS action ID if exists, otherwise return 0.
3595 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3596 struct mlx5_translated_action_handle *handle,
3599 struct mlx5_translated_action_handle *handle_end;
3600 struct mlx5_priv *priv = dev->data->dev_private;
3601 struct mlx5_shared_action_rss *shared_rss;
3604 for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3605 uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3606 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3607 uint32_t idx = act_idx &
3608 ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3610 case MLX5_INDIRECT_ACTION_TYPE_RSS:
3611 shared_rss = mlx5_ipool_get
3612 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3614 __atomic_add_fetch(&shared_rss->refcnt, 1,
3625 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3627 const struct rte_flow_item *item;
3628 unsigned int has_vlan = 0;
3630 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3631 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3637 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3638 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3639 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3640 MLX5_EXPANSION_ROOT_OUTER;
3644 * Get layer flags from the prefix flow.
3646 * Some flows may be split to several subflows, the prefix subflow gets the
3647 * match items and the suffix sub flow gets the actions.
3648 * Some actions need the user defined match item flags to get the detail for
3650 * This function helps the suffix flow to get the item layer flags from prefix
3653 * @param[in] dev_flow
3654 * Pointer the created preifx subflow.
3657 * The layers get from prefix subflow.
3659 static inline uint64_t
3660 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3662 uint64_t layers = 0;
3665 * Layers bits could be localization, but usually the compiler will
3666 * help to do the optimization work for source code.
3667 * If no decap actions, use the layers directly.
3669 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3670 return dev_flow->handle->layers;
3671 /* Convert L3 layers with decap action. */
3672 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3673 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3674 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3675 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3676 /* Convert L4 layers with decap action. */
3677 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3678 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3679 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3680 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3685 * Get metadata split action information.
3687 * @param[in] actions
3688 * Pointer to the list of actions.
3690 * Pointer to the return pointer.
3691 * @param[out] qrss_type
3692 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3693 * if no QUEUE/RSS is found.
3694 * @param[out] encap_idx
3695 * Pointer to the index of the encap action if exists, otherwise the last
3699 * Total number of actions.
3702 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3703 const struct rte_flow_action **qrss,
3706 const struct rte_flow_action_raw_encap *raw_encap;
3708 int raw_decap_idx = -1;
3711 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3712 switch (actions->type) {
3713 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3714 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3715 *encap_idx = actions_n;
3717 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3718 raw_decap_idx = actions_n;
3720 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3721 raw_encap = actions->conf;
3722 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3723 *encap_idx = raw_decap_idx != -1 ?
3724 raw_decap_idx : actions_n;
3726 case RTE_FLOW_ACTION_TYPE_QUEUE:
3727 case RTE_FLOW_ACTION_TYPE_RSS:
3735 if (*encap_idx == -1)
3736 *encap_idx = actions_n;
3737 /* Count RTE_FLOW_ACTION_TYPE_END. */
3738 return actions_n + 1;
3742 * Check if the action will change packet.
3745 * Pointer to Ethernet device.
3750 * true if action will change packet, false otherwise.
3752 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3753 enum rte_flow_action_type type)
3755 struct mlx5_priv *priv = dev->data->dev_private;
3758 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3759 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3760 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3761 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3762 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3763 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3764 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3765 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3766 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3767 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3768 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3769 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3770 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3771 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3772 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3773 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3774 case RTE_FLOW_ACTION_TYPE_SET_META:
3775 case RTE_FLOW_ACTION_TYPE_SET_TAG:
3776 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3777 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3778 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3779 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3780 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3781 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3782 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3783 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3784 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3785 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3786 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3788 case RTE_FLOW_ACTION_TYPE_FLAG:
3789 case RTE_FLOW_ACTION_TYPE_MARK:
3790 if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3800 * Check meter action from the action list.
3803 * Pointer to Ethernet device.
3804 * @param[in] actions
3805 * Pointer to the list of actions.
3806 * @param[out] has_mtr
3807 * Pointer to the meter exist flag.
3808 * @param[out] has_modify
3809 * Pointer to the flag showing there's packet change action.
3810 * @param[out] meter_id
3811 * Pointer to the meter id.
3814 * Total number of actions.
3817 flow_check_meter_action(struct rte_eth_dev *dev,
3818 const struct rte_flow_action actions[],
3819 bool *has_mtr, bool *has_modify, uint32_t *meter_id)
3821 const struct rte_flow_action_meter *mtr = NULL;
3824 MLX5_ASSERT(has_mtr);
3826 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3827 switch (actions->type) {
3828 case RTE_FLOW_ACTION_TYPE_METER:
3829 mtr = actions->conf;
3830 *meter_id = mtr->mtr_id;
3837 *has_modify |= flow_check_modify_action_type(dev,
3841 /* Count RTE_FLOW_ACTION_TYPE_END. */
3842 return actions_n + 1;
3846 * Check if the flow should be split due to hairpin.
3847 * The reason for the split is that in current HW we can't
3848 * support encap and push-vlan on Rx, so if a flow contains
3849 * these actions we move it to Tx.
3852 * Pointer to Ethernet device.
3854 * Flow rule attributes.
3855 * @param[in] actions
3856 * Associated actions (list terminated by the END action).
3859 * > 0 the number of actions and the flow should be split,
3860 * 0 when no split required.
3863 flow_check_hairpin_split(struct rte_eth_dev *dev,
3864 const struct rte_flow_attr *attr,
3865 const struct rte_flow_action actions[])
3867 int queue_action = 0;
3870 const struct rte_flow_action_queue *queue;
3871 const struct rte_flow_action_rss *rss;
3872 const struct rte_flow_action_raw_encap *raw_encap;
3873 const struct rte_eth_hairpin_conf *conf;
3877 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3878 switch (actions->type) {
3879 case RTE_FLOW_ACTION_TYPE_QUEUE:
3880 queue = actions->conf;
3883 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3884 if (conf == NULL || conf->tx_explicit != 0)
3889 case RTE_FLOW_ACTION_TYPE_RSS:
3890 rss = actions->conf;
3891 if (rss == NULL || rss->queue_num == 0)
3893 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3894 if (conf == NULL || conf->tx_explicit != 0)
3899 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3900 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3901 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3902 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3903 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3907 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3908 raw_encap = actions->conf;
3909 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3918 if (split && queue_action)
3923 /* Declare flow create/destroy prototype in advance. */
3925 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3926 const struct rte_flow_attr *attr,
3927 const struct rte_flow_item items[],
3928 const struct rte_flow_action actions[],
3929 bool external, struct rte_flow_error *error);
3932 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3936 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3937 struct mlx5_hlist_entry *entry,
3938 uint64_t key, void *cb_ctx __rte_unused)
3940 struct mlx5_flow_mreg_copy_resource *mcp_res =
3941 container_of(entry, typeof(*mcp_res), hlist_ent);
3943 return mcp_res->mark_id != key;
3946 struct mlx5_hlist_entry *
3947 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3950 struct rte_eth_dev *dev = list->ctx;
3951 struct mlx5_priv *priv = dev->data->dev_private;
3952 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3953 struct mlx5_flow_mreg_copy_resource *mcp_res;
3954 struct rte_flow_error *error = ctx->error;
3957 uint32_t mark_id = key;
3958 struct rte_flow_attr attr = {
3959 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3962 struct mlx5_rte_flow_item_tag tag_spec = {
3965 struct rte_flow_item items[] = {
3966 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3968 struct rte_flow_action_mark ftag = {
3971 struct mlx5_flow_action_copy_mreg cp_mreg = {
3975 struct rte_flow_action_jump jump = {
3976 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3978 struct rte_flow_action actions[] = {
3979 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3982 /* Fill the register fileds in the flow. */
3983 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3987 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3991 /* Provide the full width of FLAG specific value. */
3992 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3993 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3994 /* Build a new flow. */
3995 if (mark_id != MLX5_DEFAULT_COPY_ID) {
3996 items[0] = (struct rte_flow_item){
3997 .type = (enum rte_flow_item_type)
3998 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4001 items[1] = (struct rte_flow_item){
4002 .type = RTE_FLOW_ITEM_TYPE_END,
4004 actions[0] = (struct rte_flow_action){
4005 .type = (enum rte_flow_action_type)
4006 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4009 actions[1] = (struct rte_flow_action){
4010 .type = (enum rte_flow_action_type)
4011 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4014 actions[2] = (struct rte_flow_action){
4015 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4018 actions[3] = (struct rte_flow_action){
4019 .type = RTE_FLOW_ACTION_TYPE_END,
4022 /* Default rule, wildcard match. */
4023 attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4024 items[0] = (struct rte_flow_item){
4025 .type = RTE_FLOW_ITEM_TYPE_END,
4027 actions[0] = (struct rte_flow_action){
4028 .type = (enum rte_flow_action_type)
4029 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4032 actions[1] = (struct rte_flow_action){
4033 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4036 actions[2] = (struct rte_flow_action){
4037 .type = RTE_FLOW_ACTION_TYPE_END,
4040 /* Build a new entry. */
4041 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4047 mcp_res->mark_id = mark_id;
4049 * The copy Flows are not included in any list. There
4050 * ones are referenced from other Flows and can not
4051 * be applied, removed, deleted in ardbitrary order
4052 * by list traversing.
4054 mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
4055 actions, false, error);
4056 if (!mcp_res->rix_flow) {
4057 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4060 return &mcp_res->hlist_ent;
4064 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4066 * As mark_id is unique, if there's already a registered flow for the mark_id,
4067 * return by increasing the reference counter of the resource. Otherwise, create
4068 * the resource (mcp_res) and flow.
4071 * - If ingress port is ANY and reg_c[1] is mark_id,
4072 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4074 * For default flow (zero mark_id), flow is like,
4075 * - If ingress port is ANY,
4076 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
4079 * Pointer to Ethernet device.
4081 * ID of MARK action, zero means default flow for META.
4083 * Perform verbose error reporting if not NULL.
4086 * Associated resource on success, NULL otherwise and rte_errno is set.
4088 static struct mlx5_flow_mreg_copy_resource *
4089 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4090 struct rte_flow_error *error)
4092 struct mlx5_priv *priv = dev->data->dev_private;
4093 struct mlx5_hlist_entry *entry;
4094 struct mlx5_flow_cb_ctx ctx = {
4099 /* Check if already registered. */
4100 MLX5_ASSERT(priv->mreg_cp_tbl);
4101 entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4104 return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4109 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
4111 struct mlx5_flow_mreg_copy_resource *mcp_res =
4112 container_of(entry, typeof(*mcp_res), hlist_ent);
4113 struct rte_eth_dev *dev = list->ctx;
4114 struct mlx5_priv *priv = dev->data->dev_private;
4116 MLX5_ASSERT(mcp_res->rix_flow);
4117 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
4118 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4122 * Release flow in RX_CP_TBL.
4125 * Pointer to Ethernet device.
4127 * Parent flow for wich copying is provided.
4130 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4131 struct rte_flow *flow)
4133 struct mlx5_flow_mreg_copy_resource *mcp_res;
4134 struct mlx5_priv *priv = dev->data->dev_private;
4136 if (!flow->rix_mreg_copy)
4138 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4139 flow->rix_mreg_copy);
4140 if (!mcp_res || !priv->mreg_cp_tbl)
4142 MLX5_ASSERT(mcp_res->rix_flow);
4143 mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4144 flow->rix_mreg_copy = 0;
4148 * Remove the default copy action from RX_CP_TBL.
4150 * This functions is called in the mlx5_dev_start(). No thread safe
4154 * Pointer to Ethernet device.
4157 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4159 struct mlx5_hlist_entry *entry;
4160 struct mlx5_priv *priv = dev->data->dev_private;
4162 /* Check if default flow is registered. */
4163 if (!priv->mreg_cp_tbl)
4165 entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4166 MLX5_DEFAULT_COPY_ID, NULL);
4169 mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4173 * Add the default copy action in in RX_CP_TBL.
4175 * This functions is called in the mlx5_dev_start(). No thread safe
4179 * Pointer to Ethernet device.
4181 * Perform verbose error reporting if not NULL.
4184 * 0 for success, negative value otherwise and rte_errno is set.
4187 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4188 struct rte_flow_error *error)
4190 struct mlx5_priv *priv = dev->data->dev_private;
4191 struct mlx5_flow_mreg_copy_resource *mcp_res;
4193 /* Check whether extensive metadata feature is engaged. */
4194 if (!priv->config.dv_flow_en ||
4195 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4196 !mlx5_flow_ext_mreg_supported(dev) ||
4197 !priv->sh->dv_regc0_mask)
4200 * Add default mreg copy flow may be called multiple time, but
4201 * only be called once in stop. Avoid register it twice.
4203 if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4205 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4212 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4214 * All the flow having Q/RSS action should be split by
4215 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4216 * performs the following,
4217 * - CQE->flow_tag := reg_c[1] (MARK)
4218 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4219 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4220 * but there should be a flow per each MARK ID set by MARK action.
4222 * For the aforementioned reason, if there's a MARK action in flow's action
4223 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4224 * the MARK ID to CQE's flow_tag like,
4225 * - If reg_c[1] is mark_id,
4226 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4228 * For SET_META action which stores value in reg_c[0], as the destination is
4229 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4230 * MARK ID means the default flow. The default flow looks like,
4231 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4234 * Pointer to Ethernet device.
4236 * Pointer to flow structure.
4237 * @param[in] actions
4238 * Pointer to the list of actions.
4240 * Perform verbose error reporting if not NULL.
4243 * 0 on success, negative value otherwise and rte_errno is set.
4246 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4247 struct rte_flow *flow,
4248 const struct rte_flow_action *actions,
4249 struct rte_flow_error *error)
4251 struct mlx5_priv *priv = dev->data->dev_private;
4252 struct mlx5_dev_config *config = &priv->config;
4253 struct mlx5_flow_mreg_copy_resource *mcp_res;
4254 const struct rte_flow_action_mark *mark;
4256 /* Check whether extensive metadata feature is engaged. */
4257 if (!config->dv_flow_en ||
4258 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4259 !mlx5_flow_ext_mreg_supported(dev) ||
4260 !priv->sh->dv_regc0_mask)
4262 /* Find MARK action. */
4263 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4264 switch (actions->type) {
4265 case RTE_FLOW_ACTION_TYPE_FLAG:
4266 mcp_res = flow_mreg_add_copy_action
4267 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4270 flow->rix_mreg_copy = mcp_res->idx;
4272 case RTE_FLOW_ACTION_TYPE_MARK:
4273 mark = (const struct rte_flow_action_mark *)
4276 flow_mreg_add_copy_action(dev, mark->id, error);
4279 flow->rix_mreg_copy = mcp_res->idx;
4288 #define MLX5_MAX_SPLIT_ACTIONS 24
4289 #define MLX5_MAX_SPLIT_ITEMS 24
4292 * Split the hairpin flow.
4293 * Since HW can't support encap and push-vlan on Rx, we move these
4295 * If the count action is after the encap then we also
4296 * move the count action. in this case the count will also measure
4300 * Pointer to Ethernet device.
4301 * @param[in] actions
4302 * Associated actions (list terminated by the END action).
4303 * @param[out] actions_rx
4305 * @param[out] actions_tx
4307 * @param[out] pattern_tx
4308 * The pattern items for the Tx flow.
4309 * @param[out] flow_id
4310 * The flow ID connected to this flow.
4316 flow_hairpin_split(struct rte_eth_dev *dev,
4317 const struct rte_flow_action actions[],
4318 struct rte_flow_action actions_rx[],
4319 struct rte_flow_action actions_tx[],
4320 struct rte_flow_item pattern_tx[],
4323 const struct rte_flow_action_raw_encap *raw_encap;
4324 const struct rte_flow_action_raw_decap *raw_decap;
4325 struct mlx5_rte_flow_action_set_tag *set_tag;
4326 struct rte_flow_action *tag_action;
4327 struct mlx5_rte_flow_item_tag *tag_item;
4328 struct rte_flow_item *item;
4332 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4333 switch (actions->type) {
4334 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4335 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4336 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4337 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4338 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4339 rte_memcpy(actions_tx, actions,
4340 sizeof(struct rte_flow_action));
4343 case RTE_FLOW_ACTION_TYPE_COUNT:
4345 rte_memcpy(actions_tx, actions,
4346 sizeof(struct rte_flow_action));
4349 rte_memcpy(actions_rx, actions,
4350 sizeof(struct rte_flow_action));
4354 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4355 raw_encap = actions->conf;
4356 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4357 memcpy(actions_tx, actions,
4358 sizeof(struct rte_flow_action));
4362 rte_memcpy(actions_rx, actions,
4363 sizeof(struct rte_flow_action));
4367 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4368 raw_decap = actions->conf;
4369 if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4370 memcpy(actions_tx, actions,
4371 sizeof(struct rte_flow_action));
4374 rte_memcpy(actions_rx, actions,
4375 sizeof(struct rte_flow_action));
4380 rte_memcpy(actions_rx, actions,
4381 sizeof(struct rte_flow_action));
4386 /* Add set meta action and end action for the Rx flow. */
4387 tag_action = actions_rx;
4388 tag_action->type = (enum rte_flow_action_type)
4389 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4391 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4393 set_tag = (void *)actions_rx;
4394 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4395 .id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4398 MLX5_ASSERT(set_tag->id > REG_NON);
4399 tag_action->conf = set_tag;
4400 /* Create Tx item list. */
4401 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4402 addr = (void *)&pattern_tx[2];
4404 item->type = (enum rte_flow_item_type)
4405 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4406 tag_item = (void *)addr;
4407 tag_item->data = flow_id;
4408 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4409 MLX5_ASSERT(set_tag->id > REG_NON);
4410 item->spec = tag_item;
4411 addr += sizeof(struct mlx5_rte_flow_item_tag);
4412 tag_item = (void *)addr;
4413 tag_item->data = UINT32_MAX;
4414 tag_item->id = UINT16_MAX;
4415 item->mask = tag_item;
4418 item->type = RTE_FLOW_ITEM_TYPE_END;
4423 * The last stage of splitting chain, just creates the subflow
4424 * without any modification.
4427 * Pointer to Ethernet device.
4429 * Parent flow structure pointer.
4430 * @param[in, out] sub_flow
4431 * Pointer to return the created subflow, may be NULL.
4433 * Flow rule attributes.
4435 * Pattern specification (list terminated by the END pattern item).
4436 * @param[in] actions
4437 * Associated actions (list terminated by the END action).
4438 * @param[in] flow_split_info
4439 * Pointer to flow split info structure.
4441 * Perform verbose error reporting if not NULL.
4443 * 0 on success, negative value otherwise
4446 flow_create_split_inner(struct rte_eth_dev *dev,
4447 struct rte_flow *flow,
4448 struct mlx5_flow **sub_flow,
4449 const struct rte_flow_attr *attr,
4450 const struct rte_flow_item items[],
4451 const struct rte_flow_action actions[],
4452 struct mlx5_flow_split_info *flow_split_info,
4453 struct rte_flow_error *error)
4455 struct mlx5_flow *dev_flow;
4457 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4458 flow_split_info->flow_idx, error);
4461 dev_flow->flow = flow;
4462 dev_flow->external = flow_split_info->external;
4463 dev_flow->skip_scale = flow_split_info->skip_scale;
4464 /* Subflow object was created, we must include one in the list. */
4465 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4466 dev_flow->handle, next);
4468 * If dev_flow is as one of the suffix flow, some actions in suffix
4469 * flow may need some user defined item layer flags, and pass the
4470 * Metadate rxq mark flag to suffix flow as well.
4472 if (flow_split_info->prefix_layers)
4473 dev_flow->handle->layers = flow_split_info->prefix_layers;
4474 if (flow_split_info->prefix_mark)
4475 dev_flow->handle->mark = 1;
4477 *sub_flow = dev_flow;
4478 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4479 dev_flow->dv.table_id = flow_split_info->table_id;
4481 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4485 * Get the sub policy of a meter.
4488 * Pointer to Ethernet device.
4490 * Parent flow structure pointer.
4491 * @param[in] policy_id;
4494 * Flow rule attributes.
4496 * Pattern specification (list terminated by the END pattern item).
4498 * Perform verbose error reporting if not NULL.
4501 * Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4503 static struct mlx5_flow_meter_sub_policy *
4504 get_meter_sub_policy(struct rte_eth_dev *dev,
4505 struct rte_flow *flow,
4507 const struct rte_flow_attr *attr,
4508 const struct rte_flow_item items[],
4509 struct rte_flow_error *error)
4511 struct mlx5_flow_meter_policy *policy;
4512 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4514 policy = mlx5_flow_meter_policy_find(dev, policy_id, NULL);
4516 rte_flow_error_set(error, EINVAL,
4517 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4518 "Failed to find Meter Policy.");
4521 if (policy->is_rss) {
4522 struct mlx5_flow_workspace *wks =
4523 mlx5_flow_get_thread_workspace();
4524 struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4525 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4530 * This is a tmp dev_flow,
4531 * no need to register any matcher for it in translate.
4533 wks->skip_matcher_reg = 1;
4534 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4535 struct mlx5_flow dev_flow = {0};
4536 struct mlx5_flow_handle dev_handle = { {0} };
4537 const void *rss_act = policy->act_cnt[i].rss->conf;
4538 struct rte_flow_action rss_actions[2] = {
4540 .type = RTE_FLOW_ACTION_TYPE_RSS,
4544 .type = RTE_FLOW_ACTION_TYPE_END,
4549 dev_flow.handle = &dev_handle;
4550 dev_flow.ingress = attr->ingress;
4551 dev_flow.flow = flow;
4552 dev_flow.external = 0;
4553 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4554 dev_flow.dv.transfer = attr->transfer;
4556 /* Translate RSS action to get rss hash fields. */
4557 if (flow_drv_translate(dev, &dev_flow, attr,
4558 items, rss_actions, error))
4560 rss_desc_v[i] = wks->rss_desc;
4561 rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4562 rss_desc_v[i].hash_fields = dev_flow.hash_fields;
4563 rss_desc_v[i].queue_num = rss_desc_v[i].hash_fields ?
4564 rss_desc_v[i].queue_num : 1;
4565 rss_desc[i] = &rss_desc_v[i];
4567 sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4568 flow, policy, rss_desc);
4570 enum mlx5_meter_domain mtr_domain =
4571 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4572 attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4573 MLX5_MTR_DOMAIN_INGRESS;
4574 sub_policy = policy->sub_policys[mtr_domain][0];
4577 rte_flow_error_set(error, EINVAL,
4578 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4579 "Failed to get meter sub-policy.");
4587 * Split the meter flow.
4589 * As meter flow will split to three sub flow, other than meter
4590 * action, the other actions make sense to only meter accepts
4591 * the packet. If it need to be dropped, no other additional
4592 * actions should be take.
4594 * One kind of special action which decapsulates the L3 tunnel
4595 * header will be in the prefix sub flow, as not to take the
4596 * L3 tunnel header into account.
4599 * Pointer to Ethernet device.
4601 * Parent flow structure pointer.
4603 * Pointer to flow meter structure.
4605 * Flow rule attributes.
4607 * Pattern specification (list terminated by the END pattern item).
4608 * @param[out] sfx_items
4609 * Suffix flow match items (list terminated by the END pattern item).
4610 * @param[in] actions
4611 * Associated actions (list terminated by the END action).
4612 * @param[out] actions_sfx
4613 * Suffix flow actions.
4614 * @param[out] actions_pre
4615 * Prefix flow actions.
4616 * @param[out] mtr_flow_id
4617 * Pointer to meter flow id.
4619 * Perform verbose error reporting if not NULL.
4622 * 0 on success, a negative errno value otherwise and rte_errno is set.
4625 flow_meter_split_prep(struct rte_eth_dev *dev,
4626 struct rte_flow *flow,
4627 struct mlx5_flow_meter_info *fm,
4628 const struct rte_flow_attr *attr,
4629 const struct rte_flow_item items[],
4630 struct rte_flow_item sfx_items[],
4631 const struct rte_flow_action actions[],
4632 struct rte_flow_action actions_sfx[],
4633 struct rte_flow_action actions_pre[],
4634 uint32_t *mtr_flow_id,
4635 struct rte_flow_error *error)
4637 struct mlx5_priv *priv = dev->data->dev_private;
4638 struct rte_flow_action *tag_action = NULL;
4639 struct rte_flow_item *tag_item;
4640 struct mlx5_rte_flow_action_set_tag *set_tag;
4641 const struct rte_flow_action_raw_encap *raw_encap;
4642 const struct rte_flow_action_raw_decap *raw_decap;
4643 struct mlx5_rte_flow_item_tag *tag_item_spec;
4644 struct mlx5_rte_flow_item_tag *tag_item_mask;
4645 uint32_t tag_id = 0;
4646 bool copy_vlan = false;
4647 struct rte_flow_action *hw_mtr_action;
4648 struct rte_flow_action *action_pre_head = NULL;
4649 bool mtr_first = priv->sh->meter_aso_en &&
4651 (attr->transfer && priv->representor_id != UINT16_MAX));
4652 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4653 uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4654 MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4655 uint32_t flow_id = 0;
4656 uint32_t flow_id_reversed = 0;
4657 uint8_t flow_id_bits = 0;
4660 /* For ASO meter, meter must be before tag in TX direction. */
4662 action_pre_head = actions_pre++;
4663 /* Leave space for tag action. */
4664 tag_action = actions_pre++;
4666 /* Prepare the actions for prefix and suffix flow. */
4667 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4668 struct rte_flow_action *action_cur = NULL;
4670 switch (actions->type) {
4671 case RTE_FLOW_ACTION_TYPE_METER:
4673 action_cur = action_pre_head;
4675 /* Leave space for tag action. */
4676 tag_action = actions_pre++;
4677 action_cur = actions_pre++;
4680 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4681 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4682 action_cur = actions_pre++;
4684 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4685 raw_encap = actions->conf;
4686 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4687 action_cur = actions_pre++;
4689 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4690 raw_decap = actions->conf;
4691 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4692 action_cur = actions_pre++;
4694 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4695 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4702 action_cur = (fm->def_policy) ?
4703 actions_sfx++ : actions_pre++;
4704 memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4706 /* Add end action to the actions. */
4707 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4708 if (priv->sh->meter_aso_en) {
4710 * For ASO meter, need to add an extra jump action explicitly,
4711 * to jump from meter to policer table.
4713 struct mlx5_flow_meter_sub_policy *sub_policy;
4714 struct mlx5_flow_tbl_data_entry *tbl_data;
4716 if (!fm->def_policy) {
4717 sub_policy = get_meter_sub_policy(dev, flow,
4718 fm->policy_id, attr,
4723 enum mlx5_meter_domain mtr_domain =
4724 attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4725 attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4726 MLX5_MTR_DOMAIN_INGRESS;
4729 &priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
4731 tbl_data = container_of(sub_policy->tbl_rsc,
4732 struct mlx5_flow_tbl_data_entry, tbl);
4733 hw_mtr_action = actions_pre++;
4734 hw_mtr_action->type = (enum rte_flow_action_type)
4735 MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
4736 hw_mtr_action->conf = tbl_data->jump.action;
4738 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4741 return rte_flow_error_set(error, ENOMEM,
4742 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4743 "No tag action space.");
4745 tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
4748 /* Only default-policy Meter creates mtr flow id. */
4749 if (fm->def_policy) {
4750 mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4752 return rte_flow_error_set(error, ENOMEM,
4753 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4754 "Failed to allocate meter flow id.");
4755 flow_id = tag_id - 1;
4756 flow_id_bits = MLX5_REG_BITS - __builtin_clz(flow_id);
4757 flow_id_bits = flow_id_bits ? flow_id_bits : 1;
4758 if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
4760 mlx5_ipool_free(fm->flow_ipool, tag_id);
4761 return rte_flow_error_set(error, EINVAL,
4762 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4763 "Meter flow id exceeds max limit.");
4765 if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
4766 priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
4768 /* Prepare the suffix subflow items. */
4769 tag_item = sfx_items++;
4770 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4771 int item_type = items->type;
4773 switch (item_type) {
4774 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4775 memcpy(sfx_items, items, sizeof(*sfx_items));
4778 case RTE_FLOW_ITEM_TYPE_VLAN:
4780 memcpy(sfx_items, items, sizeof(*sfx_items));
4782 * Convert to internal match item, it is used
4783 * for vlan push and set vid.
4785 sfx_items->type = (enum rte_flow_item_type)
4786 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4794 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4796 /* Build tag actions and items for meter_id/meter flow_id. */
4797 set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
4798 tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4799 tag_item_mask = tag_item_spec + 1;
4800 /* Both flow_id and meter_id share the same register. */
4801 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
4802 .id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
4804 .offset = mtr_id_offset,
4805 .length = mtr_reg_bits,
4806 .data = flow->meter,
4809 * The color Reg bits used by flow_id are growing from
4810 * msb to lsb, so must do bit reverse for flow_id val in RegC.
4812 for (shift = 0; shift < flow_id_bits; shift++)
4813 flow_id_reversed = (flow_id_reversed << 1) |
4814 ((flow_id >> shift) & 0x1);
4816 flow_id_reversed << (mtr_reg_bits - flow_id_bits);
4817 tag_item_spec->id = set_tag->id;
4818 tag_item_spec->data = set_tag->data << mtr_id_offset;
4819 tag_item_mask->data = UINT32_MAX << mtr_id_offset;
4820 tag_action->type = (enum rte_flow_action_type)
4821 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4822 tag_action->conf = set_tag;
4823 tag_item->type = (enum rte_flow_item_type)
4824 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4825 tag_item->spec = tag_item_spec;
4826 tag_item->last = NULL;
4827 tag_item->mask = tag_item_mask;
4830 *mtr_flow_id = tag_id;
4835 * Split action list having QUEUE/RSS for metadata register copy.
4837 * Once Q/RSS action is detected in user's action list, the flow action
4838 * should be split in order to copy metadata registers, which will happen in
4840 * - CQE->flow_tag := reg_c[1] (MARK)
4841 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4842 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4843 * This is because the last action of each flow must be a terminal action
4844 * (QUEUE, RSS or DROP).
4846 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4847 * stored and kept in the mlx5_flow structure per each sub_flow.
4849 * The Q/RSS action is replaced with,
4850 * - SET_TAG, setting the allocated flow ID to reg_c[2].
4851 * And the following JUMP action is added at the end,
4852 * - JUMP, to RX_CP_TBL.
4854 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4855 * flow_create_split_metadata() routine. The flow will look like,
4856 * - If flow ID matches (reg_c[2]), perform Q/RSS.
4859 * Pointer to Ethernet device.
4860 * @param[out] split_actions
4861 * Pointer to store split actions to jump to CP_TBL.
4862 * @param[in] actions
4863 * Pointer to the list of original flow actions.
4865 * Pointer to the Q/RSS action.
4866 * @param[in] actions_n
4867 * Number of original actions.
4869 * Perform verbose error reporting if not NULL.
4872 * non-zero unique flow_id on success, otherwise 0 and
4873 * error/rte_error are set.
4876 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4877 struct rte_flow_action *split_actions,
4878 const struct rte_flow_action *actions,
4879 const struct rte_flow_action *qrss,
4880 int actions_n, struct rte_flow_error *error)
4882 struct mlx5_priv *priv = dev->data->dev_private;
4883 struct mlx5_rte_flow_action_set_tag *set_tag;
4884 struct rte_flow_action_jump *jump;
4885 const int qrss_idx = qrss - actions;
4886 uint32_t flow_id = 0;
4890 * Given actions will be split
4891 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4892 * - Add jump to mreg CP_TBL.
4893 * As a result, there will be one more action.
4896 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4897 set_tag = (void *)(split_actions + actions_n);
4899 * If tag action is not set to void(it means we are not the meter
4900 * suffix flow), add the tag action. Since meter suffix flow already
4901 * has the tag added.
4903 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4905 * Allocate the new subflow ID. This one is unique within
4906 * device and not shared with representors. Otherwise,
4907 * we would have to resolve multi-thread access synch
4908 * issue. Each flow on the shared device is appended
4909 * with source vport identifier, so the resulting
4910 * flows will be unique in the shared (by master and
4911 * representors) domain even if they have coinciding
4914 mlx5_ipool_malloc(priv->sh->ipool
4915 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4917 return rte_flow_error_set(error, ENOMEM,
4918 RTE_FLOW_ERROR_TYPE_ACTION,
4919 NULL, "can't allocate id "
4920 "for split Q/RSS subflow");
4921 /* Internal SET_TAG action to set flow ID. */
4922 *set_tag = (struct mlx5_rte_flow_action_set_tag){
4925 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4929 /* Construct new actions array. */
4930 /* Replace QUEUE/RSS action. */
4931 split_actions[qrss_idx] = (struct rte_flow_action){
4932 .type = (enum rte_flow_action_type)
4933 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4937 /* JUMP action to jump to mreg copy table (CP_TBL). */
4938 jump = (void *)(set_tag + 1);
4939 *jump = (struct rte_flow_action_jump){
4940 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4942 split_actions[actions_n - 2] = (struct rte_flow_action){
4943 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4946 split_actions[actions_n - 1] = (struct rte_flow_action){
4947 .type = RTE_FLOW_ACTION_TYPE_END,
4953 * Extend the given action list for Tx metadata copy.
4955 * Copy the given action list to the ext_actions and add flow metadata register
4956 * copy action in order to copy reg_a set by WQE to reg_c[0].
4958 * @param[out] ext_actions
4959 * Pointer to the extended action list.
4960 * @param[in] actions
4961 * Pointer to the list of actions.
4962 * @param[in] actions_n
4963 * Number of actions in the list.
4965 * Perform verbose error reporting if not NULL.
4966 * @param[in] encap_idx
4967 * The encap action inndex.
4970 * 0 on success, negative value otherwise
4973 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4974 struct rte_flow_action *ext_actions,
4975 const struct rte_flow_action *actions,
4976 int actions_n, struct rte_flow_error *error,
4979 struct mlx5_flow_action_copy_mreg *cp_mreg =
4980 (struct mlx5_flow_action_copy_mreg *)
4981 (ext_actions + actions_n + 1);
4984 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4988 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4993 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4994 if (encap_idx == actions_n - 1) {
4995 ext_actions[actions_n - 1] = (struct rte_flow_action){
4996 .type = (enum rte_flow_action_type)
4997 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5000 ext_actions[actions_n] = (struct rte_flow_action){
5001 .type = RTE_FLOW_ACTION_TYPE_END,
5004 ext_actions[encap_idx] = (struct rte_flow_action){
5005 .type = (enum rte_flow_action_type)
5006 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5009 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5010 sizeof(*ext_actions) * (actions_n - encap_idx));
5016 * Check the match action from the action list.
5018 * @param[in] actions
5019 * Pointer to the list of actions.
5021 * Flow rule attributes.
5023 * The action to be check if exist.
5024 * @param[out] match_action_pos
5025 * Pointer to the position of the matched action if exists, otherwise is -1.
5026 * @param[out] qrss_action_pos
5027 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5028 * @param[out] modify_after_mirror
5029 * Pointer to the flag of modify action after FDB mirroring.
5032 * > 0 the total number of actions.
5033 * 0 if not found match action in action list.
5036 flow_check_match_action(const struct rte_flow_action actions[],
5037 const struct rte_flow_attr *attr,
5038 enum rte_flow_action_type action,
5039 int *match_action_pos, int *qrss_action_pos,
5040 int *modify_after_mirror)
5042 const struct rte_flow_action_sample *sample;
5049 *match_action_pos = -1;
5050 *qrss_action_pos = -1;
5051 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5052 if (actions->type == action) {
5054 *match_action_pos = actions_n;
5056 switch (actions->type) {
5057 case RTE_FLOW_ACTION_TYPE_QUEUE:
5058 case RTE_FLOW_ACTION_TYPE_RSS:
5059 *qrss_action_pos = actions_n;
5061 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5062 sample = actions->conf;
5063 ratio = sample->ratio;
5064 sub_type = ((const struct rte_flow_action *)
5065 (sample->actions))->type;
5066 if (ratio == 1 && attr->transfer)
5069 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5070 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5071 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5072 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5073 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5074 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5075 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5076 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5077 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5078 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5079 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5080 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5081 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5082 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5083 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5084 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5085 case RTE_FLOW_ACTION_TYPE_FLAG:
5086 case RTE_FLOW_ACTION_TYPE_MARK:
5087 case RTE_FLOW_ACTION_TYPE_SET_META:
5088 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5089 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5090 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5091 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5092 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5093 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5094 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5095 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5096 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5098 *modify_after_mirror = 1;
5105 if (flag && fdb_mirror && !*modify_after_mirror) {
5106 /* FDB mirroring uses the destination array to implement
5107 * instead of FLOW_SAMPLER object.
5109 if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5112 /* Count RTE_FLOW_ACTION_TYPE_END. */
5113 return flag ? actions_n + 1 : 0;
5116 #define SAMPLE_SUFFIX_ITEM 2
5119 * Split the sample flow.
5121 * As sample flow will split to two sub flow, sample flow with
5122 * sample action, the other actions will move to new suffix flow.
5124 * Also add unique tag id with tag action in the sample flow,
5125 * the same tag id will be as match in the suffix flow.
5128 * Pointer to Ethernet device.
5129 * @param[in] add_tag
5130 * Add extra tag action flag.
5131 * @param[out] sfx_items
5132 * Suffix flow match items (list terminated by the END pattern item).
5133 * @param[in] actions
5134 * Associated actions (list terminated by the END action).
5135 * @param[out] actions_sfx
5136 * Suffix flow actions.
5137 * @param[out] actions_pre
5138 * Prefix flow actions.
5139 * @param[in] actions_n
5140 * The total number of actions.
5141 * @param[in] sample_action_pos
5142 * The sample action position.
5143 * @param[in] qrss_action_pos
5144 * The Queue/RSS action position.
5145 * @param[in] jump_table
5146 * Add extra jump action flag.
5148 * Perform verbose error reporting if not NULL.
5151 * 0 on success, or unique flow_id, a negative errno value
5152 * otherwise and rte_errno is set.
5155 flow_sample_split_prep(struct rte_eth_dev *dev,
5157 struct rte_flow_item sfx_items[],
5158 const struct rte_flow_action actions[],
5159 struct rte_flow_action actions_sfx[],
5160 struct rte_flow_action actions_pre[],
5162 int sample_action_pos,
5163 int qrss_action_pos,
5165 struct rte_flow_error *error)
5167 struct mlx5_priv *priv = dev->data->dev_private;
5168 struct mlx5_rte_flow_action_set_tag *set_tag;
5169 struct mlx5_rte_flow_item_tag *tag_spec;
5170 struct mlx5_rte_flow_item_tag *tag_mask;
5171 struct rte_flow_action_jump *jump_action;
5172 uint32_t tag_id = 0;
5174 int append_index = 0;
5177 if (sample_action_pos < 0)
5178 return rte_flow_error_set(error, EINVAL,
5179 RTE_FLOW_ERROR_TYPE_ACTION,
5180 NULL, "invalid position of sample "
5182 /* Prepare the actions for prefix and suffix flow. */
5183 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5184 index = qrss_action_pos;
5185 /* Put the preceding the Queue/RSS action into prefix flow. */
5187 memcpy(actions_pre, actions,
5188 sizeof(struct rte_flow_action) * index);
5189 /* Put others preceding the sample action into prefix flow. */
5190 if (sample_action_pos > index + 1)
5191 memcpy(actions_pre + index, actions + index + 1,
5192 sizeof(struct rte_flow_action) *
5193 (sample_action_pos - index - 1));
5194 index = sample_action_pos - 1;
5195 /* Put Queue/RSS action into Suffix flow. */
5196 memcpy(actions_sfx, actions + qrss_action_pos,
5197 sizeof(struct rte_flow_action));
5200 index = sample_action_pos;
5202 memcpy(actions_pre, actions,
5203 sizeof(struct rte_flow_action) * index);
5205 /* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5206 * For CX6DX and above, metadata registers Cx preserve their value,
5207 * add an extra tag action for NIC-RX and E-Switch Domain.
5210 /* Prepare the prefix tag action. */
5212 set_tag = (void *)(actions_pre + actions_n + append_index);
5213 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5216 mlx5_ipool_malloc(priv->sh->ipool
5217 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5218 *set_tag = (struct mlx5_rte_flow_action_set_tag) {
5222 /* Prepare the suffix subflow items. */
5223 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5224 tag_spec->data = tag_id;
5225 tag_spec->id = set_tag->id;
5226 tag_mask = tag_spec + 1;
5227 tag_mask->data = UINT32_MAX;
5228 sfx_items[0] = (struct rte_flow_item){
5229 .type = (enum rte_flow_item_type)
5230 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5235 sfx_items[1] = (struct rte_flow_item){
5236 .type = (enum rte_flow_item_type)
5237 RTE_FLOW_ITEM_TYPE_END,
5239 /* Prepare the tag action in prefix subflow. */
5240 actions_pre[index++] =
5241 (struct rte_flow_action){
5242 .type = (enum rte_flow_action_type)
5243 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5247 memcpy(actions_pre + index, actions + sample_action_pos,
5248 sizeof(struct rte_flow_action));
5250 /* For the modify action after the sample action in E-Switch mirroring,
5251 * Add the extra jump action in prefix subflow and jump into the next
5252 * table, then do the modify action in the new table.
5255 /* Prepare the prefix jump action. */
5257 jump_action = (void *)(actions_pre + actions_n + append_index);
5258 jump_action->group = jump_table;
5259 actions_pre[index++] =
5260 (struct rte_flow_action){
5261 .type = (enum rte_flow_action_type)
5262 RTE_FLOW_ACTION_TYPE_JUMP,
5263 .conf = jump_action,
5266 actions_pre[index] = (struct rte_flow_action){
5267 .type = (enum rte_flow_action_type)
5268 RTE_FLOW_ACTION_TYPE_END,
5270 /* Put the actions after sample into Suffix flow. */
5271 memcpy(actions_sfx, actions + sample_action_pos + 1,
5272 sizeof(struct rte_flow_action) *
5273 (actions_n - sample_action_pos - 1));
5278 * The splitting for metadata feature.
5280 * - Q/RSS action on NIC Rx should be split in order to pass by
5281 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5282 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5284 * - All the actions on NIC Tx should have a mreg copy action to
5285 * copy reg_a from WQE to reg_c[0].
5288 * Pointer to Ethernet device.
5290 * Parent flow structure pointer.
5292 * Flow rule attributes.
5294 * Pattern specification (list terminated by the END pattern item).
5295 * @param[in] actions
5296 * Associated actions (list terminated by the END action).
5297 * @param[in] flow_split_info
5298 * Pointer to flow split info structure.
5300 * Perform verbose error reporting if not NULL.
5302 * 0 on success, negative value otherwise
5305 flow_create_split_metadata(struct rte_eth_dev *dev,
5306 struct rte_flow *flow,
5307 const struct rte_flow_attr *attr,
5308 const struct rte_flow_item items[],
5309 const struct rte_flow_action actions[],
5310 struct mlx5_flow_split_info *flow_split_info,
5311 struct rte_flow_error *error)
5313 struct mlx5_priv *priv = dev->data->dev_private;
5314 struct mlx5_dev_config *config = &priv->config;
5315 const struct rte_flow_action *qrss = NULL;
5316 struct rte_flow_action *ext_actions = NULL;
5317 struct mlx5_flow *dev_flow = NULL;
5318 uint32_t qrss_id = 0;
5325 /* Check whether extensive metadata feature is engaged. */
5326 if (!config->dv_flow_en ||
5327 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5328 !mlx5_flow_ext_mreg_supported(dev))
5329 return flow_create_split_inner(dev, flow, NULL, attr, items,
5330 actions, flow_split_info, error);
5331 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5334 /* Exclude hairpin flows from splitting. */
5335 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5336 const struct rte_flow_action_queue *queue;
5339 if (mlx5_rxq_get_type(dev, queue->index) ==
5340 MLX5_RXQ_TYPE_HAIRPIN)
5342 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5343 const struct rte_flow_action_rss *rss;
5346 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5347 MLX5_RXQ_TYPE_HAIRPIN)
5352 /* Check if it is in meter suffix table. */
5353 mtr_sfx = attr->group == (attr->transfer ?
5354 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5355 MLX5_FLOW_TABLE_LEVEL_METER);
5357 * Q/RSS action on NIC Rx should be split in order to pass by
5358 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5359 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5361 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5362 sizeof(struct rte_flow_action_set_tag) +
5363 sizeof(struct rte_flow_action_jump);
5364 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5367 return rte_flow_error_set(error, ENOMEM,
5368 RTE_FLOW_ERROR_TYPE_ACTION,
5369 NULL, "no memory to split "
5372 * If we are the suffix flow of meter, tag already exist.
5373 * Set the tag action to void.
5376 ext_actions[qrss - actions].type =
5377 RTE_FLOW_ACTION_TYPE_VOID;
5379 ext_actions[qrss - actions].type =
5380 (enum rte_flow_action_type)
5381 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5383 * Create the new actions list with removed Q/RSS action
5384 * and appended set tag and jump to register copy table
5385 * (RX_CP_TBL). We should preallocate unique tag ID here
5386 * in advance, because it is needed for set tag action.
5388 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5389 qrss, actions_n, error);
5390 if (!mtr_sfx && !qrss_id) {
5394 } else if (attr->egress && !attr->transfer) {
5396 * All the actions on NIC Tx should have a metadata register
5397 * copy action to copy reg_a from WQE to reg_c[meta]
5399 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5400 sizeof(struct mlx5_flow_action_copy_mreg);
5401 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5404 return rte_flow_error_set(error, ENOMEM,
5405 RTE_FLOW_ERROR_TYPE_ACTION,
5406 NULL, "no memory to split "
5408 /* Create the action list appended with copy register. */
5409 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5410 actions_n, error, encap_idx);
5414 /* Add the unmodified original or prefix subflow. */
5415 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5416 items, ext_actions ? ext_actions :
5417 actions, flow_split_info, error);
5420 MLX5_ASSERT(dev_flow);
5422 const struct rte_flow_attr q_attr = {
5423 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5426 /* Internal PMD action to set register. */
5427 struct mlx5_rte_flow_item_tag q_tag_spec = {
5431 struct rte_flow_item q_items[] = {
5433 .type = (enum rte_flow_item_type)
5434 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5435 .spec = &q_tag_spec,
5440 .type = RTE_FLOW_ITEM_TYPE_END,
5443 struct rte_flow_action q_actions[] = {
5449 .type = RTE_FLOW_ACTION_TYPE_END,
5452 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5455 * Configure the tag item only if there is no meter subflow.
5456 * Since tag is already marked in the meter suffix subflow
5457 * we can just use the meter suffix items as is.
5460 /* Not meter subflow. */
5461 MLX5_ASSERT(!mtr_sfx);
5463 * Put unique id in prefix flow due to it is destroyed
5464 * after suffix flow and id will be freed after there
5465 * is no actual flows with this id and identifier
5466 * reallocation becomes possible (for example, for
5467 * other flows in other threads).
5469 dev_flow->handle->split_flow_id = qrss_id;
5470 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5474 q_tag_spec.id = ret;
5477 /* Add suffix subflow to execute Q/RSS. */
5478 flow_split_info->prefix_layers = layers;
5479 flow_split_info->prefix_mark = 0;
5480 ret = flow_create_split_inner(dev, flow, &dev_flow,
5481 &q_attr, mtr_sfx ? items :
5483 flow_split_info, error);
5486 /* qrss ID should be freed if failed. */
5488 MLX5_ASSERT(dev_flow);
5493 * We do not destroy the partially created sub_flows in case of error.
5494 * These ones are included into parent flow list and will be destroyed
5495 * by flow_drv_destroy.
5497 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5499 mlx5_free(ext_actions);
5504 * Create meter internal drop flow with the original pattern.
5507 * Pointer to Ethernet device.
5509 * Parent flow structure pointer.
5511 * Flow rule attributes.
5513 * Pattern specification (list terminated by the END pattern item).
5514 * @param[in] flow_split_info
5515 * Pointer to flow split info structure.
5517 * Pointer to flow meter structure.
5519 * Perform verbose error reporting if not NULL.
5521 * 0 on success, negative value otherwise
5524 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5525 struct rte_flow *flow,
5526 const struct rte_flow_attr *attr,
5527 const struct rte_flow_item items[],
5528 struct mlx5_flow_split_info *flow_split_info,
5529 struct mlx5_flow_meter_info *fm,
5530 struct rte_flow_error *error)
5532 struct mlx5_flow *dev_flow = NULL;
5533 struct rte_flow_attr drop_attr = *attr;
5534 struct rte_flow_action drop_actions[3];
5535 struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5537 MLX5_ASSERT(fm->drop_cnt);
5538 drop_actions[0].type =
5539 (enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5540 drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5541 drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5542 drop_actions[1].conf = NULL;
5543 drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5544 drop_actions[2].conf = NULL;
5545 drop_split_info.external = false;
5546 drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5547 drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5548 drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5549 return flow_create_split_inner(dev, flow, &dev_flow,
5550 &drop_attr, items, drop_actions,
5551 &drop_split_info, error);
5555 * The splitting for meter feature.
5557 * - The meter flow will be split to two flows as prefix and
5558 * suffix flow. The packets make sense only it pass the prefix
5561 * - Reg_C_5 is used for the packet to match betweend prefix and
5565 * Pointer to Ethernet device.
5567 * Parent flow structure pointer.
5569 * Flow rule attributes.
5571 * Pattern specification (list terminated by the END pattern item).
5572 * @param[in] actions
5573 * Associated actions (list terminated by the END action).
5574 * @param[in] flow_split_info
5575 * Pointer to flow split info structure.
5577 * Perform verbose error reporting if not NULL.
5579 * 0 on success, negative value otherwise
5582 flow_create_split_meter(struct rte_eth_dev *dev,
5583 struct rte_flow *flow,
5584 const struct rte_flow_attr *attr,
5585 const struct rte_flow_item items[],
5586 const struct rte_flow_action actions[],
5587 struct mlx5_flow_split_info *flow_split_info,
5588 struct rte_flow_error *error)
5590 struct mlx5_priv *priv = dev->data->dev_private;
5591 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5592 struct rte_flow_action *sfx_actions = NULL;
5593 struct rte_flow_action *pre_actions = NULL;
5594 struct rte_flow_item *sfx_items = NULL;
5595 struct mlx5_flow *dev_flow = NULL;
5596 struct rte_flow_attr sfx_attr = *attr;
5597 struct mlx5_flow_meter_info *fm = NULL;
5598 uint8_t skip_scale_restore;
5599 bool has_mtr = false;
5600 bool has_modify = false;
5601 bool set_mtr_reg = true;
5602 uint32_t meter_id = 0;
5603 uint32_t mtr_idx = 0;
5604 uint32_t mtr_flow_id = 0;
5611 actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5612 &has_modify, &meter_id);
5615 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5617 return rte_flow_error_set(error, EINVAL,
5618 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5619 NULL, "Meter not found.");
5621 fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5623 return rte_flow_error_set(error, EINVAL,
5624 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5625 NULL, "Meter not found.");
5626 ret = mlx5_flow_meter_attach(priv, fm,
5630 flow->meter = mtr_idx;
5635 * If it isn't default-policy Meter, and
5636 * 1. There's no action in flow to change
5637 * packet (modify/encap/decap etc.), OR
5638 * 2. No drop count needed for this meter.
5639 * no need to use regC to save meter id anymore.
5641 if (!fm->def_policy && (!has_modify || !fm->drop_cnt))
5642 set_mtr_reg = false;
5643 /* Prefix actions: meter, decap, encap, tag, jump, end. */
5644 act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5645 sizeof(struct mlx5_rte_flow_action_set_tag);
5646 /* Suffix items: tag, vlan, port id, end. */
5647 #define METER_SUFFIX_ITEM 4
5648 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5649 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5650 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5653 return rte_flow_error_set(error, ENOMEM,
5654 RTE_FLOW_ERROR_TYPE_ACTION,
5655 NULL, "no memory to split "
5657 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5659 /* There's no suffix flow for meter of non-default policy. */
5660 if (!fm->def_policy)
5661 pre_actions = sfx_actions + 1;
5663 pre_actions = sfx_actions + actions_n;
5664 ret = flow_meter_split_prep(dev, flow, fm, &sfx_attr,
5665 items, sfx_items, actions,
5666 sfx_actions, pre_actions,
5667 (set_mtr_reg ? &mtr_flow_id : NULL),
5673 /* Add the prefix subflow. */
5674 flow_split_info->prefix_mark = 0;
5675 skip_scale_restore = flow_split_info->skip_scale;
5676 flow_split_info->skip_scale |=
5677 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5678 ret = flow_create_split_inner(dev, flow, &dev_flow,
5679 attr, items, pre_actions,
5680 flow_split_info, error);
5681 flow_split_info->skip_scale = skip_scale_restore;
5684 mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
5689 dev_flow->handle->split_flow_id = mtr_flow_id;
5690 dev_flow->handle->is_meter_flow_id = 1;
5692 if (!fm->def_policy) {
5693 if (!set_mtr_reg && fm->drop_cnt)
5695 flow_meter_create_drop_flow_with_org_pattern(dev, flow,
5701 /* Setting the sfx group atrr. */
5702 sfx_attr.group = sfx_attr.transfer ?
5703 (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5704 MLX5_FLOW_TABLE_LEVEL_METER;
5705 flow_split_info->prefix_layers =
5706 flow_get_prefix_layer_flags(dev_flow);
5707 flow_split_info->prefix_mark = dev_flow->handle->mark;
5708 flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
5710 /* Add the prefix subflow. */
5711 ret = flow_create_split_metadata(dev, flow,
5712 &sfx_attr, sfx_items ?
5714 sfx_actions ? sfx_actions : actions,
5715 flow_split_info, error);
5718 mlx5_free(sfx_actions);
5723 * The splitting for sample feature.
5725 * Once Sample action is detected in the action list, the flow actions should
5726 * be split into prefix sub flow and suffix sub flow.
5728 * The original items remain in the prefix sub flow, all actions preceding the
5729 * sample action and the sample action itself will be copied to the prefix
5730 * sub flow, the actions following the sample action will be copied to the
5731 * suffix sub flow, Queue action always be located in the suffix sub flow.
5733 * In order to make the packet from prefix sub flow matches with suffix sub
5734 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5735 * flow uses tag item with the unique flow id.
5738 * Pointer to Ethernet device.
5740 * Parent flow structure pointer.
5742 * Flow rule attributes.
5744 * Pattern specification (list terminated by the END pattern item).
5745 * @param[in] actions
5746 * Associated actions (list terminated by the END action).
5747 * @param[in] flow_split_info
5748 * Pointer to flow split info structure.
5750 * Perform verbose error reporting if not NULL.
5752 * 0 on success, negative value otherwise
5755 flow_create_split_sample(struct rte_eth_dev *dev,
5756 struct rte_flow *flow,
5757 const struct rte_flow_attr *attr,
5758 const struct rte_flow_item items[],
5759 const struct rte_flow_action actions[],
5760 struct mlx5_flow_split_info *flow_split_info,
5761 struct rte_flow_error *error)
5763 struct mlx5_priv *priv = dev->data->dev_private;
5764 struct rte_flow_action *sfx_actions = NULL;
5765 struct rte_flow_action *pre_actions = NULL;
5766 struct rte_flow_item *sfx_items = NULL;
5767 struct mlx5_flow *dev_flow = NULL;
5768 struct rte_flow_attr sfx_attr = *attr;
5769 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5770 struct mlx5_flow_dv_sample_resource *sample_res;
5771 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5772 struct mlx5_flow_tbl_resource *sfx_tbl;
5776 uint32_t fdb_tx = 0;
5779 int sample_action_pos;
5780 int qrss_action_pos;
5782 int modify_after_mirror = 0;
5783 uint16_t jump_table = 0;
5784 const uint32_t next_ft_step = 1;
5787 if (priv->sampler_en)
5788 actions_n = flow_check_match_action(actions, attr,
5789 RTE_FLOW_ACTION_TYPE_SAMPLE,
5790 &sample_action_pos, &qrss_action_pos,
5791 &modify_after_mirror);
5793 /* The prefix actions must includes sample, tag, end. */
5794 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5795 + sizeof(struct mlx5_rte_flow_action_set_tag);
5796 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5797 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5798 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5799 item_size), 0, SOCKET_ID_ANY);
5801 return rte_flow_error_set(error, ENOMEM,
5802 RTE_FLOW_ERROR_TYPE_ACTION,
5803 NULL, "no memory to split "
5805 /* The representor_id is -1 for uplink. */
5806 fdb_tx = (attr->transfer && priv->representor_id != -1);
5808 * When reg_c_preserve is set, metadata registers Cx preserve
5809 * their value even through packet duplication.
5811 add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5813 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5815 if (modify_after_mirror)
5816 jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5818 pre_actions = sfx_actions + actions_n;
5819 tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5820 actions, sfx_actions,
5821 pre_actions, actions_n,
5823 qrss_action_pos, jump_table,
5825 if (tag_id < 0 || (add_tag && !tag_id)) {
5829 if (modify_after_mirror)
5830 flow_split_info->skip_scale =
5831 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5832 /* Add the prefix subflow. */
5833 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5835 flow_split_info, error);
5840 dev_flow->handle->split_flow_id = tag_id;
5841 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5842 if (!modify_after_mirror) {
5843 /* Set the sfx group attr. */
5844 sample_res = (struct mlx5_flow_dv_sample_resource *)
5845 dev_flow->dv.sample_res;
5846 sfx_tbl = (struct mlx5_flow_tbl_resource *)
5847 sample_res->normal_path_tbl;
5848 sfx_tbl_data = container_of(sfx_tbl,
5849 struct mlx5_flow_tbl_data_entry,
5851 sfx_attr.group = sfx_attr.transfer ?
5852 (sfx_tbl_data->level - 1) : sfx_tbl_data->level;
5854 MLX5_ASSERT(attr->transfer);
5855 sfx_attr.group = jump_table;
5857 flow_split_info->prefix_layers =
5858 flow_get_prefix_layer_flags(dev_flow);
5859 flow_split_info->prefix_mark = dev_flow->handle->mark;
5860 /* Suffix group level already be scaled with factor, set
5861 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5862 * again in translation.
5864 flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5867 /* Add the suffix subflow. */
5868 ret = flow_create_split_meter(dev, flow, &sfx_attr,
5869 sfx_items ? sfx_items : items,
5870 sfx_actions ? sfx_actions : actions,
5871 flow_split_info, error);
5874 mlx5_free(sfx_actions);
5879 * Split the flow to subflow set. The splitters might be linked
5880 * in the chain, like this:
5881 * flow_create_split_outer() calls:
5882 * flow_create_split_meter() calls:
5883 * flow_create_split_metadata(meter_subflow_0) calls:
5884 * flow_create_split_inner(metadata_subflow_0)
5885 * flow_create_split_inner(metadata_subflow_1)
5886 * flow_create_split_inner(metadata_subflow_2)
5887 * flow_create_split_metadata(meter_subflow_1) calls:
5888 * flow_create_split_inner(metadata_subflow_0)
5889 * flow_create_split_inner(metadata_subflow_1)
5890 * flow_create_split_inner(metadata_subflow_2)
5892 * This provide flexible way to add new levels of flow splitting.
5893 * The all of successfully created subflows are included to the
5894 * parent flow dev_flow list.
5897 * Pointer to Ethernet device.
5899 * Parent flow structure pointer.
5901 * Flow rule attributes.
5903 * Pattern specification (list terminated by the END pattern item).
5904 * @param[in] actions
5905 * Associated actions (list terminated by the END action).
5906 * @param[in] flow_split_info
5907 * Pointer to flow split info structure.
5909 * Perform verbose error reporting if not NULL.
5911 * 0 on success, negative value otherwise
5914 flow_create_split_outer(struct rte_eth_dev *dev,
5915 struct rte_flow *flow,
5916 const struct rte_flow_attr *attr,
5917 const struct rte_flow_item items[],
5918 const struct rte_flow_action actions[],
5919 struct mlx5_flow_split_info *flow_split_info,
5920 struct rte_flow_error *error)
5924 ret = flow_create_split_sample(dev, flow, attr, items,
5925 actions, flow_split_info, error);
5926 MLX5_ASSERT(ret <= 0);
5930 static struct mlx5_flow_tunnel *
5931 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5932 const struct rte_flow_attr *attr,
5933 const struct rte_flow_item items[],
5934 const struct rte_flow_action actions[])
5936 struct mlx5_flow_tunnel *tunnel;
5938 #pragma GCC diagnostic push
5939 #pragma GCC diagnostic ignored "-Wcast-qual"
5940 if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5941 tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5942 else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5943 tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5946 #pragma GCC diagnostic pop
5952 * Adjust flow RSS workspace if needed.
5955 * Pointer to thread flow work space.
5957 * Pointer to RSS descriptor.
5958 * @param[in] nrssq_num
5959 * New RSS queue number.
5962 * 0 on success, -1 otherwise and rte_errno is set.
5965 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5966 struct mlx5_flow_rss_desc *rss_desc,
5969 if (likely(nrssq_num <= wks->rssq_num))
5971 rss_desc->queue = realloc(rss_desc->queue,
5972 sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5973 if (!rss_desc->queue) {
5977 wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5982 * Create a flow and add it to @p list.
5985 * Pointer to Ethernet device.
5987 * Pointer to a TAILQ flow list. If this parameter NULL,
5988 * no list insertion occurred, flow is just created,
5989 * this is caller's responsibility to track the
5992 * Flow rule attributes.
5994 * Pattern specification (list terminated by the END pattern item).
5995 * @param[in] actions
5996 * Associated actions (list terminated by the END action).
5997 * @param[in] external
5998 * This flow rule is created by request external to PMD.
6000 * Perform verbose error reporting if not NULL.
6003 * A flow index on success, 0 otherwise and rte_errno is set.
6006 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
6007 const struct rte_flow_attr *attr,
6008 const struct rte_flow_item items[],
6009 const struct rte_flow_action original_actions[],
6010 bool external, struct rte_flow_error *error)
6012 struct mlx5_priv *priv = dev->data->dev_private;
6013 struct rte_flow *flow = NULL;
6014 struct mlx5_flow *dev_flow;
6015 const struct rte_flow_action_rss *rss = NULL;
6016 struct mlx5_translated_action_handle
6017 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6018 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6020 struct mlx5_flow_expand_rss buf;
6021 uint8_t buffer[2048];
6024 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6025 uint8_t buffer[2048];
6028 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6029 uint8_t buffer[2048];
6030 } actions_hairpin_tx;
6032 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6033 uint8_t buffer[2048];
6035 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6036 struct mlx5_flow_rss_desc *rss_desc;
6037 const struct rte_flow_action *p_actions_rx;
6041 struct rte_flow_attr attr_tx = { .priority = 0 };
6042 const struct rte_flow_action *actions;
6043 struct rte_flow_action *translated_actions = NULL;
6044 struct mlx5_flow_tunnel *tunnel;
6045 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6046 struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6047 struct mlx5_flow_split_info flow_split_info = {
6048 .external = !!external,
6058 rss_desc = &wks->rss_desc;
6059 ret = flow_action_handles_translate(dev, original_actions,
6062 &translated_actions, error);
6064 MLX5_ASSERT(translated_actions == NULL);
6067 actions = translated_actions ? translated_actions : original_actions;
6068 p_actions_rx = actions;
6069 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6070 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6071 external, hairpin_flow, error);
6073 goto error_before_hairpin_split;
6074 flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
6077 goto error_before_hairpin_split;
6079 if (hairpin_flow > 0) {
6080 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6082 goto error_before_hairpin_split;
6084 flow_hairpin_split(dev, actions, actions_rx.actions,
6085 actions_hairpin_tx.actions, items_tx.items,
6087 p_actions_rx = actions_rx.actions;
6089 flow_split_info.flow_idx = idx;
6090 flow->drv_type = flow_get_drv_type(dev, attr);
6091 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6092 flow->drv_type < MLX5_FLOW_TYPE_MAX);
6093 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6094 /* RSS Action only works on NIC RX domain */
6095 if (attr->ingress && !attr->transfer)
6096 rss = flow_get_rss_action(dev, p_actions_rx);
6098 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6101 * The following information is required by
6102 * mlx5_flow_hashfields_adjust() in advance.
6104 rss_desc->level = rss->level;
6105 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6106 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6108 flow->dev_handles = 0;
6109 if (rss && rss->types) {
6110 unsigned int graph_root;
6112 graph_root = find_graph_root(items, rss->level);
6113 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6115 mlx5_support_expansion, graph_root);
6116 MLX5_ASSERT(ret > 0 &&
6117 (unsigned int)ret < sizeof(expand_buffer.buffer));
6120 buf->entry[0].pattern = (void *)(uintptr_t)items;
6122 rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6124 for (i = 0; i < buf->entries; ++i) {
6125 /* Initialize flow split data. */
6126 flow_split_info.prefix_layers = 0;
6127 flow_split_info.prefix_mark = 0;
6128 flow_split_info.skip_scale = 0;
6130 * The splitter may create multiple dev_flows,
6131 * depending on configuration. In the simplest
6132 * case it just creates unmodified original flow.
6134 ret = flow_create_split_outer(dev, flow, attr,
6135 buf->entry[i].pattern,
6136 p_actions_rx, &flow_split_info,
6140 if (is_flow_tunnel_steer_rule(dev, attr,
6141 buf->entry[i].pattern,
6143 ret = flow_tunnel_add_default_miss(dev, flow, attr,
6149 mlx5_free(default_miss_ctx.queue);
6154 /* Create the tx flow. */
6156 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6157 attr_tx.ingress = 0;
6159 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6160 actions_hairpin_tx.actions,
6164 dev_flow->flow = flow;
6165 dev_flow->external = 0;
6166 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6167 dev_flow->handle, next);
6168 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6170 actions_hairpin_tx.actions, error);
6175 * Update the metadata register copy table. If extensive
6176 * metadata feature is enabled and registers are supported
6177 * we might create the extra rte_flow for each unique
6178 * MARK/FLAG action ID.
6180 * The table is updated for ingress Flows only, because
6181 * the egress Flows belong to the different device and
6182 * copy table should be updated in peer NIC Rx domain.
6184 if (attr->ingress &&
6185 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6186 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6191 * If the flow is external (from application) OR device is started,
6192 * OR mreg discover, then apply immediately.
6194 if (external || dev->data->dev_started ||
6195 (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6196 attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6197 ret = flow_drv_apply(dev, flow, error);
6202 rte_spinlock_lock(&priv->flow_list_lock);
6203 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
6205 rte_spinlock_unlock(&priv->flow_list_lock);
6207 flow_rxq_flags_set(dev, flow);
6208 rte_free(translated_actions);
6209 tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
6212 flow->tunnel_id = tunnel->tunnel_id;
6213 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6214 mlx5_free(default_miss_ctx.queue);
6216 mlx5_flow_pop_thread_workspace();
6220 ret = rte_errno; /* Save rte_errno before cleanup. */
6221 flow_mreg_del_copy_action(dev, flow);
6222 flow_drv_destroy(dev, flow);
6223 if (rss_desc->shared_rss)
6224 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6226 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6227 rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6228 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
6229 rte_errno = ret; /* Restore rte_errno. */
6232 mlx5_flow_pop_thread_workspace();
6233 error_before_hairpin_split:
6234 rte_free(translated_actions);
6239 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6240 * incoming packets to table 1.
6242 * Other flow rules, requested for group n, will be created in
6243 * e-switch table n+1.
6244 * Jump action to e-switch group n will be created to group n+1.
6246 * Used when working in switchdev mode, to utilise advantages of table 1
6250 * Pointer to Ethernet device.
6253 * Pointer to flow on success, NULL otherwise and rte_errno is set.
6256 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6258 const struct rte_flow_attr attr = {
6265 const struct rte_flow_item pattern = {
6266 .type = RTE_FLOW_ITEM_TYPE_END,
6268 struct rte_flow_action_jump jump = {
6271 const struct rte_flow_action actions[] = {
6273 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6277 .type = RTE_FLOW_ACTION_TYPE_END,
6280 struct mlx5_priv *priv = dev->data->dev_private;
6281 struct rte_flow_error error;
6283 return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
6285 actions, false, &error);
6289 * Validate a flow supported by the NIC.
6291 * @see rte_flow_validate()
6295 mlx5_flow_validate(struct rte_eth_dev *dev,
6296 const struct rte_flow_attr *attr,
6297 const struct rte_flow_item items[],
6298 const struct rte_flow_action original_actions[],
6299 struct rte_flow_error *error)
6302 struct mlx5_translated_action_handle
6303 indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6304 int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6305 const struct rte_flow_action *actions;
6306 struct rte_flow_action *translated_actions = NULL;
6307 int ret = flow_action_handles_translate(dev, original_actions,
6310 &translated_actions, error);
6314 actions = translated_actions ? translated_actions : original_actions;
6315 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6316 ret = flow_drv_validate(dev, attr, items, actions,
6317 true, hairpin_flow, error);
6318 rte_free(translated_actions);
6325 * @see rte_flow_create()
6329 mlx5_flow_create(struct rte_eth_dev *dev,
6330 const struct rte_flow_attr *attr,
6331 const struct rte_flow_item items[],
6332 const struct rte_flow_action actions[],
6333 struct rte_flow_error *error)
6335 struct mlx5_priv *priv = dev->data->dev_private;
6338 * If the device is not started yet, it is not allowed to created a
6339 * flow from application. PMD default flows and traffic control flows
6342 if (unlikely(!dev->data->dev_started)) {
6343 DRV_LOG(DEBUG, "port %u is not started when "
6344 "inserting a flow", dev->data->port_id);
6345 rte_flow_error_set(error, ENODEV,
6346 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6348 "port not started");
6352 return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
6353 attr, items, actions, true, error);
6357 * Destroy a flow in a list.
6360 * Pointer to Ethernet device.
6362 * Pointer to the Indexed flow list. If this parameter NULL,
6363 * there is no flow removal from the list. Be noted that as
6364 * flow is add to the indexed list, memory of the indexed
6365 * list points to maybe changed as flow destroyed.
6366 * @param[in] flow_idx
6367 * Index of flow to destroy.
6370 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
6373 struct mlx5_priv *priv = dev->data->dev_private;
6374 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6375 [MLX5_IPOOL_RTE_FLOW], flow_idx);
6380 * Update RX queue flags only if port is started, otherwise it is
6383 if (dev->data->dev_started)
6384 flow_rxq_flags_trim(dev, flow);
6385 flow_drv_destroy(dev, flow);
6387 rte_spinlock_lock(&priv->flow_list_lock);
6388 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
6389 flow_idx, flow, next);
6390 rte_spinlock_unlock(&priv->flow_list_lock);
6393 struct mlx5_flow_tunnel *tunnel;
6395 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6397 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6398 mlx5_flow_tunnel_free(dev, tunnel);
6400 flow_mreg_del_copy_action(dev, flow);
6401 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6405 * Destroy all flows.
6408 * Pointer to Ethernet device.
6410 * Pointer to the Indexed flow list.
6412 * If flushing is called avtively.
6415 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
6417 uint32_t num_flushed = 0;
6420 flow_list_destroy(dev, list, *list);
6424 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6425 dev->data->port_id, num_flushed);
6430 * Stop all default actions for flows.
6433 * Pointer to Ethernet device.
6436 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6438 flow_mreg_del_default_copy_action(dev);
6439 flow_rxq_flags_clear(dev);
6443 * Start all default actions for flows.
6446 * Pointer to Ethernet device.
6448 * 0 on success, a negative errno value otherwise and rte_errno is set.
6451 mlx5_flow_start_default(struct rte_eth_dev *dev)
6453 struct rte_flow_error error;
6455 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6456 return flow_mreg_add_default_copy_action(dev, &error);
6460 * Release key of thread specific flow workspace data.
6463 flow_release_workspace(void *data)
6465 struct mlx5_flow_workspace *wks = data;
6466 struct mlx5_flow_workspace *next;
6470 free(wks->rss_desc.queue);
6477 * Get thread specific current flow workspace.
6479 * @return pointer to thread specific flow workspace data, NULL on error.
6481 struct mlx5_flow_workspace*
6482 mlx5_flow_get_thread_workspace(void)
6484 struct mlx5_flow_workspace *data;
6486 data = mlx5_flow_os_get_specific_workspace();
6487 MLX5_ASSERT(data && data->inuse);
6488 if (!data || !data->inuse)
6489 DRV_LOG(ERR, "flow workspace not initialized.");
6494 * Allocate and init new flow workspace.
6496 * @return pointer to flow workspace data, NULL on error.
6498 static struct mlx5_flow_workspace*
6499 flow_alloc_thread_workspace(void)
6501 struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6504 DRV_LOG(ERR, "Failed to allocate flow workspace "
6508 data->rss_desc.queue = calloc(1,
6509 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6510 if (!data->rss_desc.queue)
6512 data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6515 if (data->rss_desc.queue)
6516 free(data->rss_desc.queue);
6522 * Get new thread specific flow workspace.
6524 * If current workspace inuse, create new one and set as current.
6526 * @return pointer to thread specific flow workspace data, NULL on error.
6528 static struct mlx5_flow_workspace*
6529 mlx5_flow_push_thread_workspace(void)
6531 struct mlx5_flow_workspace *curr;
6532 struct mlx5_flow_workspace *data;
6534 curr = mlx5_flow_os_get_specific_workspace();
6536 data = flow_alloc_thread_workspace();
6539 } else if (!curr->inuse) {
6541 } else if (curr->next) {
6544 data = flow_alloc_thread_workspace();
6552 /* Set as current workspace */
6553 if (mlx5_flow_os_set_specific_workspace(data))
6554 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6559 * Close current thread specific flow workspace.
6561 * If previous workspace available, set it as current.
6563 * @return pointer to thread specific flow workspace data, NULL on error.
6566 mlx5_flow_pop_thread_workspace(void)
6568 struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6573 DRV_LOG(ERR, "Failed to close unused flow workspace.");
6579 if (mlx5_flow_os_set_specific_workspace(data->prev))
6580 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6584 * Verify the flow list is empty
6587 * Pointer to Ethernet device.
6589 * @return the number of flows not released.
6592 mlx5_flow_verify(struct rte_eth_dev *dev)
6594 struct mlx5_priv *priv = dev->data->dev_private;
6595 struct rte_flow *flow;
6599 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6601 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6602 dev->data->port_id, (void *)flow);
6609 * Enable default hairpin egress flow.
6612 * Pointer to Ethernet device.
6617 * 0 on success, a negative errno value otherwise and rte_errno is set.
6620 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6623 struct mlx5_priv *priv = dev->data->dev_private;
6624 const struct rte_flow_attr attr = {
6628 struct mlx5_rte_flow_item_tx_queue queue_spec = {
6631 struct mlx5_rte_flow_item_tx_queue queue_mask = {
6632 .queue = UINT32_MAX,
6634 struct rte_flow_item items[] = {
6636 .type = (enum rte_flow_item_type)
6637 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6638 .spec = &queue_spec,
6640 .mask = &queue_mask,
6643 .type = RTE_FLOW_ITEM_TYPE_END,
6646 struct rte_flow_action_jump jump = {
6647 .group = MLX5_HAIRPIN_TX_TABLE,
6649 struct rte_flow_action actions[2];
6651 struct rte_flow_error error;
6653 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6654 actions[0].conf = &jump;
6655 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6656 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6657 &attr, items, actions, false, &error);
6660 "Failed to create ctrl flow: rte_errno(%d),"
6661 " type(%d), message(%s)",
6662 rte_errno, error.type,
6663 error.message ? error.message : " (no stated reason)");
6670 * Enable a control flow configured from the control plane.
6673 * Pointer to Ethernet device.
6675 * An Ethernet flow spec to apply.
6677 * An Ethernet flow mask to apply.
6679 * A VLAN flow spec to apply.
6681 * A VLAN flow mask to apply.
6684 * 0 on success, a negative errno value otherwise and rte_errno is set.
6687 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6688 struct rte_flow_item_eth *eth_spec,
6689 struct rte_flow_item_eth *eth_mask,
6690 struct rte_flow_item_vlan *vlan_spec,
6691 struct rte_flow_item_vlan *vlan_mask)
6693 struct mlx5_priv *priv = dev->data->dev_private;
6694 const struct rte_flow_attr attr = {
6696 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6698 struct rte_flow_item items[] = {
6700 .type = RTE_FLOW_ITEM_TYPE_ETH,
6706 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6707 RTE_FLOW_ITEM_TYPE_END,
6713 .type = RTE_FLOW_ITEM_TYPE_END,
6716 uint16_t queue[priv->reta_idx_n];
6717 struct rte_flow_action_rss action_rss = {
6718 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6720 .types = priv->rss_conf.rss_hf,
6721 .key_len = priv->rss_conf.rss_key_len,
6722 .queue_num = priv->reta_idx_n,
6723 .key = priv->rss_conf.rss_key,
6726 struct rte_flow_action actions[] = {
6728 .type = RTE_FLOW_ACTION_TYPE_RSS,
6729 .conf = &action_rss,
6732 .type = RTE_FLOW_ACTION_TYPE_END,
6736 struct rte_flow_error error;
6739 if (!priv->reta_idx_n || !priv->rxqs_n) {
6742 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6743 action_rss.types = 0;
6744 for (i = 0; i != priv->reta_idx_n; ++i)
6745 queue[i] = (*priv->reta_idx)[i];
6746 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6747 &attr, items, actions, false, &error);
6754 * Enable a flow control configured from the control plane.
6757 * Pointer to Ethernet device.
6759 * An Ethernet flow spec to apply.
6761 * An Ethernet flow mask to apply.
6764 * 0 on success, a negative errno value otherwise and rte_errno is set.
6767 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6768 struct rte_flow_item_eth *eth_spec,
6769 struct rte_flow_item_eth *eth_mask)
6771 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6775 * Create default miss flow rule matching lacp traffic
6778 * Pointer to Ethernet device.
6780 * An Ethernet flow spec to apply.
6783 * 0 on success, a negative errno value otherwise and rte_errno is set.
6786 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6788 struct mlx5_priv *priv = dev->data->dev_private;
6790 * The LACP matching is done by only using ether type since using
6791 * a multicast dst mac causes kernel to give low priority to this flow.
6793 static const struct rte_flow_item_eth lacp_spec = {
6794 .type = RTE_BE16(0x8809),
6796 static const struct rte_flow_item_eth lacp_mask = {
6799 const struct rte_flow_attr attr = {
6802 struct rte_flow_item items[] = {
6804 .type = RTE_FLOW_ITEM_TYPE_ETH,
6809 .type = RTE_FLOW_ITEM_TYPE_END,
6812 struct rte_flow_action actions[] = {
6814 .type = (enum rte_flow_action_type)
6815 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6818 .type = RTE_FLOW_ACTION_TYPE_END,
6821 struct rte_flow_error error;
6822 uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6823 &attr, items, actions, false, &error);
6833 * @see rte_flow_destroy()
6837 mlx5_flow_destroy(struct rte_eth_dev *dev,
6838 struct rte_flow *flow,
6839 struct rte_flow_error *error __rte_unused)
6841 struct mlx5_priv *priv = dev->data->dev_private;
6843 flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6848 * Destroy all flows.
6850 * @see rte_flow_flush()
6854 mlx5_flow_flush(struct rte_eth_dev *dev,
6855 struct rte_flow_error *error __rte_unused)
6857 struct mlx5_priv *priv = dev->data->dev_private;
6859 mlx5_flow_list_flush(dev, &priv->flows, false);
6866 * @see rte_flow_isolate()
6870 mlx5_flow_isolate(struct rte_eth_dev *dev,
6872 struct rte_flow_error *error)
6874 struct mlx5_priv *priv = dev->data->dev_private;
6876 if (dev->data->dev_started) {
6877 rte_flow_error_set(error, EBUSY,
6878 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6880 "port must be stopped first");
6883 priv->isolated = !!enable;
6885 dev->dev_ops = &mlx5_dev_ops_isolate;
6887 dev->dev_ops = &mlx5_dev_ops;
6889 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6890 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6898 * @see rte_flow_query()
6902 flow_drv_query(struct rte_eth_dev *dev,
6904 const struct rte_flow_action *actions,
6906 struct rte_flow_error *error)
6908 struct mlx5_priv *priv = dev->data->dev_private;
6909 const struct mlx5_flow_driver_ops *fops;
6910 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6911 [MLX5_IPOOL_RTE_FLOW],
6913 enum mlx5_flow_drv_type ftype;
6916 return rte_flow_error_set(error, ENOENT,
6917 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6919 "invalid flow handle");
6921 ftype = flow->drv_type;
6922 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6923 fops = flow_get_drv_ops(ftype);
6925 return fops->query(dev, flow, actions, data, error);
6931 * @see rte_flow_query()
6935 mlx5_flow_query(struct rte_eth_dev *dev,
6936 struct rte_flow *flow,
6937 const struct rte_flow_action *actions,
6939 struct rte_flow_error *error)
6943 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6951 * Get rte_flow callbacks.
6954 * Pointer to Ethernet device structure.
6956 * Pointer to operation-specific structure.
6961 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
6962 const struct rte_flow_ops **ops)
6964 *ops = &mlx5_flow_ops;
6969 * Validate meter policy actions.
6970 * Dispatcher for action type specific validation.
6973 * Pointer to the Ethernet device structure.
6975 * The meter policy action object to validate.
6977 * Attributes of flow to determine steering domain.
6978 * @param[out] is_rss
6980 * @param[out] domain_bitmap
6982 * @param[out] is_def_policy
6983 * Is default policy or not.
6985 * Perform verbose error reporting if not NULL. Initialized in case of
6989 * 0 on success, otherwise negative errno value.
6992 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
6993 const struct rte_flow_action *actions[RTE_COLORS],
6994 struct rte_flow_attr *attr,
6996 uint8_t *domain_bitmap,
6997 bool *is_def_policy,
6998 struct rte_mtr_error *error)
7000 const struct mlx5_flow_driver_ops *fops;
7002 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7003 return fops->validate_mtr_acts(dev, actions, attr,
7004 is_rss, domain_bitmap, is_def_policy, error);
7008 * Destroy the meter table set.
7011 * Pointer to Ethernet device.
7012 * @param[in] mtr_policy
7013 * Meter policy struct.
7016 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7017 struct mlx5_flow_meter_policy *mtr_policy)
7019 const struct mlx5_flow_driver_ops *fops;
7021 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7022 fops->destroy_mtr_acts(dev, mtr_policy);
7026 * Create policy action, lock free,
7027 * (mutex should be acquired by caller).
7028 * Dispatcher for action type specific call.
7031 * Pointer to the Ethernet device structure.
7032 * @param[in] mtr_policy
7033 * Meter policy struct.
7035 * Action specification used to create meter actions.
7037 * Perform verbose error reporting if not NULL. Initialized in case of
7041 * 0 on success, otherwise negative errno value.
7044 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7045 struct mlx5_flow_meter_policy *mtr_policy,
7046 const struct rte_flow_action *actions[RTE_COLORS],
7047 struct rte_mtr_error *error)
7049 const struct mlx5_flow_driver_ops *fops;
7051 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7052 return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7056 * Create policy rules, lock free,
7057 * (mutex should be acquired by caller).
7058 * Dispatcher for action type specific call.
7061 * Pointer to the Ethernet device structure.
7062 * @param[in] mtr_policy
7063 * Meter policy struct.
7066 * 0 on success, -1 otherwise.
7069 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7070 struct mlx5_flow_meter_policy *mtr_policy)
7072 const struct mlx5_flow_driver_ops *fops;
7074 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7075 return fops->create_policy_rules(dev, mtr_policy);
7079 * Destroy policy rules, lock free,
7080 * (mutex should be acquired by caller).
7081 * Dispatcher for action type specific call.
7084 * Pointer to the Ethernet device structure.
7085 * @param[in] mtr_policy
7086 * Meter policy struct.
7089 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7090 struct mlx5_flow_meter_policy *mtr_policy)
7092 const struct mlx5_flow_driver_ops *fops;
7094 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7095 fops->destroy_policy_rules(dev, mtr_policy);
7099 * Destroy the default policy table set.
7102 * Pointer to Ethernet device.
7105 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7107 const struct mlx5_flow_driver_ops *fops;
7109 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7110 fops->destroy_def_policy(dev);
7114 * Destroy the default policy table set.
7117 * Pointer to Ethernet device.
7120 * 0 on success, -1 otherwise.
7123 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7125 const struct mlx5_flow_driver_ops *fops;
7127 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7128 return fops->create_def_policy(dev);
7132 * Create the needed meter and suffix tables.
7135 * Pointer to Ethernet device.
7138 * 0 on success, -1 otherwise.
7141 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7142 struct mlx5_flow_meter_info *fm,
7144 uint8_t domain_bitmap)
7146 const struct mlx5_flow_driver_ops *fops;
7148 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7149 return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7153 * Destroy the meter table set.
7156 * Pointer to Ethernet device.
7158 * Pointer to the meter table set.
7161 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7162 struct mlx5_flow_meter_info *fm)
7164 const struct mlx5_flow_driver_ops *fops;
7166 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7167 fops->destroy_mtr_tbls(dev, fm);
7171 * Destroy the global meter drop table.
7174 * Pointer to Ethernet device.
7177 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7179 const struct mlx5_flow_driver_ops *fops;
7181 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7182 fops->destroy_mtr_drop_tbls(dev);
7186 * Allocate the needed aso flow meter id.
7189 * Pointer to Ethernet device.
7192 * Index to aso flow meter on success, NULL otherwise.
7195 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7197 const struct mlx5_flow_driver_ops *fops;
7199 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7200 return fops->create_meter(dev);
7204 * Free the aso flow meter id.
7207 * Pointer to Ethernet device.
7208 * @param[in] mtr_idx
7209 * Index to aso flow meter to be free.
7215 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7217 const struct mlx5_flow_driver_ops *fops;
7219 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7220 fops->free_meter(dev, mtr_idx);
7224 * Allocate a counter.
7227 * Pointer to Ethernet device structure.
7230 * Index to allocated counter on success, 0 otherwise.
7233 mlx5_counter_alloc(struct rte_eth_dev *dev)
7235 const struct mlx5_flow_driver_ops *fops;
7236 struct rte_flow_attr attr = { .transfer = 0 };
7238 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7239 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7240 return fops->counter_alloc(dev);
7243 "port %u counter allocate is not supported.",
7244 dev->data->port_id);
7252 * Pointer to Ethernet device structure.
7254 * Index to counter to be free.
7257 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7259 const struct mlx5_flow_driver_ops *fops;
7260 struct rte_flow_attr attr = { .transfer = 0 };
7262 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7263 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7264 fops->counter_free(dev, cnt);
7268 "port %u counter free is not supported.",
7269 dev->data->port_id);
7273 * Query counter statistics.
7276 * Pointer to Ethernet device structure.
7278 * Index to counter to query.
7280 * Set to clear counter statistics.
7282 * The counter hits packets number to save.
7284 * The counter hits bytes number to save.
7287 * 0 on success, a negative errno value otherwise.
7290 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7291 bool clear, uint64_t *pkts, uint64_t *bytes)
7293 const struct mlx5_flow_driver_ops *fops;
7294 struct rte_flow_attr attr = { .transfer = 0 };
7296 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7297 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7298 return fops->counter_query(dev, cnt, clear, pkts, bytes);
7301 "port %u counter query is not supported.",
7302 dev->data->port_id);
7307 * Allocate a new memory for the counter values wrapped by all the needed
7311 * Pointer to mlx5_dev_ctx_shared object.
7314 * 0 on success, a negative errno value otherwise.
7317 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7319 struct mlx5_devx_mkey_attr mkey_attr;
7320 struct mlx5_counter_stats_mem_mng *mem_mng;
7321 volatile struct flow_counter_stats *raw_data;
7322 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7323 int size = (sizeof(struct flow_counter_stats) *
7324 MLX5_COUNTERS_PER_POOL +
7325 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7326 sizeof(struct mlx5_counter_stats_mem_mng);
7327 size_t pgsize = rte_mem_page_size();
7331 if (pgsize == (size_t)-1) {
7332 DRV_LOG(ERR, "Failed to get mem page size");
7336 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7341 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7342 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7343 mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7344 IBV_ACCESS_LOCAL_WRITE);
7345 if (!mem_mng->umem) {
7350 mkey_attr.addr = (uintptr_t)mem;
7351 mkey_attr.size = size;
7352 mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7353 mkey_attr.pd = sh->pdn;
7354 mkey_attr.log_entity_size = 0;
7355 mkey_attr.pg_access = 0;
7356 mkey_attr.klm_array = NULL;
7357 mkey_attr.klm_num = 0;
7358 mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7359 mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7360 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7362 mlx5_os_umem_dereg(mem_mng->umem);
7367 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7368 raw_data = (volatile struct flow_counter_stats *)mem;
7369 for (i = 0; i < raws_n; ++i) {
7370 mem_mng->raws[i].mem_mng = mem_mng;
7371 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7373 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7374 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7375 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7377 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7378 sh->cmng.mem_mng = mem_mng;
7383 * Set the statistic memory to the new counter pool.
7386 * Pointer to mlx5_dev_ctx_shared object.
7388 * Pointer to the pool to set the statistic memory.
7391 * 0 on success, a negative errno value otherwise.
7394 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7395 struct mlx5_flow_counter_pool *pool)
7397 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7398 /* Resize statistic memory once used out. */
7399 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7400 mlx5_flow_create_counter_stat_mem_mng(sh)) {
7401 DRV_LOG(ERR, "Cannot resize counter stat mem.");
7404 rte_spinlock_lock(&pool->sl);
7405 pool->raw = cmng->mem_mng->raws + pool->index %
7406 MLX5_CNT_CONTAINER_RESIZE;
7407 rte_spinlock_unlock(&pool->sl);
7408 pool->raw_hw = NULL;
7412 #define MLX5_POOL_QUERY_FREQ_US 1000000
7415 * Set the periodic procedure for triggering asynchronous batch queries for all
7416 * the counter pools.
7419 * Pointer to mlx5_dev_ctx_shared object.
7422 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7424 uint32_t pools_n, us;
7426 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7427 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7428 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7429 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7430 sh->cmng.query_thread_on = 0;
7431 DRV_LOG(ERR, "Cannot reinitialize query alarm");
7433 sh->cmng.query_thread_on = 1;
7438 * The periodic procedure for triggering asynchronous batch queries for all the
7439 * counter pools. This function is probably called by the host thread.
7442 * The parameter for the alarm process.
7445 mlx5_flow_query_alarm(void *arg)
7447 struct mlx5_dev_ctx_shared *sh = arg;
7449 uint16_t pool_index = sh->cmng.pool_index;
7450 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7451 struct mlx5_flow_counter_pool *pool;
7454 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7456 rte_spinlock_lock(&cmng->pool_update_sl);
7457 pool = cmng->pools[pool_index];
7458 n_valid = cmng->n_valid;
7459 rte_spinlock_unlock(&cmng->pool_update_sl);
7460 /* Set the statistic memory to the new created pool. */
7461 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7464 /* There is a pool query in progress. */
7467 LIST_FIRST(&sh->cmng.free_stat_raws);
7469 /* No free counter statistics raw memory. */
7472 * Identify the counters released between query trigger and query
7473 * handle more efficiently. The counter released in this gap period
7474 * should wait for a new round of query as the new arrived packets
7475 * will not be taken into account.
7478 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7479 MLX5_COUNTERS_PER_POOL,
7481 pool->raw_hw->mem_mng->dm->id,
7485 (uint64_t)(uintptr_t)pool);
7487 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7488 " %d", pool->min_dcs->id);
7489 pool->raw_hw = NULL;
7492 LIST_REMOVE(pool->raw_hw, next);
7493 sh->cmng.pending_queries++;
7495 if (pool_index >= n_valid)
7498 sh->cmng.pool_index = pool_index;
7499 mlx5_set_query_alarm(sh);
7503 * Check and callback event for new aged flow in the counter pool
7506 * Pointer to mlx5_dev_ctx_shared object.
7508 * Pointer to Current counter pool.
7511 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7512 struct mlx5_flow_counter_pool *pool)
7514 struct mlx5_priv *priv;
7515 struct mlx5_flow_counter *cnt;
7516 struct mlx5_age_info *age_info;
7517 struct mlx5_age_param *age_param;
7518 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7519 struct mlx5_counter_stats_raw *prev = pool->raw;
7520 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7521 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7522 uint16_t expected = AGE_CANDIDATE;
7525 pool->time_of_last_age_check = curr_time;
7526 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7527 cnt = MLX5_POOL_GET_CNT(pool, i);
7528 age_param = MLX5_CNT_TO_AGE(cnt);
7529 if (__atomic_load_n(&age_param->state,
7530 __ATOMIC_RELAXED) != AGE_CANDIDATE)
7532 if (cur->data[i].hits != prev->data[i].hits) {
7533 __atomic_store_n(&age_param->sec_since_last_hit, 0,
7537 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7539 __ATOMIC_RELAXED) <= age_param->timeout)
7542 * Hold the lock first, or if between the
7543 * state AGE_TMOUT and tailq operation the
7544 * release happened, the release procedure
7545 * may delete a non-existent tailq node.
7547 priv = rte_eth_devices[age_param->port_id].data->dev_private;
7548 age_info = GET_PORT_AGE_INFO(priv);
7549 rte_spinlock_lock(&age_info->aged_sl);
7550 if (__atomic_compare_exchange_n(&age_param->state, &expected,
7553 __ATOMIC_RELAXED)) {
7554 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7555 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7557 rte_spinlock_unlock(&age_info->aged_sl);
7559 mlx5_age_event_prepare(sh);
7563 * Handler for the HW respond about ready values from an asynchronous batch
7564 * query. This function is probably called by the host thread.
7567 * The pointer to the shared device context.
7568 * @param[in] async_id
7569 * The Devx async ID.
7571 * The status of the completion.
7574 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7575 uint64_t async_id, int status)
7577 struct mlx5_flow_counter_pool *pool =
7578 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7579 struct mlx5_counter_stats_raw *raw_to_free;
7580 uint8_t query_gen = pool->query_gen ^ 1;
7581 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7582 enum mlx5_counter_type cnt_type =
7583 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7584 MLX5_COUNTER_TYPE_ORIGIN;
7586 if (unlikely(status)) {
7587 raw_to_free = pool->raw_hw;
7589 raw_to_free = pool->raw;
7591 mlx5_flow_aging_check(sh, pool);
7592 rte_spinlock_lock(&pool->sl);
7593 pool->raw = pool->raw_hw;
7594 rte_spinlock_unlock(&pool->sl);
7595 /* Be sure the new raw counters data is updated in memory. */
7597 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7598 rte_spinlock_lock(&cmng->csl[cnt_type]);
7599 TAILQ_CONCAT(&cmng->counters[cnt_type],
7600 &pool->counters[query_gen], next);
7601 rte_spinlock_unlock(&cmng->csl[cnt_type]);
7604 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7605 pool->raw_hw = NULL;
7606 sh->cmng.pending_queries--;
7610 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7611 const struct flow_grp_info *grp_info,
7612 struct rte_flow_error *error)
7614 if (grp_info->transfer && grp_info->external &&
7615 grp_info->fdb_def_rule) {
7616 if (group == UINT32_MAX)
7617 return rte_flow_error_set
7619 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7621 "group index not supported");
7626 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7631 * Translate the rte_flow group index to HW table value.
7633 * If tunnel offload is disabled, all group ids converted to flow table
7634 * id using the standard method.
7635 * If tunnel offload is enabled, group id can be converted using the
7636 * standard or tunnel conversion method. Group conversion method
7637 * selection depends on flags in `grp_info` parameter:
7638 * - Internal (grp_info.external == 0) groups conversion uses the
7640 * - Group ids in JUMP action converted with the tunnel conversion.
7641 * - Group id in rule attribute conversion depends on a rule type and
7643 * ** non zero group attributes converted with the tunnel method
7644 * ** zero group attribute in non-tunnel rule is converted using the
7645 * standard method - there's only one root table
7646 * ** zero group attribute in steer tunnel rule is converted with the
7647 * standard method - single root table
7648 * ** zero group attribute in match tunnel rule is a special OvS
7649 * case: that value is used for portability reasons. That group
7650 * id is converted with the tunnel conversion method.
7655 * PMD tunnel offload object
7657 * rte_flow group index value.
7660 * @param[in] grp_info
7661 * flags used for conversion
7663 * Pointer to error structure.
7666 * 0 on success, a negative errno value otherwise and rte_errno is set.
7669 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7670 const struct mlx5_flow_tunnel *tunnel,
7671 uint32_t group, uint32_t *table,
7672 const struct flow_grp_info *grp_info,
7673 struct rte_flow_error *error)
7676 bool standard_translation;
7678 if (!grp_info->skip_scale && grp_info->external &&
7679 group < MLX5_MAX_TABLES_EXTERNAL)
7680 group *= MLX5_FLOW_TABLE_FACTOR;
7681 if (is_tunnel_offload_active(dev)) {
7682 standard_translation = !grp_info->external ||
7683 grp_info->std_tbl_fix;
7685 standard_translation = true;
7688 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7689 dev->data->port_id, group, grp_info->transfer,
7690 grp_info->external, grp_info->fdb_def_rule,
7691 standard_translation ? "STANDARD" : "TUNNEL");
7692 if (standard_translation)
7693 ret = flow_group_to_table(dev->data->port_id, group, table,
7696 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7703 * Discover availability of metadata reg_c's.
7705 * Iteratively use test flows to check availability.
7708 * Pointer to the Ethernet device structure.
7711 * 0 on success, a negative errno value otherwise and rte_errno is set.
7714 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7716 struct mlx5_priv *priv = dev->data->dev_private;
7717 struct mlx5_dev_config *config = &priv->config;
7718 enum modify_reg idx;
7721 /* reg_c[0] and reg_c[1] are reserved. */
7722 config->flow_mreg_c[n++] = REG_C_0;
7723 config->flow_mreg_c[n++] = REG_C_1;
7724 /* Discover availability of other reg_c's. */
7725 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7726 struct rte_flow_attr attr = {
7727 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7728 .priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7731 struct rte_flow_item items[] = {
7733 .type = RTE_FLOW_ITEM_TYPE_END,
7736 struct rte_flow_action actions[] = {
7738 .type = (enum rte_flow_action_type)
7739 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7740 .conf = &(struct mlx5_flow_action_copy_mreg){
7746 .type = RTE_FLOW_ACTION_TYPE_JUMP,
7747 .conf = &(struct rte_flow_action_jump){
7748 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7752 .type = RTE_FLOW_ACTION_TYPE_END,
7756 struct rte_flow *flow;
7757 struct rte_flow_error error;
7759 if (!config->dv_flow_en)
7761 /* Create internal flow, validation skips copy action. */
7762 flow_idx = flow_list_create(dev, NULL, &attr, items,
7763 actions, false, &error);
7764 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7768 config->flow_mreg_c[n++] = idx;
7769 flow_list_destroy(dev, NULL, flow_idx);
7771 for (; n < MLX5_MREG_C_NUM; ++n)
7772 config->flow_mreg_c[n] = REG_NON;
7777 * Dump flow raw hw data to file
7780 * The pointer to Ethernet device.
7782 * A pointer to a file for output.
7784 * Perform verbose error reporting if not NULL. PMDs initialize this
7785 * structure in case of error only.
7787 * 0 on success, a nagative value otherwise.
7790 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
7792 struct rte_flow_error *error __rte_unused)
7794 struct mlx5_priv *priv = dev->data->dev_private;
7795 struct mlx5_dev_ctx_shared *sh = priv->sh;
7796 uint32_t handle_idx;
7798 struct mlx5_flow_handle *dh;
7799 struct rte_flow *flow;
7801 if (!priv->config.dv_flow_en) {
7802 if (fputs("device dv flow disabled\n", file) <= 0)
7809 return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
7811 sh->tx_domain, file);
7813 flow = mlx5_ipool_get(priv->sh->ipool
7814 [MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
7818 handle_idx = flow->dev_handles;
7819 while (handle_idx) {
7820 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
7825 ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
7830 handle_idx = dh->next.next;
7836 * Get aged-out flows.
7839 * Pointer to the Ethernet device structure.
7840 * @param[in] context
7841 * The address of an array of pointers to the aged-out flows contexts.
7842 * @param[in] nb_countexts
7843 * The length of context array pointers.
7845 * Perform verbose error reporting if not NULL. Initialized in case of
7849 * how many contexts get in success, otherwise negative errno value.
7850 * if nb_contexts is 0, return the amount of all aged contexts.
7851 * if nb_contexts is not 0 , return the amount of aged flows reported
7852 * in the context array.
7855 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7856 uint32_t nb_contexts, struct rte_flow_error *error)
7858 const struct mlx5_flow_driver_ops *fops;
7859 struct rte_flow_attr attr = { .transfer = 0 };
7861 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7862 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7863 return fops->get_aged_flows(dev, contexts, nb_contexts,
7867 "port %u get aged flows is not supported.",
7868 dev->data->port_id);
7872 /* Wrapper for driver action_validate op callback */
7874 flow_drv_action_validate(struct rte_eth_dev *dev,
7875 const struct rte_flow_indir_action_conf *conf,
7876 const struct rte_flow_action *action,
7877 const struct mlx5_flow_driver_ops *fops,
7878 struct rte_flow_error *error)
7880 static const char err_msg[] = "indirect action validation unsupported";
7882 if (!fops->action_validate) {
7883 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7884 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7888 return fops->action_validate(dev, conf, action, error);
7892 * Destroys the shared action by handle.
7895 * Pointer to Ethernet device structure.
7897 * Handle for the indirect action object to be destroyed.
7899 * Perform verbose error reporting if not NULL. PMDs initialize this
7900 * structure in case of error only.
7903 * 0 on success, a negative errno value otherwise and rte_errno is set.
7905 * @note: wrapper for driver action_create op callback.
7908 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
7909 struct rte_flow_action_handle *handle,
7910 struct rte_flow_error *error)
7912 static const char err_msg[] = "indirect action destruction unsupported";
7913 struct rte_flow_attr attr = { .transfer = 0 };
7914 const struct mlx5_flow_driver_ops *fops =
7915 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7917 if (!fops->action_destroy) {
7918 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7919 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7923 return fops->action_destroy(dev, handle, error);
7926 /* Wrapper for driver action_destroy op callback */
7928 flow_drv_action_update(struct rte_eth_dev *dev,
7929 struct rte_flow_action_handle *handle,
7931 const struct mlx5_flow_driver_ops *fops,
7932 struct rte_flow_error *error)
7934 static const char err_msg[] = "indirect action update unsupported";
7936 if (!fops->action_update) {
7937 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7938 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7942 return fops->action_update(dev, handle, update, error);
7945 /* Wrapper for driver action_destroy op callback */
7947 flow_drv_action_query(struct rte_eth_dev *dev,
7948 const struct rte_flow_action_handle *handle,
7950 const struct mlx5_flow_driver_ops *fops,
7951 struct rte_flow_error *error)
7953 static const char err_msg[] = "indirect action query unsupported";
7955 if (!fops->action_query) {
7956 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7957 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7961 return fops->action_query(dev, handle, data, error);
7965 * Create indirect action for reuse in multiple flow rules.
7968 * Pointer to Ethernet device structure.
7970 * Pointer to indirect action object configuration.
7972 * Action configuration for indirect action object creation.
7974 * Perform verbose error reporting if not NULL. PMDs initialize this
7975 * structure in case of error only.
7977 * A valid handle in case of success, NULL otherwise and rte_errno is set.
7979 static struct rte_flow_action_handle *
7980 mlx5_action_handle_create(struct rte_eth_dev *dev,
7981 const struct rte_flow_indir_action_conf *conf,
7982 const struct rte_flow_action *action,
7983 struct rte_flow_error *error)
7985 static const char err_msg[] = "indirect action creation unsupported";
7986 struct rte_flow_attr attr = { .transfer = 0 };
7987 const struct mlx5_flow_driver_ops *fops =
7988 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7990 if (flow_drv_action_validate(dev, conf, action, fops, error))
7992 if (!fops->action_create) {
7993 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7994 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7998 return fops->action_create(dev, conf, action, error);
8002 * Updates inplace the indirect action configuration pointed by *handle*
8003 * with the configuration provided as *update* argument.
8004 * The update of the indirect action configuration effects all flow rules
8005 * reusing the action via handle.
8008 * Pointer to Ethernet device structure.
8010 * Handle for the indirect action to be updated.
8012 * Action specification used to modify the action pointed by handle.
8013 * *update* could be of same type with the action pointed by the *handle*
8014 * handle argument, or some other structures like a wrapper, depending on
8015 * the indirect action type.
8017 * Perform verbose error reporting if not NULL. PMDs initialize this
8018 * structure in case of error only.
8021 * 0 on success, a negative errno value otherwise and rte_errno is set.
8024 mlx5_action_handle_update(struct rte_eth_dev *dev,
8025 struct rte_flow_action_handle *handle,
8027 struct rte_flow_error *error)
8029 struct rte_flow_attr attr = { .transfer = 0 };
8030 const struct mlx5_flow_driver_ops *fops =
8031 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8034 ret = flow_drv_action_validate(dev, NULL,
8035 (const struct rte_flow_action *)update, fops, error);
8038 return flow_drv_action_update(dev, handle, update, fops,
8043 * Query the indirect action by handle.
8045 * This function allows retrieving action-specific data such as counters.
8046 * Data is gathered by special action which may be present/referenced in
8047 * more than one flow rule definition.
8049 * see @RTE_FLOW_ACTION_TYPE_COUNT
8052 * Pointer to Ethernet device structure.
8054 * Handle for the indirect action to query.
8055 * @param[in, out] data
8056 * Pointer to storage for the associated query data type.
8058 * Perform verbose error reporting if not NULL. PMDs initialize this
8059 * structure in case of error only.
8062 * 0 on success, a negative errno value otherwise and rte_errno is set.
8065 mlx5_action_handle_query(struct rte_eth_dev *dev,
8066 const struct rte_flow_action_handle *handle,
8068 struct rte_flow_error *error)
8070 struct rte_flow_attr attr = { .transfer = 0 };
8071 const struct mlx5_flow_driver_ops *fops =
8072 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8074 return flow_drv_action_query(dev, handle, data, fops, error);
8078 * Destroy all indirect actions (shared RSS).
8081 * Pointer to Ethernet device.
8084 * 0 on success, a negative errno value otherwise and rte_errno is set.
8087 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8089 struct rte_flow_error error;
8090 struct mlx5_priv *priv = dev->data->dev_private;
8091 struct mlx5_shared_action_rss *shared_rss;
8095 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8096 priv->rss_shared_actions, idx, shared_rss, next) {
8097 ret |= mlx5_action_handle_destroy(dev,
8098 (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8103 #ifndef HAVE_MLX5DV_DR
8104 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8106 #define MLX5_DOMAIN_SYNC_FLOW \
8107 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8110 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8112 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8113 const struct mlx5_flow_driver_ops *fops;
8115 struct rte_flow_attr attr = { .transfer = 0 };
8117 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8118 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8125 * tunnel offload functionalilty is defined for DV environment only
8127 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8129 union tunnel_offload_mark {
8132 uint32_t app_reserve:8;
8133 uint32_t table_id:15;
8134 uint32_t transfer:1;
8135 uint32_t _unused_:8;
8140 mlx5_access_tunnel_offload_db
8141 (struct rte_eth_dev *dev,
8142 bool (*match)(struct rte_eth_dev *,
8143 struct mlx5_flow_tunnel *, const void *),
8144 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8145 void (*miss)(struct rte_eth_dev *, void *),
8146 void *ctx, bool lock_op);
8149 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8150 struct rte_flow *flow,
8151 const struct rte_flow_attr *attr,
8152 const struct rte_flow_action *app_actions,
8154 struct tunnel_default_miss_ctx *ctx,
8155 struct rte_flow_error *error)
8157 struct mlx5_priv *priv = dev->data->dev_private;
8158 struct mlx5_flow *dev_flow;
8159 struct rte_flow_attr miss_attr = *attr;
8160 const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
8161 const struct rte_flow_item miss_items[2] = {
8163 .type = RTE_FLOW_ITEM_TYPE_ETH,
8169 .type = RTE_FLOW_ITEM_TYPE_END,
8175 union tunnel_offload_mark mark_id;
8176 struct rte_flow_action_mark miss_mark;
8177 struct rte_flow_action miss_actions[3] = {
8178 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8179 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
8181 const struct rte_flow_action_jump *jump_data;
8182 uint32_t i, flow_table = 0; /* prevent compilation warning */
8183 struct flow_grp_info grp_info = {
8185 .transfer = attr->transfer,
8186 .fdb_def_rule = !!priv->fdb_def_rule,
8191 if (!attr->transfer) {
8194 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8195 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8196 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8199 return rte_flow_error_set
8201 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8202 NULL, "invalid default miss RSS");
8203 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8204 ctx->action_rss.level = 0,
8205 ctx->action_rss.types = priv->rss_conf.rss_hf,
8206 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8207 ctx->action_rss.queue_num = priv->reta_idx_n,
8208 ctx->action_rss.key = priv->rss_conf.rss_key,
8209 ctx->action_rss.queue = ctx->queue;
8210 if (!priv->reta_idx_n || !priv->rxqs_n)
8211 return rte_flow_error_set
8213 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8214 NULL, "invalid port configuration");
8215 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8216 ctx->action_rss.types = 0;
8217 for (i = 0; i != priv->reta_idx_n; ++i)
8218 ctx->queue[i] = (*priv->reta_idx)[i];
8220 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8221 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8223 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8224 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8225 jump_data = app_actions->conf;
8226 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8227 miss_attr.group = jump_data->group;
8228 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8229 &flow_table, &grp_info, error);
8231 return rte_flow_error_set(error, EINVAL,
8232 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8233 NULL, "invalid tunnel id");
8234 mark_id.app_reserve = 0;
8235 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8236 mark_id.transfer = !!attr->transfer;
8237 mark_id._unused_ = 0;
8238 miss_mark.id = mark_id.val;
8239 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8240 miss_items, miss_actions, flow_idx, error);
8243 dev_flow->flow = flow;
8244 dev_flow->external = true;
8245 dev_flow->tunnel = tunnel;
8246 /* Subflow object was created, we must include one in the list. */
8247 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8248 dev_flow->handle, next);
8250 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8251 dev->data->port_id, tunnel->app_tunnel.type,
8252 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8253 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8254 miss_actions, error);
8256 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8262 static const struct mlx5_flow_tbl_data_entry *
8263 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8265 struct mlx5_priv *priv = dev->data->dev_private;
8266 struct mlx5_dev_ctx_shared *sh = priv->sh;
8267 struct mlx5_hlist_entry *he;
8268 union tunnel_offload_mark mbits = { .val = mark };
8269 union mlx5_flow_tbl_key table_key = {
8271 .level = tunnel_id_to_flow_tbl(mbits.table_id),
8275 .is_fdb = !!mbits.transfer,
8279 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
8281 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8285 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
8286 struct mlx5_hlist_entry *entry)
8288 struct mlx5_dev_ctx_shared *sh = list->ctx;
8289 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8291 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8292 tunnel_flow_tbl_to_id(tte->flow_table));
8297 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
8298 struct mlx5_hlist_entry *entry,
8299 uint64_t key, void *cb_ctx __rte_unused)
8301 union tunnel_tbl_key tbl = {
8304 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8306 return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8309 static struct mlx5_hlist_entry *
8310 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
8311 void *ctx __rte_unused)
8313 struct mlx5_dev_ctx_shared *sh = list->ctx;
8314 struct tunnel_tbl_entry *tte;
8315 union tunnel_tbl_key tbl = {
8319 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8324 mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8326 if (tte->flow_table >= MLX5_MAX_TABLES) {
8327 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8329 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8332 } else if (!tte->flow_table) {
8335 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8336 tte->tunnel_id = tbl.tunnel_id;
8337 tte->group = tbl.group;
8346 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8347 const struct mlx5_flow_tunnel *tunnel,
8348 uint32_t group, uint32_t *table,
8349 struct rte_flow_error *error)
8351 struct mlx5_hlist_entry *he;
8352 struct tunnel_tbl_entry *tte;
8353 union tunnel_tbl_key key = {
8354 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8357 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8358 struct mlx5_hlist *group_hash;
8360 group_hash = tunnel ? tunnel->groups : thub->groups;
8361 he = mlx5_hlist_register(group_hash, key.val, NULL);
8363 return rte_flow_error_set(error, EINVAL,
8364 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8366 "tunnel group index not supported");
8367 tte = container_of(he, typeof(*tte), hash);
8368 *table = tte->flow_table;
8369 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8370 dev->data->port_id, key.tunnel_id, group, *table);
8375 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8376 struct mlx5_flow_tunnel *tunnel)
8378 struct mlx5_priv *priv = dev->data->dev_private;
8379 struct mlx5_indexed_pool *ipool;
8381 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8382 dev->data->port_id, tunnel->tunnel_id);
8383 LIST_REMOVE(tunnel, chain);
8384 mlx5_hlist_destroy(tunnel->groups);
8385 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8386 mlx5_ipool_free(ipool, tunnel->tunnel_id);
8390 mlx5_access_tunnel_offload_db
8391 (struct rte_eth_dev *dev,
8392 bool (*match)(struct rte_eth_dev *,
8393 struct mlx5_flow_tunnel *, const void *),
8394 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8395 void (*miss)(struct rte_eth_dev *, void *),
8396 void *ctx, bool lock_op)
8398 bool verdict = false;
8399 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8400 struct mlx5_flow_tunnel *tunnel;
8402 rte_spinlock_lock(&thub->sl);
8403 LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8404 verdict = match(dev, tunnel, (const void *)ctx);
8409 rte_spinlock_unlock(&thub->sl);
8411 hit(dev, tunnel, ctx);
8412 if (!verdict && miss)
8415 rte_spinlock_unlock(&thub->sl);
8420 struct tunnel_db_find_tunnel_id_ctx {
8422 struct mlx5_flow_tunnel *tunnel;
8426 find_tunnel_id_match(struct rte_eth_dev *dev,
8427 struct mlx5_flow_tunnel *tunnel, const void *x)
8429 const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8432 return tunnel->tunnel_id == ctx->tunnel_id;
8436 find_tunnel_id_hit(struct rte_eth_dev *dev,
8437 struct mlx5_flow_tunnel *tunnel, void *x)
8439 struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8441 ctx->tunnel = tunnel;
8444 static struct mlx5_flow_tunnel *
8445 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8447 struct tunnel_db_find_tunnel_id_ctx ctx = {
8451 mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8452 find_tunnel_id_hit, NULL, &ctx, true);
8457 static struct mlx5_flow_tunnel *
8458 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8459 const struct rte_flow_tunnel *app_tunnel)
8461 struct mlx5_priv *priv = dev->data->dev_private;
8462 struct mlx5_indexed_pool *ipool;
8463 struct mlx5_flow_tunnel *tunnel;
8466 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8467 tunnel = mlx5_ipool_zmalloc(ipool, &id);
8470 if (id >= MLX5_MAX_TUNNELS) {
8471 mlx5_ipool_free(ipool, id);
8472 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8475 tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
8476 mlx5_flow_tunnel_grp2tbl_create_cb,
8477 mlx5_flow_tunnel_grp2tbl_match_cb,
8478 mlx5_flow_tunnel_grp2tbl_remove_cb);
8479 if (!tunnel->groups) {
8480 mlx5_ipool_free(ipool, id);
8483 tunnel->groups->ctx = priv->sh;
8484 /* initiate new PMD tunnel */
8485 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8486 tunnel->tunnel_id = id;
8487 tunnel->action.type = (typeof(tunnel->action.type))
8488 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8489 tunnel->action.conf = tunnel;
8490 tunnel->item.type = (typeof(tunnel->item.type))
8491 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8492 tunnel->item.spec = tunnel;
8493 tunnel->item.last = NULL;
8494 tunnel->item.mask = NULL;
8496 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8497 dev->data->port_id, tunnel->tunnel_id);
8502 struct tunnel_db_get_tunnel_ctx {
8503 const struct rte_flow_tunnel *app_tunnel;
8504 struct mlx5_flow_tunnel *tunnel;
8507 static bool get_tunnel_match(struct rte_eth_dev *dev,
8508 struct mlx5_flow_tunnel *tunnel, const void *x)
8510 const struct tunnel_db_get_tunnel_ctx *ctx = x;
8513 return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
8514 sizeof(*ctx->app_tunnel));
8517 static void get_tunnel_hit(struct rte_eth_dev *dev,
8518 struct mlx5_flow_tunnel *tunnel, void *x)
8520 /* called under tunnel spinlock protection */
8521 struct tunnel_db_get_tunnel_ctx *ctx = x;
8525 ctx->tunnel = tunnel;
8528 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8530 /* called under tunnel spinlock protection */
8531 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8532 struct tunnel_db_get_tunnel_ctx *ctx = x;
8534 rte_spinlock_unlock(&thub->sl);
8535 ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8536 rte_spinlock_lock(&thub->sl);
8538 ctx->tunnel->refctn = 1;
8539 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8545 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8546 const struct rte_flow_tunnel *app_tunnel,
8547 struct mlx5_flow_tunnel **tunnel)
8549 struct tunnel_db_get_tunnel_ctx ctx = {
8550 .app_tunnel = app_tunnel,
8553 mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8554 get_tunnel_miss, &ctx, true);
8555 *tunnel = ctx.tunnel;
8556 return ctx.tunnel ? 0 : -ENOMEM;
8559 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8561 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8565 if (!LIST_EMPTY(&thub->tunnels))
8566 DRV_LOG(WARNING, "port %u tunnels present", port_id);
8567 mlx5_hlist_destroy(thub->groups);
8571 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8574 struct mlx5_flow_tunnel_hub *thub;
8576 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8580 LIST_INIT(&thub->tunnels);
8581 rte_spinlock_init(&thub->sl);
8582 thub->groups = mlx5_hlist_create("flow groups",
8583 rte_align32pow2(MLX5_MAX_TABLES), 0,
8584 0, mlx5_flow_tunnel_grp2tbl_create_cb,
8585 mlx5_flow_tunnel_grp2tbl_match_cb,
8586 mlx5_flow_tunnel_grp2tbl_remove_cb);
8587 if (!thub->groups) {
8591 thub->groups->ctx = sh;
8592 sh->tunnel_hub = thub;
8598 mlx5_hlist_destroy(thub->groups);
8605 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
8606 struct rte_flow_tunnel *tunnel,
8607 const char *err_msg)
8610 if (!is_tunnel_offload_active(dev)) {
8611 err_msg = "tunnel offload was not activated";
8613 } else if (!tunnel) {
8614 err_msg = "no application tunnel";
8618 switch (tunnel->type) {
8620 err_msg = "unsupported tunnel type";
8622 case RTE_FLOW_ITEM_TYPE_VXLAN:
8631 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
8632 struct rte_flow_tunnel *app_tunnel,
8633 struct rte_flow_action **actions,
8634 uint32_t *num_of_actions,
8635 struct rte_flow_error *error)
8638 struct mlx5_flow_tunnel *tunnel;
8639 const char *err_msg = NULL;
8640 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8643 return rte_flow_error_set(error, EINVAL,
8644 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8646 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8648 return rte_flow_error_set(error, ret,
8649 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8650 "failed to initialize pmd tunnel");
8652 *actions = &tunnel->action;
8653 *num_of_actions = 1;
8658 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
8659 struct rte_flow_tunnel *app_tunnel,
8660 struct rte_flow_item **items,
8661 uint32_t *num_of_items,
8662 struct rte_flow_error *error)
8665 struct mlx5_flow_tunnel *tunnel;
8666 const char *err_msg = NULL;
8667 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8670 return rte_flow_error_set(error, EINVAL,
8671 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8673 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8675 return rte_flow_error_set(error, ret,
8676 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8677 "failed to initialize pmd tunnel");
8679 *items = &tunnel->item;
8684 struct tunnel_db_element_release_ctx {
8685 struct rte_flow_item *items;
8686 struct rte_flow_action *actions;
8687 uint32_t num_elements;
8688 struct rte_flow_error *error;
8693 tunnel_element_release_match(struct rte_eth_dev *dev,
8694 struct mlx5_flow_tunnel *tunnel, const void *x)
8696 const struct tunnel_db_element_release_ctx *ctx = x;
8699 if (ctx->num_elements != 1)
8701 else if (ctx->items)
8702 return ctx->items == &tunnel->item;
8703 else if (ctx->actions)
8704 return ctx->actions == &tunnel->action;
8710 tunnel_element_release_hit(struct rte_eth_dev *dev,
8711 struct mlx5_flow_tunnel *tunnel, void *x)
8713 struct tunnel_db_element_release_ctx *ctx = x;
8715 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8716 mlx5_flow_tunnel_free(dev, tunnel);
8720 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8722 struct tunnel_db_element_release_ctx *ctx = x;
8724 ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8725 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8726 "invalid argument");
8730 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8731 struct rte_flow_item *pmd_items,
8732 uint32_t num_items, struct rte_flow_error *err)
8734 struct tunnel_db_element_release_ctx ctx = {
8737 .num_elements = num_items,
8741 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8742 tunnel_element_release_hit,
8743 tunnel_element_release_miss, &ctx, false);
8749 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8750 struct rte_flow_action *pmd_actions,
8751 uint32_t num_actions, struct rte_flow_error *err)
8753 struct tunnel_db_element_release_ctx ctx = {
8755 .actions = pmd_actions,
8756 .num_elements = num_actions,
8760 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8761 tunnel_element_release_hit,
8762 tunnel_element_release_miss, &ctx, false);
8768 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8770 struct rte_flow_restore_info *info,
8771 struct rte_flow_error *err)
8773 uint64_t ol_flags = m->ol_flags;
8774 const struct mlx5_flow_tbl_data_entry *tble;
8775 const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8777 if (!is_tunnel_offload_active(dev)) {
8782 if ((ol_flags & mask) != mask)
8784 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8786 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8787 dev->data->port_id, m->hash.fdir.hi);
8790 MLX5_ASSERT(tble->tunnel);
8791 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8792 info->group_id = tble->group_id;
8793 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8794 RTE_FLOW_RESTORE_INFO_GROUP_ID |
8795 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8800 return rte_flow_error_set(err, EINVAL,
8801 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8802 "failed to get restore info");
8805 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8807 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8808 __rte_unused struct rte_flow_tunnel *app_tunnel,
8809 __rte_unused struct rte_flow_action **actions,
8810 __rte_unused uint32_t *num_of_actions,
8811 __rte_unused struct rte_flow_error *error)
8817 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8818 __rte_unused struct rte_flow_tunnel *app_tunnel,
8819 __rte_unused struct rte_flow_item **items,
8820 __rte_unused uint32_t *num_of_items,
8821 __rte_unused struct rte_flow_error *error)
8827 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8828 __rte_unused struct rte_flow_item *pmd_items,
8829 __rte_unused uint32_t num_items,
8830 __rte_unused struct rte_flow_error *err)
8836 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8837 __rte_unused struct rte_flow_action *pmd_action,
8838 __rte_unused uint32_t num_actions,
8839 __rte_unused struct rte_flow_error *err)
8845 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8846 __rte_unused struct rte_mbuf *m,
8847 __rte_unused struct rte_flow_restore_info *i,
8848 __rte_unused struct rte_flow_error *err)
8854 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8855 __rte_unused struct rte_flow *flow,
8856 __rte_unused const struct rte_flow_attr *attr,
8857 __rte_unused const struct rte_flow_action *actions,
8858 __rte_unused uint32_t flow_idx,
8859 __rte_unused struct tunnel_default_miss_ctx *ctx,
8860 __rte_unused struct rte_flow_error *error)
8865 static struct mlx5_flow_tunnel *
8866 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8867 __rte_unused uint32_t id)
8873 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8874 __rte_unused struct mlx5_flow_tunnel *tunnel)
8879 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8880 __rte_unused const struct mlx5_flow_tunnel *t,
8881 __rte_unused uint32_t group,
8882 __rte_unused uint32_t *table,
8883 struct rte_flow_error *error)
8885 return rte_flow_error_set(error, ENOTSUP,
8886 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8887 "tunnel offload requires DV support");
8891 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8892 __rte_unused uint16_t port_id)
8895 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */