1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
6 #include <netinet/in.h>
13 #include <rte_common.h>
14 #include <rte_ether.h>
15 #include <ethdev_driver.h>
16 #include <rte_eal_paging.h>
18 #include <rte_cycles.h>
19 #include <rte_flow_driver.h>
20 #include <rte_malloc.h>
23 #include <mlx5_glue.h>
24 #include <mlx5_devx_cmds.h>
26 #include <mlx5_malloc.h>
28 #include "mlx5_defs.h"
30 #include "mlx5_flow.h"
31 #include "mlx5_flow_os.h"
32 #include "mlx5_rxtx.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;
388 /* no expanded flows but we have missed item, create one rule for it */
389 if (buf->entries == 1 && missed != 0) {
391 lsize += elt * sizeof(*item) + user_pattern_size;
393 buf->entry[buf->entries].priority = 1;
394 buf->entry[buf->entries].pattern = addr;
396 flow_items[0].type = missed_item.type;
397 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
398 rte_memcpy(addr, buf->entry[0].pattern,
400 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
401 rte_memcpy(addr, flow_items, elt * sizeof(*item));
407 enum mlx5_expansion {
409 MLX5_EXPANSION_ROOT_OUTER,
410 MLX5_EXPANSION_ROOT_ETH_VLAN,
411 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
412 MLX5_EXPANSION_OUTER_ETH,
413 MLX5_EXPANSION_OUTER_ETH_VLAN,
414 MLX5_EXPANSION_OUTER_VLAN,
415 MLX5_EXPANSION_OUTER_IPV4,
416 MLX5_EXPANSION_OUTER_IPV4_UDP,
417 MLX5_EXPANSION_OUTER_IPV4_TCP,
418 MLX5_EXPANSION_OUTER_IPV6,
419 MLX5_EXPANSION_OUTER_IPV6_UDP,
420 MLX5_EXPANSION_OUTER_IPV6_TCP,
421 MLX5_EXPANSION_VXLAN,
422 MLX5_EXPANSION_VXLAN_GPE,
426 MLX5_EXPANSION_ETH_VLAN,
429 MLX5_EXPANSION_IPV4_UDP,
430 MLX5_EXPANSION_IPV4_TCP,
432 MLX5_EXPANSION_IPV6_UDP,
433 MLX5_EXPANSION_IPV6_TCP,
436 /** Supported expansion of items. */
437 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
438 [MLX5_EXPANSION_ROOT] = {
439 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
441 MLX5_EXPANSION_IPV6),
442 .type = RTE_FLOW_ITEM_TYPE_END,
444 [MLX5_EXPANSION_ROOT_OUTER] = {
445 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
446 MLX5_EXPANSION_OUTER_IPV4,
447 MLX5_EXPANSION_OUTER_IPV6),
448 .type = RTE_FLOW_ITEM_TYPE_END,
450 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
451 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
452 .type = RTE_FLOW_ITEM_TYPE_END,
454 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
455 .next = MLX5_FLOW_EXPAND_RSS_NEXT
456 (MLX5_EXPANSION_OUTER_ETH_VLAN),
457 .type = RTE_FLOW_ITEM_TYPE_END,
459 [MLX5_EXPANSION_OUTER_ETH] = {
460 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
461 MLX5_EXPANSION_OUTER_IPV6,
462 MLX5_EXPANSION_MPLS),
463 .type = RTE_FLOW_ITEM_TYPE_ETH,
466 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
467 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
468 .type = RTE_FLOW_ITEM_TYPE_ETH,
471 [MLX5_EXPANSION_OUTER_VLAN] = {
472 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
473 MLX5_EXPANSION_OUTER_IPV6),
474 .type = RTE_FLOW_ITEM_TYPE_VLAN,
476 [MLX5_EXPANSION_OUTER_IPV4] = {
477 .next = MLX5_FLOW_EXPAND_RSS_NEXT
478 (MLX5_EXPANSION_OUTER_IPV4_UDP,
479 MLX5_EXPANSION_OUTER_IPV4_TCP,
482 MLX5_EXPANSION_IPV6),
483 .type = RTE_FLOW_ITEM_TYPE_IPV4,
484 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
485 ETH_RSS_NONFRAG_IPV4_OTHER,
487 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
488 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
489 MLX5_EXPANSION_VXLAN_GPE),
490 .type = RTE_FLOW_ITEM_TYPE_UDP,
491 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
493 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
494 .type = RTE_FLOW_ITEM_TYPE_TCP,
495 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
497 [MLX5_EXPANSION_OUTER_IPV6] = {
498 .next = MLX5_FLOW_EXPAND_RSS_NEXT
499 (MLX5_EXPANSION_OUTER_IPV6_UDP,
500 MLX5_EXPANSION_OUTER_IPV6_TCP,
502 MLX5_EXPANSION_IPV6),
503 .type = RTE_FLOW_ITEM_TYPE_IPV6,
504 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
505 ETH_RSS_NONFRAG_IPV6_OTHER,
507 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
508 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
509 MLX5_EXPANSION_VXLAN_GPE),
510 .type = RTE_FLOW_ITEM_TYPE_UDP,
511 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
513 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
514 .type = RTE_FLOW_ITEM_TYPE_TCP,
515 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
517 [MLX5_EXPANSION_VXLAN] = {
518 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
520 MLX5_EXPANSION_IPV6),
521 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
523 [MLX5_EXPANSION_VXLAN_GPE] = {
524 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
526 MLX5_EXPANSION_IPV6),
527 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
529 [MLX5_EXPANSION_GRE] = {
530 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
531 .type = RTE_FLOW_ITEM_TYPE_GRE,
533 [MLX5_EXPANSION_MPLS] = {
534 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
535 MLX5_EXPANSION_IPV6),
536 .type = RTE_FLOW_ITEM_TYPE_MPLS,
538 [MLX5_EXPANSION_ETH] = {
539 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
540 MLX5_EXPANSION_IPV6),
541 .type = RTE_FLOW_ITEM_TYPE_ETH,
543 [MLX5_EXPANSION_ETH_VLAN] = {
544 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
545 .type = RTE_FLOW_ITEM_TYPE_ETH,
547 [MLX5_EXPANSION_VLAN] = {
548 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
549 MLX5_EXPANSION_IPV6),
550 .type = RTE_FLOW_ITEM_TYPE_VLAN,
552 [MLX5_EXPANSION_IPV4] = {
553 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
554 MLX5_EXPANSION_IPV4_TCP),
555 .type = RTE_FLOW_ITEM_TYPE_IPV4,
556 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
557 ETH_RSS_NONFRAG_IPV4_OTHER,
559 [MLX5_EXPANSION_IPV4_UDP] = {
560 .type = RTE_FLOW_ITEM_TYPE_UDP,
561 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
563 [MLX5_EXPANSION_IPV4_TCP] = {
564 .type = RTE_FLOW_ITEM_TYPE_TCP,
565 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
567 [MLX5_EXPANSION_IPV6] = {
568 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
569 MLX5_EXPANSION_IPV6_TCP),
570 .type = RTE_FLOW_ITEM_TYPE_IPV6,
571 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
572 ETH_RSS_NONFRAG_IPV6_OTHER,
574 [MLX5_EXPANSION_IPV6_UDP] = {
575 .type = RTE_FLOW_ITEM_TYPE_UDP,
576 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
578 [MLX5_EXPANSION_IPV6_TCP] = {
579 .type = RTE_FLOW_ITEM_TYPE_TCP,
580 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
584 static struct rte_flow_shared_action *
585 mlx5_shared_action_create(struct rte_eth_dev *dev,
586 const struct rte_flow_shared_action_conf *conf,
587 const struct rte_flow_action *action,
588 struct rte_flow_error *error);
589 static int mlx5_shared_action_destroy
590 (struct rte_eth_dev *dev,
591 struct rte_flow_shared_action *shared_action,
592 struct rte_flow_error *error);
593 static int mlx5_shared_action_update
594 (struct rte_eth_dev *dev,
595 struct rte_flow_shared_action *shared_action,
596 const struct rte_flow_action *action,
597 struct rte_flow_error *error);
598 static int mlx5_shared_action_query
599 (struct rte_eth_dev *dev,
600 const struct rte_flow_shared_action *action,
602 struct rte_flow_error *error);
604 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
605 struct rte_flow_tunnel *app_tunnel,
606 struct rte_flow_action **actions,
607 uint32_t *num_of_actions,
608 struct rte_flow_error *error);
610 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
611 struct rte_flow_tunnel *app_tunnel,
612 struct rte_flow_item **items,
613 uint32_t *num_of_items,
614 struct rte_flow_error *error);
616 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
617 struct rte_flow_item *pmd_items,
618 uint32_t num_items, struct rte_flow_error *err);
620 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
621 struct rte_flow_action *pmd_actions,
622 uint32_t num_actions,
623 struct rte_flow_error *err);
625 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
627 struct rte_flow_restore_info *info,
628 struct rte_flow_error *err);
630 static const struct rte_flow_ops mlx5_flow_ops = {
631 .validate = mlx5_flow_validate,
632 .create = mlx5_flow_create,
633 .destroy = mlx5_flow_destroy,
634 .flush = mlx5_flow_flush,
635 .isolate = mlx5_flow_isolate,
636 .query = mlx5_flow_query,
637 .dev_dump = mlx5_flow_dev_dump,
638 .get_aged_flows = mlx5_flow_get_aged_flows,
639 .shared_action_create = mlx5_shared_action_create,
640 .shared_action_destroy = mlx5_shared_action_destroy,
641 .shared_action_update = mlx5_shared_action_update,
642 .shared_action_query = mlx5_shared_action_query,
643 .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
644 .tunnel_match = mlx5_flow_tunnel_match,
645 .tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
646 .tunnel_item_release = mlx5_flow_tunnel_item_release,
647 .get_restore_info = mlx5_flow_tunnel_get_restore_info,
650 /* Tunnel information. */
651 struct mlx5_flow_tunnel_info {
652 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
653 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
656 static struct mlx5_flow_tunnel_info tunnels_info[] = {
658 .tunnel = MLX5_FLOW_LAYER_VXLAN,
659 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
662 .tunnel = MLX5_FLOW_LAYER_GENEVE,
663 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
666 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
667 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
670 .tunnel = MLX5_FLOW_LAYER_GRE,
671 .ptype = RTE_PTYPE_TUNNEL_GRE,
674 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
675 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
678 .tunnel = MLX5_FLOW_LAYER_MPLS,
679 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
682 .tunnel = MLX5_FLOW_LAYER_NVGRE,
683 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
686 .tunnel = MLX5_FLOW_LAYER_IPIP,
687 .ptype = RTE_PTYPE_TUNNEL_IP,
690 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
691 .ptype = RTE_PTYPE_TUNNEL_IP,
694 .tunnel = MLX5_FLOW_LAYER_GTP,
695 .ptype = RTE_PTYPE_TUNNEL_GTPU,
702 * Translate tag ID to register.
705 * Pointer to the Ethernet device structure.
707 * The feature that request the register.
709 * The request register ID.
711 * Error description in case of any.
714 * The request register on success, a negative errno
715 * value otherwise and rte_errno is set.
718 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
719 enum mlx5_feature_name feature,
721 struct rte_flow_error *error)
723 struct mlx5_priv *priv = dev->data->dev_private;
724 struct mlx5_dev_config *config = &priv->config;
725 enum modify_reg start_reg;
726 bool skip_mtr_reg = false;
729 case MLX5_HAIRPIN_RX:
731 case MLX5_HAIRPIN_TX:
733 case MLX5_METADATA_RX:
734 switch (config->dv_xmeta_en) {
735 case MLX5_XMETA_MODE_LEGACY:
737 case MLX5_XMETA_MODE_META16:
739 case MLX5_XMETA_MODE_META32:
743 case MLX5_METADATA_TX:
745 case MLX5_METADATA_FDB:
746 switch (config->dv_xmeta_en) {
747 case MLX5_XMETA_MODE_LEGACY:
749 case MLX5_XMETA_MODE_META16:
751 case MLX5_XMETA_MODE_META32:
756 switch (config->dv_xmeta_en) {
757 case MLX5_XMETA_MODE_LEGACY:
759 case MLX5_XMETA_MODE_META16:
761 case MLX5_XMETA_MODE_META32:
767 * If meter color and flow match share one register, flow match
768 * should use the meter color register for match.
770 if (priv->mtr_reg_share)
771 return priv->mtr_color_reg;
773 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
776 case MLX5_ASO_FLOW_HIT: /* Both features use the same REG_C. */
777 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
778 return priv->mtr_color_reg;
781 * Metadata COPY_MARK register using is in meter suffix sub
782 * flow while with meter. It's safe to share the same register.
784 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
787 * If meter is enable, it will engage the register for color
788 * match and flow match. If meter color match is not using the
789 * REG_C_2, need to skip the REG_C_x be used by meter color
791 * If meter is disable, free to use all available registers.
793 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
794 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
795 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
796 if (id > (uint32_t)(REG_C_7 - start_reg))
797 return rte_flow_error_set(error, EINVAL,
798 RTE_FLOW_ERROR_TYPE_ITEM,
799 NULL, "invalid tag id");
800 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
801 return rte_flow_error_set(error, ENOTSUP,
802 RTE_FLOW_ERROR_TYPE_ITEM,
803 NULL, "unsupported tag id");
805 * This case means meter is using the REG_C_x great than 2.
806 * Take care not to conflict with meter color REG_C_x.
807 * If the available index REG_C_y >= REG_C_x, skip the
810 if (skip_mtr_reg && config->flow_mreg_c
811 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
812 if (id >= (uint32_t)(REG_C_7 - start_reg))
813 return rte_flow_error_set(error, EINVAL,
814 RTE_FLOW_ERROR_TYPE_ITEM,
815 NULL, "invalid tag id");
816 if (config->flow_mreg_c
817 [id + 1 + start_reg - REG_C_0] != REG_NON)
818 return config->flow_mreg_c
819 [id + 1 + start_reg - REG_C_0];
820 return rte_flow_error_set(error, ENOTSUP,
821 RTE_FLOW_ERROR_TYPE_ITEM,
822 NULL, "unsupported tag id");
824 return config->flow_mreg_c[id + start_reg - REG_C_0];
827 return rte_flow_error_set(error, EINVAL,
828 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
829 NULL, "invalid feature name");
833 * Check extensive flow metadata register support.
836 * Pointer to rte_eth_dev structure.
839 * True if device supports extensive flow metadata register, otherwise false.
842 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
844 struct mlx5_priv *priv = dev->data->dev_private;
845 struct mlx5_dev_config *config = &priv->config;
848 * Having available reg_c can be regarded inclusively as supporting
849 * extensive flow metadata register, which could mean,
850 * - metadata register copy action by modify header.
851 * - 16 modify header actions is supported.
852 * - reg_c's are preserved across different domain (FDB and NIC) on
853 * packet loopback by flow lookup miss.
855 return config->flow_mreg_c[2] != REG_NON;
859 * Verify the @p item specifications (spec, last, mask) are compatible with the
863 * Item specification.
865 * @p item->mask or flow default bit-masks.
866 * @param[in] nic_mask
867 * Bit-masks covering supported fields by the NIC to compare with user mask.
869 * Bit-masks size in bytes.
870 * @param[in] range_accepted
871 * True if range of values is accepted for specific fields, false otherwise.
873 * Pointer to error structure.
876 * 0 on success, a negative errno value otherwise and rte_errno is set.
879 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
881 const uint8_t *nic_mask,
884 struct rte_flow_error *error)
888 MLX5_ASSERT(nic_mask);
889 for (i = 0; i < size; ++i)
890 if ((nic_mask[i] | mask[i]) != nic_mask[i])
891 return rte_flow_error_set(error, ENOTSUP,
892 RTE_FLOW_ERROR_TYPE_ITEM,
894 "mask enables non supported"
896 if (!item->spec && (item->mask || item->last))
897 return rte_flow_error_set(error, EINVAL,
898 RTE_FLOW_ERROR_TYPE_ITEM, item,
899 "mask/last without a spec is not"
901 if (item->spec && item->last && !range_accepted) {
907 for (i = 0; i < size; ++i) {
908 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
909 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
911 ret = memcmp(spec, last, size);
913 return rte_flow_error_set(error, EINVAL,
914 RTE_FLOW_ERROR_TYPE_ITEM,
916 "range is not valid");
922 * Adjust the hash fields according to the @p flow information.
924 * @param[in] dev_flow.
925 * Pointer to the mlx5_flow.
927 * 1 when the hash field is for a tunnel item.
928 * @param[in] layer_types
930 * @param[in] hash_fields
934 * The hash fields that should be used.
937 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
938 int tunnel __rte_unused, uint64_t layer_types,
939 uint64_t hash_fields)
941 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
942 int rss_request_inner = rss_desc->level >= 2;
944 /* Check RSS hash level for tunnel. */
945 if (tunnel && rss_request_inner)
946 hash_fields |= IBV_RX_HASH_INNER;
947 else if (tunnel || rss_request_inner)
950 /* Check if requested layer matches RSS hash fields. */
951 if (!(rss_desc->types & layer_types))
957 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
958 * if several tunnel rules are used on this queue, the tunnel ptype will be
962 * Rx queue to update.
965 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
968 uint32_t tunnel_ptype = 0;
970 /* Look up for the ptype to use. */
971 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
972 if (!rxq_ctrl->flow_tunnels_n[i])
975 tunnel_ptype = tunnels_info[i].ptype;
981 rxq_ctrl->rxq.tunnel = tunnel_ptype;
985 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
989 * Pointer to the Ethernet device structure.
990 * @param[in] dev_handle
991 * Pointer to device flow handle structure.
994 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
995 struct mlx5_flow_handle *dev_handle)
997 struct mlx5_priv *priv = dev->data->dev_private;
998 const int mark = dev_handle->mark;
999 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1000 struct mlx5_ind_table_obj *ind_tbl = NULL;
1003 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1004 struct mlx5_hrxq *hrxq;
1006 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1007 dev_handle->rix_hrxq);
1009 ind_tbl = hrxq->ind_table;
1010 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1011 struct mlx5_shared_action_rss *shared_rss;
1013 shared_rss = mlx5_ipool_get
1014 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1015 dev_handle->rix_srss);
1017 ind_tbl = shared_rss->ind_tbl;
1021 for (i = 0; i != ind_tbl->queues_n; ++i) {
1022 int idx = ind_tbl->queues[i];
1023 struct mlx5_rxq_ctrl *rxq_ctrl =
1024 container_of((*priv->rxqs)[idx],
1025 struct mlx5_rxq_ctrl, rxq);
1028 * To support metadata register copy on Tx loopback,
1029 * this must be always enabled (metadata may arive
1030 * from other port - not from local flows only.
1032 if (priv->config.dv_flow_en &&
1033 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1034 mlx5_flow_ext_mreg_supported(dev)) {
1035 rxq_ctrl->rxq.mark = 1;
1036 rxq_ctrl->flow_mark_n = 1;
1038 rxq_ctrl->rxq.mark = 1;
1039 rxq_ctrl->flow_mark_n++;
1044 /* Increase the counter matching the flow. */
1045 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1046 if ((tunnels_info[j].tunnel &
1047 dev_handle->layers) ==
1048 tunnels_info[j].tunnel) {
1049 rxq_ctrl->flow_tunnels_n[j]++;
1053 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1059 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1062 * Pointer to the Ethernet device structure.
1064 * Pointer to flow structure.
1067 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1069 struct mlx5_priv *priv = dev->data->dev_private;
1070 uint32_t handle_idx;
1071 struct mlx5_flow_handle *dev_handle;
1073 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1074 handle_idx, dev_handle, next)
1075 flow_drv_rxq_flags_set(dev, dev_handle);
1079 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1080 * device flow if no other flow uses it with the same kind of request.
1083 * Pointer to Ethernet device.
1084 * @param[in] dev_handle
1085 * Pointer to the device flow handle structure.
1088 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1089 struct mlx5_flow_handle *dev_handle)
1091 struct mlx5_priv *priv = dev->data->dev_private;
1092 const int mark = dev_handle->mark;
1093 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1094 struct mlx5_ind_table_obj *ind_tbl = NULL;
1097 if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1098 struct mlx5_hrxq *hrxq;
1100 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1101 dev_handle->rix_hrxq);
1103 ind_tbl = hrxq->ind_table;
1104 } else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1105 struct mlx5_shared_action_rss *shared_rss;
1107 shared_rss = mlx5_ipool_get
1108 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1109 dev_handle->rix_srss);
1111 ind_tbl = shared_rss->ind_tbl;
1115 MLX5_ASSERT(dev->data->dev_started);
1116 for (i = 0; i != ind_tbl->queues_n; ++i) {
1117 int idx = ind_tbl->queues[i];
1118 struct mlx5_rxq_ctrl *rxq_ctrl =
1119 container_of((*priv->rxqs)[idx],
1120 struct mlx5_rxq_ctrl, rxq);
1122 if (priv->config.dv_flow_en &&
1123 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1124 mlx5_flow_ext_mreg_supported(dev)) {
1125 rxq_ctrl->rxq.mark = 1;
1126 rxq_ctrl->flow_mark_n = 1;
1128 rxq_ctrl->flow_mark_n--;
1129 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1134 /* Decrease the counter matching the flow. */
1135 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1136 if ((tunnels_info[j].tunnel &
1137 dev_handle->layers) ==
1138 tunnels_info[j].tunnel) {
1139 rxq_ctrl->flow_tunnels_n[j]--;
1143 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1149 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1150 * @p flow if no other flow uses it with the same kind of request.
1153 * Pointer to Ethernet device.
1155 * Pointer to the flow.
1158 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1160 struct mlx5_priv *priv = dev->data->dev_private;
1161 uint32_t handle_idx;
1162 struct mlx5_flow_handle *dev_handle;
1164 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1165 handle_idx, dev_handle, next)
1166 flow_drv_rxq_flags_trim(dev, dev_handle);
1170 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1173 * Pointer to Ethernet device.
1176 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1178 struct mlx5_priv *priv = dev->data->dev_private;
1181 for (i = 0; i != priv->rxqs_n; ++i) {
1182 struct mlx5_rxq_ctrl *rxq_ctrl;
1185 if (!(*priv->rxqs)[i])
1187 rxq_ctrl = container_of((*priv->rxqs)[i],
1188 struct mlx5_rxq_ctrl, rxq);
1189 rxq_ctrl->flow_mark_n = 0;
1190 rxq_ctrl->rxq.mark = 0;
1191 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1192 rxq_ctrl->flow_tunnels_n[j] = 0;
1193 rxq_ctrl->rxq.tunnel = 0;
1198 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1201 * Pointer to the Ethernet device structure.
1204 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1206 struct mlx5_priv *priv = dev->data->dev_private;
1207 struct mlx5_rxq_data *data;
1210 for (i = 0; i != priv->rxqs_n; ++i) {
1211 if (!(*priv->rxqs)[i])
1213 data = (*priv->rxqs)[i];
1214 if (!rte_flow_dynf_metadata_avail()) {
1215 data->dynf_meta = 0;
1216 data->flow_meta_mask = 0;
1217 data->flow_meta_offset = -1;
1219 data->dynf_meta = 1;
1220 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1221 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1227 * return a pointer to the desired action in the list of actions.
1229 * @param[in] actions
1230 * The list of actions to search the action in.
1232 * The action to find.
1235 * Pointer to the action in the list, if found. NULL otherwise.
1237 const struct rte_flow_action *
1238 mlx5_flow_find_action(const struct rte_flow_action *actions,
1239 enum rte_flow_action_type action)
1241 if (actions == NULL)
1243 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1244 if (actions->type == action)
1250 * Validate the flag action.
1252 * @param[in] action_flags
1253 * Bit-fields that holds the actions detected until now.
1255 * Attributes of flow that includes this action.
1257 * Pointer to error structure.
1260 * 0 on success, a negative errno value otherwise and rte_errno is set.
1263 mlx5_flow_validate_action_flag(uint64_t action_flags,
1264 const struct rte_flow_attr *attr,
1265 struct rte_flow_error *error)
1267 if (action_flags & MLX5_FLOW_ACTION_MARK)
1268 return rte_flow_error_set(error, EINVAL,
1269 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1270 "can't mark and flag in same flow");
1271 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1272 return rte_flow_error_set(error, EINVAL,
1273 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1275 " actions in same flow");
1277 return rte_flow_error_set(error, ENOTSUP,
1278 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1279 "flag action not supported for "
1285 * Validate the mark action.
1288 * Pointer to the queue action.
1289 * @param[in] action_flags
1290 * Bit-fields that holds the actions detected until now.
1292 * Attributes of flow that includes this action.
1294 * Pointer to error structure.
1297 * 0 on success, a negative errno value otherwise and rte_errno is set.
1300 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1301 uint64_t action_flags,
1302 const struct rte_flow_attr *attr,
1303 struct rte_flow_error *error)
1305 const struct rte_flow_action_mark *mark = action->conf;
1308 return rte_flow_error_set(error, EINVAL,
1309 RTE_FLOW_ERROR_TYPE_ACTION,
1311 "configuration cannot be null");
1312 if (mark->id >= MLX5_FLOW_MARK_MAX)
1313 return rte_flow_error_set(error, EINVAL,
1314 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1316 "mark id must in 0 <= id < "
1317 RTE_STR(MLX5_FLOW_MARK_MAX));
1318 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1319 return rte_flow_error_set(error, EINVAL,
1320 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1321 "can't flag and mark in same flow");
1322 if (action_flags & MLX5_FLOW_ACTION_MARK)
1323 return rte_flow_error_set(error, EINVAL,
1324 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1325 "can't have 2 mark actions in same"
1328 return rte_flow_error_set(error, ENOTSUP,
1329 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1330 "mark action not supported for "
1336 * Validate the drop action.
1338 * @param[in] action_flags
1339 * Bit-fields that holds the actions detected until now.
1341 * Attributes of flow that includes this action.
1343 * Pointer to error structure.
1346 * 0 on success, a negative errno value otherwise and rte_errno is set.
1349 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1350 const struct rte_flow_attr *attr,
1351 struct rte_flow_error *error)
1354 return rte_flow_error_set(error, ENOTSUP,
1355 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1356 "drop action not supported for "
1362 * Validate the queue action.
1365 * Pointer to the queue action.
1366 * @param[in] action_flags
1367 * Bit-fields that holds the actions detected until now.
1369 * Pointer to the Ethernet device structure.
1371 * Attributes of flow that includes this action.
1373 * Pointer to error structure.
1376 * 0 on success, a negative errno value otherwise and rte_errno is set.
1379 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1380 uint64_t action_flags,
1381 struct rte_eth_dev *dev,
1382 const struct rte_flow_attr *attr,
1383 struct rte_flow_error *error)
1385 struct mlx5_priv *priv = dev->data->dev_private;
1386 const struct rte_flow_action_queue *queue = action->conf;
1388 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1389 return rte_flow_error_set(error, EINVAL,
1390 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1391 "can't have 2 fate actions in"
1394 return rte_flow_error_set(error, EINVAL,
1395 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1396 NULL, "No Rx queues configured");
1397 if (queue->index >= priv->rxqs_n)
1398 return rte_flow_error_set(error, EINVAL,
1399 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1401 "queue index out of range");
1402 if (!(*priv->rxqs)[queue->index])
1403 return rte_flow_error_set(error, EINVAL,
1404 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1406 "queue is not configured");
1408 return rte_flow_error_set(error, ENOTSUP,
1409 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1410 "queue action not supported for "
1416 * Validate the rss action.
1419 * Pointer to the Ethernet device structure.
1421 * Pointer to the queue action.
1423 * Pointer to error structure.
1426 * 0 on success, a negative errno value otherwise and rte_errno is set.
1429 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1430 const struct rte_flow_action *action,
1431 struct rte_flow_error *error)
1433 struct mlx5_priv *priv = dev->data->dev_private;
1434 const struct rte_flow_action_rss *rss = action->conf;
1435 enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1438 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1439 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1440 return rte_flow_error_set(error, ENOTSUP,
1441 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1443 "RSS hash function not supported");
1444 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1449 return rte_flow_error_set(error, ENOTSUP,
1450 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1452 "tunnel RSS is not supported");
1453 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1454 if (rss->key_len == 0 && rss->key != NULL)
1455 return rte_flow_error_set(error, ENOTSUP,
1456 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1458 "RSS hash key length 0");
1459 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1460 return rte_flow_error_set(error, ENOTSUP,
1461 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1463 "RSS hash key too small");
1464 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1465 return rte_flow_error_set(error, ENOTSUP,
1466 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1468 "RSS hash key too large");
1469 if (rss->queue_num > priv->config.ind_table_max_size)
1470 return rte_flow_error_set(error, ENOTSUP,
1471 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1473 "number of queues too large");
1474 if (rss->types & MLX5_RSS_HF_MASK)
1475 return rte_flow_error_set(error, ENOTSUP,
1476 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1478 "some RSS protocols are not"
1480 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1481 !(rss->types & ETH_RSS_IP))
1482 return rte_flow_error_set(error, EINVAL,
1483 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1484 "L3 partial RSS requested but L3 RSS"
1485 " type not specified");
1486 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1487 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1488 return rte_flow_error_set(error, EINVAL,
1489 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1490 "L4 partial RSS requested but L4 RSS"
1491 " type not specified");
1493 return rte_flow_error_set(error, EINVAL,
1494 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1495 NULL, "No Rx queues configured");
1496 if (!rss->queue_num)
1497 return rte_flow_error_set(error, EINVAL,
1498 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1499 NULL, "No queues configured");
1500 for (i = 0; i != rss->queue_num; ++i) {
1501 struct mlx5_rxq_ctrl *rxq_ctrl;
1503 if (rss->queue[i] >= priv->rxqs_n)
1504 return rte_flow_error_set
1506 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1507 &rss->queue[i], "queue index out of range");
1508 if (!(*priv->rxqs)[rss->queue[i]])
1509 return rte_flow_error_set
1510 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1511 &rss->queue[i], "queue is not configured");
1512 rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1513 struct mlx5_rxq_ctrl, rxq);
1515 rxq_type = rxq_ctrl->type;
1516 if (rxq_type != rxq_ctrl->type)
1517 return rte_flow_error_set
1518 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1520 "combining hairpin and regular RSS queues is not supported");
1526 * Validate the rss action.
1529 * Pointer to the queue action.
1530 * @param[in] action_flags
1531 * Bit-fields that holds the actions detected until now.
1533 * Pointer to the Ethernet device structure.
1535 * Attributes of flow that includes this action.
1536 * @param[in] item_flags
1537 * Items that were detected.
1539 * Pointer to error structure.
1542 * 0 on success, a negative errno value otherwise and rte_errno is set.
1545 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1546 uint64_t action_flags,
1547 struct rte_eth_dev *dev,
1548 const struct rte_flow_attr *attr,
1549 uint64_t item_flags,
1550 struct rte_flow_error *error)
1552 const struct rte_flow_action_rss *rss = action->conf;
1553 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1556 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1557 return rte_flow_error_set(error, EINVAL,
1558 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1559 "can't have 2 fate actions"
1561 ret = mlx5_validate_action_rss(dev, action, error);
1565 return rte_flow_error_set(error, ENOTSUP,
1566 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1567 "rss action not supported for "
1569 if (rss->level > 1 && !tunnel)
1570 return rte_flow_error_set(error, EINVAL,
1571 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1572 "inner RSS is not supported for "
1573 "non-tunnel flows");
1574 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1575 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1576 return rte_flow_error_set(error, EINVAL,
1577 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1578 "RSS on eCPRI is not supported now");
1584 * Validate the default miss action.
1586 * @param[in] action_flags
1587 * Bit-fields that holds the actions detected until now.
1589 * Pointer to error structure.
1592 * 0 on success, a negative errno value otherwise and rte_errno is set.
1595 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1596 const struct rte_flow_attr *attr,
1597 struct rte_flow_error *error)
1599 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1600 return rte_flow_error_set(error, EINVAL,
1601 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1602 "can't have 2 fate actions in"
1605 return rte_flow_error_set(error, ENOTSUP,
1606 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1607 "default miss action not supported "
1610 return rte_flow_error_set(error, ENOTSUP,
1611 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1612 "only group 0 is supported");
1614 return rte_flow_error_set(error, ENOTSUP,
1615 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1616 NULL, "transfer is not supported");
1621 * Validate the count action.
1624 * Pointer to the Ethernet device structure.
1626 * Attributes of flow that includes this action.
1628 * Pointer to error structure.
1631 * 0 on success, a negative errno value otherwise and rte_errno is set.
1634 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1635 const struct rte_flow_attr *attr,
1636 struct rte_flow_error *error)
1639 return rte_flow_error_set(error, ENOTSUP,
1640 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1641 "count action not supported for "
1647 * Verify the @p attributes will be correctly understood by the NIC and store
1648 * them in the @p flow if everything is correct.
1651 * Pointer to the Ethernet device structure.
1652 * @param[in] attributes
1653 * Pointer to flow attributes
1655 * Pointer to error structure.
1658 * 0 on success, a negative errno value otherwise and rte_errno is set.
1661 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1662 const struct rte_flow_attr *attributes,
1663 struct rte_flow_error *error)
1665 struct mlx5_priv *priv = dev->data->dev_private;
1666 uint32_t priority_max = priv->config.flow_prio - 1;
1668 if (attributes->group)
1669 return rte_flow_error_set(error, ENOTSUP,
1670 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1671 NULL, "groups is not supported");
1672 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1673 attributes->priority >= priority_max)
1674 return rte_flow_error_set(error, ENOTSUP,
1675 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1676 NULL, "priority out of range");
1677 if (attributes->egress)
1678 return rte_flow_error_set(error, ENOTSUP,
1679 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1680 "egress is not supported");
1681 if (attributes->transfer && !priv->config.dv_esw_en)
1682 return rte_flow_error_set(error, ENOTSUP,
1683 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1684 NULL, "transfer is not supported");
1685 if (!attributes->ingress)
1686 return rte_flow_error_set(error, EINVAL,
1687 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1689 "ingress attribute is mandatory");
1694 * Validate ICMP6 item.
1697 * Item specification.
1698 * @param[in] item_flags
1699 * Bit-fields that holds the items detected until now.
1700 * @param[in] ext_vlan_sup
1701 * Whether extended VLAN features are supported or not.
1703 * Pointer to error structure.
1706 * 0 on success, a negative errno value otherwise and rte_errno is set.
1709 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1710 uint64_t item_flags,
1711 uint8_t target_protocol,
1712 struct rte_flow_error *error)
1714 const struct rte_flow_item_icmp6 *mask = item->mask;
1715 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1716 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1717 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1718 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1719 MLX5_FLOW_LAYER_OUTER_L4;
1722 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1723 return rte_flow_error_set(error, EINVAL,
1724 RTE_FLOW_ERROR_TYPE_ITEM, item,
1725 "protocol filtering not compatible"
1726 " with ICMP6 layer");
1727 if (!(item_flags & l3m))
1728 return rte_flow_error_set(error, EINVAL,
1729 RTE_FLOW_ERROR_TYPE_ITEM, item,
1730 "IPv6 is mandatory to filter on"
1732 if (item_flags & l4m)
1733 return rte_flow_error_set(error, EINVAL,
1734 RTE_FLOW_ERROR_TYPE_ITEM, item,
1735 "multiple L4 layers not supported");
1737 mask = &rte_flow_item_icmp6_mask;
1738 ret = mlx5_flow_item_acceptable
1739 (item, (const uint8_t *)mask,
1740 (const uint8_t *)&rte_flow_item_icmp6_mask,
1741 sizeof(struct rte_flow_item_icmp6),
1742 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1749 * Validate ICMP item.
1752 * Item specification.
1753 * @param[in] item_flags
1754 * Bit-fields that holds the items detected until now.
1756 * Pointer to error structure.
1759 * 0 on success, a negative errno value otherwise and rte_errno is set.
1762 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1763 uint64_t item_flags,
1764 uint8_t target_protocol,
1765 struct rte_flow_error *error)
1767 const struct rte_flow_item_icmp *mask = item->mask;
1768 const struct rte_flow_item_icmp nic_mask = {
1769 .hdr.icmp_type = 0xff,
1770 .hdr.icmp_code = 0xff,
1771 .hdr.icmp_ident = RTE_BE16(0xffff),
1772 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1774 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1775 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1776 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1777 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1778 MLX5_FLOW_LAYER_OUTER_L4;
1781 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1782 return rte_flow_error_set(error, EINVAL,
1783 RTE_FLOW_ERROR_TYPE_ITEM, item,
1784 "protocol filtering not compatible"
1785 " with ICMP layer");
1786 if (!(item_flags & l3m))
1787 return rte_flow_error_set(error, EINVAL,
1788 RTE_FLOW_ERROR_TYPE_ITEM, item,
1789 "IPv4 is mandatory to filter"
1791 if (item_flags & l4m)
1792 return rte_flow_error_set(error, EINVAL,
1793 RTE_FLOW_ERROR_TYPE_ITEM, item,
1794 "multiple L4 layers not supported");
1797 ret = mlx5_flow_item_acceptable
1798 (item, (const uint8_t *)mask,
1799 (const uint8_t *)&nic_mask,
1800 sizeof(struct rte_flow_item_icmp),
1801 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1808 * Validate Ethernet item.
1811 * Item specification.
1812 * @param[in] item_flags
1813 * Bit-fields that holds the items detected until now.
1815 * Pointer to error structure.
1818 * 0 on success, a negative errno value otherwise and rte_errno is set.
1821 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1822 uint64_t item_flags, bool ext_vlan_sup,
1823 struct rte_flow_error *error)
1825 const struct rte_flow_item_eth *mask = item->mask;
1826 const struct rte_flow_item_eth nic_mask = {
1827 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1828 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1829 .type = RTE_BE16(0xffff),
1830 .has_vlan = ext_vlan_sup ? 1 : 0,
1833 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1834 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1835 MLX5_FLOW_LAYER_OUTER_L2;
1837 if (item_flags & ethm)
1838 return rte_flow_error_set(error, ENOTSUP,
1839 RTE_FLOW_ERROR_TYPE_ITEM, item,
1840 "multiple L2 layers not supported");
1841 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1842 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1843 return rte_flow_error_set(error, EINVAL,
1844 RTE_FLOW_ERROR_TYPE_ITEM, item,
1845 "L2 layer should not follow "
1847 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1848 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1849 return rte_flow_error_set(error, EINVAL,
1850 RTE_FLOW_ERROR_TYPE_ITEM, item,
1851 "L2 layer should not follow VLAN");
1853 mask = &rte_flow_item_eth_mask;
1854 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1855 (const uint8_t *)&nic_mask,
1856 sizeof(struct rte_flow_item_eth),
1857 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1862 * Validate VLAN item.
1865 * Item specification.
1866 * @param[in] item_flags
1867 * Bit-fields that holds the items detected until now.
1869 * Ethernet device flow is being created on.
1871 * Pointer to error structure.
1874 * 0 on success, a negative errno value otherwise and rte_errno is set.
1877 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1878 uint64_t item_flags,
1879 struct rte_eth_dev *dev,
1880 struct rte_flow_error *error)
1882 const struct rte_flow_item_vlan *spec = item->spec;
1883 const struct rte_flow_item_vlan *mask = item->mask;
1884 const struct rte_flow_item_vlan nic_mask = {
1885 .tci = RTE_BE16(UINT16_MAX),
1886 .inner_type = RTE_BE16(UINT16_MAX),
1888 uint16_t vlan_tag = 0;
1889 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1891 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1892 MLX5_FLOW_LAYER_INNER_L4) :
1893 (MLX5_FLOW_LAYER_OUTER_L3 |
1894 MLX5_FLOW_LAYER_OUTER_L4);
1895 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1896 MLX5_FLOW_LAYER_OUTER_VLAN;
1898 if (item_flags & vlanm)
1899 return rte_flow_error_set(error, EINVAL,
1900 RTE_FLOW_ERROR_TYPE_ITEM, item,
1901 "multiple VLAN layers not supported");
1902 else if ((item_flags & l34m) != 0)
1903 return rte_flow_error_set(error, EINVAL,
1904 RTE_FLOW_ERROR_TYPE_ITEM, item,
1905 "VLAN cannot follow L3/L4 layer");
1907 mask = &rte_flow_item_vlan_mask;
1908 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1909 (const uint8_t *)&nic_mask,
1910 sizeof(struct rte_flow_item_vlan),
1911 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1914 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1915 struct mlx5_priv *priv = dev->data->dev_private;
1917 if (priv->vmwa_context) {
1919 * Non-NULL context means we have a virtual machine
1920 * and SR-IOV enabled, we have to create VLAN interface
1921 * to make hypervisor to setup E-Switch vport
1922 * context correctly. We avoid creating the multiple
1923 * VLAN interfaces, so we cannot support VLAN tag mask.
1925 return rte_flow_error_set(error, EINVAL,
1926 RTE_FLOW_ERROR_TYPE_ITEM,
1928 "VLAN tag mask is not"
1929 " supported in virtual"
1934 vlan_tag = spec->tci;
1935 vlan_tag &= mask->tci;
1938 * From verbs perspective an empty VLAN is equivalent
1939 * to a packet without VLAN layer.
1942 return rte_flow_error_set(error, EINVAL,
1943 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1945 "VLAN cannot be empty");
1950 * Validate IPV4 item.
1953 * Item specification.
1954 * @param[in] item_flags
1955 * Bit-fields that holds the items detected until now.
1956 * @param[in] last_item
1957 * Previous validated item in the pattern items.
1958 * @param[in] ether_type
1959 * Type in the ethernet layer header (including dot1q).
1960 * @param[in] acc_mask
1961 * Acceptable mask, if NULL default internal default mask
1962 * will be used to check whether item fields are supported.
1963 * @param[in] range_accepted
1964 * True if range of values is accepted for specific fields, false otherwise.
1966 * Pointer to error structure.
1969 * 0 on success, a negative errno value otherwise and rte_errno is set.
1972 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1973 uint64_t item_flags,
1975 uint16_t ether_type,
1976 const struct rte_flow_item_ipv4 *acc_mask,
1977 bool range_accepted,
1978 struct rte_flow_error *error)
1980 const struct rte_flow_item_ipv4 *mask = item->mask;
1981 const struct rte_flow_item_ipv4 *spec = item->spec;
1982 const struct rte_flow_item_ipv4 nic_mask = {
1984 .src_addr = RTE_BE32(0xffffffff),
1985 .dst_addr = RTE_BE32(0xffffffff),
1986 .type_of_service = 0xff,
1987 .next_proto_id = 0xff,
1990 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1991 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1992 MLX5_FLOW_LAYER_OUTER_L3;
1993 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1994 MLX5_FLOW_LAYER_OUTER_L4;
1996 uint8_t next_proto = 0xFF;
1997 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1998 MLX5_FLOW_LAYER_OUTER_VLAN |
1999 MLX5_FLOW_LAYER_INNER_VLAN);
2001 if ((last_item & l2_vlan) && ether_type &&
2002 ether_type != RTE_ETHER_TYPE_IPV4)
2003 return rte_flow_error_set(error, EINVAL,
2004 RTE_FLOW_ERROR_TYPE_ITEM, item,
2005 "IPv4 cannot follow L2/VLAN layer "
2006 "which ether type is not IPv4");
2007 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2009 next_proto = mask->hdr.next_proto_id &
2010 spec->hdr.next_proto_id;
2011 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2012 return rte_flow_error_set(error, EINVAL,
2013 RTE_FLOW_ERROR_TYPE_ITEM,
2018 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2019 return rte_flow_error_set(error, EINVAL,
2020 RTE_FLOW_ERROR_TYPE_ITEM, item,
2021 "wrong tunnel type - IPv6 specified "
2022 "but IPv4 item provided");
2023 if (item_flags & l3m)
2024 return rte_flow_error_set(error, ENOTSUP,
2025 RTE_FLOW_ERROR_TYPE_ITEM, item,
2026 "multiple L3 layers not supported");
2027 else if (item_flags & l4m)
2028 return rte_flow_error_set(error, EINVAL,
2029 RTE_FLOW_ERROR_TYPE_ITEM, item,
2030 "L3 cannot follow an L4 layer.");
2031 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2032 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2033 return rte_flow_error_set(error, EINVAL,
2034 RTE_FLOW_ERROR_TYPE_ITEM, item,
2035 "L3 cannot follow an NVGRE layer.");
2037 mask = &rte_flow_item_ipv4_mask;
2038 else if (mask->hdr.next_proto_id != 0 &&
2039 mask->hdr.next_proto_id != 0xff)
2040 return rte_flow_error_set(error, EINVAL,
2041 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2042 "partial mask is not supported"
2044 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2045 acc_mask ? (const uint8_t *)acc_mask
2046 : (const uint8_t *)&nic_mask,
2047 sizeof(struct rte_flow_item_ipv4),
2048 range_accepted, error);
2055 * Validate IPV6 item.
2058 * Item specification.
2059 * @param[in] item_flags
2060 * Bit-fields that holds the items detected until now.
2061 * @param[in] last_item
2062 * Previous validated item in the pattern items.
2063 * @param[in] ether_type
2064 * Type in the ethernet layer header (including dot1q).
2065 * @param[in] acc_mask
2066 * Acceptable mask, if NULL default internal default mask
2067 * will be used to check whether item fields are supported.
2069 * Pointer to error structure.
2072 * 0 on success, a negative errno value otherwise and rte_errno is set.
2075 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2076 uint64_t item_flags,
2078 uint16_t ether_type,
2079 const struct rte_flow_item_ipv6 *acc_mask,
2080 struct rte_flow_error *error)
2082 const struct rte_flow_item_ipv6 *mask = item->mask;
2083 const struct rte_flow_item_ipv6 *spec = item->spec;
2084 const struct rte_flow_item_ipv6 nic_mask = {
2087 "\xff\xff\xff\xff\xff\xff\xff\xff"
2088 "\xff\xff\xff\xff\xff\xff\xff\xff",
2090 "\xff\xff\xff\xff\xff\xff\xff\xff"
2091 "\xff\xff\xff\xff\xff\xff\xff\xff",
2092 .vtc_flow = RTE_BE32(0xffffffff),
2096 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2097 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2098 MLX5_FLOW_LAYER_OUTER_L3;
2099 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2100 MLX5_FLOW_LAYER_OUTER_L4;
2102 uint8_t next_proto = 0xFF;
2103 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2104 MLX5_FLOW_LAYER_OUTER_VLAN |
2105 MLX5_FLOW_LAYER_INNER_VLAN);
2107 if ((last_item & l2_vlan) && ether_type &&
2108 ether_type != RTE_ETHER_TYPE_IPV6)
2109 return rte_flow_error_set(error, EINVAL,
2110 RTE_FLOW_ERROR_TYPE_ITEM, item,
2111 "IPv6 cannot follow L2/VLAN layer "
2112 "which ether type is not IPv6");
2113 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2114 next_proto = spec->hdr.proto;
2115 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2116 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2117 return rte_flow_error_set(error, EINVAL,
2118 RTE_FLOW_ERROR_TYPE_ITEM,
2123 if (next_proto == IPPROTO_HOPOPTS ||
2124 next_proto == IPPROTO_ROUTING ||
2125 next_proto == IPPROTO_FRAGMENT ||
2126 next_proto == IPPROTO_ESP ||
2127 next_proto == IPPROTO_AH ||
2128 next_proto == IPPROTO_DSTOPTS)
2129 return rte_flow_error_set(error, EINVAL,
2130 RTE_FLOW_ERROR_TYPE_ITEM, item,
2131 "IPv6 proto (next header) should "
2132 "not be set as extension header");
2133 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2134 return rte_flow_error_set(error, EINVAL,
2135 RTE_FLOW_ERROR_TYPE_ITEM, item,
2136 "wrong tunnel type - IPv4 specified "
2137 "but IPv6 item provided");
2138 if (item_flags & l3m)
2139 return rte_flow_error_set(error, ENOTSUP,
2140 RTE_FLOW_ERROR_TYPE_ITEM, item,
2141 "multiple L3 layers not supported");
2142 else if (item_flags & l4m)
2143 return rte_flow_error_set(error, EINVAL,
2144 RTE_FLOW_ERROR_TYPE_ITEM, item,
2145 "L3 cannot follow an L4 layer.");
2146 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2147 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2148 return rte_flow_error_set(error, EINVAL,
2149 RTE_FLOW_ERROR_TYPE_ITEM, item,
2150 "L3 cannot follow an NVGRE layer.");
2152 mask = &rte_flow_item_ipv6_mask;
2153 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2154 acc_mask ? (const uint8_t *)acc_mask
2155 : (const uint8_t *)&nic_mask,
2156 sizeof(struct rte_flow_item_ipv6),
2157 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2164 * Validate UDP item.
2167 * Item specification.
2168 * @param[in] item_flags
2169 * Bit-fields that holds the items detected until now.
2170 * @param[in] target_protocol
2171 * The next protocol in the previous item.
2172 * @param[in] flow_mask
2173 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2175 * Pointer to error structure.
2178 * 0 on success, a negative errno value otherwise and rte_errno is set.
2181 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2182 uint64_t item_flags,
2183 uint8_t target_protocol,
2184 struct rte_flow_error *error)
2186 const struct rte_flow_item_udp *mask = item->mask;
2187 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2188 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2189 MLX5_FLOW_LAYER_OUTER_L3;
2190 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2191 MLX5_FLOW_LAYER_OUTER_L4;
2194 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2195 return rte_flow_error_set(error, EINVAL,
2196 RTE_FLOW_ERROR_TYPE_ITEM, item,
2197 "protocol filtering not compatible"
2199 if (!(item_flags & l3m))
2200 return rte_flow_error_set(error, EINVAL,
2201 RTE_FLOW_ERROR_TYPE_ITEM, item,
2202 "L3 is mandatory to filter on L4");
2203 if (item_flags & l4m)
2204 return rte_flow_error_set(error, EINVAL,
2205 RTE_FLOW_ERROR_TYPE_ITEM, item,
2206 "multiple L4 layers not supported");
2208 mask = &rte_flow_item_udp_mask;
2209 ret = mlx5_flow_item_acceptable
2210 (item, (const uint8_t *)mask,
2211 (const uint8_t *)&rte_flow_item_udp_mask,
2212 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2220 * Validate TCP item.
2223 * Item specification.
2224 * @param[in] item_flags
2225 * Bit-fields that holds the items detected until now.
2226 * @param[in] target_protocol
2227 * The next protocol in the previous item.
2229 * Pointer to error structure.
2232 * 0 on success, a negative errno value otherwise and rte_errno is set.
2235 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2236 uint64_t item_flags,
2237 uint8_t target_protocol,
2238 const struct rte_flow_item_tcp *flow_mask,
2239 struct rte_flow_error *error)
2241 const struct rte_flow_item_tcp *mask = item->mask;
2242 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2243 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2244 MLX5_FLOW_LAYER_OUTER_L3;
2245 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2246 MLX5_FLOW_LAYER_OUTER_L4;
2249 MLX5_ASSERT(flow_mask);
2250 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2251 return rte_flow_error_set(error, EINVAL,
2252 RTE_FLOW_ERROR_TYPE_ITEM, item,
2253 "protocol filtering not compatible"
2255 if (!(item_flags & l3m))
2256 return rte_flow_error_set(error, EINVAL,
2257 RTE_FLOW_ERROR_TYPE_ITEM, item,
2258 "L3 is mandatory to filter on L4");
2259 if (item_flags & l4m)
2260 return rte_flow_error_set(error, EINVAL,
2261 RTE_FLOW_ERROR_TYPE_ITEM, item,
2262 "multiple L4 layers not supported");
2264 mask = &rte_flow_item_tcp_mask;
2265 ret = mlx5_flow_item_acceptable
2266 (item, (const uint8_t *)mask,
2267 (const uint8_t *)flow_mask,
2268 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2276 * Validate VXLAN item.
2279 * Item specification.
2280 * @param[in] item_flags
2281 * Bit-fields that holds the items detected until now.
2282 * @param[in] target_protocol
2283 * The next protocol in the previous item.
2285 * Pointer to error structure.
2288 * 0 on success, a negative errno value otherwise and rte_errno is set.
2291 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2292 uint64_t item_flags,
2293 struct rte_flow_error *error)
2295 const struct rte_flow_item_vxlan *spec = item->spec;
2296 const struct rte_flow_item_vxlan *mask = item->mask;
2301 } id = { .vlan_id = 0, };
2304 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2305 return rte_flow_error_set(error, ENOTSUP,
2306 RTE_FLOW_ERROR_TYPE_ITEM, item,
2307 "multiple tunnel layers not"
2310 * Verify only UDPv4 is present as defined in
2311 * https://tools.ietf.org/html/rfc7348
2313 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2314 return rte_flow_error_set(error, EINVAL,
2315 RTE_FLOW_ERROR_TYPE_ITEM, item,
2316 "no outer UDP layer found");
2318 mask = &rte_flow_item_vxlan_mask;
2319 ret = mlx5_flow_item_acceptable
2320 (item, (const uint8_t *)mask,
2321 (const uint8_t *)&rte_flow_item_vxlan_mask,
2322 sizeof(struct rte_flow_item_vxlan),
2323 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2327 memcpy(&id.vni[1], spec->vni, 3);
2328 memcpy(&id.vni[1], mask->vni, 3);
2330 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2331 return rte_flow_error_set(error, ENOTSUP,
2332 RTE_FLOW_ERROR_TYPE_ITEM, item,
2333 "VXLAN tunnel must be fully defined");
2338 * Validate VXLAN_GPE item.
2341 * Item specification.
2342 * @param[in] item_flags
2343 * Bit-fields that holds the items detected until now.
2345 * Pointer to the private data structure.
2346 * @param[in] target_protocol
2347 * The next protocol in the previous item.
2349 * Pointer to error structure.
2352 * 0 on success, a negative errno value otherwise and rte_errno is set.
2355 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2356 uint64_t item_flags,
2357 struct rte_eth_dev *dev,
2358 struct rte_flow_error *error)
2360 struct mlx5_priv *priv = dev->data->dev_private;
2361 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2362 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2367 } id = { .vlan_id = 0, };
2369 if (!priv->config.l3_vxlan_en)
2370 return rte_flow_error_set(error, ENOTSUP,
2371 RTE_FLOW_ERROR_TYPE_ITEM, item,
2372 "L3 VXLAN is not enabled by device"
2373 " parameter and/or not configured in"
2375 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2376 return rte_flow_error_set(error, ENOTSUP,
2377 RTE_FLOW_ERROR_TYPE_ITEM, item,
2378 "multiple tunnel layers not"
2381 * Verify only UDPv4 is present as defined in
2382 * https://tools.ietf.org/html/rfc7348
2384 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2385 return rte_flow_error_set(error, EINVAL,
2386 RTE_FLOW_ERROR_TYPE_ITEM, item,
2387 "no outer UDP layer found");
2389 mask = &rte_flow_item_vxlan_gpe_mask;
2390 ret = mlx5_flow_item_acceptable
2391 (item, (const uint8_t *)mask,
2392 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2393 sizeof(struct rte_flow_item_vxlan_gpe),
2394 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2399 return rte_flow_error_set(error, ENOTSUP,
2400 RTE_FLOW_ERROR_TYPE_ITEM,
2402 "VxLAN-GPE protocol"
2404 memcpy(&id.vni[1], spec->vni, 3);
2405 memcpy(&id.vni[1], mask->vni, 3);
2407 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2408 return rte_flow_error_set(error, ENOTSUP,
2409 RTE_FLOW_ERROR_TYPE_ITEM, item,
2410 "VXLAN-GPE tunnel must be fully"
2415 * Validate GRE Key item.
2418 * Item specification.
2419 * @param[in] item_flags
2420 * Bit flags to mark detected items.
2421 * @param[in] gre_item
2422 * Pointer to gre_item
2424 * Pointer to error structure.
2427 * 0 on success, a negative errno value otherwise and rte_errno is set.
2430 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2431 uint64_t item_flags,
2432 const struct rte_flow_item *gre_item,
2433 struct rte_flow_error *error)
2435 const rte_be32_t *mask = item->mask;
2437 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2438 const struct rte_flow_item_gre *gre_spec;
2439 const struct rte_flow_item_gre *gre_mask;
2441 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2442 return rte_flow_error_set(error, ENOTSUP,
2443 RTE_FLOW_ERROR_TYPE_ITEM, item,
2444 "Multiple GRE key not support");
2445 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2446 return rte_flow_error_set(error, ENOTSUP,
2447 RTE_FLOW_ERROR_TYPE_ITEM, item,
2448 "No preceding GRE header");
2449 if (item_flags & MLX5_FLOW_LAYER_INNER)
2450 return rte_flow_error_set(error, ENOTSUP,
2451 RTE_FLOW_ERROR_TYPE_ITEM, item,
2452 "GRE key following a wrong item");
2453 gre_mask = gre_item->mask;
2455 gre_mask = &rte_flow_item_gre_mask;
2456 gre_spec = gre_item->spec;
2457 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2458 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2459 return rte_flow_error_set(error, EINVAL,
2460 RTE_FLOW_ERROR_TYPE_ITEM, item,
2461 "Key bit must be on");
2464 mask = &gre_key_default_mask;
2465 ret = mlx5_flow_item_acceptable
2466 (item, (const uint8_t *)mask,
2467 (const uint8_t *)&gre_key_default_mask,
2468 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2473 * Validate GRE item.
2476 * Item specification.
2477 * @param[in] item_flags
2478 * Bit flags to mark detected items.
2479 * @param[in] target_protocol
2480 * The next protocol in the previous item.
2482 * Pointer to error structure.
2485 * 0 on success, a negative errno value otherwise and rte_errno is set.
2488 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2489 uint64_t item_flags,
2490 uint8_t target_protocol,
2491 struct rte_flow_error *error)
2493 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2494 const struct rte_flow_item_gre *mask = item->mask;
2496 const struct rte_flow_item_gre nic_mask = {
2497 .c_rsvd0_ver = RTE_BE16(0xB000),
2498 .protocol = RTE_BE16(UINT16_MAX),
2501 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2502 return rte_flow_error_set(error, EINVAL,
2503 RTE_FLOW_ERROR_TYPE_ITEM, item,
2504 "protocol filtering not compatible"
2505 " with this GRE layer");
2506 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2507 return rte_flow_error_set(error, ENOTSUP,
2508 RTE_FLOW_ERROR_TYPE_ITEM, item,
2509 "multiple tunnel layers not"
2511 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2512 return rte_flow_error_set(error, ENOTSUP,
2513 RTE_FLOW_ERROR_TYPE_ITEM, item,
2514 "L3 Layer is missing");
2516 mask = &rte_flow_item_gre_mask;
2517 ret = mlx5_flow_item_acceptable
2518 (item, (const uint8_t *)mask,
2519 (const uint8_t *)&nic_mask,
2520 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2524 #ifndef HAVE_MLX5DV_DR
2525 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2526 if (spec && (spec->protocol & mask->protocol))
2527 return rte_flow_error_set(error, ENOTSUP,
2528 RTE_FLOW_ERROR_TYPE_ITEM, item,
2529 "without MPLS support the"
2530 " specification cannot be used for"
2538 * Validate Geneve item.
2541 * Item specification.
2542 * @param[in] itemFlags
2543 * Bit-fields that holds the items detected until now.
2545 * Pointer to the private data structure.
2547 * Pointer to error structure.
2550 * 0 on success, a negative errno value otherwise and rte_errno is set.
2554 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2555 uint64_t item_flags,
2556 struct rte_eth_dev *dev,
2557 struct rte_flow_error *error)
2559 struct mlx5_priv *priv = dev->data->dev_private;
2560 const struct rte_flow_item_geneve *spec = item->spec;
2561 const struct rte_flow_item_geneve *mask = item->mask;
2564 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2565 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2566 const struct rte_flow_item_geneve nic_mask = {
2567 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2568 .vni = "\xff\xff\xff",
2569 .protocol = RTE_BE16(UINT16_MAX),
2572 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2573 return rte_flow_error_set(error, ENOTSUP,
2574 RTE_FLOW_ERROR_TYPE_ITEM, item,
2575 "L3 Geneve is not enabled by device"
2576 " parameter and/or not configured in"
2578 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2579 return rte_flow_error_set(error, ENOTSUP,
2580 RTE_FLOW_ERROR_TYPE_ITEM, item,
2581 "multiple tunnel layers not"
2584 * Verify only UDPv4 is present as defined in
2585 * https://tools.ietf.org/html/rfc7348
2587 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2588 return rte_flow_error_set(error, EINVAL,
2589 RTE_FLOW_ERROR_TYPE_ITEM, item,
2590 "no outer UDP layer found");
2592 mask = &rte_flow_item_geneve_mask;
2593 ret = mlx5_flow_item_acceptable
2594 (item, (const uint8_t *)mask,
2595 (const uint8_t *)&nic_mask,
2596 sizeof(struct rte_flow_item_geneve),
2597 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2601 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2602 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2603 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2604 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2605 return rte_flow_error_set(error, ENOTSUP,
2606 RTE_FLOW_ERROR_TYPE_ITEM,
2608 "Geneve protocol unsupported"
2609 " fields are being used");
2610 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2611 return rte_flow_error_set
2613 RTE_FLOW_ERROR_TYPE_ITEM,
2615 "Unsupported Geneve options length");
2617 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2618 return rte_flow_error_set
2620 RTE_FLOW_ERROR_TYPE_ITEM, item,
2621 "Geneve tunnel must be fully defined");
2626 * Validate Geneve TLV option item.
2629 * Item specification.
2630 * @param[in] last_item
2631 * Previous validated item in the pattern items.
2632 * @param[in] geneve_item
2633 * Previous GENEVE item specification.
2635 * Pointer to the rte_eth_dev structure.
2637 * Pointer to error structure.
2640 * 0 on success, a negative errno value otherwise and rte_errno is set.
2643 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2645 const struct rte_flow_item *geneve_item,
2646 struct rte_eth_dev *dev,
2647 struct rte_flow_error *error)
2649 struct mlx5_priv *priv = dev->data->dev_private;
2650 struct mlx5_dev_ctx_shared *sh = priv->sh;
2651 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2652 struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2653 uint8_t data_max_supported =
2654 hca_attr->max_geneve_tlv_option_data_len * 4;
2655 struct mlx5_dev_config *config = &priv->config;
2656 const struct rte_flow_item_geneve *geneve_spec;
2657 const struct rte_flow_item_geneve *geneve_mask;
2658 const struct rte_flow_item_geneve_opt *spec = item->spec;
2659 const struct rte_flow_item_geneve_opt *mask = item->mask;
2661 unsigned int data_len;
2662 uint8_t tlv_option_len;
2663 uint16_t optlen_m, optlen_v;
2664 const struct rte_flow_item_geneve_opt full_mask = {
2665 .option_class = RTE_BE16(0xffff),
2666 .option_type = 0xff,
2671 mask = &rte_flow_item_geneve_opt_mask;
2673 return rte_flow_error_set
2674 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2675 "Geneve TLV opt class/type/length must be specified");
2676 if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2677 return rte_flow_error_set
2678 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2679 "Geneve TLV opt length exceeeds the limit (31)");
2680 /* Check if class type and length masks are full. */
2681 if (full_mask.option_class != mask->option_class ||
2682 full_mask.option_type != mask->option_type ||
2683 full_mask.option_len != (mask->option_len & full_mask.option_len))
2684 return rte_flow_error_set
2685 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2686 "Geneve TLV opt class/type/length masks must be full");
2687 /* Check if length is supported */
2688 if ((uint32_t)spec->option_len >
2689 config->hca_attr.max_geneve_tlv_option_data_len)
2690 return rte_flow_error_set
2691 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2692 "Geneve TLV opt length not supported");
2693 if (config->hca_attr.max_geneve_tlv_options > 1)
2695 "max_geneve_tlv_options supports more than 1 option");
2696 /* Check GENEVE item preceding. */
2697 if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2698 return rte_flow_error_set
2699 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2700 "Geneve opt item must be preceded with Geneve item");
2701 geneve_spec = geneve_item->spec;
2702 geneve_mask = geneve_item->mask ? geneve_item->mask :
2703 &rte_flow_item_geneve_mask;
2704 /* Check if GENEVE TLV option size doesn't exceed option length */
2705 if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2706 geneve_spec->ver_opt_len_o_c_rsvd0)) {
2707 tlv_option_len = spec->option_len & mask->option_len;
2708 optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2709 optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2710 optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2711 optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2712 if ((optlen_v & optlen_m) <= tlv_option_len)
2713 return rte_flow_error_set
2714 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2715 "GENEVE TLV option length exceeds optlen");
2717 /* Check if length is 0 or data is 0. */
2718 if (spec->data == NULL || spec->option_len == 0)
2719 return rte_flow_error_set
2720 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2721 "Geneve TLV opt with zero data/length not supported");
2722 /* Check not all data & mask are 0. */
2723 data_len = spec->option_len * 4;
2724 if (mask->data == NULL) {
2725 for (i = 0; i < data_len; i++)
2729 return rte_flow_error_set(error, ENOTSUP,
2730 RTE_FLOW_ERROR_TYPE_ITEM, item,
2731 "Can't match on Geneve option data 0");
2733 for (i = 0; i < data_len; i++)
2734 if (spec->data[i] & mask->data[i])
2737 return rte_flow_error_set(error, ENOTSUP,
2738 RTE_FLOW_ERROR_TYPE_ITEM, item,
2739 "Can't match on Geneve option data and mask 0");
2740 /* Check data mask supported. */
2741 for (i = data_max_supported; i < data_len ; i++)
2743 return rte_flow_error_set(error, ENOTSUP,
2744 RTE_FLOW_ERROR_TYPE_ITEM, item,
2745 "Data mask is of unsupported size");
2747 /* Check GENEVE option is supported in NIC. */
2748 if (!config->hca_attr.geneve_tlv_opt)
2749 return rte_flow_error_set
2750 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2751 "Geneve TLV opt not supported");
2752 /* Check if we already have geneve option with different type/class. */
2753 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2754 geneve_opt_resource = sh->geneve_tlv_option_resource;
2755 if (geneve_opt_resource != NULL)
2756 if (geneve_opt_resource->option_class != spec->option_class ||
2757 geneve_opt_resource->option_type != spec->option_type ||
2758 geneve_opt_resource->length != spec->option_len) {
2759 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2760 return rte_flow_error_set(error, ENOTSUP,
2761 RTE_FLOW_ERROR_TYPE_ITEM, item,
2762 "Only one Geneve TLV option supported");
2764 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2769 * Validate MPLS item.
2772 * Pointer to the rte_eth_dev structure.
2774 * Item specification.
2775 * @param[in] item_flags
2776 * Bit-fields that holds the items detected until now.
2777 * @param[in] prev_layer
2778 * The protocol layer indicated in previous item.
2780 * Pointer to error structure.
2783 * 0 on success, a negative errno value otherwise and rte_errno is set.
2786 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2787 const struct rte_flow_item *item __rte_unused,
2788 uint64_t item_flags __rte_unused,
2789 uint64_t prev_layer __rte_unused,
2790 struct rte_flow_error *error)
2792 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2793 const struct rte_flow_item_mpls *mask = item->mask;
2794 struct mlx5_priv *priv = dev->data->dev_private;
2797 if (!priv->config.mpls_en)
2798 return rte_flow_error_set(error, ENOTSUP,
2799 RTE_FLOW_ERROR_TYPE_ITEM, item,
2800 "MPLS not supported or"
2801 " disabled in firmware"
2803 /* MPLS over IP, UDP, GRE is allowed */
2804 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2805 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2806 MLX5_FLOW_LAYER_GRE |
2807 MLX5_FLOW_LAYER_GRE_KEY)))
2808 return rte_flow_error_set(error, EINVAL,
2809 RTE_FLOW_ERROR_TYPE_ITEM, item,
2810 "protocol filtering not compatible"
2811 " with MPLS layer");
2812 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2813 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2814 !(item_flags & MLX5_FLOW_LAYER_GRE))
2815 return rte_flow_error_set(error, ENOTSUP,
2816 RTE_FLOW_ERROR_TYPE_ITEM, item,
2817 "multiple tunnel layers not"
2820 mask = &rte_flow_item_mpls_mask;
2821 ret = mlx5_flow_item_acceptable
2822 (item, (const uint8_t *)mask,
2823 (const uint8_t *)&rte_flow_item_mpls_mask,
2824 sizeof(struct rte_flow_item_mpls),
2825 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2830 return rte_flow_error_set(error, ENOTSUP,
2831 RTE_FLOW_ERROR_TYPE_ITEM, item,
2832 "MPLS is not supported by Verbs, please"
2838 * Validate NVGRE item.
2841 * Item specification.
2842 * @param[in] item_flags
2843 * Bit flags to mark detected items.
2844 * @param[in] target_protocol
2845 * The next protocol in the previous item.
2847 * Pointer to error structure.
2850 * 0 on success, a negative errno value otherwise and rte_errno is set.
2853 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2854 uint64_t item_flags,
2855 uint8_t target_protocol,
2856 struct rte_flow_error *error)
2858 const struct rte_flow_item_nvgre *mask = item->mask;
2861 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2862 return rte_flow_error_set(error, EINVAL,
2863 RTE_FLOW_ERROR_TYPE_ITEM, item,
2864 "protocol filtering not compatible"
2865 " with this GRE layer");
2866 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2867 return rte_flow_error_set(error, ENOTSUP,
2868 RTE_FLOW_ERROR_TYPE_ITEM, item,
2869 "multiple tunnel layers not"
2871 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2872 return rte_flow_error_set(error, ENOTSUP,
2873 RTE_FLOW_ERROR_TYPE_ITEM, item,
2874 "L3 Layer is missing");
2876 mask = &rte_flow_item_nvgre_mask;
2877 ret = mlx5_flow_item_acceptable
2878 (item, (const uint8_t *)mask,
2879 (const uint8_t *)&rte_flow_item_nvgre_mask,
2880 sizeof(struct rte_flow_item_nvgre),
2881 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2888 * Validate eCPRI item.
2891 * Item specification.
2892 * @param[in] item_flags
2893 * Bit-fields that holds the items detected until now.
2894 * @param[in] last_item
2895 * Previous validated item in the pattern items.
2896 * @param[in] ether_type
2897 * Type in the ethernet layer header (including dot1q).
2898 * @param[in] acc_mask
2899 * Acceptable mask, if NULL default internal default mask
2900 * will be used to check whether item fields are supported.
2902 * Pointer to error structure.
2905 * 0 on success, a negative errno value otherwise and rte_errno is set.
2908 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2909 uint64_t item_flags,
2911 uint16_t ether_type,
2912 const struct rte_flow_item_ecpri *acc_mask,
2913 struct rte_flow_error *error)
2915 const struct rte_flow_item_ecpri *mask = item->mask;
2916 const struct rte_flow_item_ecpri nic_mask = {
2920 RTE_BE32(((const struct rte_ecpri_common_hdr) {
2924 .dummy[0] = 0xFFFFFFFF,
2927 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2928 MLX5_FLOW_LAYER_OUTER_VLAN);
2929 struct rte_flow_item_ecpri mask_lo;
2931 if (!(last_item & outer_l2_vlan) &&
2932 last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2933 return rte_flow_error_set(error, EINVAL,
2934 RTE_FLOW_ERROR_TYPE_ITEM, item,
2935 "eCPRI can only follow L2/VLAN layer or UDP layer");
2936 if ((last_item & outer_l2_vlan) && ether_type &&
2937 ether_type != RTE_ETHER_TYPE_ECPRI)
2938 return rte_flow_error_set(error, EINVAL,
2939 RTE_FLOW_ERROR_TYPE_ITEM, item,
2940 "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2941 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2942 return rte_flow_error_set(error, EINVAL,
2943 RTE_FLOW_ERROR_TYPE_ITEM, item,
2944 "eCPRI with tunnel is not supported right now");
2945 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2946 return rte_flow_error_set(error, ENOTSUP,
2947 RTE_FLOW_ERROR_TYPE_ITEM, item,
2948 "multiple L3 layers not supported");
2949 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2950 return rte_flow_error_set(error, EINVAL,
2951 RTE_FLOW_ERROR_TYPE_ITEM, item,
2952 "eCPRI cannot coexist with a TCP layer");
2953 /* In specification, eCPRI could be over UDP layer. */
2954 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2955 return rte_flow_error_set(error, EINVAL,
2956 RTE_FLOW_ERROR_TYPE_ITEM, item,
2957 "eCPRI over UDP layer is not yet supported right now");
2958 /* Mask for type field in common header could be zero. */
2960 mask = &rte_flow_item_ecpri_mask;
2961 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2962 /* Input mask is in big-endian format. */
2963 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2964 return rte_flow_error_set(error, EINVAL,
2965 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2966 "partial mask is not supported for protocol");
2967 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2968 return rte_flow_error_set(error, EINVAL,
2969 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2970 "message header mask must be after a type mask");
2971 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2972 acc_mask ? (const uint8_t *)acc_mask
2973 : (const uint8_t *)&nic_mask,
2974 sizeof(struct rte_flow_item_ecpri),
2975 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2979 * Release resource related QUEUE/RSS action split.
2982 * Pointer to Ethernet device.
2984 * Flow to release id's from.
2987 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2988 struct rte_flow *flow)
2990 struct mlx5_priv *priv = dev->data->dev_private;
2991 uint32_t handle_idx;
2992 struct mlx5_flow_handle *dev_handle;
2994 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2995 handle_idx, dev_handle, next)
2996 if (dev_handle->split_flow_id)
2997 mlx5_ipool_free(priv->sh->ipool
2998 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2999 dev_handle->split_flow_id);
3003 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3004 const struct rte_flow_attr *attr __rte_unused,
3005 const struct rte_flow_item items[] __rte_unused,
3006 const struct rte_flow_action actions[] __rte_unused,
3007 bool external __rte_unused,
3008 int hairpin __rte_unused,
3009 struct rte_flow_error *error)
3011 return rte_flow_error_set(error, ENOTSUP,
3012 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3015 static struct mlx5_flow *
3016 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3017 const struct rte_flow_attr *attr __rte_unused,
3018 const struct rte_flow_item items[] __rte_unused,
3019 const struct rte_flow_action actions[] __rte_unused,
3020 struct rte_flow_error *error)
3022 rte_flow_error_set(error, ENOTSUP,
3023 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3028 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3029 struct mlx5_flow *dev_flow __rte_unused,
3030 const struct rte_flow_attr *attr __rte_unused,
3031 const struct rte_flow_item items[] __rte_unused,
3032 const struct rte_flow_action actions[] __rte_unused,
3033 struct rte_flow_error *error)
3035 return rte_flow_error_set(error, ENOTSUP,
3036 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3040 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3041 struct rte_flow *flow __rte_unused,
3042 struct rte_flow_error *error)
3044 return rte_flow_error_set(error, ENOTSUP,
3045 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3049 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3050 struct rte_flow *flow __rte_unused)
3055 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3056 struct rte_flow *flow __rte_unused)
3061 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3062 struct rte_flow *flow __rte_unused,
3063 const struct rte_flow_action *actions __rte_unused,
3064 void *data __rte_unused,
3065 struct rte_flow_error *error)
3067 return rte_flow_error_set(error, ENOTSUP,
3068 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3072 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3073 uint32_t domains __rte_unused,
3074 uint32_t flags __rte_unused)
3079 /* Void driver to protect from null pointer reference. */
3080 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3081 .validate = flow_null_validate,
3082 .prepare = flow_null_prepare,
3083 .translate = flow_null_translate,
3084 .apply = flow_null_apply,
3085 .remove = flow_null_remove,
3086 .destroy = flow_null_destroy,
3087 .query = flow_null_query,
3088 .sync_domain = flow_null_sync_domain,
3092 * Select flow driver type according to flow attributes and device
3096 * Pointer to the dev structure.
3098 * Pointer to the flow attributes.
3101 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3103 static enum mlx5_flow_drv_type
3104 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3106 struct mlx5_priv *priv = dev->data->dev_private;
3107 /* The OS can determine first a specific flow type (DV, VERBS) */
3108 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3110 if (type != MLX5_FLOW_TYPE_MAX)
3112 /* If no OS specific type - continue with DV/VERBS selection */
3113 if (attr->transfer && priv->config.dv_esw_en)
3114 type = MLX5_FLOW_TYPE_DV;
3115 if (!attr->transfer)
3116 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3117 MLX5_FLOW_TYPE_VERBS;
3121 #define flow_get_drv_ops(type) flow_drv_ops[type]
3124 * Flow driver validation API. This abstracts calling driver specific functions.
3125 * The type of flow driver is determined according to flow attributes.
3128 * Pointer to the dev structure.
3130 * Pointer to the flow attributes.
3132 * Pointer to the list of items.
3133 * @param[in] actions
3134 * Pointer to the list of actions.
3135 * @param[in] external
3136 * This flow rule is created by request external to PMD.
3137 * @param[in] hairpin
3138 * Number of hairpin TX actions, 0 means classic flow.
3140 * Pointer to the error structure.
3143 * 0 on success, a negative errno value otherwise and rte_errno is set.
3146 flow_drv_validate(struct rte_eth_dev *dev,
3147 const struct rte_flow_attr *attr,
3148 const struct rte_flow_item items[],
3149 const struct rte_flow_action actions[],
3150 bool external, int hairpin, struct rte_flow_error *error)
3152 const struct mlx5_flow_driver_ops *fops;
3153 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3155 fops = flow_get_drv_ops(type);
3156 return fops->validate(dev, attr, items, actions, external,
3161 * Flow driver preparation API. This abstracts calling driver specific
3162 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3163 * calculates the size of memory required for device flow, allocates the memory,
3164 * initializes the device flow and returns the pointer.
3167 * This function initializes device flow structure such as dv or verbs in
3168 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3169 * rest. For example, adding returning device flow to flow->dev_flow list and
3170 * setting backward reference to the flow should be done out of this function.
3171 * layers field is not filled either.
3174 * Pointer to the dev structure.
3176 * Pointer to the flow attributes.
3178 * Pointer to the list of items.
3179 * @param[in] actions
3180 * Pointer to the list of actions.
3181 * @param[in] flow_idx
3182 * This memory pool index to the flow.
3184 * Pointer to the error structure.
3187 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3189 static inline struct mlx5_flow *
3190 flow_drv_prepare(struct rte_eth_dev *dev,
3191 const struct rte_flow *flow,
3192 const struct rte_flow_attr *attr,
3193 const struct rte_flow_item items[],
3194 const struct rte_flow_action actions[],
3196 struct rte_flow_error *error)
3198 const struct mlx5_flow_driver_ops *fops;
3199 enum mlx5_flow_drv_type type = flow->drv_type;
3200 struct mlx5_flow *mlx5_flow = NULL;
3202 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3203 fops = flow_get_drv_ops(type);
3204 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3206 mlx5_flow->flow_idx = flow_idx;
3211 * Flow driver translation API. This abstracts calling driver specific
3212 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3213 * translates a generic flow into a driver flow. flow_drv_prepare() must
3217 * dev_flow->layers could be filled as a result of parsing during translation
3218 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3219 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3220 * flow->actions could be overwritten even though all the expanded dev_flows
3221 * have the same actions.
3224 * Pointer to the rte dev structure.
3225 * @param[in, out] dev_flow
3226 * Pointer to the mlx5 flow.
3228 * Pointer to the flow attributes.
3230 * Pointer to the list of items.
3231 * @param[in] actions
3232 * Pointer to the list of actions.
3234 * Pointer to the error structure.
3237 * 0 on success, a negative errno value otherwise and rte_errno is set.
3240 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3241 const struct rte_flow_attr *attr,
3242 const struct rte_flow_item items[],
3243 const struct rte_flow_action actions[],
3244 struct rte_flow_error *error)
3246 const struct mlx5_flow_driver_ops *fops;
3247 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3249 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3250 fops = flow_get_drv_ops(type);
3251 return fops->translate(dev, dev_flow, attr, items, actions, error);
3255 * Flow driver apply API. This abstracts calling driver specific functions.
3256 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3257 * translated driver flows on to device. flow_drv_translate() must precede.
3260 * Pointer to Ethernet device structure.
3261 * @param[in, out] flow
3262 * Pointer to flow structure.
3264 * Pointer to error structure.
3267 * 0 on success, a negative errno value otherwise and rte_errno is set.
3270 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3271 struct rte_flow_error *error)
3273 const struct mlx5_flow_driver_ops *fops;
3274 enum mlx5_flow_drv_type type = flow->drv_type;
3276 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3277 fops = flow_get_drv_ops(type);
3278 return fops->apply(dev, flow, error);
3282 * Flow driver destroy API. This abstracts calling driver specific functions.
3283 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3284 * on device and releases resources of the flow.
3287 * Pointer to Ethernet device.
3288 * @param[in, out] flow
3289 * Pointer to flow structure.
3292 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3294 const struct mlx5_flow_driver_ops *fops;
3295 enum mlx5_flow_drv_type type = flow->drv_type;
3297 flow_mreg_split_qrss_release(dev, flow);
3298 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3299 fops = flow_get_drv_ops(type);
3300 fops->destroy(dev, flow);
3304 * Get RSS action from the action list.
3306 * @param[in] actions
3307 * Pointer to the list of actions.
3310 * Pointer to the RSS action if exist, else return NULL.
3312 static const struct rte_flow_action_rss*
3313 flow_get_rss_action(const struct rte_flow_action actions[])
3315 const struct rte_flow_action_rss *rss = NULL;
3317 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3318 switch (actions->type) {
3319 case RTE_FLOW_ACTION_TYPE_RSS:
3320 rss = actions->conf;
3322 case RTE_FLOW_ACTION_TYPE_SAMPLE:
3324 const struct rte_flow_action_sample *sample =
3326 const struct rte_flow_action *act = sample->actions;
3327 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3328 if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3340 * Get ASO age action by index.
3343 * Pointer to the Ethernet device structure.
3344 * @param[in] age_idx
3345 * Index to the ASO age action.
3348 * The specified ASO age action.
3350 struct mlx5_aso_age_action*
3351 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3353 uint16_t pool_idx = age_idx & UINT16_MAX;
3354 uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3355 struct mlx5_priv *priv = dev->data->dev_private;
3356 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3357 struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3359 return &pool->actions[offset - 1];
3362 /* maps shared action to translated non shared in some actions array */
3363 struct mlx5_translated_shared_action {
3364 struct rte_flow_shared_action *action; /**< Shared action */
3365 int index; /**< Index in related array of rte_flow_action */
3369 * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3370 * non shared action if translation possible.
3371 * This functionality used to run same execution path for both shared & non
3372 * shared actions on flow create. All necessary preparations for shared
3373 * action handling should be preformed on *shared* actions list returned
3377 * Pointer to Ethernet device.
3378 * @param[in] actions
3379 * List of actions to translate.
3380 * @param[out] shared
3381 * List to store translated shared actions.
3382 * @param[in, out] shared_n
3383 * Size of *shared* array. On return should be updated with number of shared
3384 * actions retrieved from the *actions* list.
3385 * @param[out] translated_actions
3386 * List of actions where all shared actions were translated to non shared
3387 * if possible. NULL if no translation took place.
3389 * Pointer to the error structure.
3392 * 0 on success, a negative errno value otherwise and rte_errno is set.
3395 flow_shared_actions_translate(struct rte_eth_dev *dev,
3396 const struct rte_flow_action actions[],
3397 struct mlx5_translated_shared_action *shared,
3399 struct rte_flow_action **translated_actions,
3400 struct rte_flow_error *error)
3402 struct mlx5_priv *priv = dev->data->dev_private;
3403 struct rte_flow_action *translated = NULL;
3404 size_t actions_size;
3407 struct mlx5_translated_shared_action *shared_end = NULL;
3409 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3410 if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3412 if (copied_n == *shared_n) {
3413 return rte_flow_error_set
3414 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3415 NULL, "too many shared actions");
3417 rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3418 sizeof(actions[n].conf));
3419 shared[copied_n].index = n;
3423 *shared_n = copied_n;
3426 actions_size = sizeof(struct rte_flow_action) * n;
3427 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3432 memcpy(translated, actions, actions_size);
3433 for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3434 struct mlx5_shared_action_rss *shared_rss;
3435 uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3436 uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3437 uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3441 case MLX5_SHARED_ACTION_TYPE_RSS:
3442 shared_rss = mlx5_ipool_get
3443 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3444 translated[shared->index].type =
3445 RTE_FLOW_ACTION_TYPE_RSS;
3446 translated[shared->index].conf =
3447 &shared_rss->origin;
3449 case MLX5_SHARED_ACTION_TYPE_AGE:
3450 if (priv->sh->flow_hit_aso_en) {
3451 translated[shared->index].type =
3452 (enum rte_flow_action_type)
3453 MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3454 translated[shared->index].conf =
3455 (void *)(uintptr_t)idx;
3460 mlx5_free(translated);
3461 return rte_flow_error_set
3462 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3463 NULL, "invalid shared action type");
3466 *translated_actions = translated;
3471 * Get Shared RSS action from the action list.
3474 * Pointer to Ethernet device.
3476 * Pointer to the list of actions.
3477 * @param[in] shared_n
3478 * Actions list length.
3481 * The MLX5 RSS action ID if exists, otherwise return 0.
3484 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3485 struct mlx5_translated_shared_action *shared,
3488 struct mlx5_translated_shared_action *shared_end;
3489 struct mlx5_priv *priv = dev->data->dev_private;
3490 struct mlx5_shared_action_rss *shared_rss;
3493 for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3494 uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3495 uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3496 uint32_t idx = act_idx &
3497 ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3499 case MLX5_SHARED_ACTION_TYPE_RSS:
3500 shared_rss = mlx5_ipool_get
3501 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3503 __atomic_add_fetch(&shared_rss->refcnt, 1,
3514 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3516 const struct rte_flow_item *item;
3517 unsigned int has_vlan = 0;
3519 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3520 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3526 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3527 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3528 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3529 MLX5_EXPANSION_ROOT_OUTER;
3533 * Get layer flags from the prefix flow.
3535 * Some flows may be split to several subflows, the prefix subflow gets the
3536 * match items and the suffix sub flow gets the actions.
3537 * Some actions need the user defined match item flags to get the detail for
3539 * This function helps the suffix flow to get the item layer flags from prefix
3542 * @param[in] dev_flow
3543 * Pointer the created preifx subflow.
3546 * The layers get from prefix subflow.
3548 static inline uint64_t
3549 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3551 uint64_t layers = 0;
3554 * Layers bits could be localization, but usually the compiler will
3555 * help to do the optimization work for source code.
3556 * If no decap actions, use the layers directly.
3558 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3559 return dev_flow->handle->layers;
3560 /* Convert L3 layers with decap action. */
3561 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3562 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3563 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3564 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3565 /* Convert L4 layers with decap action. */
3566 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3567 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3568 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3569 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3574 * Get metadata split action information.
3576 * @param[in] actions
3577 * Pointer to the list of actions.
3579 * Pointer to the return pointer.
3580 * @param[out] qrss_type
3581 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3582 * if no QUEUE/RSS is found.
3583 * @param[out] encap_idx
3584 * Pointer to the index of the encap action if exists, otherwise the last
3588 * Total number of actions.
3591 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3592 const struct rte_flow_action **qrss,
3595 const struct rte_flow_action_raw_encap *raw_encap;
3597 int raw_decap_idx = -1;
3600 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3601 switch (actions->type) {
3602 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3603 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3604 *encap_idx = actions_n;
3606 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3607 raw_decap_idx = actions_n;
3609 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3610 raw_encap = actions->conf;
3611 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3612 *encap_idx = raw_decap_idx != -1 ?
3613 raw_decap_idx : actions_n;
3615 case RTE_FLOW_ACTION_TYPE_QUEUE:
3616 case RTE_FLOW_ACTION_TYPE_RSS:
3624 if (*encap_idx == -1)
3625 *encap_idx = actions_n;
3626 /* Count RTE_FLOW_ACTION_TYPE_END. */
3627 return actions_n + 1;
3631 * Check meter action from the action list.
3633 * @param[in] actions
3634 * Pointer to the list of actions.
3636 * Pointer to the meter exist flag.
3639 * Total number of actions.
3642 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3648 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3649 switch (actions->type) {
3650 case RTE_FLOW_ACTION_TYPE_METER:
3658 /* Count RTE_FLOW_ACTION_TYPE_END. */
3659 return actions_n + 1;
3663 * Check if the flow should be split due to hairpin.
3664 * The reason for the split is that in current HW we can't
3665 * support encap and push-vlan on Rx, so if a flow contains
3666 * these actions we move it to Tx.
3669 * Pointer to Ethernet device.
3671 * Flow rule attributes.
3672 * @param[in] actions
3673 * Associated actions (list terminated by the END action).
3676 * > 0 the number of actions and the flow should be split,
3677 * 0 when no split required.
3680 flow_check_hairpin_split(struct rte_eth_dev *dev,
3681 const struct rte_flow_attr *attr,
3682 const struct rte_flow_action actions[])
3684 int queue_action = 0;
3687 const struct rte_flow_action_queue *queue;
3688 const struct rte_flow_action_rss *rss;
3689 const struct rte_flow_action_raw_encap *raw_encap;
3690 const struct rte_eth_hairpin_conf *conf;
3694 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3695 switch (actions->type) {
3696 case RTE_FLOW_ACTION_TYPE_QUEUE:
3697 queue = actions->conf;
3700 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3701 if (conf == NULL || conf->tx_explicit != 0)
3706 case RTE_FLOW_ACTION_TYPE_RSS:
3707 rss = actions->conf;
3708 if (rss == NULL || rss->queue_num == 0)
3710 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3711 if (conf == NULL || conf->tx_explicit != 0)
3716 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3717 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3718 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3719 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3720 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3724 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3725 raw_encap = actions->conf;
3726 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3735 if (split && queue_action)
3740 /* Declare flow create/destroy prototype in advance. */
3742 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3743 const struct rte_flow_attr *attr,
3744 const struct rte_flow_item items[],
3745 const struct rte_flow_action actions[],
3746 bool external, struct rte_flow_error *error);
3749 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3753 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3754 struct mlx5_hlist_entry *entry,
3755 uint64_t key, void *cb_ctx __rte_unused)
3757 struct mlx5_flow_mreg_copy_resource *mcp_res =
3758 container_of(entry, typeof(*mcp_res), hlist_ent);
3760 return mcp_res->mark_id != key;
3763 struct mlx5_hlist_entry *
3764 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3767 struct rte_eth_dev *dev = list->ctx;
3768 struct mlx5_priv *priv = dev->data->dev_private;
3769 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3770 struct mlx5_flow_mreg_copy_resource *mcp_res;
3771 struct rte_flow_error *error = ctx->error;
3774 uint32_t mark_id = key;
3775 struct rte_flow_attr attr = {
3776 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3779 struct mlx5_rte_flow_item_tag tag_spec = {
3782 struct rte_flow_item items[] = {
3783 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3785 struct rte_flow_action_mark ftag = {
3788 struct mlx5_flow_action_copy_mreg cp_mreg = {
3792 struct rte_flow_action_jump jump = {
3793 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3795 struct rte_flow_action actions[] = {
3796 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3799 /* Fill the register fileds in the flow. */
3800 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3804 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3808 /* Provide the full width of FLAG specific value. */
3809 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3810 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3811 /* Build a new flow. */
3812 if (mark_id != MLX5_DEFAULT_COPY_ID) {
3813 items[0] = (struct rte_flow_item){
3814 .type = (enum rte_flow_item_type)
3815 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3818 items[1] = (struct rte_flow_item){
3819 .type = RTE_FLOW_ITEM_TYPE_END,
3821 actions[0] = (struct rte_flow_action){
3822 .type = (enum rte_flow_action_type)
3823 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3826 actions[1] = (struct rte_flow_action){
3827 .type = (enum rte_flow_action_type)
3828 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3831 actions[2] = (struct rte_flow_action){
3832 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3835 actions[3] = (struct rte_flow_action){
3836 .type = RTE_FLOW_ACTION_TYPE_END,
3839 /* Default rule, wildcard match. */
3840 attr.priority = MLX5_FLOW_PRIO_RSVD;
3841 items[0] = (struct rte_flow_item){
3842 .type = RTE_FLOW_ITEM_TYPE_END,
3844 actions[0] = (struct rte_flow_action){
3845 .type = (enum rte_flow_action_type)
3846 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3849 actions[1] = (struct rte_flow_action){
3850 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3853 actions[2] = (struct rte_flow_action){
3854 .type = RTE_FLOW_ACTION_TYPE_END,
3857 /* Build a new entry. */
3858 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3864 mcp_res->mark_id = mark_id;
3866 * The copy Flows are not included in any list. There
3867 * ones are referenced from other Flows and can not
3868 * be applied, removed, deleted in ardbitrary order
3869 * by list traversing.
3871 mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3872 actions, false, error);
3873 if (!mcp_res->rix_flow) {
3874 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3877 return &mcp_res->hlist_ent;
3881 * Add a flow of copying flow metadata registers in RX_CP_TBL.
3883 * As mark_id is unique, if there's already a registered flow for the mark_id,
3884 * return by increasing the reference counter of the resource. Otherwise, create
3885 * the resource (mcp_res) and flow.
3888 * - If ingress port is ANY and reg_c[1] is mark_id,
3889 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3891 * For default flow (zero mark_id), flow is like,
3892 * - If ingress port is ANY,
3893 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
3896 * Pointer to Ethernet device.
3898 * ID of MARK action, zero means default flow for META.
3900 * Perform verbose error reporting if not NULL.
3903 * Associated resource on success, NULL otherwise and rte_errno is set.
3905 static struct mlx5_flow_mreg_copy_resource *
3906 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3907 struct rte_flow_error *error)
3909 struct mlx5_priv *priv = dev->data->dev_private;
3910 struct mlx5_hlist_entry *entry;
3911 struct mlx5_flow_cb_ctx ctx = {
3916 /* Check if already registered. */
3917 MLX5_ASSERT(priv->mreg_cp_tbl);
3918 entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3921 return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3926 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3928 struct mlx5_flow_mreg_copy_resource *mcp_res =
3929 container_of(entry, typeof(*mcp_res), hlist_ent);
3930 struct rte_eth_dev *dev = list->ctx;
3931 struct mlx5_priv *priv = dev->data->dev_private;
3933 MLX5_ASSERT(mcp_res->rix_flow);
3934 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3935 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3939 * Release flow in RX_CP_TBL.
3942 * Pointer to Ethernet device.
3944 * Parent flow for wich copying is provided.
3947 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3948 struct rte_flow *flow)
3950 struct mlx5_flow_mreg_copy_resource *mcp_res;
3951 struct mlx5_priv *priv = dev->data->dev_private;
3953 if (!flow->rix_mreg_copy)
3955 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3956 flow->rix_mreg_copy);
3957 if (!mcp_res || !priv->mreg_cp_tbl)
3959 MLX5_ASSERT(mcp_res->rix_flow);
3960 mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3961 flow->rix_mreg_copy = 0;
3965 * Remove the default copy action from RX_CP_TBL.
3967 * This functions is called in the mlx5_dev_start(). No thread safe
3971 * Pointer to Ethernet device.
3974 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3976 struct mlx5_hlist_entry *entry;
3977 struct mlx5_priv *priv = dev->data->dev_private;
3979 /* Check if default flow is registered. */
3980 if (!priv->mreg_cp_tbl)
3982 entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
3983 MLX5_DEFAULT_COPY_ID, NULL);
3986 mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
3990 * Add the default copy action in in RX_CP_TBL.
3992 * This functions is called in the mlx5_dev_start(). No thread safe
3996 * Pointer to Ethernet device.
3998 * Perform verbose error reporting if not NULL.
4001 * 0 for success, negative value otherwise and rte_errno is set.
4004 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4005 struct rte_flow_error *error)
4007 struct mlx5_priv *priv = dev->data->dev_private;
4008 struct mlx5_flow_mreg_copy_resource *mcp_res;
4010 /* Check whether extensive metadata feature is engaged. */
4011 if (!priv->config.dv_flow_en ||
4012 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4013 !mlx5_flow_ext_mreg_supported(dev) ||
4014 !priv->sh->dv_regc0_mask)
4017 * Add default mreg copy flow may be called multiple time, but
4018 * only be called once in stop. Avoid register it twice.
4020 if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4022 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4029 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4031 * All the flow having Q/RSS action should be split by
4032 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4033 * performs the following,
4034 * - CQE->flow_tag := reg_c[1] (MARK)
4035 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4036 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4037 * but there should be a flow per each MARK ID set by MARK action.
4039 * For the aforementioned reason, if there's a MARK action in flow's action
4040 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4041 * the MARK ID to CQE's flow_tag like,
4042 * - If reg_c[1] is mark_id,
4043 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4045 * For SET_META action which stores value in reg_c[0], as the destination is
4046 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4047 * MARK ID means the default flow. The default flow looks like,
4048 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4051 * Pointer to Ethernet device.
4053 * Pointer to flow structure.
4054 * @param[in] actions
4055 * Pointer to the list of actions.
4057 * Perform verbose error reporting if not NULL.
4060 * 0 on success, negative value otherwise and rte_errno is set.
4063 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4064 struct rte_flow *flow,
4065 const struct rte_flow_action *actions,
4066 struct rte_flow_error *error)
4068 struct mlx5_priv *priv = dev->data->dev_private;
4069 struct mlx5_dev_config *config = &priv->config;
4070 struct mlx5_flow_mreg_copy_resource *mcp_res;
4071 const struct rte_flow_action_mark *mark;
4073 /* Check whether extensive metadata feature is engaged. */
4074 if (!config->dv_flow_en ||
4075 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4076 !mlx5_flow_ext_mreg_supported(dev) ||
4077 !priv->sh->dv_regc0_mask)
4079 /* Find MARK action. */
4080 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4081 switch (actions->type) {
4082 case RTE_FLOW_ACTION_TYPE_FLAG:
4083 mcp_res = flow_mreg_add_copy_action
4084 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4087 flow->rix_mreg_copy = mcp_res->idx;
4089 case RTE_FLOW_ACTION_TYPE_MARK:
4090 mark = (const struct rte_flow_action_mark *)
4093 flow_mreg_add_copy_action(dev, mark->id, error);
4096 flow->rix_mreg_copy = mcp_res->idx;
4105 #define MLX5_MAX_SPLIT_ACTIONS 24
4106 #define MLX5_MAX_SPLIT_ITEMS 24
4109 * Split the hairpin flow.
4110 * Since HW can't support encap and push-vlan on Rx, we move these
4112 * If the count action is after the encap then we also
4113 * move the count action. in this case the count will also measure
4117 * Pointer to Ethernet device.
4118 * @param[in] actions
4119 * Associated actions (list terminated by the END action).
4120 * @param[out] actions_rx
4122 * @param[out] actions_tx
4124 * @param[out] pattern_tx
4125 * The pattern items for the Tx flow.
4126 * @param[out] flow_id
4127 * The flow ID connected to this flow.
4133 flow_hairpin_split(struct rte_eth_dev *dev,
4134 const struct rte_flow_action actions[],
4135 struct rte_flow_action actions_rx[],
4136 struct rte_flow_action actions_tx[],
4137 struct rte_flow_item pattern_tx[],
4140 const struct rte_flow_action_raw_encap *raw_encap;
4141 const struct rte_flow_action_raw_decap *raw_decap;
4142 struct mlx5_rte_flow_action_set_tag *set_tag;
4143 struct rte_flow_action *tag_action;
4144 struct mlx5_rte_flow_item_tag *tag_item;
4145 struct rte_flow_item *item;
4149 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4150 switch (actions->type) {
4151 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4152 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4153 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4154 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4155 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4156 rte_memcpy(actions_tx, actions,
4157 sizeof(struct rte_flow_action));
4160 case RTE_FLOW_ACTION_TYPE_COUNT:
4162 rte_memcpy(actions_tx, actions,
4163 sizeof(struct rte_flow_action));
4166 rte_memcpy(actions_rx, actions,
4167 sizeof(struct rte_flow_action));
4171 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4172 raw_encap = actions->conf;
4173 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4174 memcpy(actions_tx, actions,
4175 sizeof(struct rte_flow_action));
4179 rte_memcpy(actions_rx, actions,
4180 sizeof(struct rte_flow_action));
4184 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4185 raw_decap = actions->conf;
4186 if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4187 memcpy(actions_tx, actions,
4188 sizeof(struct rte_flow_action));
4191 rte_memcpy(actions_rx, actions,
4192 sizeof(struct rte_flow_action));
4197 rte_memcpy(actions_rx, actions,
4198 sizeof(struct rte_flow_action));
4203 /* Add set meta action and end action for the Rx flow. */
4204 tag_action = actions_rx;
4205 tag_action->type = (enum rte_flow_action_type)
4206 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4208 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4210 set_tag = (void *)actions_rx;
4211 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4212 MLX5_ASSERT(set_tag->id > REG_NON);
4213 set_tag->data = flow_id;
4214 tag_action->conf = set_tag;
4215 /* Create Tx item list. */
4216 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4217 addr = (void *)&pattern_tx[2];
4219 item->type = (enum rte_flow_item_type)
4220 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4221 tag_item = (void *)addr;
4222 tag_item->data = flow_id;
4223 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4224 MLX5_ASSERT(set_tag->id > REG_NON);
4225 item->spec = tag_item;
4226 addr += sizeof(struct mlx5_rte_flow_item_tag);
4227 tag_item = (void *)addr;
4228 tag_item->data = UINT32_MAX;
4229 tag_item->id = UINT16_MAX;
4230 item->mask = tag_item;
4233 item->type = RTE_FLOW_ITEM_TYPE_END;
4238 * The last stage of splitting chain, just creates the subflow
4239 * without any modification.
4242 * Pointer to Ethernet device.
4244 * Parent flow structure pointer.
4245 * @param[in, out] sub_flow
4246 * Pointer to return the created subflow, may be NULL.
4248 * Flow rule attributes.
4250 * Pattern specification (list terminated by the END pattern item).
4251 * @param[in] actions
4252 * Associated actions (list terminated by the END action).
4253 * @param[in] flow_split_info
4254 * Pointer to flow split info structure.
4256 * Perform verbose error reporting if not NULL.
4258 * 0 on success, negative value otherwise
4261 flow_create_split_inner(struct rte_eth_dev *dev,
4262 struct rte_flow *flow,
4263 struct mlx5_flow **sub_flow,
4264 const struct rte_flow_attr *attr,
4265 const struct rte_flow_item items[],
4266 const struct rte_flow_action actions[],
4267 struct mlx5_flow_split_info *flow_split_info,
4268 struct rte_flow_error *error)
4270 struct mlx5_flow *dev_flow;
4272 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4273 flow_split_info->flow_idx, error);
4276 dev_flow->flow = flow;
4277 dev_flow->external = flow_split_info->external;
4278 dev_flow->skip_scale = flow_split_info->skip_scale;
4279 /* Subflow object was created, we must include one in the list. */
4280 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4281 dev_flow->handle, next);
4283 * If dev_flow is as one of the suffix flow, some actions in suffix
4284 * flow may need some user defined item layer flags, and pass the
4285 * Metadate rxq mark flag to suffix flow as well.
4287 if (flow_split_info->prefix_layers)
4288 dev_flow->handle->layers = flow_split_info->prefix_layers;
4289 if (flow_split_info->prefix_mark)
4290 dev_flow->handle->mark = 1;
4292 *sub_flow = dev_flow;
4293 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4297 * Split the meter flow.
4299 * As meter flow will split to three sub flow, other than meter
4300 * action, the other actions make sense to only meter accepts
4301 * the packet. If it need to be dropped, no other additional
4302 * actions should be take.
4304 * One kind of special action which decapsulates the L3 tunnel
4305 * header will be in the prefix sub flow, as not to take the
4306 * L3 tunnel header into account.
4309 * Pointer to Ethernet device.
4311 * Pattern specification (list terminated by the END pattern item).
4312 * @param[out] sfx_items
4313 * Suffix flow match items (list terminated by the END pattern item).
4314 * @param[in] actions
4315 * Associated actions (list terminated by the END action).
4316 * @param[out] actions_sfx
4317 * Suffix flow actions.
4318 * @param[out] actions_pre
4319 * Prefix flow actions.
4320 * @param[out] pattern_sfx
4321 * The pattern items for the suffix flow.
4322 * @param[out] tag_sfx
4323 * Pointer to suffix flow tag.
4329 flow_meter_split_prep(struct rte_eth_dev *dev,
4330 const struct rte_flow_item items[],
4331 struct rte_flow_item sfx_items[],
4332 const struct rte_flow_action actions[],
4333 struct rte_flow_action actions_sfx[],
4334 struct rte_flow_action actions_pre[])
4336 struct mlx5_priv *priv = dev->data->dev_private;
4337 struct rte_flow_action *tag_action = NULL;
4338 struct rte_flow_item *tag_item;
4339 struct mlx5_rte_flow_action_set_tag *set_tag;
4340 struct rte_flow_error error;
4341 const struct rte_flow_action_raw_encap *raw_encap;
4342 const struct rte_flow_action_raw_decap *raw_decap;
4343 struct mlx5_rte_flow_item_tag *tag_spec;
4344 struct mlx5_rte_flow_item_tag *tag_mask;
4345 uint32_t tag_id = 0;
4346 bool copy_vlan = false;
4348 /* Prepare the actions for prefix and suffix flow. */
4349 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4350 struct rte_flow_action **action_cur = NULL;
4352 switch (actions->type) {
4353 case RTE_FLOW_ACTION_TYPE_METER:
4354 /* Add the extra tag action first. */
4355 tag_action = actions_pre;
4356 tag_action->type = (enum rte_flow_action_type)
4357 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4359 action_cur = &actions_pre;
4361 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4362 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4363 action_cur = &actions_pre;
4365 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4366 raw_encap = actions->conf;
4367 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4368 action_cur = &actions_pre;
4370 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4371 raw_decap = actions->conf;
4372 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4373 action_cur = &actions_pre;
4375 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4376 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4383 action_cur = &actions_sfx;
4384 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4387 /* Add end action to the actions. */
4388 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4389 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4392 set_tag = (void *)actions_pre;
4393 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4394 mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4396 if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4397 DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4398 dev->data->port_id);
4399 mlx5_ipool_free(priv->sh->ipool
4400 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4402 } else if (!tag_id) {
4405 set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4407 tag_action->conf = set_tag;
4408 /* Prepare the suffix subflow items. */
4409 tag_item = sfx_items++;
4410 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4411 int item_type = items->type;
4413 switch (item_type) {
4414 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4415 memcpy(sfx_items, items, sizeof(*sfx_items));
4418 case RTE_FLOW_ITEM_TYPE_VLAN:
4420 memcpy(sfx_items, items, sizeof(*sfx_items));
4422 * Convert to internal match item, it is used
4423 * for vlan push and set vid.
4425 sfx_items->type = (enum rte_flow_item_type)
4426 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4434 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4436 tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4437 tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4438 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4439 tag_mask = tag_spec + 1;
4440 tag_mask->data = 0xffffff00;
4441 tag_item->type = (enum rte_flow_item_type)
4442 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4443 tag_item->spec = tag_spec;
4444 tag_item->last = NULL;
4445 tag_item->mask = tag_mask;
4450 * Split action list having QUEUE/RSS for metadata register copy.
4452 * Once Q/RSS action is detected in user's action list, the flow action
4453 * should be split in order to copy metadata registers, which will happen in
4455 * - CQE->flow_tag := reg_c[1] (MARK)
4456 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4457 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4458 * This is because the last action of each flow must be a terminal action
4459 * (QUEUE, RSS or DROP).
4461 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4462 * stored and kept in the mlx5_flow structure per each sub_flow.
4464 * The Q/RSS action is replaced with,
4465 * - SET_TAG, setting the allocated flow ID to reg_c[2].
4466 * And the following JUMP action is added at the end,
4467 * - JUMP, to RX_CP_TBL.
4469 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4470 * flow_create_split_metadata() routine. The flow will look like,
4471 * - If flow ID matches (reg_c[2]), perform Q/RSS.
4474 * Pointer to Ethernet device.
4475 * @param[out] split_actions
4476 * Pointer to store split actions to jump to CP_TBL.
4477 * @param[in] actions
4478 * Pointer to the list of original flow actions.
4480 * Pointer to the Q/RSS action.
4481 * @param[in] actions_n
4482 * Number of original actions.
4484 * Perform verbose error reporting if not NULL.
4487 * non-zero unique flow_id on success, otherwise 0 and
4488 * error/rte_error are set.
4491 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4492 struct rte_flow_action *split_actions,
4493 const struct rte_flow_action *actions,
4494 const struct rte_flow_action *qrss,
4495 int actions_n, struct rte_flow_error *error)
4497 struct mlx5_priv *priv = dev->data->dev_private;
4498 struct mlx5_rte_flow_action_set_tag *set_tag;
4499 struct rte_flow_action_jump *jump;
4500 const int qrss_idx = qrss - actions;
4501 uint32_t flow_id = 0;
4505 * Given actions will be split
4506 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4507 * - Add jump to mreg CP_TBL.
4508 * As a result, there will be one more action.
4511 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4512 set_tag = (void *)(split_actions + actions_n);
4514 * If tag action is not set to void(it means we are not the meter
4515 * suffix flow), add the tag action. Since meter suffix flow already
4516 * has the tag added.
4518 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4520 * Allocate the new subflow ID. This one is unique within
4521 * device and not shared with representors. Otherwise,
4522 * we would have to resolve multi-thread access synch
4523 * issue. Each flow on the shared device is appended
4524 * with source vport identifier, so the resulting
4525 * flows will be unique in the shared (by master and
4526 * representors) domain even if they have coinciding
4529 mlx5_ipool_malloc(priv->sh->ipool
4530 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4532 return rte_flow_error_set(error, ENOMEM,
4533 RTE_FLOW_ERROR_TYPE_ACTION,
4534 NULL, "can't allocate id "
4535 "for split Q/RSS subflow");
4536 /* Internal SET_TAG action to set flow ID. */
4537 *set_tag = (struct mlx5_rte_flow_action_set_tag){
4540 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4544 /* Construct new actions array. */
4545 /* Replace QUEUE/RSS action. */
4546 split_actions[qrss_idx] = (struct rte_flow_action){
4547 .type = (enum rte_flow_action_type)
4548 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4552 /* JUMP action to jump to mreg copy table (CP_TBL). */
4553 jump = (void *)(set_tag + 1);
4554 *jump = (struct rte_flow_action_jump){
4555 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4557 split_actions[actions_n - 2] = (struct rte_flow_action){
4558 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4561 split_actions[actions_n - 1] = (struct rte_flow_action){
4562 .type = RTE_FLOW_ACTION_TYPE_END,
4568 * Extend the given action list for Tx metadata copy.
4570 * Copy the given action list to the ext_actions and add flow metadata register
4571 * copy action in order to copy reg_a set by WQE to reg_c[0].
4573 * @param[out] ext_actions
4574 * Pointer to the extended action list.
4575 * @param[in] actions
4576 * Pointer to the list of actions.
4577 * @param[in] actions_n
4578 * Number of actions in the list.
4580 * Perform verbose error reporting if not NULL.
4581 * @param[in] encap_idx
4582 * The encap action inndex.
4585 * 0 on success, negative value otherwise
4588 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4589 struct rte_flow_action *ext_actions,
4590 const struct rte_flow_action *actions,
4591 int actions_n, struct rte_flow_error *error,
4594 struct mlx5_flow_action_copy_mreg *cp_mreg =
4595 (struct mlx5_flow_action_copy_mreg *)
4596 (ext_actions + actions_n + 1);
4599 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4603 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4608 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4609 if (encap_idx == actions_n - 1) {
4610 ext_actions[actions_n - 1] = (struct rte_flow_action){
4611 .type = (enum rte_flow_action_type)
4612 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4615 ext_actions[actions_n] = (struct rte_flow_action){
4616 .type = RTE_FLOW_ACTION_TYPE_END,
4619 ext_actions[encap_idx] = (struct rte_flow_action){
4620 .type = (enum rte_flow_action_type)
4621 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4624 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4625 sizeof(*ext_actions) * (actions_n - encap_idx));
4631 * Check the match action from the action list.
4633 * @param[in] actions
4634 * Pointer to the list of actions.
4636 * Flow rule attributes.
4638 * The action to be check if exist.
4639 * @param[out] match_action_pos
4640 * Pointer to the position of the matched action if exists, otherwise is -1.
4641 * @param[out] qrss_action_pos
4642 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4643 * @param[out] modify_after_mirror
4644 * Pointer to the flag of modify action after FDB mirroring.
4647 * > 0 the total number of actions.
4648 * 0 if not found match action in action list.
4651 flow_check_match_action(const struct rte_flow_action actions[],
4652 const struct rte_flow_attr *attr,
4653 enum rte_flow_action_type action,
4654 int *match_action_pos, int *qrss_action_pos,
4655 int *modify_after_mirror)
4657 const struct rte_flow_action_sample *sample;
4664 *match_action_pos = -1;
4665 *qrss_action_pos = -1;
4666 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4667 if (actions->type == action) {
4669 *match_action_pos = actions_n;
4671 switch (actions->type) {
4672 case RTE_FLOW_ACTION_TYPE_QUEUE:
4673 case RTE_FLOW_ACTION_TYPE_RSS:
4674 *qrss_action_pos = actions_n;
4676 case RTE_FLOW_ACTION_TYPE_SAMPLE:
4677 sample = actions->conf;
4678 ratio = sample->ratio;
4679 sub_type = ((const struct rte_flow_action *)
4680 (sample->actions))->type;
4681 if (ratio == 1 && attr->transfer)
4684 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4685 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4686 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4687 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4688 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4689 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4690 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4691 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4692 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4693 case RTE_FLOW_ACTION_TYPE_SET_TTL:
4694 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4695 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4696 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4697 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4698 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4699 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4700 case RTE_FLOW_ACTION_TYPE_FLAG:
4701 case RTE_FLOW_ACTION_TYPE_MARK:
4702 case RTE_FLOW_ACTION_TYPE_SET_META:
4703 case RTE_FLOW_ACTION_TYPE_SET_TAG:
4705 *modify_after_mirror = 1;
4712 if (flag && fdb_mirror && !*modify_after_mirror) {
4713 /* FDB mirroring uses the destination array to implement
4714 * instead of FLOW_SAMPLER object.
4716 if (sub_type != RTE_FLOW_ACTION_TYPE_END)
4719 /* Count RTE_FLOW_ACTION_TYPE_END. */
4720 return flag ? actions_n + 1 : 0;
4723 #define SAMPLE_SUFFIX_ITEM 2
4726 * Split the sample flow.
4728 * As sample flow will split to two sub flow, sample flow with
4729 * sample action, the other actions will move to new suffix flow.
4731 * Also add unique tag id with tag action in the sample flow,
4732 * the same tag id will be as match in the suffix flow.
4735 * Pointer to Ethernet device.
4736 * @param[in] add_tag
4737 * Add extra tag action flag.
4738 * @param[out] sfx_items
4739 * Suffix flow match items (list terminated by the END pattern item).
4740 * @param[in] actions
4741 * Associated actions (list terminated by the END action).
4742 * @param[out] actions_sfx
4743 * Suffix flow actions.
4744 * @param[out] actions_pre
4745 * Prefix flow actions.
4746 * @param[in] actions_n
4747 * The total number of actions.
4748 * @param[in] sample_action_pos
4749 * The sample action position.
4750 * @param[in] qrss_action_pos
4751 * The Queue/RSS action position.
4752 * @param[in] jump_table
4753 * Add extra jump action flag.
4755 * Perform verbose error reporting if not NULL.
4758 * 0 on success, or unique flow_id, a negative errno value
4759 * otherwise and rte_errno is set.
4762 flow_sample_split_prep(struct rte_eth_dev *dev,
4764 struct rte_flow_item sfx_items[],
4765 const struct rte_flow_action actions[],
4766 struct rte_flow_action actions_sfx[],
4767 struct rte_flow_action actions_pre[],
4769 int sample_action_pos,
4770 int qrss_action_pos,
4772 struct rte_flow_error *error)
4774 struct mlx5_priv *priv = dev->data->dev_private;
4775 struct mlx5_rte_flow_action_set_tag *set_tag;
4776 struct mlx5_rte_flow_item_tag *tag_spec;
4777 struct mlx5_rte_flow_item_tag *tag_mask;
4778 struct rte_flow_action_jump *jump_action;
4779 uint32_t tag_id = 0;
4781 int append_index = 0;
4784 if (sample_action_pos < 0)
4785 return rte_flow_error_set(error, EINVAL,
4786 RTE_FLOW_ERROR_TYPE_ACTION,
4787 NULL, "invalid position of sample "
4789 /* Prepare the actions for prefix and suffix flow. */
4790 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4791 index = qrss_action_pos;
4792 /* Put the preceding the Queue/RSS action into prefix flow. */
4794 memcpy(actions_pre, actions,
4795 sizeof(struct rte_flow_action) * index);
4796 /* Put others preceding the sample action into prefix flow. */
4797 if (sample_action_pos > index + 1)
4798 memcpy(actions_pre + index, actions + index + 1,
4799 sizeof(struct rte_flow_action) *
4800 (sample_action_pos - index - 1));
4801 index = sample_action_pos - 1;
4802 /* Put Queue/RSS action into Suffix flow. */
4803 memcpy(actions_sfx, actions + qrss_action_pos,
4804 sizeof(struct rte_flow_action));
4807 index = sample_action_pos;
4809 memcpy(actions_pre, actions,
4810 sizeof(struct rte_flow_action) * index);
4812 /* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
4813 * For CX6DX and above, metadata registers Cx preserve their value,
4814 * add an extra tag action for NIC-RX and E-Switch Domain.
4817 /* Prepare the prefix tag action. */
4819 set_tag = (void *)(actions_pre + actions_n + append_index);
4820 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4824 mlx5_ipool_malloc(priv->sh->ipool
4825 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4826 set_tag->data = tag_id;
4827 /* Prepare the suffix subflow items. */
4828 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4829 tag_spec->data = tag_id;
4830 tag_spec->id = set_tag->id;
4831 tag_mask = tag_spec + 1;
4832 tag_mask->data = UINT32_MAX;
4833 sfx_items[0] = (struct rte_flow_item){
4834 .type = (enum rte_flow_item_type)
4835 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4840 sfx_items[1] = (struct rte_flow_item){
4841 .type = (enum rte_flow_item_type)
4842 RTE_FLOW_ITEM_TYPE_END,
4844 /* Prepare the tag action in prefix subflow. */
4845 actions_pre[index++] =
4846 (struct rte_flow_action){
4847 .type = (enum rte_flow_action_type)
4848 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4852 memcpy(actions_pre + index, actions + sample_action_pos,
4853 sizeof(struct rte_flow_action));
4855 /* For the modify action after the sample action in E-Switch mirroring,
4856 * Add the extra jump action in prefix subflow and jump into the next
4857 * table, then do the modify action in the new table.
4860 /* Prepare the prefix jump action. */
4862 jump_action = (void *)(actions_pre + actions_n + append_index);
4863 jump_action->group = jump_table;
4864 actions_pre[index++] =
4865 (struct rte_flow_action){
4866 .type = (enum rte_flow_action_type)
4867 RTE_FLOW_ACTION_TYPE_JUMP,
4868 .conf = jump_action,
4871 actions_pre[index] = (struct rte_flow_action){
4872 .type = (enum rte_flow_action_type)
4873 RTE_FLOW_ACTION_TYPE_END,
4875 /* Put the actions after sample into Suffix flow. */
4876 memcpy(actions_sfx, actions + sample_action_pos + 1,
4877 sizeof(struct rte_flow_action) *
4878 (actions_n - sample_action_pos - 1));
4883 * The splitting for metadata feature.
4885 * - Q/RSS action on NIC Rx should be split in order to pass by
4886 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4887 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4889 * - All the actions on NIC Tx should have a mreg copy action to
4890 * copy reg_a from WQE to reg_c[0].
4893 * Pointer to Ethernet device.
4895 * Parent flow structure pointer.
4897 * Flow rule attributes.
4899 * Pattern specification (list terminated by the END pattern item).
4900 * @param[in] actions
4901 * Associated actions (list terminated by the END action).
4902 * @param[in] flow_split_info
4903 * Pointer to flow split info structure.
4905 * Perform verbose error reporting if not NULL.
4907 * 0 on success, negative value otherwise
4910 flow_create_split_metadata(struct rte_eth_dev *dev,
4911 struct rte_flow *flow,
4912 const struct rte_flow_attr *attr,
4913 const struct rte_flow_item items[],
4914 const struct rte_flow_action actions[],
4915 struct mlx5_flow_split_info *flow_split_info,
4916 struct rte_flow_error *error)
4918 struct mlx5_priv *priv = dev->data->dev_private;
4919 struct mlx5_dev_config *config = &priv->config;
4920 const struct rte_flow_action *qrss = NULL;
4921 struct rte_flow_action *ext_actions = NULL;
4922 struct mlx5_flow *dev_flow = NULL;
4923 uint32_t qrss_id = 0;
4930 /* Check whether extensive metadata feature is engaged. */
4931 if (!config->dv_flow_en ||
4932 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4933 !mlx5_flow_ext_mreg_supported(dev))
4934 return flow_create_split_inner(dev, flow, NULL, attr, items,
4935 actions, flow_split_info, error);
4936 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4939 /* Exclude hairpin flows from splitting. */
4940 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4941 const struct rte_flow_action_queue *queue;
4944 if (mlx5_rxq_get_type(dev, queue->index) ==
4945 MLX5_RXQ_TYPE_HAIRPIN)
4947 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4948 const struct rte_flow_action_rss *rss;
4951 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4952 MLX5_RXQ_TYPE_HAIRPIN)
4957 /* Check if it is in meter suffix table. */
4958 mtr_sfx = attr->group == (attr->transfer ?
4959 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4960 MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4962 * Q/RSS action on NIC Rx should be split in order to pass by
4963 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4964 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4966 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4967 sizeof(struct rte_flow_action_set_tag) +
4968 sizeof(struct rte_flow_action_jump);
4969 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4972 return rte_flow_error_set(error, ENOMEM,
4973 RTE_FLOW_ERROR_TYPE_ACTION,
4974 NULL, "no memory to split "
4977 * If we are the suffix flow of meter, tag already exist.
4978 * Set the tag action to void.
4981 ext_actions[qrss - actions].type =
4982 RTE_FLOW_ACTION_TYPE_VOID;
4984 ext_actions[qrss - actions].type =
4985 (enum rte_flow_action_type)
4986 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4988 * Create the new actions list with removed Q/RSS action
4989 * and appended set tag and jump to register copy table
4990 * (RX_CP_TBL). We should preallocate unique tag ID here
4991 * in advance, because it is needed for set tag action.
4993 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4994 qrss, actions_n, error);
4995 if (!mtr_sfx && !qrss_id) {
4999 } else if (attr->egress && !attr->transfer) {
5001 * All the actions on NIC Tx should have a metadata register
5002 * copy action to copy reg_a from WQE to reg_c[meta]
5004 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5005 sizeof(struct mlx5_flow_action_copy_mreg);
5006 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5009 return rte_flow_error_set(error, ENOMEM,
5010 RTE_FLOW_ERROR_TYPE_ACTION,
5011 NULL, "no memory to split "
5013 /* Create the action list appended with copy register. */
5014 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5015 actions_n, error, encap_idx);
5019 /* Add the unmodified original or prefix subflow. */
5020 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5021 items, ext_actions ? ext_actions :
5022 actions, flow_split_info, error);
5025 MLX5_ASSERT(dev_flow);
5027 const struct rte_flow_attr q_attr = {
5028 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5031 /* Internal PMD action to set register. */
5032 struct mlx5_rte_flow_item_tag q_tag_spec = {
5036 struct rte_flow_item q_items[] = {
5038 .type = (enum rte_flow_item_type)
5039 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5040 .spec = &q_tag_spec,
5045 .type = RTE_FLOW_ITEM_TYPE_END,
5048 struct rte_flow_action q_actions[] = {
5054 .type = RTE_FLOW_ACTION_TYPE_END,
5057 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5060 * Configure the tag item only if there is no meter subflow.
5061 * Since tag is already marked in the meter suffix subflow
5062 * we can just use the meter suffix items as is.
5065 /* Not meter subflow. */
5066 MLX5_ASSERT(!mtr_sfx);
5068 * Put unique id in prefix flow due to it is destroyed
5069 * after suffix flow and id will be freed after there
5070 * is no actual flows with this id and identifier
5071 * reallocation becomes possible (for example, for
5072 * other flows in other threads).
5074 dev_flow->handle->split_flow_id = qrss_id;
5075 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5079 q_tag_spec.id = ret;
5082 /* Add suffix subflow to execute Q/RSS. */
5083 flow_split_info->prefix_layers = layers;
5084 flow_split_info->prefix_mark = 0;
5085 ret = flow_create_split_inner(dev, flow, &dev_flow,
5086 &q_attr, mtr_sfx ? items :
5088 flow_split_info, error);
5091 /* qrss ID should be freed if failed. */
5093 MLX5_ASSERT(dev_flow);
5098 * We do not destroy the partially created sub_flows in case of error.
5099 * These ones are included into parent flow list and will be destroyed
5100 * by flow_drv_destroy.
5102 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5104 mlx5_free(ext_actions);
5109 * The splitting for meter feature.
5111 * - The meter flow will be split to two flows as prefix and
5112 * suffix flow. The packets make sense only it pass the prefix
5115 * - Reg_C_5 is used for the packet to match betweend prefix and
5119 * Pointer to Ethernet device.
5121 * Parent flow structure pointer.
5123 * Flow rule attributes.
5125 * Pattern specification (list terminated by the END pattern item).
5126 * @param[in] actions
5127 * Associated actions (list terminated by the END action).
5128 * @param[in] flow_split_info
5129 * Pointer to flow split info structure.
5131 * Perform verbose error reporting if not NULL.
5133 * 0 on success, negative value otherwise
5136 flow_create_split_meter(struct rte_eth_dev *dev,
5137 struct rte_flow *flow,
5138 const struct rte_flow_attr *attr,
5139 const struct rte_flow_item items[],
5140 const struct rte_flow_action actions[],
5141 struct mlx5_flow_split_info *flow_split_info,
5142 struct rte_flow_error *error)
5144 struct mlx5_priv *priv = dev->data->dev_private;
5145 struct rte_flow_action *sfx_actions = NULL;
5146 struct rte_flow_action *pre_actions = NULL;
5147 struct rte_flow_item *sfx_items = NULL;
5148 struct mlx5_flow *dev_flow = NULL;
5149 struct rte_flow_attr sfx_attr = *attr;
5151 uint32_t mtr_tag_id = 0;
5158 actions_n = flow_check_meter_action(actions, &mtr);
5160 /* The five prefix actions: meter, decap, encap, tag, end. */
5161 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5162 sizeof(struct mlx5_rte_flow_action_set_tag);
5163 /* tag, vlan, port id, end. */
5164 #define METER_SUFFIX_ITEM 4
5165 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5166 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5167 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5170 return rte_flow_error_set(error, ENOMEM,
5171 RTE_FLOW_ERROR_TYPE_ACTION,
5172 NULL, "no memory to split "
5174 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5176 pre_actions = sfx_actions + actions_n;
5177 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5178 actions, sfx_actions,
5184 /* Add the prefix subflow. */
5185 flow_split_info->prefix_mark = 0;
5186 ret = flow_create_split_inner(dev, flow, &dev_flow,
5187 attr, items, pre_actions,
5188 flow_split_info, error);
5193 dev_flow->handle->split_flow_id = mtr_tag_id;
5194 /* Setting the sfx group atrr. */
5195 sfx_attr.group = sfx_attr.transfer ?
5196 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5197 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5198 flow_split_info->prefix_layers =
5199 flow_get_prefix_layer_flags(dev_flow);
5200 flow_split_info->prefix_mark = dev_flow->handle->mark;
5202 /* Add the prefix subflow. */
5203 ret = flow_create_split_metadata(dev, flow,
5204 &sfx_attr, sfx_items ?
5206 sfx_actions ? sfx_actions : actions,
5207 flow_split_info, error);
5210 mlx5_free(sfx_actions);
5215 * The splitting for sample feature.
5217 * Once Sample action is detected in the action list, the flow actions should
5218 * be split into prefix sub flow and suffix sub flow.
5220 * The original items remain in the prefix sub flow, all actions preceding the
5221 * sample action and the sample action itself will be copied to the prefix
5222 * sub flow, the actions following the sample action will be copied to the
5223 * suffix sub flow, Queue action always be located in the suffix sub flow.
5225 * In order to make the packet from prefix sub flow matches with suffix sub
5226 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5227 * flow uses tag item with the unique flow id.
5230 * Pointer to Ethernet device.
5232 * Parent flow structure pointer.
5234 * Flow rule attributes.
5236 * Pattern specification (list terminated by the END pattern item).
5237 * @param[in] actions
5238 * Associated actions (list terminated by the END action).
5239 * @param[in] flow_split_info
5240 * Pointer to flow split info structure.
5242 * Perform verbose error reporting if not NULL.
5244 * 0 on success, negative value otherwise
5247 flow_create_split_sample(struct rte_eth_dev *dev,
5248 struct rte_flow *flow,
5249 const struct rte_flow_attr *attr,
5250 const struct rte_flow_item items[],
5251 const struct rte_flow_action actions[],
5252 struct mlx5_flow_split_info *flow_split_info,
5253 struct rte_flow_error *error)
5255 struct mlx5_priv *priv = dev->data->dev_private;
5256 struct rte_flow_action *sfx_actions = NULL;
5257 struct rte_flow_action *pre_actions = NULL;
5258 struct rte_flow_item *sfx_items = NULL;
5259 struct mlx5_flow *dev_flow = NULL;
5260 struct rte_flow_attr sfx_attr = *attr;
5261 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5262 struct mlx5_flow_dv_sample_resource *sample_res;
5263 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5264 struct mlx5_flow_tbl_resource *sfx_tbl;
5268 uint32_t fdb_tx = 0;
5271 int sample_action_pos;
5272 int qrss_action_pos;
5274 int modify_after_mirror = 0;
5275 uint16_t jump_table = 0;
5276 const uint32_t next_ft_step = 1;
5279 if (priv->sampler_en)
5280 actions_n = flow_check_match_action(actions, attr,
5281 RTE_FLOW_ACTION_TYPE_SAMPLE,
5282 &sample_action_pos, &qrss_action_pos,
5283 &modify_after_mirror);
5285 /* The prefix actions must includes sample, tag, end. */
5286 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5287 + sizeof(struct mlx5_rte_flow_action_set_tag);
5288 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5289 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5290 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5291 item_size), 0, SOCKET_ID_ANY);
5293 return rte_flow_error_set(error, ENOMEM,
5294 RTE_FLOW_ERROR_TYPE_ACTION,
5295 NULL, "no memory to split "
5297 /* The representor_id is -1 for uplink. */
5298 fdb_tx = (attr->transfer && priv->representor_id != -1);
5300 * When reg_c_preserve is set, metadata registers Cx preserve
5301 * their value even through packet duplication.
5303 add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5305 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5307 if (modify_after_mirror)
5308 jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5310 pre_actions = sfx_actions + actions_n;
5311 tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5312 actions, sfx_actions,
5313 pre_actions, actions_n,
5315 qrss_action_pos, jump_table,
5317 if (tag_id < 0 || (add_tag && !tag_id)) {
5321 if (modify_after_mirror)
5322 flow_split_info->skip_scale =
5323 1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5324 /* Add the prefix subflow. */
5325 ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5327 flow_split_info, error);
5332 dev_flow->handle->split_flow_id = tag_id;
5333 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5334 if (!modify_after_mirror) {
5335 /* Set the sfx group attr. */
5336 sample_res = (struct mlx5_flow_dv_sample_resource *)
5337 dev_flow->dv.sample_res;
5338 sfx_tbl = (struct mlx5_flow_tbl_resource *)
5339 sample_res->normal_path_tbl;
5340 sfx_tbl_data = container_of(sfx_tbl,
5341 struct mlx5_flow_tbl_data_entry,
5343 sfx_attr.group = sfx_attr.transfer ?
5344 (sfx_tbl_data->table_id - 1) :
5345 sfx_tbl_data->table_id;
5347 MLX5_ASSERT(attr->transfer);
5348 sfx_attr.group = jump_table;
5350 flow_split_info->prefix_layers =
5351 flow_get_prefix_layer_flags(dev_flow);
5352 flow_split_info->prefix_mark = dev_flow->handle->mark;
5353 /* Suffix group level already be scaled with factor, set
5354 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5355 * again in translation.
5357 flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5360 /* Add the suffix subflow. */
5361 ret = flow_create_split_meter(dev, flow, &sfx_attr,
5362 sfx_items ? sfx_items : items,
5363 sfx_actions ? sfx_actions : actions,
5364 flow_split_info, error);
5367 mlx5_free(sfx_actions);
5372 * Split the flow to subflow set. The splitters might be linked
5373 * in the chain, like this:
5374 * flow_create_split_outer() calls:
5375 * flow_create_split_meter() calls:
5376 * flow_create_split_metadata(meter_subflow_0) calls:
5377 * flow_create_split_inner(metadata_subflow_0)
5378 * flow_create_split_inner(metadata_subflow_1)
5379 * flow_create_split_inner(metadata_subflow_2)
5380 * flow_create_split_metadata(meter_subflow_1) calls:
5381 * flow_create_split_inner(metadata_subflow_0)
5382 * flow_create_split_inner(metadata_subflow_1)
5383 * flow_create_split_inner(metadata_subflow_2)
5385 * This provide flexible way to add new levels of flow splitting.
5386 * The all of successfully created subflows are included to the
5387 * parent flow dev_flow list.
5390 * Pointer to Ethernet device.
5392 * Parent flow structure pointer.
5394 * Flow rule attributes.
5396 * Pattern specification (list terminated by the END pattern item).
5397 * @param[in] actions
5398 * Associated actions (list terminated by the END action).
5399 * @param[in] flow_split_info
5400 * Pointer to flow split info structure.
5402 * Perform verbose error reporting if not NULL.
5404 * 0 on success, negative value otherwise
5407 flow_create_split_outer(struct rte_eth_dev *dev,
5408 struct rte_flow *flow,
5409 const struct rte_flow_attr *attr,
5410 const struct rte_flow_item items[],
5411 const struct rte_flow_action actions[],
5412 struct mlx5_flow_split_info *flow_split_info,
5413 struct rte_flow_error *error)
5417 ret = flow_create_split_sample(dev, flow, attr, items,
5418 actions, flow_split_info, error);
5419 MLX5_ASSERT(ret <= 0);
5423 static struct mlx5_flow_tunnel *
5424 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5425 const struct rte_flow_attr *attr,
5426 const struct rte_flow_item items[],
5427 const struct rte_flow_action actions[])
5429 struct mlx5_flow_tunnel *tunnel;
5431 #pragma GCC diagnostic push
5432 #pragma GCC diagnostic ignored "-Wcast-qual"
5433 if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5434 tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5435 else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5436 tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5439 #pragma GCC diagnostic pop
5445 * Adjust flow RSS workspace if needed.
5448 * Pointer to thread flow work space.
5450 * Pointer to RSS descriptor.
5451 * @param[in] nrssq_num
5452 * New RSS queue number.
5455 * 0 on success, -1 otherwise and rte_errno is set.
5458 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5459 struct mlx5_flow_rss_desc *rss_desc,
5462 if (likely(nrssq_num <= wks->rssq_num))
5464 rss_desc->queue = realloc(rss_desc->queue,
5465 sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5466 if (!rss_desc->queue) {
5470 wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5475 * Create a flow and add it to @p list.
5478 * Pointer to Ethernet device.
5480 * Pointer to a TAILQ flow list. If this parameter NULL,
5481 * no list insertion occurred, flow is just created,
5482 * this is caller's responsibility to track the
5485 * Flow rule attributes.
5487 * Pattern specification (list terminated by the END pattern item).
5488 * @param[in] actions
5489 * Associated actions (list terminated by the END action).
5490 * @param[in] external
5491 * This flow rule is created by request external to PMD.
5493 * Perform verbose error reporting if not NULL.
5496 * A flow index on success, 0 otherwise and rte_errno is set.
5499 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5500 const struct rte_flow_attr *attr,
5501 const struct rte_flow_item items[],
5502 const struct rte_flow_action original_actions[],
5503 bool external, struct rte_flow_error *error)
5505 struct mlx5_priv *priv = dev->data->dev_private;
5506 struct rte_flow *flow = NULL;
5507 struct mlx5_flow *dev_flow;
5508 const struct rte_flow_action_rss *rss = NULL;
5509 struct mlx5_translated_shared_action
5510 shared_actions[MLX5_MAX_SHARED_ACTIONS];
5511 int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5513 struct mlx5_flow_expand_rss buf;
5514 uint8_t buffer[2048];
5517 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5518 uint8_t buffer[2048];
5521 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5522 uint8_t buffer[2048];
5523 } actions_hairpin_tx;
5525 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5526 uint8_t buffer[2048];
5528 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5529 struct mlx5_flow_rss_desc *rss_desc;
5530 const struct rte_flow_action *p_actions_rx;
5534 struct rte_flow_attr attr_tx = { .priority = 0 };
5535 const struct rte_flow_action *actions;
5536 struct rte_flow_action *translated_actions = NULL;
5537 struct mlx5_flow_tunnel *tunnel;
5538 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5539 struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5540 struct mlx5_flow_split_info flow_split_info = {
5541 .external = !!external,
5550 rss_desc = &wks->rss_desc;
5551 ret = flow_shared_actions_translate(dev, original_actions,
5554 &translated_actions, error);
5556 MLX5_ASSERT(translated_actions == NULL);
5559 actions = translated_actions ? translated_actions : original_actions;
5560 p_actions_rx = actions;
5561 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5562 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5563 external, hairpin_flow, error);
5565 goto error_before_hairpin_split;
5566 flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5569 goto error_before_hairpin_split;
5571 if (hairpin_flow > 0) {
5572 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5574 goto error_before_hairpin_split;
5576 flow_hairpin_split(dev, actions, actions_rx.actions,
5577 actions_hairpin_tx.actions, items_tx.items,
5579 p_actions_rx = actions_rx.actions;
5581 flow_split_info.flow_idx = idx;
5582 flow->drv_type = flow_get_drv_type(dev, attr);
5583 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5584 flow->drv_type < MLX5_FLOW_TYPE_MAX);
5585 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5586 /* RSS Action only works on NIC RX domain */
5587 if (attr->ingress && !attr->transfer)
5588 rss = flow_get_rss_action(p_actions_rx);
5590 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5593 * The following information is required by
5594 * mlx5_flow_hashfields_adjust() in advance.
5596 rss_desc->level = rss->level;
5597 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5598 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5600 flow->dev_handles = 0;
5601 if (rss && rss->types) {
5602 unsigned int graph_root;
5604 graph_root = find_graph_root(items, rss->level);
5605 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5607 mlx5_support_expansion, graph_root);
5608 MLX5_ASSERT(ret > 0 &&
5609 (unsigned int)ret < sizeof(expand_buffer.buffer));
5612 buf->entry[0].pattern = (void *)(uintptr_t)items;
5614 rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5616 for (i = 0; i < buf->entries; ++i) {
5617 /* Initialize flow split data. */
5618 flow_split_info.prefix_layers = 0;
5619 flow_split_info.prefix_mark = 0;
5620 flow_split_info.skip_scale = 0;
5622 * The splitter may create multiple dev_flows,
5623 * depending on configuration. In the simplest
5624 * case it just creates unmodified original flow.
5626 ret = flow_create_split_outer(dev, flow, attr,
5627 buf->entry[i].pattern,
5628 p_actions_rx, &flow_split_info,
5632 if (is_flow_tunnel_steer_rule(dev, attr,
5633 buf->entry[i].pattern,
5635 ret = flow_tunnel_add_default_miss(dev, flow, attr,
5641 mlx5_free(default_miss_ctx.queue);
5646 /* Create the tx flow. */
5648 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5649 attr_tx.ingress = 0;
5651 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5652 actions_hairpin_tx.actions,
5656 dev_flow->flow = flow;
5657 dev_flow->external = 0;
5658 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5659 dev_flow->handle, next);
5660 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5662 actions_hairpin_tx.actions, error);
5667 * Update the metadata register copy table. If extensive
5668 * metadata feature is enabled and registers are supported
5669 * we might create the extra rte_flow for each unique
5670 * MARK/FLAG action ID.
5672 * The table is updated for ingress Flows only, because
5673 * the egress Flows belong to the different device and
5674 * copy table should be updated in peer NIC Rx domain.
5676 if (attr->ingress &&
5677 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5678 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5683 * If the flow is external (from application) OR device is started,
5684 * OR mreg discover, then apply immediately.
5686 if (external || dev->data->dev_started ||
5687 (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5688 attr->priority == MLX5_FLOW_PRIO_RSVD)) {
5689 ret = flow_drv_apply(dev, flow, error);
5694 rte_spinlock_lock(&priv->flow_list_lock);
5695 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5697 rte_spinlock_unlock(&priv->flow_list_lock);
5699 flow_rxq_flags_set(dev, flow);
5700 rte_free(translated_actions);
5701 tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5704 flow->tunnel_id = tunnel->tunnel_id;
5705 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5706 mlx5_free(default_miss_ctx.queue);
5708 mlx5_flow_pop_thread_workspace();
5712 ret = rte_errno; /* Save rte_errno before cleanup. */
5713 flow_mreg_del_copy_action(dev, flow);
5714 flow_drv_destroy(dev, flow);
5715 if (rss_desc->shared_rss)
5716 __atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5718 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5719 rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5720 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5721 rte_errno = ret; /* Restore rte_errno. */
5724 mlx5_flow_pop_thread_workspace();
5725 error_before_hairpin_split:
5726 rte_free(translated_actions);
5731 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5732 * incoming packets to table 1.
5734 * Other flow rules, requested for group n, will be created in
5735 * e-switch table n+1.
5736 * Jump action to e-switch group n will be created to group n+1.
5738 * Used when working in switchdev mode, to utilise advantages of table 1
5742 * Pointer to Ethernet device.
5745 * Pointer to flow on success, NULL otherwise and rte_errno is set.
5748 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5750 const struct rte_flow_attr attr = {
5757 const struct rte_flow_item pattern = {
5758 .type = RTE_FLOW_ITEM_TYPE_END,
5760 struct rte_flow_action_jump jump = {
5763 const struct rte_flow_action actions[] = {
5765 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5769 .type = RTE_FLOW_ACTION_TYPE_END,
5772 struct mlx5_priv *priv = dev->data->dev_private;
5773 struct rte_flow_error error;
5775 return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5777 actions, false, &error);
5781 * Validate a flow supported by the NIC.
5783 * @see rte_flow_validate()
5787 mlx5_flow_validate(struct rte_eth_dev *dev,
5788 const struct rte_flow_attr *attr,
5789 const struct rte_flow_item items[],
5790 const struct rte_flow_action original_actions[],
5791 struct rte_flow_error *error)
5794 struct mlx5_translated_shared_action
5795 shared_actions[MLX5_MAX_SHARED_ACTIONS];
5796 int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5797 const struct rte_flow_action *actions;
5798 struct rte_flow_action *translated_actions = NULL;
5799 int ret = flow_shared_actions_translate(dev, original_actions,
5802 &translated_actions, error);
5806 actions = translated_actions ? translated_actions : original_actions;
5807 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5808 ret = flow_drv_validate(dev, attr, items, actions,
5809 true, hairpin_flow, error);
5810 rte_free(translated_actions);
5817 * @see rte_flow_create()
5821 mlx5_flow_create(struct rte_eth_dev *dev,
5822 const struct rte_flow_attr *attr,
5823 const struct rte_flow_item items[],
5824 const struct rte_flow_action actions[],
5825 struct rte_flow_error *error)
5827 struct mlx5_priv *priv = dev->data->dev_private;
5830 * If the device is not started yet, it is not allowed to created a
5831 * flow from application. PMD default flows and traffic control flows
5834 if (unlikely(!dev->data->dev_started)) {
5835 DRV_LOG(DEBUG, "port %u is not started when "
5836 "inserting a flow", dev->data->port_id);
5837 rte_flow_error_set(error, ENODEV,
5838 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5840 "port not started");
5844 return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5845 attr, items, actions, true, error);
5849 * Destroy a flow in a list.
5852 * Pointer to Ethernet device.
5854 * Pointer to the Indexed flow list. If this parameter NULL,
5855 * there is no flow removal from the list. Be noted that as
5856 * flow is add to the indexed list, memory of the indexed
5857 * list points to maybe changed as flow destroyed.
5858 * @param[in] flow_idx
5859 * Index of flow to destroy.
5862 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5865 struct mlx5_priv *priv = dev->data->dev_private;
5866 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5867 [MLX5_IPOOL_RTE_FLOW], flow_idx);
5872 * Update RX queue flags only if port is started, otherwise it is
5875 if (dev->data->dev_started)
5876 flow_rxq_flags_trim(dev, flow);
5877 flow_drv_destroy(dev, flow);
5879 rte_spinlock_lock(&priv->flow_list_lock);
5880 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5881 flow_idx, flow, next);
5882 rte_spinlock_unlock(&priv->flow_list_lock);
5885 struct mlx5_flow_tunnel *tunnel;
5887 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5889 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5890 mlx5_flow_tunnel_free(dev, tunnel);
5892 flow_mreg_del_copy_action(dev, flow);
5893 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5897 * Destroy all flows.
5900 * Pointer to Ethernet device.
5902 * Pointer to the Indexed flow list.
5904 * If flushing is called avtively.
5907 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5909 uint32_t num_flushed = 0;
5912 flow_list_destroy(dev, list, *list);
5916 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5917 dev->data->port_id, num_flushed);
5922 * Stop all default actions for flows.
5925 * Pointer to Ethernet device.
5928 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5930 flow_mreg_del_default_copy_action(dev);
5931 flow_rxq_flags_clear(dev);
5935 * Start all default actions for flows.
5938 * Pointer to Ethernet device.
5940 * 0 on success, a negative errno value otherwise and rte_errno is set.
5943 mlx5_flow_start_default(struct rte_eth_dev *dev)
5945 struct rte_flow_error error;
5947 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5948 return flow_mreg_add_default_copy_action(dev, &error);
5952 * Release key of thread specific flow workspace data.
5955 flow_release_workspace(void *data)
5957 struct mlx5_flow_workspace *wks = data;
5958 struct mlx5_flow_workspace *next;
5962 free(wks->rss_desc.queue);
5969 * Get thread specific current flow workspace.
5971 * @return pointer to thread specific flow workspace data, NULL on error.
5973 struct mlx5_flow_workspace*
5974 mlx5_flow_get_thread_workspace(void)
5976 struct mlx5_flow_workspace *data;
5978 data = mlx5_flow_os_get_specific_workspace();
5979 MLX5_ASSERT(data && data->inuse);
5980 if (!data || !data->inuse)
5981 DRV_LOG(ERR, "flow workspace not initialized.");
5986 * Allocate and init new flow workspace.
5988 * @return pointer to flow workspace data, NULL on error.
5990 static struct mlx5_flow_workspace*
5991 flow_alloc_thread_workspace(void)
5993 struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
5996 DRV_LOG(ERR, "Failed to allocate flow workspace "
6000 data->rss_desc.queue = calloc(1,
6001 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6002 if (!data->rss_desc.queue)
6004 data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6007 if (data->rss_desc.queue)
6008 free(data->rss_desc.queue);
6014 * Get new thread specific flow workspace.
6016 * If current workspace inuse, create new one and set as current.
6018 * @return pointer to thread specific flow workspace data, NULL on error.
6020 static struct mlx5_flow_workspace*
6021 mlx5_flow_push_thread_workspace(void)
6023 struct mlx5_flow_workspace *curr;
6024 struct mlx5_flow_workspace *data;
6026 curr = mlx5_flow_os_get_specific_workspace();
6028 data = flow_alloc_thread_workspace();
6031 } else if (!curr->inuse) {
6033 } else if (curr->next) {
6036 data = flow_alloc_thread_workspace();
6044 /* Set as current workspace */
6045 if (mlx5_flow_os_set_specific_workspace(data))
6046 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6051 * Close current thread specific flow workspace.
6053 * If previous workspace available, set it as current.
6055 * @return pointer to thread specific flow workspace data, NULL on error.
6058 mlx5_flow_pop_thread_workspace(void)
6060 struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6065 DRV_LOG(ERR, "Failed to close unused flow workspace.");
6071 if (mlx5_flow_os_set_specific_workspace(data->prev))
6072 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6076 * Verify the flow list is empty
6079 * Pointer to Ethernet device.
6081 * @return the number of flows not released.
6084 mlx5_flow_verify(struct rte_eth_dev *dev)
6086 struct mlx5_priv *priv = dev->data->dev_private;
6087 struct rte_flow *flow;
6091 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6093 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6094 dev->data->port_id, (void *)flow);
6101 * Enable default hairpin egress flow.
6104 * Pointer to Ethernet device.
6109 * 0 on success, a negative errno value otherwise and rte_errno is set.
6112 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6115 struct mlx5_priv *priv = dev->data->dev_private;
6116 const struct rte_flow_attr attr = {
6120 struct mlx5_rte_flow_item_tx_queue queue_spec = {
6123 struct mlx5_rte_flow_item_tx_queue queue_mask = {
6124 .queue = UINT32_MAX,
6126 struct rte_flow_item items[] = {
6128 .type = (enum rte_flow_item_type)
6129 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6130 .spec = &queue_spec,
6132 .mask = &queue_mask,
6135 .type = RTE_FLOW_ITEM_TYPE_END,
6138 struct rte_flow_action_jump jump = {
6139 .group = MLX5_HAIRPIN_TX_TABLE,
6141 struct rte_flow_action actions[2];
6143 struct rte_flow_error error;
6145 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6146 actions[0].conf = &jump;
6147 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6148 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6149 &attr, items, actions, false, &error);
6152 "Failed to create ctrl flow: rte_errno(%d),"
6153 " type(%d), message(%s)",
6154 rte_errno, error.type,
6155 error.message ? error.message : " (no stated reason)");
6162 * Enable a control flow configured from the control plane.
6165 * Pointer to Ethernet device.
6167 * An Ethernet flow spec to apply.
6169 * An Ethernet flow mask to apply.
6171 * A VLAN flow spec to apply.
6173 * A VLAN flow mask to apply.
6176 * 0 on success, a negative errno value otherwise and rte_errno is set.
6179 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6180 struct rte_flow_item_eth *eth_spec,
6181 struct rte_flow_item_eth *eth_mask,
6182 struct rte_flow_item_vlan *vlan_spec,
6183 struct rte_flow_item_vlan *vlan_mask)
6185 struct mlx5_priv *priv = dev->data->dev_private;
6186 const struct rte_flow_attr attr = {
6188 .priority = MLX5_FLOW_PRIO_RSVD,
6190 struct rte_flow_item items[] = {
6192 .type = RTE_FLOW_ITEM_TYPE_ETH,
6198 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6199 RTE_FLOW_ITEM_TYPE_END,
6205 .type = RTE_FLOW_ITEM_TYPE_END,
6208 uint16_t queue[priv->reta_idx_n];
6209 struct rte_flow_action_rss action_rss = {
6210 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6212 .types = priv->rss_conf.rss_hf,
6213 .key_len = priv->rss_conf.rss_key_len,
6214 .queue_num = priv->reta_idx_n,
6215 .key = priv->rss_conf.rss_key,
6218 struct rte_flow_action actions[] = {
6220 .type = RTE_FLOW_ACTION_TYPE_RSS,
6221 .conf = &action_rss,
6224 .type = RTE_FLOW_ACTION_TYPE_END,
6228 struct rte_flow_error error;
6231 if (!priv->reta_idx_n || !priv->rxqs_n) {
6234 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6235 action_rss.types = 0;
6236 for (i = 0; i != priv->reta_idx_n; ++i)
6237 queue[i] = (*priv->reta_idx)[i];
6238 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6239 &attr, items, actions, false, &error);
6246 * Enable a flow control configured from the control plane.
6249 * Pointer to Ethernet device.
6251 * An Ethernet flow spec to apply.
6253 * An Ethernet flow mask to apply.
6256 * 0 on success, a negative errno value otherwise and rte_errno is set.
6259 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6260 struct rte_flow_item_eth *eth_spec,
6261 struct rte_flow_item_eth *eth_mask)
6263 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6267 * Create default miss flow rule matching lacp traffic
6270 * Pointer to Ethernet device.
6272 * An Ethernet flow spec to apply.
6275 * 0 on success, a negative errno value otherwise and rte_errno is set.
6278 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6280 struct mlx5_priv *priv = dev->data->dev_private;
6282 * The LACP matching is done by only using ether type since using
6283 * a multicast dst mac causes kernel to give low priority to this flow.
6285 static const struct rte_flow_item_eth lacp_spec = {
6286 .type = RTE_BE16(0x8809),
6288 static const struct rte_flow_item_eth lacp_mask = {
6291 const struct rte_flow_attr attr = {
6294 struct rte_flow_item items[] = {
6296 .type = RTE_FLOW_ITEM_TYPE_ETH,
6301 .type = RTE_FLOW_ITEM_TYPE_END,
6304 struct rte_flow_action actions[] = {
6306 .type = (enum rte_flow_action_type)
6307 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6310 .type = RTE_FLOW_ACTION_TYPE_END,
6313 struct rte_flow_error error;
6314 uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6315 &attr, items, actions, false, &error);
6325 * @see rte_flow_destroy()
6329 mlx5_flow_destroy(struct rte_eth_dev *dev,
6330 struct rte_flow *flow,
6331 struct rte_flow_error *error __rte_unused)
6333 struct mlx5_priv *priv = dev->data->dev_private;
6335 flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6340 * Destroy all flows.
6342 * @see rte_flow_flush()
6346 mlx5_flow_flush(struct rte_eth_dev *dev,
6347 struct rte_flow_error *error __rte_unused)
6349 struct mlx5_priv *priv = dev->data->dev_private;
6351 mlx5_flow_list_flush(dev, &priv->flows, false);
6358 * @see rte_flow_isolate()
6362 mlx5_flow_isolate(struct rte_eth_dev *dev,
6364 struct rte_flow_error *error)
6366 struct mlx5_priv *priv = dev->data->dev_private;
6368 if (dev->data->dev_started) {
6369 rte_flow_error_set(error, EBUSY,
6370 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6372 "port must be stopped first");
6375 priv->isolated = !!enable;
6377 dev->dev_ops = &mlx5_dev_ops_isolate;
6379 dev->dev_ops = &mlx5_dev_ops;
6381 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6382 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6390 * @see rte_flow_query()
6394 flow_drv_query(struct rte_eth_dev *dev,
6396 const struct rte_flow_action *actions,
6398 struct rte_flow_error *error)
6400 struct mlx5_priv *priv = dev->data->dev_private;
6401 const struct mlx5_flow_driver_ops *fops;
6402 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6403 [MLX5_IPOOL_RTE_FLOW],
6405 enum mlx5_flow_drv_type ftype;
6408 return rte_flow_error_set(error, ENOENT,
6409 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6411 "invalid flow handle");
6413 ftype = flow->drv_type;
6414 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6415 fops = flow_get_drv_ops(ftype);
6417 return fops->query(dev, flow, actions, data, error);
6423 * @see rte_flow_query()
6427 mlx5_flow_query(struct rte_eth_dev *dev,
6428 struct rte_flow *flow,
6429 const struct rte_flow_action *actions,
6431 struct rte_flow_error *error)
6435 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6443 * Manage filter operations.
6446 * Pointer to Ethernet device structure.
6447 * @param filter_type
6450 * Operation to perform.
6452 * Pointer to operation-specific structure.
6455 * 0 on success, a negative errno value otherwise and rte_errno is set.
6458 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6459 enum rte_filter_type filter_type,
6460 enum rte_filter_op filter_op,
6463 switch (filter_type) {
6464 case RTE_ETH_FILTER_GENERIC:
6465 if (filter_op != RTE_ETH_FILTER_GET) {
6469 *(const void **)arg = &mlx5_flow_ops;
6472 DRV_LOG(ERR, "port %u filter type (%d) not supported",
6473 dev->data->port_id, filter_type);
6474 rte_errno = ENOTSUP;
6481 * Create the needed meter and suffix tables.
6484 * Pointer to Ethernet device.
6486 * Pointer to the flow meter.
6489 * Pointer to table set on success, NULL otherwise.
6491 struct mlx5_meter_domains_infos *
6492 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6493 const struct mlx5_flow_meter *fm)
6495 const struct mlx5_flow_driver_ops *fops;
6497 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6498 return fops->create_mtr_tbls(dev, fm);
6502 * Destroy the meter table set.
6505 * Pointer to Ethernet device.
6507 * Pointer to the meter table set.
6513 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6514 struct mlx5_meter_domains_infos *tbls)
6516 const struct mlx5_flow_driver_ops *fops;
6518 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6519 return fops->destroy_mtr_tbls(dev, tbls);
6523 * Create policer rules.
6526 * Pointer to Ethernet device.
6528 * Pointer to flow meter structure.
6530 * Pointer to flow attributes.
6533 * 0 on success, -1 otherwise.
6536 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6537 struct mlx5_flow_meter *fm,
6538 const struct rte_flow_attr *attr)
6540 const struct mlx5_flow_driver_ops *fops;
6542 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6543 return fops->create_policer_rules(dev, fm, attr);
6547 * Destroy policer rules.
6550 * Pointer to flow meter structure.
6552 * Pointer to flow attributes.
6555 * 0 on success, -1 otherwise.
6558 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6559 struct mlx5_flow_meter *fm,
6560 const struct rte_flow_attr *attr)
6562 const struct mlx5_flow_driver_ops *fops;
6564 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6565 return fops->destroy_policer_rules(dev, fm, attr);
6569 * Allocate a counter.
6572 * Pointer to Ethernet device structure.
6575 * Index to allocated counter on success, 0 otherwise.
6578 mlx5_counter_alloc(struct rte_eth_dev *dev)
6580 const struct mlx5_flow_driver_ops *fops;
6581 struct rte_flow_attr attr = { .transfer = 0 };
6583 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6584 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6585 return fops->counter_alloc(dev);
6588 "port %u counter allocate is not supported.",
6589 dev->data->port_id);
6597 * Pointer to Ethernet device structure.
6599 * Index to counter to be free.
6602 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6604 const struct mlx5_flow_driver_ops *fops;
6605 struct rte_flow_attr attr = { .transfer = 0 };
6607 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6608 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6609 fops->counter_free(dev, cnt);
6613 "port %u counter free is not supported.",
6614 dev->data->port_id);
6618 * Query counter statistics.
6621 * Pointer to Ethernet device structure.
6623 * Index to counter to query.
6625 * Set to clear counter statistics.
6627 * The counter hits packets number to save.
6629 * The counter hits bytes number to save.
6632 * 0 on success, a negative errno value otherwise.
6635 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6636 bool clear, uint64_t *pkts, uint64_t *bytes)
6638 const struct mlx5_flow_driver_ops *fops;
6639 struct rte_flow_attr attr = { .transfer = 0 };
6641 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6642 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6643 return fops->counter_query(dev, cnt, clear, pkts, bytes);
6646 "port %u counter query is not supported.",
6647 dev->data->port_id);
6652 * Allocate a new memory for the counter values wrapped by all the needed
6656 * Pointer to mlx5_dev_ctx_shared object.
6659 * 0 on success, a negative errno value otherwise.
6662 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6664 struct mlx5_devx_mkey_attr mkey_attr;
6665 struct mlx5_counter_stats_mem_mng *mem_mng;
6666 volatile struct flow_counter_stats *raw_data;
6667 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6668 int size = (sizeof(struct flow_counter_stats) *
6669 MLX5_COUNTERS_PER_POOL +
6670 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6671 sizeof(struct mlx5_counter_stats_mem_mng);
6672 size_t pgsize = rte_mem_page_size();
6676 if (pgsize == (size_t)-1) {
6677 DRV_LOG(ERR, "Failed to get mem page size");
6681 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6686 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6687 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6688 mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
6689 IBV_ACCESS_LOCAL_WRITE);
6690 if (!mem_mng->umem) {
6695 mkey_attr.addr = (uintptr_t)mem;
6696 mkey_attr.size = size;
6697 mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6698 mkey_attr.pd = sh->pdn;
6699 mkey_attr.log_entity_size = 0;
6700 mkey_attr.pg_access = 0;
6701 mkey_attr.klm_array = NULL;
6702 mkey_attr.klm_num = 0;
6703 mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6704 mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6705 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6707 mlx5_os_umem_dereg(mem_mng->umem);
6712 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6713 raw_data = (volatile struct flow_counter_stats *)mem;
6714 for (i = 0; i < raws_n; ++i) {
6715 mem_mng->raws[i].mem_mng = mem_mng;
6716 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6718 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6719 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6720 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6722 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6723 sh->cmng.mem_mng = mem_mng;
6728 * Set the statistic memory to the new counter pool.
6731 * Pointer to mlx5_dev_ctx_shared object.
6733 * Pointer to the pool to set the statistic memory.
6736 * 0 on success, a negative errno value otherwise.
6739 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6740 struct mlx5_flow_counter_pool *pool)
6742 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6743 /* Resize statistic memory once used out. */
6744 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6745 mlx5_flow_create_counter_stat_mem_mng(sh)) {
6746 DRV_LOG(ERR, "Cannot resize counter stat mem.");
6749 rte_spinlock_lock(&pool->sl);
6750 pool->raw = cmng->mem_mng->raws + pool->index %
6751 MLX5_CNT_CONTAINER_RESIZE;
6752 rte_spinlock_unlock(&pool->sl);
6753 pool->raw_hw = NULL;
6757 #define MLX5_POOL_QUERY_FREQ_US 1000000
6760 * Set the periodic procedure for triggering asynchronous batch queries for all
6761 * the counter pools.
6764 * Pointer to mlx5_dev_ctx_shared object.
6767 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6769 uint32_t pools_n, us;
6771 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6772 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6773 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6774 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6775 sh->cmng.query_thread_on = 0;
6776 DRV_LOG(ERR, "Cannot reinitialize query alarm");
6778 sh->cmng.query_thread_on = 1;
6783 * The periodic procedure for triggering asynchronous batch queries for all the
6784 * counter pools. This function is probably called by the host thread.
6787 * The parameter for the alarm process.
6790 mlx5_flow_query_alarm(void *arg)
6792 struct mlx5_dev_ctx_shared *sh = arg;
6794 uint16_t pool_index = sh->cmng.pool_index;
6795 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6796 struct mlx5_flow_counter_pool *pool;
6799 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6801 rte_spinlock_lock(&cmng->pool_update_sl);
6802 pool = cmng->pools[pool_index];
6803 n_valid = cmng->n_valid;
6804 rte_spinlock_unlock(&cmng->pool_update_sl);
6805 /* Set the statistic memory to the new created pool. */
6806 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6809 /* There is a pool query in progress. */
6812 LIST_FIRST(&sh->cmng.free_stat_raws);
6814 /* No free counter statistics raw memory. */
6817 * Identify the counters released between query trigger and query
6818 * handle more efficiently. The counter released in this gap period
6819 * should wait for a new round of query as the new arrived packets
6820 * will not be taken into account.
6823 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6824 MLX5_COUNTERS_PER_POOL,
6826 pool->raw_hw->mem_mng->dm->id,
6830 (uint64_t)(uintptr_t)pool);
6832 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6833 " %d", pool->min_dcs->id);
6834 pool->raw_hw = NULL;
6837 LIST_REMOVE(pool->raw_hw, next);
6838 sh->cmng.pending_queries++;
6840 if (pool_index >= n_valid)
6843 sh->cmng.pool_index = pool_index;
6844 mlx5_set_query_alarm(sh);
6848 * Check and callback event for new aged flow in the counter pool
6851 * Pointer to mlx5_dev_ctx_shared object.
6853 * Pointer to Current counter pool.
6856 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6857 struct mlx5_flow_counter_pool *pool)
6859 struct mlx5_priv *priv;
6860 struct mlx5_flow_counter *cnt;
6861 struct mlx5_age_info *age_info;
6862 struct mlx5_age_param *age_param;
6863 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6864 struct mlx5_counter_stats_raw *prev = pool->raw;
6865 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6866 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6867 uint16_t expected = AGE_CANDIDATE;
6870 pool->time_of_last_age_check = curr_time;
6871 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6872 cnt = MLX5_POOL_GET_CNT(pool, i);
6873 age_param = MLX5_CNT_TO_AGE(cnt);
6874 if (__atomic_load_n(&age_param->state,
6875 __ATOMIC_RELAXED) != AGE_CANDIDATE)
6877 if (cur->data[i].hits != prev->data[i].hits) {
6878 __atomic_store_n(&age_param->sec_since_last_hit, 0,
6882 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6884 __ATOMIC_RELAXED) <= age_param->timeout)
6887 * Hold the lock first, or if between the
6888 * state AGE_TMOUT and tailq operation the
6889 * release happened, the release procedure
6890 * may delete a non-existent tailq node.
6892 priv = rte_eth_devices[age_param->port_id].data->dev_private;
6893 age_info = GET_PORT_AGE_INFO(priv);
6894 rte_spinlock_lock(&age_info->aged_sl);
6895 if (__atomic_compare_exchange_n(&age_param->state, &expected,
6898 __ATOMIC_RELAXED)) {
6899 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6900 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6902 rte_spinlock_unlock(&age_info->aged_sl);
6904 mlx5_age_event_prepare(sh);
6908 * Handler for the HW respond about ready values from an asynchronous batch
6909 * query. This function is probably called by the host thread.
6912 * The pointer to the shared device context.
6913 * @param[in] async_id
6914 * The Devx async ID.
6916 * The status of the completion.
6919 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6920 uint64_t async_id, int status)
6922 struct mlx5_flow_counter_pool *pool =
6923 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6924 struct mlx5_counter_stats_raw *raw_to_free;
6925 uint8_t query_gen = pool->query_gen ^ 1;
6926 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6927 enum mlx5_counter_type cnt_type =
6928 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6929 MLX5_COUNTER_TYPE_ORIGIN;
6931 if (unlikely(status)) {
6932 raw_to_free = pool->raw_hw;
6934 raw_to_free = pool->raw;
6936 mlx5_flow_aging_check(sh, pool);
6937 rte_spinlock_lock(&pool->sl);
6938 pool->raw = pool->raw_hw;
6939 rte_spinlock_unlock(&pool->sl);
6940 /* Be sure the new raw counters data is updated in memory. */
6942 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6943 rte_spinlock_lock(&cmng->csl[cnt_type]);
6944 TAILQ_CONCAT(&cmng->counters[cnt_type],
6945 &pool->counters[query_gen], next);
6946 rte_spinlock_unlock(&cmng->csl[cnt_type]);
6949 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6950 pool->raw_hw = NULL;
6951 sh->cmng.pending_queries--;
6955 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
6956 const struct flow_grp_info *grp_info,
6957 struct rte_flow_error *error)
6959 if (grp_info->transfer && grp_info->external &&
6960 grp_info->fdb_def_rule) {
6961 if (group == UINT32_MAX)
6962 return rte_flow_error_set
6964 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6966 "group index not supported");
6971 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
6976 * Translate the rte_flow group index to HW table value.
6978 * If tunnel offload is disabled, all group ids converted to flow table
6979 * id using the standard method.
6980 * If tunnel offload is enabled, group id can be converted using the
6981 * standard or tunnel conversion method. Group conversion method
6982 * selection depends on flags in `grp_info` parameter:
6983 * - Internal (grp_info.external == 0) groups conversion uses the
6985 * - Group ids in JUMP action converted with the tunnel conversion.
6986 * - Group id in rule attribute conversion depends on a rule type and
6988 * ** non zero group attributes converted with the tunnel method
6989 * ** zero group attribute in non-tunnel rule is converted using the
6990 * standard method - there's only one root table
6991 * ** zero group attribute in steer tunnel rule is converted with the
6992 * standard method - single root table
6993 * ** zero group attribute in match tunnel rule is a special OvS
6994 * case: that value is used for portability reasons. That group
6995 * id is converted with the tunnel conversion method.
7000 * PMD tunnel offload object
7002 * rte_flow group index value.
7005 * @param[in] grp_info
7006 * flags used for conversion
7008 * Pointer to error structure.
7011 * 0 on success, a negative errno value otherwise and rte_errno is set.
7014 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7015 const struct mlx5_flow_tunnel *tunnel,
7016 uint32_t group, uint32_t *table,
7017 const struct flow_grp_info *grp_info,
7018 struct rte_flow_error *error)
7021 bool standard_translation;
7023 if (!grp_info->skip_scale && grp_info->external &&
7024 group < MLX5_MAX_TABLES_EXTERNAL)
7025 group *= MLX5_FLOW_TABLE_FACTOR;
7026 if (is_tunnel_offload_active(dev)) {
7027 standard_translation = !grp_info->external ||
7028 grp_info->std_tbl_fix;
7030 standard_translation = true;
7033 "port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7034 dev->data->port_id, group, grp_info->transfer,
7035 grp_info->external, grp_info->fdb_def_rule,
7036 standard_translation ? "STANDARD" : "TUNNEL");
7037 if (standard_translation)
7038 ret = flow_group_to_table(dev->data->port_id, group, table,
7041 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7048 * Discover availability of metadata reg_c's.
7050 * Iteratively use test flows to check availability.
7053 * Pointer to the Ethernet device structure.
7056 * 0 on success, a negative errno value otherwise and rte_errno is set.
7059 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7061 struct mlx5_priv *priv = dev->data->dev_private;
7062 struct mlx5_dev_config *config = &priv->config;
7063 enum modify_reg idx;
7066 /* reg_c[0] and reg_c[1] are reserved. */
7067 config->flow_mreg_c[n++] = REG_C_0;
7068 config->flow_mreg_c[n++] = REG_C_1;
7069 /* Discover availability of other reg_c's. */
7070 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7071 struct rte_flow_attr attr = {
7072 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7073 .priority = MLX5_FLOW_PRIO_RSVD,
7076 struct rte_flow_item items[] = {
7078 .type = RTE_FLOW_ITEM_TYPE_END,
7081 struct rte_flow_action actions[] = {
7083 .type = (enum rte_flow_action_type)
7084 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7085 .conf = &(struct mlx5_flow_action_copy_mreg){
7091 .type = RTE_FLOW_ACTION_TYPE_JUMP,
7092 .conf = &(struct rte_flow_action_jump){
7093 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7097 .type = RTE_FLOW_ACTION_TYPE_END,
7101 struct rte_flow *flow;
7102 struct rte_flow_error error;
7104 if (!config->dv_flow_en)
7106 /* Create internal flow, validation skips copy action. */
7107 flow_idx = flow_list_create(dev, NULL, &attr, items,
7108 actions, false, &error);
7109 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7113 config->flow_mreg_c[n++] = idx;
7114 flow_list_destroy(dev, NULL, flow_idx);
7116 for (; n < MLX5_MREG_C_NUM; ++n)
7117 config->flow_mreg_c[n] = REG_NON;
7122 * Dump flow raw hw data to file
7125 * The pointer to Ethernet device.
7127 * A pointer to a file for output.
7129 * Perform verbose error reporting if not NULL. PMDs initialize this
7130 * structure in case of error only.
7132 * 0 on success, a nagative value otherwise.
7135 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7137 struct rte_flow_error *error __rte_unused)
7139 struct mlx5_priv *priv = dev->data->dev_private;
7140 struct mlx5_dev_ctx_shared *sh = priv->sh;
7142 if (!priv->config.dv_flow_en) {
7143 if (fputs("device dv flow disabled\n", file) <= 0)
7147 return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7148 sh->tx_domain, file);
7152 * Get aged-out flows.
7155 * Pointer to the Ethernet device structure.
7156 * @param[in] context
7157 * The address of an array of pointers to the aged-out flows contexts.
7158 * @param[in] nb_countexts
7159 * The length of context array pointers.
7161 * Perform verbose error reporting if not NULL. Initialized in case of
7165 * how many contexts get in success, otherwise negative errno value.
7166 * if nb_contexts is 0, return the amount of all aged contexts.
7167 * if nb_contexts is not 0 , return the amount of aged flows reported
7168 * in the context array.
7171 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7172 uint32_t nb_contexts, struct rte_flow_error *error)
7174 const struct mlx5_flow_driver_ops *fops;
7175 struct rte_flow_attr attr = { .transfer = 0 };
7177 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7178 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7179 return fops->get_aged_flows(dev, contexts, nb_contexts,
7183 "port %u get aged flows is not supported.",
7184 dev->data->port_id);
7188 /* Wrapper for driver action_validate op callback */
7190 flow_drv_action_validate(struct rte_eth_dev *dev,
7191 const struct rte_flow_shared_action_conf *conf,
7192 const struct rte_flow_action *action,
7193 const struct mlx5_flow_driver_ops *fops,
7194 struct rte_flow_error *error)
7196 static const char err_msg[] = "shared action validation unsupported";
7198 if (!fops->action_validate) {
7199 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7200 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7204 return fops->action_validate(dev, conf, action, error);
7208 * Destroys the shared action by handle.
7211 * Pointer to Ethernet device structure.
7213 * Handle for the shared action to be destroyed.
7215 * Perform verbose error reporting if not NULL. PMDs initialize this
7216 * structure in case of error only.
7219 * 0 on success, a negative errno value otherwise and rte_errno is set.
7221 * @note: wrapper for driver action_create op callback.
7224 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7225 struct rte_flow_shared_action *action,
7226 struct rte_flow_error *error)
7228 static const char err_msg[] = "shared action destruction unsupported";
7229 struct rte_flow_attr attr = { .transfer = 0 };
7230 const struct mlx5_flow_driver_ops *fops =
7231 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7233 if (!fops->action_destroy) {
7234 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7235 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7239 return fops->action_destroy(dev, action, error);
7242 /* Wrapper for driver action_destroy op callback */
7244 flow_drv_action_update(struct rte_eth_dev *dev,
7245 struct rte_flow_shared_action *action,
7246 const void *action_conf,
7247 const struct mlx5_flow_driver_ops *fops,
7248 struct rte_flow_error *error)
7250 static const char err_msg[] = "shared action update unsupported";
7252 if (!fops->action_update) {
7253 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7254 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7258 return fops->action_update(dev, action, action_conf, error);
7261 /* Wrapper for driver action_destroy op callback */
7263 flow_drv_action_query(struct rte_eth_dev *dev,
7264 const struct rte_flow_shared_action *action,
7266 const struct mlx5_flow_driver_ops *fops,
7267 struct rte_flow_error *error)
7269 static const char err_msg[] = "shared action query unsupported";
7271 if (!fops->action_query) {
7272 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7273 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7277 return fops->action_query(dev, action, data, error);
7281 * Create shared action for reuse in multiple flow rules.
7284 * Pointer to Ethernet device structure.
7286 * Action configuration for shared action creation.
7288 * Perform verbose error reporting if not NULL. PMDs initialize this
7289 * structure in case of error only.
7291 * A valid handle in case of success, NULL otherwise and rte_errno is set.
7293 static struct rte_flow_shared_action *
7294 mlx5_shared_action_create(struct rte_eth_dev *dev,
7295 const struct rte_flow_shared_action_conf *conf,
7296 const struct rte_flow_action *action,
7297 struct rte_flow_error *error)
7299 static const char err_msg[] = "shared action creation unsupported";
7300 struct rte_flow_attr attr = { .transfer = 0 };
7301 const struct mlx5_flow_driver_ops *fops =
7302 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7304 if (flow_drv_action_validate(dev, conf, action, fops, error))
7306 if (!fops->action_create) {
7307 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7308 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7312 return fops->action_create(dev, conf, action, error);
7316 * Updates inplace the shared action configuration pointed by *action* handle
7317 * with the configuration provided as *action* argument.
7318 * The update of the shared action configuration effects all flow rules reusing
7319 * the action via handle.
7322 * Pointer to Ethernet device structure.
7323 * @param[in] shared_action
7324 * Handle for the shared action to be updated.
7326 * Action specification used to modify the action pointed by handle.
7327 * *action* should be of same type with the action pointed by the *action*
7328 * handle argument, otherwise considered as invalid.
7330 * Perform verbose error reporting if not NULL. PMDs initialize this
7331 * structure in case of error only.
7334 * 0 on success, a negative errno value otherwise and rte_errno is set.
7337 mlx5_shared_action_update(struct rte_eth_dev *dev,
7338 struct rte_flow_shared_action *shared_action,
7339 const struct rte_flow_action *action,
7340 struct rte_flow_error *error)
7342 struct rte_flow_attr attr = { .transfer = 0 };
7343 const struct mlx5_flow_driver_ops *fops =
7344 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7347 ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7350 return flow_drv_action_update(dev, shared_action, action->conf, fops,
7355 * Query the shared action by handle.
7357 * This function allows retrieving action-specific data such as counters.
7358 * Data is gathered by special action which may be present/referenced in
7359 * more than one flow rule definition.
7361 * \see RTE_FLOW_ACTION_TYPE_COUNT
7364 * Pointer to Ethernet device structure.
7366 * Handle for the shared action to query.
7367 * @param[in, out] data
7368 * Pointer to storage for the associated query data type.
7370 * Perform verbose error reporting if not NULL. PMDs initialize this
7371 * structure in case of error only.
7374 * 0 on success, a negative errno value otherwise and rte_errno is set.
7377 mlx5_shared_action_query(struct rte_eth_dev *dev,
7378 const struct rte_flow_shared_action *action,
7380 struct rte_flow_error *error)
7382 struct rte_flow_attr attr = { .transfer = 0 };
7383 const struct mlx5_flow_driver_ops *fops =
7384 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7386 return flow_drv_action_query(dev, action, data, fops, error);
7390 * Destroy all shared actions.
7393 * Pointer to Ethernet device.
7396 * 0 on success, a negative errno value otherwise and rte_errno is set.
7399 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7401 struct rte_flow_error error;
7402 struct mlx5_priv *priv = dev->data->dev_private;
7403 struct mlx5_shared_action_rss *action;
7407 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7408 priv->rss_shared_actions, idx, action, next) {
7409 ret |= mlx5_shared_action_destroy(dev,
7410 (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7415 #ifndef HAVE_MLX5DV_DR
7416 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7418 #define MLX5_DOMAIN_SYNC_FLOW \
7419 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7422 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7424 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7425 const struct mlx5_flow_driver_ops *fops;
7427 struct rte_flow_attr attr = { .transfer = 0 };
7429 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7430 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7437 * tunnel offload functionalilty is defined for DV environment only
7439 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7441 union tunnel_offload_mark {
7444 uint32_t app_reserve:8;
7445 uint32_t table_id:15;
7446 uint32_t transfer:1;
7447 uint32_t _unused_:8;
7452 mlx5_access_tunnel_offload_db
7453 (struct rte_eth_dev *dev,
7454 bool (*match)(struct rte_eth_dev *,
7455 struct mlx5_flow_tunnel *, const void *),
7456 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7457 void (*miss)(struct rte_eth_dev *, void *),
7458 void *ctx, bool lock_op);
7461 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
7462 struct rte_flow *flow,
7463 const struct rte_flow_attr *attr,
7464 const struct rte_flow_action *app_actions,
7466 struct tunnel_default_miss_ctx *ctx,
7467 struct rte_flow_error *error)
7469 struct mlx5_priv *priv = dev->data->dev_private;
7470 struct mlx5_flow *dev_flow;
7471 struct rte_flow_attr miss_attr = *attr;
7472 const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
7473 const struct rte_flow_item miss_items[2] = {
7475 .type = RTE_FLOW_ITEM_TYPE_ETH,
7481 .type = RTE_FLOW_ITEM_TYPE_END,
7487 union tunnel_offload_mark mark_id;
7488 struct rte_flow_action_mark miss_mark;
7489 struct rte_flow_action miss_actions[3] = {
7490 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
7491 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
7493 const struct rte_flow_action_jump *jump_data;
7494 uint32_t i, flow_table = 0; /* prevent compilation warning */
7495 struct flow_grp_info grp_info = {
7497 .transfer = attr->transfer,
7498 .fdb_def_rule = !!priv->fdb_def_rule,
7503 if (!attr->transfer) {
7506 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
7507 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
7508 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
7511 return rte_flow_error_set
7513 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7514 NULL, "invalid default miss RSS");
7515 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7516 ctx->action_rss.level = 0,
7517 ctx->action_rss.types = priv->rss_conf.rss_hf,
7518 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
7519 ctx->action_rss.queue_num = priv->reta_idx_n,
7520 ctx->action_rss.key = priv->rss_conf.rss_key,
7521 ctx->action_rss.queue = ctx->queue;
7522 if (!priv->reta_idx_n || !priv->rxqs_n)
7523 return rte_flow_error_set
7525 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7526 NULL, "invalid port configuration");
7527 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7528 ctx->action_rss.types = 0;
7529 for (i = 0; i != priv->reta_idx_n; ++i)
7530 ctx->queue[i] = (*priv->reta_idx)[i];
7532 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
7533 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
7535 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
7536 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
7537 jump_data = app_actions->conf;
7538 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
7539 miss_attr.group = jump_data->group;
7540 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
7541 &flow_table, &grp_info, error);
7543 return rte_flow_error_set(error, EINVAL,
7544 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7545 NULL, "invalid tunnel id");
7546 mark_id.app_reserve = 0;
7547 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
7548 mark_id.transfer = !!attr->transfer;
7549 mark_id._unused_ = 0;
7550 miss_mark.id = mark_id.val;
7551 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
7552 miss_items, miss_actions, flow_idx, error);
7555 dev_flow->flow = flow;
7556 dev_flow->external = true;
7557 dev_flow->tunnel = tunnel;
7558 /* Subflow object was created, we must include one in the list. */
7559 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
7560 dev_flow->handle, next);
7562 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
7563 dev->data->port_id, tunnel->app_tunnel.type,
7564 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
7565 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
7566 miss_actions, error);
7568 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
7574 static const struct mlx5_flow_tbl_data_entry *
7575 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7577 struct mlx5_priv *priv = dev->data->dev_private;
7578 struct mlx5_dev_ctx_shared *sh = priv->sh;
7579 struct mlx5_hlist_entry *he;
7580 union tunnel_offload_mark mbits = { .val = mark };
7581 union mlx5_flow_tbl_key table_key = {
7583 .table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7585 .domain = !!mbits.transfer,
7589 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7591 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7595 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
7596 struct mlx5_hlist_entry *entry)
7598 struct mlx5_dev_ctx_shared *sh = list->ctx;
7599 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7601 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7602 tunnel_flow_tbl_to_id(tte->flow_table));
7607 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
7608 struct mlx5_hlist_entry *entry,
7609 uint64_t key, void *cb_ctx __rte_unused)
7611 union tunnel_tbl_key tbl = {
7614 struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7616 return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
7619 static struct mlx5_hlist_entry *
7620 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
7621 void *ctx __rte_unused)
7623 struct mlx5_dev_ctx_shared *sh = list->ctx;
7624 struct tunnel_tbl_entry *tte;
7625 union tunnel_tbl_key tbl = {
7629 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7634 mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7636 if (tte->flow_table >= MLX5_MAX_TABLES) {
7637 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7639 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7642 } else if (!tte->flow_table) {
7645 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7646 tte->tunnel_id = tbl.tunnel_id;
7647 tte->group = tbl.group;
7656 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7657 const struct mlx5_flow_tunnel *tunnel,
7658 uint32_t group, uint32_t *table,
7659 struct rte_flow_error *error)
7661 struct mlx5_hlist_entry *he;
7662 struct tunnel_tbl_entry *tte;
7663 union tunnel_tbl_key key = {
7664 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7667 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7668 struct mlx5_hlist *group_hash;
7670 group_hash = tunnel ? tunnel->groups : thub->groups;
7671 he = mlx5_hlist_register(group_hash, key.val, NULL);
7673 return rte_flow_error_set(error, EINVAL,
7674 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7676 "tunnel group index not supported");
7677 tte = container_of(he, typeof(*tte), hash);
7678 *table = tte->flow_table;
7679 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7680 dev->data->port_id, key.tunnel_id, group, *table);
7685 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7686 struct mlx5_flow_tunnel *tunnel)
7688 struct mlx5_priv *priv = dev->data->dev_private;
7689 struct mlx5_indexed_pool *ipool;
7691 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7692 dev->data->port_id, tunnel->tunnel_id);
7693 LIST_REMOVE(tunnel, chain);
7694 mlx5_hlist_destroy(tunnel->groups);
7695 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7696 mlx5_ipool_free(ipool, tunnel->tunnel_id);
7700 mlx5_access_tunnel_offload_db
7701 (struct rte_eth_dev *dev,
7702 bool (*match)(struct rte_eth_dev *,
7703 struct mlx5_flow_tunnel *, const void *),
7704 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7705 void (*miss)(struct rte_eth_dev *, void *),
7706 void *ctx, bool lock_op)
7708 bool verdict = false;
7709 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7710 struct mlx5_flow_tunnel *tunnel;
7712 rte_spinlock_lock(&thub->sl);
7713 LIST_FOREACH(tunnel, &thub->tunnels, chain) {
7714 verdict = match(dev, tunnel, (const void *)ctx);
7719 rte_spinlock_unlock(&thub->sl);
7721 hit(dev, tunnel, ctx);
7722 if (!verdict && miss)
7725 rte_spinlock_unlock(&thub->sl);
7730 struct tunnel_db_find_tunnel_id_ctx {
7732 struct mlx5_flow_tunnel *tunnel;
7736 find_tunnel_id_match(struct rte_eth_dev *dev,
7737 struct mlx5_flow_tunnel *tunnel, const void *x)
7739 const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7742 return tunnel->tunnel_id == ctx->tunnel_id;
7746 find_tunnel_id_hit(struct rte_eth_dev *dev,
7747 struct mlx5_flow_tunnel *tunnel, void *x)
7749 struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7751 ctx->tunnel = tunnel;
7754 static struct mlx5_flow_tunnel *
7755 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7757 struct tunnel_db_find_tunnel_id_ctx ctx = {
7761 mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
7762 find_tunnel_id_hit, NULL, &ctx, true);
7767 static struct mlx5_flow_tunnel *
7768 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7769 const struct rte_flow_tunnel *app_tunnel)
7771 struct mlx5_priv *priv = dev->data->dev_private;
7772 struct mlx5_indexed_pool *ipool;
7773 struct mlx5_flow_tunnel *tunnel;
7776 ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7777 tunnel = mlx5_ipool_zmalloc(ipool, &id);
7780 if (id >= MLX5_MAX_TUNNELS) {
7781 mlx5_ipool_free(ipool, id);
7782 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7785 tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7786 mlx5_flow_tunnel_grp2tbl_create_cb,
7787 mlx5_flow_tunnel_grp2tbl_match_cb,
7788 mlx5_flow_tunnel_grp2tbl_remove_cb);
7789 if (!tunnel->groups) {
7790 mlx5_ipool_free(ipool, id);
7793 tunnel->groups->ctx = priv->sh;
7794 /* initiate new PMD tunnel */
7795 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7796 tunnel->tunnel_id = id;
7797 tunnel->action.type = (typeof(tunnel->action.type))
7798 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7799 tunnel->action.conf = tunnel;
7800 tunnel->item.type = (typeof(tunnel->item.type))
7801 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7802 tunnel->item.spec = tunnel;
7803 tunnel->item.last = NULL;
7804 tunnel->item.mask = NULL;
7806 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7807 dev->data->port_id, tunnel->tunnel_id);
7812 struct tunnel_db_get_tunnel_ctx {
7813 const struct rte_flow_tunnel *app_tunnel;
7814 struct mlx5_flow_tunnel *tunnel;
7817 static bool get_tunnel_match(struct rte_eth_dev *dev,
7818 struct mlx5_flow_tunnel *tunnel, const void *x)
7820 const struct tunnel_db_get_tunnel_ctx *ctx = x;
7823 return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
7824 sizeof(*ctx->app_tunnel));
7827 static void get_tunnel_hit(struct rte_eth_dev *dev,
7828 struct mlx5_flow_tunnel *tunnel, void *x)
7830 /* called under tunnel spinlock protection */
7831 struct tunnel_db_get_tunnel_ctx *ctx = x;
7835 ctx->tunnel = tunnel;
7838 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
7840 /* called under tunnel spinlock protection */
7841 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7842 struct tunnel_db_get_tunnel_ctx *ctx = x;
7844 rte_spinlock_unlock(&thub->sl);
7845 ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
7846 ctx->tunnel->refctn = 1;
7847 rte_spinlock_lock(&thub->sl);
7849 LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
7854 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7855 const struct rte_flow_tunnel *app_tunnel,
7856 struct mlx5_flow_tunnel **tunnel)
7858 struct tunnel_db_get_tunnel_ctx ctx = {
7859 .app_tunnel = app_tunnel,
7862 mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
7863 get_tunnel_miss, &ctx, true);
7864 *tunnel = ctx.tunnel;
7865 return ctx.tunnel ? 0 : -ENOMEM;
7868 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7870 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7874 if (!LIST_EMPTY(&thub->tunnels))
7875 DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
7876 mlx5_hlist_destroy(thub->groups);
7880 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7883 struct mlx5_flow_tunnel_hub *thub;
7885 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7889 LIST_INIT(&thub->tunnels);
7890 rte_spinlock_init(&thub->sl);
7891 thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
7892 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7893 mlx5_flow_tunnel_grp2tbl_match_cb,
7894 mlx5_flow_tunnel_grp2tbl_remove_cb);
7895 if (!thub->groups) {
7899 thub->groups->ctx = sh;
7900 sh->tunnel_hub = thub;
7906 mlx5_hlist_destroy(thub->groups);
7913 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
7914 struct rte_flow_tunnel *tunnel,
7915 const char *err_msg)
7918 if (!is_tunnel_offload_active(dev)) {
7919 err_msg = "tunnel offload was not activated";
7921 } else if (!tunnel) {
7922 err_msg = "no application tunnel";
7926 switch (tunnel->type) {
7928 err_msg = "unsupported tunnel type";
7930 case RTE_FLOW_ITEM_TYPE_VXLAN:
7939 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
7940 struct rte_flow_tunnel *app_tunnel,
7941 struct rte_flow_action **actions,
7942 uint32_t *num_of_actions,
7943 struct rte_flow_error *error)
7946 struct mlx5_flow_tunnel *tunnel;
7947 const char *err_msg = NULL;
7948 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7951 return rte_flow_error_set(error, EINVAL,
7952 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7954 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7956 return rte_flow_error_set(error, ret,
7957 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7958 "failed to initialize pmd tunnel");
7960 *actions = &tunnel->action;
7961 *num_of_actions = 1;
7966 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
7967 struct rte_flow_tunnel *app_tunnel,
7968 struct rte_flow_item **items,
7969 uint32_t *num_of_items,
7970 struct rte_flow_error *error)
7973 struct mlx5_flow_tunnel *tunnel;
7974 const char *err_msg = NULL;
7975 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7978 return rte_flow_error_set(error, EINVAL,
7979 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7981 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7983 return rte_flow_error_set(error, ret,
7984 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7985 "failed to initialize pmd tunnel");
7987 *items = &tunnel->item;
7992 struct tunnel_db_element_release_ctx {
7993 struct rte_flow_item *items;
7994 struct rte_flow_action *actions;
7995 uint32_t num_elements;
7996 struct rte_flow_error *error;
8001 tunnel_element_release_match(struct rte_eth_dev *dev,
8002 struct mlx5_flow_tunnel *tunnel, const void *x)
8004 const struct tunnel_db_element_release_ctx *ctx = x;
8007 if (ctx->num_elements != 1)
8009 else if (ctx->items)
8010 return ctx->items == &tunnel->item;
8011 else if (ctx->actions)
8012 return ctx->actions == &tunnel->action;
8018 tunnel_element_release_hit(struct rte_eth_dev *dev,
8019 struct mlx5_flow_tunnel *tunnel, void *x)
8021 struct tunnel_db_element_release_ctx *ctx = x;
8023 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8024 mlx5_flow_tunnel_free(dev, tunnel);
8028 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8030 struct tunnel_db_element_release_ctx *ctx = x;
8032 ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8033 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8034 "invalid argument");
8038 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8039 struct rte_flow_item *pmd_items,
8040 uint32_t num_items, struct rte_flow_error *err)
8042 struct tunnel_db_element_release_ctx ctx = {
8045 .num_elements = num_items,
8049 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8050 tunnel_element_release_hit,
8051 tunnel_element_release_miss, &ctx, false);
8057 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8058 struct rte_flow_action *pmd_actions,
8059 uint32_t num_actions, struct rte_flow_error *err)
8061 struct tunnel_db_element_release_ctx ctx = {
8063 .actions = pmd_actions,
8064 .num_elements = num_actions,
8068 mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8069 tunnel_element_release_hit,
8070 tunnel_element_release_miss, &ctx, false);
8076 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8078 struct rte_flow_restore_info *info,
8079 struct rte_flow_error *err)
8081 uint64_t ol_flags = m->ol_flags;
8082 const struct mlx5_flow_tbl_data_entry *tble;
8083 const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8085 if (!is_tunnel_offload_active(dev)) {
8090 if ((ol_flags & mask) != mask)
8092 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8094 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8095 dev->data->port_id, m->hash.fdir.hi);
8098 MLX5_ASSERT(tble->tunnel);
8099 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8100 info->group_id = tble->group_id;
8101 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8102 RTE_FLOW_RESTORE_INFO_GROUP_ID |
8103 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8108 return rte_flow_error_set(err, EINVAL,
8109 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8110 "failed to get restore info");
8113 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8115 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8116 __rte_unused struct rte_flow_tunnel *app_tunnel,
8117 __rte_unused struct rte_flow_action **actions,
8118 __rte_unused uint32_t *num_of_actions,
8119 __rte_unused struct rte_flow_error *error)
8125 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8126 __rte_unused struct rte_flow_tunnel *app_tunnel,
8127 __rte_unused struct rte_flow_item **items,
8128 __rte_unused uint32_t *num_of_items,
8129 __rte_unused struct rte_flow_error *error)
8135 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8136 __rte_unused struct rte_flow_item *pmd_items,
8137 __rte_unused uint32_t num_items,
8138 __rte_unused struct rte_flow_error *err)
8144 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8145 __rte_unused struct rte_flow_action *pmd_action,
8146 __rte_unused uint32_t num_actions,
8147 __rte_unused struct rte_flow_error *err)
8153 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8154 __rte_unused struct rte_mbuf *m,
8155 __rte_unused struct rte_flow_restore_info *i,
8156 __rte_unused struct rte_flow_error *err)
8162 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8163 __rte_unused struct rte_flow *flow,
8164 __rte_unused const struct rte_flow_attr *attr,
8165 __rte_unused const struct rte_flow_action *actions,
8166 __rte_unused uint32_t flow_idx,
8167 __rte_unused struct tunnel_default_miss_ctx *ctx,
8168 __rte_unused struct rte_flow_error *error)
8173 static struct mlx5_flow_tunnel *
8174 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8175 __rte_unused uint32_t id)
8181 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8182 __rte_unused struct mlx5_flow_tunnel *tunnel)
8187 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8188 __rte_unused const struct mlx5_flow_tunnel *t,
8189 __rte_unused uint32_t group,
8190 __rte_unused uint32_t *table,
8191 struct rte_flow_error *error)
8193 return rte_flow_error_set(error, ENOTSUP,
8194 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8195 "tunnel offload requires DV support");
8199 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8200 __rte_unused uint16_t port_id)
8203 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */