1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018 Mellanox Technologies, Ltd
12 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #pragma GCC diagnostic ignored "-Wpedantic"
16 #include <infiniband/verbs.h>
18 #pragma GCC diagnostic error "-Wpedantic"
21 #include <rte_common.h>
22 #include <rte_ether.h>
23 #include <rte_ethdev_driver.h>
25 #include <rte_flow_driver.h>
26 #include <rte_malloc.h>
31 #include "mlx5_defs.h"
32 #include "mlx5_glue.h"
33 #include "mlx5_flow.h"
35 #include "mlx5_rxtx.h"
37 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
39 #ifndef HAVE_IBV_FLOW_DEVX_COUNTERS
40 #define MLX5DV_FLOW_ACTION_COUNTERS_DEVX 0
43 #ifndef HAVE_MLX5DV_DR_ESWITCH
44 #ifndef MLX5DV_FLOW_TABLE_TYPE_FDB
45 #define MLX5DV_FLOW_TABLE_TYPE_FDB 0
49 #ifndef HAVE_MLX5DV_DR
50 #define MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL 1
66 * Initialize flow attributes structure according to flow items' types.
69 * Pointer to item specification.
71 * Pointer to flow attributes structure.
74 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr)
76 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
78 case RTE_FLOW_ITEM_TYPE_IPV4:
81 case RTE_FLOW_ITEM_TYPE_IPV6:
84 case RTE_FLOW_ITEM_TYPE_UDP:
87 case RTE_FLOW_ITEM_TYPE_TCP:
97 struct field_modify_info {
98 uint32_t size; /* Size of field in protocol header, in bytes. */
99 uint32_t offset; /* Offset of field in protocol header, in bytes. */
100 enum mlx5_modification_field id;
103 struct field_modify_info modify_eth[] = {
104 {4, 0, MLX5_MODI_OUT_DMAC_47_16},
105 {2, 4, MLX5_MODI_OUT_DMAC_15_0},
106 {4, 6, MLX5_MODI_OUT_SMAC_47_16},
107 {2, 10, MLX5_MODI_OUT_SMAC_15_0},
111 struct field_modify_info modify_ipv4[] = {
112 {1, 8, MLX5_MODI_OUT_IPV4_TTL},
113 {4, 12, MLX5_MODI_OUT_SIPV4},
114 {4, 16, MLX5_MODI_OUT_DIPV4},
118 struct field_modify_info modify_ipv6[] = {
119 {1, 7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
120 {4, 8, MLX5_MODI_OUT_SIPV6_127_96},
121 {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
122 {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
123 {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
124 {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
125 {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
126 {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
127 {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
131 struct field_modify_info modify_udp[] = {
132 {2, 0, MLX5_MODI_OUT_UDP_SPORT},
133 {2, 2, MLX5_MODI_OUT_UDP_DPORT},
137 struct field_modify_info modify_tcp[] = {
138 {2, 0, MLX5_MODI_OUT_TCP_SPORT},
139 {2, 2, MLX5_MODI_OUT_TCP_DPORT},
140 {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
141 {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
146 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item, uint64_t *flags)
148 uint8_t next_protocol = 0xFF;
150 if (item->mask != NULL) {
151 switch (item->type) {
152 case RTE_FLOW_ITEM_TYPE_IPV4:
154 ((const struct rte_flow_item_ipv4 *)
155 (item->spec))->hdr.next_proto_id;
157 ((const struct rte_flow_item_ipv4 *)
158 (item->mask))->hdr.next_proto_id;
160 case RTE_FLOW_ITEM_TYPE_IPV6:
162 ((const struct rte_flow_item_ipv6 *)
163 (item->spec))->hdr.proto;
165 ((const struct rte_flow_item_ipv6 *)
166 (item->mask))->hdr.proto;
172 if (next_protocol == IPPROTO_IPIP)
173 *flags |= MLX5_FLOW_LAYER_IPIP;
174 if (next_protocol == IPPROTO_IPV6)
175 *flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
179 * Acquire the synchronizing object to protect multithreaded access
180 * to shared dv context. Lock occurs only if context is actually
181 * shared, i.e. we have multiport IB device and representors are
185 * Pointer to the rte_eth_dev structure.
188 flow_d_shared_lock(struct rte_eth_dev *dev)
190 struct mlx5_priv *priv = dev->data->dev_private;
191 struct mlx5_ibv_shared *sh = priv->sh;
193 if (sh->dv_refcnt > 1) {
196 ret = pthread_mutex_lock(&sh->dv_mutex);
203 flow_d_shared_unlock(struct rte_eth_dev *dev)
205 struct mlx5_priv *priv = dev->data->dev_private;
206 struct mlx5_ibv_shared *sh = priv->sh;
208 if (sh->dv_refcnt > 1) {
211 ret = pthread_mutex_unlock(&sh->dv_mutex);
218 * Convert modify-header action to DV specification.
221 * Pointer to item specification.
223 * Pointer to field modification information.
224 * @param[in,out] resource
225 * Pointer to the modify-header resource.
227 * Type of modification.
229 * Pointer to the error structure.
232 * 0 on success, a negative errno value otherwise and rte_errno is set.
235 flow_dv_convert_modify_action(struct rte_flow_item *item,
236 struct field_modify_info *field,
237 struct mlx5_flow_dv_modify_hdr_resource *resource,
239 struct rte_flow_error *error)
241 uint32_t i = resource->actions_num;
242 struct mlx5_modification_cmd *actions = resource->actions;
243 const uint8_t *spec = item->spec;
244 const uint8_t *mask = item->mask;
247 while (field->size) {
249 /* Generate modify command for each mask segment. */
250 memcpy(&set, &mask[field->offset], field->size);
252 if (i >= MLX5_MODIFY_NUM)
253 return rte_flow_error_set(error, EINVAL,
254 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
255 "too many items to modify");
256 actions[i].action_type = type;
257 actions[i].field = field->id;
258 actions[i].length = field->size ==
259 4 ? 0 : field->size * 8;
260 rte_memcpy(&actions[i].data[4 - field->size],
261 &spec[field->offset], field->size);
262 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
265 if (resource->actions_num != i)
266 resource->actions_num = i;
269 if (!resource->actions_num)
270 return rte_flow_error_set(error, EINVAL,
271 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
272 "invalid modification flow item");
277 * Convert modify-header set IPv4 address action to DV specification.
279 * @param[in,out] resource
280 * Pointer to the modify-header resource.
282 * Pointer to action specification.
284 * Pointer to the error structure.
287 * 0 on success, a negative errno value otherwise and rte_errno is set.
290 flow_dv_convert_action_modify_ipv4
291 (struct mlx5_flow_dv_modify_hdr_resource *resource,
292 const struct rte_flow_action *action,
293 struct rte_flow_error *error)
295 const struct rte_flow_action_set_ipv4 *conf =
296 (const struct rte_flow_action_set_ipv4 *)(action->conf);
297 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
298 struct rte_flow_item_ipv4 ipv4;
299 struct rte_flow_item_ipv4 ipv4_mask;
301 memset(&ipv4, 0, sizeof(ipv4));
302 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
303 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
304 ipv4.hdr.src_addr = conf->ipv4_addr;
305 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
307 ipv4.hdr.dst_addr = conf->ipv4_addr;
308 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
311 item.mask = &ipv4_mask;
312 return flow_dv_convert_modify_action(&item, modify_ipv4, resource,
313 MLX5_MODIFICATION_TYPE_SET, error);
317 * Convert modify-header set IPv6 address action to DV specification.
319 * @param[in,out] resource
320 * Pointer to the modify-header resource.
322 * Pointer to action specification.
324 * Pointer to the error structure.
327 * 0 on success, a negative errno value otherwise and rte_errno is set.
330 flow_dv_convert_action_modify_ipv6
331 (struct mlx5_flow_dv_modify_hdr_resource *resource,
332 const struct rte_flow_action *action,
333 struct rte_flow_error *error)
335 const struct rte_flow_action_set_ipv6 *conf =
336 (const struct rte_flow_action_set_ipv6 *)(action->conf);
337 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
338 struct rte_flow_item_ipv6 ipv6;
339 struct rte_flow_item_ipv6 ipv6_mask;
341 memset(&ipv6, 0, sizeof(ipv6));
342 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
343 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
344 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
345 sizeof(ipv6.hdr.src_addr));
346 memcpy(&ipv6_mask.hdr.src_addr,
347 &rte_flow_item_ipv6_mask.hdr.src_addr,
348 sizeof(ipv6.hdr.src_addr));
350 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
351 sizeof(ipv6.hdr.dst_addr));
352 memcpy(&ipv6_mask.hdr.dst_addr,
353 &rte_flow_item_ipv6_mask.hdr.dst_addr,
354 sizeof(ipv6.hdr.dst_addr));
357 item.mask = &ipv6_mask;
358 return flow_dv_convert_modify_action(&item, modify_ipv6, resource,
359 MLX5_MODIFICATION_TYPE_SET, error);
363 * Convert modify-header set MAC address action to DV specification.
365 * @param[in,out] resource
366 * Pointer to the modify-header resource.
368 * Pointer to action specification.
370 * Pointer to the error structure.
373 * 0 on success, a negative errno value otherwise and rte_errno is set.
376 flow_dv_convert_action_modify_mac
377 (struct mlx5_flow_dv_modify_hdr_resource *resource,
378 const struct rte_flow_action *action,
379 struct rte_flow_error *error)
381 const struct rte_flow_action_set_mac *conf =
382 (const struct rte_flow_action_set_mac *)(action->conf);
383 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
384 struct rte_flow_item_eth eth;
385 struct rte_flow_item_eth eth_mask;
387 memset(ð, 0, sizeof(eth));
388 memset(ð_mask, 0, sizeof(eth_mask));
389 if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
390 memcpy(ð.src.addr_bytes, &conf->mac_addr,
391 sizeof(eth.src.addr_bytes));
392 memcpy(ð_mask.src.addr_bytes,
393 &rte_flow_item_eth_mask.src.addr_bytes,
394 sizeof(eth_mask.src.addr_bytes));
396 memcpy(ð.dst.addr_bytes, &conf->mac_addr,
397 sizeof(eth.dst.addr_bytes));
398 memcpy(ð_mask.dst.addr_bytes,
399 &rte_flow_item_eth_mask.dst.addr_bytes,
400 sizeof(eth_mask.dst.addr_bytes));
403 item.mask = ð_mask;
404 return flow_dv_convert_modify_action(&item, modify_eth, resource,
405 MLX5_MODIFICATION_TYPE_SET, error);
409 * Convert modify-header set TP action to DV specification.
411 * @param[in,out] resource
412 * Pointer to the modify-header resource.
414 * Pointer to action specification.
416 * Pointer to rte_flow_item objects list.
418 * Pointer to flow attributes structure.
420 * Pointer to the error structure.
423 * 0 on success, a negative errno value otherwise and rte_errno is set.
426 flow_dv_convert_action_modify_tp
427 (struct mlx5_flow_dv_modify_hdr_resource *resource,
428 const struct rte_flow_action *action,
429 const struct rte_flow_item *items,
430 union flow_dv_attr *attr,
431 struct rte_flow_error *error)
433 const struct rte_flow_action_set_tp *conf =
434 (const struct rte_flow_action_set_tp *)(action->conf);
435 struct rte_flow_item item;
436 struct rte_flow_item_udp udp;
437 struct rte_flow_item_udp udp_mask;
438 struct rte_flow_item_tcp tcp;
439 struct rte_flow_item_tcp tcp_mask;
440 struct field_modify_info *field;
443 flow_dv_attr_init(items, attr);
445 memset(&udp, 0, sizeof(udp));
446 memset(&udp_mask, 0, sizeof(udp_mask));
447 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
448 udp.hdr.src_port = conf->port;
449 udp_mask.hdr.src_port =
450 rte_flow_item_udp_mask.hdr.src_port;
452 udp.hdr.dst_port = conf->port;
453 udp_mask.hdr.dst_port =
454 rte_flow_item_udp_mask.hdr.dst_port;
456 item.type = RTE_FLOW_ITEM_TYPE_UDP;
458 item.mask = &udp_mask;
462 memset(&tcp, 0, sizeof(tcp));
463 memset(&tcp_mask, 0, sizeof(tcp_mask));
464 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
465 tcp.hdr.src_port = conf->port;
466 tcp_mask.hdr.src_port =
467 rte_flow_item_tcp_mask.hdr.src_port;
469 tcp.hdr.dst_port = conf->port;
470 tcp_mask.hdr.dst_port =
471 rte_flow_item_tcp_mask.hdr.dst_port;
473 item.type = RTE_FLOW_ITEM_TYPE_TCP;
475 item.mask = &tcp_mask;
478 return flow_dv_convert_modify_action(&item, field, resource,
479 MLX5_MODIFICATION_TYPE_SET, error);
483 * Convert modify-header set TTL action to DV specification.
485 * @param[in,out] resource
486 * Pointer to the modify-header resource.
488 * Pointer to action specification.
490 * Pointer to rte_flow_item objects list.
492 * Pointer to flow attributes structure.
494 * Pointer to the error structure.
497 * 0 on success, a negative errno value otherwise and rte_errno is set.
500 flow_dv_convert_action_modify_ttl
501 (struct mlx5_flow_dv_modify_hdr_resource *resource,
502 const struct rte_flow_action *action,
503 const struct rte_flow_item *items,
504 union flow_dv_attr *attr,
505 struct rte_flow_error *error)
507 const struct rte_flow_action_set_ttl *conf =
508 (const struct rte_flow_action_set_ttl *)(action->conf);
509 struct rte_flow_item item;
510 struct rte_flow_item_ipv4 ipv4;
511 struct rte_flow_item_ipv4 ipv4_mask;
512 struct rte_flow_item_ipv6 ipv6;
513 struct rte_flow_item_ipv6 ipv6_mask;
514 struct field_modify_info *field;
517 flow_dv_attr_init(items, attr);
519 memset(&ipv4, 0, sizeof(ipv4));
520 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
521 ipv4.hdr.time_to_live = conf->ttl_value;
522 ipv4_mask.hdr.time_to_live = 0xFF;
523 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
525 item.mask = &ipv4_mask;
529 memset(&ipv6, 0, sizeof(ipv6));
530 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
531 ipv6.hdr.hop_limits = conf->ttl_value;
532 ipv6_mask.hdr.hop_limits = 0xFF;
533 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
535 item.mask = &ipv6_mask;
538 return flow_dv_convert_modify_action(&item, field, resource,
539 MLX5_MODIFICATION_TYPE_SET, error);
543 * Convert modify-header decrement TTL action to DV specification.
545 * @param[in,out] resource
546 * Pointer to the modify-header resource.
548 * Pointer to action specification.
550 * Pointer to rte_flow_item objects list.
552 * Pointer to flow attributes structure.
554 * Pointer to the error structure.
557 * 0 on success, a negative errno value otherwise and rte_errno is set.
560 flow_dv_convert_action_modify_dec_ttl
561 (struct mlx5_flow_dv_modify_hdr_resource *resource,
562 const struct rte_flow_item *items,
563 union flow_dv_attr *attr,
564 struct rte_flow_error *error)
566 struct rte_flow_item item;
567 struct rte_flow_item_ipv4 ipv4;
568 struct rte_flow_item_ipv4 ipv4_mask;
569 struct rte_flow_item_ipv6 ipv6;
570 struct rte_flow_item_ipv6 ipv6_mask;
571 struct field_modify_info *field;
574 flow_dv_attr_init(items, attr);
576 memset(&ipv4, 0, sizeof(ipv4));
577 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
578 ipv4.hdr.time_to_live = 0xFF;
579 ipv4_mask.hdr.time_to_live = 0xFF;
580 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
582 item.mask = &ipv4_mask;
586 memset(&ipv6, 0, sizeof(ipv6));
587 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
588 ipv6.hdr.hop_limits = 0xFF;
589 ipv6_mask.hdr.hop_limits = 0xFF;
590 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
592 item.mask = &ipv6_mask;
595 return flow_dv_convert_modify_action(&item, field, resource,
596 MLX5_MODIFICATION_TYPE_ADD, error);
600 * Convert modify-header increment/decrement TCP Sequence number
601 * to DV specification.
603 * @param[in,out] resource
604 * Pointer to the modify-header resource.
606 * Pointer to action specification.
608 * Pointer to the error structure.
611 * 0 on success, a negative errno value otherwise and rte_errno is set.
614 flow_dv_convert_action_modify_tcp_seq
615 (struct mlx5_flow_dv_modify_hdr_resource *resource,
616 const struct rte_flow_action *action,
617 struct rte_flow_error *error)
619 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
620 uint64_t value = rte_be_to_cpu_32(*conf);
621 struct rte_flow_item item;
622 struct rte_flow_item_tcp tcp;
623 struct rte_flow_item_tcp tcp_mask;
625 memset(&tcp, 0, sizeof(tcp));
626 memset(&tcp_mask, 0, sizeof(tcp_mask));
627 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
629 * The HW has no decrement operation, only increment operation.
630 * To simulate decrement X from Y using increment operation
631 * we need to add UINT32_MAX X times to Y.
632 * Each adding of UINT32_MAX decrements Y by 1.
635 tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
636 tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
637 item.type = RTE_FLOW_ITEM_TYPE_TCP;
639 item.mask = &tcp_mask;
640 return flow_dv_convert_modify_action(&item, modify_tcp, resource,
641 MLX5_MODIFICATION_TYPE_ADD, error);
645 * Convert modify-header increment/decrement TCP Acknowledgment number
646 * to DV specification.
648 * @param[in,out] resource
649 * Pointer to the modify-header resource.
651 * Pointer to action specification.
653 * Pointer to the error structure.
656 * 0 on success, a negative errno value otherwise and rte_errno is set.
659 flow_dv_convert_action_modify_tcp_ack
660 (struct mlx5_flow_dv_modify_hdr_resource *resource,
661 const struct rte_flow_action *action,
662 struct rte_flow_error *error)
664 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
665 uint64_t value = rte_be_to_cpu_32(*conf);
666 struct rte_flow_item item;
667 struct rte_flow_item_tcp tcp;
668 struct rte_flow_item_tcp tcp_mask;
670 memset(&tcp, 0, sizeof(tcp));
671 memset(&tcp_mask, 0, sizeof(tcp_mask));
672 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
674 * The HW has no decrement operation, only increment operation.
675 * To simulate decrement X from Y using increment operation
676 * we need to add UINT32_MAX X times to Y.
677 * Each adding of UINT32_MAX decrements Y by 1.
680 tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
681 tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
682 item.type = RTE_FLOW_ITEM_TYPE_TCP;
684 item.mask = &tcp_mask;
685 return flow_dv_convert_modify_action(&item, modify_tcp, resource,
686 MLX5_MODIFICATION_TYPE_ADD, error);
690 * Validate META item.
693 * Pointer to the rte_eth_dev structure.
695 * Item specification.
697 * Attributes of flow that includes this item.
699 * Pointer to error structure.
702 * 0 on success, a negative errno value otherwise and rte_errno is set.
705 flow_dv_validate_item_meta(struct rte_eth_dev *dev,
706 const struct rte_flow_item *item,
707 const struct rte_flow_attr *attr,
708 struct rte_flow_error *error)
710 const struct rte_flow_item_meta *spec = item->spec;
711 const struct rte_flow_item_meta *mask = item->mask;
712 const struct rte_flow_item_meta nic_mask = {
713 .data = RTE_BE32(UINT32_MAX)
716 uint64_t offloads = dev->data->dev_conf.txmode.offloads;
718 if (!(offloads & DEV_TX_OFFLOAD_MATCH_METADATA))
719 return rte_flow_error_set(error, EPERM,
720 RTE_FLOW_ERROR_TYPE_ITEM,
722 "match on metadata offload "
723 "configuration is off for this port");
725 return rte_flow_error_set(error, EINVAL,
726 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
728 "data cannot be empty");
730 return rte_flow_error_set(error, EINVAL,
731 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
733 "data cannot be zero");
735 mask = &rte_flow_item_meta_mask;
736 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
737 (const uint8_t *)&nic_mask,
738 sizeof(struct rte_flow_item_meta),
743 return rte_flow_error_set(error, ENOTSUP,
744 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
746 "pattern not supported for ingress");
751 * Validate vport item.
754 * Pointer to the rte_eth_dev structure.
756 * Item specification.
758 * Attributes of flow that includes this item.
759 * @param[in] item_flags
760 * Bit-fields that holds the items detected until now.
762 * Pointer to error structure.
765 * 0 on success, a negative errno value otherwise and rte_errno is set.
768 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
769 const struct rte_flow_item *item,
770 const struct rte_flow_attr *attr,
772 struct rte_flow_error *error)
774 const struct rte_flow_item_port_id *spec = item->spec;
775 const struct rte_flow_item_port_id *mask = item->mask;
776 const struct rte_flow_item_port_id switch_mask = {
779 uint16_t esw_domain_id;
780 uint16_t item_port_esw_domain_id;
784 return rte_flow_error_set(error, EINVAL,
785 RTE_FLOW_ERROR_TYPE_ITEM,
787 "match on port id is valid only"
788 " when transfer flag is enabled");
789 if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
790 return rte_flow_error_set(error, ENOTSUP,
791 RTE_FLOW_ERROR_TYPE_ITEM, item,
792 "multiple source ports are not"
796 if (mask->id != 0xffffffff)
797 return rte_flow_error_set(error, ENOTSUP,
798 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
800 "no support for partial mask on"
802 ret = mlx5_flow_item_acceptable
803 (item, (const uint8_t *)mask,
804 (const uint8_t *)&rte_flow_item_port_id_mask,
805 sizeof(struct rte_flow_item_port_id),
811 ret = mlx5_port_to_eswitch_info(spec->id, &item_port_esw_domain_id,
814 return rte_flow_error_set(error, -ret,
815 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
816 "failed to obtain E-Switch info for"
818 ret = mlx5_port_to_eswitch_info(dev->data->port_id,
819 &esw_domain_id, NULL);
821 return rte_flow_error_set(error, -ret,
822 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
824 "failed to obtain E-Switch info");
825 if (item_port_esw_domain_id != esw_domain_id)
826 return rte_flow_error_set(error, -ret,
827 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
828 "cannot match on a port from a"
829 " different E-Switch");
834 * Validate count action.
839 * Pointer to error structure.
842 * 0 on success, a negative errno value otherwise and rte_errno is set.
845 flow_dv_validate_action_count(struct rte_eth_dev *dev,
846 struct rte_flow_error *error)
848 struct mlx5_priv *priv = dev->data->dev_private;
850 if (!priv->config.devx)
852 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
856 return rte_flow_error_set
858 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
860 "count action not supported");
864 * Validate the L2 encap action.
866 * @param[in] action_flags
867 * Holds the actions detected until now.
869 * Pointer to the encap action.
871 * Pointer to flow attributes
873 * Pointer to error structure.
876 * 0 on success, a negative errno value otherwise and rte_errno is set.
879 flow_dv_validate_action_l2_encap(uint64_t action_flags,
880 const struct rte_flow_action *action,
881 const struct rte_flow_attr *attr,
882 struct rte_flow_error *error)
885 return rte_flow_error_set(error, EINVAL,
886 RTE_FLOW_ERROR_TYPE_ACTION, action,
887 "configuration cannot be null");
888 if (action_flags & MLX5_FLOW_ACTION_DROP)
889 return rte_flow_error_set(error, EINVAL,
890 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
891 "can't drop and encap in same flow");
892 if (action_flags & (MLX5_FLOW_ENCAP_ACTIONS | MLX5_FLOW_DECAP_ACTIONS))
893 return rte_flow_error_set(error, EINVAL,
894 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
895 "can only have a single encap or"
896 " decap action in a flow");
897 if (!attr->transfer && attr->ingress)
898 return rte_flow_error_set(error, ENOTSUP,
899 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
901 "encap action not supported for "
907 * Validate the L2 decap action.
909 * @param[in] action_flags
910 * Holds the actions detected until now.
912 * Pointer to flow attributes
914 * Pointer to error structure.
917 * 0 on success, a negative errno value otherwise and rte_errno is set.
920 flow_dv_validate_action_l2_decap(uint64_t action_flags,
921 const struct rte_flow_attr *attr,
922 struct rte_flow_error *error)
924 if (action_flags & MLX5_FLOW_ACTION_DROP)
925 return rte_flow_error_set(error, EINVAL,
926 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
927 "can't drop and decap in same flow");
928 if (action_flags & (MLX5_FLOW_ENCAP_ACTIONS | MLX5_FLOW_DECAP_ACTIONS))
929 return rte_flow_error_set(error, EINVAL,
930 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
931 "can only have a single encap or"
932 " decap action in a flow");
933 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
934 return rte_flow_error_set(error, EINVAL,
935 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
936 "can't have decap action after"
939 return rte_flow_error_set(error, ENOTSUP,
940 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
942 "decap action not supported for "
948 * Validate the raw encap action.
950 * @param[in] action_flags
951 * Holds the actions detected until now.
953 * Pointer to the encap action.
955 * Pointer to flow attributes
957 * Pointer to error structure.
960 * 0 on success, a negative errno value otherwise and rte_errno is set.
963 flow_dv_validate_action_raw_encap(uint64_t action_flags,
964 const struct rte_flow_action *action,
965 const struct rte_flow_attr *attr,
966 struct rte_flow_error *error)
969 return rte_flow_error_set(error, EINVAL,
970 RTE_FLOW_ERROR_TYPE_ACTION, action,
971 "configuration cannot be null");
972 if (action_flags & MLX5_FLOW_ACTION_DROP)
973 return rte_flow_error_set(error, EINVAL,
974 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
975 "can't drop and encap in same flow");
976 if (action_flags & MLX5_FLOW_ENCAP_ACTIONS)
977 return rte_flow_error_set(error, EINVAL,
978 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
979 "can only have a single encap"
980 " action in a flow");
981 /* encap without preceding decap is not supported for ingress */
982 if (!attr->transfer && attr->ingress &&
983 !(action_flags & MLX5_FLOW_ACTION_RAW_DECAP))
984 return rte_flow_error_set(error, ENOTSUP,
985 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
987 "encap action not supported for "
993 * Validate the raw decap action.
995 * @param[in] action_flags
996 * Holds the actions detected until now.
998 * Pointer to the encap action.
1000 * Pointer to flow attributes
1002 * Pointer to error structure.
1005 * 0 on success, a negative errno value otherwise and rte_errno is set.
1008 flow_dv_validate_action_raw_decap(uint64_t action_flags,
1009 const struct rte_flow_action *action,
1010 const struct rte_flow_attr *attr,
1011 struct rte_flow_error *error)
1013 if (action_flags & MLX5_FLOW_ACTION_DROP)
1014 return rte_flow_error_set(error, EINVAL,
1015 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1016 "can't drop and decap in same flow");
1017 if (action_flags & MLX5_FLOW_ENCAP_ACTIONS)
1018 return rte_flow_error_set(error, EINVAL,
1019 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1020 "can't have encap action before"
1022 if (action_flags & MLX5_FLOW_DECAP_ACTIONS)
1023 return rte_flow_error_set(error, EINVAL,
1024 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1025 "can only have a single decap"
1026 " action in a flow");
1027 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
1028 return rte_flow_error_set(error, EINVAL,
1029 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1030 "can't have decap action after"
1032 /* decap action is valid on egress only if it is followed by encap */
1034 for (; action->type != RTE_FLOW_ACTION_TYPE_END &&
1035 action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP;
1038 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP)
1039 return rte_flow_error_set
1041 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
1042 NULL, "decap action not supported"
1049 * Find existing encap/decap resource or create and register a new one.
1051 * @param dev[in, out]
1052 * Pointer to rte_eth_dev structure.
1053 * @param[in, out] resource
1054 * Pointer to encap/decap resource.
1055 * @parm[in, out] dev_flow
1056 * Pointer to the dev_flow.
1058 * pointer to error structure.
1061 * 0 on success otherwise -errno and errno is set.
1064 flow_dv_encap_decap_resource_register
1065 (struct rte_eth_dev *dev,
1066 struct mlx5_flow_dv_encap_decap_resource *resource,
1067 struct mlx5_flow *dev_flow,
1068 struct rte_flow_error *error)
1070 struct mlx5_priv *priv = dev->data->dev_private;
1071 struct mlx5_ibv_shared *sh = priv->sh;
1072 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
1073 struct rte_flow *flow = dev_flow->flow;
1074 struct mlx5dv_dr_domain *domain;
1076 resource->flags = flow->group ? 0 : 1;
1077 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
1078 domain = sh->fdb_domain;
1079 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
1080 domain = sh->rx_domain;
1082 domain = sh->tx_domain;
1084 /* Lookup a matching resource from cache. */
1085 LIST_FOREACH(cache_resource, &sh->encaps_decaps, next) {
1086 if (resource->reformat_type == cache_resource->reformat_type &&
1087 resource->ft_type == cache_resource->ft_type &&
1088 resource->flags == cache_resource->flags &&
1089 resource->size == cache_resource->size &&
1090 !memcmp((const void *)resource->buf,
1091 (const void *)cache_resource->buf,
1093 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
1094 (void *)cache_resource,
1095 rte_atomic32_read(&cache_resource->refcnt));
1096 rte_atomic32_inc(&cache_resource->refcnt);
1097 dev_flow->dv.encap_decap = cache_resource;
1101 /* Register new encap/decap resource. */
1102 cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
1103 if (!cache_resource)
1104 return rte_flow_error_set(error, ENOMEM,
1105 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1106 "cannot allocate resource memory");
1107 *cache_resource = *resource;
1108 cache_resource->verbs_action =
1109 mlx5_glue->dv_create_flow_action_packet_reformat
1110 (sh->ctx, cache_resource->reformat_type,
1111 cache_resource->ft_type, domain, cache_resource->flags,
1112 cache_resource->size,
1113 (cache_resource->size ? cache_resource->buf : NULL));
1114 if (!cache_resource->verbs_action) {
1115 rte_free(cache_resource);
1116 return rte_flow_error_set(error, ENOMEM,
1117 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1118 NULL, "cannot create action");
1120 rte_atomic32_init(&cache_resource->refcnt);
1121 rte_atomic32_inc(&cache_resource->refcnt);
1122 LIST_INSERT_HEAD(&sh->encaps_decaps, cache_resource, next);
1123 dev_flow->dv.encap_decap = cache_resource;
1124 DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
1125 (void *)cache_resource,
1126 rte_atomic32_read(&cache_resource->refcnt));
1131 * Find existing table jump resource or create and register a new one.
1133 * @param dev[in, out]
1134 * Pointer to rte_eth_dev structure.
1135 * @param[in, out] resource
1136 * Pointer to jump table resource.
1137 * @parm[in, out] dev_flow
1138 * Pointer to the dev_flow.
1140 * pointer to error structure.
1143 * 0 on success otherwise -errno and errno is set.
1146 flow_dv_jump_tbl_resource_register
1147 (struct rte_eth_dev *dev,
1148 struct mlx5_flow_dv_jump_tbl_resource *resource,
1149 struct mlx5_flow *dev_flow,
1150 struct rte_flow_error *error)
1152 struct mlx5_priv *priv = dev->data->dev_private;
1153 struct mlx5_ibv_shared *sh = priv->sh;
1154 struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
1156 /* Lookup a matching resource from cache. */
1157 LIST_FOREACH(cache_resource, &sh->jump_tbl, next) {
1158 if (resource->tbl == cache_resource->tbl) {
1159 DRV_LOG(DEBUG, "jump table resource resource %p: refcnt %d++",
1160 (void *)cache_resource,
1161 rte_atomic32_read(&cache_resource->refcnt));
1162 rte_atomic32_inc(&cache_resource->refcnt);
1163 dev_flow->dv.jump = cache_resource;
1167 /* Register new jump table resource. */
1168 cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
1169 if (!cache_resource)
1170 return rte_flow_error_set(error, ENOMEM,
1171 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1172 "cannot allocate resource memory");
1173 *cache_resource = *resource;
1174 cache_resource->action =
1175 mlx5_glue->dr_create_flow_action_dest_flow_tbl
1176 (resource->tbl->obj);
1177 if (!cache_resource->action) {
1178 rte_free(cache_resource);
1179 return rte_flow_error_set(error, ENOMEM,
1180 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1181 NULL, "cannot create action");
1183 rte_atomic32_init(&cache_resource->refcnt);
1184 rte_atomic32_inc(&cache_resource->refcnt);
1185 LIST_INSERT_HEAD(&sh->jump_tbl, cache_resource, next);
1186 dev_flow->dv.jump = cache_resource;
1187 DRV_LOG(DEBUG, "new jump table resource %p: refcnt %d++",
1188 (void *)cache_resource,
1189 rte_atomic32_read(&cache_resource->refcnt));
1194 * Find existing table port ID resource or create and register a new one.
1196 * @param dev[in, out]
1197 * Pointer to rte_eth_dev structure.
1198 * @param[in, out] resource
1199 * Pointer to port ID action resource.
1200 * @parm[in, out] dev_flow
1201 * Pointer to the dev_flow.
1203 * pointer to error structure.
1206 * 0 on success otherwise -errno and errno is set.
1209 flow_dv_port_id_action_resource_register
1210 (struct rte_eth_dev *dev,
1211 struct mlx5_flow_dv_port_id_action_resource *resource,
1212 struct mlx5_flow *dev_flow,
1213 struct rte_flow_error *error)
1215 struct mlx5_priv *priv = dev->data->dev_private;
1216 struct mlx5_ibv_shared *sh = priv->sh;
1217 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
1219 /* Lookup a matching resource from cache. */
1220 LIST_FOREACH(cache_resource, &sh->port_id_action_list, next) {
1221 if (resource->port_id == cache_resource->port_id) {
1222 DRV_LOG(DEBUG, "port id action resource resource %p: "
1224 (void *)cache_resource,
1225 rte_atomic32_read(&cache_resource->refcnt));
1226 rte_atomic32_inc(&cache_resource->refcnt);
1227 dev_flow->dv.port_id_action = cache_resource;
1231 /* Register new port id action resource. */
1232 cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
1233 if (!cache_resource)
1234 return rte_flow_error_set(error, ENOMEM,
1235 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1236 "cannot allocate resource memory");
1237 *cache_resource = *resource;
1238 cache_resource->action =
1239 mlx5_glue->dr_create_flow_action_dest_vport
1240 (priv->sh->fdb_domain, resource->port_id);
1241 if (!cache_resource->action) {
1242 rte_free(cache_resource);
1243 return rte_flow_error_set(error, ENOMEM,
1244 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1245 NULL, "cannot create action");
1247 rte_atomic32_init(&cache_resource->refcnt);
1248 rte_atomic32_inc(&cache_resource->refcnt);
1249 LIST_INSERT_HEAD(&sh->port_id_action_list, cache_resource, next);
1250 dev_flow->dv.port_id_action = cache_resource;
1251 DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
1252 (void *)cache_resource,
1253 rte_atomic32_read(&cache_resource->refcnt));
1258 * Get the size of specific rte_flow_item_type
1260 * @param[in] item_type
1261 * Tested rte_flow_item_type.
1264 * sizeof struct item_type, 0 if void or irrelevant.
1267 flow_dv_get_item_len(const enum rte_flow_item_type item_type)
1271 switch (item_type) {
1272 case RTE_FLOW_ITEM_TYPE_ETH:
1273 retval = sizeof(struct rte_flow_item_eth);
1275 case RTE_FLOW_ITEM_TYPE_VLAN:
1276 retval = sizeof(struct rte_flow_item_vlan);
1278 case RTE_FLOW_ITEM_TYPE_IPV4:
1279 retval = sizeof(struct rte_flow_item_ipv4);
1281 case RTE_FLOW_ITEM_TYPE_IPV6:
1282 retval = sizeof(struct rte_flow_item_ipv6);
1284 case RTE_FLOW_ITEM_TYPE_UDP:
1285 retval = sizeof(struct rte_flow_item_udp);
1287 case RTE_FLOW_ITEM_TYPE_TCP:
1288 retval = sizeof(struct rte_flow_item_tcp);
1290 case RTE_FLOW_ITEM_TYPE_VXLAN:
1291 retval = sizeof(struct rte_flow_item_vxlan);
1293 case RTE_FLOW_ITEM_TYPE_GRE:
1294 retval = sizeof(struct rte_flow_item_gre);
1296 case RTE_FLOW_ITEM_TYPE_NVGRE:
1297 retval = sizeof(struct rte_flow_item_nvgre);
1299 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1300 retval = sizeof(struct rte_flow_item_vxlan_gpe);
1302 case RTE_FLOW_ITEM_TYPE_MPLS:
1303 retval = sizeof(struct rte_flow_item_mpls);
1305 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
1313 #define MLX5_ENCAP_IPV4_VERSION 0x40
1314 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
1315 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
1316 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
1317 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
1318 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
1319 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
1322 * Convert the encap action data from list of rte_flow_item to raw buffer
1325 * Pointer to rte_flow_item objects list.
1327 * Pointer to the output buffer.
1329 * Pointer to the output buffer size.
1331 * Pointer to the error structure.
1334 * 0 on success, a negative errno value otherwise and rte_errno is set.
1337 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
1338 size_t *size, struct rte_flow_error *error)
1340 struct rte_ether_hdr *eth = NULL;
1341 struct rte_vlan_hdr *vlan = NULL;
1342 struct rte_ipv4_hdr *ipv4 = NULL;
1343 struct rte_ipv6_hdr *ipv6 = NULL;
1344 struct rte_udp_hdr *udp = NULL;
1345 struct rte_vxlan_hdr *vxlan = NULL;
1346 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
1347 struct rte_gre_hdr *gre = NULL;
1349 size_t temp_size = 0;
1352 return rte_flow_error_set(error, EINVAL,
1353 RTE_FLOW_ERROR_TYPE_ACTION,
1354 NULL, "invalid empty data");
1355 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1356 len = flow_dv_get_item_len(items->type);
1357 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
1358 return rte_flow_error_set(error, EINVAL,
1359 RTE_FLOW_ERROR_TYPE_ACTION,
1360 (void *)items->type,
1361 "items total size is too big"
1362 " for encap action");
1363 rte_memcpy((void *)&buf[temp_size], items->spec, len);
1364 switch (items->type) {
1365 case RTE_FLOW_ITEM_TYPE_ETH:
1366 eth = (struct rte_ether_hdr *)&buf[temp_size];
1368 case RTE_FLOW_ITEM_TYPE_VLAN:
1369 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
1371 return rte_flow_error_set(error, EINVAL,
1372 RTE_FLOW_ERROR_TYPE_ACTION,
1373 (void *)items->type,
1374 "eth header not found");
1375 if (!eth->ether_type)
1376 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
1378 case RTE_FLOW_ITEM_TYPE_IPV4:
1379 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
1381 return rte_flow_error_set(error, EINVAL,
1382 RTE_FLOW_ERROR_TYPE_ACTION,
1383 (void *)items->type,
1384 "neither eth nor vlan"
1386 if (vlan && !vlan->eth_proto)
1387 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
1388 else if (eth && !eth->ether_type)
1389 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
1390 if (!ipv4->version_ihl)
1391 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
1392 MLX5_ENCAP_IPV4_IHL_MIN;
1393 if (!ipv4->time_to_live)
1394 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
1396 case RTE_FLOW_ITEM_TYPE_IPV6:
1397 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
1399 return rte_flow_error_set(error, EINVAL,
1400 RTE_FLOW_ERROR_TYPE_ACTION,
1401 (void *)items->type,
1402 "neither eth nor vlan"
1404 if (vlan && !vlan->eth_proto)
1405 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
1406 else if (eth && !eth->ether_type)
1407 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
1408 if (!ipv6->vtc_flow)
1410 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
1411 if (!ipv6->hop_limits)
1412 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
1414 case RTE_FLOW_ITEM_TYPE_UDP:
1415 udp = (struct rte_udp_hdr *)&buf[temp_size];
1417 return rte_flow_error_set(error, EINVAL,
1418 RTE_FLOW_ERROR_TYPE_ACTION,
1419 (void *)items->type,
1420 "ip header not found");
1421 if (ipv4 && !ipv4->next_proto_id)
1422 ipv4->next_proto_id = IPPROTO_UDP;
1423 else if (ipv6 && !ipv6->proto)
1424 ipv6->proto = IPPROTO_UDP;
1426 case RTE_FLOW_ITEM_TYPE_VXLAN:
1427 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
1429 return rte_flow_error_set(error, EINVAL,
1430 RTE_FLOW_ERROR_TYPE_ACTION,
1431 (void *)items->type,
1432 "udp header not found");
1434 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
1435 if (!vxlan->vx_flags)
1437 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
1439 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1440 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
1442 return rte_flow_error_set(error, EINVAL,
1443 RTE_FLOW_ERROR_TYPE_ACTION,
1444 (void *)items->type,
1445 "udp header not found");
1446 if (!vxlan_gpe->proto)
1447 return rte_flow_error_set(error, EINVAL,
1448 RTE_FLOW_ERROR_TYPE_ACTION,
1449 (void *)items->type,
1450 "next protocol not found");
1453 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
1454 if (!vxlan_gpe->vx_flags)
1455 vxlan_gpe->vx_flags =
1456 MLX5_ENCAP_VXLAN_GPE_FLAGS;
1458 case RTE_FLOW_ITEM_TYPE_GRE:
1459 case RTE_FLOW_ITEM_TYPE_NVGRE:
1460 gre = (struct rte_gre_hdr *)&buf[temp_size];
1462 return rte_flow_error_set(error, EINVAL,
1463 RTE_FLOW_ERROR_TYPE_ACTION,
1464 (void *)items->type,
1465 "next protocol not found");
1467 return rte_flow_error_set(error, EINVAL,
1468 RTE_FLOW_ERROR_TYPE_ACTION,
1469 (void *)items->type,
1470 "ip header not found");
1471 if (ipv4 && !ipv4->next_proto_id)
1472 ipv4->next_proto_id = IPPROTO_GRE;
1473 else if (ipv6 && !ipv6->proto)
1474 ipv6->proto = IPPROTO_GRE;
1476 case RTE_FLOW_ITEM_TYPE_VOID:
1479 return rte_flow_error_set(error, EINVAL,
1480 RTE_FLOW_ERROR_TYPE_ACTION,
1481 (void *)items->type,
1482 "unsupported item type");
1492 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
1494 struct rte_ether_hdr *eth = NULL;
1495 struct rte_vlan_hdr *vlan = NULL;
1496 struct rte_ipv6_hdr *ipv6 = NULL;
1497 struct rte_udp_hdr *udp = NULL;
1501 eth = (struct rte_ether_hdr *)data;
1502 next_hdr = (char *)(eth + 1);
1503 proto = RTE_BE16(eth->ether_type);
1506 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
1507 next_hdr += sizeof(struct rte_vlan_hdr);
1508 vlan = (struct rte_vlan_hdr *)next_hdr;
1509 proto = RTE_BE16(vlan->eth_proto);
1512 /* HW calculates IPv4 csum. no need to proceed */
1513 if (proto == RTE_ETHER_TYPE_IPV4)
1516 /* non IPv4/IPv6 header. not supported */
1517 if (proto != RTE_ETHER_TYPE_IPV6) {
1518 return rte_flow_error_set(error, ENOTSUP,
1519 RTE_FLOW_ERROR_TYPE_ACTION,
1520 NULL, "Cannot offload non IPv4/IPv6");
1523 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
1525 /* ignore non UDP */
1526 if (ipv6->proto != IPPROTO_UDP)
1529 udp = (struct rte_udp_hdr *)(ipv6 + 1);
1530 udp->dgram_cksum = 0;
1536 * Convert L2 encap action to DV specification.
1539 * Pointer to rte_eth_dev structure.
1541 * Pointer to action structure.
1542 * @param[in, out] dev_flow
1543 * Pointer to the mlx5_flow.
1544 * @param[in] transfer
1545 * Mark if the flow is E-Switch flow.
1547 * Pointer to the error structure.
1550 * 0 on success, a negative errno value otherwise and rte_errno is set.
1553 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
1554 const struct rte_flow_action *action,
1555 struct mlx5_flow *dev_flow,
1557 struct rte_flow_error *error)
1559 const struct rte_flow_item *encap_data;
1560 const struct rte_flow_action_raw_encap *raw_encap_data;
1561 struct mlx5_flow_dv_encap_decap_resource res = {
1563 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
1564 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
1565 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
1568 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
1570 (const struct rte_flow_action_raw_encap *)action->conf;
1571 res.size = raw_encap_data->size;
1572 memcpy(res.buf, raw_encap_data->data, res.size);
1573 if (flow_dv_zero_encap_udp_csum(res.buf, error))
1576 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
1578 ((const struct rte_flow_action_vxlan_encap *)
1579 action->conf)->definition;
1582 ((const struct rte_flow_action_nvgre_encap *)
1583 action->conf)->definition;
1584 if (flow_dv_convert_encap_data(encap_data, res.buf,
1588 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
1589 return rte_flow_error_set(error, EINVAL,
1590 RTE_FLOW_ERROR_TYPE_ACTION,
1591 NULL, "can't create L2 encap action");
1596 * Convert L2 decap action to DV specification.
1599 * Pointer to rte_eth_dev structure.
1600 * @param[in, out] dev_flow
1601 * Pointer to the mlx5_flow.
1602 * @param[in] transfer
1603 * Mark if the flow is E-Switch flow.
1605 * Pointer to the error structure.
1608 * 0 on success, a negative errno value otherwise and rte_errno is set.
1611 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
1612 struct mlx5_flow *dev_flow,
1614 struct rte_flow_error *error)
1616 struct mlx5_flow_dv_encap_decap_resource res = {
1619 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
1620 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
1621 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
1624 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
1625 return rte_flow_error_set(error, EINVAL,
1626 RTE_FLOW_ERROR_TYPE_ACTION,
1627 NULL, "can't create L2 decap action");
1632 * Convert raw decap/encap (L3 tunnel) action to DV specification.
1635 * Pointer to rte_eth_dev structure.
1637 * Pointer to action structure.
1638 * @param[in, out] dev_flow
1639 * Pointer to the mlx5_flow.
1641 * Pointer to the flow attributes.
1643 * Pointer to the error structure.
1646 * 0 on success, a negative errno value otherwise and rte_errno is set.
1649 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
1650 const struct rte_flow_action *action,
1651 struct mlx5_flow *dev_flow,
1652 const struct rte_flow_attr *attr,
1653 struct rte_flow_error *error)
1655 const struct rte_flow_action_raw_encap *encap_data;
1656 struct mlx5_flow_dv_encap_decap_resource res;
1658 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
1659 res.size = encap_data->size;
1660 memcpy(res.buf, encap_data->data, res.size);
1661 res.reformat_type = attr->egress ?
1662 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL :
1663 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2;
1665 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
1667 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
1668 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
1669 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
1670 return rte_flow_error_set(error, EINVAL,
1671 RTE_FLOW_ERROR_TYPE_ACTION,
1672 NULL, "can't create encap action");
1677 * Validate the modify-header actions.
1679 * @param[in] action_flags
1680 * Holds the actions detected until now.
1682 * Pointer to the modify action.
1684 * Pointer to error structure.
1687 * 0 on success, a negative errno value otherwise and rte_errno is set.
1690 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
1691 const struct rte_flow_action *action,
1692 struct rte_flow_error *error)
1694 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
1695 return rte_flow_error_set(error, EINVAL,
1696 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1697 NULL, "action configuration not set");
1698 if (action_flags & MLX5_FLOW_ENCAP_ACTIONS)
1699 return rte_flow_error_set(error, EINVAL,
1700 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1701 "can't have encap action before"
1707 * Validate the modify-header MAC address actions.
1709 * @param[in] action_flags
1710 * Holds the actions detected until now.
1712 * Pointer to the modify action.
1713 * @param[in] item_flags
1714 * Holds the items detected.
1716 * Pointer to error structure.
1719 * 0 on success, a negative errno value otherwise and rte_errno is set.
1722 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
1723 const struct rte_flow_action *action,
1724 const uint64_t item_flags,
1725 struct rte_flow_error *error)
1729 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1731 if (!(item_flags & MLX5_FLOW_LAYER_L2))
1732 return rte_flow_error_set(error, EINVAL,
1733 RTE_FLOW_ERROR_TYPE_ACTION,
1735 "no L2 item in pattern");
1741 * Validate the modify-header IPv4 address actions.
1743 * @param[in] action_flags
1744 * Holds the actions detected until now.
1746 * Pointer to the modify action.
1747 * @param[in] item_flags
1748 * Holds the items detected.
1750 * Pointer to error structure.
1753 * 0 on success, a negative errno value otherwise and rte_errno is set.
1756 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
1757 const struct rte_flow_action *action,
1758 const uint64_t item_flags,
1759 struct rte_flow_error *error)
1763 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1765 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
1766 return rte_flow_error_set(error, EINVAL,
1767 RTE_FLOW_ERROR_TYPE_ACTION,
1769 "no ipv4 item in pattern");
1775 * Validate the modify-header IPv6 address actions.
1777 * @param[in] action_flags
1778 * Holds the actions detected until now.
1780 * Pointer to the modify action.
1781 * @param[in] item_flags
1782 * Holds the items detected.
1784 * Pointer to error structure.
1787 * 0 on success, a negative errno value otherwise and rte_errno is set.
1790 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
1791 const struct rte_flow_action *action,
1792 const uint64_t item_flags,
1793 struct rte_flow_error *error)
1797 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1799 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
1800 return rte_flow_error_set(error, EINVAL,
1801 RTE_FLOW_ERROR_TYPE_ACTION,
1803 "no ipv6 item in pattern");
1809 * Validate the modify-header TP actions.
1811 * @param[in] action_flags
1812 * Holds the actions detected until now.
1814 * Pointer to the modify action.
1815 * @param[in] item_flags
1816 * Holds the items detected.
1818 * Pointer to error structure.
1821 * 0 on success, a negative errno value otherwise and rte_errno is set.
1824 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
1825 const struct rte_flow_action *action,
1826 const uint64_t item_flags,
1827 struct rte_flow_error *error)
1831 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1833 if (!(item_flags & MLX5_FLOW_LAYER_L4))
1834 return rte_flow_error_set(error, EINVAL,
1835 RTE_FLOW_ERROR_TYPE_ACTION,
1836 NULL, "no transport layer "
1843 * Validate the modify-header actions of increment/decrement
1844 * TCP Sequence-number.
1846 * @param[in] action_flags
1847 * Holds the actions detected until now.
1849 * Pointer to the modify action.
1850 * @param[in] item_flags
1851 * Holds the items detected.
1853 * Pointer to error structure.
1856 * 0 on success, a negative errno value otherwise and rte_errno is set.
1859 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
1860 const struct rte_flow_action *action,
1861 const uint64_t item_flags,
1862 struct rte_flow_error *error)
1866 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1868 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP))
1869 return rte_flow_error_set(error, EINVAL,
1870 RTE_FLOW_ERROR_TYPE_ACTION,
1871 NULL, "no TCP item in"
1873 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
1874 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
1875 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
1876 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
1877 return rte_flow_error_set(error, EINVAL,
1878 RTE_FLOW_ERROR_TYPE_ACTION,
1880 "cannot decrease and increase"
1881 " TCP sequence number"
1882 " at the same time");
1888 * Validate the modify-header actions of increment/decrement
1889 * TCP Acknowledgment number.
1891 * @param[in] action_flags
1892 * Holds the actions detected until now.
1894 * Pointer to the modify action.
1895 * @param[in] item_flags
1896 * Holds the items detected.
1898 * Pointer to error structure.
1901 * 0 on success, a negative errno value otherwise and rte_errno is set.
1904 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
1905 const struct rte_flow_action *action,
1906 const uint64_t item_flags,
1907 struct rte_flow_error *error)
1911 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1913 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP))
1914 return rte_flow_error_set(error, EINVAL,
1915 RTE_FLOW_ERROR_TYPE_ACTION,
1916 NULL, "no TCP item in"
1918 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
1919 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
1920 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
1921 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
1922 return rte_flow_error_set(error, EINVAL,
1923 RTE_FLOW_ERROR_TYPE_ACTION,
1925 "cannot decrease and increase"
1926 " TCP acknowledgment number"
1927 " at the same time");
1933 * Validate the modify-header TTL actions.
1935 * @param[in] action_flags
1936 * Holds the actions detected until now.
1938 * Pointer to the modify action.
1939 * @param[in] item_flags
1940 * Holds the items detected.
1942 * Pointer to error structure.
1945 * 0 on success, a negative errno value otherwise and rte_errno is set.
1948 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
1949 const struct rte_flow_action *action,
1950 const uint64_t item_flags,
1951 struct rte_flow_error *error)
1955 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
1957 if (!(item_flags & MLX5_FLOW_LAYER_L3))
1958 return rte_flow_error_set(error, EINVAL,
1959 RTE_FLOW_ERROR_TYPE_ACTION,
1961 "no IP protocol in pattern");
1967 * Validate jump action.
1970 * Pointer to the modify action.
1972 * The group of the current flow.
1974 * Pointer to error structure.
1977 * 0 on success, a negative errno value otherwise and rte_errno is set.
1980 flow_dv_validate_action_jump(const struct rte_flow_action *action,
1982 struct rte_flow_error *error)
1984 if (action->type != RTE_FLOW_ACTION_TYPE_JUMP && !action->conf)
1985 return rte_flow_error_set(error, EINVAL,
1986 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1987 NULL, "action configuration not set");
1988 if (group >= ((const struct rte_flow_action_jump *)action->conf)->group)
1989 return rte_flow_error_set(error, EINVAL,
1990 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1991 "target group must be higher then"
1992 " the current flow group");
1997 * Validate the port_id action.
2000 * Pointer to rte_eth_dev structure.
2001 * @param[in] action_flags
2002 * Bit-fields that holds the actions detected until now.
2004 * Port_id RTE action structure.
2006 * Attributes of flow that includes this action.
2008 * Pointer to error structure.
2011 * 0 on success, a negative errno value otherwise and rte_errno is set.
2014 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
2015 uint64_t action_flags,
2016 const struct rte_flow_action *action,
2017 const struct rte_flow_attr *attr,
2018 struct rte_flow_error *error)
2020 const struct rte_flow_action_port_id *port_id;
2022 uint16_t esw_domain_id;
2023 uint16_t act_port_domain_id;
2026 if (!attr->transfer)
2027 return rte_flow_error_set(error, ENOTSUP,
2028 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2030 "port id action is valid in transfer"
2032 if (!action || !action->conf)
2033 return rte_flow_error_set(error, ENOTSUP,
2034 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2036 "port id action parameters must be"
2038 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
2039 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
2040 return rte_flow_error_set(error, EINVAL,
2041 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2042 "can have only one fate actions in"
2044 ret = mlx5_port_to_eswitch_info(dev->data->port_id,
2045 &esw_domain_id, NULL);
2047 return rte_flow_error_set(error, -ret,
2048 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2050 "failed to obtain E-Switch info");
2051 port_id = action->conf;
2052 port = port_id->original ? dev->data->port_id : port_id->id;
2053 ret = mlx5_port_to_eswitch_info(port, &act_port_domain_id, NULL);
2055 return rte_flow_error_set
2057 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
2058 "failed to obtain E-Switch port id for port");
2059 if (act_port_domain_id != esw_domain_id)
2060 return rte_flow_error_set
2062 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2063 "port does not belong to"
2064 " E-Switch being configured");
2069 * Find existing modify-header resource or create and register a new one.
2071 * @param dev[in, out]
2072 * Pointer to rte_eth_dev structure.
2073 * @param[in, out] resource
2074 * Pointer to modify-header resource.
2075 * @parm[in, out] dev_flow
2076 * Pointer to the dev_flow.
2078 * pointer to error structure.
2081 * 0 on success otherwise -errno and errno is set.
2084 flow_dv_modify_hdr_resource_register
2085 (struct rte_eth_dev *dev,
2086 struct mlx5_flow_dv_modify_hdr_resource *resource,
2087 struct mlx5_flow *dev_flow,
2088 struct rte_flow_error *error)
2090 struct mlx5_priv *priv = dev->data->dev_private;
2091 struct mlx5_ibv_shared *sh = priv->sh;
2092 struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
2093 struct mlx5dv_dr_domain *ns;
2095 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2096 ns = sh->fdb_domain;
2097 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
2102 dev_flow->flow->group ? 0 : MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
2103 /* Lookup a matching resource from cache. */
2104 LIST_FOREACH(cache_resource, &sh->modify_cmds, next) {
2105 if (resource->ft_type == cache_resource->ft_type &&
2106 resource->actions_num == cache_resource->actions_num &&
2107 resource->flags == cache_resource->flags &&
2108 !memcmp((const void *)resource->actions,
2109 (const void *)cache_resource->actions,
2110 (resource->actions_num *
2111 sizeof(resource->actions[0])))) {
2112 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
2113 (void *)cache_resource,
2114 rte_atomic32_read(&cache_resource->refcnt));
2115 rte_atomic32_inc(&cache_resource->refcnt);
2116 dev_flow->dv.modify_hdr = cache_resource;
2120 /* Register new modify-header resource. */
2121 cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
2122 if (!cache_resource)
2123 return rte_flow_error_set(error, ENOMEM,
2124 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2125 "cannot allocate resource memory");
2126 *cache_resource = *resource;
2127 cache_resource->verbs_action =
2128 mlx5_glue->dv_create_flow_action_modify_header
2129 (sh->ctx, cache_resource->ft_type,
2130 ns, cache_resource->flags,
2131 cache_resource->actions_num *
2132 sizeof(cache_resource->actions[0]),
2133 (uint64_t *)cache_resource->actions);
2134 if (!cache_resource->verbs_action) {
2135 rte_free(cache_resource);
2136 return rte_flow_error_set(error, ENOMEM,
2137 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2138 NULL, "cannot create action");
2140 rte_atomic32_init(&cache_resource->refcnt);
2141 rte_atomic32_inc(&cache_resource->refcnt);
2142 LIST_INSERT_HEAD(&sh->modify_cmds, cache_resource, next);
2143 dev_flow->dv.modify_hdr = cache_resource;
2144 DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
2145 (void *)cache_resource,
2146 rte_atomic32_read(&cache_resource->refcnt));
2150 #define MLX5_CNT_CONTAINER_RESIZE 64
2151 #define MLX5_CNT_CONTAINER(priv, batch) (&(priv)->sh->cmng.ccont[batch])
2154 * Get a pool by a counter.
2157 * Pointer to the counter.
2162 static struct mlx5_flow_counter_pool *
2163 flow_dv_counter_pool_get(struct mlx5_flow_counter *cnt)
2166 cnt -= cnt->dcs->id % MLX5_COUNTERS_PER_POOL;
2167 return (struct mlx5_flow_counter_pool *)cnt - 1;
2173 * Get a pool by devx counter ID.
2176 * Pointer to the counter container.
2178 * The counter devx ID.
2181 * The counter pool pointer if exists, NULL otherwise,
2183 static struct mlx5_flow_counter_pool *
2184 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
2186 struct mlx5_flow_counter_pool *pool;
2188 TAILQ_FOREACH(pool, &cont->pool_list, next) {
2189 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
2190 MLX5_COUNTERS_PER_POOL;
2192 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
2199 * Allocate a new memory for the counter values wrapped by all the needed
2203 * Pointer to the Ethernet device structure.
2205 * The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
2208 * The new memory management pointer on success, otherwise NULL and rte_errno
2211 static struct mlx5_counter_stats_mem_mng *
2212 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
2214 struct mlx5_ibv_shared *sh = ((struct mlx5_priv *)
2215 (dev->data->dev_private))->sh;
2216 struct mlx5dv_pd dv_pd;
2217 struct mlx5dv_obj dv_obj;
2218 struct mlx5_devx_mkey_attr mkey_attr;
2219 struct mlx5_counter_stats_mem_mng *mem_mng;
2220 volatile struct flow_counter_stats *raw_data;
2221 int size = (sizeof(struct flow_counter_stats) *
2222 MLX5_COUNTERS_PER_POOL +
2223 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
2224 sizeof(struct mlx5_counter_stats_mem_mng);
2225 uint8_t *mem = rte_calloc(__func__, 1, size, sysconf(_SC_PAGESIZE));
2232 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
2233 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
2234 mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
2235 IBV_ACCESS_LOCAL_WRITE);
2236 if (!mem_mng->umem) {
2241 dv_obj.pd.in = sh->pd;
2242 dv_obj.pd.out = &dv_pd;
2243 mlx5_glue->dv_init_obj(&dv_obj, MLX5DV_OBJ_PD);
2244 mkey_attr.addr = (uintptr_t)mem;
2245 mkey_attr.size = size;
2246 mkey_attr.umem_id = mem_mng->umem->umem_id;
2247 mkey_attr.pd = dv_pd.pdn;
2248 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
2250 mlx5_glue->devx_umem_dereg(mem_mng->umem);
2255 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
2256 raw_data = (volatile struct flow_counter_stats *)mem;
2257 for (i = 0; i < raws_n; ++i) {
2258 mem_mng->raws[i].mem_mng = mem_mng;
2259 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
2261 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
2266 * Resize a counter container.
2269 * Pointer to the Ethernet device structure.
2271 * Whether the pool is for counter that was allocated by batch command.
2274 * The container pointer on success, otherwise NULL and rte_errno is set.
2276 static struct mlx5_pools_container *
2277 flow_dv_container_resize(struct rte_eth_dev *dev, uint32_t batch)
2279 struct mlx5_priv *priv = dev->data->dev_private;
2280 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv, batch);
2281 struct mlx5_counter_stats_mem_mng *mem_mng;
2282 uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
2283 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
2284 struct mlx5_flow_counter_pool **new_pools = rte_calloc(__func__, 1,
2290 mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
2291 MLX5_CNT_CONTAINER_RESIZE);
2293 rte_free(new_pools);
2297 memcpy(new_pools, cont->pools,
2298 cont->n * sizeof(struct mlx5_flow_counter_pool *));
2299 rte_free(cont->pools);
2301 TAILQ_INIT(&cont->pool_list);
2303 cont->pools = new_pools;
2305 cont->init_mem_mng = mem_mng;
2310 * Query a devx flow counter.
2313 * Pointer to the Ethernet device structure.
2315 * Pointer to the flow counter.
2317 * The statistics value of packets.
2319 * The statistics value of bytes.
2322 * 0 on success, otherwise a negative errno value and rte_errno is set.
2325 _flow_dv_query_count(struct rte_eth_dev *dev __rte_unused,
2326 struct mlx5_flow_counter *cnt, uint64_t *pkts,
2329 struct mlx5_flow_counter_pool *pool =
2330 flow_dv_counter_pool_get(cnt);
2331 uint16_t offset = pool->min_dcs->id % MLX5_COUNTERS_PER_POOL;
2332 int ret = mlx5_devx_cmd_flow_counter_query
2333 (pool->min_dcs, 0, MLX5_COUNTERS_PER_POOL - offset, NULL,
2334 NULL, pool->raw->mem_mng->dm->id,
2335 (void *)(uintptr_t)(pool->raw->data +
2339 DRV_LOG(ERR, "Failed to trigger synchronous"
2340 " query for dcs ID %d\n",
2344 offset = cnt - &pool->counters_raw[0];
2345 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
2346 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
2351 * Create and initialize a new counter pool.
2354 * Pointer to the Ethernet device structure.
2356 * The devX counter handle.
2358 * Whether the pool is for counter that was allocated by batch command.
2361 * A new pool pointer on success, NULL otherwise and rte_errno is set.
2363 static struct mlx5_flow_counter_pool *
2364 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
2367 struct mlx5_priv *priv = dev->data->dev_private;
2368 struct mlx5_flow_counter_pool *pool;
2369 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv, batch);
2372 if (cont->n == cont->n_valid) {
2373 cont = flow_dv_container_resize(dev, batch);
2377 size = sizeof(*pool) + MLX5_COUNTERS_PER_POOL *
2378 sizeof(struct mlx5_flow_counter);
2379 pool = rte_calloc(__func__, 1, size, 0);
2384 pool->min_dcs = dcs;
2385 pool->raw = cont->init_mem_mng->raws + cont->n_valid %
2386 MLX5_CNT_CONTAINER_RESIZE;
2387 TAILQ_INIT(&pool->counters);
2388 TAILQ_INSERT_TAIL(&cont->pool_list, pool, next);
2389 cont->pools[cont->n_valid] = pool;
2395 * Prepare a new counter and/or a new counter pool.
2398 * Pointer to the Ethernet device structure.
2399 * @param[out] cnt_free
2400 * Where to put the pointer of a new counter.
2402 * Whether the pool is for counter that was allocated by batch command.
2405 * The free counter pool pointer and @p cnt_free is set on success,
2406 * NULL otherwise and rte_errno is set.
2408 static struct mlx5_flow_counter_pool *
2409 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
2410 struct mlx5_flow_counter **cnt_free,
2413 struct mlx5_priv *priv = dev->data->dev_private;
2414 struct mlx5_flow_counter_pool *pool;
2415 struct mlx5_devx_obj *dcs = NULL;
2416 struct mlx5_flow_counter *cnt;
2420 /* bulk_bitmap must be 0 for single counter allocation. */
2421 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
2424 pool = flow_dv_find_pool_by_id(MLX5_CNT_CONTAINER(priv, batch),
2427 pool = flow_dv_pool_create(dev, dcs, batch);
2429 mlx5_devx_cmd_destroy(dcs);
2432 } else if (dcs->id < pool->min_dcs->id) {
2433 pool->min_dcs->id = dcs->id;
2435 cnt = &pool->counters_raw[dcs->id % MLX5_COUNTERS_PER_POOL];
2436 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
2441 /* bulk_bitmap is in 128 counters units. */
2442 if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
2443 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
2445 rte_errno = ENODATA;
2448 pool = flow_dv_pool_create(dev, dcs, batch);
2450 mlx5_devx_cmd_destroy(dcs);
2453 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
2454 cnt = &pool->counters_raw[i];
2456 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
2458 *cnt_free = &pool->counters_raw[0];
2463 * Search for existed shared counter.
2466 * Pointer to the relevant counter pool container.
2468 * The shared counter ID to search.
2471 * NULL if not existed, otherwise pointer to the shared counter.
2473 static struct mlx5_flow_counter *
2474 flow_dv_counter_shared_search(struct mlx5_pools_container *cont,
2477 static struct mlx5_flow_counter *cnt;
2478 struct mlx5_flow_counter_pool *pool;
2481 TAILQ_FOREACH(pool, &cont->pool_list, next) {
2482 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
2483 cnt = &pool->counters_raw[i];
2484 if (cnt->ref_cnt && cnt->shared && cnt->id == id)
2492 * Allocate a flow counter.
2495 * Pointer to the Ethernet device structure.
2497 * Indicate if this counter is shared with other flows.
2499 * Counter identifier.
2501 * Counter flow group.
2504 * pointer to flow counter on success, NULL otherwise and rte_errno is set.
2506 static struct mlx5_flow_counter *
2507 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
2510 struct mlx5_priv *priv = dev->data->dev_private;
2511 struct mlx5_flow_counter_pool *pool = NULL;
2512 struct mlx5_flow_counter *cnt_free = NULL;
2514 * Currently group 0 flow counter cannot be assigned to a flow if it is
2515 * not the first one in the batch counter allocation, so it is better
2516 * to allocate counters one by one for these flows in a separate
2518 * A counter can be shared between different groups so need to take
2519 * shared counters from the single container.
2521 uint32_t batch = (group && !shared) ? 1 : 0;
2522 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv, batch);
2524 if (!priv->config.devx) {
2525 rte_errno = ENOTSUP;
2529 cnt_free = flow_dv_counter_shared_search(cont, id);
2531 if (cnt_free->ref_cnt + 1 == 0) {
2535 cnt_free->ref_cnt++;
2539 /* Pools which has a free counters are in the start. */
2540 pool = TAILQ_FIRST(&cont->pool_list);
2542 cnt_free = TAILQ_FIRST(&pool->counters);
2544 pool = flow_dv_counter_pool_prepare(dev, &cnt_free, batch);
2548 cnt_free->batch = batch;
2549 /* Create a DV counter action only in the first time usage. */
2550 if (!cnt_free->action) {
2552 struct mlx5_devx_obj *dcs;
2555 offset = cnt_free - &pool->counters_raw[0];
2556 dcs = pool->min_dcs;
2559 dcs = cnt_free->dcs;
2561 cnt_free->action = mlx5_glue->dv_create_flow_action_counter
2563 if (!cnt_free->action) {
2568 /* Update the counter reset values. */
2569 if (_flow_dv_query_count(dev, cnt_free, &cnt_free->hits,
2572 cnt_free->shared = shared;
2573 cnt_free->ref_cnt = 1;
2575 TAILQ_REMOVE(&pool->counters, cnt_free, next);
2576 if (TAILQ_EMPTY(&pool->counters)) {
2577 /* Move the pool to the end of the container pool list. */
2578 TAILQ_REMOVE(&cont->pool_list, pool, next);
2579 TAILQ_INSERT_TAIL(&cont->pool_list, pool, next);
2585 * Release a flow counter.
2588 * Pointer to the Ethernet device structure.
2589 * @param[in] counter
2590 * Pointer to the counter handler.
2593 flow_dv_counter_release(struct rte_eth_dev *dev __rte_unused,
2594 struct mlx5_flow_counter *counter)
2598 if (--counter->ref_cnt == 0) {
2599 struct mlx5_flow_counter_pool *pool =
2600 flow_dv_counter_pool_get(counter);
2602 /* Put the counter in the end - the earliest one. */
2603 TAILQ_INSERT_TAIL(&pool->counters, counter, next);
2608 * Verify the @p attributes will be correctly understood by the NIC and store
2609 * them in the @p flow if everything is correct.
2612 * Pointer to dev struct.
2613 * @param[in] attributes
2614 * Pointer to flow attributes
2616 * Pointer to error structure.
2619 * 0 on success, a negative errno value otherwise and rte_errno is set.
2622 flow_dv_validate_attributes(struct rte_eth_dev *dev,
2623 const struct rte_flow_attr *attributes,
2624 struct rte_flow_error *error)
2626 struct mlx5_priv *priv = dev->data->dev_private;
2627 uint32_t priority_max = priv->config.flow_prio - 1;
2629 #ifndef HAVE_MLX5DV_DR
2630 if (attributes->group)
2631 return rte_flow_error_set(error, ENOTSUP,
2632 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
2634 "groups is not supported");
2636 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
2637 attributes->priority >= priority_max)
2638 return rte_flow_error_set(error, ENOTSUP,
2639 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
2641 "priority out of range");
2642 if (attributes->transfer) {
2643 if (!priv->config.dv_esw_en)
2644 return rte_flow_error_set
2646 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2647 "E-Switch dr is not supported");
2648 if (!(priv->representor || priv->master))
2649 return rte_flow_error_set
2650 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2651 NULL, "E-Switch configurationd can only be"
2652 " done by a master or a representor device");
2653 if (attributes->egress)
2654 return rte_flow_error_set
2656 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
2657 "egress is not supported");
2658 if (attributes->group >= MLX5_MAX_TABLES_FDB)
2659 return rte_flow_error_set
2661 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
2662 NULL, "group must be smaller than "
2663 RTE_STR(MLX5_MAX_FDB_TABLES));
2665 if (!(attributes->egress ^ attributes->ingress))
2666 return rte_flow_error_set(error, ENOTSUP,
2667 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
2668 "must specify exactly one of "
2669 "ingress or egress");
2674 * Internal validation function. For validating both actions and items.
2677 * Pointer to the rte_eth_dev structure.
2679 * Pointer to the flow attributes.
2681 * Pointer to the list of items.
2682 * @param[in] actions
2683 * Pointer to the list of actions.
2685 * Pointer to the error structure.
2688 * 0 on success, a negative errno value otherwise and rte_errno is set.
2691 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
2692 const struct rte_flow_item items[],
2693 const struct rte_flow_action actions[],
2694 struct rte_flow_error *error)
2697 uint64_t action_flags = 0;
2698 uint64_t item_flags = 0;
2699 uint64_t last_item = 0;
2700 uint8_t next_protocol = 0xff;
2702 const struct rte_flow_item *gre_item = NULL;
2703 struct rte_flow_item_tcp nic_tcp_mask = {
2706 .src_port = RTE_BE16(UINT16_MAX),
2707 .dst_port = RTE_BE16(UINT16_MAX),
2713 ret = flow_dv_validate_attributes(dev, attr, error);
2716 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2717 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2718 switch (items->type) {
2719 case RTE_FLOW_ITEM_TYPE_VOID:
2721 case RTE_FLOW_ITEM_TYPE_PORT_ID:
2722 ret = flow_dv_validate_item_port_id
2723 (dev, items, attr, item_flags, error);
2726 last_item |= MLX5_FLOW_ITEM_PORT_ID;
2728 case RTE_FLOW_ITEM_TYPE_ETH:
2729 ret = mlx5_flow_validate_item_eth(items, item_flags,
2733 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
2734 MLX5_FLOW_LAYER_OUTER_L2;
2736 case RTE_FLOW_ITEM_TYPE_VLAN:
2737 ret = mlx5_flow_validate_item_vlan(items, item_flags,
2741 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2742 MLX5_FLOW_LAYER_OUTER_VLAN;
2744 case RTE_FLOW_ITEM_TYPE_IPV4:
2745 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
2749 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2750 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2751 if (items->mask != NULL &&
2752 ((const struct rte_flow_item_ipv4 *)
2753 items->mask)->hdr.next_proto_id) {
2755 ((const struct rte_flow_item_ipv4 *)
2756 (items->spec))->hdr.next_proto_id;
2758 ((const struct rte_flow_item_ipv4 *)
2759 (items->mask))->hdr.next_proto_id;
2761 /* Reset for inner layer. */
2762 next_protocol = 0xff;
2764 mlx5_flow_tunnel_ip_check(items, &last_item);
2766 case RTE_FLOW_ITEM_TYPE_IPV6:
2767 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
2771 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2772 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2773 if (items->mask != NULL &&
2774 ((const struct rte_flow_item_ipv6 *)
2775 items->mask)->hdr.proto) {
2777 ((const struct rte_flow_item_ipv6 *)
2778 items->spec)->hdr.proto;
2780 ((const struct rte_flow_item_ipv6 *)
2781 items->mask)->hdr.proto;
2783 /* Reset for inner layer. */
2784 next_protocol = 0xff;
2786 mlx5_flow_tunnel_ip_check(items, &last_item);
2788 case RTE_FLOW_ITEM_TYPE_TCP:
2789 ret = mlx5_flow_validate_item_tcp
2796 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
2797 MLX5_FLOW_LAYER_OUTER_L4_TCP;
2799 case RTE_FLOW_ITEM_TYPE_UDP:
2800 ret = mlx5_flow_validate_item_udp(items, item_flags,
2805 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
2806 MLX5_FLOW_LAYER_OUTER_L4_UDP;
2808 case RTE_FLOW_ITEM_TYPE_GRE:
2809 case RTE_FLOW_ITEM_TYPE_NVGRE:
2810 ret = mlx5_flow_validate_item_gre(items, item_flags,
2811 next_protocol, error);
2815 last_item = MLX5_FLOW_LAYER_GRE;
2817 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
2818 ret = mlx5_flow_validate_item_gre_key
2819 (items, item_flags, gre_item, error);
2822 item_flags |= MLX5_FLOW_LAYER_GRE_KEY;
2824 case RTE_FLOW_ITEM_TYPE_VXLAN:
2825 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
2829 last_item = MLX5_FLOW_LAYER_VXLAN;
2831 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2832 ret = mlx5_flow_validate_item_vxlan_gpe(items,
2837 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
2839 case RTE_FLOW_ITEM_TYPE_MPLS:
2840 ret = mlx5_flow_validate_item_mpls(dev, items,
2845 last_item = MLX5_FLOW_LAYER_MPLS;
2847 case RTE_FLOW_ITEM_TYPE_META:
2848 ret = flow_dv_validate_item_meta(dev, items, attr,
2852 last_item = MLX5_FLOW_ITEM_METADATA;
2854 case RTE_FLOW_ITEM_TYPE_ICMP:
2855 ret = mlx5_flow_validate_item_icmp(items, item_flags,
2860 item_flags |= MLX5_FLOW_LAYER_ICMP;
2862 case RTE_FLOW_ITEM_TYPE_ICMP6:
2863 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
2868 item_flags |= MLX5_FLOW_LAYER_ICMP6;
2871 return rte_flow_error_set(error, ENOTSUP,
2872 RTE_FLOW_ERROR_TYPE_ITEM,
2873 NULL, "item not supported");
2875 item_flags |= last_item;
2877 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2878 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
2879 return rte_flow_error_set(error, ENOTSUP,
2880 RTE_FLOW_ERROR_TYPE_ACTION,
2881 actions, "too many actions");
2882 switch (actions->type) {
2883 case RTE_FLOW_ACTION_TYPE_VOID:
2885 case RTE_FLOW_ACTION_TYPE_PORT_ID:
2886 ret = flow_dv_validate_action_port_id(dev,
2893 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
2896 case RTE_FLOW_ACTION_TYPE_FLAG:
2897 ret = mlx5_flow_validate_action_flag(action_flags,
2901 action_flags |= MLX5_FLOW_ACTION_FLAG;
2904 case RTE_FLOW_ACTION_TYPE_MARK:
2905 ret = mlx5_flow_validate_action_mark(actions,
2910 action_flags |= MLX5_FLOW_ACTION_MARK;
2913 case RTE_FLOW_ACTION_TYPE_DROP:
2914 ret = mlx5_flow_validate_action_drop(action_flags,
2918 action_flags |= MLX5_FLOW_ACTION_DROP;
2921 case RTE_FLOW_ACTION_TYPE_QUEUE:
2922 ret = mlx5_flow_validate_action_queue(actions,
2927 action_flags |= MLX5_FLOW_ACTION_QUEUE;
2930 case RTE_FLOW_ACTION_TYPE_RSS:
2931 ret = mlx5_flow_validate_action_rss(actions,
2937 action_flags |= MLX5_FLOW_ACTION_RSS;
2940 case RTE_FLOW_ACTION_TYPE_COUNT:
2941 ret = flow_dv_validate_action_count(dev, error);
2944 action_flags |= MLX5_FLOW_ACTION_COUNT;
2947 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2948 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2949 ret = flow_dv_validate_action_l2_encap(action_flags,
2954 action_flags |= actions->type ==
2955 RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP ?
2956 MLX5_FLOW_ACTION_VXLAN_ENCAP :
2957 MLX5_FLOW_ACTION_NVGRE_ENCAP;
2960 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
2961 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
2962 ret = flow_dv_validate_action_l2_decap(action_flags,
2966 action_flags |= actions->type ==
2967 RTE_FLOW_ACTION_TYPE_VXLAN_DECAP ?
2968 MLX5_FLOW_ACTION_VXLAN_DECAP :
2969 MLX5_FLOW_ACTION_NVGRE_DECAP;
2972 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2973 ret = flow_dv_validate_action_raw_encap(action_flags,
2978 action_flags |= MLX5_FLOW_ACTION_RAW_ENCAP;
2981 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2982 ret = flow_dv_validate_action_raw_decap(action_flags,
2987 action_flags |= MLX5_FLOW_ACTION_RAW_DECAP;
2990 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
2991 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
2992 ret = flow_dv_validate_action_modify_mac(action_flags,
2998 /* Count all modify-header actions as one action. */
2999 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3001 action_flags |= actions->type ==
3002 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
3003 MLX5_FLOW_ACTION_SET_MAC_SRC :
3004 MLX5_FLOW_ACTION_SET_MAC_DST;
3007 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3008 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3009 ret = flow_dv_validate_action_modify_ipv4(action_flags,
3015 /* Count all modify-header actions as one action. */
3016 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3018 action_flags |= actions->type ==
3019 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
3020 MLX5_FLOW_ACTION_SET_IPV4_SRC :
3021 MLX5_FLOW_ACTION_SET_IPV4_DST;
3023 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3024 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3025 ret = flow_dv_validate_action_modify_ipv6(action_flags,
3031 /* Count all modify-header actions as one action. */
3032 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3034 action_flags |= actions->type ==
3035 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
3036 MLX5_FLOW_ACTION_SET_IPV6_SRC :
3037 MLX5_FLOW_ACTION_SET_IPV6_DST;
3039 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3040 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3041 ret = flow_dv_validate_action_modify_tp(action_flags,
3047 /* Count all modify-header actions as one action. */
3048 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3050 action_flags |= actions->type ==
3051 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
3052 MLX5_FLOW_ACTION_SET_TP_SRC :
3053 MLX5_FLOW_ACTION_SET_TP_DST;
3055 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3056 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3057 ret = flow_dv_validate_action_modify_ttl(action_flags,
3063 /* Count all modify-header actions as one action. */
3064 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3066 action_flags |= actions->type ==
3067 RTE_FLOW_ACTION_TYPE_SET_TTL ?
3068 MLX5_FLOW_ACTION_SET_TTL :
3069 MLX5_FLOW_ACTION_DEC_TTL;
3071 case RTE_FLOW_ACTION_TYPE_JUMP:
3072 ret = flow_dv_validate_action_jump(actions,
3073 attr->group, error);
3077 action_flags |= MLX5_FLOW_ACTION_JUMP;
3079 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3080 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3081 ret = flow_dv_validate_action_modify_tcp_seq
3088 /* Count all modify-header actions as one action. */
3089 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3091 action_flags |= actions->type ==
3092 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
3093 MLX5_FLOW_ACTION_INC_TCP_SEQ :
3094 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
3096 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3097 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3098 ret = flow_dv_validate_action_modify_tcp_ack
3105 /* Count all modify-header actions as one action. */
3106 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
3108 action_flags |= actions->type ==
3109 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
3110 MLX5_FLOW_ACTION_INC_TCP_ACK :
3111 MLX5_FLOW_ACTION_DEC_TCP_ACK;
3114 return rte_flow_error_set(error, ENOTSUP,
3115 RTE_FLOW_ERROR_TYPE_ACTION,
3117 "action not supported");
3120 /* Eswitch has few restrictions on using items and actions */
3121 if (attr->transfer) {
3122 if (action_flags & MLX5_FLOW_ACTION_FLAG)
3123 return rte_flow_error_set(error, ENOTSUP,
3124 RTE_FLOW_ERROR_TYPE_ACTION,
3126 "unsupported action FLAG");
3127 if (action_flags & MLX5_FLOW_ACTION_MARK)
3128 return rte_flow_error_set(error, ENOTSUP,
3129 RTE_FLOW_ERROR_TYPE_ACTION,
3131 "unsupported action MARK");
3132 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
3133 return rte_flow_error_set(error, ENOTSUP,
3134 RTE_FLOW_ERROR_TYPE_ACTION,
3136 "unsupported action QUEUE");
3137 if (action_flags & MLX5_FLOW_ACTION_RSS)
3138 return rte_flow_error_set(error, ENOTSUP,
3139 RTE_FLOW_ERROR_TYPE_ACTION,
3141 "unsupported action RSS");
3142 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3143 return rte_flow_error_set(error, EINVAL,
3144 RTE_FLOW_ERROR_TYPE_ACTION,
3146 "no fate action is found");
3148 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
3149 return rte_flow_error_set(error, EINVAL,
3150 RTE_FLOW_ERROR_TYPE_ACTION,
3152 "no fate action is found");
3158 * Internal preparation function. Allocates the DV flow size,
3159 * this size is constant.
3162 * Pointer to the flow attributes.
3164 * Pointer to the list of items.
3165 * @param[in] actions
3166 * Pointer to the list of actions.
3168 * Pointer to the error structure.
3171 * Pointer to mlx5_flow object on success,
3172 * otherwise NULL and rte_errno is set.
3174 static struct mlx5_flow *
3175 flow_dv_prepare(const struct rte_flow_attr *attr __rte_unused,
3176 const struct rte_flow_item items[] __rte_unused,
3177 const struct rte_flow_action actions[] __rte_unused,
3178 struct rte_flow_error *error)
3180 uint32_t size = sizeof(struct mlx5_flow);
3181 struct mlx5_flow *flow;
3183 flow = rte_calloc(__func__, 1, size, 0);
3185 rte_flow_error_set(error, ENOMEM,
3186 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3187 "not enough memory to create flow");
3190 flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
3196 * Sanity check for match mask and value. Similar to check_valid_spec() in
3197 * kernel driver. If unmasked bit is present in value, it returns failure.
3200 * pointer to match mask buffer.
3201 * @param match_value
3202 * pointer to match value buffer.
3205 * 0 if valid, -EINVAL otherwise.
3208 flow_dv_check_valid_spec(void *match_mask, void *match_value)
3210 uint8_t *m = match_mask;
3211 uint8_t *v = match_value;
3214 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
3217 "match_value differs from match_criteria"
3218 " %p[%u] != %p[%u]",
3219 match_value, i, match_mask, i);
3228 * Add Ethernet item to matcher and to the value.
3230 * @param[in, out] matcher
3232 * @param[in, out] key
3233 * Flow matcher value.
3235 * Flow pattern to translate.
3237 * Item is inner pattern.
3240 flow_dv_translate_item_eth(void *matcher, void *key,
3241 const struct rte_flow_item *item, int inner)
3243 const struct rte_flow_item_eth *eth_m = item->mask;
3244 const struct rte_flow_item_eth *eth_v = item->spec;
3245 const struct rte_flow_item_eth nic_mask = {
3246 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
3247 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
3248 .type = RTE_BE16(0xffff),
3260 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3262 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3264 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3266 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3268 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
3269 ð_m->dst, sizeof(eth_m->dst));
3270 /* The value must be in the range of the mask. */
3271 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
3272 for (i = 0; i < sizeof(eth_m->dst); ++i)
3273 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
3274 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
3275 ð_m->src, sizeof(eth_m->src));
3276 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
3277 /* The value must be in the range of the mask. */
3278 for (i = 0; i < sizeof(eth_m->dst); ++i)
3279 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
3280 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
3281 rte_be_to_cpu_16(eth_m->type));
3282 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, ethertype);
3283 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
3287 * Add VLAN item to matcher and to the value.
3289 * @param[in, out] matcher
3291 * @param[in, out] key
3292 * Flow matcher value.
3294 * Flow pattern to translate.
3296 * Item is inner pattern.
3299 flow_dv_translate_item_vlan(void *matcher, void *key,
3300 const struct rte_flow_item *item,
3303 const struct rte_flow_item_vlan *vlan_m = item->mask;
3304 const struct rte_flow_item_vlan *vlan_v = item->spec;
3305 const struct rte_flow_item_vlan nic_mask = {
3306 .tci = RTE_BE16(0x0fff),
3307 .inner_type = RTE_BE16(0xffff),
3319 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3321 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3323 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3325 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3327 tci_m = rte_be_to_cpu_16(vlan_m->tci);
3328 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
3329 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
3330 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
3331 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
3332 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
3333 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
3334 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
3335 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
3336 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
3340 * Add IPV4 item to matcher and to the value.
3342 * @param[in, out] matcher
3344 * @param[in, out] key
3345 * Flow matcher value.
3347 * Flow pattern to translate.
3349 * Item is inner pattern.
3351 * The group to insert the rule.
3354 flow_dv_translate_item_ipv4(void *matcher, void *key,
3355 const struct rte_flow_item *item,
3356 int inner, uint32_t group)
3358 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
3359 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
3360 const struct rte_flow_item_ipv4 nic_mask = {
3362 .src_addr = RTE_BE32(0xffffffff),
3363 .dst_addr = RTE_BE32(0xffffffff),
3364 .type_of_service = 0xff,
3365 .next_proto_id = 0xff,
3375 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3377 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3379 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3381 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3384 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
3386 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x4);
3387 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 4);
3392 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
3393 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
3394 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
3395 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
3396 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
3397 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
3398 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
3399 src_ipv4_src_ipv6.ipv4_layout.ipv4);
3400 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
3401 src_ipv4_src_ipv6.ipv4_layout.ipv4);
3402 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
3403 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
3404 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
3405 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
3406 ipv4_m->hdr.type_of_service);
3407 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
3408 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
3409 ipv4_m->hdr.type_of_service >> 2);
3410 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
3411 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
3412 ipv4_m->hdr.next_proto_id);
3413 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
3414 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
3418 * Add IPV6 item to matcher and to the value.
3420 * @param[in, out] matcher
3422 * @param[in, out] key
3423 * Flow matcher value.
3425 * Flow pattern to translate.
3427 * Item is inner pattern.
3429 * The group to insert the rule.
3432 flow_dv_translate_item_ipv6(void *matcher, void *key,
3433 const struct rte_flow_item *item,
3434 int inner, uint32_t group)
3436 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
3437 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
3438 const struct rte_flow_item_ipv6 nic_mask = {
3441 "\xff\xff\xff\xff\xff\xff\xff\xff"
3442 "\xff\xff\xff\xff\xff\xff\xff\xff",
3444 "\xff\xff\xff\xff\xff\xff\xff\xff"
3445 "\xff\xff\xff\xff\xff\xff\xff\xff",
3446 .vtc_flow = RTE_BE32(0xffffffff),
3453 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3454 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3463 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3465 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3467 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3469 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3472 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
3474 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x6);
3475 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 6);
3480 size = sizeof(ipv6_m->hdr.dst_addr);
3481 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
3482 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
3483 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
3484 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
3485 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
3486 for (i = 0; i < size; ++i)
3487 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
3488 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
3489 src_ipv4_src_ipv6.ipv6_layout.ipv6);
3490 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
3491 src_ipv4_src_ipv6.ipv6_layout.ipv6);
3492 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
3493 for (i = 0; i < size; ++i)
3494 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
3496 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
3497 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
3498 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
3499 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
3500 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
3501 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
3504 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
3506 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
3509 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
3511 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
3515 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
3517 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
3518 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
3522 * Add TCP item to matcher and to the value.
3524 * @param[in, out] matcher
3526 * @param[in, out] key
3527 * Flow matcher value.
3529 * Flow pattern to translate.
3531 * Item is inner pattern.
3534 flow_dv_translate_item_tcp(void *matcher, void *key,
3535 const struct rte_flow_item *item,
3538 const struct rte_flow_item_tcp *tcp_m = item->mask;
3539 const struct rte_flow_item_tcp *tcp_v = item->spec;
3544 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3546 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3548 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3550 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3552 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
3553 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
3557 tcp_m = &rte_flow_item_tcp_mask;
3558 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
3559 rte_be_to_cpu_16(tcp_m->hdr.src_port));
3560 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
3561 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
3562 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
3563 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
3564 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
3565 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
3566 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
3567 tcp_m->hdr.tcp_flags);
3568 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
3569 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
3573 * Add UDP item to matcher and to the value.
3575 * @param[in, out] matcher
3577 * @param[in, out] key
3578 * Flow matcher value.
3580 * Flow pattern to translate.
3582 * Item is inner pattern.
3585 flow_dv_translate_item_udp(void *matcher, void *key,
3586 const struct rte_flow_item *item,
3589 const struct rte_flow_item_udp *udp_m = item->mask;
3590 const struct rte_flow_item_udp *udp_v = item->spec;
3595 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3597 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3599 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3601 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3603 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
3604 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
3608 udp_m = &rte_flow_item_udp_mask;
3609 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
3610 rte_be_to_cpu_16(udp_m->hdr.src_port));
3611 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
3612 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
3613 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
3614 rte_be_to_cpu_16(udp_m->hdr.dst_port));
3615 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
3616 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
3620 * Add GRE optional Key item to matcher and to the value.
3622 * @param[in, out] matcher
3624 * @param[in, out] key
3625 * Flow matcher value.
3627 * Flow pattern to translate.
3629 * Item is inner pattern.
3632 flow_dv_translate_item_gre_key(void *matcher, void *key,
3633 const struct rte_flow_item *item)
3635 const rte_be32_t *key_m = item->mask;
3636 const rte_be32_t *key_v = item->spec;
3637 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3638 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3639 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
3644 key_m = &gre_key_default_mask;
3645 /* GRE K bit must be on and should already be validated */
3646 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
3647 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
3648 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
3649 rte_be_to_cpu_32(*key_m) >> 8);
3650 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
3651 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
3652 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
3653 rte_be_to_cpu_32(*key_m) & 0xFF);
3654 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
3655 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
3659 * Add GRE item to matcher and to the value.
3661 * @param[in, out] matcher
3663 * @param[in, out] key
3664 * Flow matcher value.
3666 * Flow pattern to translate.
3668 * Item is inner pattern.
3671 flow_dv_translate_item_gre(void *matcher, void *key,
3672 const struct rte_flow_item *item,
3675 const struct rte_flow_item_gre *gre_m = item->mask;
3676 const struct rte_flow_item_gre *gre_v = item->spec;
3679 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3680 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3687 uint16_t s_present:1;
3688 uint16_t k_present:1;
3689 uint16_t rsvd_bit1:1;
3690 uint16_t c_present:1;
3694 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
3697 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3699 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3701 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3703 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3705 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
3706 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
3710 gre_m = &rte_flow_item_gre_mask;
3711 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
3712 rte_be_to_cpu_16(gre_m->protocol));
3713 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
3714 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
3715 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
3716 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
3717 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
3718 gre_crks_rsvd0_ver_m.c_present);
3719 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
3720 gre_crks_rsvd0_ver_v.c_present &
3721 gre_crks_rsvd0_ver_m.c_present);
3722 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
3723 gre_crks_rsvd0_ver_m.k_present);
3724 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
3725 gre_crks_rsvd0_ver_v.k_present &
3726 gre_crks_rsvd0_ver_m.k_present);
3727 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
3728 gre_crks_rsvd0_ver_m.s_present);
3729 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
3730 gre_crks_rsvd0_ver_v.s_present &
3731 gre_crks_rsvd0_ver_m.s_present);
3735 * Add NVGRE item to matcher and to the value.
3737 * @param[in, out] matcher
3739 * @param[in, out] key
3740 * Flow matcher value.
3742 * Flow pattern to translate.
3744 * Item is inner pattern.
3747 flow_dv_translate_item_nvgre(void *matcher, void *key,
3748 const struct rte_flow_item *item,
3751 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
3752 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
3753 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3754 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3755 const char *tni_flow_id_m = (const char *)nvgre_m->tni;
3756 const char *tni_flow_id_v = (const char *)nvgre_v->tni;
3762 flow_dv_translate_item_gre(matcher, key, item, inner);
3766 nvgre_m = &rte_flow_item_nvgre_mask;
3767 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
3768 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
3769 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
3770 memcpy(gre_key_m, tni_flow_id_m, size);
3771 for (i = 0; i < size; ++i)
3772 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
3776 * Add VXLAN item to matcher and to the value.
3778 * @param[in, out] matcher
3780 * @param[in, out] key
3781 * Flow matcher value.
3783 * Flow pattern to translate.
3785 * Item is inner pattern.
3788 flow_dv_translate_item_vxlan(void *matcher, void *key,
3789 const struct rte_flow_item *item,
3792 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
3793 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
3796 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3797 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3805 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3807 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
3809 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
3811 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3813 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
3814 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
3815 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
3816 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
3817 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
3822 vxlan_m = &rte_flow_item_vxlan_mask;
3823 size = sizeof(vxlan_m->vni);
3824 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
3825 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
3826 memcpy(vni_m, vxlan_m->vni, size);
3827 for (i = 0; i < size; ++i)
3828 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
3832 * Add MPLS item to matcher and to the value.
3834 * @param[in, out] matcher
3836 * @param[in, out] key
3837 * Flow matcher value.
3839 * Flow pattern to translate.
3840 * @param[in] prev_layer
3841 * The protocol layer indicated in previous item.
3843 * Item is inner pattern.
3846 flow_dv_translate_item_mpls(void *matcher, void *key,
3847 const struct rte_flow_item *item,
3848 uint64_t prev_layer,
3851 const uint32_t *in_mpls_m = item->mask;
3852 const uint32_t *in_mpls_v = item->spec;
3853 uint32_t *out_mpls_m = 0;
3854 uint32_t *out_mpls_v = 0;
3855 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3856 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3857 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
3859 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
3860 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
3861 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
3863 switch (prev_layer) {
3864 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
3865 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
3866 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
3867 MLX5_UDP_PORT_MPLS);
3869 case MLX5_FLOW_LAYER_GRE:
3870 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
3871 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
3872 RTE_ETHER_TYPE_MPLS);
3875 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
3876 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
3883 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
3884 switch (prev_layer) {
3885 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
3887 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
3888 outer_first_mpls_over_udp);
3890 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
3891 outer_first_mpls_over_udp);
3893 case MLX5_FLOW_LAYER_GRE:
3895 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
3896 outer_first_mpls_over_gre);
3898 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
3899 outer_first_mpls_over_gre);
3902 /* Inner MPLS not over GRE is not supported. */
3905 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
3909 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
3915 if (out_mpls_m && out_mpls_v) {
3916 *out_mpls_m = *in_mpls_m;
3917 *out_mpls_v = *in_mpls_v & *in_mpls_m;
3922 * Add META item to matcher
3924 * @param[in, out] matcher
3926 * @param[in, out] key
3927 * Flow matcher value.
3929 * Flow pattern to translate.
3931 * Item is inner pattern.
3934 flow_dv_translate_item_meta(void *matcher, void *key,
3935 const struct rte_flow_item *item)
3937 const struct rte_flow_item_meta *meta_m;
3938 const struct rte_flow_item_meta *meta_v;
3940 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
3942 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
3944 meta_m = (const void *)item->mask;
3946 meta_m = &rte_flow_item_meta_mask;
3947 meta_v = (const void *)item->spec;
3949 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a,
3950 rte_be_to_cpu_32(meta_m->data));
3951 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a,
3952 rte_be_to_cpu_32(meta_v->data & meta_m->data));
3957 * Add source vport match to the specified matcher.
3959 * @param[in, out] matcher
3961 * @param[in, out] key
3962 * Flow matcher value.
3964 * Source vport value to match
3969 flow_dv_translate_item_source_vport(void *matcher, void *key,
3970 int16_t port, uint16_t mask)
3972 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
3973 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
3975 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
3976 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
3980 * Translate port-id item to eswitch match on port-id.
3983 * The devich to configure through.
3984 * @param[in, out] matcher
3986 * @param[in, out] key
3987 * Flow matcher value.
3989 * Flow pattern to translate.
3992 * 0 on success, a negative errno value otherwise.
3995 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
3996 void *key, const struct rte_flow_item *item)
3998 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
3999 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
4000 uint16_t mask, val, id;
4003 mask = pid_m ? pid_m->id : 0xffff;
4004 id = pid_v ? pid_v->id : dev->data->port_id;
4005 ret = mlx5_port_to_eswitch_info(id, NULL, &val);
4008 flow_dv_translate_item_source_vport(matcher, key, val, mask);
4013 * Add ICMP6 item to matcher and to the value.
4015 * @param[in, out] matcher
4017 * @param[in, out] key
4018 * Flow matcher value.
4020 * Flow pattern to translate.
4022 * Item is inner pattern.
4025 flow_dv_translate_item_icmp6(void *matcher, void *key,
4026 const struct rte_flow_item *item,
4029 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
4030 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
4033 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
4035 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
4037 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
4039 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
4041 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
4043 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
4045 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
4046 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
4050 icmp6_m = &rte_flow_item_icmp6_mask;
4051 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
4052 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
4053 icmp6_v->type & icmp6_m->type);
4054 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
4055 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
4056 icmp6_v->code & icmp6_m->code);
4060 * Add ICMP item to matcher and to the value.
4062 * @param[in, out] matcher
4064 * @param[in, out] key
4065 * Flow matcher value.
4067 * Flow pattern to translate.
4069 * Item is inner pattern.
4072 flow_dv_translate_item_icmp(void *matcher, void *key,
4073 const struct rte_flow_item *item,
4076 const struct rte_flow_item_icmp *icmp_m = item->mask;
4077 const struct rte_flow_item_icmp *icmp_v = item->spec;
4080 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
4082 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
4084 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
4086 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
4088 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
4090 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
4092 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
4093 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
4097 icmp_m = &rte_flow_item_icmp_mask;
4098 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
4099 icmp_m->hdr.icmp_type);
4100 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
4101 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
4102 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
4103 icmp_m->hdr.icmp_code);
4104 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
4105 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
4108 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
4110 #define HEADER_IS_ZERO(match_criteria, headers) \
4111 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
4112 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
4115 * Calculate flow matcher enable bitmap.
4117 * @param match_criteria
4118 * Pointer to flow matcher criteria.
4121 * Bitmap of enabled fields.
4124 flow_dv_matcher_enable(uint32_t *match_criteria)
4126 uint8_t match_criteria_enable;
4128 match_criteria_enable =
4129 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
4130 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
4131 match_criteria_enable |=
4132 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
4133 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
4134 match_criteria_enable |=
4135 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
4136 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
4137 match_criteria_enable |=
4138 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
4139 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
4140 #ifdef HAVE_MLX5DV_DR
4141 match_criteria_enable |=
4142 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
4143 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
4145 return match_criteria_enable;
4152 * @param dev[in, out]
4153 * Pointer to rte_eth_dev structure.
4154 * @param[in] table_id
4157 * Direction of the table.
4158 * @param[in] transfer
4159 * E-Switch or NIC flow.
4161 * pointer to error structure.
4164 * Returns tables resource based on the index, NULL in case of failed.
4166 static struct mlx5_flow_tbl_resource *
4167 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
4168 uint32_t table_id, uint8_t egress,
4170 struct rte_flow_error *error)
4172 struct mlx5_priv *priv = dev->data->dev_private;
4173 struct mlx5_ibv_shared *sh = priv->sh;
4174 struct mlx5_flow_tbl_resource *tbl;
4176 #ifdef HAVE_MLX5DV_DR
4178 tbl = &sh->fdb_tbl[table_id];
4180 tbl->obj = mlx5_glue->dr_create_flow_tbl
4181 (sh->fdb_domain, table_id);
4182 } else if (egress) {
4183 tbl = &sh->tx_tbl[table_id];
4185 tbl->obj = mlx5_glue->dr_create_flow_tbl
4186 (sh->tx_domain, table_id);
4188 tbl = &sh->rx_tbl[table_id];
4190 tbl->obj = mlx5_glue->dr_create_flow_tbl
4191 (sh->rx_domain, table_id);
4194 rte_flow_error_set(error, ENOMEM,
4195 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4196 NULL, "cannot create table");
4199 rte_atomic32_inc(&tbl->refcnt);
4205 return &sh->fdb_tbl[table_id];
4207 return &sh->tx_tbl[table_id];
4209 return &sh->rx_tbl[table_id];
4214 * Release a flow table.
4217 * Table resource to be released.
4220 * Returns 0 if table was released, else return 1;
4223 flow_dv_tbl_resource_release(struct mlx5_flow_tbl_resource *tbl)
4227 if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
4228 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
4236 * Register the flow matcher.
4238 * @param dev[in, out]
4239 * Pointer to rte_eth_dev structure.
4240 * @param[in, out] matcher
4241 * Pointer to flow matcher.
4242 * @parm[in, out] dev_flow
4243 * Pointer to the dev_flow.
4245 * pointer to error structure.
4248 * 0 on success otherwise -errno and errno is set.
4251 flow_dv_matcher_register(struct rte_eth_dev *dev,
4252 struct mlx5_flow_dv_matcher *matcher,
4253 struct mlx5_flow *dev_flow,
4254 struct rte_flow_error *error)
4256 struct mlx5_priv *priv = dev->data->dev_private;
4257 struct mlx5_ibv_shared *sh = priv->sh;
4258 struct mlx5_flow_dv_matcher *cache_matcher;
4259 struct mlx5dv_flow_matcher_attr dv_attr = {
4260 .type = IBV_FLOW_ATTR_NORMAL,
4261 .match_mask = (void *)&matcher->mask,
4263 struct mlx5_flow_tbl_resource *tbl = NULL;
4265 /* Lookup from cache. */
4266 LIST_FOREACH(cache_matcher, &sh->matchers, next) {
4267 if (matcher->crc == cache_matcher->crc &&
4268 matcher->priority == cache_matcher->priority &&
4269 matcher->egress == cache_matcher->egress &&
4270 matcher->group == cache_matcher->group &&
4271 matcher->transfer == cache_matcher->transfer &&
4272 !memcmp((const void *)matcher->mask.buf,
4273 (const void *)cache_matcher->mask.buf,
4274 cache_matcher->mask.size)) {
4276 "priority %hd use %s matcher %p: refcnt %d++",
4277 cache_matcher->priority,
4278 cache_matcher->egress ? "tx" : "rx",
4279 (void *)cache_matcher,
4280 rte_atomic32_read(&cache_matcher->refcnt));
4281 rte_atomic32_inc(&cache_matcher->refcnt);
4282 dev_flow->dv.matcher = cache_matcher;
4286 /* Register new matcher. */
4287 cache_matcher = rte_calloc(__func__, 1, sizeof(*cache_matcher), 0);
4289 return rte_flow_error_set(error, ENOMEM,
4290 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4291 "cannot allocate matcher memory");
4292 tbl = flow_dv_tbl_resource_get(dev, matcher->group * MLX5_GROUP_FACTOR,
4293 matcher->egress, matcher->transfer,
4296 rte_free(cache_matcher);
4297 return rte_flow_error_set(error, ENOMEM,
4298 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4299 NULL, "cannot create table");
4301 *cache_matcher = *matcher;
4302 dv_attr.match_criteria_enable =
4303 flow_dv_matcher_enable(cache_matcher->mask.buf);
4304 dv_attr.priority = matcher->priority;
4305 if (matcher->egress)
4306 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
4307 cache_matcher->matcher_object =
4308 mlx5_glue->dv_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj);
4309 if (!cache_matcher->matcher_object) {
4310 rte_free(cache_matcher);
4311 #ifdef HAVE_MLX5DV_DR
4312 flow_dv_tbl_resource_release(tbl);
4314 return rte_flow_error_set(error, ENOMEM,
4315 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4316 NULL, "cannot create matcher");
4318 rte_atomic32_inc(&cache_matcher->refcnt);
4319 LIST_INSERT_HEAD(&sh->matchers, cache_matcher, next);
4320 dev_flow->dv.matcher = cache_matcher;
4321 DRV_LOG(DEBUG, "priority %hd new %s matcher %p: refcnt %d",
4322 cache_matcher->priority,
4323 cache_matcher->egress ? "tx" : "rx", (void *)cache_matcher,
4324 rte_atomic32_read(&cache_matcher->refcnt));
4325 rte_atomic32_inc(&tbl->refcnt);
4330 * Find existing tag resource or create and register a new one.
4332 * @param dev[in, out]
4333 * Pointer to rte_eth_dev structure.
4334 * @param[in, out] resource
4335 * Pointer to tag resource.
4336 * @parm[in, out] dev_flow
4337 * Pointer to the dev_flow.
4339 * pointer to error structure.
4342 * 0 on success otherwise -errno and errno is set.
4345 flow_dv_tag_resource_register
4346 (struct rte_eth_dev *dev,
4347 struct mlx5_flow_dv_tag_resource *resource,
4348 struct mlx5_flow *dev_flow,
4349 struct rte_flow_error *error)
4351 struct mlx5_priv *priv = dev->data->dev_private;
4352 struct mlx5_ibv_shared *sh = priv->sh;
4353 struct mlx5_flow_dv_tag_resource *cache_resource;
4355 /* Lookup a matching resource from cache. */
4356 LIST_FOREACH(cache_resource, &sh->tags, next) {
4357 if (resource->tag == cache_resource->tag) {
4358 DRV_LOG(DEBUG, "tag resource %p: refcnt %d++",
4359 (void *)cache_resource,
4360 rte_atomic32_read(&cache_resource->refcnt));
4361 rte_atomic32_inc(&cache_resource->refcnt);
4362 dev_flow->flow->tag_resource = cache_resource;
4366 /* Register new resource. */
4367 cache_resource = rte_calloc(__func__, 1, sizeof(*cache_resource), 0);
4368 if (!cache_resource)
4369 return rte_flow_error_set(error, ENOMEM,
4370 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4371 "cannot allocate resource memory");
4372 *cache_resource = *resource;
4373 cache_resource->action = mlx5_glue->dv_create_flow_action_tag
4375 if (!cache_resource->action) {
4376 rte_free(cache_resource);
4377 return rte_flow_error_set(error, ENOMEM,
4378 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4379 NULL, "cannot create action");
4381 rte_atomic32_init(&cache_resource->refcnt);
4382 rte_atomic32_inc(&cache_resource->refcnt);
4383 LIST_INSERT_HEAD(&sh->tags, cache_resource, next);
4384 dev_flow->flow->tag_resource = cache_resource;
4385 DRV_LOG(DEBUG, "new tag resource %p: refcnt %d++",
4386 (void *)cache_resource,
4387 rte_atomic32_read(&cache_resource->refcnt));
4395 * Pointer to Ethernet device.
4397 * Pointer to mlx5_flow.
4400 * 1 while a reference on it exists, 0 when freed.
4403 flow_dv_tag_release(struct rte_eth_dev *dev,
4404 struct mlx5_flow_dv_tag_resource *tag)
4407 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
4408 dev->data->port_id, (void *)tag,
4409 rte_atomic32_read(&tag->refcnt));
4410 if (rte_atomic32_dec_and_test(&tag->refcnt)) {
4411 claim_zero(mlx5_glue->destroy_flow_action(tag->action));
4412 LIST_REMOVE(tag, next);
4413 DRV_LOG(DEBUG, "port %u tag %p: removed",
4414 dev->data->port_id, (void *)tag);
4422 * Translate port ID action to vport.
4425 * Pointer to rte_eth_dev structure.
4427 * Pointer to the port ID action.
4428 * @param[out] dst_port_id
4429 * The target port ID.
4431 * Pointer to the error structure.
4434 * 0 on success, a negative errno value otherwise and rte_errno is set.
4437 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
4438 const struct rte_flow_action *action,
4439 uint32_t *dst_port_id,
4440 struct rte_flow_error *error)
4445 const struct rte_flow_action_port_id *conf =
4446 (const struct rte_flow_action_port_id *)action->conf;
4448 port = conf->original ? dev->data->port_id : conf->id;
4449 ret = mlx5_port_to_eswitch_info(port, NULL, &port_id);
4451 return rte_flow_error_set(error, -ret,
4452 RTE_FLOW_ERROR_TYPE_ACTION,
4454 "No eswitch info was found for port");
4455 *dst_port_id = port_id;
4460 * Fill the flow with DV spec.
4463 * Pointer to rte_eth_dev structure.
4464 * @param[in, out] dev_flow
4465 * Pointer to the sub flow.
4467 * Pointer to the flow attributes.
4469 * Pointer to the list of items.
4470 * @param[in] actions
4471 * Pointer to the list of actions.
4473 * Pointer to the error structure.
4476 * 0 on success, a negative errno value otherwise and rte_errno is set.
4479 flow_dv_translate(struct rte_eth_dev *dev,
4480 struct mlx5_flow *dev_flow,
4481 const struct rte_flow_attr *attr,
4482 const struct rte_flow_item items[],
4483 const struct rte_flow_action actions[],
4484 struct rte_flow_error *error)
4486 struct mlx5_priv *priv = dev->data->dev_private;
4487 struct rte_flow *flow = dev_flow->flow;
4488 uint64_t item_flags = 0;
4489 uint64_t last_item = 0;
4490 uint64_t action_flags = 0;
4491 uint64_t priority = attr->priority;
4492 struct mlx5_flow_dv_matcher matcher = {
4494 .size = sizeof(matcher.mask.buf),
4498 bool actions_end = false;
4499 struct mlx5_flow_dv_modify_hdr_resource res = {
4500 .ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
4501 MLX5DV_FLOW_TABLE_TYPE_NIC_RX
4503 union flow_dv_attr flow_attr = { .attr = 0 };
4504 struct mlx5_flow_dv_tag_resource tag_resource;
4505 uint32_t modify_action_position = UINT32_MAX;
4506 void *match_mask = matcher.mask.buf;
4507 void *match_value = dev_flow->dv.value.buf;
4509 flow->group = attr->group;
4511 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
4512 if (priority == MLX5_FLOW_PRIO_RSVD)
4513 priority = priv->config.flow_prio - 1;
4514 for (; !actions_end ; actions++) {
4515 const struct rte_flow_action_queue *queue;
4516 const struct rte_flow_action_rss *rss;
4517 const struct rte_flow_action *action = actions;
4518 const struct rte_flow_action_count *count = action->conf;
4519 const uint8_t *rss_key;
4520 const struct rte_flow_action_jump *jump_data;
4521 struct mlx5_flow_dv_jump_tbl_resource jump_tbl_resource;
4522 struct mlx5_flow_tbl_resource *tbl;
4523 uint32_t port_id = 0;
4524 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
4526 switch (actions->type) {
4527 case RTE_FLOW_ACTION_TYPE_VOID:
4529 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4530 if (flow_dv_translate_action_port_id(dev, action,
4533 port_id_resource.port_id = port_id;
4534 if (flow_dv_port_id_action_resource_register
4535 (dev, &port_id_resource, dev_flow, error))
4537 dev_flow->dv.actions[actions_n++] =
4538 dev_flow->dv.port_id_action->action;
4539 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4541 case RTE_FLOW_ACTION_TYPE_FLAG:
4543 mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
4544 if (!flow->tag_resource)
4545 if (flow_dv_tag_resource_register
4546 (dev, &tag_resource, dev_flow, error))
4548 dev_flow->dv.actions[actions_n++] =
4549 flow->tag_resource->action;
4550 action_flags |= MLX5_FLOW_ACTION_FLAG;
4552 case RTE_FLOW_ACTION_TYPE_MARK:
4553 tag_resource.tag = mlx5_flow_mark_set
4554 (((const struct rte_flow_action_mark *)
4555 (actions->conf))->id);
4556 if (!flow->tag_resource)
4557 if (flow_dv_tag_resource_register
4558 (dev, &tag_resource, dev_flow, error))
4560 dev_flow->dv.actions[actions_n++] =
4561 flow->tag_resource->action;
4562 action_flags |= MLX5_FLOW_ACTION_MARK;
4564 case RTE_FLOW_ACTION_TYPE_DROP:
4565 action_flags |= MLX5_FLOW_ACTION_DROP;
4567 case RTE_FLOW_ACTION_TYPE_QUEUE:
4568 queue = actions->conf;
4569 flow->rss.queue_num = 1;
4570 (*flow->queue)[0] = queue->index;
4571 action_flags |= MLX5_FLOW_ACTION_QUEUE;
4573 case RTE_FLOW_ACTION_TYPE_RSS:
4574 rss = actions->conf;
4576 memcpy((*flow->queue), rss->queue,
4577 rss->queue_num * sizeof(uint16_t));
4578 flow->rss.queue_num = rss->queue_num;
4579 /* NULL RSS key indicates default RSS key. */
4580 rss_key = !rss->key ? rss_hash_default_key : rss->key;
4581 memcpy(flow->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
4582 /* RSS type 0 indicates default RSS type ETH_RSS_IP. */
4583 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4584 flow->rss.level = rss->level;
4585 action_flags |= MLX5_FLOW_ACTION_RSS;
4587 case RTE_FLOW_ACTION_TYPE_COUNT:
4588 if (!priv->config.devx) {
4589 rte_errno = ENOTSUP;
4592 flow->counter = flow_dv_counter_alloc(dev,
4596 if (flow->counter == NULL)
4598 dev_flow->dv.actions[actions_n++] =
4599 flow->counter->action;
4600 action_flags |= MLX5_FLOW_ACTION_COUNT;
4603 if (rte_errno == ENOTSUP)
4604 return rte_flow_error_set
4606 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4608 "count action not supported");
4610 return rte_flow_error_set
4612 RTE_FLOW_ERROR_TYPE_ACTION,
4614 "cannot create counter"
4616 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4617 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4618 if (flow_dv_create_action_l2_encap(dev, actions,
4623 dev_flow->dv.actions[actions_n++] =
4624 dev_flow->dv.encap_decap->verbs_action;
4625 action_flags |= actions->type ==
4626 RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP ?
4627 MLX5_FLOW_ACTION_VXLAN_ENCAP :
4628 MLX5_FLOW_ACTION_NVGRE_ENCAP;
4630 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4631 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4632 if (flow_dv_create_action_l2_decap(dev, dev_flow,
4636 dev_flow->dv.actions[actions_n++] =
4637 dev_flow->dv.encap_decap->verbs_action;
4638 action_flags |= actions->type ==
4639 RTE_FLOW_ACTION_TYPE_VXLAN_DECAP ?
4640 MLX5_FLOW_ACTION_VXLAN_DECAP :
4641 MLX5_FLOW_ACTION_NVGRE_DECAP;
4643 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4644 /* Handle encap with preceding decap. */
4645 if (action_flags & MLX5_FLOW_ACTION_RAW_DECAP) {
4646 if (flow_dv_create_action_raw_encap
4647 (dev, actions, dev_flow, attr, error))
4649 dev_flow->dv.actions[actions_n++] =
4650 dev_flow->dv.encap_decap->verbs_action;
4652 /* Handle encap without preceding decap. */
4653 if (flow_dv_create_action_l2_encap
4654 (dev, actions, dev_flow, attr->transfer,
4657 dev_flow->dv.actions[actions_n++] =
4658 dev_flow->dv.encap_decap->verbs_action;
4660 action_flags |= MLX5_FLOW_ACTION_RAW_ENCAP;
4662 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4663 /* Check if this decap is followed by encap. */
4664 for (; action->type != RTE_FLOW_ACTION_TYPE_END &&
4665 action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP;
4668 /* Handle decap only if it isn't followed by encap. */
4669 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
4670 if (flow_dv_create_action_l2_decap
4671 (dev, dev_flow, attr->transfer, error))
4673 dev_flow->dv.actions[actions_n++] =
4674 dev_flow->dv.encap_decap->verbs_action;
4676 /* If decap is followed by encap, handle it at encap. */
4677 action_flags |= MLX5_FLOW_ACTION_RAW_DECAP;
4679 case RTE_FLOW_ACTION_TYPE_JUMP:
4680 jump_data = action->conf;
4681 tbl = flow_dv_tbl_resource_get(dev, jump_data->group *
4684 attr->transfer, error);
4686 return rte_flow_error_set
4688 RTE_FLOW_ERROR_TYPE_ACTION,
4690 "cannot create jump action.");
4691 jump_tbl_resource.tbl = tbl;
4692 if (flow_dv_jump_tbl_resource_register
4693 (dev, &jump_tbl_resource, dev_flow, error)) {
4694 flow_dv_tbl_resource_release(tbl);
4695 return rte_flow_error_set
4697 RTE_FLOW_ERROR_TYPE_ACTION,
4699 "cannot create jump action.");
4701 dev_flow->dv.actions[actions_n++] =
4702 dev_flow->dv.jump->action;
4703 action_flags |= MLX5_FLOW_ACTION_JUMP;
4705 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4706 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4707 if (flow_dv_convert_action_modify_mac(&res, actions,
4710 action_flags |= actions->type ==
4711 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
4712 MLX5_FLOW_ACTION_SET_MAC_SRC :
4713 MLX5_FLOW_ACTION_SET_MAC_DST;
4715 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4716 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4717 if (flow_dv_convert_action_modify_ipv4(&res, actions,
4720 action_flags |= actions->type ==
4721 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
4722 MLX5_FLOW_ACTION_SET_IPV4_SRC :
4723 MLX5_FLOW_ACTION_SET_IPV4_DST;
4725 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4726 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4727 if (flow_dv_convert_action_modify_ipv6(&res, actions,
4730 action_flags |= actions->type ==
4731 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
4732 MLX5_FLOW_ACTION_SET_IPV6_SRC :
4733 MLX5_FLOW_ACTION_SET_IPV6_DST;
4735 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4736 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4737 if (flow_dv_convert_action_modify_tp(&res, actions,
4741 action_flags |= actions->type ==
4742 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
4743 MLX5_FLOW_ACTION_SET_TP_SRC :
4744 MLX5_FLOW_ACTION_SET_TP_DST;
4746 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4747 if (flow_dv_convert_action_modify_dec_ttl(&res, items,
4751 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
4753 case RTE_FLOW_ACTION_TYPE_SET_TTL:
4754 if (flow_dv_convert_action_modify_ttl(&res, actions,
4758 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
4760 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4761 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4762 if (flow_dv_convert_action_modify_tcp_seq(&res, actions,
4765 action_flags |= actions->type ==
4766 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
4767 MLX5_FLOW_ACTION_INC_TCP_SEQ :
4768 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
4771 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4772 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4773 if (flow_dv_convert_action_modify_tcp_ack(&res, actions,
4776 action_flags |= actions->type ==
4777 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
4778 MLX5_FLOW_ACTION_INC_TCP_ACK :
4779 MLX5_FLOW_ACTION_DEC_TCP_ACK;
4781 case RTE_FLOW_ACTION_TYPE_END:
4783 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS) {
4784 /* create modify action if needed. */
4785 if (flow_dv_modify_hdr_resource_register
4790 dev_flow->dv.actions[modify_action_position] =
4791 dev_flow->dv.modify_hdr->verbs_action;
4797 if ((action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS) &&
4798 modify_action_position == UINT32_MAX)
4799 modify_action_position = actions_n++;
4801 dev_flow->dv.actions_n = actions_n;
4802 flow->actions = action_flags;
4803 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4804 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
4806 switch (items->type) {
4807 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4808 flow_dv_translate_item_port_id(dev, match_mask,
4809 match_value, items);
4810 last_item = MLX5_FLOW_ITEM_PORT_ID;
4812 case RTE_FLOW_ITEM_TYPE_ETH:
4813 flow_dv_translate_item_eth(match_mask, match_value,
4815 matcher.priority = MLX5_PRIORITY_MAP_L2;
4816 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
4817 MLX5_FLOW_LAYER_OUTER_L2;
4819 case RTE_FLOW_ITEM_TYPE_VLAN:
4820 flow_dv_translate_item_vlan(match_mask, match_value,
4822 matcher.priority = MLX5_PRIORITY_MAP_L2;
4823 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
4824 MLX5_FLOW_LAYER_INNER_VLAN) :
4825 (MLX5_FLOW_LAYER_OUTER_L2 |
4826 MLX5_FLOW_LAYER_OUTER_VLAN);
4828 case RTE_FLOW_ITEM_TYPE_IPV4:
4829 flow_dv_translate_item_ipv4(match_mask, match_value,
4830 items, tunnel, attr->group);
4831 matcher.priority = MLX5_PRIORITY_MAP_L3;
4832 dev_flow->dv.hash_fields |=
4833 mlx5_flow_hashfields_adjust
4835 MLX5_IPV4_LAYER_TYPES,
4836 MLX5_IPV4_IBV_RX_HASH);
4837 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
4838 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4839 mlx5_flow_tunnel_ip_check(items, &last_item);
4841 case RTE_FLOW_ITEM_TYPE_IPV6:
4842 flow_dv_translate_item_ipv6(match_mask, match_value,
4843 items, tunnel, attr->group);
4844 matcher.priority = MLX5_PRIORITY_MAP_L3;
4845 dev_flow->dv.hash_fields |=
4846 mlx5_flow_hashfields_adjust
4848 MLX5_IPV6_LAYER_TYPES,
4849 MLX5_IPV6_IBV_RX_HASH);
4850 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
4851 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4852 mlx5_flow_tunnel_ip_check(items, &last_item);
4854 case RTE_FLOW_ITEM_TYPE_TCP:
4855 flow_dv_translate_item_tcp(match_mask, match_value,
4857 matcher.priority = MLX5_PRIORITY_MAP_L4;
4858 dev_flow->dv.hash_fields |=
4859 mlx5_flow_hashfields_adjust
4860 (dev_flow, tunnel, ETH_RSS_TCP,
4861 IBV_RX_HASH_SRC_PORT_TCP |
4862 IBV_RX_HASH_DST_PORT_TCP);
4863 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
4864 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4866 case RTE_FLOW_ITEM_TYPE_UDP:
4867 flow_dv_translate_item_udp(match_mask, match_value,
4869 matcher.priority = MLX5_PRIORITY_MAP_L4;
4870 dev_flow->dv.hash_fields |=
4871 mlx5_flow_hashfields_adjust
4872 (dev_flow, tunnel, ETH_RSS_UDP,
4873 IBV_RX_HASH_SRC_PORT_UDP |
4874 IBV_RX_HASH_DST_PORT_UDP);
4875 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
4876 MLX5_FLOW_LAYER_OUTER_L4_UDP;
4878 case RTE_FLOW_ITEM_TYPE_GRE:
4879 flow_dv_translate_item_gre(match_mask, match_value,
4881 last_item = MLX5_FLOW_LAYER_GRE;
4883 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
4884 flow_dv_translate_item_gre_key(match_mask,
4885 match_value, items);
4886 item_flags |= MLX5_FLOW_LAYER_GRE_KEY;
4888 case RTE_FLOW_ITEM_TYPE_NVGRE:
4889 flow_dv_translate_item_nvgre(match_mask, match_value,
4891 last_item = MLX5_FLOW_LAYER_GRE;
4893 case RTE_FLOW_ITEM_TYPE_VXLAN:
4894 flow_dv_translate_item_vxlan(match_mask, match_value,
4896 last_item = MLX5_FLOW_LAYER_VXLAN;
4898 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
4899 flow_dv_translate_item_vxlan(match_mask, match_value,
4901 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
4903 case RTE_FLOW_ITEM_TYPE_MPLS:
4904 flow_dv_translate_item_mpls(match_mask, match_value,
4905 items, last_item, tunnel);
4906 last_item = MLX5_FLOW_LAYER_MPLS;
4908 case RTE_FLOW_ITEM_TYPE_META:
4909 flow_dv_translate_item_meta(match_mask, match_value,
4911 last_item = MLX5_FLOW_ITEM_METADATA;
4913 case RTE_FLOW_ITEM_TYPE_ICMP:
4914 flow_dv_translate_item_icmp(match_mask, match_value,
4916 item_flags |= MLX5_FLOW_LAYER_ICMP;
4918 case RTE_FLOW_ITEM_TYPE_ICMP6:
4919 flow_dv_translate_item_icmp6(match_mask, match_value,
4921 item_flags |= MLX5_FLOW_LAYER_ICMP6;
4926 item_flags |= last_item;
4929 * In case of ingress traffic when E-Switch mode is enabled,
4930 * we have two cases where we need to set the source port manually.
4931 * The first one, is in case of Nic steering rule, and the second is
4932 * E-Switch rule where no port_id item was found. In both cases
4933 * the source port is set according the current port in use.
4935 if ((attr->ingress && !(item_flags & MLX5_FLOW_ITEM_PORT_ID)) &&
4936 (priv->representor || priv->master)) {
4937 if (flow_dv_translate_item_port_id(dev, match_mask,
4941 assert(!flow_dv_check_valid_spec(matcher.mask.buf,
4942 dev_flow->dv.value.buf));
4943 dev_flow->layers = item_flags;
4944 /* Register matcher. */
4945 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
4947 matcher.priority = mlx5_flow_adjust_priority(dev, priority,
4949 matcher.egress = attr->egress;
4950 matcher.group = attr->group;
4951 matcher.transfer = attr->transfer;
4952 if (flow_dv_matcher_register(dev, &matcher, dev_flow, error))
4958 * Apply the flow to the NIC.
4961 * Pointer to the Ethernet device structure.
4962 * @param[in, out] flow
4963 * Pointer to flow structure.
4965 * Pointer to error structure.
4968 * 0 on success, a negative errno value otherwise and rte_errno is set.
4971 flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
4972 struct rte_flow_error *error)
4974 struct mlx5_flow_dv *dv;
4975 struct mlx5_flow *dev_flow;
4976 struct mlx5_priv *priv = dev->data->dev_private;
4980 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
4983 if (flow->actions & MLX5_FLOW_ACTION_DROP) {
4984 if (flow->transfer) {
4985 dv->actions[n++] = priv->sh->esw_drop_action;
4987 dv->hrxq = mlx5_hrxq_drop_new(dev);
4991 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4993 "cannot get drop hash queue");
4996 dv->actions[n++] = dv->hrxq->action;
4998 } else if (flow->actions &
4999 (MLX5_FLOW_ACTION_QUEUE | MLX5_FLOW_ACTION_RSS)) {
5000 struct mlx5_hrxq *hrxq;
5002 hrxq = mlx5_hrxq_get(dev, flow->key,
5003 MLX5_RSS_HASH_KEY_LEN,
5006 flow->rss.queue_num);
5008 hrxq = mlx5_hrxq_new
5009 (dev, flow->key, MLX5_RSS_HASH_KEY_LEN,
5010 dv->hash_fields, (*flow->queue),
5011 flow->rss.queue_num,
5012 !!(dev_flow->layers &
5013 MLX5_FLOW_LAYER_TUNNEL));
5017 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5018 "cannot get hash queue");
5022 dv->actions[n++] = dv->hrxq->action;
5025 mlx5_glue->dv_create_flow(dv->matcher->matcher_object,
5026 (void *)&dv->value, n,
5029 rte_flow_error_set(error, errno,
5030 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5032 "hardware refuses to create flow");
5038 err = rte_errno; /* Save rte_errno before cleanup. */
5039 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
5040 struct mlx5_flow_dv *dv = &dev_flow->dv;
5042 if (flow->actions & MLX5_FLOW_ACTION_DROP)
5043 mlx5_hrxq_drop_release(dev);
5045 mlx5_hrxq_release(dev, dv->hrxq);
5049 rte_errno = err; /* Restore rte_errno. */
5054 * Release the flow matcher.
5057 * Pointer to Ethernet device.
5059 * Pointer to mlx5_flow.
5062 * 1 while a reference on it exists, 0 when freed.
5065 flow_dv_matcher_release(struct rte_eth_dev *dev,
5066 struct mlx5_flow *flow)
5068 struct mlx5_flow_dv_matcher *matcher = flow->dv.matcher;
5069 struct mlx5_priv *priv = dev->data->dev_private;
5070 struct mlx5_ibv_shared *sh = priv->sh;
5071 struct mlx5_flow_tbl_resource *tbl;
5073 assert(matcher->matcher_object);
5074 DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
5075 dev->data->port_id, (void *)matcher,
5076 rte_atomic32_read(&matcher->refcnt));
5077 if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
5078 claim_zero(mlx5_glue->dv_destroy_flow_matcher
5079 (matcher->matcher_object));
5080 LIST_REMOVE(matcher, next);
5081 if (matcher->egress)
5082 tbl = &sh->tx_tbl[matcher->group];
5084 tbl = &sh->rx_tbl[matcher->group];
5085 flow_dv_tbl_resource_release(tbl);
5087 DRV_LOG(DEBUG, "port %u matcher %p: removed",
5088 dev->data->port_id, (void *)matcher);
5095 * Release an encap/decap resource.
5098 * Pointer to mlx5_flow.
5101 * 1 while a reference on it exists, 0 when freed.
5104 flow_dv_encap_decap_resource_release(struct mlx5_flow *flow)
5106 struct mlx5_flow_dv_encap_decap_resource *cache_resource =
5107 flow->dv.encap_decap;
5109 assert(cache_resource->verbs_action);
5110 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
5111 (void *)cache_resource,
5112 rte_atomic32_read(&cache_resource->refcnt));
5113 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
5114 claim_zero(mlx5_glue->destroy_flow_action
5115 (cache_resource->verbs_action));
5116 LIST_REMOVE(cache_resource, next);
5117 rte_free(cache_resource);
5118 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
5119 (void *)cache_resource);
5126 * Release an jump to table action resource.
5129 * Pointer to mlx5_flow.
5132 * 1 while a reference on it exists, 0 when freed.
5135 flow_dv_jump_tbl_resource_release(struct mlx5_flow *flow)
5137 struct mlx5_flow_dv_jump_tbl_resource *cache_resource =
5140 assert(cache_resource->action);
5141 DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
5142 (void *)cache_resource,
5143 rte_atomic32_read(&cache_resource->refcnt));
5144 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
5145 claim_zero(mlx5_glue->destroy_flow_action
5146 (cache_resource->action));
5147 LIST_REMOVE(cache_resource, next);
5148 flow_dv_tbl_resource_release(cache_resource->tbl);
5149 rte_free(cache_resource);
5150 DRV_LOG(DEBUG, "jump table resource %p: removed",
5151 (void *)cache_resource);
5158 * Release a modify-header resource.
5161 * Pointer to mlx5_flow.
5164 * 1 while a reference on it exists, 0 when freed.
5167 flow_dv_modify_hdr_resource_release(struct mlx5_flow *flow)
5169 struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
5170 flow->dv.modify_hdr;
5172 assert(cache_resource->verbs_action);
5173 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
5174 (void *)cache_resource,
5175 rte_atomic32_read(&cache_resource->refcnt));
5176 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
5177 claim_zero(mlx5_glue->destroy_flow_action
5178 (cache_resource->verbs_action));
5179 LIST_REMOVE(cache_resource, next);
5180 rte_free(cache_resource);
5181 DRV_LOG(DEBUG, "modify-header resource %p: removed",
5182 (void *)cache_resource);
5189 * Release port ID action resource.
5192 * Pointer to mlx5_flow.
5195 * 1 while a reference on it exists, 0 when freed.
5198 flow_dv_port_id_action_resource_release(struct mlx5_flow *flow)
5200 struct mlx5_flow_dv_port_id_action_resource *cache_resource =
5201 flow->dv.port_id_action;
5203 assert(cache_resource->action);
5204 DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
5205 (void *)cache_resource,
5206 rte_atomic32_read(&cache_resource->refcnt));
5207 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
5208 claim_zero(mlx5_glue->destroy_flow_action
5209 (cache_resource->action));
5210 LIST_REMOVE(cache_resource, next);
5211 rte_free(cache_resource);
5212 DRV_LOG(DEBUG, "port id action resource %p: removed",
5213 (void *)cache_resource);
5220 * Remove the flow from the NIC but keeps it in memory.
5223 * Pointer to Ethernet device.
5224 * @param[in, out] flow
5225 * Pointer to flow structure.
5228 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
5230 struct mlx5_flow_dv *dv;
5231 struct mlx5_flow *dev_flow;
5235 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
5238 claim_zero(mlx5_glue->dv_destroy_flow(dv->flow));
5242 if (flow->actions & MLX5_FLOW_ACTION_DROP)
5243 mlx5_hrxq_drop_release(dev);
5245 mlx5_hrxq_release(dev, dv->hrxq);
5252 * Remove the flow from the NIC and the memory.
5255 * Pointer to the Ethernet device structure.
5256 * @param[in, out] flow
5257 * Pointer to flow structure.
5260 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
5262 struct mlx5_flow *dev_flow;
5266 flow_dv_remove(dev, flow);
5267 if (flow->counter) {
5268 flow_dv_counter_release(dev, flow->counter);
5269 flow->counter = NULL;
5271 if (flow->tag_resource) {
5272 flow_dv_tag_release(dev, flow->tag_resource);
5273 flow->tag_resource = NULL;
5275 while (!LIST_EMPTY(&flow->dev_flows)) {
5276 dev_flow = LIST_FIRST(&flow->dev_flows);
5277 LIST_REMOVE(dev_flow, next);
5278 if (dev_flow->dv.matcher)
5279 flow_dv_matcher_release(dev, dev_flow);
5280 if (dev_flow->dv.encap_decap)
5281 flow_dv_encap_decap_resource_release(dev_flow);
5282 if (dev_flow->dv.modify_hdr)
5283 flow_dv_modify_hdr_resource_release(dev_flow);
5284 if (dev_flow->dv.jump)
5285 flow_dv_jump_tbl_resource_release(dev_flow);
5286 if (dev_flow->dv.port_id_action)
5287 flow_dv_port_id_action_resource_release(dev_flow);
5293 * Query a dv flow rule for its statistics via devx.
5296 * Pointer to Ethernet device.
5298 * Pointer to the sub flow.
5300 * data retrieved by the query.
5302 * Perform verbose error reporting if not NULL.
5305 * 0 on success, a negative errno value otherwise and rte_errno is set.
5308 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
5309 void *data, struct rte_flow_error *error)
5311 struct mlx5_priv *priv = dev->data->dev_private;
5312 struct rte_flow_query_count *qc = data;
5314 if (!priv->config.devx)
5315 return rte_flow_error_set(error, ENOTSUP,
5316 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5318 "counters are not supported");
5319 if (flow->counter) {
5320 uint64_t pkts, bytes;
5321 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
5325 return rte_flow_error_set(error, -err,
5326 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5327 NULL, "cannot read counters");
5330 qc->hits = pkts - flow->counter->hits;
5331 qc->bytes = bytes - flow->counter->bytes;
5333 flow->counter->hits = pkts;
5334 flow->counter->bytes = bytes;
5338 return rte_flow_error_set(error, EINVAL,
5339 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5341 "counters are not available");
5347 * @see rte_flow_query()
5351 flow_dv_query(struct rte_eth_dev *dev,
5352 struct rte_flow *flow __rte_unused,
5353 const struct rte_flow_action *actions __rte_unused,
5354 void *data __rte_unused,
5355 struct rte_flow_error *error __rte_unused)
5359 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5360 switch (actions->type) {
5361 case RTE_FLOW_ACTION_TYPE_VOID:
5363 case RTE_FLOW_ACTION_TYPE_COUNT:
5364 ret = flow_dv_query_count(dev, flow, data, error);
5367 return rte_flow_error_set(error, ENOTSUP,
5368 RTE_FLOW_ERROR_TYPE_ACTION,
5370 "action not supported");
5377 * Mutex-protected thunk to flow_dv_translate().
5380 flow_d_translate(struct rte_eth_dev *dev,
5381 struct mlx5_flow *dev_flow,
5382 const struct rte_flow_attr *attr,
5383 const struct rte_flow_item items[],
5384 const struct rte_flow_action actions[],
5385 struct rte_flow_error *error)
5389 flow_d_shared_lock(dev);
5390 ret = flow_dv_translate(dev, dev_flow, attr, items, actions, error);
5391 flow_d_shared_unlock(dev);
5396 * Mutex-protected thunk to flow_dv_apply().
5399 flow_d_apply(struct rte_eth_dev *dev,
5400 struct rte_flow *flow,
5401 struct rte_flow_error *error)
5405 flow_d_shared_lock(dev);
5406 ret = flow_dv_apply(dev, flow, error);
5407 flow_d_shared_unlock(dev);
5412 * Mutex-protected thunk to flow_dv_remove().
5415 flow_d_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
5417 flow_d_shared_lock(dev);
5418 flow_dv_remove(dev, flow);
5419 flow_d_shared_unlock(dev);
5423 * Mutex-protected thunk to flow_dv_destroy().
5426 flow_d_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
5428 flow_d_shared_lock(dev);
5429 flow_dv_destroy(dev, flow);
5430 flow_d_shared_unlock(dev);
5433 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
5434 .validate = flow_dv_validate,
5435 .prepare = flow_dv_prepare,
5436 .translate = flow_d_translate,
5437 .apply = flow_d_apply,
5438 .remove = flow_d_remove,
5439 .destroy = flow_d_destroy,
5440 .query = flow_dv_query,
5443 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */