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 <rte_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 static struct mlx5_flow_tunnel *
37 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
39 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
40 static const struct mlx5_flow_tbl_data_entry *
41 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark);
43 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
44 const struct rte_flow_tunnel *app_tunnel,
45 struct mlx5_flow_tunnel **tunnel);
48 /** Device flow drivers. */
49 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
51 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
53 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
54 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
55 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
56 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
58 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
59 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
62 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
63 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
68 /** Node object of input graph for mlx5_flow_expand_rss(). */
69 struct mlx5_flow_expand_node {
70 const int *const next;
72 * List of next node indexes. Index 0 is interpreted as a terminator.
74 const enum rte_flow_item_type type;
75 /**< Pattern item type of current node. */
78 * RSS types bit-field associated with this node
79 * (see ETH_RSS_* definitions).
83 /** Object returned by mlx5_flow_expand_rss(). */
84 struct mlx5_flow_expand_rss {
86 /**< Number of entries @p patterns and @p priorities. */
88 struct rte_flow_item *pattern; /**< Expanded pattern array. */
89 uint32_t priority; /**< Priority offset for each expansion. */
93 static enum rte_flow_item_type
94 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
96 enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
97 uint16_t ether_type = 0;
98 uint16_t ether_type_m;
99 uint8_t ip_next_proto = 0;
100 uint8_t ip_next_proto_m;
102 if (item == NULL || item->spec == NULL)
104 switch (item->type) {
105 case RTE_FLOW_ITEM_TYPE_ETH:
107 ether_type_m = ((const struct rte_flow_item_eth *)
110 ether_type_m = rte_flow_item_eth_mask.type;
111 if (ether_type_m != RTE_BE16(0xFFFF))
113 ether_type = ((const struct rte_flow_item_eth *)
115 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
116 ret = RTE_FLOW_ITEM_TYPE_IPV4;
117 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
118 ret = RTE_FLOW_ITEM_TYPE_IPV6;
119 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
120 ret = RTE_FLOW_ITEM_TYPE_VLAN;
122 ret = RTE_FLOW_ITEM_TYPE_END;
124 case RTE_FLOW_ITEM_TYPE_VLAN:
126 ether_type_m = ((const struct rte_flow_item_vlan *)
127 (item->mask))->inner_type;
129 ether_type_m = rte_flow_item_vlan_mask.inner_type;
130 if (ether_type_m != RTE_BE16(0xFFFF))
132 ether_type = ((const struct rte_flow_item_vlan *)
133 (item->spec))->inner_type;
134 if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
135 ret = RTE_FLOW_ITEM_TYPE_IPV4;
136 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
137 ret = RTE_FLOW_ITEM_TYPE_IPV6;
138 else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
139 ret = RTE_FLOW_ITEM_TYPE_VLAN;
141 ret = RTE_FLOW_ITEM_TYPE_END;
143 case RTE_FLOW_ITEM_TYPE_IPV4:
145 ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
146 (item->mask))->hdr.next_proto_id;
149 rte_flow_item_ipv4_mask.hdr.next_proto_id;
150 if (ip_next_proto_m != 0xFF)
152 ip_next_proto = ((const struct rte_flow_item_ipv4 *)
153 (item->spec))->hdr.next_proto_id;
154 if (ip_next_proto == IPPROTO_UDP)
155 ret = RTE_FLOW_ITEM_TYPE_UDP;
156 else if (ip_next_proto == IPPROTO_TCP)
157 ret = RTE_FLOW_ITEM_TYPE_TCP;
158 else if (ip_next_proto == IPPROTO_IP)
159 ret = RTE_FLOW_ITEM_TYPE_IPV4;
160 else if (ip_next_proto == IPPROTO_IPV6)
161 ret = RTE_FLOW_ITEM_TYPE_IPV6;
163 ret = RTE_FLOW_ITEM_TYPE_END;
165 case RTE_FLOW_ITEM_TYPE_IPV6:
167 ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
168 (item->mask))->hdr.proto;
171 rte_flow_item_ipv6_mask.hdr.proto;
172 if (ip_next_proto_m != 0xFF)
174 ip_next_proto = ((const struct rte_flow_item_ipv6 *)
175 (item->spec))->hdr.proto;
176 if (ip_next_proto == IPPROTO_UDP)
177 ret = RTE_FLOW_ITEM_TYPE_UDP;
178 else if (ip_next_proto == IPPROTO_TCP)
179 ret = RTE_FLOW_ITEM_TYPE_TCP;
180 else if (ip_next_proto == IPPROTO_IP)
181 ret = RTE_FLOW_ITEM_TYPE_IPV4;
182 else if (ip_next_proto == IPPROTO_IPV6)
183 ret = RTE_FLOW_ITEM_TYPE_IPV6;
185 ret = RTE_FLOW_ITEM_TYPE_END;
188 ret = RTE_FLOW_ITEM_TYPE_VOID;
195 * Expand RSS flows into several possible flows according to the RSS hash
196 * fields requested and the driver capabilities.
199 * Buffer to store the result expansion.
201 * Buffer size in bytes. If 0, @p buf can be NULL.
205 * RSS types to expand (see ETH_RSS_* definitions).
207 * Input graph to expand @p pattern according to @p types.
208 * @param[in] graph_root_index
209 * Index of root node in @p graph, typically 0.
212 * A positive value representing the size of @p buf in bytes regardless of
213 * @p size on success, a negative errno value otherwise and rte_errno is
214 * set, the following errors are defined:
216 * -E2BIG: graph-depth @p graph is too deep.
219 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
220 const struct rte_flow_item *pattern, uint64_t types,
221 const struct mlx5_flow_expand_node graph[],
222 int graph_root_index)
225 const struct rte_flow_item *item;
226 const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
227 const int *next_node;
228 const int *stack[elt_n];
230 struct rte_flow_item flow_items[elt_n];
233 size_t user_pattern_size = 0;
235 const struct mlx5_flow_expand_node *next = NULL;
236 struct rte_flow_item missed_item;
239 const struct rte_flow_item *last_item = NULL;
241 memset(&missed_item, 0, sizeof(missed_item));
242 lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
243 elt_n * sizeof(buf->entry[0]);
245 buf->entry[0].priority = 0;
246 buf->entry[0].pattern = (void *)&buf->entry[elt_n];
248 addr = buf->entry[0].pattern;
250 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
251 if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
253 for (i = 0; node->next && node->next[i]; ++i) {
254 next = &graph[node->next[i]];
255 if (next->type == item->type)
260 user_pattern_size += sizeof(*item);
262 user_pattern_size += sizeof(*item); /* Handle END item. */
263 lsize += user_pattern_size;
264 /* Copy the user pattern in the first entry of the buffer. */
266 rte_memcpy(addr, pattern, user_pattern_size);
267 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
270 /* Start expanding. */
271 memset(flow_items, 0, sizeof(flow_items));
272 user_pattern_size -= sizeof(*item);
274 * Check if the last valid item has spec set, need complete pattern,
275 * and the pattern can be used for expansion.
277 missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
278 if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
279 /* Item type END indicates expansion is not required. */
282 if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
285 for (i = 0; node->next && node->next[i]; ++i) {
286 next = &graph[node->next[i]];
287 if (next->type == missed_item.type) {
288 flow_items[0].type = missed_item.type;
289 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
295 if (next && missed) {
296 elt = 2; /* missed item + item end. */
298 lsize += elt * sizeof(*item) + user_pattern_size;
299 if ((node->rss_types & types) && lsize <= size) {
300 buf->entry[buf->entries].priority = 1;
301 buf->entry[buf->entries].pattern = addr;
303 rte_memcpy(addr, buf->entry[0].pattern,
305 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
306 rte_memcpy(addr, flow_items, elt * sizeof(*item));
307 addr = (void *)(((uintptr_t)addr) +
308 elt * sizeof(*item));
311 memset(flow_items, 0, sizeof(flow_items));
312 next_node = node->next;
313 stack[stack_pos] = next_node;
314 node = next_node ? &graph[*next_node] : NULL;
316 flow_items[stack_pos].type = node->type;
317 if (node->rss_types & types) {
319 * compute the number of items to copy from the
320 * expansion and copy it.
321 * When the stack_pos is 0, there are 1 element in it,
322 * plus the addition END item.
325 flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
326 lsize += elt * sizeof(*item) + user_pattern_size;
328 size_t n = elt * sizeof(*item);
330 buf->entry[buf->entries].priority =
331 stack_pos + 1 + missed;
332 buf->entry[buf->entries].pattern = addr;
334 rte_memcpy(addr, buf->entry[0].pattern,
336 addr = (void *)(((uintptr_t)addr) +
338 rte_memcpy(addr, &missed_item,
339 missed * sizeof(*item));
340 addr = (void *)(((uintptr_t)addr) +
341 missed * sizeof(*item));
342 rte_memcpy(addr, flow_items, n);
343 addr = (void *)(((uintptr_t)addr) + n);
348 next_node = node->next;
349 if (stack_pos++ == elt_n) {
353 stack[stack_pos] = next_node;
354 } else if (*(next_node + 1)) {
355 /* Follow up with the next possibility. */
358 /* Move to the next path. */
360 next_node = stack[--stack_pos];
362 stack[stack_pos] = next_node;
364 node = *next_node ? &graph[*next_node] : NULL;
366 /* no expanded flows but we have missed item, create one rule for it */
367 if (buf->entries == 1 && missed != 0) {
369 lsize += elt * sizeof(*item) + user_pattern_size;
371 buf->entry[buf->entries].priority = 1;
372 buf->entry[buf->entries].pattern = addr;
374 flow_items[0].type = missed_item.type;
375 flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
376 rte_memcpy(addr, buf->entry[0].pattern,
378 addr = (void *)(((uintptr_t)addr) + user_pattern_size);
379 rte_memcpy(addr, flow_items, elt * sizeof(*item));
380 addr = (void *)(((uintptr_t)addr) +
381 elt * sizeof(*item));
387 enum mlx5_expansion {
389 MLX5_EXPANSION_ROOT_OUTER,
390 MLX5_EXPANSION_ROOT_ETH_VLAN,
391 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
392 MLX5_EXPANSION_OUTER_ETH,
393 MLX5_EXPANSION_OUTER_ETH_VLAN,
394 MLX5_EXPANSION_OUTER_VLAN,
395 MLX5_EXPANSION_OUTER_IPV4,
396 MLX5_EXPANSION_OUTER_IPV4_UDP,
397 MLX5_EXPANSION_OUTER_IPV4_TCP,
398 MLX5_EXPANSION_OUTER_IPV6,
399 MLX5_EXPANSION_OUTER_IPV6_UDP,
400 MLX5_EXPANSION_OUTER_IPV6_TCP,
401 MLX5_EXPANSION_VXLAN,
402 MLX5_EXPANSION_VXLAN_GPE,
406 MLX5_EXPANSION_ETH_VLAN,
409 MLX5_EXPANSION_IPV4_UDP,
410 MLX5_EXPANSION_IPV4_TCP,
412 MLX5_EXPANSION_IPV6_UDP,
413 MLX5_EXPANSION_IPV6_TCP,
416 /** Supported expansion of items. */
417 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
418 [MLX5_EXPANSION_ROOT] = {
419 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
421 MLX5_EXPANSION_IPV6),
422 .type = RTE_FLOW_ITEM_TYPE_END,
424 [MLX5_EXPANSION_ROOT_OUTER] = {
425 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
426 MLX5_EXPANSION_OUTER_IPV4,
427 MLX5_EXPANSION_OUTER_IPV6),
428 .type = RTE_FLOW_ITEM_TYPE_END,
430 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
431 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
432 .type = RTE_FLOW_ITEM_TYPE_END,
434 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
435 .next = MLX5_FLOW_EXPAND_RSS_NEXT
436 (MLX5_EXPANSION_OUTER_ETH_VLAN),
437 .type = RTE_FLOW_ITEM_TYPE_END,
439 [MLX5_EXPANSION_OUTER_ETH] = {
440 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
441 MLX5_EXPANSION_OUTER_IPV6,
442 MLX5_EXPANSION_MPLS),
443 .type = RTE_FLOW_ITEM_TYPE_ETH,
446 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
447 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
448 .type = RTE_FLOW_ITEM_TYPE_ETH,
451 [MLX5_EXPANSION_OUTER_VLAN] = {
452 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
453 MLX5_EXPANSION_OUTER_IPV6),
454 .type = RTE_FLOW_ITEM_TYPE_VLAN,
456 [MLX5_EXPANSION_OUTER_IPV4] = {
457 .next = MLX5_FLOW_EXPAND_RSS_NEXT
458 (MLX5_EXPANSION_OUTER_IPV4_UDP,
459 MLX5_EXPANSION_OUTER_IPV4_TCP,
462 MLX5_EXPANSION_IPV6),
463 .type = RTE_FLOW_ITEM_TYPE_IPV4,
464 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
465 ETH_RSS_NONFRAG_IPV4_OTHER,
467 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
468 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
469 MLX5_EXPANSION_VXLAN_GPE),
470 .type = RTE_FLOW_ITEM_TYPE_UDP,
471 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
473 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
474 .type = RTE_FLOW_ITEM_TYPE_TCP,
475 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
477 [MLX5_EXPANSION_OUTER_IPV6] = {
478 .next = MLX5_FLOW_EXPAND_RSS_NEXT
479 (MLX5_EXPANSION_OUTER_IPV6_UDP,
480 MLX5_EXPANSION_OUTER_IPV6_TCP,
482 MLX5_EXPANSION_IPV6),
483 .type = RTE_FLOW_ITEM_TYPE_IPV6,
484 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
485 ETH_RSS_NONFRAG_IPV6_OTHER,
487 [MLX5_EXPANSION_OUTER_IPV6_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_IPV6_UDP,
493 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
494 .type = RTE_FLOW_ITEM_TYPE_TCP,
495 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
497 [MLX5_EXPANSION_VXLAN] = {
498 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
500 MLX5_EXPANSION_IPV6),
501 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
503 [MLX5_EXPANSION_VXLAN_GPE] = {
504 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
506 MLX5_EXPANSION_IPV6),
507 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
509 [MLX5_EXPANSION_GRE] = {
510 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
511 .type = RTE_FLOW_ITEM_TYPE_GRE,
513 [MLX5_EXPANSION_MPLS] = {
514 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
515 MLX5_EXPANSION_IPV6),
516 .type = RTE_FLOW_ITEM_TYPE_MPLS,
518 [MLX5_EXPANSION_ETH] = {
519 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
520 MLX5_EXPANSION_IPV6),
521 .type = RTE_FLOW_ITEM_TYPE_ETH,
523 [MLX5_EXPANSION_ETH_VLAN] = {
524 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
525 .type = RTE_FLOW_ITEM_TYPE_ETH,
527 [MLX5_EXPANSION_VLAN] = {
528 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
529 MLX5_EXPANSION_IPV6),
530 .type = RTE_FLOW_ITEM_TYPE_VLAN,
532 [MLX5_EXPANSION_IPV4] = {
533 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
534 MLX5_EXPANSION_IPV4_TCP),
535 .type = RTE_FLOW_ITEM_TYPE_IPV4,
536 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
537 ETH_RSS_NONFRAG_IPV4_OTHER,
539 [MLX5_EXPANSION_IPV4_UDP] = {
540 .type = RTE_FLOW_ITEM_TYPE_UDP,
541 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
543 [MLX5_EXPANSION_IPV4_TCP] = {
544 .type = RTE_FLOW_ITEM_TYPE_TCP,
545 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
547 [MLX5_EXPANSION_IPV6] = {
548 .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
549 MLX5_EXPANSION_IPV6_TCP),
550 .type = RTE_FLOW_ITEM_TYPE_IPV6,
551 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
552 ETH_RSS_NONFRAG_IPV6_OTHER,
554 [MLX5_EXPANSION_IPV6_UDP] = {
555 .type = RTE_FLOW_ITEM_TYPE_UDP,
556 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
558 [MLX5_EXPANSION_IPV6_TCP] = {
559 .type = RTE_FLOW_ITEM_TYPE_TCP,
560 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
564 static struct rte_flow_shared_action *
565 mlx5_shared_action_create(struct rte_eth_dev *dev,
566 const struct rte_flow_shared_action_conf *conf,
567 const struct rte_flow_action *action,
568 struct rte_flow_error *error);
569 static int mlx5_shared_action_destroy
570 (struct rte_eth_dev *dev,
571 struct rte_flow_shared_action *shared_action,
572 struct rte_flow_error *error);
573 static int mlx5_shared_action_update
574 (struct rte_eth_dev *dev,
575 struct rte_flow_shared_action *shared_action,
576 const struct rte_flow_action *action,
577 struct rte_flow_error *error);
578 static int mlx5_shared_action_query
579 (struct rte_eth_dev *dev,
580 const struct rte_flow_shared_action *action,
582 struct rte_flow_error *error);
584 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
585 struct rte_flow_tunnel *tunnel,
589 if (!is_tunnel_offload_active(dev)) {
590 err_msg = "tunnel offload was not activated";
592 } else if (!tunnel) {
593 err_msg = "no application tunnel";
597 switch (tunnel->type) {
599 err_msg = "unsupported tunnel type";
601 case RTE_FLOW_ITEM_TYPE_VXLAN:
611 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
612 struct rte_flow_tunnel *app_tunnel,
613 struct rte_flow_action **actions,
614 uint32_t *num_of_actions,
615 struct rte_flow_error *error)
618 struct mlx5_flow_tunnel *tunnel;
619 const char *err_msg = NULL;
620 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
623 return rte_flow_error_set(error, EINVAL,
624 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
626 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
628 return rte_flow_error_set(error, ret,
629 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
630 "failed to initialize pmd tunnel");
632 *actions = &tunnel->action;
638 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
639 struct rte_flow_tunnel *app_tunnel,
640 struct rte_flow_item **items,
641 uint32_t *num_of_items,
642 struct rte_flow_error *error)
645 struct mlx5_flow_tunnel *tunnel;
646 const char *err_msg = NULL;
647 bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
650 return rte_flow_error_set(error, EINVAL,
651 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
653 ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
655 return rte_flow_error_set(error, ret,
656 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
657 "failed to initialize pmd tunnel");
659 *items = &tunnel->item;
665 mlx5_flow_item_release(struct rte_eth_dev *dev,
666 struct rte_flow_item *pmd_items,
667 uint32_t num_items, struct rte_flow_error *err)
669 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
670 struct mlx5_flow_tunnel *tun;
672 LIST_FOREACH(tun, &thub->tunnels, chain) {
673 if (&tun->item == pmd_items)
676 if (!tun || num_items != 1)
677 return rte_flow_error_set(err, EINVAL,
678 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
680 if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
681 mlx5_flow_tunnel_free(dev, tun);
686 mlx5_flow_action_release(struct rte_eth_dev *dev,
687 struct rte_flow_action *pmd_actions,
688 uint32_t num_actions, struct rte_flow_error *err)
690 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
691 struct mlx5_flow_tunnel *tun;
693 LIST_FOREACH(tun, &thub->tunnels, chain) {
694 if (&tun->action == pmd_actions)
697 if (!tun || num_actions != 1)
698 return rte_flow_error_set(err, EINVAL,
699 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
701 if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
702 mlx5_flow_tunnel_free(dev, tun);
708 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
710 struct rte_flow_restore_info *info,
711 struct rte_flow_error *err)
713 uint64_t ol_flags = m->ol_flags;
714 const struct mlx5_flow_tbl_data_entry *tble;
715 const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
717 if ((ol_flags & mask) != mask)
719 tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
721 DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
722 dev->data->port_id, m->hash.fdir.hi);
725 MLX5_ASSERT(tble->tunnel);
726 memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
727 info->group_id = tble->group_id;
728 info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
729 RTE_FLOW_RESTORE_INFO_GROUP_ID |
730 RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
735 return rte_flow_error_set(err, EINVAL,
736 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
737 "failed to get restore info");
740 static const struct rte_flow_ops mlx5_flow_ops = {
741 .validate = mlx5_flow_validate,
742 .create = mlx5_flow_create,
743 .destroy = mlx5_flow_destroy,
744 .flush = mlx5_flow_flush,
745 .isolate = mlx5_flow_isolate,
746 .query = mlx5_flow_query,
747 .dev_dump = mlx5_flow_dev_dump,
748 .get_aged_flows = mlx5_flow_get_aged_flows,
749 .shared_action_create = mlx5_shared_action_create,
750 .shared_action_destroy = mlx5_shared_action_destroy,
751 .shared_action_update = mlx5_shared_action_update,
752 .shared_action_query = mlx5_shared_action_query,
753 .tunnel_decap_set = mlx5_flow_tunnel_decap_set,
754 .tunnel_match = mlx5_flow_tunnel_match,
755 .tunnel_action_decap_release = mlx5_flow_action_release,
756 .tunnel_item_release = mlx5_flow_item_release,
757 .get_restore_info = mlx5_flow_tunnel_get_restore_info,
760 /* Convert FDIR request to Generic flow. */
762 struct rte_flow_attr attr;
763 struct rte_flow_item items[4];
764 struct rte_flow_item_eth l2;
765 struct rte_flow_item_eth l2_mask;
767 struct rte_flow_item_ipv4 ipv4;
768 struct rte_flow_item_ipv6 ipv6;
771 struct rte_flow_item_ipv4 ipv4;
772 struct rte_flow_item_ipv6 ipv6;
775 struct rte_flow_item_udp udp;
776 struct rte_flow_item_tcp tcp;
779 struct rte_flow_item_udp udp;
780 struct rte_flow_item_tcp tcp;
782 struct rte_flow_action actions[2];
783 struct rte_flow_action_queue queue;
786 /* Tunnel information. */
787 struct mlx5_flow_tunnel_info {
788 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
789 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
792 static struct mlx5_flow_tunnel_info tunnels_info[] = {
794 .tunnel = MLX5_FLOW_LAYER_VXLAN,
795 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
798 .tunnel = MLX5_FLOW_LAYER_GENEVE,
799 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
802 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
803 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
806 .tunnel = MLX5_FLOW_LAYER_GRE,
807 .ptype = RTE_PTYPE_TUNNEL_GRE,
810 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
811 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
814 .tunnel = MLX5_FLOW_LAYER_MPLS,
815 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
818 .tunnel = MLX5_FLOW_LAYER_NVGRE,
819 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
822 .tunnel = MLX5_FLOW_LAYER_IPIP,
823 .ptype = RTE_PTYPE_TUNNEL_IP,
826 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
827 .ptype = RTE_PTYPE_TUNNEL_IP,
830 .tunnel = MLX5_FLOW_LAYER_GTP,
831 .ptype = RTE_PTYPE_TUNNEL_GTPU,
835 /* Key of thread specific flow workspace data. */
836 static pthread_key_t key_workspace;
838 /* Thread specific flow workspace data once initialization data. */
839 static pthread_once_t key_workspace_init;
843 * Translate tag ID to register.
846 * Pointer to the Ethernet device structure.
848 * The feature that request the register.
850 * The request register ID.
852 * Error description in case of any.
855 * The request register on success, a negative errno
856 * value otherwise and rte_errno is set.
859 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
860 enum mlx5_feature_name feature,
862 struct rte_flow_error *error)
864 struct mlx5_priv *priv = dev->data->dev_private;
865 struct mlx5_dev_config *config = &priv->config;
866 enum modify_reg start_reg;
867 bool skip_mtr_reg = false;
870 case MLX5_HAIRPIN_RX:
872 case MLX5_HAIRPIN_TX:
874 case MLX5_METADATA_RX:
875 switch (config->dv_xmeta_en) {
876 case MLX5_XMETA_MODE_LEGACY:
878 case MLX5_XMETA_MODE_META16:
880 case MLX5_XMETA_MODE_META32:
884 case MLX5_METADATA_TX:
886 case MLX5_METADATA_FDB:
887 switch (config->dv_xmeta_en) {
888 case MLX5_XMETA_MODE_LEGACY:
890 case MLX5_XMETA_MODE_META16:
892 case MLX5_XMETA_MODE_META32:
897 switch (config->dv_xmeta_en) {
898 case MLX5_XMETA_MODE_LEGACY:
900 case MLX5_XMETA_MODE_META16:
902 case MLX5_XMETA_MODE_META32:
908 * If meter color and flow match share one register, flow match
909 * should use the meter color register for match.
911 if (priv->mtr_reg_share)
912 return priv->mtr_color_reg;
914 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
917 MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
918 return priv->mtr_color_reg;
921 * Metadata COPY_MARK register using is in meter suffix sub
922 * flow while with meter. It's safe to share the same register.
924 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
927 * If meter is enable, it will engage the register for color
928 * match and flow match. If meter color match is not using the
929 * REG_C_2, need to skip the REG_C_x be used by meter color
931 * If meter is disable, free to use all available registers.
933 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
934 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
935 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
936 if (id > (REG_C_7 - start_reg))
937 return rte_flow_error_set(error, EINVAL,
938 RTE_FLOW_ERROR_TYPE_ITEM,
939 NULL, "invalid tag id");
940 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
941 return rte_flow_error_set(error, ENOTSUP,
942 RTE_FLOW_ERROR_TYPE_ITEM,
943 NULL, "unsupported tag id");
945 * This case means meter is using the REG_C_x great than 2.
946 * Take care not to conflict with meter color REG_C_x.
947 * If the available index REG_C_y >= REG_C_x, skip the
950 if (skip_mtr_reg && config->flow_mreg_c
951 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
952 if (id >= (REG_C_7 - start_reg))
953 return rte_flow_error_set(error, EINVAL,
954 RTE_FLOW_ERROR_TYPE_ITEM,
955 NULL, "invalid tag id");
956 if (config->flow_mreg_c
957 [id + 1 + start_reg - REG_C_0] != REG_NON)
958 return config->flow_mreg_c
959 [id + 1 + start_reg - REG_C_0];
960 return rte_flow_error_set(error, ENOTSUP,
961 RTE_FLOW_ERROR_TYPE_ITEM,
962 NULL, "unsupported tag id");
964 return config->flow_mreg_c[id + start_reg - REG_C_0];
967 return rte_flow_error_set(error, EINVAL,
968 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
969 NULL, "invalid feature name");
973 * Check extensive flow metadata register support.
976 * Pointer to rte_eth_dev structure.
979 * True if device supports extensive flow metadata register, otherwise false.
982 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
984 struct mlx5_priv *priv = dev->data->dev_private;
985 struct mlx5_dev_config *config = &priv->config;
988 * Having available reg_c can be regarded inclusively as supporting
989 * extensive flow metadata register, which could mean,
990 * - metadata register copy action by modify header.
991 * - 16 modify header actions is supported.
992 * - reg_c's are preserved across different domain (FDB and NIC) on
993 * packet loopback by flow lookup miss.
995 return config->flow_mreg_c[2] != REG_NON;
999 * Verify the @p item specifications (spec, last, mask) are compatible with the
1003 * Item specification.
1005 * @p item->mask or flow default bit-masks.
1006 * @param[in] nic_mask
1007 * Bit-masks covering supported fields by the NIC to compare with user mask.
1009 * Bit-masks size in bytes.
1010 * @param[in] range_accepted
1011 * True if range of values is accepted for specific fields, false otherwise.
1013 * Pointer to error structure.
1016 * 0 on success, a negative errno value otherwise and rte_errno is set.
1019 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1020 const uint8_t *mask,
1021 const uint8_t *nic_mask,
1023 bool range_accepted,
1024 struct rte_flow_error *error)
1028 MLX5_ASSERT(nic_mask);
1029 for (i = 0; i < size; ++i)
1030 if ((nic_mask[i] | mask[i]) != nic_mask[i])
1031 return rte_flow_error_set(error, ENOTSUP,
1032 RTE_FLOW_ERROR_TYPE_ITEM,
1034 "mask enables non supported"
1036 if (!item->spec && (item->mask || item->last))
1037 return rte_flow_error_set(error, EINVAL,
1038 RTE_FLOW_ERROR_TYPE_ITEM, item,
1039 "mask/last without a spec is not"
1041 if (item->spec && item->last && !range_accepted) {
1047 for (i = 0; i < size; ++i) {
1048 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1049 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1051 ret = memcmp(spec, last, size);
1053 return rte_flow_error_set(error, EINVAL,
1054 RTE_FLOW_ERROR_TYPE_ITEM,
1056 "range is not valid");
1062 * Adjust the hash fields according to the @p flow information.
1064 * @param[in] dev_flow.
1065 * Pointer to the mlx5_flow.
1067 * 1 when the hash field is for a tunnel item.
1068 * @param[in] layer_types
1070 * @param[in] hash_fields
1074 * The hash fields that should be used.
1077 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1078 int tunnel __rte_unused, uint64_t layer_types,
1079 uint64_t hash_fields)
1081 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1082 int rss_request_inner = rss_desc->level >= 2;
1084 /* Check RSS hash level for tunnel. */
1085 if (tunnel && rss_request_inner)
1086 hash_fields |= IBV_RX_HASH_INNER;
1087 else if (tunnel || rss_request_inner)
1090 /* Check if requested layer matches RSS hash fields. */
1091 if (!(rss_desc->types & layer_types))
1097 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
1098 * if several tunnel rules are used on this queue, the tunnel ptype will be
1102 * Rx queue to update.
1105 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1108 uint32_t tunnel_ptype = 0;
1110 /* Look up for the ptype to use. */
1111 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1112 if (!rxq_ctrl->flow_tunnels_n[i])
1114 if (!tunnel_ptype) {
1115 tunnel_ptype = tunnels_info[i].ptype;
1121 rxq_ctrl->rxq.tunnel = tunnel_ptype;
1125 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1129 * Pointer to the Ethernet device structure.
1130 * @param[in] dev_handle
1131 * Pointer to device flow handle structure.
1134 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1135 struct mlx5_flow_handle *dev_handle)
1137 struct mlx5_priv *priv = dev->data->dev_private;
1138 const int mark = dev_handle->mark;
1139 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1140 struct mlx5_hrxq *hrxq;
1143 if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1145 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1146 dev_handle->rix_hrxq);
1149 for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1150 int idx = hrxq->ind_table->queues[i];
1151 struct mlx5_rxq_ctrl *rxq_ctrl =
1152 container_of((*priv->rxqs)[idx],
1153 struct mlx5_rxq_ctrl, rxq);
1156 * To support metadata register copy on Tx loopback,
1157 * this must be always enabled (metadata may arive
1158 * from other port - not from local flows only.
1160 if (priv->config.dv_flow_en &&
1161 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1162 mlx5_flow_ext_mreg_supported(dev)) {
1163 rxq_ctrl->rxq.mark = 1;
1164 rxq_ctrl->flow_mark_n = 1;
1166 rxq_ctrl->rxq.mark = 1;
1167 rxq_ctrl->flow_mark_n++;
1172 /* Increase the counter matching the flow. */
1173 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1174 if ((tunnels_info[j].tunnel &
1175 dev_handle->layers) ==
1176 tunnels_info[j].tunnel) {
1177 rxq_ctrl->flow_tunnels_n[j]++;
1181 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1187 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1190 * Pointer to the Ethernet device structure.
1192 * Pointer to flow structure.
1195 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1197 struct mlx5_priv *priv = dev->data->dev_private;
1198 uint32_t handle_idx;
1199 struct mlx5_flow_handle *dev_handle;
1201 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1202 handle_idx, dev_handle, next)
1203 flow_drv_rxq_flags_set(dev, dev_handle);
1207 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1208 * device flow if no other flow uses it with the same kind of request.
1211 * Pointer to Ethernet device.
1212 * @param[in] dev_handle
1213 * Pointer to the device flow handle structure.
1216 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1217 struct mlx5_flow_handle *dev_handle)
1219 struct mlx5_priv *priv = dev->data->dev_private;
1220 const int mark = dev_handle->mark;
1221 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1222 struct mlx5_hrxq *hrxq;
1225 if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1227 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1228 dev_handle->rix_hrxq);
1231 MLX5_ASSERT(dev->data->dev_started);
1232 for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1233 int idx = hrxq->ind_table->queues[i];
1234 struct mlx5_rxq_ctrl *rxq_ctrl =
1235 container_of((*priv->rxqs)[idx],
1236 struct mlx5_rxq_ctrl, rxq);
1238 if (priv->config.dv_flow_en &&
1239 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1240 mlx5_flow_ext_mreg_supported(dev)) {
1241 rxq_ctrl->rxq.mark = 1;
1242 rxq_ctrl->flow_mark_n = 1;
1244 rxq_ctrl->flow_mark_n--;
1245 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1250 /* Decrease the counter matching the flow. */
1251 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1252 if ((tunnels_info[j].tunnel &
1253 dev_handle->layers) ==
1254 tunnels_info[j].tunnel) {
1255 rxq_ctrl->flow_tunnels_n[j]--;
1259 flow_rxq_tunnel_ptype_update(rxq_ctrl);
1265 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1266 * @p flow if no other flow uses it with the same kind of request.
1269 * Pointer to Ethernet device.
1271 * Pointer to the flow.
1274 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1276 struct mlx5_priv *priv = dev->data->dev_private;
1277 uint32_t handle_idx;
1278 struct mlx5_flow_handle *dev_handle;
1280 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1281 handle_idx, dev_handle, next)
1282 flow_drv_rxq_flags_trim(dev, dev_handle);
1286 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1289 * Pointer to Ethernet device.
1292 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1294 struct mlx5_priv *priv = dev->data->dev_private;
1297 for (i = 0; i != priv->rxqs_n; ++i) {
1298 struct mlx5_rxq_ctrl *rxq_ctrl;
1301 if (!(*priv->rxqs)[i])
1303 rxq_ctrl = container_of((*priv->rxqs)[i],
1304 struct mlx5_rxq_ctrl, rxq);
1305 rxq_ctrl->flow_mark_n = 0;
1306 rxq_ctrl->rxq.mark = 0;
1307 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1308 rxq_ctrl->flow_tunnels_n[j] = 0;
1309 rxq_ctrl->rxq.tunnel = 0;
1314 * Set the Rx queue dynamic metadata (mask and offset) for a flow
1317 * Pointer to the Ethernet device structure.
1320 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1322 struct mlx5_priv *priv = dev->data->dev_private;
1323 struct mlx5_rxq_data *data;
1326 for (i = 0; i != priv->rxqs_n; ++i) {
1327 if (!(*priv->rxqs)[i])
1329 data = (*priv->rxqs)[i];
1330 if (!rte_flow_dynf_metadata_avail()) {
1331 data->dynf_meta = 0;
1332 data->flow_meta_mask = 0;
1333 data->flow_meta_offset = -1;
1335 data->dynf_meta = 1;
1336 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1337 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1343 * return a pointer to the desired action in the list of actions.
1345 * @param[in] actions
1346 * The list of actions to search the action in.
1348 * The action to find.
1351 * Pointer to the action in the list, if found. NULL otherwise.
1353 const struct rte_flow_action *
1354 mlx5_flow_find_action(const struct rte_flow_action *actions,
1355 enum rte_flow_action_type action)
1357 if (actions == NULL)
1359 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1360 if (actions->type == action)
1366 * Validate the flag action.
1368 * @param[in] action_flags
1369 * Bit-fields that holds the actions detected until now.
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_flag(uint64_t action_flags,
1380 const struct rte_flow_attr *attr,
1381 struct rte_flow_error *error)
1383 if (action_flags & MLX5_FLOW_ACTION_MARK)
1384 return rte_flow_error_set(error, EINVAL,
1385 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1386 "can't mark and flag in same flow");
1387 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1388 return rte_flow_error_set(error, EINVAL,
1389 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1391 " actions in same flow");
1393 return rte_flow_error_set(error, ENOTSUP,
1394 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1395 "flag action not supported for "
1401 * Validate the mark action.
1404 * Pointer to the queue action.
1405 * @param[in] action_flags
1406 * Bit-fields that holds the actions detected until now.
1408 * Attributes of flow that includes this action.
1410 * Pointer to error structure.
1413 * 0 on success, a negative errno value otherwise and rte_errno is set.
1416 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1417 uint64_t action_flags,
1418 const struct rte_flow_attr *attr,
1419 struct rte_flow_error *error)
1421 const struct rte_flow_action_mark *mark = action->conf;
1424 return rte_flow_error_set(error, EINVAL,
1425 RTE_FLOW_ERROR_TYPE_ACTION,
1427 "configuration cannot be null");
1428 if (mark->id >= MLX5_FLOW_MARK_MAX)
1429 return rte_flow_error_set(error, EINVAL,
1430 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1432 "mark id must in 0 <= id < "
1433 RTE_STR(MLX5_FLOW_MARK_MAX));
1434 if (action_flags & MLX5_FLOW_ACTION_FLAG)
1435 return rte_flow_error_set(error, EINVAL,
1436 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1437 "can't flag and mark in same flow");
1438 if (action_flags & MLX5_FLOW_ACTION_MARK)
1439 return rte_flow_error_set(error, EINVAL,
1440 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1441 "can't have 2 mark actions in same"
1444 return rte_flow_error_set(error, ENOTSUP,
1445 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1446 "mark action not supported for "
1452 * Validate the drop action.
1454 * @param[in] action_flags
1455 * Bit-fields that holds the actions detected until now.
1457 * Attributes of flow that includes this action.
1459 * Pointer to error structure.
1462 * 0 on success, a negative errno value otherwise and rte_errno is set.
1465 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1466 const struct rte_flow_attr *attr,
1467 struct rte_flow_error *error)
1470 return rte_flow_error_set(error, ENOTSUP,
1471 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1472 "drop action not supported for "
1478 * Validate the queue action.
1481 * Pointer to the queue action.
1482 * @param[in] action_flags
1483 * Bit-fields that holds the actions detected until now.
1485 * Pointer to the Ethernet device structure.
1487 * Attributes of flow that includes this action.
1489 * Pointer to error structure.
1492 * 0 on success, a negative errno value otherwise and rte_errno is set.
1495 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1496 uint64_t action_flags,
1497 struct rte_eth_dev *dev,
1498 const struct rte_flow_attr *attr,
1499 struct rte_flow_error *error)
1501 struct mlx5_priv *priv = dev->data->dev_private;
1502 const struct rte_flow_action_queue *queue = action->conf;
1504 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1505 return rte_flow_error_set(error, EINVAL,
1506 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1507 "can't have 2 fate actions in"
1510 return rte_flow_error_set(error, EINVAL,
1511 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1512 NULL, "No Rx queues configured");
1513 if (queue->index >= priv->rxqs_n)
1514 return rte_flow_error_set(error, EINVAL,
1515 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1517 "queue index out of range");
1518 if (!(*priv->rxqs)[queue->index])
1519 return rte_flow_error_set(error, EINVAL,
1520 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1522 "queue is not configured");
1524 return rte_flow_error_set(error, ENOTSUP,
1525 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1526 "queue action not supported for "
1532 * Validate the rss action.
1535 * Pointer to the Ethernet device structure.
1537 * Pointer to the queue action.
1539 * Pointer to error structure.
1542 * 0 on success, a negative errno value otherwise and rte_errno is set.
1545 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1546 const struct rte_flow_action *action,
1547 struct rte_flow_error *error)
1549 struct mlx5_priv *priv = dev->data->dev_private;
1550 const struct rte_flow_action_rss *rss = action->conf;
1553 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1554 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1555 return rte_flow_error_set(error, ENOTSUP,
1556 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1558 "RSS hash function not supported");
1559 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1564 return rte_flow_error_set(error, ENOTSUP,
1565 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1567 "tunnel RSS is not supported");
1568 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1569 if (rss->key_len == 0 && rss->key != NULL)
1570 return rte_flow_error_set(error, ENOTSUP,
1571 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1573 "RSS hash key length 0");
1574 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1575 return rte_flow_error_set(error, ENOTSUP,
1576 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1578 "RSS hash key too small");
1579 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1580 return rte_flow_error_set(error, ENOTSUP,
1581 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1583 "RSS hash key too large");
1584 if (rss->queue_num > priv->config.ind_table_max_size)
1585 return rte_flow_error_set(error, ENOTSUP,
1586 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1588 "number of queues too large");
1589 if (rss->types & MLX5_RSS_HF_MASK)
1590 return rte_flow_error_set(error, ENOTSUP,
1591 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1593 "some RSS protocols are not"
1595 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1596 !(rss->types & ETH_RSS_IP))
1597 return rte_flow_error_set(error, EINVAL,
1598 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1599 "L3 partial RSS requested but L3 RSS"
1600 " type not specified");
1601 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1602 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1603 return rte_flow_error_set(error, EINVAL,
1604 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1605 "L4 partial RSS requested but L4 RSS"
1606 " type not specified");
1608 return rte_flow_error_set(error, EINVAL,
1609 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1610 NULL, "No Rx queues configured");
1611 if (!rss->queue_num)
1612 return rte_flow_error_set(error, EINVAL,
1613 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1614 NULL, "No queues configured");
1615 for (i = 0; i != rss->queue_num; ++i) {
1616 if (rss->queue[i] >= priv->rxqs_n)
1617 return rte_flow_error_set
1619 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1620 &rss->queue[i], "queue index out of range");
1621 if (!(*priv->rxqs)[rss->queue[i]])
1622 return rte_flow_error_set
1623 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1624 &rss->queue[i], "queue is not configured");
1630 * Validate the rss action.
1633 * Pointer to the queue action.
1634 * @param[in] action_flags
1635 * Bit-fields that holds the actions detected until now.
1637 * Pointer to the Ethernet device structure.
1639 * Attributes of flow that includes this action.
1640 * @param[in] item_flags
1641 * Items that were detected.
1643 * Pointer to error structure.
1646 * 0 on success, a negative errno value otherwise and rte_errno is set.
1649 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1650 uint64_t action_flags,
1651 struct rte_eth_dev *dev,
1652 const struct rte_flow_attr *attr,
1653 uint64_t item_flags,
1654 struct rte_flow_error *error)
1656 const struct rte_flow_action_rss *rss = action->conf;
1657 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1660 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1661 return rte_flow_error_set(error, EINVAL,
1662 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1663 "can't have 2 fate actions"
1665 ret = mlx5_validate_action_rss(dev, action, error);
1669 return rte_flow_error_set(error, ENOTSUP,
1670 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1671 "rss action not supported for "
1673 if (rss->level > 1 && !tunnel)
1674 return rte_flow_error_set(error, EINVAL,
1675 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1676 "inner RSS is not supported for "
1677 "non-tunnel flows");
1678 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1679 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1680 return rte_flow_error_set(error, EINVAL,
1681 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1682 "RSS on eCPRI is not supported now");
1688 * Validate the default miss action.
1690 * @param[in] action_flags
1691 * Bit-fields that holds the actions detected until now.
1693 * Pointer to error structure.
1696 * 0 on success, a negative errno value otherwise and rte_errno is set.
1699 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1700 const struct rte_flow_attr *attr,
1701 struct rte_flow_error *error)
1703 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1704 return rte_flow_error_set(error, EINVAL,
1705 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1706 "can't have 2 fate actions in"
1709 return rte_flow_error_set(error, ENOTSUP,
1710 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1711 "default miss action not supported "
1714 return rte_flow_error_set(error, ENOTSUP,
1715 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1716 "only group 0 is supported");
1718 return rte_flow_error_set(error, ENOTSUP,
1719 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1720 NULL, "transfer is not supported");
1725 * Validate the count action.
1728 * Pointer to the Ethernet device structure.
1730 * Attributes of flow that includes this action.
1732 * Pointer to error structure.
1735 * 0 on success, a negative errno value otherwise and rte_errno is set.
1738 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1739 const struct rte_flow_attr *attr,
1740 struct rte_flow_error *error)
1743 return rte_flow_error_set(error, ENOTSUP,
1744 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1745 "count action not supported for "
1751 * Verify the @p attributes will be correctly understood by the NIC and store
1752 * them in the @p flow if everything is correct.
1755 * Pointer to the Ethernet device structure.
1756 * @param[in] attributes
1757 * Pointer to flow attributes
1759 * Pointer to error structure.
1762 * 0 on success, a negative errno value otherwise and rte_errno is set.
1765 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1766 const struct rte_flow_attr *attributes,
1767 struct rte_flow_error *error)
1769 struct mlx5_priv *priv = dev->data->dev_private;
1770 uint32_t priority_max = priv->config.flow_prio - 1;
1772 if (attributes->group)
1773 return rte_flow_error_set(error, ENOTSUP,
1774 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1775 NULL, "groups is not supported");
1776 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1777 attributes->priority >= priority_max)
1778 return rte_flow_error_set(error, ENOTSUP,
1779 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1780 NULL, "priority out of range");
1781 if (attributes->egress)
1782 return rte_flow_error_set(error, ENOTSUP,
1783 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1784 "egress is not supported");
1785 if (attributes->transfer && !priv->config.dv_esw_en)
1786 return rte_flow_error_set(error, ENOTSUP,
1787 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1788 NULL, "transfer is not supported");
1789 if (!attributes->ingress)
1790 return rte_flow_error_set(error, EINVAL,
1791 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1793 "ingress attribute is mandatory");
1798 * Validate ICMP6 item.
1801 * Item specification.
1802 * @param[in] item_flags
1803 * Bit-fields that holds the items detected until now.
1804 * @param[in] ext_vlan_sup
1805 * Whether extended VLAN features are supported or not.
1807 * Pointer to error structure.
1810 * 0 on success, a negative errno value otherwise and rte_errno is set.
1813 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1814 uint64_t item_flags,
1815 uint8_t target_protocol,
1816 struct rte_flow_error *error)
1818 const struct rte_flow_item_icmp6 *mask = item->mask;
1819 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1820 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1821 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1822 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1823 MLX5_FLOW_LAYER_OUTER_L4;
1826 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1827 return rte_flow_error_set(error, EINVAL,
1828 RTE_FLOW_ERROR_TYPE_ITEM, item,
1829 "protocol filtering not compatible"
1830 " with ICMP6 layer");
1831 if (!(item_flags & l3m))
1832 return rte_flow_error_set(error, EINVAL,
1833 RTE_FLOW_ERROR_TYPE_ITEM, item,
1834 "IPv6 is mandatory to filter on"
1836 if (item_flags & l4m)
1837 return rte_flow_error_set(error, EINVAL,
1838 RTE_FLOW_ERROR_TYPE_ITEM, item,
1839 "multiple L4 layers not supported");
1841 mask = &rte_flow_item_icmp6_mask;
1842 ret = mlx5_flow_item_acceptable
1843 (item, (const uint8_t *)mask,
1844 (const uint8_t *)&rte_flow_item_icmp6_mask,
1845 sizeof(struct rte_flow_item_icmp6),
1846 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1853 * Validate ICMP item.
1856 * Item specification.
1857 * @param[in] item_flags
1858 * Bit-fields that holds the items detected until now.
1860 * Pointer to error structure.
1863 * 0 on success, a negative errno value otherwise and rte_errno is set.
1866 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1867 uint64_t item_flags,
1868 uint8_t target_protocol,
1869 struct rte_flow_error *error)
1871 const struct rte_flow_item_icmp *mask = item->mask;
1872 const struct rte_flow_item_icmp nic_mask = {
1873 .hdr.icmp_type = 0xff,
1874 .hdr.icmp_code = 0xff,
1875 .hdr.icmp_ident = RTE_BE16(0xffff),
1876 .hdr.icmp_seq_nb = RTE_BE16(0xffff),
1878 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1879 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1880 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1881 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1882 MLX5_FLOW_LAYER_OUTER_L4;
1885 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1886 return rte_flow_error_set(error, EINVAL,
1887 RTE_FLOW_ERROR_TYPE_ITEM, item,
1888 "protocol filtering not compatible"
1889 " with ICMP layer");
1890 if (!(item_flags & l3m))
1891 return rte_flow_error_set(error, EINVAL,
1892 RTE_FLOW_ERROR_TYPE_ITEM, item,
1893 "IPv4 is mandatory to filter"
1895 if (item_flags & l4m)
1896 return rte_flow_error_set(error, EINVAL,
1897 RTE_FLOW_ERROR_TYPE_ITEM, item,
1898 "multiple L4 layers not supported");
1901 ret = mlx5_flow_item_acceptable
1902 (item, (const uint8_t *)mask,
1903 (const uint8_t *)&nic_mask,
1904 sizeof(struct rte_flow_item_icmp),
1905 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1912 * Validate Ethernet item.
1915 * Item specification.
1916 * @param[in] item_flags
1917 * Bit-fields that holds the items detected until now.
1919 * Pointer to error structure.
1922 * 0 on success, a negative errno value otherwise and rte_errno is set.
1925 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1926 uint64_t item_flags, bool ext_vlan_sup,
1927 struct rte_flow_error *error)
1929 const struct rte_flow_item_eth *mask = item->mask;
1930 const struct rte_flow_item_eth nic_mask = {
1931 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1932 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1933 .type = RTE_BE16(0xffff),
1934 .has_vlan = ext_vlan_sup ? 1 : 0,
1937 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1938 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1939 MLX5_FLOW_LAYER_OUTER_L2;
1941 if (item_flags & ethm)
1942 return rte_flow_error_set(error, ENOTSUP,
1943 RTE_FLOW_ERROR_TYPE_ITEM, item,
1944 "multiple L2 layers not supported");
1945 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1946 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1947 return rte_flow_error_set(error, EINVAL,
1948 RTE_FLOW_ERROR_TYPE_ITEM, item,
1949 "L2 layer should not follow "
1951 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1952 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1953 return rte_flow_error_set(error, EINVAL,
1954 RTE_FLOW_ERROR_TYPE_ITEM, item,
1955 "L2 layer should not follow VLAN");
1957 mask = &rte_flow_item_eth_mask;
1958 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1959 (const uint8_t *)&nic_mask,
1960 sizeof(struct rte_flow_item_eth),
1961 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1966 * Validate VLAN item.
1969 * Item specification.
1970 * @param[in] item_flags
1971 * Bit-fields that holds the items detected until now.
1973 * Ethernet device flow is being created on.
1975 * Pointer to error structure.
1978 * 0 on success, a negative errno value otherwise and rte_errno is set.
1981 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1982 uint64_t item_flags,
1983 struct rte_eth_dev *dev,
1984 struct rte_flow_error *error)
1986 const struct rte_flow_item_vlan *spec = item->spec;
1987 const struct rte_flow_item_vlan *mask = item->mask;
1988 const struct rte_flow_item_vlan nic_mask = {
1989 .tci = RTE_BE16(UINT16_MAX),
1990 .inner_type = RTE_BE16(UINT16_MAX),
1992 uint16_t vlan_tag = 0;
1993 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1995 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1996 MLX5_FLOW_LAYER_INNER_L4) :
1997 (MLX5_FLOW_LAYER_OUTER_L3 |
1998 MLX5_FLOW_LAYER_OUTER_L4);
1999 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2000 MLX5_FLOW_LAYER_OUTER_VLAN;
2002 if (item_flags & vlanm)
2003 return rte_flow_error_set(error, EINVAL,
2004 RTE_FLOW_ERROR_TYPE_ITEM, item,
2005 "multiple VLAN layers not supported");
2006 else if ((item_flags & l34m) != 0)
2007 return rte_flow_error_set(error, EINVAL,
2008 RTE_FLOW_ERROR_TYPE_ITEM, item,
2009 "VLAN cannot follow L3/L4 layer");
2011 mask = &rte_flow_item_vlan_mask;
2012 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2013 (const uint8_t *)&nic_mask,
2014 sizeof(struct rte_flow_item_vlan),
2015 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2018 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2019 struct mlx5_priv *priv = dev->data->dev_private;
2021 if (priv->vmwa_context) {
2023 * Non-NULL context means we have a virtual machine
2024 * and SR-IOV enabled, we have to create VLAN interface
2025 * to make hypervisor to setup E-Switch vport
2026 * context correctly. We avoid creating the multiple
2027 * VLAN interfaces, so we cannot support VLAN tag mask.
2029 return rte_flow_error_set(error, EINVAL,
2030 RTE_FLOW_ERROR_TYPE_ITEM,
2032 "VLAN tag mask is not"
2033 " supported in virtual"
2038 vlan_tag = spec->tci;
2039 vlan_tag &= mask->tci;
2042 * From verbs perspective an empty VLAN is equivalent
2043 * to a packet without VLAN layer.
2046 return rte_flow_error_set(error, EINVAL,
2047 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2049 "VLAN cannot be empty");
2054 * Validate IPV4 item.
2057 * Item specification.
2058 * @param[in] item_flags
2059 * Bit-fields that holds the items detected until now.
2060 * @param[in] last_item
2061 * Previous validated item in the pattern items.
2062 * @param[in] ether_type
2063 * Type in the ethernet layer header (including dot1q).
2064 * @param[in] acc_mask
2065 * Acceptable mask, if NULL default internal default mask
2066 * will be used to check whether item fields are supported.
2067 * @param[in] range_accepted
2068 * True if range of values is accepted for specific fields, false otherwise.
2070 * Pointer to error structure.
2073 * 0 on success, a negative errno value otherwise and rte_errno is set.
2076 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2077 uint64_t item_flags,
2079 uint16_t ether_type,
2080 const struct rte_flow_item_ipv4 *acc_mask,
2081 bool range_accepted,
2082 struct rte_flow_error *error)
2084 const struct rte_flow_item_ipv4 *mask = item->mask;
2085 const struct rte_flow_item_ipv4 *spec = item->spec;
2086 const struct rte_flow_item_ipv4 nic_mask = {
2088 .src_addr = RTE_BE32(0xffffffff),
2089 .dst_addr = RTE_BE32(0xffffffff),
2090 .type_of_service = 0xff,
2091 .next_proto_id = 0xff,
2094 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2095 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2096 MLX5_FLOW_LAYER_OUTER_L3;
2097 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2098 MLX5_FLOW_LAYER_OUTER_L4;
2100 uint8_t next_proto = 0xFF;
2101 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2102 MLX5_FLOW_LAYER_OUTER_VLAN |
2103 MLX5_FLOW_LAYER_INNER_VLAN);
2105 if ((last_item & l2_vlan) && ether_type &&
2106 ether_type != RTE_ETHER_TYPE_IPV4)
2107 return rte_flow_error_set(error, EINVAL,
2108 RTE_FLOW_ERROR_TYPE_ITEM, item,
2109 "IPv4 cannot follow L2/VLAN layer "
2110 "which ether type is not IPv4");
2111 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2113 next_proto = mask->hdr.next_proto_id &
2114 spec->hdr.next_proto_id;
2115 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2116 return rte_flow_error_set(error, EINVAL,
2117 RTE_FLOW_ERROR_TYPE_ITEM,
2122 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2123 return rte_flow_error_set(error, EINVAL,
2124 RTE_FLOW_ERROR_TYPE_ITEM, item,
2125 "wrong tunnel type - IPv6 specified "
2126 "but IPv4 item provided");
2127 if (item_flags & l3m)
2128 return rte_flow_error_set(error, ENOTSUP,
2129 RTE_FLOW_ERROR_TYPE_ITEM, item,
2130 "multiple L3 layers not supported");
2131 else if (item_flags & l4m)
2132 return rte_flow_error_set(error, EINVAL,
2133 RTE_FLOW_ERROR_TYPE_ITEM, item,
2134 "L3 cannot follow an L4 layer.");
2135 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2136 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2137 return rte_flow_error_set(error, EINVAL,
2138 RTE_FLOW_ERROR_TYPE_ITEM, item,
2139 "L3 cannot follow an NVGRE layer.");
2141 mask = &rte_flow_item_ipv4_mask;
2142 else if (mask->hdr.next_proto_id != 0 &&
2143 mask->hdr.next_proto_id != 0xff)
2144 return rte_flow_error_set(error, EINVAL,
2145 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2146 "partial mask is not supported"
2148 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2149 acc_mask ? (const uint8_t *)acc_mask
2150 : (const uint8_t *)&nic_mask,
2151 sizeof(struct rte_flow_item_ipv4),
2152 range_accepted, error);
2159 * Validate IPV6 item.
2162 * Item specification.
2163 * @param[in] item_flags
2164 * Bit-fields that holds the items detected until now.
2165 * @param[in] last_item
2166 * Previous validated item in the pattern items.
2167 * @param[in] ether_type
2168 * Type in the ethernet layer header (including dot1q).
2169 * @param[in] acc_mask
2170 * Acceptable mask, if NULL default internal default mask
2171 * will be used to check whether item fields are supported.
2173 * Pointer to error structure.
2176 * 0 on success, a negative errno value otherwise and rte_errno is set.
2179 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2180 uint64_t item_flags,
2182 uint16_t ether_type,
2183 const struct rte_flow_item_ipv6 *acc_mask,
2184 struct rte_flow_error *error)
2186 const struct rte_flow_item_ipv6 *mask = item->mask;
2187 const struct rte_flow_item_ipv6 *spec = item->spec;
2188 const struct rte_flow_item_ipv6 nic_mask = {
2191 "\xff\xff\xff\xff\xff\xff\xff\xff"
2192 "\xff\xff\xff\xff\xff\xff\xff\xff",
2194 "\xff\xff\xff\xff\xff\xff\xff\xff"
2195 "\xff\xff\xff\xff\xff\xff\xff\xff",
2196 .vtc_flow = RTE_BE32(0xffffffff),
2200 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2201 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2202 MLX5_FLOW_LAYER_OUTER_L3;
2203 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2204 MLX5_FLOW_LAYER_OUTER_L4;
2206 uint8_t next_proto = 0xFF;
2207 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2208 MLX5_FLOW_LAYER_OUTER_VLAN |
2209 MLX5_FLOW_LAYER_INNER_VLAN);
2211 if ((last_item & l2_vlan) && ether_type &&
2212 ether_type != RTE_ETHER_TYPE_IPV6)
2213 return rte_flow_error_set(error, EINVAL,
2214 RTE_FLOW_ERROR_TYPE_ITEM, item,
2215 "IPv6 cannot follow L2/VLAN layer "
2216 "which ether type is not IPv6");
2217 if (mask && mask->hdr.proto == UINT8_MAX && spec)
2218 next_proto = spec->hdr.proto;
2219 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2220 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2221 return rte_flow_error_set(error, EINVAL,
2222 RTE_FLOW_ERROR_TYPE_ITEM,
2227 if (next_proto == IPPROTO_HOPOPTS ||
2228 next_proto == IPPROTO_ROUTING ||
2229 next_proto == IPPROTO_FRAGMENT ||
2230 next_proto == IPPROTO_ESP ||
2231 next_proto == IPPROTO_AH ||
2232 next_proto == IPPROTO_DSTOPTS)
2233 return rte_flow_error_set(error, EINVAL,
2234 RTE_FLOW_ERROR_TYPE_ITEM, item,
2235 "IPv6 proto (next header) should "
2236 "not be set as extension header");
2237 if (item_flags & MLX5_FLOW_LAYER_IPIP)
2238 return rte_flow_error_set(error, EINVAL,
2239 RTE_FLOW_ERROR_TYPE_ITEM, item,
2240 "wrong tunnel type - IPv4 specified "
2241 "but IPv6 item provided");
2242 if (item_flags & l3m)
2243 return rte_flow_error_set(error, ENOTSUP,
2244 RTE_FLOW_ERROR_TYPE_ITEM, item,
2245 "multiple L3 layers not supported");
2246 else if (item_flags & l4m)
2247 return rte_flow_error_set(error, EINVAL,
2248 RTE_FLOW_ERROR_TYPE_ITEM, item,
2249 "L3 cannot follow an L4 layer.");
2250 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2251 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2252 return rte_flow_error_set(error, EINVAL,
2253 RTE_FLOW_ERROR_TYPE_ITEM, item,
2254 "L3 cannot follow an NVGRE layer.");
2256 mask = &rte_flow_item_ipv6_mask;
2257 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2258 acc_mask ? (const uint8_t *)acc_mask
2259 : (const uint8_t *)&nic_mask,
2260 sizeof(struct rte_flow_item_ipv6),
2261 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2268 * Validate UDP item.
2271 * Item specification.
2272 * @param[in] item_flags
2273 * Bit-fields that holds the items detected until now.
2274 * @param[in] target_protocol
2275 * The next protocol in the previous item.
2276 * @param[in] flow_mask
2277 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2279 * Pointer to error structure.
2282 * 0 on success, a negative errno value otherwise and rte_errno is set.
2285 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2286 uint64_t item_flags,
2287 uint8_t target_protocol,
2288 struct rte_flow_error *error)
2290 const struct rte_flow_item_udp *mask = item->mask;
2291 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2292 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2293 MLX5_FLOW_LAYER_OUTER_L3;
2294 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2295 MLX5_FLOW_LAYER_OUTER_L4;
2298 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2299 return rte_flow_error_set(error, EINVAL,
2300 RTE_FLOW_ERROR_TYPE_ITEM, item,
2301 "protocol filtering not compatible"
2303 if (!(item_flags & l3m))
2304 return rte_flow_error_set(error, EINVAL,
2305 RTE_FLOW_ERROR_TYPE_ITEM, item,
2306 "L3 is mandatory to filter on L4");
2307 if (item_flags & l4m)
2308 return rte_flow_error_set(error, EINVAL,
2309 RTE_FLOW_ERROR_TYPE_ITEM, item,
2310 "multiple L4 layers not supported");
2312 mask = &rte_flow_item_udp_mask;
2313 ret = mlx5_flow_item_acceptable
2314 (item, (const uint8_t *)mask,
2315 (const uint8_t *)&rte_flow_item_udp_mask,
2316 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2324 * Validate TCP item.
2327 * Item specification.
2328 * @param[in] item_flags
2329 * Bit-fields that holds the items detected until now.
2330 * @param[in] target_protocol
2331 * The next protocol in the previous item.
2333 * Pointer to error structure.
2336 * 0 on success, a negative errno value otherwise and rte_errno is set.
2339 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2340 uint64_t item_flags,
2341 uint8_t target_protocol,
2342 const struct rte_flow_item_tcp *flow_mask,
2343 struct rte_flow_error *error)
2345 const struct rte_flow_item_tcp *mask = item->mask;
2346 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2347 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2348 MLX5_FLOW_LAYER_OUTER_L3;
2349 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2350 MLX5_FLOW_LAYER_OUTER_L4;
2353 MLX5_ASSERT(flow_mask);
2354 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2355 return rte_flow_error_set(error, EINVAL,
2356 RTE_FLOW_ERROR_TYPE_ITEM, item,
2357 "protocol filtering not compatible"
2359 if (!(item_flags & l3m))
2360 return rte_flow_error_set(error, EINVAL,
2361 RTE_FLOW_ERROR_TYPE_ITEM, item,
2362 "L3 is mandatory to filter on L4");
2363 if (item_flags & l4m)
2364 return rte_flow_error_set(error, EINVAL,
2365 RTE_FLOW_ERROR_TYPE_ITEM, item,
2366 "multiple L4 layers not supported");
2368 mask = &rte_flow_item_tcp_mask;
2369 ret = mlx5_flow_item_acceptable
2370 (item, (const uint8_t *)mask,
2371 (const uint8_t *)flow_mask,
2372 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2380 * Validate VXLAN item.
2383 * Item specification.
2384 * @param[in] item_flags
2385 * Bit-fields that holds the items detected until now.
2386 * @param[in] target_protocol
2387 * The next protocol in the previous item.
2389 * Pointer to error structure.
2392 * 0 on success, a negative errno value otherwise and rte_errno is set.
2395 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2396 uint64_t item_flags,
2397 struct rte_flow_error *error)
2399 const struct rte_flow_item_vxlan *spec = item->spec;
2400 const struct rte_flow_item_vxlan *mask = item->mask;
2405 } id = { .vlan_id = 0, };
2408 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2409 return rte_flow_error_set(error, ENOTSUP,
2410 RTE_FLOW_ERROR_TYPE_ITEM, item,
2411 "multiple tunnel layers not"
2414 * Verify only UDPv4 is present as defined in
2415 * https://tools.ietf.org/html/rfc7348
2417 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2418 return rte_flow_error_set(error, EINVAL,
2419 RTE_FLOW_ERROR_TYPE_ITEM, item,
2420 "no outer UDP layer found");
2422 mask = &rte_flow_item_vxlan_mask;
2423 ret = mlx5_flow_item_acceptable
2424 (item, (const uint8_t *)mask,
2425 (const uint8_t *)&rte_flow_item_vxlan_mask,
2426 sizeof(struct rte_flow_item_vxlan),
2427 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2431 memcpy(&id.vni[1], spec->vni, 3);
2432 memcpy(&id.vni[1], mask->vni, 3);
2434 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2435 return rte_flow_error_set(error, ENOTSUP,
2436 RTE_FLOW_ERROR_TYPE_ITEM, item,
2437 "VXLAN tunnel must be fully defined");
2442 * Validate VXLAN_GPE item.
2445 * Item specification.
2446 * @param[in] item_flags
2447 * Bit-fields that holds the items detected until now.
2449 * Pointer to the private data structure.
2450 * @param[in] target_protocol
2451 * The next protocol in the previous item.
2453 * Pointer to error structure.
2456 * 0 on success, a negative errno value otherwise and rte_errno is set.
2459 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2460 uint64_t item_flags,
2461 struct rte_eth_dev *dev,
2462 struct rte_flow_error *error)
2464 struct mlx5_priv *priv = dev->data->dev_private;
2465 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2466 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2471 } id = { .vlan_id = 0, };
2473 if (!priv->config.l3_vxlan_en)
2474 return rte_flow_error_set(error, ENOTSUP,
2475 RTE_FLOW_ERROR_TYPE_ITEM, item,
2476 "L3 VXLAN is not enabled by device"
2477 " parameter and/or not configured in"
2479 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2480 return rte_flow_error_set(error, ENOTSUP,
2481 RTE_FLOW_ERROR_TYPE_ITEM, item,
2482 "multiple tunnel layers not"
2485 * Verify only UDPv4 is present as defined in
2486 * https://tools.ietf.org/html/rfc7348
2488 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2489 return rte_flow_error_set(error, EINVAL,
2490 RTE_FLOW_ERROR_TYPE_ITEM, item,
2491 "no outer UDP layer found");
2493 mask = &rte_flow_item_vxlan_gpe_mask;
2494 ret = mlx5_flow_item_acceptable
2495 (item, (const uint8_t *)mask,
2496 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2497 sizeof(struct rte_flow_item_vxlan_gpe),
2498 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2503 return rte_flow_error_set(error, ENOTSUP,
2504 RTE_FLOW_ERROR_TYPE_ITEM,
2506 "VxLAN-GPE protocol"
2508 memcpy(&id.vni[1], spec->vni, 3);
2509 memcpy(&id.vni[1], mask->vni, 3);
2511 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2512 return rte_flow_error_set(error, ENOTSUP,
2513 RTE_FLOW_ERROR_TYPE_ITEM, item,
2514 "VXLAN-GPE tunnel must be fully"
2519 * Validate GRE Key item.
2522 * Item specification.
2523 * @param[in] item_flags
2524 * Bit flags to mark detected items.
2525 * @param[in] gre_item
2526 * Pointer to gre_item
2528 * Pointer to error structure.
2531 * 0 on success, a negative errno value otherwise and rte_errno is set.
2534 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2535 uint64_t item_flags,
2536 const struct rte_flow_item *gre_item,
2537 struct rte_flow_error *error)
2539 const rte_be32_t *mask = item->mask;
2541 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2542 const struct rte_flow_item_gre *gre_spec;
2543 const struct rte_flow_item_gre *gre_mask;
2545 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2546 return rte_flow_error_set(error, ENOTSUP,
2547 RTE_FLOW_ERROR_TYPE_ITEM, item,
2548 "Multiple GRE key not support");
2549 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2550 return rte_flow_error_set(error, ENOTSUP,
2551 RTE_FLOW_ERROR_TYPE_ITEM, item,
2552 "No preceding GRE header");
2553 if (item_flags & MLX5_FLOW_LAYER_INNER)
2554 return rte_flow_error_set(error, ENOTSUP,
2555 RTE_FLOW_ERROR_TYPE_ITEM, item,
2556 "GRE key following a wrong item");
2557 gre_mask = gre_item->mask;
2559 gre_mask = &rte_flow_item_gre_mask;
2560 gre_spec = gre_item->spec;
2561 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2562 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2563 return rte_flow_error_set(error, EINVAL,
2564 RTE_FLOW_ERROR_TYPE_ITEM, item,
2565 "Key bit must be on");
2568 mask = &gre_key_default_mask;
2569 ret = mlx5_flow_item_acceptable
2570 (item, (const uint8_t *)mask,
2571 (const uint8_t *)&gre_key_default_mask,
2572 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2577 * Validate GRE item.
2580 * Item specification.
2581 * @param[in] item_flags
2582 * Bit flags to mark detected items.
2583 * @param[in] target_protocol
2584 * The next protocol in the previous item.
2586 * Pointer to error structure.
2589 * 0 on success, a negative errno value otherwise and rte_errno is set.
2592 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2593 uint64_t item_flags,
2594 uint8_t target_protocol,
2595 struct rte_flow_error *error)
2597 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2598 const struct rte_flow_item_gre *mask = item->mask;
2600 const struct rte_flow_item_gre nic_mask = {
2601 .c_rsvd0_ver = RTE_BE16(0xB000),
2602 .protocol = RTE_BE16(UINT16_MAX),
2605 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2606 return rte_flow_error_set(error, EINVAL,
2607 RTE_FLOW_ERROR_TYPE_ITEM, item,
2608 "protocol filtering not compatible"
2609 " with this GRE layer");
2610 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2611 return rte_flow_error_set(error, ENOTSUP,
2612 RTE_FLOW_ERROR_TYPE_ITEM, item,
2613 "multiple tunnel layers not"
2615 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2616 return rte_flow_error_set(error, ENOTSUP,
2617 RTE_FLOW_ERROR_TYPE_ITEM, item,
2618 "L3 Layer is missing");
2620 mask = &rte_flow_item_gre_mask;
2621 ret = mlx5_flow_item_acceptable
2622 (item, (const uint8_t *)mask,
2623 (const uint8_t *)&nic_mask,
2624 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2628 #ifndef HAVE_MLX5DV_DR
2629 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2630 if (spec && (spec->protocol & mask->protocol))
2631 return rte_flow_error_set(error, ENOTSUP,
2632 RTE_FLOW_ERROR_TYPE_ITEM, item,
2633 "without MPLS support the"
2634 " specification cannot be used for"
2642 * Validate Geneve item.
2645 * Item specification.
2646 * @param[in] itemFlags
2647 * Bit-fields that holds the items detected until now.
2649 * Pointer to the private data structure.
2651 * Pointer to error structure.
2654 * 0 on success, a negative errno value otherwise and rte_errno is set.
2658 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2659 uint64_t item_flags,
2660 struct rte_eth_dev *dev,
2661 struct rte_flow_error *error)
2663 struct mlx5_priv *priv = dev->data->dev_private;
2664 const struct rte_flow_item_geneve *spec = item->spec;
2665 const struct rte_flow_item_geneve *mask = item->mask;
2668 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2669 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2670 const struct rte_flow_item_geneve nic_mask = {
2671 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2672 .vni = "\xff\xff\xff",
2673 .protocol = RTE_BE16(UINT16_MAX),
2676 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2677 return rte_flow_error_set(error, ENOTSUP,
2678 RTE_FLOW_ERROR_TYPE_ITEM, item,
2679 "L3 Geneve is not enabled by device"
2680 " parameter and/or not configured in"
2682 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2683 return rte_flow_error_set(error, ENOTSUP,
2684 RTE_FLOW_ERROR_TYPE_ITEM, item,
2685 "multiple tunnel layers not"
2688 * Verify only UDPv4 is present as defined in
2689 * https://tools.ietf.org/html/rfc7348
2691 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2692 return rte_flow_error_set(error, EINVAL,
2693 RTE_FLOW_ERROR_TYPE_ITEM, item,
2694 "no outer UDP layer found");
2696 mask = &rte_flow_item_geneve_mask;
2697 ret = mlx5_flow_item_acceptable
2698 (item, (const uint8_t *)mask,
2699 (const uint8_t *)&nic_mask,
2700 sizeof(struct rte_flow_item_geneve),
2701 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2705 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2706 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2707 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2708 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2709 return rte_flow_error_set(error, ENOTSUP,
2710 RTE_FLOW_ERROR_TYPE_ITEM,
2712 "Geneve protocol unsupported"
2713 " fields are being used");
2714 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2715 return rte_flow_error_set
2717 RTE_FLOW_ERROR_TYPE_ITEM,
2719 "Unsupported Geneve options length");
2721 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2722 return rte_flow_error_set
2724 RTE_FLOW_ERROR_TYPE_ITEM, item,
2725 "Geneve tunnel must be fully defined");
2730 * Validate MPLS item.
2733 * Pointer to the rte_eth_dev structure.
2735 * Item specification.
2736 * @param[in] item_flags
2737 * Bit-fields that holds the items detected until now.
2738 * @param[in] prev_layer
2739 * The protocol layer indicated in previous item.
2741 * Pointer to error structure.
2744 * 0 on success, a negative errno value otherwise and rte_errno is set.
2747 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2748 const struct rte_flow_item *item __rte_unused,
2749 uint64_t item_flags __rte_unused,
2750 uint64_t prev_layer __rte_unused,
2751 struct rte_flow_error *error)
2753 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2754 const struct rte_flow_item_mpls *mask = item->mask;
2755 struct mlx5_priv *priv = dev->data->dev_private;
2758 if (!priv->config.mpls_en)
2759 return rte_flow_error_set(error, ENOTSUP,
2760 RTE_FLOW_ERROR_TYPE_ITEM, item,
2761 "MPLS not supported or"
2762 " disabled in firmware"
2764 /* MPLS over IP, UDP, GRE is allowed */
2765 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2766 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2767 MLX5_FLOW_LAYER_GRE)))
2768 return rte_flow_error_set(error, EINVAL,
2769 RTE_FLOW_ERROR_TYPE_ITEM, item,
2770 "protocol filtering not compatible"
2771 " with MPLS layer");
2772 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2773 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2774 !(item_flags & MLX5_FLOW_LAYER_GRE))
2775 return rte_flow_error_set(error, ENOTSUP,
2776 RTE_FLOW_ERROR_TYPE_ITEM, item,
2777 "multiple tunnel layers not"
2780 mask = &rte_flow_item_mpls_mask;
2781 ret = mlx5_flow_item_acceptable
2782 (item, (const uint8_t *)mask,
2783 (const uint8_t *)&rte_flow_item_mpls_mask,
2784 sizeof(struct rte_flow_item_mpls),
2785 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2790 return rte_flow_error_set(error, ENOTSUP,
2791 RTE_FLOW_ERROR_TYPE_ITEM, item,
2792 "MPLS is not supported by Verbs, please"
2798 * Validate NVGRE item.
2801 * Item specification.
2802 * @param[in] item_flags
2803 * Bit flags to mark detected items.
2804 * @param[in] target_protocol
2805 * The next protocol in the previous item.
2807 * Pointer to error structure.
2810 * 0 on success, a negative errno value otherwise and rte_errno is set.
2813 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2814 uint64_t item_flags,
2815 uint8_t target_protocol,
2816 struct rte_flow_error *error)
2818 const struct rte_flow_item_nvgre *mask = item->mask;
2821 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2822 return rte_flow_error_set(error, EINVAL,
2823 RTE_FLOW_ERROR_TYPE_ITEM, item,
2824 "protocol filtering not compatible"
2825 " with this GRE layer");
2826 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2827 return rte_flow_error_set(error, ENOTSUP,
2828 RTE_FLOW_ERROR_TYPE_ITEM, item,
2829 "multiple tunnel layers not"
2831 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2832 return rte_flow_error_set(error, ENOTSUP,
2833 RTE_FLOW_ERROR_TYPE_ITEM, item,
2834 "L3 Layer is missing");
2836 mask = &rte_flow_item_nvgre_mask;
2837 ret = mlx5_flow_item_acceptable
2838 (item, (const uint8_t *)mask,
2839 (const uint8_t *)&rte_flow_item_nvgre_mask,
2840 sizeof(struct rte_flow_item_nvgre),
2841 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2848 * Validate eCPRI item.
2851 * Item specification.
2852 * @param[in] item_flags
2853 * Bit-fields that holds the items detected until now.
2854 * @param[in] last_item
2855 * Previous validated item in the pattern items.
2856 * @param[in] ether_type
2857 * Type in the ethernet layer header (including dot1q).
2858 * @param[in] acc_mask
2859 * Acceptable mask, if NULL default internal default mask
2860 * will be used to check whether item fields are supported.
2862 * Pointer to error structure.
2865 * 0 on success, a negative errno value otherwise and rte_errno is set.
2868 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2869 uint64_t item_flags,
2871 uint16_t ether_type,
2872 const struct rte_flow_item_ecpri *acc_mask,
2873 struct rte_flow_error *error)
2875 const struct rte_flow_item_ecpri *mask = item->mask;
2876 const struct rte_flow_item_ecpri nic_mask = {
2880 RTE_BE32(((const struct rte_ecpri_common_hdr) {
2884 .dummy[0] = 0xFFFFFFFF,
2887 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2888 MLX5_FLOW_LAYER_OUTER_VLAN);
2889 struct rte_flow_item_ecpri mask_lo;
2891 if ((last_item & outer_l2_vlan) && ether_type &&
2892 ether_type != RTE_ETHER_TYPE_ECPRI)
2893 return rte_flow_error_set(error, EINVAL,
2894 RTE_FLOW_ERROR_TYPE_ITEM, item,
2895 "eCPRI cannot follow L2/VLAN layer "
2896 "which ether type is not 0xAEFE.");
2897 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2898 return rte_flow_error_set(error, EINVAL,
2899 RTE_FLOW_ERROR_TYPE_ITEM, item,
2900 "eCPRI with tunnel is not supported "
2902 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2903 return rte_flow_error_set(error, ENOTSUP,
2904 RTE_FLOW_ERROR_TYPE_ITEM, item,
2905 "multiple L3 layers not supported");
2906 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2907 return rte_flow_error_set(error, EINVAL,
2908 RTE_FLOW_ERROR_TYPE_ITEM, item,
2909 "eCPRI cannot follow a TCP layer.");
2910 /* In specification, eCPRI could be over UDP layer. */
2911 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2912 return rte_flow_error_set(error, EINVAL,
2913 RTE_FLOW_ERROR_TYPE_ITEM, item,
2914 "eCPRI over UDP layer is not yet "
2915 "supported right now.");
2916 /* Mask for type field in common header could be zero. */
2918 mask = &rte_flow_item_ecpri_mask;
2919 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2920 /* Input mask is in big-endian format. */
2921 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2922 return rte_flow_error_set(error, EINVAL,
2923 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2924 "partial mask is not supported "
2926 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2927 return rte_flow_error_set(error, EINVAL,
2928 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2929 "message header mask must be after "
2931 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2932 acc_mask ? (const uint8_t *)acc_mask
2933 : (const uint8_t *)&nic_mask,
2934 sizeof(struct rte_flow_item_ecpri),
2935 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2939 * Release resource related QUEUE/RSS action split.
2942 * Pointer to Ethernet device.
2944 * Flow to release id's from.
2947 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2948 struct rte_flow *flow)
2950 struct mlx5_priv *priv = dev->data->dev_private;
2951 uint32_t handle_idx;
2952 struct mlx5_flow_handle *dev_handle;
2954 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2955 handle_idx, dev_handle, next)
2956 if (dev_handle->split_flow_id)
2957 mlx5_ipool_free(priv->sh->ipool
2958 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2959 dev_handle->split_flow_id);
2963 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2964 const struct rte_flow_attr *attr __rte_unused,
2965 const struct rte_flow_item items[] __rte_unused,
2966 const struct rte_flow_action actions[] __rte_unused,
2967 bool external __rte_unused,
2968 int hairpin __rte_unused,
2969 struct rte_flow_error *error)
2971 return rte_flow_error_set(error, ENOTSUP,
2972 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2975 static struct mlx5_flow *
2976 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2977 const struct rte_flow_attr *attr __rte_unused,
2978 const struct rte_flow_item items[] __rte_unused,
2979 const struct rte_flow_action actions[] __rte_unused,
2980 struct rte_flow_error *error)
2982 rte_flow_error_set(error, ENOTSUP,
2983 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2988 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2989 struct mlx5_flow *dev_flow __rte_unused,
2990 const struct rte_flow_attr *attr __rte_unused,
2991 const struct rte_flow_item items[] __rte_unused,
2992 const struct rte_flow_action actions[] __rte_unused,
2993 struct rte_flow_error *error)
2995 return rte_flow_error_set(error, ENOTSUP,
2996 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3000 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3001 struct rte_flow *flow __rte_unused,
3002 struct rte_flow_error *error)
3004 return rte_flow_error_set(error, ENOTSUP,
3005 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3009 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3010 struct rte_flow *flow __rte_unused)
3015 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3016 struct rte_flow *flow __rte_unused)
3021 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3022 struct rte_flow *flow __rte_unused,
3023 const struct rte_flow_action *actions __rte_unused,
3024 void *data __rte_unused,
3025 struct rte_flow_error *error)
3027 return rte_flow_error_set(error, ENOTSUP,
3028 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3032 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3033 uint32_t domains __rte_unused,
3034 uint32_t flags __rte_unused)
3039 /* Void driver to protect from null pointer reference. */
3040 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3041 .validate = flow_null_validate,
3042 .prepare = flow_null_prepare,
3043 .translate = flow_null_translate,
3044 .apply = flow_null_apply,
3045 .remove = flow_null_remove,
3046 .destroy = flow_null_destroy,
3047 .query = flow_null_query,
3048 .sync_domain = flow_null_sync_domain,
3052 * Select flow driver type according to flow attributes and device
3056 * Pointer to the dev structure.
3058 * Pointer to the flow attributes.
3061 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3063 static enum mlx5_flow_drv_type
3064 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3066 struct mlx5_priv *priv = dev->data->dev_private;
3067 /* The OS can determine first a specific flow type (DV, VERBS) */
3068 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3070 if (type != MLX5_FLOW_TYPE_MAX)
3072 /* If no OS specific type - continue with DV/VERBS selection */
3073 if (attr->transfer && priv->config.dv_esw_en)
3074 type = MLX5_FLOW_TYPE_DV;
3075 if (!attr->transfer)
3076 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3077 MLX5_FLOW_TYPE_VERBS;
3081 #define flow_get_drv_ops(type) flow_drv_ops[type]
3084 * Flow driver validation API. This abstracts calling driver specific functions.
3085 * The type of flow driver is determined according to flow attributes.
3088 * Pointer to the dev structure.
3090 * Pointer to the flow attributes.
3092 * Pointer to the list of items.
3093 * @param[in] actions
3094 * Pointer to the list of actions.
3095 * @param[in] external
3096 * This flow rule is created by request external to PMD.
3097 * @param[in] hairpin
3098 * Number of hairpin TX actions, 0 means classic flow.
3100 * Pointer to the error structure.
3103 * 0 on success, a negative errno value otherwise and rte_errno is set.
3106 flow_drv_validate(struct rte_eth_dev *dev,
3107 const struct rte_flow_attr *attr,
3108 const struct rte_flow_item items[],
3109 const struct rte_flow_action actions[],
3110 bool external, int hairpin, struct rte_flow_error *error)
3112 const struct mlx5_flow_driver_ops *fops;
3113 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3115 fops = flow_get_drv_ops(type);
3116 return fops->validate(dev, attr, items, actions, external,
3121 * Flow driver preparation API. This abstracts calling driver specific
3122 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3123 * calculates the size of memory required for device flow, allocates the memory,
3124 * initializes the device flow and returns the pointer.
3127 * This function initializes device flow structure such as dv or verbs in
3128 * struct mlx5_flow. However, it is caller's responsibility to initialize the
3129 * rest. For example, adding returning device flow to flow->dev_flow list and
3130 * setting backward reference to the flow should be done out of this function.
3131 * layers field is not filled either.
3134 * Pointer to the dev structure.
3136 * Pointer to the flow attributes.
3138 * Pointer to the list of items.
3139 * @param[in] actions
3140 * Pointer to the list of actions.
3141 * @param[in] flow_idx
3142 * This memory pool index to the flow.
3144 * Pointer to the error structure.
3147 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
3149 static inline struct mlx5_flow *
3150 flow_drv_prepare(struct rte_eth_dev *dev,
3151 const struct rte_flow *flow,
3152 const struct rte_flow_attr *attr,
3153 const struct rte_flow_item items[],
3154 const struct rte_flow_action actions[],
3156 struct rte_flow_error *error)
3158 const struct mlx5_flow_driver_ops *fops;
3159 enum mlx5_flow_drv_type type = flow->drv_type;
3160 struct mlx5_flow *mlx5_flow = NULL;
3162 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3163 fops = flow_get_drv_ops(type);
3164 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3166 mlx5_flow->flow_idx = flow_idx;
3171 * Flow driver translation API. This abstracts calling driver specific
3172 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3173 * translates a generic flow into a driver flow. flow_drv_prepare() must
3177 * dev_flow->layers could be filled as a result of parsing during translation
3178 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3179 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3180 * flow->actions could be overwritten even though all the expanded dev_flows
3181 * have the same actions.
3184 * Pointer to the rte dev structure.
3185 * @param[in, out] dev_flow
3186 * Pointer to the mlx5 flow.
3188 * Pointer to the flow attributes.
3190 * Pointer to the list of items.
3191 * @param[in] actions
3192 * Pointer to the list of actions.
3194 * Pointer to the error structure.
3197 * 0 on success, a negative errno value otherwise and rte_errno is set.
3200 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3201 const struct rte_flow_attr *attr,
3202 const struct rte_flow_item items[],
3203 const struct rte_flow_action actions[],
3204 struct rte_flow_error *error)
3206 const struct mlx5_flow_driver_ops *fops;
3207 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3209 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3210 fops = flow_get_drv_ops(type);
3211 return fops->translate(dev, dev_flow, attr, items, actions, error);
3215 * Flow driver apply API. This abstracts calling driver specific functions.
3216 * Parent flow (rte_flow) should have driver type (drv_type). It applies
3217 * translated driver flows on to device. flow_drv_translate() must precede.
3220 * Pointer to Ethernet device structure.
3221 * @param[in, out] flow
3222 * Pointer to flow structure.
3224 * Pointer to error structure.
3227 * 0 on success, a negative errno value otherwise and rte_errno is set.
3230 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3231 struct rte_flow_error *error)
3233 const struct mlx5_flow_driver_ops *fops;
3234 enum mlx5_flow_drv_type type = flow->drv_type;
3236 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3237 fops = flow_get_drv_ops(type);
3238 return fops->apply(dev, flow, error);
3242 * Flow driver remove API. This abstracts calling driver specific functions.
3243 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3244 * on device. All the resources of the flow should be freed by calling
3245 * flow_drv_destroy().
3248 * Pointer to Ethernet device.
3249 * @param[in, out] flow
3250 * Pointer to flow structure.
3253 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
3255 const struct mlx5_flow_driver_ops *fops;
3256 enum mlx5_flow_drv_type type = flow->drv_type;
3258 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3259 fops = flow_get_drv_ops(type);
3260 fops->remove(dev, flow);
3264 * Flow driver destroy API. This abstracts calling driver specific functions.
3265 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3266 * on device and releases resources of the flow.
3269 * Pointer to Ethernet device.
3270 * @param[in, out] flow
3271 * Pointer to flow structure.
3274 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3276 const struct mlx5_flow_driver_ops *fops;
3277 enum mlx5_flow_drv_type type = flow->drv_type;
3279 flow_mreg_split_qrss_release(dev, flow);
3280 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3281 fops = flow_get_drv_ops(type);
3282 fops->destroy(dev, flow);
3286 * Get RSS action from the action list.
3288 * @param[in] actions
3289 * Pointer to the list of actions.
3292 * Pointer to the RSS action if exist, else return NULL.
3294 static const struct rte_flow_action_rss*
3295 flow_get_rss_action(const struct rte_flow_action actions[])
3297 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3298 switch (actions->type) {
3299 case RTE_FLOW_ACTION_TYPE_RSS:
3300 return (const struct rte_flow_action_rss *)
3309 /* maps shared action to translated non shared in some actions array */
3310 struct mlx5_translated_shared_action {
3311 struct rte_flow_shared_action *action; /**< Shared action */
3312 int index; /**< Index in related array of rte_flow_action */
3316 * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3317 * non shared action if translation possible.
3318 * This functionality used to run same execution path for both shared & non
3319 * shared actions on flow create. All necessary preparations for shared
3320 * action handling should be preformed on *shared* actions list returned
3323 * @param[in] actions
3324 * List of actions to translate.
3325 * @param[out] shared
3326 * List to store translated shared actions.
3327 * @param[in, out] shared_n
3328 * Size of *shared* array. On return should be updated with number of shared
3329 * actions retrieved from the *actions* list.
3330 * @param[out] translated_actions
3331 * List of actions where all shared actions were translated to non shared
3332 * if possible. NULL if no translation took place.
3334 * Pointer to the error structure.
3337 * 0 on success, a negative errno value otherwise and rte_errno is set.
3340 flow_shared_actions_translate(const struct rte_flow_action actions[],
3341 struct mlx5_translated_shared_action *shared,
3343 struct rte_flow_action **translated_actions,
3344 struct rte_flow_error *error)
3346 struct rte_flow_action *translated = NULL;
3347 size_t actions_size;
3350 struct mlx5_translated_shared_action *shared_end = NULL;
3352 for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3353 if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3355 if (copied_n == *shared_n) {
3356 return rte_flow_error_set
3357 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3358 NULL, "too many shared actions");
3360 rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3361 sizeof(actions[n].conf));
3362 shared[copied_n].index = n;
3366 *shared_n = copied_n;
3369 actions_size = sizeof(struct rte_flow_action) * n;
3370 translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3375 memcpy(translated, actions, actions_size);
3376 for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3377 const struct rte_flow_shared_action *shared_action;
3379 shared_action = shared->action;
3380 switch (shared_action->type) {
3381 case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
3382 translated[shared->index].type =
3383 RTE_FLOW_ACTION_TYPE_RSS;
3384 translated[shared->index].conf =
3385 &shared_action->rss.origin;
3388 mlx5_free(translated);
3389 return rte_flow_error_set
3390 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3391 NULL, "invalid shared action type");
3394 *translated_actions = translated;
3399 * Get Shared RSS action from the action list.
3402 * Pointer to the list of actions.
3403 * @param[in] shared_n
3404 * Actions list length.
3407 * Pointer to the MLX5 RSS action if exists, otherwise return NULL.
3409 static struct mlx5_shared_action_rss *
3410 flow_get_shared_rss_action(struct mlx5_translated_shared_action *shared,
3413 struct mlx5_translated_shared_action *shared_end;
3415 for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3416 struct rte_flow_shared_action *shared_action;
3418 shared_action = shared->action;
3419 switch (shared_action->type) {
3420 case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
3421 __atomic_add_fetch(&shared_action->refcnt, 1,
3423 return &shared_action->rss;
3431 struct rte_flow_shared_action *
3432 mlx5_flow_get_shared_rss(struct rte_flow *flow)
3434 if (flow->shared_rss)
3435 return container_of(flow->shared_rss,
3436 struct rte_flow_shared_action, rss);
3442 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3444 const struct rte_flow_item *item;
3445 unsigned int has_vlan = 0;
3447 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3448 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3454 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3455 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3456 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3457 MLX5_EXPANSION_ROOT_OUTER;
3461 * Get layer flags from the prefix flow.
3463 * Some flows may be split to several subflows, the prefix subflow gets the
3464 * match items and the suffix sub flow gets the actions.
3465 * Some actions need the user defined match item flags to get the detail for
3467 * This function helps the suffix flow to get the item layer flags from prefix
3470 * @param[in] dev_flow
3471 * Pointer the created preifx subflow.
3474 * The layers get from prefix subflow.
3476 static inline uint64_t
3477 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3479 uint64_t layers = 0;
3482 * Layers bits could be localization, but usually the compiler will
3483 * help to do the optimization work for source code.
3484 * If no decap actions, use the layers directly.
3486 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3487 return dev_flow->handle->layers;
3488 /* Convert L3 layers with decap action. */
3489 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3490 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3491 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3492 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3493 /* Convert L4 layers with decap action. */
3494 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3495 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3496 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3497 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3502 * Get metadata split action information.
3504 * @param[in] actions
3505 * Pointer to the list of actions.
3507 * Pointer to the return pointer.
3508 * @param[out] qrss_type
3509 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3510 * if no QUEUE/RSS is found.
3511 * @param[out] encap_idx
3512 * Pointer to the index of the encap action if exists, otherwise the last
3516 * Total number of actions.
3519 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3520 const struct rte_flow_action **qrss,
3523 const struct rte_flow_action_raw_encap *raw_encap;
3525 int raw_decap_idx = -1;
3528 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3529 switch (actions->type) {
3530 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3531 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3532 *encap_idx = actions_n;
3534 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3535 raw_decap_idx = actions_n;
3537 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3538 raw_encap = actions->conf;
3539 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3540 *encap_idx = raw_decap_idx != -1 ?
3541 raw_decap_idx : actions_n;
3543 case RTE_FLOW_ACTION_TYPE_QUEUE:
3544 case RTE_FLOW_ACTION_TYPE_RSS:
3552 if (*encap_idx == -1)
3553 *encap_idx = actions_n;
3554 /* Count RTE_FLOW_ACTION_TYPE_END. */
3555 return actions_n + 1;
3559 * Check meter action from the action list.
3561 * @param[in] actions
3562 * Pointer to the list of actions.
3564 * Pointer to the meter exist flag.
3567 * Total number of actions.
3570 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3576 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3577 switch (actions->type) {
3578 case RTE_FLOW_ACTION_TYPE_METER:
3586 /* Count RTE_FLOW_ACTION_TYPE_END. */
3587 return actions_n + 1;
3591 * Check if the flow should be split due to hairpin.
3592 * The reason for the split is that in current HW we can't
3593 * support encap and push-vlan on Rx, so if a flow contains
3594 * these actions we move it to Tx.
3597 * Pointer to Ethernet device.
3599 * Flow rule attributes.
3600 * @param[in] actions
3601 * Associated actions (list terminated by the END action).
3604 * > 0 the number of actions and the flow should be split,
3605 * 0 when no split required.
3608 flow_check_hairpin_split(struct rte_eth_dev *dev,
3609 const struct rte_flow_attr *attr,
3610 const struct rte_flow_action actions[])
3612 int queue_action = 0;
3615 const struct rte_flow_action_queue *queue;
3616 const struct rte_flow_action_rss *rss;
3617 const struct rte_flow_action_raw_encap *raw_encap;
3618 const struct rte_eth_hairpin_conf *conf;
3622 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3623 switch (actions->type) {
3624 case RTE_FLOW_ACTION_TYPE_QUEUE:
3625 queue = actions->conf;
3628 conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3629 if (conf != NULL && !!conf->tx_explicit)
3634 case RTE_FLOW_ACTION_TYPE_RSS:
3635 rss = actions->conf;
3636 if (rss == NULL || rss->queue_num == 0)
3638 conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3639 if (conf != NULL && !!conf->tx_explicit)
3644 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3645 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3646 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3647 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3648 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3652 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3653 raw_encap = actions->conf;
3654 if (raw_encap->size >
3655 (sizeof(struct rte_flow_item_eth) +
3656 sizeof(struct rte_flow_item_ipv4)))
3665 if (split && queue_action)
3670 /* Declare flow create/destroy prototype in advance. */
3672 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3673 const struct rte_flow_attr *attr,
3674 const struct rte_flow_item items[],
3675 const struct rte_flow_action actions[],
3676 bool external, struct rte_flow_error *error);
3679 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3683 * Add a flow of copying flow metadata registers in RX_CP_TBL.
3685 * As mark_id is unique, if there's already a registered flow for the mark_id,
3686 * return by increasing the reference counter of the resource. Otherwise, create
3687 * the resource (mcp_res) and flow.
3690 * - If ingress port is ANY and reg_c[1] is mark_id,
3691 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3693 * For default flow (zero mark_id), flow is like,
3694 * - If ingress port is ANY,
3695 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
3698 * Pointer to Ethernet device.
3700 * ID of MARK action, zero means default flow for META.
3702 * Perform verbose error reporting if not NULL.
3705 * Associated resource on success, NULL otherwise and rte_errno is set.
3707 static struct mlx5_flow_mreg_copy_resource *
3708 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3709 struct rte_flow_error *error)
3711 struct mlx5_priv *priv = dev->data->dev_private;
3712 struct rte_flow_attr attr = {
3713 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3716 struct mlx5_rte_flow_item_tag tag_spec = {
3719 struct rte_flow_item items[] = {
3720 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3722 struct rte_flow_action_mark ftag = {
3725 struct mlx5_flow_action_copy_mreg cp_mreg = {
3729 struct rte_flow_action_jump jump = {
3730 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3732 struct rte_flow_action actions[] = {
3733 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3735 struct mlx5_flow_mreg_copy_resource *mcp_res;
3739 /* Fill the register fileds in the flow. */
3740 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3744 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3748 /* Check if already registered. */
3749 MLX5_ASSERT(priv->mreg_cp_tbl);
3750 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3752 /* For non-default rule. */
3753 if (mark_id != MLX5_DEFAULT_COPY_ID)
3755 MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3756 mcp_res->refcnt == 1);
3759 /* Provide the full width of FLAG specific value. */
3760 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3761 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3762 /* Build a new flow. */
3763 if (mark_id != MLX5_DEFAULT_COPY_ID) {
3764 items[0] = (struct rte_flow_item){
3765 .type = (enum rte_flow_item_type)
3766 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3769 items[1] = (struct rte_flow_item){
3770 .type = RTE_FLOW_ITEM_TYPE_END,
3772 actions[0] = (struct rte_flow_action){
3773 .type = (enum rte_flow_action_type)
3774 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3777 actions[1] = (struct rte_flow_action){
3778 .type = (enum rte_flow_action_type)
3779 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3782 actions[2] = (struct rte_flow_action){
3783 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3786 actions[3] = (struct rte_flow_action){
3787 .type = RTE_FLOW_ACTION_TYPE_END,
3790 /* Default rule, wildcard match. */
3791 attr.priority = MLX5_FLOW_PRIO_RSVD;
3792 items[0] = (struct rte_flow_item){
3793 .type = RTE_FLOW_ITEM_TYPE_END,
3795 actions[0] = (struct rte_flow_action){
3796 .type = (enum rte_flow_action_type)
3797 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3800 actions[1] = (struct rte_flow_action){
3801 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3804 actions[2] = (struct rte_flow_action){
3805 .type = RTE_FLOW_ACTION_TYPE_END,
3808 /* Build a new entry. */
3809 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3816 * The copy Flows are not included in any list. There
3817 * ones are referenced from other Flows and can not
3818 * be applied, removed, deleted in ardbitrary order
3819 * by list traversing.
3821 mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3822 actions, false, error);
3823 if (!mcp_res->rix_flow)
3826 mcp_res->hlist_ent.key = mark_id;
3827 ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3828 &mcp_res->hlist_ent);
3834 if (mcp_res->rix_flow)
3835 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3836 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3841 * Release flow in RX_CP_TBL.
3844 * Pointer to Ethernet device.
3846 * Parent flow for wich copying is provided.
3849 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3850 struct rte_flow *flow)
3852 struct mlx5_flow_mreg_copy_resource *mcp_res;
3853 struct mlx5_priv *priv = dev->data->dev_private;
3855 if (!flow->rix_mreg_copy)
3857 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3858 flow->rix_mreg_copy);
3859 if (!mcp_res || !priv->mreg_cp_tbl)
3861 if (flow->copy_applied) {
3862 MLX5_ASSERT(mcp_res->appcnt);
3863 flow->copy_applied = 0;
3865 if (!mcp_res->appcnt) {
3866 struct rte_flow *mcp_flow = mlx5_ipool_get
3867 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3871 flow_drv_remove(dev, mcp_flow);
3875 * We do not check availability of metadata registers here,
3876 * because copy resources are not allocated in this case.
3878 if (--mcp_res->refcnt)
3880 MLX5_ASSERT(mcp_res->rix_flow);
3881 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3882 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3883 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3884 flow->rix_mreg_copy = 0;
3888 * Start flow in RX_CP_TBL.
3891 * Pointer to Ethernet device.
3893 * Parent flow for wich copying is provided.
3896 * 0 on success, a negative errno value otherwise and rte_errno is set.
3899 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3900 struct rte_flow *flow)
3902 struct mlx5_flow_mreg_copy_resource *mcp_res;
3903 struct mlx5_priv *priv = dev->data->dev_private;
3906 if (!flow->rix_mreg_copy || flow->copy_applied)
3908 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3909 flow->rix_mreg_copy);
3912 if (!mcp_res->appcnt) {
3913 struct rte_flow *mcp_flow = mlx5_ipool_get
3914 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3918 ret = flow_drv_apply(dev, mcp_flow, NULL);
3924 flow->copy_applied = 1;
3929 * Stop flow in RX_CP_TBL.
3932 * Pointer to Ethernet device.
3934 * Parent flow for wich copying is provided.
3937 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3938 struct rte_flow *flow)
3940 struct mlx5_flow_mreg_copy_resource *mcp_res;
3941 struct mlx5_priv *priv = dev->data->dev_private;
3943 if (!flow->rix_mreg_copy || !flow->copy_applied)
3945 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3946 flow->rix_mreg_copy);
3949 MLX5_ASSERT(mcp_res->appcnt);
3951 flow->copy_applied = 0;
3952 if (!mcp_res->appcnt) {
3953 struct rte_flow *mcp_flow = mlx5_ipool_get
3954 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3958 flow_drv_remove(dev, mcp_flow);
3963 * Remove the default copy action from RX_CP_TBL.
3966 * Pointer to Ethernet device.
3969 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3971 struct mlx5_flow_mreg_copy_resource *mcp_res;
3972 struct mlx5_priv *priv = dev->data->dev_private;
3974 /* Check if default flow is registered. */
3975 if (!priv->mreg_cp_tbl)
3977 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3978 MLX5_DEFAULT_COPY_ID);
3981 MLX5_ASSERT(mcp_res->rix_flow);
3982 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3983 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3984 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3988 * Add the default copy action in in RX_CP_TBL.
3991 * Pointer to Ethernet device.
3993 * Perform verbose error reporting if not NULL.
3996 * 0 for success, negative value otherwise and rte_errno is set.
3999 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4000 struct rte_flow_error *error)
4002 struct mlx5_priv *priv = dev->data->dev_private;
4003 struct mlx5_flow_mreg_copy_resource *mcp_res;
4005 /* Check whether extensive metadata feature is engaged. */
4006 if (!priv->config.dv_flow_en ||
4007 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4008 !mlx5_flow_ext_mreg_supported(dev) ||
4009 !priv->sh->dv_regc0_mask)
4011 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4018 * Add a flow of copying flow metadata registers in RX_CP_TBL.
4020 * All the flow having Q/RSS action should be split by
4021 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4022 * performs the following,
4023 * - CQE->flow_tag := reg_c[1] (MARK)
4024 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4025 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4026 * but there should be a flow per each MARK ID set by MARK action.
4028 * For the aforementioned reason, if there's a MARK action in flow's action
4029 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4030 * the MARK ID to CQE's flow_tag like,
4031 * - If reg_c[1] is mark_id,
4032 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4034 * For SET_META action which stores value in reg_c[0], as the destination is
4035 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4036 * MARK ID means the default flow. The default flow looks like,
4037 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4040 * Pointer to Ethernet device.
4042 * Pointer to flow structure.
4043 * @param[in] actions
4044 * Pointer to the list of actions.
4046 * Perform verbose error reporting if not NULL.
4049 * 0 on success, negative value otherwise and rte_errno is set.
4052 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4053 struct rte_flow *flow,
4054 const struct rte_flow_action *actions,
4055 struct rte_flow_error *error)
4057 struct mlx5_priv *priv = dev->data->dev_private;
4058 struct mlx5_dev_config *config = &priv->config;
4059 struct mlx5_flow_mreg_copy_resource *mcp_res;
4060 const struct rte_flow_action_mark *mark;
4062 /* Check whether extensive metadata feature is engaged. */
4063 if (!config->dv_flow_en ||
4064 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4065 !mlx5_flow_ext_mreg_supported(dev) ||
4066 !priv->sh->dv_regc0_mask)
4068 /* Find MARK action. */
4069 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4070 switch (actions->type) {
4071 case RTE_FLOW_ACTION_TYPE_FLAG:
4072 mcp_res = flow_mreg_add_copy_action
4073 (dev, MLX5_FLOW_MARK_DEFAULT, error);
4076 flow->rix_mreg_copy = mcp_res->idx;
4077 if (dev->data->dev_started) {
4079 flow->copy_applied = 1;
4082 case RTE_FLOW_ACTION_TYPE_MARK:
4083 mark = (const struct rte_flow_action_mark *)
4086 flow_mreg_add_copy_action(dev, mark->id, error);
4089 flow->rix_mreg_copy = mcp_res->idx;
4090 if (dev->data->dev_started) {
4092 flow->copy_applied = 1;
4102 #define MLX5_MAX_SPLIT_ACTIONS 24
4103 #define MLX5_MAX_SPLIT_ITEMS 24
4106 * Split the hairpin flow.
4107 * Since HW can't support encap and push-vlan on Rx, we move these
4109 * If the count action is after the encap then we also
4110 * move the count action. in this case the count will also measure
4114 * Pointer to Ethernet device.
4115 * @param[in] actions
4116 * Associated actions (list terminated by the END action).
4117 * @param[out] actions_rx
4119 * @param[out] actions_tx
4121 * @param[out] pattern_tx
4122 * The pattern items for the Tx flow.
4123 * @param[out] flow_id
4124 * The flow ID connected to this flow.
4130 flow_hairpin_split(struct rte_eth_dev *dev,
4131 const struct rte_flow_action actions[],
4132 struct rte_flow_action actions_rx[],
4133 struct rte_flow_action actions_tx[],
4134 struct rte_flow_item pattern_tx[],
4137 const struct rte_flow_action_raw_encap *raw_encap;
4138 const struct rte_flow_action_raw_decap *raw_decap;
4139 struct mlx5_rte_flow_action_set_tag *set_tag;
4140 struct rte_flow_action *tag_action;
4141 struct mlx5_rte_flow_item_tag *tag_item;
4142 struct rte_flow_item *item;
4146 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4147 switch (actions->type) {
4148 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4149 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4150 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4151 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4152 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4153 rte_memcpy(actions_tx, actions,
4154 sizeof(struct rte_flow_action));
4157 case RTE_FLOW_ACTION_TYPE_COUNT:
4159 rte_memcpy(actions_tx, actions,
4160 sizeof(struct rte_flow_action));
4163 rte_memcpy(actions_rx, actions,
4164 sizeof(struct rte_flow_action));
4168 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4169 raw_encap = actions->conf;
4170 if (raw_encap->size >
4171 (sizeof(struct rte_flow_item_eth) +
4172 sizeof(struct rte_flow_item_ipv4))) {
4173 memcpy(actions_tx, actions,
4174 sizeof(struct rte_flow_action));
4178 rte_memcpy(actions_rx, actions,
4179 sizeof(struct rte_flow_action));
4183 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4184 raw_decap = actions->conf;
4185 if (raw_decap->size <
4186 (sizeof(struct rte_flow_item_eth) +
4187 sizeof(struct rte_flow_item_ipv4))) {
4188 memcpy(actions_tx, actions,
4189 sizeof(struct rte_flow_action));
4192 rte_memcpy(actions_rx, actions,
4193 sizeof(struct rte_flow_action));
4198 rte_memcpy(actions_rx, actions,
4199 sizeof(struct rte_flow_action));
4204 /* Add set meta action and end action for the Rx flow. */
4205 tag_action = actions_rx;
4206 tag_action->type = (enum rte_flow_action_type)
4207 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4209 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4211 set_tag = (void *)actions_rx;
4212 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4213 MLX5_ASSERT(set_tag->id > REG_NON);
4214 set_tag->data = flow_id;
4215 tag_action->conf = set_tag;
4216 /* Create Tx item list. */
4217 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4218 addr = (void *)&pattern_tx[2];
4220 item->type = (enum rte_flow_item_type)
4221 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4222 tag_item = (void *)addr;
4223 tag_item->data = flow_id;
4224 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4225 MLX5_ASSERT(set_tag->id > REG_NON);
4226 item->spec = tag_item;
4227 addr += sizeof(struct mlx5_rte_flow_item_tag);
4228 tag_item = (void *)addr;
4229 tag_item->data = UINT32_MAX;
4230 tag_item->id = UINT16_MAX;
4231 item->mask = tag_item;
4234 item->type = RTE_FLOW_ITEM_TYPE_END;
4239 union tunnel_offload_mark {
4242 uint32_t app_reserve:8;
4243 uint32_t table_id:15;
4244 uint32_t transfer:1;
4245 uint32_t _unused_:8;
4249 struct tunnel_default_miss_ctx {
4253 struct rte_flow_action_rss action_rss;
4254 struct rte_flow_action_queue miss_queue;
4255 struct rte_flow_action_jump miss_jump;
4261 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
4262 struct rte_flow *flow,
4263 const struct rte_flow_attr *attr,
4264 const struct rte_flow_action *app_actions,
4266 struct tunnel_default_miss_ctx *ctx,
4267 struct rte_flow_error *error)
4269 struct mlx5_priv *priv = dev->data->dev_private;
4270 struct mlx5_flow *dev_flow;
4271 struct rte_flow_attr miss_attr = *attr;
4272 const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
4273 const struct rte_flow_item miss_items[2] = {
4275 .type = RTE_FLOW_ITEM_TYPE_ETH,
4281 .type = RTE_FLOW_ITEM_TYPE_END,
4287 union tunnel_offload_mark mark_id;
4288 struct rte_flow_action_mark miss_mark;
4289 struct rte_flow_action miss_actions[3] = {
4290 [0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
4291 [2] = { .type = RTE_FLOW_ACTION_TYPE_END, .conf = NULL }
4293 const struct rte_flow_action_jump *jump_data;
4294 uint32_t i, flow_table = 0; /* prevent compilation warning */
4295 struct flow_grp_info grp_info = {
4297 .transfer = attr->transfer,
4298 .fdb_def_rule = !!priv->fdb_def_rule,
4303 if (!attr->transfer) {
4306 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
4307 q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
4308 ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
4311 return rte_flow_error_set
4313 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4314 NULL, "invalid default miss RSS");
4315 ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4316 ctx->action_rss.level = 0,
4317 ctx->action_rss.types = priv->rss_conf.rss_hf,
4318 ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
4319 ctx->action_rss.queue_num = priv->reta_idx_n,
4320 ctx->action_rss.key = priv->rss_conf.rss_key,
4321 ctx->action_rss.queue = ctx->queue;
4322 if (!priv->reta_idx_n || !priv->rxqs_n)
4323 return rte_flow_error_set
4325 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4326 NULL, "invalid port configuration");
4327 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4328 ctx->action_rss.types = 0;
4329 for (i = 0; i != priv->reta_idx_n; ++i)
4330 ctx->queue[i] = (*priv->reta_idx)[i];
4332 miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
4333 ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
4335 miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
4336 for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
4337 jump_data = app_actions->conf;
4338 miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
4339 miss_attr.group = jump_data->group;
4340 ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
4341 &flow_table, grp_info, error);
4343 return rte_flow_error_set(error, EINVAL,
4344 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4345 NULL, "invalid tunnel id");
4346 mark_id.app_reserve = 0;
4347 mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
4348 mark_id.transfer = !!attr->transfer;
4349 mark_id._unused_ = 0;
4350 miss_mark.id = mark_id.val;
4351 dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
4352 miss_items, miss_actions, flow_idx, error);
4355 dev_flow->flow = flow;
4356 dev_flow->external = true;
4357 dev_flow->tunnel = tunnel;
4358 /* Subflow object was created, we must include one in the list. */
4359 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4360 dev_flow->handle, next);
4362 "port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
4363 dev->data->port_id, tunnel->app_tunnel.type,
4364 tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
4365 ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
4366 miss_actions, error);
4368 ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
4375 * The last stage of splitting chain, just creates the subflow
4376 * without any modification.
4379 * Pointer to Ethernet device.
4381 * Parent flow structure pointer.
4382 * @param[in, out] sub_flow
4383 * Pointer to return the created subflow, may be NULL.
4384 * @param[in] prefix_layers
4385 * Prefix subflow layers, may be 0.
4386 * @param[in] prefix_mark
4387 * Prefix subflow mark flag, may be 0.
4389 * Flow rule attributes.
4391 * Pattern specification (list terminated by the END pattern item).
4392 * @param[in] actions
4393 * Associated actions (list terminated by the END action).
4394 * @param[in] external
4395 * This flow rule is created by request external to PMD.
4396 * @param[in] flow_idx
4397 * This memory pool index to the flow.
4399 * Perform verbose error reporting if not NULL.
4401 * 0 on success, negative value otherwise
4404 flow_create_split_inner(struct rte_eth_dev *dev,
4405 struct rte_flow *flow,
4406 struct mlx5_flow **sub_flow,
4407 uint64_t prefix_layers,
4408 uint32_t prefix_mark,
4409 const struct rte_flow_attr *attr,
4410 const struct rte_flow_item items[],
4411 const struct rte_flow_action actions[],
4412 bool external, uint32_t flow_idx,
4413 struct rte_flow_error *error)
4415 struct mlx5_flow *dev_flow;
4417 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4421 dev_flow->flow = flow;
4422 dev_flow->external = external;
4423 /* Subflow object was created, we must include one in the list. */
4424 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4425 dev_flow->handle, next);
4427 * If dev_flow is as one of the suffix flow, some actions in suffix
4428 * flow may need some user defined item layer flags, and pass the
4429 * Metadate rxq mark flag to suffix flow as well.
4432 dev_flow->handle->layers = prefix_layers;
4434 dev_flow->handle->mark = 1;
4436 *sub_flow = dev_flow;
4437 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4441 * Split the meter flow.
4443 * As meter flow will split to three sub flow, other than meter
4444 * action, the other actions make sense to only meter accepts
4445 * the packet. If it need to be dropped, no other additional
4446 * actions should be take.
4448 * One kind of special action which decapsulates the L3 tunnel
4449 * header will be in the prefix sub flow, as not to take the
4450 * L3 tunnel header into account.
4453 * Pointer to Ethernet device.
4455 * Pattern specification (list terminated by the END pattern item).
4456 * @param[out] sfx_items
4457 * Suffix flow match items (list terminated by the END pattern item).
4458 * @param[in] actions
4459 * Associated actions (list terminated by the END action).
4460 * @param[out] actions_sfx
4461 * Suffix flow actions.
4462 * @param[out] actions_pre
4463 * Prefix flow actions.
4464 * @param[out] pattern_sfx
4465 * The pattern items for the suffix flow.
4466 * @param[out] tag_sfx
4467 * Pointer to suffix flow tag.
4473 flow_meter_split_prep(struct rte_eth_dev *dev,
4474 const struct rte_flow_item items[],
4475 struct rte_flow_item sfx_items[],
4476 const struct rte_flow_action actions[],
4477 struct rte_flow_action actions_sfx[],
4478 struct rte_flow_action actions_pre[])
4480 struct mlx5_priv *priv = dev->data->dev_private;
4481 struct rte_flow_action *tag_action = NULL;
4482 struct rte_flow_item *tag_item;
4483 struct mlx5_rte_flow_action_set_tag *set_tag;
4484 struct rte_flow_error error;
4485 const struct rte_flow_action_raw_encap *raw_encap;
4486 const struct rte_flow_action_raw_decap *raw_decap;
4487 struct mlx5_rte_flow_item_tag *tag_spec;
4488 struct mlx5_rte_flow_item_tag *tag_mask;
4489 uint32_t tag_id = 0;
4490 bool copy_vlan = false;
4492 /* Prepare the actions for prefix and suffix flow. */
4493 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4494 struct rte_flow_action **action_cur = NULL;
4496 switch (actions->type) {
4497 case RTE_FLOW_ACTION_TYPE_METER:
4498 /* Add the extra tag action first. */
4499 tag_action = actions_pre;
4500 tag_action->type = (enum rte_flow_action_type)
4501 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4503 action_cur = &actions_pre;
4505 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4506 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4507 action_cur = &actions_pre;
4509 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4510 raw_encap = actions->conf;
4511 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4512 action_cur = &actions_pre;
4514 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4515 raw_decap = actions->conf;
4516 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4517 action_cur = &actions_pre;
4519 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4520 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4527 action_cur = &actions_sfx;
4528 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4531 /* Add end action to the actions. */
4532 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4533 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4536 set_tag = (void *)actions_pre;
4537 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4538 mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4540 if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4541 DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4542 dev->data->port_id);
4543 mlx5_ipool_free(priv->sh->ipool
4544 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4546 } else if (!tag_id) {
4549 set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4551 tag_action->conf = set_tag;
4552 /* Prepare the suffix subflow items. */
4553 tag_item = sfx_items++;
4554 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4555 int item_type = items->type;
4557 switch (item_type) {
4558 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4559 memcpy(sfx_items, items, sizeof(*sfx_items));
4562 case RTE_FLOW_ITEM_TYPE_VLAN:
4564 memcpy(sfx_items, items, sizeof(*sfx_items));
4566 * Convert to internal match item, it is used
4567 * for vlan push and set vid.
4569 sfx_items->type = (enum rte_flow_item_type)
4570 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4578 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4580 tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4581 tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4582 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4583 tag_mask = tag_spec + 1;
4584 tag_mask->data = 0xffffff00;
4585 tag_item->type = (enum rte_flow_item_type)
4586 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4587 tag_item->spec = tag_spec;
4588 tag_item->last = NULL;
4589 tag_item->mask = tag_mask;
4594 * Split action list having QUEUE/RSS for metadata register copy.
4596 * Once Q/RSS action is detected in user's action list, the flow action
4597 * should be split in order to copy metadata registers, which will happen in
4599 * - CQE->flow_tag := reg_c[1] (MARK)
4600 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4601 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4602 * This is because the last action of each flow must be a terminal action
4603 * (QUEUE, RSS or DROP).
4605 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4606 * stored and kept in the mlx5_flow structure per each sub_flow.
4608 * The Q/RSS action is replaced with,
4609 * - SET_TAG, setting the allocated flow ID to reg_c[2].
4610 * And the following JUMP action is added at the end,
4611 * - JUMP, to RX_CP_TBL.
4613 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4614 * flow_create_split_metadata() routine. The flow will look like,
4615 * - If flow ID matches (reg_c[2]), perform Q/RSS.
4618 * Pointer to Ethernet device.
4619 * @param[out] split_actions
4620 * Pointer to store split actions to jump to CP_TBL.
4621 * @param[in] actions
4622 * Pointer to the list of original flow actions.
4624 * Pointer to the Q/RSS action.
4625 * @param[in] actions_n
4626 * Number of original actions.
4628 * Perform verbose error reporting if not NULL.
4631 * non-zero unique flow_id on success, otherwise 0 and
4632 * error/rte_error are set.
4635 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4636 struct rte_flow_action *split_actions,
4637 const struct rte_flow_action *actions,
4638 const struct rte_flow_action *qrss,
4639 int actions_n, struct rte_flow_error *error)
4641 struct mlx5_priv *priv = dev->data->dev_private;
4642 struct mlx5_rte_flow_action_set_tag *set_tag;
4643 struct rte_flow_action_jump *jump;
4644 const int qrss_idx = qrss - actions;
4645 uint32_t flow_id = 0;
4649 * Given actions will be split
4650 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4651 * - Add jump to mreg CP_TBL.
4652 * As a result, there will be one more action.
4655 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4656 set_tag = (void *)(split_actions + actions_n);
4658 * If tag action is not set to void(it means we are not the meter
4659 * suffix flow), add the tag action. Since meter suffix flow already
4660 * has the tag added.
4662 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4664 * Allocate the new subflow ID. This one is unique within
4665 * device and not shared with representors. Otherwise,
4666 * we would have to resolve multi-thread access synch
4667 * issue. Each flow on the shared device is appended
4668 * with source vport identifier, so the resulting
4669 * flows will be unique in the shared (by master and
4670 * representors) domain even if they have coinciding
4673 mlx5_ipool_malloc(priv->sh->ipool
4674 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4676 return rte_flow_error_set(error, ENOMEM,
4677 RTE_FLOW_ERROR_TYPE_ACTION,
4678 NULL, "can't allocate id "
4679 "for split Q/RSS subflow");
4680 /* Internal SET_TAG action to set flow ID. */
4681 *set_tag = (struct mlx5_rte_flow_action_set_tag){
4684 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4688 /* Construct new actions array. */
4689 /* Replace QUEUE/RSS action. */
4690 split_actions[qrss_idx] = (struct rte_flow_action){
4691 .type = (enum rte_flow_action_type)
4692 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4696 /* JUMP action to jump to mreg copy table (CP_TBL). */
4697 jump = (void *)(set_tag + 1);
4698 *jump = (struct rte_flow_action_jump){
4699 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4701 split_actions[actions_n - 2] = (struct rte_flow_action){
4702 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4705 split_actions[actions_n - 1] = (struct rte_flow_action){
4706 .type = RTE_FLOW_ACTION_TYPE_END,
4712 * Extend the given action list for Tx metadata copy.
4714 * Copy the given action list to the ext_actions and add flow metadata register
4715 * copy action in order to copy reg_a set by WQE to reg_c[0].
4717 * @param[out] ext_actions
4718 * Pointer to the extended action list.
4719 * @param[in] actions
4720 * Pointer to the list of actions.
4721 * @param[in] actions_n
4722 * Number of actions in the list.
4724 * Perform verbose error reporting if not NULL.
4725 * @param[in] encap_idx
4726 * The encap action inndex.
4729 * 0 on success, negative value otherwise
4732 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4733 struct rte_flow_action *ext_actions,
4734 const struct rte_flow_action *actions,
4735 int actions_n, struct rte_flow_error *error,
4738 struct mlx5_flow_action_copy_mreg *cp_mreg =
4739 (struct mlx5_flow_action_copy_mreg *)
4740 (ext_actions + actions_n + 1);
4743 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4747 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4752 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4753 if (encap_idx == actions_n - 1) {
4754 ext_actions[actions_n - 1] = (struct rte_flow_action){
4755 .type = (enum rte_flow_action_type)
4756 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4759 ext_actions[actions_n] = (struct rte_flow_action){
4760 .type = RTE_FLOW_ACTION_TYPE_END,
4763 ext_actions[encap_idx] = (struct rte_flow_action){
4764 .type = (enum rte_flow_action_type)
4765 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4768 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4769 sizeof(*ext_actions) * (actions_n - encap_idx));
4775 * Check the match action from the action list.
4777 * @param[in] actions
4778 * Pointer to the list of actions.
4780 * Flow rule attributes.
4782 * The action to be check if exist.
4783 * @param[out] match_action_pos
4784 * Pointer to the position of the matched action if exists, otherwise is -1.
4785 * @param[out] qrss_action_pos
4786 * Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4789 * > 0 the total number of actions.
4790 * 0 if not found match action in action list.
4793 flow_check_match_action(const struct rte_flow_action actions[],
4794 const struct rte_flow_attr *attr,
4795 enum rte_flow_action_type action,
4796 int *match_action_pos, int *qrss_action_pos)
4798 const struct rte_flow_action_sample *sample;
4805 *match_action_pos = -1;
4806 *qrss_action_pos = -1;
4807 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4808 if (actions->type == action) {
4810 *match_action_pos = actions_n;
4812 if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4813 actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4814 *qrss_action_pos = actions_n;
4815 if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4817 if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4818 sample = actions->conf;
4819 ratio = sample->ratio;
4820 sub_type = ((const struct rte_flow_action *)
4821 (sample->actions))->type;
4825 if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4827 /* JUMP Action not support for Mirroring;
4828 * Mirroring support multi-destination;
4830 if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4834 /* Count RTE_FLOW_ACTION_TYPE_END. */
4835 return flag ? actions_n + 1 : 0;
4838 #define SAMPLE_SUFFIX_ITEM 2
4841 * Split the sample flow.
4843 * As sample flow will split to two sub flow, sample flow with
4844 * sample action, the other actions will move to new suffix flow.
4846 * Also add unique tag id with tag action in the sample flow,
4847 * the same tag id will be as match in the suffix flow.
4850 * Pointer to Ethernet device.
4852 * FDB egress flow flag.
4853 * @param[out] sfx_items
4854 * Suffix flow match items (list terminated by the END pattern item).
4855 * @param[in] actions
4856 * Associated actions (list terminated by the END action).
4857 * @param[out] actions_sfx
4858 * Suffix flow actions.
4859 * @param[out] actions_pre
4860 * Prefix flow actions.
4861 * @param[in] actions_n
4862 * The total number of actions.
4863 * @param[in] sample_action_pos
4864 * The sample action position.
4865 * @param[in] qrss_action_pos
4866 * The Queue/RSS action position.
4868 * Perform verbose error reporting if not NULL.
4871 * 0 on success, or unique flow_id, a negative errno value
4872 * otherwise and rte_errno is set.
4875 flow_sample_split_prep(struct rte_eth_dev *dev,
4877 struct rte_flow_item sfx_items[],
4878 const struct rte_flow_action actions[],
4879 struct rte_flow_action actions_sfx[],
4880 struct rte_flow_action actions_pre[],
4882 int sample_action_pos,
4883 int qrss_action_pos,
4884 struct rte_flow_error *error)
4886 struct mlx5_priv *priv = dev->data->dev_private;
4887 struct mlx5_rte_flow_action_set_tag *set_tag;
4888 struct mlx5_rte_flow_item_tag *tag_spec;
4889 struct mlx5_rte_flow_item_tag *tag_mask;
4890 uint32_t tag_id = 0;
4894 if (sample_action_pos < 0)
4895 return rte_flow_error_set(error, EINVAL,
4896 RTE_FLOW_ERROR_TYPE_ACTION,
4897 NULL, "invalid position of sample "
4900 /* Prepare the prefix tag action. */
4901 set_tag = (void *)(actions_pre + actions_n + 1);
4902 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4906 mlx5_ipool_malloc(priv->sh->ipool
4907 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4908 set_tag->data = tag_id;
4909 /* Prepare the suffix subflow items. */
4910 tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4911 tag_spec->data = tag_id;
4912 tag_spec->id = set_tag->id;
4913 tag_mask = tag_spec + 1;
4914 tag_mask->data = UINT32_MAX;
4915 sfx_items[0] = (struct rte_flow_item){
4916 .type = (enum rte_flow_item_type)
4917 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4922 sfx_items[1] = (struct rte_flow_item){
4923 .type = (enum rte_flow_item_type)
4924 RTE_FLOW_ITEM_TYPE_END,
4927 /* Prepare the actions for prefix and suffix flow. */
4928 if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4929 index = qrss_action_pos;
4930 /* Put the preceding the Queue/RSS action into prefix flow. */
4932 memcpy(actions_pre, actions,
4933 sizeof(struct rte_flow_action) * index);
4934 /* Put others preceding the sample action into prefix flow. */
4935 if (sample_action_pos > index + 1)
4936 memcpy(actions_pre + index, actions + index + 1,
4937 sizeof(struct rte_flow_action) *
4938 (sample_action_pos - index - 1));
4939 index = sample_action_pos - 1;
4940 /* Put Queue/RSS action into Suffix flow. */
4941 memcpy(actions_sfx, actions + qrss_action_pos,
4942 sizeof(struct rte_flow_action));
4945 index = sample_action_pos;
4947 memcpy(actions_pre, actions,
4948 sizeof(struct rte_flow_action) * index);
4950 /* Add the extra tag action for NIC-RX and E-Switch ingress. */
4952 actions_pre[index++] =
4953 (struct rte_flow_action){
4954 .type = (enum rte_flow_action_type)
4955 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4959 memcpy(actions_pre + index, actions + sample_action_pos,
4960 sizeof(struct rte_flow_action));
4962 actions_pre[index] = (struct rte_flow_action){
4963 .type = (enum rte_flow_action_type)
4964 RTE_FLOW_ACTION_TYPE_END,
4966 /* Put the actions after sample into Suffix flow. */
4967 memcpy(actions_sfx, actions + sample_action_pos + 1,
4968 sizeof(struct rte_flow_action) *
4969 (actions_n - sample_action_pos - 1));
4974 * The splitting for metadata feature.
4976 * - Q/RSS action on NIC Rx should be split in order to pass by
4977 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4978 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4980 * - All the actions on NIC Tx should have a mreg copy action to
4981 * copy reg_a from WQE to reg_c[0].
4984 * Pointer to Ethernet device.
4986 * Parent flow structure pointer.
4987 * @param[in] prefix_layers
4988 * Prefix flow layer flags.
4989 * @param[in] prefix_mark
4990 * Prefix subflow mark flag, may be 0.
4992 * Flow rule attributes.
4994 * Pattern specification (list terminated by the END pattern item).
4995 * @param[in] actions
4996 * Associated actions (list terminated by the END action).
4997 * @param[in] external
4998 * This flow rule is created by request external to PMD.
4999 * @param[in] flow_idx
5000 * This memory pool index to the flow.
5002 * Perform verbose error reporting if not NULL.
5004 * 0 on success, negative value otherwise
5007 flow_create_split_metadata(struct rte_eth_dev *dev,
5008 struct rte_flow *flow,
5009 uint64_t prefix_layers,
5010 uint32_t prefix_mark,
5011 const struct rte_flow_attr *attr,
5012 const struct rte_flow_item items[],
5013 const struct rte_flow_action actions[],
5014 bool external, uint32_t flow_idx,
5015 struct rte_flow_error *error)
5017 struct mlx5_priv *priv = dev->data->dev_private;
5018 struct mlx5_dev_config *config = &priv->config;
5019 const struct rte_flow_action *qrss = NULL;
5020 struct rte_flow_action *ext_actions = NULL;
5021 struct mlx5_flow *dev_flow = NULL;
5022 uint32_t qrss_id = 0;
5029 /* Check whether extensive metadata feature is engaged. */
5030 if (!config->dv_flow_en ||
5031 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5032 !mlx5_flow_ext_mreg_supported(dev))
5033 return flow_create_split_inner(dev, flow, NULL, prefix_layers,
5034 prefix_mark, attr, items,
5035 actions, external, flow_idx,
5037 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5040 /* Exclude hairpin flows from splitting. */
5041 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5042 const struct rte_flow_action_queue *queue;
5045 if (mlx5_rxq_get_type(dev, queue->index) ==
5046 MLX5_RXQ_TYPE_HAIRPIN)
5048 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5049 const struct rte_flow_action_rss *rss;
5052 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5053 MLX5_RXQ_TYPE_HAIRPIN)
5058 /* Check if it is in meter suffix table. */
5059 mtr_sfx = attr->group == (attr->transfer ?
5060 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5061 MLX5_FLOW_TABLE_LEVEL_SUFFIX);
5063 * Q/RSS action on NIC Rx should be split in order to pass by
5064 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5065 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5067 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5068 sizeof(struct rte_flow_action_set_tag) +
5069 sizeof(struct rte_flow_action_jump);
5070 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5073 return rte_flow_error_set(error, ENOMEM,
5074 RTE_FLOW_ERROR_TYPE_ACTION,
5075 NULL, "no memory to split "
5078 * If we are the suffix flow of meter, tag already exist.
5079 * Set the tag action to void.
5082 ext_actions[qrss - actions].type =
5083 RTE_FLOW_ACTION_TYPE_VOID;
5085 ext_actions[qrss - actions].type =
5086 (enum rte_flow_action_type)
5087 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5089 * Create the new actions list with removed Q/RSS action
5090 * and appended set tag and jump to register copy table
5091 * (RX_CP_TBL). We should preallocate unique tag ID here
5092 * in advance, because it is needed for set tag action.
5094 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5095 qrss, actions_n, error);
5096 if (!mtr_sfx && !qrss_id) {
5100 } else if (attr->egress && !attr->transfer) {
5102 * All the actions on NIC Tx should have a metadata register
5103 * copy action to copy reg_a from WQE to reg_c[meta]
5105 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5106 sizeof(struct mlx5_flow_action_copy_mreg);
5107 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5110 return rte_flow_error_set(error, ENOMEM,
5111 RTE_FLOW_ERROR_TYPE_ACTION,
5112 NULL, "no memory to split "
5114 /* Create the action list appended with copy register. */
5115 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5116 actions_n, error, encap_idx);
5120 /* Add the unmodified original or prefix subflow. */
5121 ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers,
5123 items, ext_actions ? ext_actions :
5124 actions, external, flow_idx, error);
5127 MLX5_ASSERT(dev_flow);
5129 const struct rte_flow_attr q_attr = {
5130 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5133 /* Internal PMD action to set register. */
5134 struct mlx5_rte_flow_item_tag q_tag_spec = {
5138 struct rte_flow_item q_items[] = {
5140 .type = (enum rte_flow_item_type)
5141 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5142 .spec = &q_tag_spec,
5147 .type = RTE_FLOW_ITEM_TYPE_END,
5150 struct rte_flow_action q_actions[] = {
5156 .type = RTE_FLOW_ACTION_TYPE_END,
5159 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5162 * Configure the tag item only if there is no meter subflow.
5163 * Since tag is already marked in the meter suffix subflow
5164 * we can just use the meter suffix items as is.
5167 /* Not meter subflow. */
5168 MLX5_ASSERT(!mtr_sfx);
5170 * Put unique id in prefix flow due to it is destroyed
5171 * after suffix flow and id will be freed after there
5172 * is no actual flows with this id and identifier
5173 * reallocation becomes possible (for example, for
5174 * other flows in other threads).
5176 dev_flow->handle->split_flow_id = qrss_id;
5177 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5181 q_tag_spec.id = ret;
5184 /* Add suffix subflow to execute Q/RSS. */
5185 ret = flow_create_split_inner(dev, flow, &dev_flow, layers, 0,
5186 &q_attr, mtr_sfx ? items :
5188 external, flow_idx, error);
5191 /* qrss ID should be freed if failed. */
5193 MLX5_ASSERT(dev_flow);
5198 * We do not destroy the partially created sub_flows in case of error.
5199 * These ones are included into parent flow list and will be destroyed
5200 * by flow_drv_destroy.
5202 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5204 mlx5_free(ext_actions);
5209 * The splitting for meter feature.
5211 * - The meter flow will be split to two flows as prefix and
5212 * suffix flow. The packets make sense only it pass the prefix
5215 * - Reg_C_5 is used for the packet to match betweend prefix and
5219 * Pointer to Ethernet device.
5221 * Parent flow structure pointer.
5222 * @param[in] prefix_layers
5223 * Prefix subflow layers, may be 0.
5224 * @param[in] prefix_mark
5225 * Prefix subflow mark flag, may be 0.
5227 * Flow rule attributes.
5229 * Pattern specification (list terminated by the END pattern item).
5230 * @param[in] actions
5231 * Associated actions (list terminated by the END action).
5232 * @param[in] external
5233 * This flow rule is created by request external to PMD.
5234 * @param[in] flow_idx
5235 * This memory pool index to the flow.
5237 * Perform verbose error reporting if not NULL.
5239 * 0 on success, negative value otherwise
5242 flow_create_split_meter(struct rte_eth_dev *dev,
5243 struct rte_flow *flow,
5244 uint64_t prefix_layers,
5245 uint32_t prefix_mark,
5246 const struct rte_flow_attr *attr,
5247 const struct rte_flow_item items[],
5248 const struct rte_flow_action actions[],
5249 bool external, uint32_t flow_idx,
5250 struct rte_flow_error *error)
5252 struct mlx5_priv *priv = dev->data->dev_private;
5253 struct rte_flow_action *sfx_actions = NULL;
5254 struct rte_flow_action *pre_actions = NULL;
5255 struct rte_flow_item *sfx_items = NULL;
5256 struct mlx5_flow *dev_flow = NULL;
5257 struct rte_flow_attr sfx_attr = *attr;
5259 uint32_t mtr_tag_id = 0;
5266 actions_n = flow_check_meter_action(actions, &mtr);
5268 /* The five prefix actions: meter, decap, encap, tag, end. */
5269 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5270 sizeof(struct mlx5_rte_flow_action_set_tag);
5271 /* tag, vlan, port id, end. */
5272 #define METER_SUFFIX_ITEM 4
5273 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5274 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5275 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5278 return rte_flow_error_set(error, ENOMEM,
5279 RTE_FLOW_ERROR_TYPE_ACTION,
5280 NULL, "no memory to split "
5282 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5284 pre_actions = sfx_actions + actions_n;
5285 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5286 actions, sfx_actions,
5292 /* Add the prefix subflow. */
5293 ret = flow_create_split_inner(dev, flow, &dev_flow,
5296 pre_actions, external,
5302 dev_flow->handle->split_flow_id = mtr_tag_id;
5303 /* Setting the sfx group atrr. */
5304 sfx_attr.group = sfx_attr.transfer ?
5305 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5306 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5308 /* Add the prefix subflow. */
5309 ret = flow_create_split_metadata(dev, flow, dev_flow ?
5310 flow_get_prefix_layer_flags(dev_flow) :
5311 prefix_layers, dev_flow ?
5312 dev_flow->handle->mark : prefix_mark,
5313 &sfx_attr, sfx_items ?
5315 sfx_actions ? sfx_actions : actions,
5316 external, flow_idx, error);
5319 mlx5_free(sfx_actions);
5324 * The splitting for sample feature.
5326 * Once Sample action is detected in the action list, the flow actions should
5327 * be split into prefix sub flow and suffix sub flow.
5329 * The original items remain in the prefix sub flow, all actions preceding the
5330 * sample action and the sample action itself will be copied to the prefix
5331 * sub flow, the actions following the sample action will be copied to the
5332 * suffix sub flow, Queue action always be located in the suffix sub flow.
5334 * In order to make the packet from prefix sub flow matches with suffix sub
5335 * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5336 * flow uses tag item with the unique flow id.
5339 * Pointer to Ethernet device.
5341 * Parent flow structure pointer.
5343 * Flow rule attributes.
5345 * Pattern specification (list terminated by the END pattern item).
5346 * @param[in] actions
5347 * Associated actions (list terminated by the END action).
5348 * @param[in] external
5349 * This flow rule is created by request external to PMD.
5350 * @param[in] flow_idx
5351 * This memory pool index to the flow.
5353 * Perform verbose error reporting if not NULL.
5355 * 0 on success, negative value otherwise
5358 flow_create_split_sample(struct rte_eth_dev *dev,
5359 struct rte_flow *flow,
5360 const struct rte_flow_attr *attr,
5361 const struct rte_flow_item items[],
5362 const struct rte_flow_action actions[],
5363 bool external, uint32_t flow_idx,
5364 struct rte_flow_error *error)
5366 struct mlx5_priv *priv = dev->data->dev_private;
5367 struct rte_flow_action *sfx_actions = NULL;
5368 struct rte_flow_action *pre_actions = NULL;
5369 struct rte_flow_item *sfx_items = NULL;
5370 struct mlx5_flow *dev_flow = NULL;
5371 struct rte_flow_attr sfx_attr = *attr;
5372 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5373 struct mlx5_flow_dv_sample_resource *sample_res;
5374 struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5375 struct mlx5_flow_tbl_resource *sfx_tbl;
5376 union mlx5_flow_tbl_key sfx_table_key;
5380 uint32_t fdb_tx = 0;
5383 int sample_action_pos;
5384 int qrss_action_pos;
5387 if (priv->sampler_en)
5388 actions_n = flow_check_match_action(actions, attr,
5389 RTE_FLOW_ACTION_TYPE_SAMPLE,
5390 &sample_action_pos, &qrss_action_pos);
5392 /* The prefix actions must includes sample, tag, end. */
5393 act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5394 + sizeof(struct mlx5_rte_flow_action_set_tag);
5395 item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5396 sizeof(struct mlx5_rte_flow_item_tag) * 2;
5397 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5398 item_size), 0, SOCKET_ID_ANY);
5400 return rte_flow_error_set(error, ENOMEM,
5401 RTE_FLOW_ERROR_TYPE_ACTION,
5402 NULL, "no memory to split "
5404 /* The representor_id is -1 for uplink. */
5405 fdb_tx = (attr->transfer && priv->representor_id != -1);
5407 sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5409 pre_actions = sfx_actions + actions_n;
5410 tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
5411 actions, sfx_actions,
5412 pre_actions, actions_n,
5414 qrss_action_pos, error);
5415 if (tag_id < 0 || (!fdb_tx && !tag_id)) {
5419 /* Add the prefix subflow. */
5420 ret = flow_create_split_inner(dev, flow, &dev_flow, 0, 0, attr,
5421 items, pre_actions, external,
5427 dev_flow->handle->split_flow_id = tag_id;
5428 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5429 /* Set the sfx group attr. */
5430 sample_res = (struct mlx5_flow_dv_sample_resource *)
5431 dev_flow->dv.sample_res;
5432 sfx_tbl = (struct mlx5_flow_tbl_resource *)
5433 sample_res->normal_path_tbl;
5434 sfx_tbl_data = container_of(sfx_tbl,
5435 struct mlx5_flow_tbl_data_entry, tbl);
5436 sfx_table_key.v64 = sfx_tbl_data->entry.key;
5437 sfx_attr.group = sfx_attr.transfer ?
5438 (sfx_table_key.table_id - 1) :
5439 sfx_table_key.table_id;
5442 /* Add the suffix subflow. */
5443 ret = flow_create_split_meter(dev, flow, dev_flow ?
5444 flow_get_prefix_layer_flags(dev_flow) : 0,
5445 dev_flow ? dev_flow->handle->mark : 0,
5446 &sfx_attr, sfx_items ? sfx_items : items,
5447 sfx_actions ? sfx_actions : actions,
5448 external, flow_idx, error);
5451 mlx5_free(sfx_actions);
5456 * Split the flow to subflow set. The splitters might be linked
5457 * in the chain, like this:
5458 * flow_create_split_outer() calls:
5459 * flow_create_split_meter() calls:
5460 * flow_create_split_metadata(meter_subflow_0) calls:
5461 * flow_create_split_inner(metadata_subflow_0)
5462 * flow_create_split_inner(metadata_subflow_1)
5463 * flow_create_split_inner(metadata_subflow_2)
5464 * flow_create_split_metadata(meter_subflow_1) calls:
5465 * flow_create_split_inner(metadata_subflow_0)
5466 * flow_create_split_inner(metadata_subflow_1)
5467 * flow_create_split_inner(metadata_subflow_2)
5469 * This provide flexible way to add new levels of flow splitting.
5470 * The all of successfully created subflows are included to the
5471 * parent flow dev_flow list.
5474 * Pointer to Ethernet device.
5476 * Parent flow structure pointer.
5478 * Flow rule attributes.
5480 * Pattern specification (list terminated by the END pattern item).
5481 * @param[in] actions
5482 * Associated actions (list terminated by the END action).
5483 * @param[in] external
5484 * This flow rule is created by request external to PMD.
5485 * @param[in] flow_idx
5486 * This memory pool index to the flow.
5488 * Perform verbose error reporting if not NULL.
5490 * 0 on success, negative value otherwise
5493 flow_create_split_outer(struct rte_eth_dev *dev,
5494 struct rte_flow *flow,
5495 const struct rte_flow_attr *attr,
5496 const struct rte_flow_item items[],
5497 const struct rte_flow_action actions[],
5498 bool external, uint32_t flow_idx,
5499 struct rte_flow_error *error)
5503 ret = flow_create_split_sample(dev, flow, attr, items,
5504 actions, external, flow_idx, error);
5505 MLX5_ASSERT(ret <= 0);
5509 static struct mlx5_flow_tunnel *
5510 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5511 const struct rte_flow_attr *attr,
5512 const struct rte_flow_item items[],
5513 const struct rte_flow_action actions[])
5515 struct mlx5_flow_tunnel *tunnel;
5517 #pragma GCC diagnostic push
5518 #pragma GCC diagnostic ignored "-Wcast-qual"
5519 if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5520 tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5521 else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5522 tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5525 #pragma GCC diagnostic pop
5531 * Adjust flow RSS workspace if needed.
5534 * Pointer to thread flow work space.
5536 * Pointer to RSS descriptor.
5537 * @param[in] nrssq_num
5538 * New RSS queue number.
5541 * 0 on success, -1 otherwise and rte_errno is set.
5544 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5545 struct mlx5_flow_rss_desc *rss_desc,
5548 bool fidx = !!wks->flow_idx;
5550 if (likely(nrssq_num <= wks->rssq_num[fidx]))
5552 rss_desc->queue = realloc(rss_desc->queue,
5553 sizeof(rss_desc->queue[0]) * RTE_ALIGN(nrssq_num, 2));
5554 if (!rss_desc->queue) {
5558 wks->rssq_num[fidx] = RTE_ALIGN(nrssq_num, 2);
5563 * Create a flow and add it to @p list.
5566 * Pointer to Ethernet device.
5568 * Pointer to a TAILQ flow list. If this parameter NULL,
5569 * no list insertion occurred, flow is just created,
5570 * this is caller's responsibility to track the
5573 * Flow rule attributes.
5575 * Pattern specification (list terminated by the END pattern item).
5576 * @param[in] actions
5577 * Associated actions (list terminated by the END action).
5578 * @param[in] external
5579 * This flow rule is created by request external to PMD.
5581 * Perform verbose error reporting if not NULL.
5584 * A flow index on success, 0 otherwise and rte_errno is set.
5587 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5588 const struct rte_flow_attr *attr,
5589 const struct rte_flow_item items[],
5590 const struct rte_flow_action original_actions[],
5591 bool external, struct rte_flow_error *error)
5593 struct mlx5_priv *priv = dev->data->dev_private;
5594 struct rte_flow *flow = NULL;
5595 struct mlx5_flow *dev_flow;
5596 const struct rte_flow_action_rss *rss;
5597 struct mlx5_translated_shared_action
5598 shared_actions[MLX5_MAX_SHARED_ACTIONS];
5599 int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5601 struct mlx5_flow_expand_rss buf;
5602 uint8_t buffer[2048];
5605 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5606 uint8_t buffer[2048];
5609 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5610 uint8_t buffer[2048];
5611 } actions_hairpin_tx;
5613 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5614 uint8_t buffer[2048];
5616 struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5617 struct mlx5_flow_rss_desc *rss_desc;
5618 const struct rte_flow_action *p_actions_rx;
5622 struct rte_flow_attr attr_tx = { .priority = 0 };
5623 struct rte_flow_attr attr_factor = {0};
5624 const struct rte_flow_action *actions;
5625 struct rte_flow_action *translated_actions = NULL;
5626 struct mlx5_flow_tunnel *tunnel;
5627 struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5628 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5629 bool fidx = !!wks->flow_idx;
5633 rss_desc = &wks->rss_desc[fidx];
5634 ret = flow_shared_actions_translate(original_actions,
5637 &translated_actions, error);
5639 MLX5_ASSERT(translated_actions == NULL);
5642 actions = translated_actions ? translated_actions : original_actions;
5643 memcpy((void *)&attr_factor, (const void *)attr, sizeof(*attr));
5644 p_actions_rx = actions;
5645 hairpin_flow = flow_check_hairpin_split(dev, &attr_factor, actions);
5646 ret = flow_drv_validate(dev, &attr_factor, items, p_actions_rx,
5647 external, hairpin_flow, error);
5649 goto error_before_hairpin_split;
5650 flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5653 goto error_before_hairpin_split;
5655 if (hairpin_flow > 0) {
5656 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5658 goto error_before_hairpin_split;
5660 flow_hairpin_split(dev, actions, actions_rx.actions,
5661 actions_hairpin_tx.actions, items_tx.items,
5663 p_actions_rx = actions_rx.actions;
5665 flow->drv_type = flow_get_drv_type(dev, &attr_factor);
5666 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5667 flow->drv_type < MLX5_FLOW_TYPE_MAX);
5668 memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5669 rss = flow_get_rss_action(p_actions_rx);
5671 if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5674 * The following information is required by
5675 * mlx5_flow_hashfields_adjust() in advance.
5677 rss_desc->level = rss->level;
5678 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5679 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5681 flow->dev_handles = 0;
5682 if (rss && rss->types) {
5683 unsigned int graph_root;
5685 graph_root = find_graph_root(items, rss->level);
5686 ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5688 mlx5_support_expansion, graph_root);
5689 MLX5_ASSERT(ret > 0 &&
5690 (unsigned int)ret < sizeof(expand_buffer.buffer));
5693 buf->entry[0].pattern = (void *)(uintptr_t)items;
5695 flow->shared_rss = flow_get_shared_rss_action(shared_actions,
5698 * Record the start index when there is a nested call. All sub-flows
5699 * need to be translated before another calling.
5700 * No need to use ping-pong buffer to save memory here.
5703 MLX5_ASSERT(!wks->flow_nested_idx);
5704 wks->flow_nested_idx = fidx;
5706 for (i = 0; i < buf->entries; ++i) {
5708 * The splitter may create multiple dev_flows,
5709 * depending on configuration. In the simplest
5710 * case it just creates unmodified original flow.
5712 ret = flow_create_split_outer(dev, flow, &attr_factor,
5713 buf->entry[i].pattern,
5714 p_actions_rx, external, idx,
5718 if (is_flow_tunnel_steer_rule(dev, attr,
5719 buf->entry[i].pattern,
5721 ret = flow_tunnel_add_default_miss(dev, flow, attr,
5727 mlx5_free(default_miss_ctx.queue);
5732 /* Create the tx flow. */
5734 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5735 attr_tx.ingress = 0;
5737 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5738 actions_hairpin_tx.actions,
5742 dev_flow->flow = flow;
5743 dev_flow->external = 0;
5744 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5745 dev_flow->handle, next);
5746 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5748 actions_hairpin_tx.actions, error);
5753 * Update the metadata register copy table. If extensive
5754 * metadata feature is enabled and registers are supported
5755 * we might create the extra rte_flow for each unique
5756 * MARK/FLAG action ID.
5758 * The table is updated for ingress Flows only, because
5759 * the egress Flows belong to the different device and
5760 * copy table should be updated in peer NIC Rx domain.
5762 if (attr_factor.ingress &&
5763 (external || attr_factor.group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5764 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5769 * If the flow is external (from application) OR device is started, then
5770 * the flow will be applied immediately.
5772 if (external || dev->data->dev_started) {
5773 ret = flow_drv_apply(dev, flow, error);
5778 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5780 flow_rxq_flags_set(dev, flow);
5781 rte_free(translated_actions);
5782 /* Nested flow creation index recovery. */
5783 wks->flow_idx = wks->flow_nested_idx;
5784 if (wks->flow_nested_idx)
5785 wks->flow_nested_idx = 0;
5786 tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5789 flow->tunnel_id = tunnel->tunnel_id;
5790 __atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5791 mlx5_free(default_miss_ctx.queue);
5796 ret = rte_errno; /* Save rte_errno before cleanup. */
5797 flow_mreg_del_copy_action(dev, flow);
5798 flow_drv_destroy(dev, flow);
5799 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5800 rte_errno = ret; /* Restore rte_errno. */
5803 wks->flow_idx = wks->flow_nested_idx;
5804 if (wks->flow_nested_idx)
5805 wks->flow_nested_idx = 0;
5806 error_before_hairpin_split:
5807 rte_free(translated_actions);
5812 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5813 * incoming packets to table 1.
5815 * Other flow rules, requested for group n, will be created in
5816 * e-switch table n+1.
5817 * Jump action to e-switch group n will be created to group n+1.
5819 * Used when working in switchdev mode, to utilise advantages of table 1
5823 * Pointer to Ethernet device.
5826 * Pointer to flow on success, NULL otherwise and rte_errno is set.
5829 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5831 const struct rte_flow_attr attr = {
5838 const struct rte_flow_item pattern = {
5839 .type = RTE_FLOW_ITEM_TYPE_END,
5841 struct rte_flow_action_jump jump = {
5844 const struct rte_flow_action actions[] = {
5846 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5850 .type = RTE_FLOW_ACTION_TYPE_END,
5853 struct mlx5_priv *priv = dev->data->dev_private;
5854 struct rte_flow_error error;
5856 return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5858 actions, false, &error);
5862 * Validate a flow supported by the NIC.
5864 * @see rte_flow_validate()
5868 mlx5_flow_validate(struct rte_eth_dev *dev,
5869 const struct rte_flow_attr *attr,
5870 const struct rte_flow_item items[],
5871 const struct rte_flow_action original_actions[],
5872 struct rte_flow_error *error)
5875 struct mlx5_translated_shared_action
5876 shared_actions[MLX5_MAX_SHARED_ACTIONS];
5877 int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5878 const struct rte_flow_action *actions;
5879 struct rte_flow_action *translated_actions = NULL;
5880 int ret = flow_shared_actions_translate(original_actions,
5883 &translated_actions, error);
5887 actions = translated_actions ? translated_actions : original_actions;
5888 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5889 ret = flow_drv_validate(dev, attr, items, actions,
5890 true, hairpin_flow, error);
5891 rte_free(translated_actions);
5898 * @see rte_flow_create()
5902 mlx5_flow_create(struct rte_eth_dev *dev,
5903 const struct rte_flow_attr *attr,
5904 const struct rte_flow_item items[],
5905 const struct rte_flow_action actions[],
5906 struct rte_flow_error *error)
5908 struct mlx5_priv *priv = dev->data->dev_private;
5911 * If the device is not started yet, it is not allowed to created a
5912 * flow from application. PMD default flows and traffic control flows
5915 if (unlikely(!dev->data->dev_started)) {
5916 DRV_LOG(DEBUG, "port %u is not started when "
5917 "inserting a flow", dev->data->port_id);
5918 rte_flow_error_set(error, ENODEV,
5919 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5921 "port not started");
5925 return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5926 attr, items, actions, true, error);
5930 * Destroy a flow in a list.
5933 * Pointer to Ethernet device.
5935 * Pointer to the Indexed flow list. If this parameter NULL,
5936 * there is no flow removal from the list. Be noted that as
5937 * flow is add to the indexed list, memory of the indexed
5938 * list points to maybe changed as flow destroyed.
5939 * @param[in] flow_idx
5940 * Index of flow to destroy.
5943 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5946 struct mlx5_priv *priv = dev->data->dev_private;
5947 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5948 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5949 [MLX5_IPOOL_RTE_FLOW], flow_idx);
5954 * Update RX queue flags only if port is started, otherwise it is
5957 if (dev->data->dev_started)
5958 flow_rxq_flags_trim(dev, flow);
5959 flow_drv_destroy(dev, flow);
5961 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5962 flow_idx, flow, next);
5963 flow_mreg_del_copy_action(dev, flow);
5965 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5966 if (priv_fdir_flow->rix_flow == flow_idx)
5969 if (priv_fdir_flow) {
5970 LIST_REMOVE(priv_fdir_flow, next);
5971 mlx5_free(priv_fdir_flow->fdir);
5972 mlx5_free(priv_fdir_flow);
5975 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5977 struct mlx5_flow_tunnel *tunnel;
5978 tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5980 if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5981 mlx5_flow_tunnel_free(dev, tunnel);
5986 * Destroy all flows.
5989 * Pointer to Ethernet device.
5991 * Pointer to the Indexed flow list.
5993 * If flushing is called avtively.
5996 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5998 uint32_t num_flushed = 0;
6001 flow_list_destroy(dev, list, *list);
6005 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6006 dev->data->port_id, num_flushed);
6014 * Pointer to Ethernet device.
6016 * Pointer to the Indexed flow list.
6019 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
6021 struct mlx5_priv *priv = dev->data->dev_private;
6022 struct rte_flow *flow = NULL;
6025 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
6027 flow_drv_remove(dev, flow);
6028 flow_mreg_stop_copy_action(dev, flow);
6030 flow_mreg_del_default_copy_action(dev);
6031 flow_rxq_flags_clear(dev);
6038 * Pointer to Ethernet device.
6040 * Pointer to the Indexed flow list.
6043 * 0 on success, a negative errno value otherwise and rte_errno is set.
6046 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
6048 struct mlx5_priv *priv = dev->data->dev_private;
6049 struct rte_flow *flow = NULL;
6050 struct rte_flow_error error;
6054 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6055 ret = flow_mreg_add_default_copy_action(dev, &error);
6058 /* Apply Flows created by application. */
6059 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
6061 ret = flow_mreg_start_copy_action(dev, flow);
6064 ret = flow_drv_apply(dev, flow, &error);
6067 flow_rxq_flags_set(dev, flow);
6071 ret = rte_errno; /* Save rte_errno before cleanup. */
6072 mlx5_flow_stop(dev, list);
6073 rte_errno = ret; /* Restore rte_errno. */
6078 * Stop all default actions for flows.
6081 * Pointer to Ethernet device.
6084 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6086 flow_mreg_del_default_copy_action(dev);
6087 flow_rxq_flags_clear(dev);
6091 * Start all default actions for flows.
6094 * Pointer to Ethernet device.
6096 * 0 on success, a negative errno value otherwise and rte_errno is set.
6099 mlx5_flow_start_default(struct rte_eth_dev *dev)
6101 struct rte_flow_error error;
6103 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6104 return flow_mreg_add_default_copy_action(dev, &error);
6108 * Release key of thread specific flow workspace data.
6111 flow_release_workspace(void *data)
6113 struct mlx5_flow_workspace *wks = data;
6117 free(wks->rss_desc[0].queue);
6118 free(wks->rss_desc[1].queue);
6123 * Initialize key of thread specific flow workspace data.
6126 flow_alloc_workspace(void)
6128 if (pthread_key_create(&key_workspace, flow_release_workspace))
6129 DRV_LOG(ERR, "Can't create flow workspace data thread key.");
6133 * Get thread specific flow workspace.
6135 * @return pointer to thread specific flowworkspace data, NULL on error.
6137 struct mlx5_flow_workspace*
6138 mlx5_flow_get_thread_workspace(void)
6140 struct mlx5_flow_workspace *data;
6142 if (pthread_once(&key_workspace_init, flow_alloc_workspace)) {
6143 DRV_LOG(ERR, "Failed to init flow workspace data thread key.");
6146 data = pthread_getspecific(key_workspace);
6148 data = calloc(1, sizeof(*data));
6150 DRV_LOG(ERR, "Failed to allocate flow workspace "
6154 data->rss_desc[0].queue = calloc(1,
6155 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6156 if (!data->rss_desc[0].queue)
6158 data->rss_desc[1].queue = calloc(1,
6159 sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6160 if (!data->rss_desc[1].queue)
6162 data->rssq_num[0] = MLX5_RSSQ_DEFAULT_NUM;
6163 data->rssq_num[1] = MLX5_RSSQ_DEFAULT_NUM;
6164 if (pthread_setspecific(key_workspace, data)) {
6165 DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6171 if (data->rss_desc[0].queue)
6172 free(data->rss_desc[0].queue);
6173 if (data->rss_desc[1].queue)
6174 free(data->rss_desc[1].queue);
6180 * Verify the flow list is empty
6183 * Pointer to Ethernet device.
6185 * @return the number of flows not released.
6188 mlx5_flow_verify(struct rte_eth_dev *dev)
6190 struct mlx5_priv *priv = dev->data->dev_private;
6191 struct rte_flow *flow;
6195 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6197 DRV_LOG(DEBUG, "port %u flow %p still referenced",
6198 dev->data->port_id, (void *)flow);
6205 * Enable default hairpin egress flow.
6208 * Pointer to Ethernet device.
6213 * 0 on success, a negative errno value otherwise and rte_errno is set.
6216 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6219 struct mlx5_priv *priv = dev->data->dev_private;
6220 const struct rte_flow_attr attr = {
6224 struct mlx5_rte_flow_item_tx_queue queue_spec = {
6227 struct mlx5_rte_flow_item_tx_queue queue_mask = {
6228 .queue = UINT32_MAX,
6230 struct rte_flow_item items[] = {
6232 .type = (enum rte_flow_item_type)
6233 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6234 .spec = &queue_spec,
6236 .mask = &queue_mask,
6239 .type = RTE_FLOW_ITEM_TYPE_END,
6242 struct rte_flow_action_jump jump = {
6243 .group = MLX5_HAIRPIN_TX_TABLE,
6245 struct rte_flow_action actions[2];
6247 struct rte_flow_error error;
6249 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6250 actions[0].conf = &jump;
6251 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6252 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6253 &attr, items, actions, false, &error);
6256 "Failed to create ctrl flow: rte_errno(%d),"
6257 " type(%d), message(%s)",
6258 rte_errno, error.type,
6259 error.message ? error.message : " (no stated reason)");
6266 * Enable a control flow configured from the control plane.
6269 * Pointer to Ethernet device.
6271 * An Ethernet flow spec to apply.
6273 * An Ethernet flow mask to apply.
6275 * A VLAN flow spec to apply.
6277 * A VLAN flow mask to apply.
6280 * 0 on success, a negative errno value otherwise and rte_errno is set.
6283 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6284 struct rte_flow_item_eth *eth_spec,
6285 struct rte_flow_item_eth *eth_mask,
6286 struct rte_flow_item_vlan *vlan_spec,
6287 struct rte_flow_item_vlan *vlan_mask)
6289 struct mlx5_priv *priv = dev->data->dev_private;
6290 const struct rte_flow_attr attr = {
6292 .priority = MLX5_FLOW_PRIO_RSVD,
6294 struct rte_flow_item items[] = {
6296 .type = RTE_FLOW_ITEM_TYPE_ETH,
6302 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6303 RTE_FLOW_ITEM_TYPE_END,
6309 .type = RTE_FLOW_ITEM_TYPE_END,
6312 uint16_t queue[priv->reta_idx_n];
6313 struct rte_flow_action_rss action_rss = {
6314 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6316 .types = priv->rss_conf.rss_hf,
6317 .key_len = priv->rss_conf.rss_key_len,
6318 .queue_num = priv->reta_idx_n,
6319 .key = priv->rss_conf.rss_key,
6322 struct rte_flow_action actions[] = {
6324 .type = RTE_FLOW_ACTION_TYPE_RSS,
6325 .conf = &action_rss,
6328 .type = RTE_FLOW_ACTION_TYPE_END,
6332 struct rte_flow_error error;
6335 if (!priv->reta_idx_n || !priv->rxqs_n) {
6338 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6339 action_rss.types = 0;
6340 for (i = 0; i != priv->reta_idx_n; ++i)
6341 queue[i] = (*priv->reta_idx)[i];
6342 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6343 &attr, items, actions, false, &error);
6350 * Enable a flow control configured from the control plane.
6353 * Pointer to Ethernet device.
6355 * An Ethernet flow spec to apply.
6357 * An Ethernet flow mask to apply.
6360 * 0 on success, a negative errno value otherwise and rte_errno is set.
6363 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6364 struct rte_flow_item_eth *eth_spec,
6365 struct rte_flow_item_eth *eth_mask)
6367 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6371 * Create default miss flow rule matching lacp traffic
6374 * Pointer to Ethernet device.
6376 * An Ethernet flow spec to apply.
6379 * 0 on success, a negative errno value otherwise and rte_errno is set.
6382 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6384 struct mlx5_priv *priv = dev->data->dev_private;
6386 * The LACP matching is done by only using ether type since using
6387 * a multicast dst mac causes kernel to give low priority to this flow.
6389 static const struct rte_flow_item_eth lacp_spec = {
6390 .type = RTE_BE16(0x8809),
6392 static const struct rte_flow_item_eth lacp_mask = {
6395 const struct rte_flow_attr attr = {
6398 struct rte_flow_item items[] = {
6400 .type = RTE_FLOW_ITEM_TYPE_ETH,
6405 .type = RTE_FLOW_ITEM_TYPE_END,
6408 struct rte_flow_action actions[] = {
6410 .type = (enum rte_flow_action_type)
6411 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6414 .type = RTE_FLOW_ACTION_TYPE_END,
6417 struct rte_flow_error error;
6418 uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6419 &attr, items, actions, false, &error);
6429 * @see rte_flow_destroy()
6433 mlx5_flow_destroy(struct rte_eth_dev *dev,
6434 struct rte_flow *flow,
6435 struct rte_flow_error *error __rte_unused)
6437 struct mlx5_priv *priv = dev->data->dev_private;
6439 flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6444 * Destroy all flows.
6446 * @see rte_flow_flush()
6450 mlx5_flow_flush(struct rte_eth_dev *dev,
6451 struct rte_flow_error *error __rte_unused)
6453 struct mlx5_priv *priv = dev->data->dev_private;
6455 mlx5_flow_list_flush(dev, &priv->flows, false);
6462 * @see rte_flow_isolate()
6466 mlx5_flow_isolate(struct rte_eth_dev *dev,
6468 struct rte_flow_error *error)
6470 struct mlx5_priv *priv = dev->data->dev_private;
6472 if (dev->data->dev_started) {
6473 rte_flow_error_set(error, EBUSY,
6474 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6476 "port must be stopped first");
6479 priv->isolated = !!enable;
6481 dev->dev_ops = &mlx5_os_dev_ops_isolate;
6483 dev->dev_ops = &mlx5_os_dev_ops;
6485 dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6486 dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6494 * @see rte_flow_query()
6498 flow_drv_query(struct rte_eth_dev *dev,
6500 const struct rte_flow_action *actions,
6502 struct rte_flow_error *error)
6504 struct mlx5_priv *priv = dev->data->dev_private;
6505 const struct mlx5_flow_driver_ops *fops;
6506 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6507 [MLX5_IPOOL_RTE_FLOW],
6509 enum mlx5_flow_drv_type ftype;
6512 return rte_flow_error_set(error, ENOENT,
6513 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6515 "invalid flow handle");
6517 ftype = flow->drv_type;
6518 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6519 fops = flow_get_drv_ops(ftype);
6521 return fops->query(dev, flow, actions, data, error);
6527 * @see rte_flow_query()
6531 mlx5_flow_query(struct rte_eth_dev *dev,
6532 struct rte_flow *flow,
6533 const struct rte_flow_action *actions,
6535 struct rte_flow_error *error)
6539 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6547 * Convert a flow director filter to a generic flow.
6550 * Pointer to Ethernet device.
6551 * @param fdir_filter
6552 * Flow director filter to add.
6554 * Generic flow parameters structure.
6557 * 0 on success, a negative errno value otherwise and rte_errno is set.
6560 flow_fdir_filter_convert(struct rte_eth_dev *dev,
6561 const struct rte_eth_fdir_filter *fdir_filter,
6562 struct mlx5_fdir *attributes)
6564 struct mlx5_priv *priv = dev->data->dev_private;
6565 const struct rte_eth_fdir_input *input = &fdir_filter->input;
6566 const struct rte_eth_fdir_masks *mask =
6567 &dev->data->dev_conf.fdir_conf.mask;
6569 /* Validate queue number. */
6570 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
6571 DRV_LOG(ERR, "port %u invalid queue number %d",
6572 dev->data->port_id, fdir_filter->action.rx_queue);
6576 attributes->attr.ingress = 1;
6577 attributes->items[0] = (struct rte_flow_item) {
6578 .type = RTE_FLOW_ITEM_TYPE_ETH,
6579 .spec = &attributes->l2,
6580 .mask = &attributes->l2_mask,
6582 switch (fdir_filter->action.behavior) {
6583 case RTE_ETH_FDIR_ACCEPT:
6584 attributes->actions[0] = (struct rte_flow_action){
6585 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
6586 .conf = &attributes->queue,
6589 case RTE_ETH_FDIR_REJECT:
6590 attributes->actions[0] = (struct rte_flow_action){
6591 .type = RTE_FLOW_ACTION_TYPE_DROP,
6595 DRV_LOG(ERR, "port %u invalid behavior %d",
6597 fdir_filter->action.behavior);
6598 rte_errno = ENOTSUP;
6601 attributes->queue.index = fdir_filter->action.rx_queue;
6603 switch (fdir_filter->input.flow_type) {
6604 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
6605 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
6606 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
6607 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
6608 .src_addr = input->flow.ip4_flow.src_ip,
6609 .dst_addr = input->flow.ip4_flow.dst_ip,
6610 .time_to_live = input->flow.ip4_flow.ttl,
6611 .type_of_service = input->flow.ip4_flow.tos,
6613 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
6614 .src_addr = mask->ipv4_mask.src_ip,
6615 .dst_addr = mask->ipv4_mask.dst_ip,
6616 .time_to_live = mask->ipv4_mask.ttl,
6617 .type_of_service = mask->ipv4_mask.tos,
6618 .next_proto_id = mask->ipv4_mask.proto,
6620 attributes->items[1] = (struct rte_flow_item){
6621 .type = RTE_FLOW_ITEM_TYPE_IPV4,
6622 .spec = &attributes->l3,
6623 .mask = &attributes->l3_mask,
6626 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
6627 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
6628 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
6629 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
6630 .hop_limits = input->flow.ipv6_flow.hop_limits,
6631 .proto = input->flow.ipv6_flow.proto,
6634 memcpy(attributes->l3.ipv6.hdr.src_addr,
6635 input->flow.ipv6_flow.src_ip,
6636 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
6637 memcpy(attributes->l3.ipv6.hdr.dst_addr,
6638 input->flow.ipv6_flow.dst_ip,
6639 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
6640 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
6641 mask->ipv6_mask.src_ip,
6642 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
6643 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
6644 mask->ipv6_mask.dst_ip,
6645 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
6646 attributes->items[1] = (struct rte_flow_item){
6647 .type = RTE_FLOW_ITEM_TYPE_IPV6,
6648 .spec = &attributes->l3,
6649 .mask = &attributes->l3_mask,
6653 DRV_LOG(ERR, "port %u invalid flow type%d",
6654 dev->data->port_id, fdir_filter->input.flow_type);
6655 rte_errno = ENOTSUP;
6659 switch (fdir_filter->input.flow_type) {
6660 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
6661 attributes->l4.udp.hdr = (struct rte_udp_hdr){
6662 .src_port = input->flow.udp4_flow.src_port,
6663 .dst_port = input->flow.udp4_flow.dst_port,
6665 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
6666 .src_port = mask->src_port_mask,
6667 .dst_port = mask->dst_port_mask,
6669 attributes->items[2] = (struct rte_flow_item){
6670 .type = RTE_FLOW_ITEM_TYPE_UDP,
6671 .spec = &attributes->l4,
6672 .mask = &attributes->l4_mask,
6675 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
6676 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
6677 .src_port = input->flow.tcp4_flow.src_port,
6678 .dst_port = input->flow.tcp4_flow.dst_port,
6680 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
6681 .src_port = mask->src_port_mask,
6682 .dst_port = mask->dst_port_mask,
6684 attributes->items[2] = (struct rte_flow_item){
6685 .type = RTE_FLOW_ITEM_TYPE_TCP,
6686 .spec = &attributes->l4,
6687 .mask = &attributes->l4_mask,
6690 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
6691 attributes->l4.udp.hdr = (struct rte_udp_hdr){
6692 .src_port = input->flow.udp6_flow.src_port,
6693 .dst_port = input->flow.udp6_flow.dst_port,
6695 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
6696 .src_port = mask->src_port_mask,
6697 .dst_port = mask->dst_port_mask,
6699 attributes->items[2] = (struct rte_flow_item){
6700 .type = RTE_FLOW_ITEM_TYPE_UDP,
6701 .spec = &attributes->l4,
6702 .mask = &attributes->l4_mask,
6705 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
6706 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
6707 .src_port = input->flow.tcp6_flow.src_port,
6708 .dst_port = input->flow.tcp6_flow.dst_port,
6710 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
6711 .src_port = mask->src_port_mask,
6712 .dst_port = mask->dst_port_mask,
6714 attributes->items[2] = (struct rte_flow_item){
6715 .type = RTE_FLOW_ITEM_TYPE_TCP,
6716 .spec = &attributes->l4,
6717 .mask = &attributes->l4_mask,
6720 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
6721 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
6724 DRV_LOG(ERR, "port %u invalid flow type%d",
6725 dev->data->port_id, fdir_filter->input.flow_type);
6726 rte_errno = ENOTSUP;
6732 #define FLOW_FDIR_CMP(f1, f2, fld) \
6733 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
6736 * Compare two FDIR flows. If items and actions are identical, the two flows are
6740 * Pointer to Ethernet device.
6742 * FDIR flow to compare.
6744 * FDIR flow to compare.
6747 * Zero on match, 1 otherwise.
6750 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
6752 if (FLOW_FDIR_CMP(f1, f2, attr) ||
6753 FLOW_FDIR_CMP(f1, f2, l2) ||
6754 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
6755 FLOW_FDIR_CMP(f1, f2, l3) ||
6756 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
6757 FLOW_FDIR_CMP(f1, f2, l4) ||
6758 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
6759 FLOW_FDIR_CMP(f1, f2, actions[0].type))
6761 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
6762 FLOW_FDIR_CMP(f1, f2, queue))
6768 * Search device flow list to find out a matched FDIR flow.
6771 * Pointer to Ethernet device.
6773 * FDIR flow to lookup.
6776 * Index of flow if found, 0 otherwise.
6779 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
6781 struct mlx5_priv *priv = dev->data->dev_private;
6782 uint32_t flow_idx = 0;
6783 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6785 MLX5_ASSERT(fdir_flow);
6786 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
6787 if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
6788 DRV_LOG(DEBUG, "port %u found FDIR flow %u",
6789 dev->data->port_id, flow_idx);
6790 flow_idx = priv_fdir_flow->rix_flow;
6798 * Add new flow director filter and store it in list.
6801 * Pointer to Ethernet device.
6802 * @param fdir_filter
6803 * Flow director filter to add.
6806 * 0 on success, a negative errno value otherwise and rte_errno is set.
6809 flow_fdir_filter_add(struct rte_eth_dev *dev,
6810 const struct rte_eth_fdir_filter *fdir_filter)
6812 struct mlx5_priv *priv = dev->data->dev_private;
6813 struct mlx5_fdir *fdir_flow;
6814 struct rte_flow *flow;
6815 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6819 fdir_flow = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*fdir_flow), 0,
6825 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
6828 flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
6833 priv_fdir_flow = mlx5_malloc(MLX5_MEM_ZERO,
6834 sizeof(struct mlx5_fdir_flow),
6836 if (!priv_fdir_flow) {
6840 flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
6841 fdir_flow->items, fdir_flow->actions, true,
6843 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6847 priv_fdir_flow->fdir = fdir_flow;
6848 priv_fdir_flow->rix_flow = flow_idx;
6849 LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
6850 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
6851 dev->data->port_id, (void *)flow);
6854 mlx5_free(priv_fdir_flow);
6855 mlx5_free(fdir_flow);
6860 * Delete specific filter.
6863 * Pointer to Ethernet device.
6864 * @param fdir_filter
6865 * Filter to be deleted.
6868 * 0 on success, a negative errno value otherwise and rte_errno is set.
6871 flow_fdir_filter_delete(struct rte_eth_dev *dev,
6872 const struct rte_eth_fdir_filter *fdir_filter)
6874 struct mlx5_priv *priv = dev->data->dev_private;
6876 struct mlx5_fdir fdir_flow = {
6879 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6882 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
6885 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
6886 /* Find the fdir in priv list */
6887 if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
6890 if (!priv_fdir_flow)
6892 LIST_REMOVE(priv_fdir_flow, next);
6893 flow_idx = priv_fdir_flow->rix_flow;
6894 flow_list_destroy(dev, &priv->flows, flow_idx);
6895 mlx5_free(priv_fdir_flow->fdir);
6896 mlx5_free(priv_fdir_flow);
6897 DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
6898 dev->data->port_id, flow_idx);
6903 * Update queue for specific filter.
6906 * Pointer to Ethernet device.
6907 * @param fdir_filter
6908 * Filter to be updated.
6911 * 0 on success, a negative errno value otherwise and rte_errno is set.
6914 flow_fdir_filter_update(struct rte_eth_dev *dev,
6915 const struct rte_eth_fdir_filter *fdir_filter)
6919 ret = flow_fdir_filter_delete(dev, fdir_filter);
6922 return flow_fdir_filter_add(dev, fdir_filter);
6926 * Flush all filters.
6929 * Pointer to Ethernet device.
6932 flow_fdir_filter_flush(struct rte_eth_dev *dev)
6934 struct mlx5_priv *priv = dev->data->dev_private;
6935 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6937 while (!LIST_EMPTY(&priv->fdir_flows)) {
6938 priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
6939 LIST_REMOVE(priv_fdir_flow, next);
6940 flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
6941 mlx5_free(priv_fdir_flow->fdir);
6942 mlx5_free(priv_fdir_flow);
6947 * Get flow director information.
6950 * Pointer to Ethernet device.
6951 * @param[out] fdir_info
6952 * Resulting flow director information.
6955 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
6957 struct rte_eth_fdir_masks *mask =
6958 &dev->data->dev_conf.fdir_conf.mask;
6960 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
6961 fdir_info->guarant_spc = 0;
6962 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
6963 fdir_info->max_flexpayload = 0;
6964 fdir_info->flow_types_mask[0] = 0;
6965 fdir_info->flex_payload_unit = 0;
6966 fdir_info->max_flex_payload_segment_num = 0;
6967 fdir_info->flex_payload_limit = 0;
6968 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
6972 * Deal with flow director operations.
6975 * Pointer to Ethernet device.
6977 * Operation to perform.
6979 * Pointer to operation-specific structure.
6982 * 0 on success, a negative errno value otherwise and rte_errno is set.
6985 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
6988 enum rte_fdir_mode fdir_mode =
6989 dev->data->dev_conf.fdir_conf.mode;
6991 if (filter_op == RTE_ETH_FILTER_NOP)
6993 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
6994 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
6995 DRV_LOG(ERR, "port %u flow director mode %d not supported",
6996 dev->data->port_id, fdir_mode);
7000 switch (filter_op) {
7001 case RTE_ETH_FILTER_ADD:
7002 return flow_fdir_filter_add(dev, arg);
7003 case RTE_ETH_FILTER_UPDATE:
7004 return flow_fdir_filter_update(dev, arg);
7005 case RTE_ETH_FILTER_DELETE:
7006 return flow_fdir_filter_delete(dev, arg);
7007 case RTE_ETH_FILTER_FLUSH:
7008 flow_fdir_filter_flush(dev);
7010 case RTE_ETH_FILTER_INFO:
7011 flow_fdir_info_get(dev, arg);
7014 DRV_LOG(DEBUG, "port %u unknown operation %u",
7015 dev->data->port_id, filter_op);
7023 * Manage filter operations.
7026 * Pointer to Ethernet device structure.
7027 * @param filter_type
7030 * Operation to perform.
7032 * Pointer to operation-specific structure.
7035 * 0 on success, a negative errno value otherwise and rte_errno is set.
7038 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
7039 enum rte_filter_type filter_type,
7040 enum rte_filter_op filter_op,
7043 switch (filter_type) {
7044 case RTE_ETH_FILTER_GENERIC:
7045 if (filter_op != RTE_ETH_FILTER_GET) {
7049 *(const void **)arg = &mlx5_flow_ops;
7051 case RTE_ETH_FILTER_FDIR:
7052 return flow_fdir_ctrl_func(dev, filter_op, arg);
7054 DRV_LOG(ERR, "port %u filter type (%d) not supported",
7055 dev->data->port_id, filter_type);
7056 rte_errno = ENOTSUP;
7063 * Create the needed meter and suffix tables.
7066 * Pointer to Ethernet device.
7068 * Pointer to the flow meter.
7071 * Pointer to table set on success, NULL otherwise.
7073 struct mlx5_meter_domains_infos *
7074 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7075 const struct mlx5_flow_meter *fm)
7077 const struct mlx5_flow_driver_ops *fops;
7079 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7080 return fops->create_mtr_tbls(dev, fm);
7084 * Destroy the meter table set.
7087 * Pointer to Ethernet device.
7089 * Pointer to the meter table set.
7095 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7096 struct mlx5_meter_domains_infos *tbls)
7098 const struct mlx5_flow_driver_ops *fops;
7100 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7101 return fops->destroy_mtr_tbls(dev, tbls);
7105 * Create policer rules.
7108 * Pointer to Ethernet device.
7110 * Pointer to flow meter structure.
7112 * Pointer to flow attributes.
7115 * 0 on success, -1 otherwise.
7118 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
7119 struct mlx5_flow_meter *fm,
7120 const struct rte_flow_attr *attr)
7122 const struct mlx5_flow_driver_ops *fops;
7124 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7125 return fops->create_policer_rules(dev, fm, attr);
7129 * Destroy policer rules.
7132 * Pointer to flow meter structure.
7134 * Pointer to flow attributes.
7137 * 0 on success, -1 otherwise.
7140 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
7141 struct mlx5_flow_meter *fm,
7142 const struct rte_flow_attr *attr)
7144 const struct mlx5_flow_driver_ops *fops;
7146 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7147 return fops->destroy_policer_rules(dev, fm, attr);
7151 * Allocate a counter.
7154 * Pointer to Ethernet device structure.
7157 * Index to allocated counter on success, 0 otherwise.
7160 mlx5_counter_alloc(struct rte_eth_dev *dev)
7162 const struct mlx5_flow_driver_ops *fops;
7163 struct rte_flow_attr attr = { .transfer = 0 };
7165 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7166 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7167 return fops->counter_alloc(dev);
7170 "port %u counter allocate is not supported.",
7171 dev->data->port_id);
7179 * Pointer to Ethernet device structure.
7181 * Index to counter to be free.
7184 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7186 const struct mlx5_flow_driver_ops *fops;
7187 struct rte_flow_attr attr = { .transfer = 0 };
7189 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7190 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7191 fops->counter_free(dev, cnt);
7195 "port %u counter free is not supported.",
7196 dev->data->port_id);
7200 * Query counter statistics.
7203 * Pointer to Ethernet device structure.
7205 * Index to counter to query.
7207 * Set to clear counter statistics.
7209 * The counter hits packets number to save.
7211 * The counter hits bytes number to save.
7214 * 0 on success, a negative errno value otherwise.
7217 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7218 bool clear, uint64_t *pkts, uint64_t *bytes)
7220 const struct mlx5_flow_driver_ops *fops;
7221 struct rte_flow_attr attr = { .transfer = 0 };
7223 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7224 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7225 return fops->counter_query(dev, cnt, clear, pkts, bytes);
7228 "port %u counter query is not supported.",
7229 dev->data->port_id);
7234 * Allocate a new memory for the counter values wrapped by all the needed
7238 * Pointer to mlx5_dev_ctx_shared object.
7241 * 0 on success, a negative errno value otherwise.
7244 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7246 struct mlx5_devx_mkey_attr mkey_attr;
7247 struct mlx5_counter_stats_mem_mng *mem_mng;
7248 volatile struct flow_counter_stats *raw_data;
7249 int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7250 int size = (sizeof(struct flow_counter_stats) *
7251 MLX5_COUNTERS_PER_POOL +
7252 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7253 sizeof(struct mlx5_counter_stats_mem_mng);
7254 size_t pgsize = rte_mem_page_size();
7258 if (pgsize == (size_t)-1) {
7259 DRV_LOG(ERR, "Failed to get mem page size");
7263 mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7268 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7269 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7270 mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
7271 IBV_ACCESS_LOCAL_WRITE);
7272 if (!mem_mng->umem) {
7277 mkey_attr.addr = (uintptr_t)mem;
7278 mkey_attr.size = size;
7279 mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7280 mkey_attr.pd = sh->pdn;
7281 mkey_attr.log_entity_size = 0;
7282 mkey_attr.pg_access = 0;
7283 mkey_attr.klm_array = NULL;
7284 mkey_attr.klm_num = 0;
7285 mkey_attr.relaxed_ordering = sh->cmng.relaxed_ordering;
7286 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7288 mlx5_glue->devx_umem_dereg(mem_mng->umem);
7293 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7294 raw_data = (volatile struct flow_counter_stats *)mem;
7295 for (i = 0; i < raws_n; ++i) {
7296 mem_mng->raws[i].mem_mng = mem_mng;
7297 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7299 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7300 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7301 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7303 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7304 sh->cmng.mem_mng = mem_mng;
7309 * Set the statistic memory to the new counter pool.
7312 * Pointer to mlx5_dev_ctx_shared object.
7314 * Pointer to the pool to set the statistic memory.
7317 * 0 on success, a negative errno value otherwise.
7320 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7321 struct mlx5_flow_counter_pool *pool)
7323 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7324 /* Resize statistic memory once used out. */
7325 if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7326 mlx5_flow_create_counter_stat_mem_mng(sh)) {
7327 DRV_LOG(ERR, "Cannot resize counter stat mem.");
7330 rte_spinlock_lock(&pool->sl);
7331 pool->raw = cmng->mem_mng->raws + pool->index %
7332 MLX5_CNT_CONTAINER_RESIZE;
7333 rte_spinlock_unlock(&pool->sl);
7334 pool->raw_hw = NULL;
7338 #define MLX5_POOL_QUERY_FREQ_US 1000000
7341 * Set the periodic procedure for triggering asynchronous batch queries for all
7342 * the counter pools.
7345 * Pointer to mlx5_dev_ctx_shared object.
7348 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7350 uint32_t pools_n, us;
7352 pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7353 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7354 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7355 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7356 sh->cmng.query_thread_on = 0;
7357 DRV_LOG(ERR, "Cannot reinitialize query alarm");
7359 sh->cmng.query_thread_on = 1;
7364 * The periodic procedure for triggering asynchronous batch queries for all the
7365 * counter pools. This function is probably called by the host thread.
7368 * The parameter for the alarm process.
7371 mlx5_flow_query_alarm(void *arg)
7373 struct mlx5_dev_ctx_shared *sh = arg;
7375 uint16_t pool_index = sh->cmng.pool_index;
7376 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7377 struct mlx5_flow_counter_pool *pool;
7380 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7382 rte_spinlock_lock(&cmng->pool_update_sl);
7383 pool = cmng->pools[pool_index];
7384 n_valid = cmng->n_valid;
7385 rte_spinlock_unlock(&cmng->pool_update_sl);
7386 /* Set the statistic memory to the new created pool. */
7387 if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7390 /* There is a pool query in progress. */
7393 LIST_FIRST(&sh->cmng.free_stat_raws);
7395 /* No free counter statistics raw memory. */
7398 * Identify the counters released between query trigger and query
7399 * handle more efficiently. The counter released in this gap period
7400 * should wait for a new round of query as the new arrived packets
7401 * will not be taken into account.
7404 ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7405 MLX5_COUNTERS_PER_POOL,
7407 pool->raw_hw->mem_mng->dm->id,
7411 (uint64_t)(uintptr_t)pool);
7413 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7414 " %d", pool->min_dcs->id);
7415 pool->raw_hw = NULL;
7418 LIST_REMOVE(pool->raw_hw, next);
7419 sh->cmng.pending_queries++;
7421 if (pool_index >= n_valid)
7424 sh->cmng.pool_index = pool_index;
7425 mlx5_set_query_alarm(sh);
7429 * Check and callback event for new aged flow in the counter pool
7432 * Pointer to mlx5_dev_ctx_shared object.
7434 * Pointer to Current counter pool.
7437 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7438 struct mlx5_flow_counter_pool *pool)
7440 struct mlx5_priv *priv;
7441 struct mlx5_flow_counter *cnt;
7442 struct mlx5_age_info *age_info;
7443 struct mlx5_age_param *age_param;
7444 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7445 struct mlx5_counter_stats_raw *prev = pool->raw;
7446 const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7447 const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7448 uint16_t expected = AGE_CANDIDATE;
7451 pool->time_of_last_age_check = curr_time;
7452 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7453 cnt = MLX5_POOL_GET_CNT(pool, i);
7454 age_param = MLX5_CNT_TO_AGE(cnt);
7455 if (__atomic_load_n(&age_param->state,
7456 __ATOMIC_RELAXED) != AGE_CANDIDATE)
7458 if (cur->data[i].hits != prev->data[i].hits) {
7459 __atomic_store_n(&age_param->sec_since_last_hit, 0,
7463 if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7465 __ATOMIC_RELAXED) <= age_param->timeout)
7468 * Hold the lock first, or if between the
7469 * state AGE_TMOUT and tailq operation the
7470 * release happened, the release procedure
7471 * may delete a non-existent tailq node.
7473 priv = rte_eth_devices[age_param->port_id].data->dev_private;
7474 age_info = GET_PORT_AGE_INFO(priv);
7475 rte_spinlock_lock(&age_info->aged_sl);
7476 if (__atomic_compare_exchange_n(&age_param->state, &expected,
7479 __ATOMIC_RELAXED)) {
7480 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7481 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7483 rte_spinlock_unlock(&age_info->aged_sl);
7485 for (i = 0; i < sh->max_port; i++) {
7486 age_info = &sh->port[i].age_info;
7487 if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
7489 if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
7490 rte_eth_dev_callback_process
7491 (&rte_eth_devices[sh->port[i].devx_ih_port_id],
7492 RTE_ETH_EVENT_FLOW_AGED, NULL);
7493 age_info->flags = 0;
7498 * Handler for the HW respond about ready values from an asynchronous batch
7499 * query. This function is probably called by the host thread.
7502 * The pointer to the shared device context.
7503 * @param[in] async_id
7504 * The Devx async ID.
7506 * The status of the completion.
7509 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7510 uint64_t async_id, int status)
7512 struct mlx5_flow_counter_pool *pool =
7513 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7514 struct mlx5_counter_stats_raw *raw_to_free;
7515 uint8_t query_gen = pool->query_gen ^ 1;
7516 struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7517 enum mlx5_counter_type cnt_type =
7518 pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7519 MLX5_COUNTER_TYPE_ORIGIN;
7521 if (unlikely(status)) {
7522 raw_to_free = pool->raw_hw;
7524 raw_to_free = pool->raw;
7526 mlx5_flow_aging_check(sh, pool);
7527 rte_spinlock_lock(&pool->sl);
7528 pool->raw = pool->raw_hw;
7529 rte_spinlock_unlock(&pool->sl);
7530 /* Be sure the new raw counters data is updated in memory. */
7532 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7533 rte_spinlock_lock(&cmng->csl[cnt_type]);
7534 TAILQ_CONCAT(&cmng->counters[cnt_type],
7535 &pool->counters[query_gen], next);
7536 rte_spinlock_unlock(&cmng->csl[cnt_type]);
7539 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7540 pool->raw_hw = NULL;
7541 sh->cmng.pending_queries--;
7544 static const struct mlx5_flow_tbl_data_entry *
7545 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7547 struct mlx5_priv *priv = dev->data->dev_private;
7548 struct mlx5_dev_ctx_shared *sh = priv->sh;
7549 struct mlx5_hlist_entry *he;
7550 union tunnel_offload_mark mbits = { .val = mark };
7551 union mlx5_flow_tbl_key table_key = {
7553 .table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7555 .domain = !!mbits.transfer,
7559 he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64);
7561 container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7565 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7566 const struct mlx5_flow_tunnel *tunnel,
7567 uint32_t group, uint32_t *table,
7568 struct rte_flow_error *error)
7570 struct mlx5_priv *priv = dev->data->dev_private;
7571 struct mlx5_hlist_entry *he;
7572 struct tunnel_tbl_entry *tte;
7573 union tunnel_tbl_key key = {
7574 .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7577 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7578 struct mlx5_hlist *group_hash;
7580 group_hash = tunnel ? tunnel->groups : thub->groups;
7581 he = mlx5_hlist_lookup(group_hash, key.val);
7583 tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7588 tte->hash.key = key.val;
7589 mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7591 if (tte->flow_table >= MLX5_MAX_TABLES) {
7592 DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7594 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7597 } else if (!tte->flow_table) {
7600 tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7601 mlx5_hlist_insert(group_hash, &tte->hash);
7603 tte = container_of(he, typeof(*tte), hash);
7605 *table = tte->flow_table;
7606 DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7607 dev->data->port_id, key.tunnel_id, group, *table);
7613 return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7614 NULL, "tunnel group index not supported");
7618 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7619 struct flow_grp_info grp_info, struct rte_flow_error *error)
7621 if (grp_info.transfer && grp_info.external && grp_info.fdb_def_rule) {
7622 if (group == UINT32_MAX)
7623 return rte_flow_error_set
7625 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7627 "group index not supported");
7632 DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7637 * Translate the rte_flow group index to HW table value.
7639 * If tunnel offload is disabled, all group ids converted to flow table
7640 * id using the standard method.
7641 * If tunnel offload is enabled, group id can be converted using the
7642 * standard or tunnel conversion method. Group conversion method
7643 * selection depends on flags in `grp_info` parameter:
7644 * - Internal (grp_info.external == 0) groups conversion uses the
7646 * - Group ids in JUMP action converted with the tunnel conversion.
7647 * - Group id in rule attribute conversion depends on a rule type and
7649 * ** non zero group attributes converted with the tunnel method
7650 * ** zero group attribute in non-tunnel rule is converted using the
7651 * standard method - there's only one root table
7652 * ** zero group attribute in steer tunnel rule is converted with the
7653 * standard method - single root table
7654 * ** zero group attribute in match tunnel rule is a special OvS
7655 * case: that value is used for portability reasons. That group
7656 * id is converted with the tunnel conversion method.
7661 * PMD tunnel offload object
7663 * rte_flow group index value.
7666 * @param[in] grp_info
7667 * flags used for conversion
7669 * Pointer to error structure.
7672 * 0 on success, a negative errno value otherwise and rte_errno is set.
7675 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7676 const struct mlx5_flow_tunnel *tunnel,
7677 uint32_t group, uint32_t *table,
7678 struct flow_grp_info grp_info,
7679 struct rte_flow_error *error)
7682 bool standard_translation;
7684 if (grp_info.external && group < MLX5_MAX_TABLES_EXTERNAL)
7685 group *= MLX5_FLOW_TABLE_FACTOR;
7686 if (is_tunnel_offload_active(dev)) {
7687 standard_translation = !grp_info.external ||
7688 grp_info.std_tbl_fix;
7690 standard_translation = true;
7693 "port %u group=%#x transfer=%d external=%d fdb_def_rule=%d translate=%s",
7694 dev->data->port_id, group, grp_info.transfer,
7695 grp_info.external, grp_info.fdb_def_rule,
7696 standard_translation ? "STANDARD" : "TUNNEL");
7697 if (standard_translation)
7698 ret = flow_group_to_table(dev->data->port_id, group, table,
7701 ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7708 * Discover availability of metadata reg_c's.
7710 * Iteratively use test flows to check availability.
7713 * Pointer to the Ethernet device structure.
7716 * 0 on success, a negative errno value otherwise and rte_errno is set.
7719 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7721 struct mlx5_priv *priv = dev->data->dev_private;
7722 struct mlx5_dev_config *config = &priv->config;
7723 enum modify_reg idx;
7726 /* reg_c[0] and reg_c[1] are reserved. */
7727 config->flow_mreg_c[n++] = REG_C_0;
7728 config->flow_mreg_c[n++] = REG_C_1;
7729 /* Discover availability of other reg_c's. */
7730 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7731 struct rte_flow_attr attr = {
7732 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7733 .priority = MLX5_FLOW_PRIO_RSVD,
7736 struct rte_flow_item items[] = {
7738 .type = RTE_FLOW_ITEM_TYPE_END,
7741 struct rte_flow_action actions[] = {
7743 .type = (enum rte_flow_action_type)
7744 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7745 .conf = &(struct mlx5_flow_action_copy_mreg){
7751 .type = RTE_FLOW_ACTION_TYPE_JUMP,
7752 .conf = &(struct rte_flow_action_jump){
7753 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7757 .type = RTE_FLOW_ACTION_TYPE_END,
7761 struct rte_flow *flow;
7762 struct rte_flow_error error;
7764 if (!config->dv_flow_en)
7766 /* Create internal flow, validation skips copy action. */
7767 flow_idx = flow_list_create(dev, NULL, &attr, items,
7768 actions, false, &error);
7769 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7773 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
7774 config->flow_mreg_c[n++] = idx;
7775 flow_list_destroy(dev, NULL, flow_idx);
7777 for (; n < MLX5_MREG_C_NUM; ++n)
7778 config->flow_mreg_c[n] = REG_NON;
7783 * Dump flow raw hw data to file
7786 * The pointer to Ethernet device.
7788 * A pointer to a file for output.
7790 * Perform verbose error reporting if not NULL. PMDs initialize this
7791 * structure in case of error only.
7793 * 0 on success, a nagative value otherwise.
7796 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7798 struct rte_flow_error *error __rte_unused)
7800 struct mlx5_priv *priv = dev->data->dev_private;
7801 struct mlx5_dev_ctx_shared *sh = priv->sh;
7803 if (!priv->config.dv_flow_en) {
7804 if (fputs("device dv flow disabled\n", file) <= 0)
7808 return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7809 sh->tx_domain, file);
7813 * Get aged-out flows.
7816 * Pointer to the Ethernet device structure.
7817 * @param[in] context
7818 * The address of an array of pointers to the aged-out flows contexts.
7819 * @param[in] nb_countexts
7820 * The length of context array pointers.
7822 * Perform verbose error reporting if not NULL. Initialized in case of
7826 * how many contexts get in success, otherwise negative errno value.
7827 * if nb_contexts is 0, return the amount of all aged contexts.
7828 * if nb_contexts is not 0 , return the amount of aged flows reported
7829 * in the context array.
7832 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7833 uint32_t nb_contexts, struct rte_flow_error *error)
7835 const struct mlx5_flow_driver_ops *fops;
7836 struct rte_flow_attr attr = { .transfer = 0 };
7838 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7839 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7840 return fops->get_aged_flows(dev, contexts, nb_contexts,
7844 "port %u get aged flows is not supported.",
7845 dev->data->port_id);
7849 /* Wrapper for driver action_validate op callback */
7851 flow_drv_action_validate(struct rte_eth_dev *dev,
7852 const struct rte_flow_shared_action_conf *conf,
7853 const struct rte_flow_action *action,
7854 const struct mlx5_flow_driver_ops *fops,
7855 struct rte_flow_error *error)
7857 static const char err_msg[] = "shared action validation unsupported";
7859 if (!fops->action_validate) {
7860 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7861 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7865 return fops->action_validate(dev, conf, action, error);
7869 * Destroys the shared action by handle.
7872 * Pointer to Ethernet device structure.
7874 * Handle for the shared action to be destroyed.
7876 * Perform verbose error reporting if not NULL. PMDs initialize this
7877 * structure in case of error only.
7880 * 0 on success, a negative errno value otherwise and rte_errno is set.
7882 * @note: wrapper for driver action_create op callback.
7885 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7886 struct rte_flow_shared_action *action,
7887 struct rte_flow_error *error)
7889 static const char err_msg[] = "shared action destruction unsupported";
7890 struct rte_flow_attr attr = { .transfer = 0 };
7891 const struct mlx5_flow_driver_ops *fops =
7892 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7894 if (!fops->action_destroy) {
7895 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7896 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7900 return fops->action_destroy(dev, action, error);
7903 /* Wrapper for driver action_destroy op callback */
7905 flow_drv_action_update(struct rte_eth_dev *dev,
7906 struct rte_flow_shared_action *action,
7907 const void *action_conf,
7908 const struct mlx5_flow_driver_ops *fops,
7909 struct rte_flow_error *error)
7911 static const char err_msg[] = "shared action update unsupported";
7913 if (!fops->action_update) {
7914 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7915 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7919 return fops->action_update(dev, action, action_conf, error);
7923 * Create shared action for reuse in multiple flow rules.
7926 * Pointer to Ethernet device structure.
7928 * Action configuration for shared action creation.
7930 * Perform verbose error reporting if not NULL. PMDs initialize this
7931 * structure in case of error only.
7933 * A valid handle in case of success, NULL otherwise and rte_errno is set.
7935 static struct rte_flow_shared_action *
7936 mlx5_shared_action_create(struct rte_eth_dev *dev,
7937 const struct rte_flow_shared_action_conf *conf,
7938 const struct rte_flow_action *action,
7939 struct rte_flow_error *error)
7941 static const char err_msg[] = "shared action creation unsupported";
7942 struct rte_flow_attr attr = { .transfer = 0 };
7943 const struct mlx5_flow_driver_ops *fops =
7944 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7946 if (flow_drv_action_validate(dev, conf, action, fops, error))
7948 if (!fops->action_create) {
7949 DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7950 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7954 return fops->action_create(dev, conf, action, error);
7958 * Updates inplace the shared action configuration pointed by *action* handle
7959 * with the configuration provided as *action* argument.
7960 * The update of the shared action configuration effects all flow rules reusing
7961 * the action via handle.
7964 * Pointer to Ethernet device structure.
7965 * @param[in] shared_action
7966 * Handle for the shared action to be updated.
7968 * Action specification used to modify the action pointed by handle.
7969 * *action* should be of same type with the action pointed by the *action*
7970 * handle argument, otherwise considered as invalid.
7972 * Perform verbose error reporting if not NULL. PMDs initialize this
7973 * structure in case of error only.
7976 * 0 on success, a negative errno value otherwise and rte_errno is set.
7979 mlx5_shared_action_update(struct rte_eth_dev *dev,
7980 struct rte_flow_shared_action *shared_action,
7981 const struct rte_flow_action *action,
7982 struct rte_flow_error *error)
7984 struct rte_flow_attr attr = { .transfer = 0 };
7985 const struct mlx5_flow_driver_ops *fops =
7986 flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7989 switch (shared_action->type) {
7990 case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
7991 if (action->type != RTE_FLOW_ACTION_TYPE_RSS) {
7992 return rte_flow_error_set(error, EINVAL,
7993 RTE_FLOW_ERROR_TYPE_ACTION,
7995 "update action type invalid");
7997 ret = flow_drv_action_validate(dev, NULL, action, fops, error);
8000 return flow_drv_action_update(dev, shared_action, action->conf,
8003 return rte_flow_error_set(error, ENOTSUP,
8004 RTE_FLOW_ERROR_TYPE_ACTION,
8006 "action type not supported");
8011 * Query the shared action by handle.
8013 * This function allows retrieving action-specific data such as counters.
8014 * Data is gathered by special action which may be present/referenced in
8015 * more than one flow rule definition.
8017 * \see RTE_FLOW_ACTION_TYPE_COUNT
8020 * Pointer to Ethernet device structure.
8022 * Handle for the shared action to query.
8023 * @param[in, out] data
8024 * Pointer to storage for the associated query data type.
8026 * Perform verbose error reporting if not NULL. PMDs initialize this
8027 * structure in case of error only.
8030 * 0 on success, a negative errno value otherwise and rte_errno is set.
8033 mlx5_shared_action_query(struct rte_eth_dev *dev,
8034 const struct rte_flow_shared_action *action,
8036 struct rte_flow_error *error)
8039 switch (action->type) {
8040 case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
8041 __atomic_load(&action->refcnt, (uint32_t *)data,
8045 return rte_flow_error_set(error, ENOTSUP,
8046 RTE_FLOW_ERROR_TYPE_ACTION,
8048 "action type not supported");
8053 * Destroy all shared actions.
8056 * Pointer to Ethernet device.
8059 * 0 on success, a negative errno value otherwise and rte_errno is set.
8062 mlx5_shared_action_flush(struct rte_eth_dev *dev)
8064 struct rte_flow_error error;
8065 struct mlx5_priv *priv = dev->data->dev_private;
8066 struct rte_flow_shared_action *action;
8069 while (!LIST_EMPTY(&priv->shared_actions)) {
8070 action = LIST_FIRST(&priv->shared_actions);
8071 ret = mlx5_shared_action_destroy(dev, action, &error);
8077 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8078 struct mlx5_flow_tunnel *tunnel)
8080 struct mlx5_priv *priv = dev->data->dev_private;
8082 DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8083 dev->data->port_id, tunnel->tunnel_id);
8084 RTE_VERIFY(!__atomic_load_n(&tunnel->refctn, __ATOMIC_RELAXED));
8085 LIST_REMOVE(tunnel, chain);
8086 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID],
8088 mlx5_hlist_destroy(tunnel->groups, NULL, NULL);
8092 static struct mlx5_flow_tunnel *
8093 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8095 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8096 struct mlx5_flow_tunnel *tun;
8098 LIST_FOREACH(tun, &thub->tunnels, chain) {
8099 if (tun->tunnel_id == id)
8106 static struct mlx5_flow_tunnel *
8107 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8108 const struct rte_flow_tunnel *app_tunnel)
8110 struct mlx5_priv *priv = dev->data->dev_private;
8111 struct mlx5_flow_tunnel *tunnel;
8114 mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
8116 if (id >= MLX5_MAX_TUNNELS) {
8117 mlx5_ipool_free(priv->sh->ipool
8118 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
8119 DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8125 * mlx5 flow tunnel is an auxlilary data structure
8126 * It's not part of IO. No need to allocate it from
8127 * huge pages pools dedicated for IO
8129 tunnel = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*tunnel),
8132 mlx5_ipool_free(priv->sh->ipool
8133 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
8136 tunnel->groups = mlx5_hlist_create("tunnel groups", 1024);
8137 if (!tunnel->groups) {
8138 mlx5_ipool_free(priv->sh->ipool
8139 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
8143 /* initiate new PMD tunnel */
8144 memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8145 tunnel->tunnel_id = id;
8146 tunnel->action.type = (typeof(tunnel->action.type))
8147 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8148 tunnel->action.conf = tunnel;
8149 tunnel->item.type = (typeof(tunnel->item.type))
8150 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8151 tunnel->item.spec = tunnel;
8152 tunnel->item.last = NULL;
8153 tunnel->item.mask = NULL;
8155 DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8156 dev->data->port_id, tunnel->tunnel_id);
8162 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8163 const struct rte_flow_tunnel *app_tunnel,
8164 struct mlx5_flow_tunnel **tunnel)
8167 struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8168 struct mlx5_flow_tunnel *tun;
8170 LIST_FOREACH(tun, &thub->tunnels, chain) {
8171 if (!memcmp(app_tunnel, &tun->app_tunnel,
8172 sizeof(*app_tunnel))) {
8179 tun = mlx5_flow_tunnel_allocate(dev, app_tunnel);
8181 LIST_INSERT_HEAD(&thub->tunnels, tun, chain);
8188 __atomic_add_fetch(&tun->refctn, 1, __ATOMIC_RELAXED);
8193 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8195 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8199 if (!LIST_EMPTY(&thub->tunnels))
8200 DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
8201 mlx5_hlist_destroy(thub->groups, NULL, NULL);
8205 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8208 struct mlx5_flow_tunnel_hub *thub;
8210 thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8214 LIST_INIT(&thub->tunnels);
8215 thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES);
8216 if (!thub->groups) {
8220 sh->tunnel_hub = thub;
8226 mlx5_hlist_destroy(thub->groups, NULL, NULL);
8232 #ifndef HAVE_MLX5DV_DR
8233 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8235 #define MLX5_DOMAIN_SYNC_FLOW \
8236 (MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8239 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8241 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8242 const struct mlx5_flow_driver_ops *fops;
8244 struct rte_flow_attr attr = { .transfer = 0 };
8246 fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8247 ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);