1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018 Mellanox Technologies, Ltd
11 #include <rte_common.h>
12 #include <rte_ether.h>
13 #include <rte_ethdev_driver.h>
15 #include <rte_flow_driver.h>
16 #include <rte_malloc.h>
17 #include <rte_cycles.h>
20 #include <rte_vxlan.h>
22 #include <rte_eal_paging.h>
25 #include <mlx5_glue.h>
26 #include <mlx5_devx_cmds.h>
28 #include <mlx5_malloc.h>
30 #include "mlx5_defs.h"
32 #include "mlx5_common_os.h"
33 #include "mlx5_flow.h"
34 #include "mlx5_flow_os.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
53 /* VLAN header definitions */
54 #define MLX5DV_FLOW_VLAN_PCP_SHIFT 13
55 #define MLX5DV_FLOW_VLAN_PCP_MASK (0x7 << MLX5DV_FLOW_VLAN_PCP_SHIFT)
56 #define MLX5DV_FLOW_VLAN_VID_MASK 0x0fff
57 #define MLX5DV_FLOW_VLAN_PCP_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK)
58 #define MLX5DV_FLOW_VLAN_VID_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_VID_MASK)
73 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
74 struct mlx5_flow_tbl_resource *tbl);
77 flow_dv_default_miss_resource_release(struct rte_eth_dev *dev);
80 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
81 uint32_t encap_decap_idx);
84 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
88 * Initialize flow attributes structure according to flow items' types.
90 * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
91 * mode. For tunnel mode, the items to be modified are the outermost ones.
94 * Pointer to item specification.
96 * Pointer to flow attributes structure.
98 * Pointer to the sub flow.
99 * @param[in] tunnel_decap
100 * Whether action is after tunnel decapsulation.
103 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
104 struct mlx5_flow *dev_flow, bool tunnel_decap)
106 uint64_t layers = dev_flow->handle->layers;
109 * If layers is already initialized, it means this dev_flow is the
110 * suffix flow, the layers flags is set by the prefix flow. Need to
111 * use the layer flags from prefix flow as the suffix flow may not
112 * have the user defined items as the flow is split.
115 if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
117 else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
119 if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
121 else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
126 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
127 uint8_t next_protocol = 0xff;
128 switch (item->type) {
129 case RTE_FLOW_ITEM_TYPE_GRE:
130 case RTE_FLOW_ITEM_TYPE_NVGRE:
131 case RTE_FLOW_ITEM_TYPE_VXLAN:
132 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
133 case RTE_FLOW_ITEM_TYPE_GENEVE:
134 case RTE_FLOW_ITEM_TYPE_MPLS:
138 case RTE_FLOW_ITEM_TYPE_IPV4:
141 if (item->mask != NULL &&
142 ((const struct rte_flow_item_ipv4 *)
143 item->mask)->hdr.next_proto_id)
145 ((const struct rte_flow_item_ipv4 *)
146 (item->spec))->hdr.next_proto_id &
147 ((const struct rte_flow_item_ipv4 *)
148 (item->mask))->hdr.next_proto_id;
149 if ((next_protocol == IPPROTO_IPIP ||
150 next_protocol == IPPROTO_IPV6) && tunnel_decap)
153 case RTE_FLOW_ITEM_TYPE_IPV6:
156 if (item->mask != NULL &&
157 ((const struct rte_flow_item_ipv6 *)
158 item->mask)->hdr.proto)
160 ((const struct rte_flow_item_ipv6 *)
161 (item->spec))->hdr.proto &
162 ((const struct rte_flow_item_ipv6 *)
163 (item->mask))->hdr.proto;
164 if ((next_protocol == IPPROTO_IPIP ||
165 next_protocol == IPPROTO_IPV6) && tunnel_decap)
168 case RTE_FLOW_ITEM_TYPE_UDP:
172 case RTE_FLOW_ITEM_TYPE_TCP:
184 * Convert rte_mtr_color to mlx5 color.
193 rte_col_2_mlx5_col(enum rte_color rcol)
196 case RTE_COLOR_GREEN:
197 return MLX5_FLOW_COLOR_GREEN;
198 case RTE_COLOR_YELLOW:
199 return MLX5_FLOW_COLOR_YELLOW;
201 return MLX5_FLOW_COLOR_RED;
205 return MLX5_FLOW_COLOR_UNDEFINED;
208 struct field_modify_info {
209 uint32_t size; /* Size of field in protocol header, in bytes. */
210 uint32_t offset; /* Offset of field in protocol header, in bytes. */
211 enum mlx5_modification_field id;
214 struct field_modify_info modify_eth[] = {
215 {4, 0, MLX5_MODI_OUT_DMAC_47_16},
216 {2, 4, MLX5_MODI_OUT_DMAC_15_0},
217 {4, 6, MLX5_MODI_OUT_SMAC_47_16},
218 {2, 10, MLX5_MODI_OUT_SMAC_15_0},
222 struct field_modify_info modify_vlan_out_first_vid[] = {
223 /* Size in bits !!! */
224 {12, 0, MLX5_MODI_OUT_FIRST_VID},
228 struct field_modify_info modify_ipv4[] = {
229 {1, 1, MLX5_MODI_OUT_IP_DSCP},
230 {1, 8, MLX5_MODI_OUT_IPV4_TTL},
231 {4, 12, MLX5_MODI_OUT_SIPV4},
232 {4, 16, MLX5_MODI_OUT_DIPV4},
236 struct field_modify_info modify_ipv6[] = {
237 {1, 0, MLX5_MODI_OUT_IP_DSCP},
238 {1, 7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
239 {4, 8, MLX5_MODI_OUT_SIPV6_127_96},
240 {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
241 {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
242 {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
243 {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
244 {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
245 {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
246 {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
250 struct field_modify_info modify_udp[] = {
251 {2, 0, MLX5_MODI_OUT_UDP_SPORT},
252 {2, 2, MLX5_MODI_OUT_UDP_DPORT},
256 struct field_modify_info modify_tcp[] = {
257 {2, 0, MLX5_MODI_OUT_TCP_SPORT},
258 {2, 2, MLX5_MODI_OUT_TCP_DPORT},
259 {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
260 {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
265 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
266 uint8_t next_protocol, uint64_t *item_flags,
269 MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
270 item->type == RTE_FLOW_ITEM_TYPE_IPV6);
271 if (next_protocol == IPPROTO_IPIP) {
272 *item_flags |= MLX5_FLOW_LAYER_IPIP;
275 if (next_protocol == IPPROTO_IPV6) {
276 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
282 * Acquire the synchronizing object to protect multithreaded access
283 * to shared dv context. Lock occurs only if context is actually
284 * shared, i.e. we have multiport IB device and representors are
288 * Pointer to the rte_eth_dev structure.
291 flow_dv_shared_lock(struct rte_eth_dev *dev)
293 struct mlx5_priv *priv = dev->data->dev_private;
294 struct mlx5_dev_ctx_shared *sh = priv->sh;
296 if (sh->dv_refcnt > 1) {
299 ret = pthread_mutex_lock(&sh->dv_mutex);
306 flow_dv_shared_unlock(struct rte_eth_dev *dev)
308 struct mlx5_priv *priv = dev->data->dev_private;
309 struct mlx5_dev_ctx_shared *sh = priv->sh;
311 if (sh->dv_refcnt > 1) {
314 ret = pthread_mutex_unlock(&sh->dv_mutex);
320 /* Update VLAN's VID/PCP based on input rte_flow_action.
323 * Pointer to struct rte_flow_action.
325 * Pointer to struct rte_vlan_hdr.
328 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
329 struct rte_vlan_hdr *vlan)
332 if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
334 ((const struct rte_flow_action_of_set_vlan_pcp *)
335 action->conf)->vlan_pcp;
336 vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
337 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
338 vlan->vlan_tci |= vlan_tci;
339 } else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
340 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
341 vlan->vlan_tci |= rte_be_to_cpu_16
342 (((const struct rte_flow_action_of_set_vlan_vid *)
343 action->conf)->vlan_vid);
348 * Fetch 1, 2, 3 or 4 byte field from the byte array
349 * and return as unsigned integer in host-endian format.
352 * Pointer to data array.
354 * Size of field to extract.
357 * converted field in host endian format.
359 static inline uint32_t
360 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
369 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
372 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
373 ret = (ret << 8) | *(data + sizeof(uint16_t));
376 ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
387 * Convert modify-header action to DV specification.
389 * Data length of each action is determined by provided field description
390 * and the item mask. Data bit offset and width of each action is determined
391 * by provided item mask.
394 * Pointer to item specification.
396 * Pointer to field modification information.
397 * For MLX5_MODIFICATION_TYPE_SET specifies destination field.
398 * For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
399 * For MLX5_MODIFICATION_TYPE_COPY specifies source field.
401 * Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
402 * Negative offset value sets the same offset as source offset.
403 * size field is ignored, value is taken from source field.
404 * @param[in,out] resource
405 * Pointer to the modify-header resource.
407 * Type of modification.
409 * Pointer to the error structure.
412 * 0 on success, a negative errno value otherwise and rte_errno is set.
415 flow_dv_convert_modify_action(struct rte_flow_item *item,
416 struct field_modify_info *field,
417 struct field_modify_info *dcopy,
418 struct mlx5_flow_dv_modify_hdr_resource *resource,
419 uint32_t type, struct rte_flow_error *error)
421 uint32_t i = resource->actions_num;
422 struct mlx5_modification_cmd *actions = resource->actions;
425 * The item and mask are provided in big-endian format.
426 * The fields should be presented as in big-endian format either.
427 * Mask must be always present, it defines the actual field width.
429 MLX5_ASSERT(item->mask);
430 MLX5_ASSERT(field->size);
437 if (i >= MLX5_MAX_MODIFY_NUM)
438 return rte_flow_error_set(error, EINVAL,
439 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
440 "too many items to modify");
441 /* Fetch variable byte size mask from the array. */
442 mask = flow_dv_fetch_field((const uint8_t *)item->mask +
443 field->offset, field->size);
448 /* Deduce actual data width in bits from mask value. */
449 off_b = rte_bsf32(mask);
450 size_b = sizeof(uint32_t) * CHAR_BIT -
451 off_b - __builtin_clz(mask);
453 size_b = size_b == sizeof(uint32_t) * CHAR_BIT ? 0 : size_b;
454 actions[i] = (struct mlx5_modification_cmd) {
460 /* Convert entire record to expected big-endian format. */
461 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
462 if (type == MLX5_MODIFICATION_TYPE_COPY) {
464 actions[i].dst_field = dcopy->id;
465 actions[i].dst_offset =
466 (int)dcopy->offset < 0 ? off_b : dcopy->offset;
467 /* Convert entire record to big-endian format. */
468 actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
470 MLX5_ASSERT(item->spec);
471 data = flow_dv_fetch_field((const uint8_t *)item->spec +
472 field->offset, field->size);
473 /* Shift out the trailing masked bits from data. */
474 data = (data & mask) >> off_b;
475 actions[i].data1 = rte_cpu_to_be_32(data);
479 } while (field->size);
480 if (resource->actions_num == i)
481 return rte_flow_error_set(error, EINVAL,
482 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
483 "invalid modification flow item");
484 resource->actions_num = i;
489 * Convert modify-header set IPv4 address action to DV specification.
491 * @param[in,out] resource
492 * Pointer to the modify-header resource.
494 * Pointer to action specification.
496 * Pointer to the error structure.
499 * 0 on success, a negative errno value otherwise and rte_errno is set.
502 flow_dv_convert_action_modify_ipv4
503 (struct mlx5_flow_dv_modify_hdr_resource *resource,
504 const struct rte_flow_action *action,
505 struct rte_flow_error *error)
507 const struct rte_flow_action_set_ipv4 *conf =
508 (const struct rte_flow_action_set_ipv4 *)(action->conf);
509 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
510 struct rte_flow_item_ipv4 ipv4;
511 struct rte_flow_item_ipv4 ipv4_mask;
513 memset(&ipv4, 0, sizeof(ipv4));
514 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
515 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
516 ipv4.hdr.src_addr = conf->ipv4_addr;
517 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
519 ipv4.hdr.dst_addr = conf->ipv4_addr;
520 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
523 item.mask = &ipv4_mask;
524 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
525 MLX5_MODIFICATION_TYPE_SET, error);
529 * Convert modify-header set IPv6 address action to DV specification.
531 * @param[in,out] resource
532 * Pointer to the modify-header resource.
534 * Pointer to action specification.
536 * Pointer to the error structure.
539 * 0 on success, a negative errno value otherwise and rte_errno is set.
542 flow_dv_convert_action_modify_ipv6
543 (struct mlx5_flow_dv_modify_hdr_resource *resource,
544 const struct rte_flow_action *action,
545 struct rte_flow_error *error)
547 const struct rte_flow_action_set_ipv6 *conf =
548 (const struct rte_flow_action_set_ipv6 *)(action->conf);
549 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
550 struct rte_flow_item_ipv6 ipv6;
551 struct rte_flow_item_ipv6 ipv6_mask;
553 memset(&ipv6, 0, sizeof(ipv6));
554 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
555 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
556 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
557 sizeof(ipv6.hdr.src_addr));
558 memcpy(&ipv6_mask.hdr.src_addr,
559 &rte_flow_item_ipv6_mask.hdr.src_addr,
560 sizeof(ipv6.hdr.src_addr));
562 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
563 sizeof(ipv6.hdr.dst_addr));
564 memcpy(&ipv6_mask.hdr.dst_addr,
565 &rte_flow_item_ipv6_mask.hdr.dst_addr,
566 sizeof(ipv6.hdr.dst_addr));
569 item.mask = &ipv6_mask;
570 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
571 MLX5_MODIFICATION_TYPE_SET, error);
575 * Convert modify-header set MAC address action to DV specification.
577 * @param[in,out] resource
578 * Pointer to the modify-header resource.
580 * Pointer to action specification.
582 * Pointer to the error structure.
585 * 0 on success, a negative errno value otherwise and rte_errno is set.
588 flow_dv_convert_action_modify_mac
589 (struct mlx5_flow_dv_modify_hdr_resource *resource,
590 const struct rte_flow_action *action,
591 struct rte_flow_error *error)
593 const struct rte_flow_action_set_mac *conf =
594 (const struct rte_flow_action_set_mac *)(action->conf);
595 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
596 struct rte_flow_item_eth eth;
597 struct rte_flow_item_eth eth_mask;
599 memset(ð, 0, sizeof(eth));
600 memset(ð_mask, 0, sizeof(eth_mask));
601 if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
602 memcpy(ð.src.addr_bytes, &conf->mac_addr,
603 sizeof(eth.src.addr_bytes));
604 memcpy(ð_mask.src.addr_bytes,
605 &rte_flow_item_eth_mask.src.addr_bytes,
606 sizeof(eth_mask.src.addr_bytes));
608 memcpy(ð.dst.addr_bytes, &conf->mac_addr,
609 sizeof(eth.dst.addr_bytes));
610 memcpy(ð_mask.dst.addr_bytes,
611 &rte_flow_item_eth_mask.dst.addr_bytes,
612 sizeof(eth_mask.dst.addr_bytes));
615 item.mask = ð_mask;
616 return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
617 MLX5_MODIFICATION_TYPE_SET, error);
621 * Convert modify-header set VLAN VID action to DV specification.
623 * @param[in,out] resource
624 * Pointer to the modify-header resource.
626 * Pointer to action specification.
628 * Pointer to the error structure.
631 * 0 on success, a negative errno value otherwise and rte_errno is set.
634 flow_dv_convert_action_modify_vlan_vid
635 (struct mlx5_flow_dv_modify_hdr_resource *resource,
636 const struct rte_flow_action *action,
637 struct rte_flow_error *error)
639 const struct rte_flow_action_of_set_vlan_vid *conf =
640 (const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
641 int i = resource->actions_num;
642 struct mlx5_modification_cmd *actions = resource->actions;
643 struct field_modify_info *field = modify_vlan_out_first_vid;
645 if (i >= MLX5_MAX_MODIFY_NUM)
646 return rte_flow_error_set(error, EINVAL,
647 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
648 "too many items to modify");
649 actions[i] = (struct mlx5_modification_cmd) {
650 .action_type = MLX5_MODIFICATION_TYPE_SET,
652 .length = field->size,
653 .offset = field->offset,
655 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
656 actions[i].data1 = conf->vlan_vid;
657 actions[i].data1 = actions[i].data1 << 16;
658 resource->actions_num = ++i;
663 * Convert modify-header set TP action to DV specification.
665 * @param[in,out] resource
666 * Pointer to the modify-header resource.
668 * Pointer to action specification.
670 * Pointer to rte_flow_item objects list.
672 * Pointer to flow attributes structure.
673 * @param[in] dev_flow
674 * Pointer to the sub flow.
675 * @param[in] tunnel_decap
676 * Whether action is after tunnel decapsulation.
678 * Pointer to the error structure.
681 * 0 on success, a negative errno value otherwise and rte_errno is set.
684 flow_dv_convert_action_modify_tp
685 (struct mlx5_flow_dv_modify_hdr_resource *resource,
686 const struct rte_flow_action *action,
687 const struct rte_flow_item *items,
688 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
689 bool tunnel_decap, struct rte_flow_error *error)
691 const struct rte_flow_action_set_tp *conf =
692 (const struct rte_flow_action_set_tp *)(action->conf);
693 struct rte_flow_item item;
694 struct rte_flow_item_udp udp;
695 struct rte_flow_item_udp udp_mask;
696 struct rte_flow_item_tcp tcp;
697 struct rte_flow_item_tcp tcp_mask;
698 struct field_modify_info *field;
701 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
703 memset(&udp, 0, sizeof(udp));
704 memset(&udp_mask, 0, sizeof(udp_mask));
705 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
706 udp.hdr.src_port = conf->port;
707 udp_mask.hdr.src_port =
708 rte_flow_item_udp_mask.hdr.src_port;
710 udp.hdr.dst_port = conf->port;
711 udp_mask.hdr.dst_port =
712 rte_flow_item_udp_mask.hdr.dst_port;
714 item.type = RTE_FLOW_ITEM_TYPE_UDP;
716 item.mask = &udp_mask;
719 MLX5_ASSERT(attr->tcp);
720 memset(&tcp, 0, sizeof(tcp));
721 memset(&tcp_mask, 0, sizeof(tcp_mask));
722 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
723 tcp.hdr.src_port = conf->port;
724 tcp_mask.hdr.src_port =
725 rte_flow_item_tcp_mask.hdr.src_port;
727 tcp.hdr.dst_port = conf->port;
728 tcp_mask.hdr.dst_port =
729 rte_flow_item_tcp_mask.hdr.dst_port;
731 item.type = RTE_FLOW_ITEM_TYPE_TCP;
733 item.mask = &tcp_mask;
736 return flow_dv_convert_modify_action(&item, field, NULL, resource,
737 MLX5_MODIFICATION_TYPE_SET, error);
741 * Convert modify-header set TTL action to DV specification.
743 * @param[in,out] resource
744 * Pointer to the modify-header resource.
746 * Pointer to action specification.
748 * Pointer to rte_flow_item objects list.
750 * Pointer to flow attributes structure.
751 * @param[in] dev_flow
752 * Pointer to the sub flow.
753 * @param[in] tunnel_decap
754 * Whether action is after tunnel decapsulation.
756 * Pointer to the error structure.
759 * 0 on success, a negative errno value otherwise and rte_errno is set.
762 flow_dv_convert_action_modify_ttl
763 (struct mlx5_flow_dv_modify_hdr_resource *resource,
764 const struct rte_flow_action *action,
765 const struct rte_flow_item *items,
766 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
767 bool tunnel_decap, struct rte_flow_error *error)
769 const struct rte_flow_action_set_ttl *conf =
770 (const struct rte_flow_action_set_ttl *)(action->conf);
771 struct rte_flow_item item;
772 struct rte_flow_item_ipv4 ipv4;
773 struct rte_flow_item_ipv4 ipv4_mask;
774 struct rte_flow_item_ipv6 ipv6;
775 struct rte_flow_item_ipv6 ipv6_mask;
776 struct field_modify_info *field;
779 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
781 memset(&ipv4, 0, sizeof(ipv4));
782 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
783 ipv4.hdr.time_to_live = conf->ttl_value;
784 ipv4_mask.hdr.time_to_live = 0xFF;
785 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
787 item.mask = &ipv4_mask;
790 MLX5_ASSERT(attr->ipv6);
791 memset(&ipv6, 0, sizeof(ipv6));
792 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
793 ipv6.hdr.hop_limits = conf->ttl_value;
794 ipv6_mask.hdr.hop_limits = 0xFF;
795 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
797 item.mask = &ipv6_mask;
800 return flow_dv_convert_modify_action(&item, field, NULL, resource,
801 MLX5_MODIFICATION_TYPE_SET, error);
805 * Convert modify-header decrement TTL action to DV specification.
807 * @param[in,out] resource
808 * Pointer to the modify-header resource.
810 * Pointer to action specification.
812 * Pointer to rte_flow_item objects list.
814 * Pointer to flow attributes structure.
815 * @param[in] dev_flow
816 * Pointer to the sub flow.
817 * @param[in] tunnel_decap
818 * Whether action is after tunnel decapsulation.
820 * Pointer to the error structure.
823 * 0 on success, a negative errno value otherwise and rte_errno is set.
826 flow_dv_convert_action_modify_dec_ttl
827 (struct mlx5_flow_dv_modify_hdr_resource *resource,
828 const struct rte_flow_item *items,
829 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
830 bool tunnel_decap, struct rte_flow_error *error)
832 struct rte_flow_item item;
833 struct rte_flow_item_ipv4 ipv4;
834 struct rte_flow_item_ipv4 ipv4_mask;
835 struct rte_flow_item_ipv6 ipv6;
836 struct rte_flow_item_ipv6 ipv6_mask;
837 struct field_modify_info *field;
840 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
842 memset(&ipv4, 0, sizeof(ipv4));
843 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
844 ipv4.hdr.time_to_live = 0xFF;
845 ipv4_mask.hdr.time_to_live = 0xFF;
846 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
848 item.mask = &ipv4_mask;
851 MLX5_ASSERT(attr->ipv6);
852 memset(&ipv6, 0, sizeof(ipv6));
853 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
854 ipv6.hdr.hop_limits = 0xFF;
855 ipv6_mask.hdr.hop_limits = 0xFF;
856 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
858 item.mask = &ipv6_mask;
861 return flow_dv_convert_modify_action(&item, field, NULL, resource,
862 MLX5_MODIFICATION_TYPE_ADD, error);
866 * Convert modify-header increment/decrement TCP Sequence number
867 * to DV specification.
869 * @param[in,out] resource
870 * Pointer to the modify-header resource.
872 * Pointer to action specification.
874 * Pointer to the error structure.
877 * 0 on success, a negative errno value otherwise and rte_errno is set.
880 flow_dv_convert_action_modify_tcp_seq
881 (struct mlx5_flow_dv_modify_hdr_resource *resource,
882 const struct rte_flow_action *action,
883 struct rte_flow_error *error)
885 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
886 uint64_t value = rte_be_to_cpu_32(*conf);
887 struct rte_flow_item item;
888 struct rte_flow_item_tcp tcp;
889 struct rte_flow_item_tcp tcp_mask;
891 memset(&tcp, 0, sizeof(tcp));
892 memset(&tcp_mask, 0, sizeof(tcp_mask));
893 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
895 * The HW has no decrement operation, only increment operation.
896 * To simulate decrement X from Y using increment operation
897 * we need to add UINT32_MAX X times to Y.
898 * Each adding of UINT32_MAX decrements Y by 1.
901 tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
902 tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
903 item.type = RTE_FLOW_ITEM_TYPE_TCP;
905 item.mask = &tcp_mask;
906 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
907 MLX5_MODIFICATION_TYPE_ADD, error);
911 * Convert modify-header increment/decrement TCP Acknowledgment number
912 * to DV specification.
914 * @param[in,out] resource
915 * Pointer to the modify-header resource.
917 * Pointer to action specification.
919 * Pointer to the error structure.
922 * 0 on success, a negative errno value otherwise and rte_errno is set.
925 flow_dv_convert_action_modify_tcp_ack
926 (struct mlx5_flow_dv_modify_hdr_resource *resource,
927 const struct rte_flow_action *action,
928 struct rte_flow_error *error)
930 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
931 uint64_t value = rte_be_to_cpu_32(*conf);
932 struct rte_flow_item item;
933 struct rte_flow_item_tcp tcp;
934 struct rte_flow_item_tcp tcp_mask;
936 memset(&tcp, 0, sizeof(tcp));
937 memset(&tcp_mask, 0, sizeof(tcp_mask));
938 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
940 * The HW has no decrement operation, only increment operation.
941 * To simulate decrement X from Y using increment operation
942 * we need to add UINT32_MAX X times to Y.
943 * Each adding of UINT32_MAX decrements Y by 1.
946 tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
947 tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
948 item.type = RTE_FLOW_ITEM_TYPE_TCP;
950 item.mask = &tcp_mask;
951 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
952 MLX5_MODIFICATION_TYPE_ADD, error);
955 static enum mlx5_modification_field reg_to_field[] = {
956 [REG_NON] = MLX5_MODI_OUT_NONE,
957 [REG_A] = MLX5_MODI_META_DATA_REG_A,
958 [REG_B] = MLX5_MODI_META_DATA_REG_B,
959 [REG_C_0] = MLX5_MODI_META_REG_C_0,
960 [REG_C_1] = MLX5_MODI_META_REG_C_1,
961 [REG_C_2] = MLX5_MODI_META_REG_C_2,
962 [REG_C_3] = MLX5_MODI_META_REG_C_3,
963 [REG_C_4] = MLX5_MODI_META_REG_C_4,
964 [REG_C_5] = MLX5_MODI_META_REG_C_5,
965 [REG_C_6] = MLX5_MODI_META_REG_C_6,
966 [REG_C_7] = MLX5_MODI_META_REG_C_7,
970 * Convert register set to DV specification.
972 * @param[in,out] resource
973 * Pointer to the modify-header resource.
975 * Pointer to action specification.
977 * Pointer to the error structure.
980 * 0 on success, a negative errno value otherwise and rte_errno is set.
983 flow_dv_convert_action_set_reg
984 (struct mlx5_flow_dv_modify_hdr_resource *resource,
985 const struct rte_flow_action *action,
986 struct rte_flow_error *error)
988 const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
989 struct mlx5_modification_cmd *actions = resource->actions;
990 uint32_t i = resource->actions_num;
992 if (i >= MLX5_MAX_MODIFY_NUM)
993 return rte_flow_error_set(error, EINVAL,
994 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
995 "too many items to modify");
996 MLX5_ASSERT(conf->id != REG_NON);
997 MLX5_ASSERT(conf->id < RTE_DIM(reg_to_field));
998 actions[i] = (struct mlx5_modification_cmd) {
999 .action_type = MLX5_MODIFICATION_TYPE_SET,
1000 .field = reg_to_field[conf->id],
1002 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
1003 actions[i].data1 = rte_cpu_to_be_32(conf->data);
1005 resource->actions_num = i;
1010 * Convert SET_TAG action to DV specification.
1013 * Pointer to the rte_eth_dev structure.
1014 * @param[in,out] resource
1015 * Pointer to the modify-header resource.
1017 * Pointer to action specification.
1019 * Pointer to the error structure.
1022 * 0 on success, a negative errno value otherwise and rte_errno is set.
1025 flow_dv_convert_action_set_tag
1026 (struct rte_eth_dev *dev,
1027 struct mlx5_flow_dv_modify_hdr_resource *resource,
1028 const struct rte_flow_action_set_tag *conf,
1029 struct rte_flow_error *error)
1031 rte_be32_t data = rte_cpu_to_be_32(conf->data);
1032 rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1033 struct rte_flow_item item = {
1037 struct field_modify_info reg_c_x[] = {
1040 enum mlx5_modification_field reg_type;
1043 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1046 MLX5_ASSERT(ret != REG_NON);
1047 MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1048 reg_type = reg_to_field[ret];
1049 MLX5_ASSERT(reg_type > 0);
1050 reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1051 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1052 MLX5_MODIFICATION_TYPE_SET, error);
1056 * Convert internal COPY_REG action to DV specification.
1059 * Pointer to the rte_eth_dev structure.
1060 * @param[in,out] res
1061 * Pointer to the modify-header resource.
1063 * Pointer to action specification.
1065 * Pointer to the error structure.
1068 * 0 on success, a negative errno value otherwise and rte_errno is set.
1071 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1072 struct mlx5_flow_dv_modify_hdr_resource *res,
1073 const struct rte_flow_action *action,
1074 struct rte_flow_error *error)
1076 const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1077 rte_be32_t mask = RTE_BE32(UINT32_MAX);
1078 struct rte_flow_item item = {
1082 struct field_modify_info reg_src[] = {
1083 {4, 0, reg_to_field[conf->src]},
1086 struct field_modify_info reg_dst = {
1088 .id = reg_to_field[conf->dst],
1090 /* Adjust reg_c[0] usage according to reported mask. */
1091 if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1092 struct mlx5_priv *priv = dev->data->dev_private;
1093 uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1095 MLX5_ASSERT(reg_c0);
1096 MLX5_ASSERT(priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY);
1097 if (conf->dst == REG_C_0) {
1098 /* Copy to reg_c[0], within mask only. */
1099 reg_dst.offset = rte_bsf32(reg_c0);
1101 * Mask is ignoring the enianness, because
1102 * there is no conversion in datapath.
1104 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1105 /* Copy from destination lower bits to reg_c[0]. */
1106 mask = reg_c0 >> reg_dst.offset;
1108 /* Copy from destination upper bits to reg_c[0]. */
1109 mask = reg_c0 << (sizeof(reg_c0) * CHAR_BIT -
1110 rte_fls_u32(reg_c0));
1113 mask = rte_cpu_to_be_32(reg_c0);
1114 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1115 /* Copy from reg_c[0] to destination lower bits. */
1118 /* Copy from reg_c[0] to destination upper bits. */
1119 reg_dst.offset = sizeof(reg_c0) * CHAR_BIT -
1120 (rte_fls_u32(reg_c0) -
1125 return flow_dv_convert_modify_action(&item,
1126 reg_src, ®_dst, res,
1127 MLX5_MODIFICATION_TYPE_COPY,
1132 * Convert MARK action to DV specification. This routine is used
1133 * in extensive metadata only and requires metadata register to be
1134 * handled. In legacy mode hardware tag resource is engaged.
1137 * Pointer to the rte_eth_dev structure.
1139 * Pointer to MARK action specification.
1140 * @param[in,out] resource
1141 * Pointer to the modify-header resource.
1143 * Pointer to the error structure.
1146 * 0 on success, a negative errno value otherwise and rte_errno is set.
1149 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1150 const struct rte_flow_action_mark *conf,
1151 struct mlx5_flow_dv_modify_hdr_resource *resource,
1152 struct rte_flow_error *error)
1154 struct mlx5_priv *priv = dev->data->dev_private;
1155 rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1156 priv->sh->dv_mark_mask);
1157 rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1158 struct rte_flow_item item = {
1162 struct field_modify_info reg_c_x[] = {
1168 return rte_flow_error_set(error, EINVAL,
1169 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1170 NULL, "zero mark action mask");
1171 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1174 MLX5_ASSERT(reg > 0);
1175 if (reg == REG_C_0) {
1176 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1177 uint32_t shl_c0 = rte_bsf32(msk_c0);
1179 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1180 mask = rte_cpu_to_be_32(mask) & msk_c0;
1181 mask = rte_cpu_to_be_32(mask << shl_c0);
1183 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1184 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1185 MLX5_MODIFICATION_TYPE_SET, error);
1189 * Get metadata register index for specified steering domain.
1192 * Pointer to the rte_eth_dev structure.
1194 * Attributes of flow to determine steering domain.
1196 * Pointer to the error structure.
1199 * positive index on success, a negative errno value otherwise
1200 * and rte_errno is set.
1202 static enum modify_reg
1203 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1204 const struct rte_flow_attr *attr,
1205 struct rte_flow_error *error)
1208 mlx5_flow_get_reg_id(dev, attr->transfer ?
1212 MLX5_METADATA_RX, 0, error);
1214 return rte_flow_error_set(error,
1215 ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1216 NULL, "unavailable "
1217 "metadata register");
1222 * Convert SET_META action to DV specification.
1225 * Pointer to the rte_eth_dev structure.
1226 * @param[in,out] resource
1227 * Pointer to the modify-header resource.
1229 * Attributes of flow that includes this item.
1231 * Pointer to action specification.
1233 * Pointer to the error structure.
1236 * 0 on success, a negative errno value otherwise and rte_errno is set.
1239 flow_dv_convert_action_set_meta
1240 (struct rte_eth_dev *dev,
1241 struct mlx5_flow_dv_modify_hdr_resource *resource,
1242 const struct rte_flow_attr *attr,
1243 const struct rte_flow_action_set_meta *conf,
1244 struct rte_flow_error *error)
1246 uint32_t data = conf->data;
1247 uint32_t mask = conf->mask;
1248 struct rte_flow_item item = {
1252 struct field_modify_info reg_c_x[] = {
1255 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1260 * In datapath code there is no endianness
1261 * coversions for perfromance reasons, all
1262 * pattern conversions are done in rte_flow.
1264 if (reg == REG_C_0) {
1265 struct mlx5_priv *priv = dev->data->dev_private;
1266 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1269 MLX5_ASSERT(msk_c0);
1270 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1271 shl_c0 = rte_bsf32(msk_c0);
1273 shl_c0 = sizeof(msk_c0) * CHAR_BIT - rte_fls_u32(msk_c0);
1277 MLX5_ASSERT(!(~msk_c0 & rte_cpu_to_be_32(mask)));
1279 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1280 /* The routine expects parameters in memory as big-endian ones. */
1281 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1282 MLX5_MODIFICATION_TYPE_SET, error);
1286 * Convert modify-header set IPv4 DSCP action to DV specification.
1288 * @param[in,out] resource
1289 * Pointer to the modify-header resource.
1291 * Pointer to action specification.
1293 * Pointer to the error structure.
1296 * 0 on success, a negative errno value otherwise and rte_errno is set.
1299 flow_dv_convert_action_modify_ipv4_dscp
1300 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1301 const struct rte_flow_action *action,
1302 struct rte_flow_error *error)
1304 const struct rte_flow_action_set_dscp *conf =
1305 (const struct rte_flow_action_set_dscp *)(action->conf);
1306 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1307 struct rte_flow_item_ipv4 ipv4;
1308 struct rte_flow_item_ipv4 ipv4_mask;
1310 memset(&ipv4, 0, sizeof(ipv4));
1311 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1312 ipv4.hdr.type_of_service = conf->dscp;
1313 ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1315 item.mask = &ipv4_mask;
1316 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1317 MLX5_MODIFICATION_TYPE_SET, error);
1321 * Convert modify-header set IPv6 DSCP action to DV specification.
1323 * @param[in,out] resource
1324 * Pointer to the modify-header resource.
1326 * Pointer to action specification.
1328 * Pointer to the error structure.
1331 * 0 on success, a negative errno value otherwise and rte_errno is set.
1334 flow_dv_convert_action_modify_ipv6_dscp
1335 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1336 const struct rte_flow_action *action,
1337 struct rte_flow_error *error)
1339 const struct rte_flow_action_set_dscp *conf =
1340 (const struct rte_flow_action_set_dscp *)(action->conf);
1341 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1342 struct rte_flow_item_ipv6 ipv6;
1343 struct rte_flow_item_ipv6 ipv6_mask;
1345 memset(&ipv6, 0, sizeof(ipv6));
1346 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1348 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1349 * rdma-core only accept the DSCP bits byte aligned start from
1350 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1351 * bits in IPv6 case as rdma-core requires byte aligned value.
1353 ipv6.hdr.vtc_flow = conf->dscp;
1354 ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1356 item.mask = &ipv6_mask;
1357 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1358 MLX5_MODIFICATION_TYPE_SET, error);
1362 * Validate MARK item.
1365 * Pointer to the rte_eth_dev structure.
1367 * Item specification.
1369 * Attributes of flow that includes this item.
1371 * Pointer to error structure.
1374 * 0 on success, a negative errno value otherwise and rte_errno is set.
1377 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1378 const struct rte_flow_item *item,
1379 const struct rte_flow_attr *attr __rte_unused,
1380 struct rte_flow_error *error)
1382 struct mlx5_priv *priv = dev->data->dev_private;
1383 struct mlx5_dev_config *config = &priv->config;
1384 const struct rte_flow_item_mark *spec = item->spec;
1385 const struct rte_flow_item_mark *mask = item->mask;
1386 const struct rte_flow_item_mark nic_mask = {
1387 .id = priv->sh->dv_mark_mask,
1391 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1392 return rte_flow_error_set(error, ENOTSUP,
1393 RTE_FLOW_ERROR_TYPE_ITEM, item,
1394 "extended metadata feature"
1396 if (!mlx5_flow_ext_mreg_supported(dev))
1397 return rte_flow_error_set(error, ENOTSUP,
1398 RTE_FLOW_ERROR_TYPE_ITEM, item,
1399 "extended metadata register"
1400 " isn't supported");
1402 return rte_flow_error_set(error, ENOTSUP,
1403 RTE_FLOW_ERROR_TYPE_ITEM, item,
1404 "extended metadata register"
1405 " isn't available");
1406 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1410 return rte_flow_error_set(error, EINVAL,
1411 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1413 "data cannot be empty");
1414 if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1415 return rte_flow_error_set(error, EINVAL,
1416 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1418 "mark id exceeds the limit");
1422 return rte_flow_error_set(error, EINVAL,
1423 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1424 "mask cannot be zero");
1426 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1427 (const uint8_t *)&nic_mask,
1428 sizeof(struct rte_flow_item_mark),
1429 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1436 * Validate META item.
1439 * Pointer to the rte_eth_dev structure.
1441 * Item specification.
1443 * Attributes of flow that includes this item.
1445 * Pointer to error structure.
1448 * 0 on success, a negative errno value otherwise and rte_errno is set.
1451 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1452 const struct rte_flow_item *item,
1453 const struct rte_flow_attr *attr,
1454 struct rte_flow_error *error)
1456 struct mlx5_priv *priv = dev->data->dev_private;
1457 struct mlx5_dev_config *config = &priv->config;
1458 const struct rte_flow_item_meta *spec = item->spec;
1459 const struct rte_flow_item_meta *mask = item->mask;
1460 struct rte_flow_item_meta nic_mask = {
1467 return rte_flow_error_set(error, EINVAL,
1468 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1470 "data cannot be empty");
1471 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1472 if (!mlx5_flow_ext_mreg_supported(dev))
1473 return rte_flow_error_set(error, ENOTSUP,
1474 RTE_FLOW_ERROR_TYPE_ITEM, item,
1475 "extended metadata register"
1476 " isn't supported");
1477 reg = flow_dv_get_metadata_reg(dev, attr, error);
1481 return rte_flow_error_set(error, ENOTSUP,
1482 RTE_FLOW_ERROR_TYPE_ITEM, item,
1486 nic_mask.data = priv->sh->dv_meta_mask;
1487 } else if (attr->transfer) {
1488 return rte_flow_error_set(error, ENOTSUP,
1489 RTE_FLOW_ERROR_TYPE_ITEM, item,
1490 "extended metadata feature "
1491 "should be enabled when "
1492 "meta item is requested "
1493 "with e-switch mode ");
1496 mask = &rte_flow_item_meta_mask;
1498 return rte_flow_error_set(error, EINVAL,
1499 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1500 "mask cannot be zero");
1502 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1503 (const uint8_t *)&nic_mask,
1504 sizeof(struct rte_flow_item_meta),
1505 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1510 * Validate TAG item.
1513 * Pointer to the rte_eth_dev structure.
1515 * Item specification.
1517 * Attributes of flow that includes this item.
1519 * Pointer to error structure.
1522 * 0 on success, a negative errno value otherwise and rte_errno is set.
1525 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
1526 const struct rte_flow_item *item,
1527 const struct rte_flow_attr *attr __rte_unused,
1528 struct rte_flow_error *error)
1530 const struct rte_flow_item_tag *spec = item->spec;
1531 const struct rte_flow_item_tag *mask = item->mask;
1532 const struct rte_flow_item_tag nic_mask = {
1533 .data = RTE_BE32(UINT32_MAX),
1538 if (!mlx5_flow_ext_mreg_supported(dev))
1539 return rte_flow_error_set(error, ENOTSUP,
1540 RTE_FLOW_ERROR_TYPE_ITEM, item,
1541 "extensive metadata register"
1542 " isn't supported");
1544 return rte_flow_error_set(error, EINVAL,
1545 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1547 "data cannot be empty");
1549 mask = &rte_flow_item_tag_mask;
1551 return rte_flow_error_set(error, EINVAL,
1552 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1553 "mask cannot be zero");
1555 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1556 (const uint8_t *)&nic_mask,
1557 sizeof(struct rte_flow_item_tag),
1558 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1561 if (mask->index != 0xff)
1562 return rte_flow_error_set(error, EINVAL,
1563 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1564 "partial mask for tag index"
1565 " is not supported");
1566 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
1569 MLX5_ASSERT(ret != REG_NON);
1574 * Validate vport item.
1577 * Pointer to the rte_eth_dev structure.
1579 * Item specification.
1581 * Attributes of flow that includes this item.
1582 * @param[in] item_flags
1583 * Bit-fields that holds the items detected until now.
1585 * Pointer to error structure.
1588 * 0 on success, a negative errno value otherwise and rte_errno is set.
1591 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
1592 const struct rte_flow_item *item,
1593 const struct rte_flow_attr *attr,
1594 uint64_t item_flags,
1595 struct rte_flow_error *error)
1597 const struct rte_flow_item_port_id *spec = item->spec;
1598 const struct rte_flow_item_port_id *mask = item->mask;
1599 const struct rte_flow_item_port_id switch_mask = {
1602 struct mlx5_priv *esw_priv;
1603 struct mlx5_priv *dev_priv;
1606 if (!attr->transfer)
1607 return rte_flow_error_set(error, EINVAL,
1608 RTE_FLOW_ERROR_TYPE_ITEM,
1610 "match on port id is valid only"
1611 " when transfer flag is enabled");
1612 if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
1613 return rte_flow_error_set(error, ENOTSUP,
1614 RTE_FLOW_ERROR_TYPE_ITEM, item,
1615 "multiple source ports are not"
1618 mask = &switch_mask;
1619 if (mask->id != 0xffffffff)
1620 return rte_flow_error_set(error, ENOTSUP,
1621 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1623 "no support for partial mask on"
1625 ret = mlx5_flow_item_acceptable
1626 (item, (const uint8_t *)mask,
1627 (const uint8_t *)&rte_flow_item_port_id_mask,
1628 sizeof(struct rte_flow_item_port_id),
1629 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1634 esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
1636 return rte_flow_error_set(error, rte_errno,
1637 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1638 "failed to obtain E-Switch info for"
1640 dev_priv = mlx5_dev_to_eswitch_info(dev);
1642 return rte_flow_error_set(error, rte_errno,
1643 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1645 "failed to obtain E-Switch info");
1646 if (esw_priv->domain_id != dev_priv->domain_id)
1647 return rte_flow_error_set(error, EINVAL,
1648 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1649 "cannot match on a port from a"
1650 " different E-Switch");
1655 * Validate VLAN item.
1658 * Item specification.
1659 * @param[in] item_flags
1660 * Bit-fields that holds the items detected until now.
1662 * Ethernet device flow is being created on.
1664 * Pointer to error structure.
1667 * 0 on success, a negative errno value otherwise and rte_errno is set.
1670 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
1671 uint64_t item_flags,
1672 struct rte_eth_dev *dev,
1673 struct rte_flow_error *error)
1675 const struct rte_flow_item_vlan *mask = item->mask;
1676 const struct rte_flow_item_vlan nic_mask = {
1677 .tci = RTE_BE16(UINT16_MAX),
1678 .inner_type = RTE_BE16(UINT16_MAX),
1680 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1682 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1683 MLX5_FLOW_LAYER_INNER_L4) :
1684 (MLX5_FLOW_LAYER_OUTER_L3 |
1685 MLX5_FLOW_LAYER_OUTER_L4);
1686 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1687 MLX5_FLOW_LAYER_OUTER_VLAN;
1689 if (item_flags & vlanm)
1690 return rte_flow_error_set(error, EINVAL,
1691 RTE_FLOW_ERROR_TYPE_ITEM, item,
1692 "multiple VLAN layers not supported");
1693 else if ((item_flags & l34m) != 0)
1694 return rte_flow_error_set(error, EINVAL,
1695 RTE_FLOW_ERROR_TYPE_ITEM, item,
1696 "VLAN cannot follow L3/L4 layer");
1698 mask = &rte_flow_item_vlan_mask;
1699 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1700 (const uint8_t *)&nic_mask,
1701 sizeof(struct rte_flow_item_vlan),
1702 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1705 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1706 struct mlx5_priv *priv = dev->data->dev_private;
1708 if (priv->vmwa_context) {
1710 * Non-NULL context means we have a virtual machine
1711 * and SR-IOV enabled, we have to create VLAN interface
1712 * to make hypervisor to setup E-Switch vport
1713 * context correctly. We avoid creating the multiple
1714 * VLAN interfaces, so we cannot support VLAN tag mask.
1716 return rte_flow_error_set(error, EINVAL,
1717 RTE_FLOW_ERROR_TYPE_ITEM,
1719 "VLAN tag mask is not"
1720 " supported in virtual"
1728 * GTP flags are contained in 1 byte of the format:
1729 * -------------------------------------------
1730 * | bit | 0 - 2 | 3 | 4 | 5 | 6 | 7 |
1731 * |-----------------------------------------|
1732 * | value | Version | PT | Res | E | S | PN |
1733 * -------------------------------------------
1735 * Matching is supported only for GTP flags E, S, PN.
1737 #define MLX5_GTP_FLAGS_MASK 0x07
1740 * Validate GTP item.
1743 * Pointer to the rte_eth_dev structure.
1745 * Item specification.
1746 * @param[in] item_flags
1747 * Bit-fields that holds the items detected until now.
1749 * Pointer to error structure.
1752 * 0 on success, a negative errno value otherwise and rte_errno is set.
1755 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1756 const struct rte_flow_item *item,
1757 uint64_t item_flags,
1758 struct rte_flow_error *error)
1760 struct mlx5_priv *priv = dev->data->dev_private;
1761 const struct rte_flow_item_gtp *spec = item->spec;
1762 const struct rte_flow_item_gtp *mask = item->mask;
1763 const struct rte_flow_item_gtp nic_mask = {
1764 .v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
1766 .teid = RTE_BE32(0xffffffff),
1769 if (!priv->config.hca_attr.tunnel_stateless_gtp)
1770 return rte_flow_error_set(error, ENOTSUP,
1771 RTE_FLOW_ERROR_TYPE_ITEM, item,
1772 "GTP support is not enabled");
1773 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1774 return rte_flow_error_set(error, ENOTSUP,
1775 RTE_FLOW_ERROR_TYPE_ITEM, item,
1776 "multiple tunnel layers not"
1778 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1779 return rte_flow_error_set(error, EINVAL,
1780 RTE_FLOW_ERROR_TYPE_ITEM, item,
1781 "no outer UDP layer found");
1783 mask = &rte_flow_item_gtp_mask;
1784 if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
1785 return rte_flow_error_set(error, ENOTSUP,
1786 RTE_FLOW_ERROR_TYPE_ITEM, item,
1787 "Match is supported for GTP"
1789 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1790 (const uint8_t *)&nic_mask,
1791 sizeof(struct rte_flow_item_gtp),
1792 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1796 * Validate IPV4 item.
1797 * Use existing validation function mlx5_flow_validate_item_ipv4(), and
1798 * add specific validation of fragment_offset field,
1801 * Item specification.
1802 * @param[in] item_flags
1803 * Bit-fields that holds the items detected until now.
1805 * Pointer to error structure.
1808 * 0 on success, a negative errno value otherwise and rte_errno is set.
1811 flow_dv_validate_item_ipv4(const struct rte_flow_item *item,
1812 uint64_t item_flags,
1814 uint16_t ether_type,
1815 struct rte_flow_error *error)
1818 const struct rte_flow_item_ipv4 *spec = item->spec;
1819 const struct rte_flow_item_ipv4 *last = item->last;
1820 const struct rte_flow_item_ipv4 *mask = item->mask;
1821 rte_be16_t fragment_offset_spec = 0;
1822 rte_be16_t fragment_offset_last = 0;
1823 const struct rte_flow_item_ipv4 nic_ipv4_mask = {
1825 .src_addr = RTE_BE32(0xffffffff),
1826 .dst_addr = RTE_BE32(0xffffffff),
1827 .type_of_service = 0xff,
1828 .fragment_offset = RTE_BE16(0xffff),
1829 .next_proto_id = 0xff,
1830 .time_to_live = 0xff,
1834 ret = mlx5_flow_validate_item_ipv4(item, item_flags, last_item,
1835 ether_type, &nic_ipv4_mask,
1836 MLX5_ITEM_RANGE_ACCEPTED, error);
1840 fragment_offset_spec = spec->hdr.fragment_offset &
1841 mask->hdr.fragment_offset;
1842 if (!fragment_offset_spec)
1845 * spec and mask are valid, enforce using full mask to make sure the
1846 * complete value is used correctly.
1848 if ((mask->hdr.fragment_offset & RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1849 != RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1850 return rte_flow_error_set(error, EINVAL,
1851 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1852 item, "must use full mask for"
1853 " fragment_offset");
1855 * Match on fragment_offset 0x2000 means MF is 1 and frag-offset is 0,
1856 * indicating this is 1st fragment of fragmented packet.
1857 * This is not yet supported in MLX5, return appropriate error message.
1859 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG))
1860 return rte_flow_error_set(error, ENOTSUP,
1861 RTE_FLOW_ERROR_TYPE_ITEM, item,
1862 "match on first fragment not "
1864 if (fragment_offset_spec && !last)
1865 return rte_flow_error_set(error, ENOTSUP,
1866 RTE_FLOW_ERROR_TYPE_ITEM, item,
1867 "specified value not supported");
1868 /* spec and last are valid, validate the specified range. */
1869 fragment_offset_last = last->hdr.fragment_offset &
1870 mask->hdr.fragment_offset;
1872 * Match on fragment_offset spec 0x2001 and last 0x3fff
1873 * means MF is 1 and frag-offset is > 0.
1874 * This packet is fragment 2nd and onward, excluding last.
1875 * This is not yet supported in MLX5, return appropriate
1878 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG + 1) &&
1879 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1880 return rte_flow_error_set(error, ENOTSUP,
1881 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1882 last, "match on following "
1883 "fragments not supported");
1885 * Match on fragment_offset spec 0x0001 and last 0x1fff
1886 * means MF is 0 and frag-offset is > 0.
1887 * This packet is last fragment of fragmented packet.
1888 * This is not yet supported in MLX5, return appropriate
1891 if (fragment_offset_spec == RTE_BE16(1) &&
1892 fragment_offset_last == RTE_BE16(RTE_IPV4_HDR_OFFSET_MASK))
1893 return rte_flow_error_set(error, ENOTSUP,
1894 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1895 last, "match on last "
1896 "fragment not supported");
1898 * Match on fragment_offset spec 0x0001 and last 0x3fff
1899 * means MF and/or frag-offset is not 0.
1900 * This is a fragmented packet.
1901 * Other range values are invalid and rejected.
1903 if (!(fragment_offset_spec == RTE_BE16(1) &&
1904 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK)))
1905 return rte_flow_error_set(error, ENOTSUP,
1906 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
1907 "specified range not supported");
1912 * Validate IPV6 fragment extension item.
1915 * Item specification.
1916 * @param[in] item_flags
1917 * Bit-fields that holds the items detected until now.
1919 * Pointer to error structure.
1922 * 0 on success, a negative errno value otherwise and rte_errno is set.
1925 flow_dv_validate_item_ipv6_frag_ext(const struct rte_flow_item *item,
1926 uint64_t item_flags,
1927 struct rte_flow_error *error)
1929 const struct rte_flow_item_ipv6_frag_ext *spec = item->spec;
1930 const struct rte_flow_item_ipv6_frag_ext *last = item->last;
1931 const struct rte_flow_item_ipv6_frag_ext *mask = item->mask;
1932 rte_be16_t frag_data_spec = 0;
1933 rte_be16_t frag_data_last = 0;
1934 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1935 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1936 MLX5_FLOW_LAYER_OUTER_L4;
1938 struct rte_flow_item_ipv6_frag_ext nic_mask = {
1940 .next_header = 0xff,
1941 .frag_data = RTE_BE16(0xffff),
1945 if (item_flags & l4m)
1946 return rte_flow_error_set(error, EINVAL,
1947 RTE_FLOW_ERROR_TYPE_ITEM, item,
1948 "ipv6 fragment extension item cannot "
1950 if ((tunnel && !(item_flags & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
1951 (!tunnel && !(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
1952 return rte_flow_error_set(error, EINVAL,
1953 RTE_FLOW_ERROR_TYPE_ITEM, item,
1954 "ipv6 fragment extension item must "
1955 "follow ipv6 item");
1957 frag_data_spec = spec->hdr.frag_data & mask->hdr.frag_data;
1958 if (!frag_data_spec)
1961 * spec and mask are valid, enforce using full mask to make sure the
1962 * complete value is used correctly.
1964 if ((mask->hdr.frag_data & RTE_BE16(RTE_IPV6_FRAG_USED_MASK)) !=
1965 RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
1966 return rte_flow_error_set(error, EINVAL,
1967 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1968 item, "must use full mask for"
1971 * Match on frag_data 0x00001 means M is 1 and frag-offset is 0.
1972 * This is 1st fragment of fragmented packet.
1974 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_MF_MASK))
1975 return rte_flow_error_set(error, ENOTSUP,
1976 RTE_FLOW_ERROR_TYPE_ITEM, item,
1977 "match on first fragment not "
1979 if (frag_data_spec && !last)
1980 return rte_flow_error_set(error, EINVAL,
1981 RTE_FLOW_ERROR_TYPE_ITEM, item,
1982 "specified value not supported");
1983 ret = mlx5_flow_item_acceptable
1984 (item, (const uint8_t *)mask,
1985 (const uint8_t *)&nic_mask,
1986 sizeof(struct rte_flow_item_ipv6_frag_ext),
1987 MLX5_ITEM_RANGE_ACCEPTED, error);
1990 /* spec and last are valid, validate the specified range. */
1991 frag_data_last = last->hdr.frag_data & mask->hdr.frag_data;
1993 * Match on frag_data spec 0x0009 and last 0xfff9
1994 * means M is 1 and frag-offset is > 0.
1995 * This packet is fragment 2nd and onward, excluding last.
1996 * This is not yet supported in MLX5, return appropriate
1999 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN |
2000 RTE_IPV6_EHDR_MF_MASK) &&
2001 frag_data_last == RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
2002 return rte_flow_error_set(error, ENOTSUP,
2003 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2004 last, "match on following "
2005 "fragments not supported");
2007 * Match on frag_data spec 0x0008 and last 0xfff8
2008 * means M is 0 and frag-offset is > 0.
2009 * This packet is last fragment of fragmented packet.
2010 * This is not yet supported in MLX5, return appropriate
2013 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN) &&
2014 frag_data_last == RTE_BE16(RTE_IPV6_EHDR_FO_MASK))
2015 return rte_flow_error_set(error, ENOTSUP,
2016 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2017 last, "match on last "
2018 "fragment not supported");
2019 /* Other range values are invalid and rejected. */
2020 return rte_flow_error_set(error, EINVAL,
2021 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
2022 "specified range not supported");
2026 * Validate the pop VLAN action.
2029 * Pointer to the rte_eth_dev structure.
2030 * @param[in] action_flags
2031 * Holds the actions detected until now.
2033 * Pointer to the pop vlan action.
2034 * @param[in] item_flags
2035 * The items found in this flow rule.
2037 * Pointer to flow attributes.
2039 * Pointer to error structure.
2042 * 0 on success, a negative errno value otherwise and rte_errno is set.
2045 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
2046 uint64_t action_flags,
2047 const struct rte_flow_action *action,
2048 uint64_t item_flags,
2049 const struct rte_flow_attr *attr,
2050 struct rte_flow_error *error)
2052 const struct mlx5_priv *priv = dev->data->dev_private;
2056 if (!priv->sh->pop_vlan_action)
2057 return rte_flow_error_set(error, ENOTSUP,
2058 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2060 "pop vlan action is not supported");
2062 return rte_flow_error_set(error, ENOTSUP,
2063 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2065 "pop vlan action not supported for "
2067 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
2068 return rte_flow_error_set(error, ENOTSUP,
2069 RTE_FLOW_ERROR_TYPE_ACTION, action,
2070 "no support for multiple VLAN "
2072 /* Pop VLAN with preceding Decap requires inner header with VLAN. */
2073 if ((action_flags & MLX5_FLOW_ACTION_DECAP) &&
2074 !(item_flags & MLX5_FLOW_LAYER_INNER_VLAN))
2075 return rte_flow_error_set(error, ENOTSUP,
2076 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2078 "cannot pop vlan after decap without "
2079 "match on inner vlan in the flow");
2080 /* Pop VLAN without preceding Decap requires outer header with VLAN. */
2081 if (!(action_flags & MLX5_FLOW_ACTION_DECAP) &&
2082 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2083 return rte_flow_error_set(error, ENOTSUP,
2084 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2086 "cannot pop vlan without a "
2087 "match on (outer) vlan in the flow");
2088 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2089 return rte_flow_error_set(error, EINVAL,
2090 RTE_FLOW_ERROR_TYPE_ACTION, action,
2091 "wrong action order, port_id should "
2092 "be after pop VLAN action");
2093 if (!attr->transfer && priv->representor)
2094 return rte_flow_error_set(error, ENOTSUP,
2095 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2096 "pop vlan action for VF representor "
2097 "not supported on NIC table");
2102 * Get VLAN default info from vlan match info.
2105 * the list of item specifications.
2107 * pointer VLAN info to fill to.
2110 * 0 on success, a negative errno value otherwise and rte_errno is set.
2113 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
2114 struct rte_vlan_hdr *vlan)
2116 const struct rte_flow_item_vlan nic_mask = {
2117 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
2118 MLX5DV_FLOW_VLAN_VID_MASK),
2119 .inner_type = RTE_BE16(0xffff),
2124 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2125 int type = items->type;
2127 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
2128 type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
2131 if (items->type != RTE_FLOW_ITEM_TYPE_END) {
2132 const struct rte_flow_item_vlan *vlan_m = items->mask;
2133 const struct rte_flow_item_vlan *vlan_v = items->spec;
2135 /* If VLAN item in pattern doesn't contain data, return here. */
2140 /* Only full match values are accepted */
2141 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
2142 MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
2143 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
2145 rte_be_to_cpu_16(vlan_v->tci &
2146 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
2148 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
2149 MLX5DV_FLOW_VLAN_VID_MASK_BE) {
2150 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
2152 rte_be_to_cpu_16(vlan_v->tci &
2153 MLX5DV_FLOW_VLAN_VID_MASK_BE);
2155 if (vlan_m->inner_type == nic_mask.inner_type)
2156 vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
2157 vlan_m->inner_type);
2162 * Validate the push VLAN action.
2165 * Pointer to the rte_eth_dev structure.
2166 * @param[in] action_flags
2167 * Holds the actions detected until now.
2168 * @param[in] item_flags
2169 * The items found in this flow rule.
2171 * Pointer to the action structure.
2173 * Pointer to flow attributes
2175 * Pointer to error structure.
2178 * 0 on success, a negative errno value otherwise and rte_errno is set.
2181 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
2182 uint64_t action_flags,
2183 const struct rte_flow_item_vlan *vlan_m,
2184 const struct rte_flow_action *action,
2185 const struct rte_flow_attr *attr,
2186 struct rte_flow_error *error)
2188 const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
2189 const struct mlx5_priv *priv = dev->data->dev_private;
2191 if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
2192 push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
2193 return rte_flow_error_set(error, EINVAL,
2194 RTE_FLOW_ERROR_TYPE_ACTION, action,
2195 "invalid vlan ethertype");
2196 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2197 return rte_flow_error_set(error, EINVAL,
2198 RTE_FLOW_ERROR_TYPE_ACTION, action,
2199 "wrong action order, port_id should "
2200 "be after push VLAN");
2201 if (!attr->transfer && priv->representor)
2202 return rte_flow_error_set(error, ENOTSUP,
2203 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2204 "push vlan action for VF representor "
2205 "not supported on NIC table");
2207 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
2208 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
2209 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
2210 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
2211 !(mlx5_flow_find_action
2212 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
2213 return rte_flow_error_set(error, EINVAL,
2214 RTE_FLOW_ERROR_TYPE_ACTION, action,
2215 "not full match mask on VLAN PCP and "
2216 "there is no of_set_vlan_pcp action, "
2217 "push VLAN action cannot figure out "
2220 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
2221 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
2222 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
2223 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
2224 !(mlx5_flow_find_action
2225 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
2226 return rte_flow_error_set(error, EINVAL,
2227 RTE_FLOW_ERROR_TYPE_ACTION, action,
2228 "not full match mask on VLAN VID and "
2229 "there is no of_set_vlan_vid action, "
2230 "push VLAN action cannot figure out "
2237 * Validate the set VLAN PCP.
2239 * @param[in] action_flags
2240 * Holds the actions detected until now.
2241 * @param[in] actions
2242 * Pointer to the list of actions remaining in the flow rule.
2244 * Pointer to error structure.
2247 * 0 on success, a negative errno value otherwise and rte_errno is set.
2250 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
2251 const struct rte_flow_action actions[],
2252 struct rte_flow_error *error)
2254 const struct rte_flow_action *action = actions;
2255 const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2257 if (conf->vlan_pcp > 7)
2258 return rte_flow_error_set(error, EINVAL,
2259 RTE_FLOW_ERROR_TYPE_ACTION, action,
2260 "VLAN PCP value is too big");
2261 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2262 return rte_flow_error_set(error, ENOTSUP,
2263 RTE_FLOW_ERROR_TYPE_ACTION, action,
2264 "set VLAN PCP action must follow "
2265 "the push VLAN action");
2266 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2267 return rte_flow_error_set(error, ENOTSUP,
2268 RTE_FLOW_ERROR_TYPE_ACTION, action,
2269 "Multiple VLAN PCP modification are "
2271 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2272 return rte_flow_error_set(error, EINVAL,
2273 RTE_FLOW_ERROR_TYPE_ACTION, action,
2274 "wrong action order, port_id should "
2275 "be after set VLAN PCP");
2280 * Validate the set VLAN VID.
2282 * @param[in] item_flags
2283 * Holds the items detected in this rule.
2284 * @param[in] action_flags
2285 * Holds the actions detected until now.
2286 * @param[in] actions
2287 * Pointer to the list of actions remaining in the flow rule.
2289 * Pointer to error structure.
2292 * 0 on success, a negative errno value otherwise and rte_errno is set.
2295 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
2296 uint64_t action_flags,
2297 const struct rte_flow_action actions[],
2298 struct rte_flow_error *error)
2300 const struct rte_flow_action *action = actions;
2301 const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
2303 if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
2304 return rte_flow_error_set(error, EINVAL,
2305 RTE_FLOW_ERROR_TYPE_ACTION, action,
2306 "VLAN VID value is too big");
2307 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2308 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2309 return rte_flow_error_set(error, ENOTSUP,
2310 RTE_FLOW_ERROR_TYPE_ACTION, action,
2311 "set VLAN VID action must follow push"
2312 " VLAN action or match on VLAN item");
2313 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
2314 return rte_flow_error_set(error, ENOTSUP,
2315 RTE_FLOW_ERROR_TYPE_ACTION, action,
2316 "Multiple VLAN VID modifications are "
2318 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2319 return rte_flow_error_set(error, EINVAL,
2320 RTE_FLOW_ERROR_TYPE_ACTION, action,
2321 "wrong action order, port_id should "
2322 "be after set VLAN VID");
2327 * Validate the FLAG action.
2330 * Pointer to the rte_eth_dev structure.
2331 * @param[in] action_flags
2332 * Holds the actions detected until now.
2334 * Pointer to flow attributes
2336 * Pointer to error structure.
2339 * 0 on success, a negative errno value otherwise and rte_errno is set.
2342 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2343 uint64_t action_flags,
2344 const struct rte_flow_attr *attr,
2345 struct rte_flow_error *error)
2347 struct mlx5_priv *priv = dev->data->dev_private;
2348 struct mlx5_dev_config *config = &priv->config;
2351 /* Fall back if no extended metadata register support. */
2352 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2353 return mlx5_flow_validate_action_flag(action_flags, attr,
2355 /* Extensive metadata mode requires registers. */
2356 if (!mlx5_flow_ext_mreg_supported(dev))
2357 return rte_flow_error_set(error, ENOTSUP,
2358 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2359 "no metadata registers "
2360 "to support flag action");
2361 if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2362 return rte_flow_error_set(error, ENOTSUP,
2363 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2364 "extended metadata register"
2365 " isn't available");
2366 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2369 MLX5_ASSERT(ret > 0);
2370 if (action_flags & MLX5_FLOW_ACTION_MARK)
2371 return rte_flow_error_set(error, EINVAL,
2372 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2373 "can't mark and flag in same flow");
2374 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2375 return rte_flow_error_set(error, EINVAL,
2376 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2378 " actions in same flow");
2383 * Validate MARK action.
2386 * Pointer to the rte_eth_dev structure.
2388 * Pointer to action.
2389 * @param[in] action_flags
2390 * Holds the actions detected until now.
2392 * Pointer to flow attributes
2394 * Pointer to error structure.
2397 * 0 on success, a negative errno value otherwise and rte_errno is set.
2400 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2401 const struct rte_flow_action *action,
2402 uint64_t action_flags,
2403 const struct rte_flow_attr *attr,
2404 struct rte_flow_error *error)
2406 struct mlx5_priv *priv = dev->data->dev_private;
2407 struct mlx5_dev_config *config = &priv->config;
2408 const struct rte_flow_action_mark *mark = action->conf;
2411 /* Fall back if no extended metadata register support. */
2412 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2413 return mlx5_flow_validate_action_mark(action, action_flags,
2415 /* Extensive metadata mode requires registers. */
2416 if (!mlx5_flow_ext_mreg_supported(dev))
2417 return rte_flow_error_set(error, ENOTSUP,
2418 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2419 "no metadata registers "
2420 "to support mark action");
2421 if (!priv->sh->dv_mark_mask)
2422 return rte_flow_error_set(error, ENOTSUP,
2423 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2424 "extended metadata register"
2425 " isn't available");
2426 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2429 MLX5_ASSERT(ret > 0);
2431 return rte_flow_error_set(error, EINVAL,
2432 RTE_FLOW_ERROR_TYPE_ACTION, action,
2433 "configuration cannot be null");
2434 if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2435 return rte_flow_error_set(error, EINVAL,
2436 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2438 "mark id exceeds the limit");
2439 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2440 return rte_flow_error_set(error, EINVAL,
2441 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2442 "can't flag and mark in same flow");
2443 if (action_flags & MLX5_FLOW_ACTION_MARK)
2444 return rte_flow_error_set(error, EINVAL,
2445 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2446 "can't have 2 mark actions in same"
2452 * Validate SET_META action.
2455 * Pointer to the rte_eth_dev structure.
2457 * Pointer to the action structure.
2458 * @param[in] action_flags
2459 * Holds the actions detected until now.
2461 * Pointer to flow attributes
2463 * Pointer to error structure.
2466 * 0 on success, a negative errno value otherwise and rte_errno is set.
2469 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2470 const struct rte_flow_action *action,
2471 uint64_t action_flags __rte_unused,
2472 const struct rte_flow_attr *attr,
2473 struct rte_flow_error *error)
2475 const struct rte_flow_action_set_meta *conf;
2476 uint32_t nic_mask = UINT32_MAX;
2479 if (!mlx5_flow_ext_mreg_supported(dev))
2480 return rte_flow_error_set(error, ENOTSUP,
2481 RTE_FLOW_ERROR_TYPE_ACTION, action,
2482 "extended metadata register"
2483 " isn't supported");
2484 reg = flow_dv_get_metadata_reg(dev, attr, error);
2487 if (reg != REG_A && reg != REG_B) {
2488 struct mlx5_priv *priv = dev->data->dev_private;
2490 nic_mask = priv->sh->dv_meta_mask;
2492 if (!(action->conf))
2493 return rte_flow_error_set(error, EINVAL,
2494 RTE_FLOW_ERROR_TYPE_ACTION, action,
2495 "configuration cannot be null");
2496 conf = (const struct rte_flow_action_set_meta *)action->conf;
2498 return rte_flow_error_set(error, EINVAL,
2499 RTE_FLOW_ERROR_TYPE_ACTION, action,
2500 "zero mask doesn't have any effect");
2501 if (conf->mask & ~nic_mask)
2502 return rte_flow_error_set(error, EINVAL,
2503 RTE_FLOW_ERROR_TYPE_ACTION, action,
2504 "meta data must be within reg C0");
2509 * Validate SET_TAG action.
2512 * Pointer to the rte_eth_dev structure.
2514 * Pointer to the action structure.
2515 * @param[in] action_flags
2516 * Holds the actions detected until now.
2518 * Pointer to flow attributes
2520 * Pointer to error structure.
2523 * 0 on success, a negative errno value otherwise and rte_errno is set.
2526 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2527 const struct rte_flow_action *action,
2528 uint64_t action_flags,
2529 const struct rte_flow_attr *attr,
2530 struct rte_flow_error *error)
2532 const struct rte_flow_action_set_tag *conf;
2533 const uint64_t terminal_action_flags =
2534 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2535 MLX5_FLOW_ACTION_RSS;
2538 if (!mlx5_flow_ext_mreg_supported(dev))
2539 return rte_flow_error_set(error, ENOTSUP,
2540 RTE_FLOW_ERROR_TYPE_ACTION, action,
2541 "extensive metadata register"
2542 " isn't supported");
2543 if (!(action->conf))
2544 return rte_flow_error_set(error, EINVAL,
2545 RTE_FLOW_ERROR_TYPE_ACTION, action,
2546 "configuration cannot be null");
2547 conf = (const struct rte_flow_action_set_tag *)action->conf;
2549 return rte_flow_error_set(error, EINVAL,
2550 RTE_FLOW_ERROR_TYPE_ACTION, action,
2551 "zero mask doesn't have any effect");
2552 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2555 if (!attr->transfer && attr->ingress &&
2556 (action_flags & terminal_action_flags))
2557 return rte_flow_error_set(error, EINVAL,
2558 RTE_FLOW_ERROR_TYPE_ACTION, action,
2559 "set_tag has no effect"
2560 " with terminal actions");
2565 * Validate count action.
2568 * Pointer to rte_eth_dev structure.
2570 * Pointer to error structure.
2573 * 0 on success, a negative errno value otherwise and rte_errno is set.
2576 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2577 struct rte_flow_error *error)
2579 struct mlx5_priv *priv = dev->data->dev_private;
2581 if (!priv->config.devx)
2583 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2587 return rte_flow_error_set
2589 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2591 "count action not supported");
2595 * Validate the L2 encap action.
2598 * Pointer to the rte_eth_dev structure.
2599 * @param[in] action_flags
2600 * Holds the actions detected until now.
2602 * Pointer to the action structure.
2604 * Pointer to flow attributes.
2606 * Pointer to error structure.
2609 * 0 on success, a negative errno value otherwise and rte_errno is set.
2612 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2613 uint64_t action_flags,
2614 const struct rte_flow_action *action,
2615 const struct rte_flow_attr *attr,
2616 struct rte_flow_error *error)
2618 const struct mlx5_priv *priv = dev->data->dev_private;
2620 if (!(action->conf))
2621 return rte_flow_error_set(error, EINVAL,
2622 RTE_FLOW_ERROR_TYPE_ACTION, action,
2623 "configuration cannot be null");
2624 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2625 return rte_flow_error_set(error, EINVAL,
2626 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2627 "can only have a single encap action "
2629 if (!attr->transfer && priv->representor)
2630 return rte_flow_error_set(error, ENOTSUP,
2631 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2632 "encap action for VF representor "
2633 "not supported on NIC table");
2638 * Validate a decap action.
2641 * Pointer to the rte_eth_dev structure.
2642 * @param[in] action_flags
2643 * Holds the actions detected until now.
2645 * Pointer to flow attributes
2647 * Pointer to error structure.
2650 * 0 on success, a negative errno value otherwise and rte_errno is set.
2653 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2654 uint64_t action_flags,
2655 const struct rte_flow_attr *attr,
2656 struct rte_flow_error *error)
2658 const struct mlx5_priv *priv = dev->data->dev_private;
2660 if (priv->config.hca_attr.scatter_fcs_w_decap_disable &&
2661 !priv->config.decap_en)
2662 return rte_flow_error_set(error, ENOTSUP,
2663 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2664 "decap is not enabled");
2665 if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2666 return rte_flow_error_set(error, ENOTSUP,
2667 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2669 MLX5_FLOW_ACTION_DECAP ? "can only "
2670 "have a single decap action" : "decap "
2671 "after encap is not supported");
2672 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2673 return rte_flow_error_set(error, EINVAL,
2674 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2675 "can't have decap action after"
2678 return rte_flow_error_set(error, ENOTSUP,
2679 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2681 "decap action not supported for "
2683 if (!attr->transfer && priv->representor)
2684 return rte_flow_error_set(error, ENOTSUP,
2685 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2686 "decap action for VF representor "
2687 "not supported on NIC table");
2691 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2694 * Validate the raw encap and decap actions.
2697 * Pointer to the rte_eth_dev structure.
2699 * Pointer to the decap action.
2701 * Pointer to the encap action.
2703 * Pointer to flow attributes
2704 * @param[in/out] action_flags
2705 * Holds the actions detected until now.
2706 * @param[out] actions_n
2707 * pointer to the number of actions counter.
2709 * Pointer to error structure.
2712 * 0 on success, a negative errno value otherwise and rte_errno is set.
2715 flow_dv_validate_action_raw_encap_decap
2716 (struct rte_eth_dev *dev,
2717 const struct rte_flow_action_raw_decap *decap,
2718 const struct rte_flow_action_raw_encap *encap,
2719 const struct rte_flow_attr *attr, uint64_t *action_flags,
2720 int *actions_n, struct rte_flow_error *error)
2722 const struct mlx5_priv *priv = dev->data->dev_private;
2725 if (encap && (!encap->size || !encap->data))
2726 return rte_flow_error_set(error, EINVAL,
2727 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2728 "raw encap data cannot be empty");
2729 if (decap && encap) {
2730 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2731 encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2734 else if (encap->size <=
2735 MLX5_ENCAPSULATION_DECISION_SIZE &&
2737 MLX5_ENCAPSULATION_DECISION_SIZE)
2740 else if (encap->size >
2741 MLX5_ENCAPSULATION_DECISION_SIZE &&
2743 MLX5_ENCAPSULATION_DECISION_SIZE)
2744 /* 2 L2 actions: encap and decap. */
2747 return rte_flow_error_set(error,
2749 RTE_FLOW_ERROR_TYPE_ACTION,
2750 NULL, "unsupported too small "
2751 "raw decap and too small raw "
2752 "encap combination");
2755 ret = flow_dv_validate_action_decap(dev, *action_flags, attr,
2759 *action_flags |= MLX5_FLOW_ACTION_DECAP;
2763 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2764 return rte_flow_error_set(error, ENOTSUP,
2765 RTE_FLOW_ERROR_TYPE_ACTION,
2767 "small raw encap size");
2768 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2769 return rte_flow_error_set(error, EINVAL,
2770 RTE_FLOW_ERROR_TYPE_ACTION,
2772 "more than one encap action");
2773 if (!attr->transfer && priv->representor)
2774 return rte_flow_error_set
2776 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2777 "encap action for VF representor "
2778 "not supported on NIC table");
2779 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
2786 * Match encap_decap resource.
2789 * Pointer to exist resource entry object.
2791 * Pointer to new encap_decap resource.
2794 * 0 on matching, -1 otherwise.
2797 flow_dv_encap_decap_resource_match(struct mlx5_hlist_entry *entry, void *ctx)
2799 struct mlx5_flow_dv_encap_decap_resource *resource;
2800 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2802 resource = (struct mlx5_flow_dv_encap_decap_resource *)ctx;
2803 cache_resource = container_of(entry,
2804 struct mlx5_flow_dv_encap_decap_resource,
2806 if (resource->entry.key == cache_resource->entry.key &&
2807 resource->reformat_type == cache_resource->reformat_type &&
2808 resource->ft_type == cache_resource->ft_type &&
2809 resource->flags == cache_resource->flags &&
2810 resource->size == cache_resource->size &&
2811 !memcmp((const void *)resource->buf,
2812 (const void *)cache_resource->buf,
2819 * Find existing encap/decap resource or create and register a new one.
2821 * @param[in, out] dev
2822 * Pointer to rte_eth_dev structure.
2823 * @param[in, out] resource
2824 * Pointer to encap/decap resource.
2825 * @parm[in, out] dev_flow
2826 * Pointer to the dev_flow.
2828 * pointer to error structure.
2831 * 0 on success otherwise -errno and errno is set.
2834 flow_dv_encap_decap_resource_register
2835 (struct rte_eth_dev *dev,
2836 struct mlx5_flow_dv_encap_decap_resource *resource,
2837 struct mlx5_flow *dev_flow,
2838 struct rte_flow_error *error)
2840 struct mlx5_priv *priv = dev->data->dev_private;
2841 struct mlx5_dev_ctx_shared *sh = priv->sh;
2842 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2843 struct mlx5dv_dr_domain *domain;
2844 struct mlx5_hlist_entry *entry;
2845 union mlx5_flow_encap_decap_key encap_decap_key = {
2847 .ft_type = resource->ft_type,
2848 .refmt_type = resource->reformat_type,
2849 .buf_size = resource->size,
2850 .table_level = !!dev_flow->dv.group,
2856 resource->flags = dev_flow->dv.group ? 0 : 1;
2857 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2858 domain = sh->fdb_domain;
2859 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2860 domain = sh->rx_domain;
2862 domain = sh->tx_domain;
2863 encap_decap_key.cksum = __rte_raw_cksum(resource->buf,
2865 resource->entry.key = encap_decap_key.v64;
2866 /* Lookup a matching resource from cache. */
2867 entry = mlx5_hlist_lookup_ex(sh->encaps_decaps, resource->entry.key,
2868 flow_dv_encap_decap_resource_match,
2871 cache_resource = container_of(entry,
2872 struct mlx5_flow_dv_encap_decap_resource, entry);
2873 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
2874 (void *)cache_resource,
2875 rte_atomic32_read(&cache_resource->refcnt));
2876 rte_atomic32_inc(&cache_resource->refcnt);
2877 dev_flow->handle->dvh.rix_encap_decap = cache_resource->idx;
2878 dev_flow->dv.encap_decap = cache_resource;
2881 /* Register new encap/decap resource. */
2882 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2883 &dev_flow->handle->dvh.rix_encap_decap);
2884 if (!cache_resource)
2885 return rte_flow_error_set(error, ENOMEM,
2886 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2887 "cannot allocate resource memory");
2888 *cache_resource = *resource;
2889 cache_resource->idx = dev_flow->handle->dvh.rix_encap_decap;
2890 ret = mlx5_flow_os_create_flow_action_packet_reformat
2891 (sh->ctx, domain, cache_resource,
2892 &cache_resource->action);
2894 mlx5_free(cache_resource);
2895 return rte_flow_error_set(error, ENOMEM,
2896 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2897 NULL, "cannot create action");
2899 rte_atomic32_init(&cache_resource->refcnt);
2900 rte_atomic32_inc(&cache_resource->refcnt);
2901 if (mlx5_hlist_insert_ex(sh->encaps_decaps, &cache_resource->entry,
2902 flow_dv_encap_decap_resource_match,
2903 (void *)cache_resource)) {
2904 claim_zero(mlx5_flow_os_destroy_flow_action
2905 (cache_resource->action));
2906 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2907 cache_resource->idx);
2908 return rte_flow_error_set(error, EEXIST,
2909 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2910 NULL, "action exist");
2912 dev_flow->dv.encap_decap = cache_resource;
2913 DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
2914 (void *)cache_resource,
2915 rte_atomic32_read(&cache_resource->refcnt));
2920 * Find existing table jump resource or create and register a new one.
2922 * @param[in, out] dev
2923 * Pointer to rte_eth_dev structure.
2924 * @param[in, out] tbl
2925 * Pointer to flow table resource.
2926 * @parm[in, out] dev_flow
2927 * Pointer to the dev_flow.
2929 * pointer to error structure.
2932 * 0 on success otherwise -errno and errno is set.
2935 flow_dv_jump_tbl_resource_register
2936 (struct rte_eth_dev *dev __rte_unused,
2937 struct mlx5_flow_tbl_resource *tbl,
2938 struct mlx5_flow *dev_flow,
2939 struct rte_flow_error *error)
2941 struct mlx5_flow_tbl_data_entry *tbl_data =
2942 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2946 cnt = rte_atomic32_read(&tbl_data->jump.refcnt);
2948 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
2949 (tbl->obj, &tbl_data->jump.action);
2951 return rte_flow_error_set(error, ENOMEM,
2952 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2953 NULL, "cannot create jump action");
2954 DRV_LOG(DEBUG, "new jump table resource %p: refcnt %d++",
2955 (void *)&tbl_data->jump, cnt);
2957 /* old jump should not make the table ref++. */
2958 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
2959 MLX5_ASSERT(tbl_data->jump.action);
2960 DRV_LOG(DEBUG, "existed jump table resource %p: refcnt %d++",
2961 (void *)&tbl_data->jump, cnt);
2963 rte_atomic32_inc(&tbl_data->jump.refcnt);
2964 dev_flow->handle->rix_jump = tbl_data->idx;
2965 dev_flow->dv.jump = &tbl_data->jump;
2970 * Find existing default miss resource or create and register a new one.
2972 * @param[in, out] dev
2973 * Pointer to rte_eth_dev structure.
2975 * pointer to error structure.
2978 * 0 on success otherwise -errno and errno is set.
2981 flow_dv_default_miss_resource_register(struct rte_eth_dev *dev,
2982 struct rte_flow_error *error)
2984 struct mlx5_priv *priv = dev->data->dev_private;
2985 struct mlx5_dev_ctx_shared *sh = priv->sh;
2986 struct mlx5_flow_default_miss_resource *cache_resource =
2988 int cnt = rte_atomic32_read(&cache_resource->refcnt);
2991 MLX5_ASSERT(cache_resource->action);
2992 cache_resource->action =
2993 mlx5_glue->dr_create_flow_action_default_miss();
2994 if (!cache_resource->action)
2995 return rte_flow_error_set(error, ENOMEM,
2996 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2997 "cannot create default miss action");
2998 DRV_LOG(DEBUG, "new default miss resource %p: refcnt %d++",
2999 (void *)cache_resource->action, cnt);
3001 rte_atomic32_inc(&cache_resource->refcnt);
3006 * Find existing table port ID resource or create and register a new one.
3008 * @param[in, out] dev
3009 * Pointer to rte_eth_dev structure.
3010 * @param[in, out] resource
3011 * Pointer to port ID action resource.
3012 * @parm[in, out] dev_flow
3013 * Pointer to the dev_flow.
3015 * pointer to error structure.
3018 * 0 on success otherwise -errno and errno is set.
3021 flow_dv_port_id_action_resource_register
3022 (struct rte_eth_dev *dev,
3023 struct mlx5_flow_dv_port_id_action_resource *resource,
3024 struct mlx5_flow *dev_flow,
3025 struct rte_flow_error *error)
3027 struct mlx5_priv *priv = dev->data->dev_private;
3028 struct mlx5_dev_ctx_shared *sh = priv->sh;
3029 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
3033 /* Lookup a matching resource from cache. */
3034 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
3035 idx, cache_resource, next) {
3036 if (resource->port_id == cache_resource->port_id) {
3037 DRV_LOG(DEBUG, "port id action resource resource %p: "
3039 (void *)cache_resource,
3040 rte_atomic32_read(&cache_resource->refcnt));
3041 rte_atomic32_inc(&cache_resource->refcnt);
3042 dev_flow->handle->rix_port_id_action = idx;
3043 dev_flow->dv.port_id_action = cache_resource;
3047 /* Register new port id action resource. */
3048 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
3049 &dev_flow->handle->rix_port_id_action);
3050 if (!cache_resource)
3051 return rte_flow_error_set(error, ENOMEM,
3052 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3053 "cannot allocate resource memory");
3054 *cache_resource = *resource;
3055 ret = mlx5_flow_os_create_flow_action_dest_port
3056 (priv->sh->fdb_domain, resource->port_id,
3057 &cache_resource->action);
3059 mlx5_free(cache_resource);
3060 return rte_flow_error_set(error, ENOMEM,
3061 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3062 NULL, "cannot create action");
3064 rte_atomic32_init(&cache_resource->refcnt);
3065 rte_atomic32_inc(&cache_resource->refcnt);
3066 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
3067 dev_flow->handle->rix_port_id_action, cache_resource,
3069 dev_flow->dv.port_id_action = cache_resource;
3070 DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
3071 (void *)cache_resource,
3072 rte_atomic32_read(&cache_resource->refcnt));
3077 * Find existing push vlan resource or create and register a new one.
3079 * @param [in, out] dev
3080 * Pointer to rte_eth_dev structure.
3081 * @param[in, out] resource
3082 * Pointer to port ID action resource.
3083 * @parm[in, out] dev_flow
3084 * Pointer to the dev_flow.
3086 * pointer to error structure.
3089 * 0 on success otherwise -errno and errno is set.
3092 flow_dv_push_vlan_action_resource_register
3093 (struct rte_eth_dev *dev,
3094 struct mlx5_flow_dv_push_vlan_action_resource *resource,
3095 struct mlx5_flow *dev_flow,
3096 struct rte_flow_error *error)
3098 struct mlx5_priv *priv = dev->data->dev_private;
3099 struct mlx5_dev_ctx_shared *sh = priv->sh;
3100 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
3101 struct mlx5dv_dr_domain *domain;
3105 /* Lookup a matching resource from cache. */
3106 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3107 sh->push_vlan_action_list, idx, cache_resource, next) {
3108 if (resource->vlan_tag == cache_resource->vlan_tag &&
3109 resource->ft_type == cache_resource->ft_type) {
3110 DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
3112 (void *)cache_resource,
3113 rte_atomic32_read(&cache_resource->refcnt));
3114 rte_atomic32_inc(&cache_resource->refcnt);
3115 dev_flow->handle->dvh.rix_push_vlan = idx;
3116 dev_flow->dv.push_vlan_res = cache_resource;
3120 /* Register new push_vlan action resource. */
3121 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3122 &dev_flow->handle->dvh.rix_push_vlan);
3123 if (!cache_resource)
3124 return rte_flow_error_set(error, ENOMEM,
3125 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3126 "cannot allocate resource memory");
3127 *cache_resource = *resource;
3128 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3129 domain = sh->fdb_domain;
3130 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
3131 domain = sh->rx_domain;
3133 domain = sh->tx_domain;
3134 ret = mlx5_flow_os_create_flow_action_push_vlan
3135 (domain, resource->vlan_tag,
3136 &cache_resource->action);
3138 mlx5_free(cache_resource);
3139 return rte_flow_error_set(error, ENOMEM,
3140 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3141 NULL, "cannot create action");
3143 rte_atomic32_init(&cache_resource->refcnt);
3144 rte_atomic32_inc(&cache_resource->refcnt);
3145 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3146 &sh->push_vlan_action_list,
3147 dev_flow->handle->dvh.rix_push_vlan,
3148 cache_resource, next);
3149 dev_flow->dv.push_vlan_res = cache_resource;
3150 DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
3151 (void *)cache_resource,
3152 rte_atomic32_read(&cache_resource->refcnt));
3156 * Get the size of specific rte_flow_item_type hdr size
3158 * @param[in] item_type
3159 * Tested rte_flow_item_type.
3162 * sizeof struct item_type, 0 if void or irrelevant.
3165 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
3169 switch (item_type) {
3170 case RTE_FLOW_ITEM_TYPE_ETH:
3171 retval = sizeof(struct rte_ether_hdr);
3173 case RTE_FLOW_ITEM_TYPE_VLAN:
3174 retval = sizeof(struct rte_vlan_hdr);
3176 case RTE_FLOW_ITEM_TYPE_IPV4:
3177 retval = sizeof(struct rte_ipv4_hdr);
3179 case RTE_FLOW_ITEM_TYPE_IPV6:
3180 retval = sizeof(struct rte_ipv6_hdr);
3182 case RTE_FLOW_ITEM_TYPE_UDP:
3183 retval = sizeof(struct rte_udp_hdr);
3185 case RTE_FLOW_ITEM_TYPE_TCP:
3186 retval = sizeof(struct rte_tcp_hdr);
3188 case RTE_FLOW_ITEM_TYPE_VXLAN:
3189 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3190 retval = sizeof(struct rte_vxlan_hdr);
3192 case RTE_FLOW_ITEM_TYPE_GRE:
3193 case RTE_FLOW_ITEM_TYPE_NVGRE:
3194 retval = sizeof(struct rte_gre_hdr);
3196 case RTE_FLOW_ITEM_TYPE_MPLS:
3197 retval = sizeof(struct rte_mpls_hdr);
3199 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
3207 #define MLX5_ENCAP_IPV4_VERSION 0x40
3208 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
3209 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
3210 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
3211 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
3212 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
3213 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
3216 * Convert the encap action data from list of rte_flow_item to raw buffer
3219 * Pointer to rte_flow_item objects list.
3221 * Pointer to the output buffer.
3223 * Pointer to the output buffer size.
3225 * Pointer to the error structure.
3228 * 0 on success, a negative errno value otherwise and rte_errno is set.
3231 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
3232 size_t *size, struct rte_flow_error *error)
3234 struct rte_ether_hdr *eth = NULL;
3235 struct rte_vlan_hdr *vlan = NULL;
3236 struct rte_ipv4_hdr *ipv4 = NULL;
3237 struct rte_ipv6_hdr *ipv6 = NULL;
3238 struct rte_udp_hdr *udp = NULL;
3239 struct rte_vxlan_hdr *vxlan = NULL;
3240 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
3241 struct rte_gre_hdr *gre = NULL;
3243 size_t temp_size = 0;
3246 return rte_flow_error_set(error, EINVAL,
3247 RTE_FLOW_ERROR_TYPE_ACTION,
3248 NULL, "invalid empty data");
3249 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3250 len = flow_dv_get_item_hdr_len(items->type);
3251 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
3252 return rte_flow_error_set(error, EINVAL,
3253 RTE_FLOW_ERROR_TYPE_ACTION,
3254 (void *)items->type,
3255 "items total size is too big"
3256 " for encap action");
3257 rte_memcpy((void *)&buf[temp_size], items->spec, len);
3258 switch (items->type) {
3259 case RTE_FLOW_ITEM_TYPE_ETH:
3260 eth = (struct rte_ether_hdr *)&buf[temp_size];
3262 case RTE_FLOW_ITEM_TYPE_VLAN:
3263 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
3265 return rte_flow_error_set(error, EINVAL,
3266 RTE_FLOW_ERROR_TYPE_ACTION,
3267 (void *)items->type,
3268 "eth header not found");
3269 if (!eth->ether_type)
3270 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
3272 case RTE_FLOW_ITEM_TYPE_IPV4:
3273 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
3275 return rte_flow_error_set(error, EINVAL,
3276 RTE_FLOW_ERROR_TYPE_ACTION,
3277 (void *)items->type,
3278 "neither eth nor vlan"
3280 if (vlan && !vlan->eth_proto)
3281 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3282 else if (eth && !eth->ether_type)
3283 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3284 if (!ipv4->version_ihl)
3285 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
3286 MLX5_ENCAP_IPV4_IHL_MIN;
3287 if (!ipv4->time_to_live)
3288 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
3290 case RTE_FLOW_ITEM_TYPE_IPV6:
3291 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
3293 return rte_flow_error_set(error, EINVAL,
3294 RTE_FLOW_ERROR_TYPE_ACTION,
3295 (void *)items->type,
3296 "neither eth nor vlan"
3298 if (vlan && !vlan->eth_proto)
3299 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3300 else if (eth && !eth->ether_type)
3301 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3302 if (!ipv6->vtc_flow)
3304 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
3305 if (!ipv6->hop_limits)
3306 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
3308 case RTE_FLOW_ITEM_TYPE_UDP:
3309 udp = (struct rte_udp_hdr *)&buf[temp_size];
3311 return rte_flow_error_set(error, EINVAL,
3312 RTE_FLOW_ERROR_TYPE_ACTION,
3313 (void *)items->type,
3314 "ip header not found");
3315 if (ipv4 && !ipv4->next_proto_id)
3316 ipv4->next_proto_id = IPPROTO_UDP;
3317 else if (ipv6 && !ipv6->proto)
3318 ipv6->proto = IPPROTO_UDP;
3320 case RTE_FLOW_ITEM_TYPE_VXLAN:
3321 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
3323 return rte_flow_error_set(error, EINVAL,
3324 RTE_FLOW_ERROR_TYPE_ACTION,
3325 (void *)items->type,
3326 "udp header not found");
3328 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
3329 if (!vxlan->vx_flags)
3331 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
3333 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3334 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
3336 return rte_flow_error_set(error, EINVAL,
3337 RTE_FLOW_ERROR_TYPE_ACTION,
3338 (void *)items->type,
3339 "udp header not found");
3340 if (!vxlan_gpe->proto)
3341 return rte_flow_error_set(error, EINVAL,
3342 RTE_FLOW_ERROR_TYPE_ACTION,
3343 (void *)items->type,
3344 "next protocol not found");
3347 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
3348 if (!vxlan_gpe->vx_flags)
3349 vxlan_gpe->vx_flags =
3350 MLX5_ENCAP_VXLAN_GPE_FLAGS;
3352 case RTE_FLOW_ITEM_TYPE_GRE:
3353 case RTE_FLOW_ITEM_TYPE_NVGRE:
3354 gre = (struct rte_gre_hdr *)&buf[temp_size];
3356 return rte_flow_error_set(error, EINVAL,
3357 RTE_FLOW_ERROR_TYPE_ACTION,
3358 (void *)items->type,
3359 "next protocol not found");
3361 return rte_flow_error_set(error, EINVAL,
3362 RTE_FLOW_ERROR_TYPE_ACTION,
3363 (void *)items->type,
3364 "ip header not found");
3365 if (ipv4 && !ipv4->next_proto_id)
3366 ipv4->next_proto_id = IPPROTO_GRE;
3367 else if (ipv6 && !ipv6->proto)
3368 ipv6->proto = IPPROTO_GRE;
3370 case RTE_FLOW_ITEM_TYPE_VOID:
3373 return rte_flow_error_set(error, EINVAL,
3374 RTE_FLOW_ERROR_TYPE_ACTION,
3375 (void *)items->type,
3376 "unsupported item type");
3386 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
3388 struct rte_ether_hdr *eth = NULL;
3389 struct rte_vlan_hdr *vlan = NULL;
3390 struct rte_ipv6_hdr *ipv6 = NULL;
3391 struct rte_udp_hdr *udp = NULL;
3395 eth = (struct rte_ether_hdr *)data;
3396 next_hdr = (char *)(eth + 1);
3397 proto = RTE_BE16(eth->ether_type);
3400 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
3401 vlan = (struct rte_vlan_hdr *)next_hdr;
3402 proto = RTE_BE16(vlan->eth_proto);
3403 next_hdr += sizeof(struct rte_vlan_hdr);
3406 /* HW calculates IPv4 csum. no need to proceed */
3407 if (proto == RTE_ETHER_TYPE_IPV4)
3410 /* non IPv4/IPv6 header. not supported */
3411 if (proto != RTE_ETHER_TYPE_IPV6) {
3412 return rte_flow_error_set(error, ENOTSUP,
3413 RTE_FLOW_ERROR_TYPE_ACTION,
3414 NULL, "Cannot offload non IPv4/IPv6");
3417 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3419 /* ignore non UDP */
3420 if (ipv6->proto != IPPROTO_UDP)
3423 udp = (struct rte_udp_hdr *)(ipv6 + 1);
3424 udp->dgram_cksum = 0;
3430 * Convert L2 encap action to DV specification.
3433 * Pointer to rte_eth_dev structure.
3435 * Pointer to action structure.
3436 * @param[in, out] dev_flow
3437 * Pointer to the mlx5_flow.
3438 * @param[in] transfer
3439 * Mark if the flow is E-Switch flow.
3441 * Pointer to the error structure.
3444 * 0 on success, a negative errno value otherwise and rte_errno is set.
3447 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3448 const struct rte_flow_action *action,
3449 struct mlx5_flow *dev_flow,
3451 struct rte_flow_error *error)
3453 const struct rte_flow_item *encap_data;
3454 const struct rte_flow_action_raw_encap *raw_encap_data;
3455 struct mlx5_flow_dv_encap_decap_resource res = {
3457 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3458 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3459 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3462 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3464 (const struct rte_flow_action_raw_encap *)action->conf;
3465 res.size = raw_encap_data->size;
3466 memcpy(res.buf, raw_encap_data->data, res.size);
3468 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3470 ((const struct rte_flow_action_vxlan_encap *)
3471 action->conf)->definition;
3474 ((const struct rte_flow_action_nvgre_encap *)
3475 action->conf)->definition;
3476 if (flow_dv_convert_encap_data(encap_data, res.buf,
3480 if (flow_dv_zero_encap_udp_csum(res.buf, error))
3482 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3483 return rte_flow_error_set(error, EINVAL,
3484 RTE_FLOW_ERROR_TYPE_ACTION,
3485 NULL, "can't create L2 encap action");
3490 * Convert L2 decap action to DV specification.
3493 * Pointer to rte_eth_dev structure.
3494 * @param[in, out] dev_flow
3495 * Pointer to the mlx5_flow.
3496 * @param[in] transfer
3497 * Mark if the flow is E-Switch flow.
3499 * Pointer to the error structure.
3502 * 0 on success, a negative errno value otherwise and rte_errno is set.
3505 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3506 struct mlx5_flow *dev_flow,
3508 struct rte_flow_error *error)
3510 struct mlx5_flow_dv_encap_decap_resource res = {
3513 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3514 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3515 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3518 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3519 return rte_flow_error_set(error, EINVAL,
3520 RTE_FLOW_ERROR_TYPE_ACTION,
3521 NULL, "can't create L2 decap action");
3526 * Convert raw decap/encap (L3 tunnel) action to DV specification.
3529 * Pointer to rte_eth_dev structure.
3531 * Pointer to action structure.
3532 * @param[in, out] dev_flow
3533 * Pointer to the mlx5_flow.
3535 * Pointer to the flow attributes.
3537 * Pointer to the error structure.
3540 * 0 on success, a negative errno value otherwise and rte_errno is set.
3543 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3544 const struct rte_flow_action *action,
3545 struct mlx5_flow *dev_flow,
3546 const struct rte_flow_attr *attr,
3547 struct rte_flow_error *error)
3549 const struct rte_flow_action_raw_encap *encap_data;
3550 struct mlx5_flow_dv_encap_decap_resource res;
3552 memset(&res, 0, sizeof(res));
3553 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3554 res.size = encap_data->size;
3555 memcpy(res.buf, encap_data->data, res.size);
3556 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3557 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3558 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3560 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3562 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3563 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3564 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3565 return rte_flow_error_set(error, EINVAL,
3566 RTE_FLOW_ERROR_TYPE_ACTION,
3567 NULL, "can't create encap action");
3572 * Create action push VLAN.
3575 * Pointer to rte_eth_dev structure.
3577 * Pointer to the flow attributes.
3579 * Pointer to the vlan to push to the Ethernet header.
3580 * @param[in, out] dev_flow
3581 * Pointer to the mlx5_flow.
3583 * Pointer to the error structure.
3586 * 0 on success, a negative errno value otherwise and rte_errno is set.
3589 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3590 const struct rte_flow_attr *attr,
3591 const struct rte_vlan_hdr *vlan,
3592 struct mlx5_flow *dev_flow,
3593 struct rte_flow_error *error)
3595 struct mlx5_flow_dv_push_vlan_action_resource res;
3597 memset(&res, 0, sizeof(res));
3599 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3602 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3604 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3605 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3606 return flow_dv_push_vlan_action_resource_register
3607 (dev, &res, dev_flow, error);
3610 static int fdb_mirror;
3613 * Validate the modify-header actions.
3615 * @param[in] action_flags
3616 * Holds the actions detected until now.
3618 * Pointer to the modify action.
3620 * Pointer to error structure.
3623 * 0 on success, a negative errno value otherwise and rte_errno is set.
3626 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3627 const struct rte_flow_action *action,
3628 struct rte_flow_error *error)
3630 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3631 return rte_flow_error_set(error, EINVAL,
3632 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3633 NULL, "action configuration not set");
3634 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3635 return rte_flow_error_set(error, EINVAL,
3636 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3637 "can't have encap action before"
3639 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3640 return rte_flow_error_set(error, EINVAL,
3641 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3642 "can't support sample action before"
3643 " modify action for E-Switch"
3649 * Validate the modify-header MAC address actions.
3651 * @param[in] action_flags
3652 * Holds the actions detected until now.
3654 * Pointer to the modify action.
3655 * @param[in] item_flags
3656 * Holds the items detected.
3658 * Pointer to error structure.
3661 * 0 on success, a negative errno value otherwise and rte_errno is set.
3664 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3665 const struct rte_flow_action *action,
3666 const uint64_t item_flags,
3667 struct rte_flow_error *error)
3671 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3673 if (!(item_flags & MLX5_FLOW_LAYER_L2))
3674 return rte_flow_error_set(error, EINVAL,
3675 RTE_FLOW_ERROR_TYPE_ACTION,
3677 "no L2 item in pattern");
3683 * Validate the modify-header IPv4 address actions.
3685 * @param[in] action_flags
3686 * Holds the actions detected until now.
3688 * Pointer to the modify action.
3689 * @param[in] item_flags
3690 * Holds the items detected.
3692 * Pointer to error structure.
3695 * 0 on success, a negative errno value otherwise and rte_errno is set.
3698 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3699 const struct rte_flow_action *action,
3700 const uint64_t item_flags,
3701 struct rte_flow_error *error)
3706 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3708 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3709 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3710 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3711 if (!(item_flags & layer))
3712 return rte_flow_error_set(error, EINVAL,
3713 RTE_FLOW_ERROR_TYPE_ACTION,
3715 "no ipv4 item in pattern");
3721 * Validate the modify-header IPv6 address actions.
3723 * @param[in] action_flags
3724 * Holds the actions detected until now.
3726 * Pointer to the modify action.
3727 * @param[in] item_flags
3728 * Holds the items detected.
3730 * Pointer to error structure.
3733 * 0 on success, a negative errno value otherwise and rte_errno is set.
3736 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3737 const struct rte_flow_action *action,
3738 const uint64_t item_flags,
3739 struct rte_flow_error *error)
3744 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3746 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3747 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3748 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3749 if (!(item_flags & layer))
3750 return rte_flow_error_set(error, EINVAL,
3751 RTE_FLOW_ERROR_TYPE_ACTION,
3753 "no ipv6 item in pattern");
3759 * Validate the modify-header TP actions.
3761 * @param[in] action_flags
3762 * Holds the actions detected until now.
3764 * Pointer to the modify action.
3765 * @param[in] item_flags
3766 * Holds the items detected.
3768 * Pointer to error structure.
3771 * 0 on success, a negative errno value otherwise and rte_errno is set.
3774 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3775 const struct rte_flow_action *action,
3776 const uint64_t item_flags,
3777 struct rte_flow_error *error)
3782 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3784 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3785 MLX5_FLOW_LAYER_INNER_L4 :
3786 MLX5_FLOW_LAYER_OUTER_L4;
3787 if (!(item_flags & layer))
3788 return rte_flow_error_set(error, EINVAL,
3789 RTE_FLOW_ERROR_TYPE_ACTION,
3790 NULL, "no transport layer "
3797 * Validate the modify-header actions of increment/decrement
3798 * TCP Sequence-number.
3800 * @param[in] action_flags
3801 * Holds the actions detected until now.
3803 * Pointer to the modify action.
3804 * @param[in] item_flags
3805 * Holds the items detected.
3807 * Pointer to error structure.
3810 * 0 on success, a negative errno value otherwise and rte_errno is set.
3813 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3814 const struct rte_flow_action *action,
3815 const uint64_t item_flags,
3816 struct rte_flow_error *error)
3821 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3823 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3824 MLX5_FLOW_LAYER_INNER_L4_TCP :
3825 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3826 if (!(item_flags & layer))
3827 return rte_flow_error_set(error, EINVAL,
3828 RTE_FLOW_ERROR_TYPE_ACTION,
3829 NULL, "no TCP item in"
3831 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3832 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3833 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3834 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3835 return rte_flow_error_set(error, EINVAL,
3836 RTE_FLOW_ERROR_TYPE_ACTION,
3838 "cannot decrease and increase"
3839 " TCP sequence number"
3840 " at the same time");
3846 * Validate the modify-header actions of increment/decrement
3847 * TCP Acknowledgment number.
3849 * @param[in] action_flags
3850 * Holds the actions detected until now.
3852 * Pointer to the modify action.
3853 * @param[in] item_flags
3854 * Holds the items detected.
3856 * Pointer to error structure.
3859 * 0 on success, a negative errno value otherwise and rte_errno is set.
3862 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3863 const struct rte_flow_action *action,
3864 const uint64_t item_flags,
3865 struct rte_flow_error *error)
3870 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3872 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3873 MLX5_FLOW_LAYER_INNER_L4_TCP :
3874 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3875 if (!(item_flags & layer))
3876 return rte_flow_error_set(error, EINVAL,
3877 RTE_FLOW_ERROR_TYPE_ACTION,
3878 NULL, "no TCP item in"
3880 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3881 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3882 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3883 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3884 return rte_flow_error_set(error, EINVAL,
3885 RTE_FLOW_ERROR_TYPE_ACTION,
3887 "cannot decrease and increase"
3888 " TCP acknowledgment number"
3889 " at the same time");
3895 * Validate the modify-header TTL actions.
3897 * @param[in] action_flags
3898 * Holds the actions detected until now.
3900 * Pointer to the modify action.
3901 * @param[in] item_flags
3902 * Holds the items detected.
3904 * Pointer to error structure.
3907 * 0 on success, a negative errno value otherwise and rte_errno is set.
3910 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3911 const struct rte_flow_action *action,
3912 const uint64_t item_flags,
3913 struct rte_flow_error *error)
3918 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3920 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3921 MLX5_FLOW_LAYER_INNER_L3 :
3922 MLX5_FLOW_LAYER_OUTER_L3;
3923 if (!(item_flags & layer))
3924 return rte_flow_error_set(error, EINVAL,
3925 RTE_FLOW_ERROR_TYPE_ACTION,
3927 "no IP protocol in pattern");
3933 * Validate jump action.
3936 * Pointer to the jump action.
3937 * @param[in] action_flags
3938 * Holds the actions detected until now.
3939 * @param[in] attributes
3940 * Pointer to flow attributes
3941 * @param[in] external
3942 * Action belongs to flow rule created by request external to PMD.
3944 * Pointer to error structure.
3947 * 0 on success, a negative errno value otherwise and rte_errno is set.
3950 flow_dv_validate_action_jump(const struct rte_flow_action *action,
3951 uint64_t action_flags,
3952 const struct rte_flow_attr *attributes,
3953 bool external, struct rte_flow_error *error)
3955 uint32_t target_group, table;
3958 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3959 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3960 return rte_flow_error_set(error, EINVAL,
3961 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3962 "can't have 2 fate actions in"
3964 if (action_flags & MLX5_FLOW_ACTION_METER)
3965 return rte_flow_error_set(error, ENOTSUP,
3966 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3967 "jump with meter not support");
3968 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3969 return rte_flow_error_set(error, EINVAL,
3970 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3971 "E-Switch mirroring can't support"
3972 " Sample action and jump action in"
3975 return rte_flow_error_set(error, EINVAL,
3976 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3977 NULL, "action configuration not set");
3979 ((const struct rte_flow_action_jump *)action->conf)->group;
3980 ret = mlx5_flow_group_to_table(attributes, external, target_group,
3981 true, &table, error);
3984 if (attributes->group == target_group)
3985 return rte_flow_error_set(error, EINVAL,
3986 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3987 "target group must be other than"
3988 " the current flow group");
3993 * Validate the port_id action.
3996 * Pointer to rte_eth_dev structure.
3997 * @param[in] action_flags
3998 * Bit-fields that holds the actions detected until now.
4000 * Port_id RTE action structure.
4002 * Attributes of flow that includes this action.
4004 * Pointer to error structure.
4007 * 0 on success, a negative errno value otherwise and rte_errno is set.
4010 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
4011 uint64_t action_flags,
4012 const struct rte_flow_action *action,
4013 const struct rte_flow_attr *attr,
4014 struct rte_flow_error *error)
4016 const struct rte_flow_action_port_id *port_id;
4017 struct mlx5_priv *act_priv;
4018 struct mlx5_priv *dev_priv;
4021 if (!attr->transfer)
4022 return rte_flow_error_set(error, ENOTSUP,
4023 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4025 "port id action is valid in transfer"
4027 if (!action || !action->conf)
4028 return rte_flow_error_set(error, ENOTSUP,
4029 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4031 "port id action parameters must be"
4033 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
4034 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
4035 return rte_flow_error_set(error, EINVAL,
4036 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4037 "can have only one fate actions in"
4039 dev_priv = mlx5_dev_to_eswitch_info(dev);
4041 return rte_flow_error_set(error, rte_errno,
4042 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4044 "failed to obtain E-Switch info");
4045 port_id = action->conf;
4046 port = port_id->original ? dev->data->port_id : port_id->id;
4047 act_priv = mlx5_port_to_eswitch_info(port, false);
4049 return rte_flow_error_set
4051 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
4052 "failed to obtain E-Switch port id for port");
4053 if (act_priv->domain_id != dev_priv->domain_id)
4054 return rte_flow_error_set
4056 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4057 "port does not belong to"
4058 " E-Switch being configured");
4063 * Get the maximum number of modify header actions.
4066 * Pointer to rte_eth_dev structure.
4068 * Flags bits to check if root level.
4071 * Max number of modify header actions device can support.
4073 static inline unsigned int
4074 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
4078 * There's no way to directly query the max capacity from FW.
4079 * The maximal value on root table should be assumed to be supported.
4081 if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
4082 return MLX5_MAX_MODIFY_NUM;
4084 return MLX5_ROOT_TBL_MODIFY_NUM;
4088 * Validate the meter action.
4091 * Pointer to rte_eth_dev structure.
4092 * @param[in] action_flags
4093 * Bit-fields that holds the actions detected until now.
4095 * Pointer to the meter action.
4097 * Attributes of flow that includes this action.
4099 * Pointer to error structure.
4102 * 0 on success, a negative errno value otherwise and rte_ernno is set.
4105 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
4106 uint64_t action_flags,
4107 const struct rte_flow_action *action,
4108 const struct rte_flow_attr *attr,
4109 struct rte_flow_error *error)
4111 struct mlx5_priv *priv = dev->data->dev_private;
4112 const struct rte_flow_action_meter *am = action->conf;
4113 struct mlx5_flow_meter *fm;
4116 return rte_flow_error_set(error, EINVAL,
4117 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4118 "meter action conf is NULL");
4120 if (action_flags & MLX5_FLOW_ACTION_METER)
4121 return rte_flow_error_set(error, ENOTSUP,
4122 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4123 "meter chaining not support");
4124 if (action_flags & MLX5_FLOW_ACTION_JUMP)
4125 return rte_flow_error_set(error, ENOTSUP,
4126 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4127 "meter with jump not support");
4129 return rte_flow_error_set(error, ENOTSUP,
4130 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4132 "meter action not supported");
4133 fm = mlx5_flow_meter_find(priv, am->mtr_id);
4135 return rte_flow_error_set(error, EINVAL,
4136 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4138 if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
4139 (!fm->ingress && !attr->ingress && attr->egress) ||
4140 (!fm->egress && !attr->egress && attr->ingress))))
4141 return rte_flow_error_set(error, EINVAL,
4142 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4143 "Flow attributes are either invalid "
4144 "or have a conflict with current "
4145 "meter attributes");
4150 * Validate the age action.
4152 * @param[in] action_flags
4153 * Holds the actions detected until now.
4155 * Pointer to the age action.
4157 * Pointer to the Ethernet device structure.
4159 * Pointer to error structure.
4162 * 0 on success, a negative errno value otherwise and rte_errno is set.
4165 flow_dv_validate_action_age(uint64_t action_flags,
4166 const struct rte_flow_action *action,
4167 struct rte_eth_dev *dev,
4168 struct rte_flow_error *error)
4170 struct mlx5_priv *priv = dev->data->dev_private;
4171 const struct rte_flow_action_age *age = action->conf;
4173 if (!priv->config.devx || priv->sh->cmng.counter_fallback)
4174 return rte_flow_error_set(error, ENOTSUP,
4175 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4177 "age action not supported");
4178 if (!(action->conf))
4179 return rte_flow_error_set(error, EINVAL,
4180 RTE_FLOW_ERROR_TYPE_ACTION, action,
4181 "configuration cannot be null");
4182 if (!(age->timeout))
4183 return rte_flow_error_set(error, EINVAL,
4184 RTE_FLOW_ERROR_TYPE_ACTION, action,
4185 "invalid timeout value 0");
4186 if (action_flags & MLX5_FLOW_ACTION_AGE)
4187 return rte_flow_error_set(error, EINVAL,
4188 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4189 "duplicate age actions set");
4194 * Validate the modify-header IPv4 DSCP actions.
4196 * @param[in] action_flags
4197 * Holds the actions detected until now.
4199 * Pointer to the modify action.
4200 * @param[in] item_flags
4201 * Holds the items detected.
4203 * Pointer to error structure.
4206 * 0 on success, a negative errno value otherwise and rte_errno is set.
4209 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
4210 const struct rte_flow_action *action,
4211 const uint64_t item_flags,
4212 struct rte_flow_error *error)
4216 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4218 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
4219 return rte_flow_error_set(error, EINVAL,
4220 RTE_FLOW_ERROR_TYPE_ACTION,
4222 "no ipv4 item in pattern");
4228 * Validate the modify-header IPv6 DSCP actions.
4230 * @param[in] action_flags
4231 * Holds the actions detected until now.
4233 * Pointer to the modify action.
4234 * @param[in] item_flags
4235 * Holds the items detected.
4237 * Pointer to error structure.
4240 * 0 on success, a negative errno value otherwise and rte_errno is set.
4243 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
4244 const struct rte_flow_action *action,
4245 const uint64_t item_flags,
4246 struct rte_flow_error *error)
4250 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4252 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
4253 return rte_flow_error_set(error, EINVAL,
4254 RTE_FLOW_ERROR_TYPE_ACTION,
4256 "no ipv6 item in pattern");
4262 * Match modify-header resource.
4265 * Pointer to exist resource entry object.
4267 * Pointer to new modify-header resource.
4270 * 0 on matching, -1 otherwise.
4273 flow_dv_modify_hdr_resource_match(struct mlx5_hlist_entry *entry, void *ctx)
4275 struct mlx5_flow_dv_modify_hdr_resource *resource;
4276 struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
4277 uint32_t actions_len;
4279 resource = (struct mlx5_flow_dv_modify_hdr_resource *)ctx;
4280 cache_resource = container_of(entry,
4281 struct mlx5_flow_dv_modify_hdr_resource,
4283 actions_len = resource->actions_num * sizeof(resource->actions[0]);
4284 if (resource->entry.key == cache_resource->entry.key &&
4285 resource->ft_type == cache_resource->ft_type &&
4286 resource->actions_num == cache_resource->actions_num &&
4287 resource->flags == cache_resource->flags &&
4288 !memcmp((const void *)resource->actions,
4289 (const void *)cache_resource->actions,
4296 * Validate the sample action.
4298 * @param[in] action_flags
4299 * Holds the actions detected until now.
4301 * Pointer to the sample action.
4303 * Pointer to the Ethernet device structure.
4305 * Attributes of flow that includes this action.
4307 * Pointer to error structure.
4310 * 0 on success, a negative errno value otherwise and rte_errno is set.
4313 flow_dv_validate_action_sample(uint64_t action_flags,
4314 const struct rte_flow_action *action,
4315 struct rte_eth_dev *dev,
4316 const struct rte_flow_attr *attr,
4317 struct rte_flow_error *error)
4319 struct mlx5_priv *priv = dev->data->dev_private;
4320 struct mlx5_dev_config *dev_conf = &priv->config;
4321 const struct rte_flow_action_sample *sample = action->conf;
4322 const struct rte_flow_action *act;
4323 uint64_t sub_action_flags = 0;
4324 uint16_t queue_index = 0xFFFF;
4330 return rte_flow_error_set(error, EINVAL,
4331 RTE_FLOW_ERROR_TYPE_ACTION, action,
4332 "configuration cannot be NULL");
4333 if (sample->ratio == 0)
4334 return rte_flow_error_set(error, EINVAL,
4335 RTE_FLOW_ERROR_TYPE_ACTION, action,
4336 "ratio value starts from 1");
4337 if (!priv->config.devx || (sample->ratio > 0 && !priv->sampler_en))
4338 return rte_flow_error_set(error, ENOTSUP,
4339 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4341 "sample action not supported");
4342 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
4343 return rte_flow_error_set(error, EINVAL,
4344 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4345 "Multiple sample actions not "
4347 if (action_flags & MLX5_FLOW_ACTION_METER)
4348 return rte_flow_error_set(error, EINVAL,
4349 RTE_FLOW_ERROR_TYPE_ACTION, action,
4350 "wrong action order, meter should "
4351 "be after sample action");
4352 if (action_flags & MLX5_FLOW_ACTION_JUMP)
4353 return rte_flow_error_set(error, EINVAL,
4354 RTE_FLOW_ERROR_TYPE_ACTION, action,
4355 "wrong action order, jump should "
4356 "be after sample action");
4357 act = sample->actions;
4358 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
4359 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
4360 return rte_flow_error_set(error, ENOTSUP,
4361 RTE_FLOW_ERROR_TYPE_ACTION,
4362 act, "too many actions");
4363 switch (act->type) {
4364 case RTE_FLOW_ACTION_TYPE_QUEUE:
4365 ret = mlx5_flow_validate_action_queue(act,
4371 queue_index = ((const struct rte_flow_action_queue *)
4372 (act->conf))->index;
4373 sub_action_flags |= MLX5_FLOW_ACTION_QUEUE;
4376 case RTE_FLOW_ACTION_TYPE_MARK:
4377 ret = flow_dv_validate_action_mark(dev, act,
4382 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4383 sub_action_flags |= MLX5_FLOW_ACTION_MARK |
4384 MLX5_FLOW_ACTION_MARK_EXT;
4386 sub_action_flags |= MLX5_FLOW_ACTION_MARK;
4389 case RTE_FLOW_ACTION_TYPE_COUNT:
4390 ret = flow_dv_validate_action_count(dev, error);
4393 sub_action_flags |= MLX5_FLOW_ACTION_COUNT;
4396 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4397 ret = flow_dv_validate_action_port_id(dev,
4404 sub_action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4407 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4408 ret = flow_dv_validate_action_raw_encap_decap
4409 (dev, NULL, act->conf, attr, &sub_action_flags,
4416 return rte_flow_error_set(error, ENOTSUP,
4417 RTE_FLOW_ERROR_TYPE_ACTION,
4419 "Doesn't support optional "
4423 if (attr->ingress && !attr->transfer) {
4424 if (!(sub_action_flags & MLX5_FLOW_ACTION_QUEUE))
4425 return rte_flow_error_set(error, EINVAL,
4426 RTE_FLOW_ERROR_TYPE_ACTION,
4428 "Ingress must has a dest "
4429 "QUEUE for Sample");
4430 } else if (attr->egress && !attr->transfer) {
4431 return rte_flow_error_set(error, ENOTSUP,
4432 RTE_FLOW_ERROR_TYPE_ACTION,
4434 "Sample Only support Ingress "
4436 } else if (sample->actions->type != RTE_FLOW_ACTION_TYPE_END) {
4437 MLX5_ASSERT(attr->transfer);
4438 if (sample->ratio > 1)
4439 return rte_flow_error_set(error, ENOTSUP,
4440 RTE_FLOW_ERROR_TYPE_ACTION,
4442 "E-Switch doesn't support "
4443 "any optional action "
4446 if (sub_action_flags & MLX5_FLOW_ACTION_QUEUE)
4447 return rte_flow_error_set(error, ENOTSUP,
4448 RTE_FLOW_ERROR_TYPE_ACTION,
4450 "unsupported action QUEUE");
4451 if (!(sub_action_flags & MLX5_FLOW_ACTION_PORT_ID))
4452 return rte_flow_error_set(error, EINVAL,
4453 RTE_FLOW_ERROR_TYPE_ACTION,
4455 "E-Switch must has a dest "
4456 "port for mirroring");
4458 /* Continue validation for Xcap actions.*/
4459 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) &&
4460 (queue_index == 0xFFFF ||
4461 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
4462 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
4463 MLX5_FLOW_XCAP_ACTIONS)
4464 return rte_flow_error_set(error, ENOTSUP,
4465 RTE_FLOW_ERROR_TYPE_ACTION,
4466 NULL, "encap and decap "
4467 "combination aren't "
4469 if (!attr->transfer && attr->ingress && (sub_action_flags &
4470 MLX5_FLOW_ACTION_ENCAP))
4471 return rte_flow_error_set(error, ENOTSUP,
4472 RTE_FLOW_ERROR_TYPE_ACTION,
4473 NULL, "encap is not supported"
4474 " for ingress traffic");
4480 * Find existing modify-header resource or create and register a new one.
4482 * @param dev[in, out]
4483 * Pointer to rte_eth_dev structure.
4484 * @param[in, out] resource
4485 * Pointer to modify-header resource.
4486 * @parm[in, out] dev_flow
4487 * Pointer to the dev_flow.
4489 * pointer to error structure.
4492 * 0 on success otherwise -errno and errno is set.
4495 flow_dv_modify_hdr_resource_register
4496 (struct rte_eth_dev *dev,
4497 struct mlx5_flow_dv_modify_hdr_resource *resource,
4498 struct mlx5_flow *dev_flow,
4499 struct rte_flow_error *error)
4501 struct mlx5_priv *priv = dev->data->dev_private;
4502 struct mlx5_dev_ctx_shared *sh = priv->sh;
4503 struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
4504 struct mlx5dv_dr_domain *ns;
4505 uint32_t actions_len;
4506 struct mlx5_hlist_entry *entry;
4507 union mlx5_flow_modify_hdr_key hdr_mod_key = {
4509 .ft_type = resource->ft_type,
4510 .actions_num = resource->actions_num,
4511 .group = dev_flow->dv.group,
4517 resource->flags = dev_flow->dv.group ? 0 :
4518 MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4519 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
4521 return rte_flow_error_set(error, EOVERFLOW,
4522 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4523 "too many modify header items");
4524 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
4525 ns = sh->fdb_domain;
4526 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
4530 /* Lookup a matching resource from cache. */
4531 actions_len = resource->actions_num * sizeof(resource->actions[0]);
4532 hdr_mod_key.cksum = __rte_raw_cksum(resource->actions, actions_len, 0);
4533 resource->entry.key = hdr_mod_key.v64;
4534 entry = mlx5_hlist_lookup_ex(sh->modify_cmds, resource->entry.key,
4535 flow_dv_modify_hdr_resource_match,
4538 cache_resource = container_of(entry,
4539 struct mlx5_flow_dv_modify_hdr_resource,
4541 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
4542 (void *)cache_resource,
4543 rte_atomic32_read(&cache_resource->refcnt));
4544 rte_atomic32_inc(&cache_resource->refcnt);
4545 dev_flow->handle->dvh.modify_hdr = cache_resource;
4549 /* Register new modify-header resource. */
4550 cache_resource = mlx5_malloc(MLX5_MEM_ZERO,
4551 sizeof(*cache_resource) + actions_len, 0,
4553 if (!cache_resource)
4554 return rte_flow_error_set(error, ENOMEM,
4555 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4556 "cannot allocate resource memory");
4557 *cache_resource = *resource;
4558 rte_memcpy(cache_resource->actions, resource->actions, actions_len);
4559 ret = mlx5_flow_os_create_flow_action_modify_header
4560 (sh->ctx, ns, cache_resource,
4561 actions_len, &cache_resource->action);
4563 mlx5_free(cache_resource);
4564 return rte_flow_error_set(error, ENOMEM,
4565 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4566 NULL, "cannot create action");
4568 rte_atomic32_init(&cache_resource->refcnt);
4569 rte_atomic32_inc(&cache_resource->refcnt);
4570 if (mlx5_hlist_insert_ex(sh->modify_cmds, &cache_resource->entry,
4571 flow_dv_modify_hdr_resource_match,
4572 (void *)cache_resource)) {
4573 claim_zero(mlx5_flow_os_destroy_flow_action
4574 (cache_resource->action));
4575 mlx5_free(cache_resource);
4576 return rte_flow_error_set(error, EEXIST,
4577 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4578 NULL, "action exist");
4580 dev_flow->handle->dvh.modify_hdr = cache_resource;
4581 DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
4582 (void *)cache_resource,
4583 rte_atomic32_read(&cache_resource->refcnt));
4588 * Get DV flow counter by index.
4591 * Pointer to the Ethernet device structure.
4593 * mlx5 flow counter index in the container.
4595 * mlx5 flow counter pool in the container,
4598 * Pointer to the counter, NULL otherwise.
4600 static struct mlx5_flow_counter *
4601 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
4603 struct mlx5_flow_counter_pool **ppool)
4605 struct mlx5_priv *priv = dev->data->dev_private;
4606 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4607 struct mlx5_flow_counter_pool *pool;
4609 /* Decrease to original index and clear shared bit. */
4610 idx = (idx - 1) & (MLX5_CNT_SHARED_OFFSET - 1);
4611 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cmng->n);
4612 pool = cmng->pools[idx / MLX5_COUNTERS_PER_POOL];
4616 return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
4620 * Check the devx counter belongs to the pool.
4623 * Pointer to the counter pool.
4625 * The counter devx ID.
4628 * True if counter belongs to the pool, false otherwise.
4631 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
4633 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
4634 MLX5_COUNTERS_PER_POOL;
4636 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
4642 * Get a pool by devx counter ID.
4645 * Pointer to the counter management.
4647 * The counter devx ID.
4650 * The counter pool pointer if exists, NULL otherwise,
4652 static struct mlx5_flow_counter_pool *
4653 flow_dv_find_pool_by_id(struct mlx5_flow_counter_mng *cmng, int id)
4656 struct mlx5_flow_counter_pool *pool = NULL;
4658 rte_spinlock_lock(&cmng->pool_update_sl);
4659 /* Check last used pool. */
4660 if (cmng->last_pool_idx != POOL_IDX_INVALID &&
4661 flow_dv_is_counter_in_pool(cmng->pools[cmng->last_pool_idx], id)) {
4662 pool = cmng->pools[cmng->last_pool_idx];
4665 /* ID out of range means no suitable pool in the container. */
4666 if (id > cmng->max_id || id < cmng->min_id)
4669 * Find the pool from the end of the container, since mostly counter
4670 * ID is sequence increasing, and the last pool should be the needed
4675 struct mlx5_flow_counter_pool *pool_tmp = cmng->pools[i];
4677 if (flow_dv_is_counter_in_pool(pool_tmp, id)) {
4683 rte_spinlock_unlock(&cmng->pool_update_sl);
4688 * Resize a counter container.
4691 * Pointer to the Ethernet device structure.
4694 * 0 on success, otherwise negative errno value and rte_errno is set.
4697 flow_dv_container_resize(struct rte_eth_dev *dev)
4699 struct mlx5_priv *priv = dev->data->dev_private;
4700 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4701 void *old_pools = cmng->pools;
4702 uint32_t resize = cmng->n + MLX5_CNT_CONTAINER_RESIZE;
4703 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4704 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
4711 memcpy(pools, old_pools, cmng->n *
4712 sizeof(struct mlx5_flow_counter_pool *));
4714 cmng->pools = pools;
4716 mlx5_free(old_pools);
4721 * Query a devx flow counter.
4724 * Pointer to the Ethernet device structure.
4726 * Index to the flow counter.
4728 * The statistics value of packets.
4730 * The statistics value of bytes.
4733 * 0 on success, otherwise a negative errno value and rte_errno is set.
4736 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4739 struct mlx5_priv *priv = dev->data->dev_private;
4740 struct mlx5_flow_counter_pool *pool = NULL;
4741 struct mlx5_flow_counter *cnt;
4744 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4746 if (priv->sh->cmng.counter_fallback)
4747 return mlx5_devx_cmd_flow_counter_query(cnt->dcs_when_active, 0,
4748 0, pkts, bytes, 0, NULL, NULL, 0);
4749 rte_spinlock_lock(&pool->sl);
4754 offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
4755 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4756 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4758 rte_spinlock_unlock(&pool->sl);
4763 * Create and initialize a new counter pool.
4766 * Pointer to the Ethernet device structure.
4768 * The devX counter handle.
4770 * Whether the pool is for counter that was allocated for aging.
4771 * @param[in/out] cont_cur
4772 * Pointer to the container pointer, it will be update in pool resize.
4775 * The pool container pointer on success, NULL otherwise and rte_errno is set.
4777 static struct mlx5_flow_counter_pool *
4778 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4781 struct mlx5_priv *priv = dev->data->dev_private;
4782 struct mlx5_flow_counter_pool *pool;
4783 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4784 bool fallback = priv->sh->cmng.counter_fallback;
4785 uint32_t size = sizeof(*pool);
4787 size += MLX5_COUNTERS_PER_POOL * MLX5_CNT_SIZE;
4788 size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * MLX5_AGE_SIZE);
4789 pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
4795 pool->is_aged = !!age;
4796 pool->query_gen = 0;
4797 pool->min_dcs = dcs;
4798 rte_spinlock_init(&pool->sl);
4799 rte_spinlock_init(&pool->csl);
4800 TAILQ_INIT(&pool->counters[0]);
4801 TAILQ_INIT(&pool->counters[1]);
4802 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
4803 rte_spinlock_lock(&cmng->pool_update_sl);
4804 pool->index = cmng->n_valid;
4805 if (pool->index == cmng->n && flow_dv_container_resize(dev)) {
4807 rte_spinlock_unlock(&cmng->pool_update_sl);
4810 cmng->pools[pool->index] = pool;
4812 if (unlikely(fallback)) {
4813 int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
4815 if (base < cmng->min_id)
4816 cmng->min_id = base;
4817 if (base > cmng->max_id)
4818 cmng->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
4819 cmng->last_pool_idx = pool->index;
4821 rte_spinlock_unlock(&cmng->pool_update_sl);
4826 * Prepare a new counter and/or a new counter pool.
4829 * Pointer to the Ethernet device structure.
4830 * @param[out] cnt_free
4831 * Where to put the pointer of a new counter.
4833 * Whether the pool is for counter that was allocated for aging.
4836 * The counter pool pointer and @p cnt_free is set on success,
4837 * NULL otherwise and rte_errno is set.
4839 static struct mlx5_flow_counter_pool *
4840 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4841 struct mlx5_flow_counter **cnt_free,
4844 struct mlx5_priv *priv = dev->data->dev_private;
4845 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4846 struct mlx5_flow_counter_pool *pool;
4847 struct mlx5_counters tmp_tq;
4848 struct mlx5_devx_obj *dcs = NULL;
4849 struct mlx5_flow_counter *cnt;
4850 enum mlx5_counter_type cnt_type =
4851 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
4852 bool fallback = priv->sh->cmng.counter_fallback;
4856 /* bulk_bitmap must be 0 for single counter allocation. */
4857 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4860 pool = flow_dv_find_pool_by_id(cmng, dcs->id);
4862 pool = flow_dv_pool_create(dev, dcs, age);
4864 mlx5_devx_cmd_destroy(dcs);
4868 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4869 cnt = MLX5_POOL_GET_CNT(pool, i);
4871 cnt->dcs_when_free = dcs;
4875 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4877 rte_errno = ENODATA;
4880 pool = flow_dv_pool_create(dev, dcs, age);
4882 mlx5_devx_cmd_destroy(dcs);
4885 TAILQ_INIT(&tmp_tq);
4886 for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
4887 cnt = MLX5_POOL_GET_CNT(pool, i);
4889 TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
4891 rte_spinlock_lock(&cmng->csl[cnt_type]);
4892 TAILQ_CONCAT(&cmng->counters[cnt_type], &tmp_tq, next);
4893 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4894 *cnt_free = MLX5_POOL_GET_CNT(pool, 0);
4895 (*cnt_free)->pool = pool;
4900 * Allocate a flow counter.
4903 * Pointer to the Ethernet device structure.
4905 * Whether the counter was allocated for aging.
4908 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4911 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t age)
4913 struct mlx5_priv *priv = dev->data->dev_private;
4914 struct mlx5_flow_counter_pool *pool = NULL;
4915 struct mlx5_flow_counter *cnt_free = NULL;
4916 bool fallback = priv->sh->cmng.counter_fallback;
4917 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4918 enum mlx5_counter_type cnt_type =
4919 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
4922 if (!priv->config.devx) {
4923 rte_errno = ENOTSUP;
4926 /* Get free counters from container. */
4927 rte_spinlock_lock(&cmng->csl[cnt_type]);
4928 cnt_free = TAILQ_FIRST(&cmng->counters[cnt_type]);
4930 TAILQ_REMOVE(&cmng->counters[cnt_type], cnt_free, next);
4931 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4932 if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free, age))
4934 pool = cnt_free->pool;
4936 cnt_free->dcs_when_active = cnt_free->dcs_when_free;
4937 /* Create a DV counter action only in the first time usage. */
4938 if (!cnt_free->action) {
4940 struct mlx5_devx_obj *dcs;
4944 offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
4945 dcs = pool->min_dcs;
4948 dcs = cnt_free->dcs_when_free;
4950 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
4957 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4958 MLX5_CNT_ARRAY_IDX(pool, cnt_free));
4959 /* Update the counter reset values. */
4960 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4963 if (!fallback && !priv->sh->cmng.query_thread_on)
4964 /* Start the asynchronous batch query by the host thread. */
4965 mlx5_set_query_alarm(priv->sh);
4969 cnt_free->pool = pool;
4971 cnt_free->dcs_when_free = cnt_free->dcs_when_active;
4972 rte_spinlock_lock(&cmng->csl[cnt_type]);
4973 TAILQ_INSERT_TAIL(&cmng->counters[cnt_type], cnt_free, next);
4974 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4980 * Allocate a shared flow counter.
4983 * Pointer to the shared counter configuration.
4985 * Pointer to save the allocated counter index.
4988 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4992 flow_dv_counter_alloc_shared_cb(void *ctx, union mlx5_l3t_data *data)
4994 struct mlx5_shared_counter_conf *conf = ctx;
4995 struct rte_eth_dev *dev = conf->dev;
4996 struct mlx5_flow_counter *cnt;
4998 data->dword = flow_dv_counter_alloc(dev, 0);
4999 data->dword |= MLX5_CNT_SHARED_OFFSET;
5000 cnt = flow_dv_counter_get_by_idx(dev, data->dword, NULL);
5001 cnt->shared_info.id = conf->id;
5006 * Get a shared flow counter.
5009 * Pointer to the Ethernet device structure.
5011 * Counter identifier.
5014 * Index to flow counter on success, 0 otherwise and rte_errno is set.
5017 flow_dv_counter_get_shared(struct rte_eth_dev *dev, uint32_t id)
5019 struct mlx5_priv *priv = dev->data->dev_private;
5020 struct mlx5_shared_counter_conf conf = {
5024 union mlx5_l3t_data data = {
5028 mlx5_l3t_prepare_entry(priv->sh->cnt_id_tbl, id, &data,
5029 flow_dv_counter_alloc_shared_cb, &conf);
5034 * Get age param from counter index.
5037 * Pointer to the Ethernet device structure.
5038 * @param[in] counter
5039 * Index to the counter handler.
5042 * The aging parameter specified for the counter index.
5044 static struct mlx5_age_param*
5045 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
5048 struct mlx5_flow_counter *cnt;
5049 struct mlx5_flow_counter_pool *pool = NULL;
5051 flow_dv_counter_get_by_idx(dev, counter, &pool);
5052 counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
5053 cnt = MLX5_POOL_GET_CNT(pool, counter);
5054 return MLX5_CNT_TO_AGE(cnt);
5058 * Remove a flow counter from aged counter list.
5061 * Pointer to the Ethernet device structure.
5062 * @param[in] counter
5063 * Index to the counter handler.
5065 * Pointer to the counter handler.
5068 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
5069 uint32_t counter, struct mlx5_flow_counter *cnt)
5071 struct mlx5_age_info *age_info;
5072 struct mlx5_age_param *age_param;
5073 struct mlx5_priv *priv = dev->data->dev_private;
5074 uint16_t expected = AGE_CANDIDATE;
5076 age_info = GET_PORT_AGE_INFO(priv);
5077 age_param = flow_dv_counter_idx_get_age(dev, counter);
5078 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
5079 AGE_FREE, false, __ATOMIC_RELAXED,
5080 __ATOMIC_RELAXED)) {
5082 * We need the lock even it is age timeout,
5083 * since counter may still in process.
5085 rte_spinlock_lock(&age_info->aged_sl);
5086 TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
5087 rte_spinlock_unlock(&age_info->aged_sl);
5088 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
5093 * Release a flow counter.
5096 * Pointer to the Ethernet device structure.
5097 * @param[in] counter
5098 * Index to the counter handler.
5101 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
5103 struct mlx5_priv *priv = dev->data->dev_private;
5104 struct mlx5_flow_counter_pool *pool = NULL;
5105 struct mlx5_flow_counter *cnt;
5106 enum mlx5_counter_type cnt_type;
5110 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
5112 if (IS_SHARED_CNT(counter) &&
5113 mlx5_l3t_clear_entry(priv->sh->cnt_id_tbl, cnt->shared_info.id))
5116 flow_dv_counter_remove_from_age(dev, counter, cnt);
5119 * Put the counter back to list to be updated in none fallback mode.
5120 * Currently, we are using two list alternately, while one is in query,
5121 * add the freed counter to the other list based on the pool query_gen
5122 * value. After query finishes, add counter the list to the global
5123 * container counter list. The list changes while query starts. In
5124 * this case, lock will not be needed as query callback and release
5125 * function both operate with the different list.
5128 if (!priv->sh->cmng.counter_fallback) {
5129 rte_spinlock_lock(&pool->csl);
5130 TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
5131 rte_spinlock_unlock(&pool->csl);
5133 cnt->dcs_when_free = cnt->dcs_when_active;
5134 cnt_type = pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
5135 MLX5_COUNTER_TYPE_ORIGIN;
5136 rte_spinlock_lock(&priv->sh->cmng.csl[cnt_type]);
5137 TAILQ_INSERT_TAIL(&priv->sh->cmng.counters[cnt_type],
5139 rte_spinlock_unlock(&priv->sh->cmng.csl[cnt_type]);
5144 * Verify the @p attributes will be correctly understood by the NIC and store
5145 * them in the @p flow if everything is correct.
5148 * Pointer to dev struct.
5149 * @param[in] attributes
5150 * Pointer to flow attributes
5151 * @param[in] external
5152 * This flow rule is created by request external to PMD.
5154 * Pointer to error structure.
5157 * - 0 on success and non root table.
5158 * - 1 on success and root table.
5159 * - a negative errno value otherwise and rte_errno is set.
5162 flow_dv_validate_attributes(struct rte_eth_dev *dev,
5163 const struct rte_flow_attr *attributes,
5164 bool external __rte_unused,
5165 struct rte_flow_error *error)
5167 struct mlx5_priv *priv = dev->data->dev_private;
5168 uint32_t priority_max = priv->config.flow_prio - 1;
5171 #ifndef HAVE_MLX5DV_DR
5172 if (attributes->group)
5173 return rte_flow_error_set(error, ENOTSUP,
5174 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5176 "groups are not supported");
5180 ret = mlx5_flow_group_to_table(attributes, external,
5181 attributes->group, !!priv->fdb_def_rule,
5186 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
5188 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
5189 attributes->priority >= priority_max)
5190 return rte_flow_error_set(error, ENOTSUP,
5191 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
5193 "priority out of range");
5194 if (attributes->transfer) {
5195 if (!priv->config.dv_esw_en)
5196 return rte_flow_error_set
5198 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5199 "E-Switch dr is not supported");
5200 if (!(priv->representor || priv->master))
5201 return rte_flow_error_set
5202 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5203 NULL, "E-Switch configuration can only be"
5204 " done by a master or a representor device");
5205 if (attributes->egress)
5206 return rte_flow_error_set
5208 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
5209 "egress is not supported");
5211 if (!(attributes->egress ^ attributes->ingress))
5212 return rte_flow_error_set(error, ENOTSUP,
5213 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
5214 "must specify exactly one of "
5215 "ingress or egress");
5220 * Internal validation function. For validating both actions and items.
5223 * Pointer to the rte_eth_dev structure.
5225 * Pointer to the flow attributes.
5227 * Pointer to the list of items.
5228 * @param[in] actions
5229 * Pointer to the list of actions.
5230 * @param[in] external
5231 * This flow rule is created by request external to PMD.
5232 * @param[in] hairpin
5233 * Number of hairpin TX actions, 0 means classic flow.
5235 * Pointer to the error structure.
5238 * 0 on success, a negative errno value otherwise and rte_errno is set.
5241 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
5242 const struct rte_flow_item items[],
5243 const struct rte_flow_action actions[],
5244 bool external, int hairpin, struct rte_flow_error *error)
5247 uint64_t action_flags = 0;
5248 uint64_t item_flags = 0;
5249 uint64_t last_item = 0;
5250 uint8_t next_protocol = 0xff;
5251 uint16_t ether_type = 0;
5253 uint8_t item_ipv6_proto = 0;
5254 const struct rte_flow_item *gre_item = NULL;
5255 const struct rte_flow_action_raw_decap *decap;
5256 const struct rte_flow_action_raw_encap *encap;
5257 const struct rte_flow_action_rss *rss;
5258 const struct rte_flow_item_tcp nic_tcp_mask = {
5261 .src_port = RTE_BE16(UINT16_MAX),
5262 .dst_port = RTE_BE16(UINT16_MAX),
5265 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
5268 "\xff\xff\xff\xff\xff\xff\xff\xff"
5269 "\xff\xff\xff\xff\xff\xff\xff\xff",
5271 "\xff\xff\xff\xff\xff\xff\xff\xff"
5272 "\xff\xff\xff\xff\xff\xff\xff\xff",
5273 .vtc_flow = RTE_BE32(0xffffffff),
5279 const struct rte_flow_item_ecpri nic_ecpri_mask = {
5283 RTE_BE32(((const struct rte_ecpri_common_hdr) {
5287 .dummy[0] = 0xffffffff,
5290 struct mlx5_priv *priv = dev->data->dev_private;
5291 struct mlx5_dev_config *dev_conf = &priv->config;
5292 uint16_t queue_index = 0xFFFF;
5293 const struct rte_flow_item_vlan *vlan_m = NULL;
5294 int16_t rw_act_num = 0;
5299 ret = flow_dv_validate_attributes(dev, attr, external, error);
5302 is_root = (uint64_t)ret;
5303 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5304 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
5305 int type = items->type;
5307 if (!mlx5_flow_os_item_supported(type))
5308 return rte_flow_error_set(error, ENOTSUP,
5309 RTE_FLOW_ERROR_TYPE_ITEM,
5310 NULL, "item not supported");
5312 case RTE_FLOW_ITEM_TYPE_VOID:
5314 case RTE_FLOW_ITEM_TYPE_PORT_ID:
5315 ret = flow_dv_validate_item_port_id
5316 (dev, items, attr, item_flags, error);
5319 last_item = MLX5_FLOW_ITEM_PORT_ID;
5321 case RTE_FLOW_ITEM_TYPE_ETH:
5322 ret = mlx5_flow_validate_item_eth(items, item_flags,
5326 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
5327 MLX5_FLOW_LAYER_OUTER_L2;
5328 if (items->mask != NULL && items->spec != NULL) {
5330 ((const struct rte_flow_item_eth *)
5333 ((const struct rte_flow_item_eth *)
5335 ether_type = rte_be_to_cpu_16(ether_type);
5340 case RTE_FLOW_ITEM_TYPE_VLAN:
5341 ret = flow_dv_validate_item_vlan(items, item_flags,
5345 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
5346 MLX5_FLOW_LAYER_OUTER_VLAN;
5347 if (items->mask != NULL && items->spec != NULL) {
5349 ((const struct rte_flow_item_vlan *)
5350 items->spec)->inner_type;
5352 ((const struct rte_flow_item_vlan *)
5353 items->mask)->inner_type;
5354 ether_type = rte_be_to_cpu_16(ether_type);
5358 /* Store outer VLAN mask for of_push_vlan action. */
5360 vlan_m = items->mask;
5362 case RTE_FLOW_ITEM_TYPE_IPV4:
5363 mlx5_flow_tunnel_ip_check(items, next_protocol,
5364 &item_flags, &tunnel);
5365 ret = flow_dv_validate_item_ipv4(items, item_flags,
5366 last_item, ether_type,
5370 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
5371 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
5372 if (items->mask != NULL &&
5373 ((const struct rte_flow_item_ipv4 *)
5374 items->mask)->hdr.next_proto_id) {
5376 ((const struct rte_flow_item_ipv4 *)
5377 (items->spec))->hdr.next_proto_id;
5379 ((const struct rte_flow_item_ipv4 *)
5380 (items->mask))->hdr.next_proto_id;
5382 /* Reset for inner layer. */
5383 next_protocol = 0xff;
5386 case RTE_FLOW_ITEM_TYPE_IPV6:
5387 mlx5_flow_tunnel_ip_check(items, next_protocol,
5388 &item_flags, &tunnel);
5389 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
5396 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
5397 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
5398 if (items->mask != NULL &&
5399 ((const struct rte_flow_item_ipv6 *)
5400 items->mask)->hdr.proto) {
5402 ((const struct rte_flow_item_ipv6 *)
5403 items->spec)->hdr.proto;
5405 ((const struct rte_flow_item_ipv6 *)
5406 items->spec)->hdr.proto;
5408 ((const struct rte_flow_item_ipv6 *)
5409 items->mask)->hdr.proto;
5411 /* Reset for inner layer. */
5412 next_protocol = 0xff;
5415 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
5416 ret = flow_dv_validate_item_ipv6_frag_ext(items,
5421 last_item = tunnel ?
5422 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
5423 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
5424 if (items->mask != NULL &&
5425 ((const struct rte_flow_item_ipv6_frag_ext *)
5426 items->mask)->hdr.next_header) {
5428 ((const struct rte_flow_item_ipv6_frag_ext *)
5429 items->spec)->hdr.next_header;
5431 ((const struct rte_flow_item_ipv6_frag_ext *)
5432 items->mask)->hdr.next_header;
5434 /* Reset for inner layer. */
5435 next_protocol = 0xff;
5438 case RTE_FLOW_ITEM_TYPE_TCP:
5439 ret = mlx5_flow_validate_item_tcp
5446 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
5447 MLX5_FLOW_LAYER_OUTER_L4_TCP;
5449 case RTE_FLOW_ITEM_TYPE_UDP:
5450 ret = mlx5_flow_validate_item_udp(items, item_flags,
5455 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
5456 MLX5_FLOW_LAYER_OUTER_L4_UDP;
5458 case RTE_FLOW_ITEM_TYPE_GRE:
5459 ret = mlx5_flow_validate_item_gre(items, item_flags,
5460 next_protocol, error);
5464 last_item = MLX5_FLOW_LAYER_GRE;
5466 case RTE_FLOW_ITEM_TYPE_NVGRE:
5467 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
5472 last_item = MLX5_FLOW_LAYER_NVGRE;
5474 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5475 ret = mlx5_flow_validate_item_gre_key
5476 (items, item_flags, gre_item, error);
5479 last_item = MLX5_FLOW_LAYER_GRE_KEY;
5481 case RTE_FLOW_ITEM_TYPE_VXLAN:
5482 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
5486 last_item = MLX5_FLOW_LAYER_VXLAN;
5488 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5489 ret = mlx5_flow_validate_item_vxlan_gpe(items,
5494 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5496 case RTE_FLOW_ITEM_TYPE_GENEVE:
5497 ret = mlx5_flow_validate_item_geneve(items,
5502 last_item = MLX5_FLOW_LAYER_GENEVE;
5504 case RTE_FLOW_ITEM_TYPE_MPLS:
5505 ret = mlx5_flow_validate_item_mpls(dev, items,
5510 last_item = MLX5_FLOW_LAYER_MPLS;
5513 case RTE_FLOW_ITEM_TYPE_MARK:
5514 ret = flow_dv_validate_item_mark(dev, items, attr,
5518 last_item = MLX5_FLOW_ITEM_MARK;
5520 case RTE_FLOW_ITEM_TYPE_META:
5521 ret = flow_dv_validate_item_meta(dev, items, attr,
5525 last_item = MLX5_FLOW_ITEM_METADATA;
5527 case RTE_FLOW_ITEM_TYPE_ICMP:
5528 ret = mlx5_flow_validate_item_icmp(items, item_flags,
5533 last_item = MLX5_FLOW_LAYER_ICMP;
5535 case RTE_FLOW_ITEM_TYPE_ICMP6:
5536 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
5541 item_ipv6_proto = IPPROTO_ICMPV6;
5542 last_item = MLX5_FLOW_LAYER_ICMP6;
5544 case RTE_FLOW_ITEM_TYPE_TAG:
5545 ret = flow_dv_validate_item_tag(dev, items,
5549 last_item = MLX5_FLOW_ITEM_TAG;
5551 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
5552 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
5554 case RTE_FLOW_ITEM_TYPE_GTP:
5555 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
5559 last_item = MLX5_FLOW_LAYER_GTP;
5561 case RTE_FLOW_ITEM_TYPE_ECPRI:
5562 /* Capacity will be checked in the translate stage. */
5563 ret = mlx5_flow_validate_item_ecpri(items, item_flags,
5570 last_item = MLX5_FLOW_LAYER_ECPRI;
5573 return rte_flow_error_set(error, ENOTSUP,
5574 RTE_FLOW_ERROR_TYPE_ITEM,
5575 NULL, "item not supported");
5577 item_flags |= last_item;
5579 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5580 int type = actions->type;
5582 if (!mlx5_flow_os_action_supported(type))
5583 return rte_flow_error_set(error, ENOTSUP,
5584 RTE_FLOW_ERROR_TYPE_ACTION,
5586 "action not supported");
5587 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5588 return rte_flow_error_set(error, ENOTSUP,
5589 RTE_FLOW_ERROR_TYPE_ACTION,
5590 actions, "too many actions");
5592 case RTE_FLOW_ACTION_TYPE_VOID:
5594 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5595 ret = flow_dv_validate_action_port_id(dev,
5602 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5605 case RTE_FLOW_ACTION_TYPE_FLAG:
5606 ret = flow_dv_validate_action_flag(dev, action_flags,
5610 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5611 /* Count all modify-header actions as one. */
5612 if (!(action_flags &
5613 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5615 action_flags |= MLX5_FLOW_ACTION_FLAG |
5616 MLX5_FLOW_ACTION_MARK_EXT;
5618 action_flags |= MLX5_FLOW_ACTION_FLAG;
5621 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5623 case RTE_FLOW_ACTION_TYPE_MARK:
5624 ret = flow_dv_validate_action_mark(dev, actions,
5629 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5630 /* Count all modify-header actions as one. */
5631 if (!(action_flags &
5632 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5634 action_flags |= MLX5_FLOW_ACTION_MARK |
5635 MLX5_FLOW_ACTION_MARK_EXT;
5637 action_flags |= MLX5_FLOW_ACTION_MARK;
5640 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5642 case RTE_FLOW_ACTION_TYPE_SET_META:
5643 ret = flow_dv_validate_action_set_meta(dev, actions,
5648 /* Count all modify-header actions as one action. */
5649 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5651 action_flags |= MLX5_FLOW_ACTION_SET_META;
5652 rw_act_num += MLX5_ACT_NUM_SET_META;
5654 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5655 ret = flow_dv_validate_action_set_tag(dev, actions,
5660 /* Count all modify-header actions as one action. */
5661 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5663 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
5664 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5666 case RTE_FLOW_ACTION_TYPE_DROP:
5667 ret = mlx5_flow_validate_action_drop(action_flags,
5671 action_flags |= MLX5_FLOW_ACTION_DROP;
5674 case RTE_FLOW_ACTION_TYPE_QUEUE:
5675 ret = mlx5_flow_validate_action_queue(actions,
5680 queue_index = ((const struct rte_flow_action_queue *)
5681 (actions->conf))->index;
5682 action_flags |= MLX5_FLOW_ACTION_QUEUE;
5685 case RTE_FLOW_ACTION_TYPE_RSS:
5686 rss = actions->conf;
5687 ret = mlx5_flow_validate_action_rss(actions,
5693 if (rss != NULL && rss->queue_num)
5694 queue_index = rss->queue[0];
5695 action_flags |= MLX5_FLOW_ACTION_RSS;
5698 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
5700 mlx5_flow_validate_action_default_miss(action_flags,
5704 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
5707 case RTE_FLOW_ACTION_TYPE_COUNT:
5708 ret = flow_dv_validate_action_count(dev, error);
5711 action_flags |= MLX5_FLOW_ACTION_COUNT;
5714 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5715 if (flow_dv_validate_action_pop_vlan(dev,
5721 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5724 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5725 ret = flow_dv_validate_action_push_vlan(dev,
5732 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5735 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5736 ret = flow_dv_validate_action_set_vlan_pcp
5737 (action_flags, actions, error);
5740 /* Count PCP with push_vlan command. */
5741 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5743 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5744 ret = flow_dv_validate_action_set_vlan_vid
5745 (item_flags, action_flags,
5749 /* Count VID with push_vlan command. */
5750 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5751 rw_act_num += MLX5_ACT_NUM_MDF_VID;
5753 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5754 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5755 ret = flow_dv_validate_action_l2_encap(dev,
5761 action_flags |= MLX5_FLOW_ACTION_ENCAP;
5764 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5765 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5766 ret = flow_dv_validate_action_decap(dev, action_flags,
5770 action_flags |= MLX5_FLOW_ACTION_DECAP;
5773 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5774 ret = flow_dv_validate_action_raw_encap_decap
5775 (dev, NULL, actions->conf, attr, &action_flags,
5780 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5781 decap = actions->conf;
5782 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5784 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5788 encap = actions->conf;
5790 ret = flow_dv_validate_action_raw_encap_decap
5792 decap ? decap : &empty_decap, encap,
5793 attr, &action_flags, &actions_n,
5798 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5799 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5800 ret = flow_dv_validate_action_modify_mac(action_flags,
5806 /* Count all modify-header actions as one action. */
5807 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5809 action_flags |= actions->type ==
5810 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5811 MLX5_FLOW_ACTION_SET_MAC_SRC :
5812 MLX5_FLOW_ACTION_SET_MAC_DST;
5814 * Even if the source and destination MAC addresses have
5815 * overlap in the header with 4B alignment, the convert
5816 * function will handle them separately and 4 SW actions
5817 * will be created. And 2 actions will be added each
5818 * time no matter how many bytes of address will be set.
5820 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5822 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5823 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5824 ret = flow_dv_validate_action_modify_ipv4(action_flags,
5830 /* Count all modify-header actions as one action. */
5831 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5833 action_flags |= actions->type ==
5834 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5835 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5836 MLX5_FLOW_ACTION_SET_IPV4_DST;
5837 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5839 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5840 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5841 ret = flow_dv_validate_action_modify_ipv6(action_flags,
5847 if (item_ipv6_proto == IPPROTO_ICMPV6)
5848 return rte_flow_error_set(error, ENOTSUP,
5849 RTE_FLOW_ERROR_TYPE_ACTION,
5851 "Can't change header "
5852 "with ICMPv6 proto");
5853 /* Count all modify-header actions as one action. */
5854 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5856 action_flags |= actions->type ==
5857 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5858 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5859 MLX5_FLOW_ACTION_SET_IPV6_DST;
5860 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5862 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5863 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5864 ret = flow_dv_validate_action_modify_tp(action_flags,
5870 /* Count all modify-header actions as one action. */
5871 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5873 action_flags |= actions->type ==
5874 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5875 MLX5_FLOW_ACTION_SET_TP_SRC :
5876 MLX5_FLOW_ACTION_SET_TP_DST;
5877 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5879 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5880 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5881 ret = flow_dv_validate_action_modify_ttl(action_flags,
5887 /* Count all modify-header actions as one action. */
5888 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5890 action_flags |= actions->type ==
5891 RTE_FLOW_ACTION_TYPE_SET_TTL ?
5892 MLX5_FLOW_ACTION_SET_TTL :
5893 MLX5_FLOW_ACTION_DEC_TTL;
5894 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5896 case RTE_FLOW_ACTION_TYPE_JUMP:
5897 ret = flow_dv_validate_action_jump(actions,
5904 action_flags |= MLX5_FLOW_ACTION_JUMP;
5906 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5907 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5908 ret = flow_dv_validate_action_modify_tcp_seq
5915 /* Count all modify-header actions as one action. */
5916 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5918 action_flags |= actions->type ==
5919 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5920 MLX5_FLOW_ACTION_INC_TCP_SEQ :
5921 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5922 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5924 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5925 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5926 ret = flow_dv_validate_action_modify_tcp_ack
5933 /* Count all modify-header actions as one action. */
5934 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5936 action_flags |= actions->type ==
5937 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5938 MLX5_FLOW_ACTION_INC_TCP_ACK :
5939 MLX5_FLOW_ACTION_DEC_TCP_ACK;
5940 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5942 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5944 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5945 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5946 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5948 case RTE_FLOW_ACTION_TYPE_METER:
5949 ret = mlx5_flow_validate_action_meter(dev,
5955 action_flags |= MLX5_FLOW_ACTION_METER;
5957 /* Meter action will add one more TAG action. */
5958 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5960 case RTE_FLOW_ACTION_TYPE_AGE:
5961 ret = flow_dv_validate_action_age(action_flags,
5966 action_flags |= MLX5_FLOW_ACTION_AGE;
5969 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5970 ret = flow_dv_validate_action_modify_ipv4_dscp
5977 /* Count all modify-header actions as one action. */
5978 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5980 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5981 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5983 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5984 ret = flow_dv_validate_action_modify_ipv6_dscp
5991 /* Count all modify-header actions as one action. */
5992 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5994 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5995 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5997 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5998 ret = flow_dv_validate_action_sample(action_flags,
6003 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
6007 return rte_flow_error_set(error, ENOTSUP,
6008 RTE_FLOW_ERROR_TYPE_ACTION,
6010 "action not supported");
6014 * Validate the drop action mutual exclusion with other actions.
6015 * Drop action is mutually-exclusive with any other action, except for
6018 if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
6019 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
6020 return rte_flow_error_set(error, EINVAL,
6021 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6022 "Drop action is mutually-exclusive "
6023 "with any other action, except for "
6025 /* Eswitch has few restrictions on using items and actions */
6026 if (attr->transfer) {
6027 if (!mlx5_flow_ext_mreg_supported(dev) &&
6028 action_flags & MLX5_FLOW_ACTION_FLAG)
6029 return rte_flow_error_set(error, ENOTSUP,
6030 RTE_FLOW_ERROR_TYPE_ACTION,
6032 "unsupported action FLAG");
6033 if (!mlx5_flow_ext_mreg_supported(dev) &&
6034 action_flags & MLX5_FLOW_ACTION_MARK)
6035 return rte_flow_error_set(error, ENOTSUP,
6036 RTE_FLOW_ERROR_TYPE_ACTION,
6038 "unsupported action MARK");
6039 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
6040 return rte_flow_error_set(error, ENOTSUP,
6041 RTE_FLOW_ERROR_TYPE_ACTION,
6043 "unsupported action QUEUE");
6044 if (action_flags & MLX5_FLOW_ACTION_RSS)
6045 return rte_flow_error_set(error, ENOTSUP,
6046 RTE_FLOW_ERROR_TYPE_ACTION,
6048 "unsupported action RSS");
6049 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
6050 return rte_flow_error_set(error, EINVAL,
6051 RTE_FLOW_ERROR_TYPE_ACTION,
6053 "no fate action is found");
6055 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
6056 return rte_flow_error_set(error, EINVAL,
6057 RTE_FLOW_ERROR_TYPE_ACTION,
6059 "no fate action is found");
6061 /* Continue validation for Xcap and VLAN actions.*/
6062 if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
6063 MLX5_FLOW_VLAN_ACTIONS)) &&
6064 (queue_index == 0xFFFF ||
6065 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
6066 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
6067 MLX5_FLOW_XCAP_ACTIONS)
6068 return rte_flow_error_set(error, ENOTSUP,
6069 RTE_FLOW_ERROR_TYPE_ACTION,
6070 NULL, "encap and decap "
6071 "combination aren't supported");
6072 if (!attr->transfer && attr->ingress) {
6073 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
6074 return rte_flow_error_set
6076 RTE_FLOW_ERROR_TYPE_ACTION,
6077 NULL, "encap is not supported"
6078 " for ingress traffic");
6079 else if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
6080 return rte_flow_error_set
6082 RTE_FLOW_ERROR_TYPE_ACTION,
6083 NULL, "push VLAN action not "
6084 "supported for ingress");
6085 else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
6086 MLX5_FLOW_VLAN_ACTIONS)
6087 return rte_flow_error_set
6089 RTE_FLOW_ERROR_TYPE_ACTION,
6090 NULL, "no support for "
6091 "multiple VLAN actions");
6094 /* Hairpin flow will add one more TAG action. */
6096 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6097 /* extra metadata enabled: one more TAG action will be add. */
6098 if (dev_conf->dv_flow_en &&
6099 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
6100 mlx5_flow_ext_mreg_supported(dev))
6101 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6102 if ((uint32_t)rw_act_num >
6103 flow_dv_modify_hdr_action_max(dev, is_root)) {
6104 return rte_flow_error_set(error, ENOTSUP,
6105 RTE_FLOW_ERROR_TYPE_ACTION,
6106 NULL, "too many header modify"
6107 " actions to support");
6113 * Internal preparation function. Allocates the DV flow size,
6114 * this size is constant.
6117 * Pointer to the rte_eth_dev structure.
6119 * Pointer to the flow attributes.
6121 * Pointer to the list of items.
6122 * @param[in] actions
6123 * Pointer to the list of actions.
6125 * Pointer to the error structure.
6128 * Pointer to mlx5_flow object on success,
6129 * otherwise NULL and rte_errno is set.
6131 static struct mlx5_flow *
6132 flow_dv_prepare(struct rte_eth_dev *dev,
6133 const struct rte_flow_attr *attr __rte_unused,
6134 const struct rte_flow_item items[] __rte_unused,
6135 const struct rte_flow_action actions[] __rte_unused,
6136 struct rte_flow_error *error)
6138 uint32_t handle_idx = 0;
6139 struct mlx5_flow *dev_flow;
6140 struct mlx5_flow_handle *dev_handle;
6141 struct mlx5_priv *priv = dev->data->dev_private;
6143 /* In case of corrupting the memory. */
6144 if (priv->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
6145 rte_flow_error_set(error, ENOSPC,
6146 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6147 "not free temporary device flow");
6150 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
6153 rte_flow_error_set(error, ENOMEM,
6154 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6155 "not enough memory to create flow handle");
6158 /* No multi-thread supporting. */
6159 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[priv->flow_idx++];
6160 dev_flow->handle = dev_handle;
6161 dev_flow->handle_idx = handle_idx;
6163 * In some old rdma-core releases, before continuing, a check of the
6164 * length of matching parameter will be done at first. It needs to use
6165 * the length without misc4 param. If the flow has misc4 support, then
6166 * the length needs to be adjusted accordingly. Each param member is
6167 * aligned with a 64B boundary naturally.
6169 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param) -
6170 MLX5_ST_SZ_BYTES(fte_match_set_misc4);
6172 * The matching value needs to be cleared to 0 before using. In the
6173 * past, it will be automatically cleared when using rte_*alloc
6174 * API. The time consumption will be almost the same as before.
6176 memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
6177 dev_flow->ingress = attr->ingress;
6178 dev_flow->dv.transfer = attr->transfer;
6182 #ifdef RTE_LIBRTE_MLX5_DEBUG
6184 * Sanity check for match mask and value. Similar to check_valid_spec() in
6185 * kernel driver. If unmasked bit is present in value, it returns failure.
6188 * pointer to match mask buffer.
6189 * @param match_value
6190 * pointer to match value buffer.
6193 * 0 if valid, -EINVAL otherwise.
6196 flow_dv_check_valid_spec(void *match_mask, void *match_value)
6198 uint8_t *m = match_mask;
6199 uint8_t *v = match_value;
6202 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
6205 "match_value differs from match_criteria"
6206 " %p[%u] != %p[%u]",
6207 match_value, i, match_mask, i);
6216 * Add match of ip_version.
6220 * @param[in] headers_v
6221 * Values header pointer.
6222 * @param[in] headers_m
6223 * Masks header pointer.
6224 * @param[in] ip_version
6225 * The IP version to set.
6228 flow_dv_set_match_ip_version(uint32_t group,
6234 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
6236 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
6238 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
6239 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
6240 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
6244 * Add Ethernet item to matcher and to the value.
6246 * @param[in, out] matcher
6248 * @param[in, out] key
6249 * Flow matcher value.
6251 * Flow pattern to translate.
6253 * Item is inner pattern.
6256 flow_dv_translate_item_eth(void *matcher, void *key,
6257 const struct rte_flow_item *item, int inner,
6260 const struct rte_flow_item_eth *eth_m = item->mask;
6261 const struct rte_flow_item_eth *eth_v = item->spec;
6262 const struct rte_flow_item_eth nic_mask = {
6263 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6264 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6265 .type = RTE_BE16(0xffff),
6277 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6279 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6281 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6283 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6285 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
6286 ð_m->dst, sizeof(eth_m->dst));
6287 /* The value must be in the range of the mask. */
6288 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
6289 for (i = 0; i < sizeof(eth_m->dst); ++i)
6290 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
6291 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
6292 ð_m->src, sizeof(eth_m->src));
6293 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
6294 /* The value must be in the range of the mask. */
6295 for (i = 0; i < sizeof(eth_m->dst); ++i)
6296 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
6298 /* When ethertype is present set mask for tagged VLAN. */
6299 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6300 /* Set value for tagged VLAN if ethertype is 802.1Q. */
6301 if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
6302 eth_v->type == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
6303 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag,
6305 /* Return here to avoid setting match on ethertype. */
6310 * HW supports match on one Ethertype, the Ethertype following the last
6311 * VLAN tag of the packet (see PRM).
6312 * Set match on ethertype only if ETH header is not followed by VLAN.
6313 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6314 * ethertype, and use ip_version field instead.
6315 * eCPRI over Ether layer will use type value 0xAEFE.
6317 if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
6318 eth_m->type == 0xFFFF) {
6319 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6320 } else if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
6321 eth_m->type == 0xFFFF) {
6322 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6324 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
6325 rte_be_to_cpu_16(eth_m->type));
6326 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6328 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
6333 * Add VLAN item to matcher and to the value.
6335 * @param[in, out] dev_flow
6337 * @param[in, out] matcher
6339 * @param[in, out] key
6340 * Flow matcher value.
6342 * Flow pattern to translate.
6344 * Item is inner pattern.
6347 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
6348 void *matcher, void *key,
6349 const struct rte_flow_item *item,
6350 int inner, uint32_t group)
6352 const struct rte_flow_item_vlan *vlan_m = item->mask;
6353 const struct rte_flow_item_vlan *vlan_v = item->spec;
6360 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6362 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6364 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6366 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6368 * This is workaround, masks are not supported,
6369 * and pre-validated.
6372 dev_flow->handle->vf_vlan.tag =
6373 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
6376 * When VLAN item exists in flow, mark packet as tagged,
6377 * even if TCI is not specified.
6379 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6380 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
6384 vlan_m = &rte_flow_item_vlan_mask;
6385 tci_m = rte_be_to_cpu_16(vlan_m->tci);
6386 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
6387 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
6388 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
6389 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
6390 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
6391 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
6392 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
6394 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6395 * ethertype, and use ip_version field instead.
6397 if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
6398 vlan_m->inner_type == 0xFFFF) {
6399 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6400 } else if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
6401 vlan_m->inner_type == 0xFFFF) {
6402 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6404 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
6405 rte_be_to_cpu_16(vlan_m->inner_type));
6406 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
6407 rte_be_to_cpu_16(vlan_m->inner_type &
6408 vlan_v->inner_type));
6413 * Add IPV4 item to matcher and to the value.
6415 * @param[in, out] matcher
6417 * @param[in, out] key
6418 * Flow matcher value.
6420 * Flow pattern to translate.
6421 * @param[in] item_flags
6422 * Bit-fields that holds the items detected until now.
6424 * Item is inner pattern.
6426 * The group to insert the rule.
6429 flow_dv_translate_item_ipv4(void *matcher, void *key,
6430 const struct rte_flow_item *item,
6431 const uint64_t item_flags,
6432 int inner, uint32_t group)
6434 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
6435 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
6436 const struct rte_flow_item_ipv4 nic_mask = {
6438 .src_addr = RTE_BE32(0xffffffff),
6439 .dst_addr = RTE_BE32(0xffffffff),
6440 .type_of_service = 0xff,
6441 .next_proto_id = 0xff,
6442 .time_to_live = 0xff,
6452 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6454 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6456 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6458 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6460 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6462 * On outer header (which must contains L2), or inner header with L2,
6463 * set cvlan_tag mask bit to mark this packet as untagged.
6464 * This should be done even if item->spec is empty.
6466 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6467 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6472 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6473 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6474 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6475 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6476 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
6477 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
6478 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6479 src_ipv4_src_ipv6.ipv4_layout.ipv4);
6480 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6481 src_ipv4_src_ipv6.ipv4_layout.ipv4);
6482 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
6483 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
6484 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
6485 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
6486 ipv4_m->hdr.type_of_service);
6487 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
6488 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
6489 ipv4_m->hdr.type_of_service >> 2);
6490 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
6491 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6492 ipv4_m->hdr.next_proto_id);
6493 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6494 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
6495 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6496 ipv4_m->hdr.time_to_live);
6497 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6498 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
6499 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
6500 !!(ipv4_m->hdr.fragment_offset));
6501 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
6502 !!(ipv4_v->hdr.fragment_offset & ipv4_m->hdr.fragment_offset));
6506 * Add IPV6 item to matcher and to the value.
6508 * @param[in, out] matcher
6510 * @param[in, out] key
6511 * Flow matcher value.
6513 * Flow pattern to translate.
6514 * @param[in] item_flags
6515 * Bit-fields that holds the items detected until now.
6517 * Item is inner pattern.
6519 * The group to insert the rule.
6522 flow_dv_translate_item_ipv6(void *matcher, void *key,
6523 const struct rte_flow_item *item,
6524 const uint64_t item_flags,
6525 int inner, uint32_t group)
6527 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
6528 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
6529 const struct rte_flow_item_ipv6 nic_mask = {
6532 "\xff\xff\xff\xff\xff\xff\xff\xff"
6533 "\xff\xff\xff\xff\xff\xff\xff\xff",
6535 "\xff\xff\xff\xff\xff\xff\xff\xff"
6536 "\xff\xff\xff\xff\xff\xff\xff\xff",
6537 .vtc_flow = RTE_BE32(0xffffffff),
6544 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6545 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6554 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6556 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6558 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6560 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6562 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6564 * On outer header (which must contains L2), or inner header with L2,
6565 * set cvlan_tag mask bit to mark this packet as untagged.
6566 * This should be done even if item->spec is empty.
6568 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6569 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6574 size = sizeof(ipv6_m->hdr.dst_addr);
6575 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6576 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6577 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6578 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6579 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
6580 for (i = 0; i < size; ++i)
6581 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
6582 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6583 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6584 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6585 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6586 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
6587 for (i = 0; i < size; ++i)
6588 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
6590 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
6591 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
6592 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
6593 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
6594 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
6595 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
6598 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
6600 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
6603 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
6605 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
6609 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6611 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6612 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
6614 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6615 ipv6_m->hdr.hop_limits);
6616 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6617 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
6618 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
6619 !!(ipv6_m->has_frag_ext));
6620 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
6621 !!(ipv6_v->has_frag_ext & ipv6_m->has_frag_ext));
6625 * Add IPV6 fragment extension item to matcher and to the value.
6627 * @param[in, out] matcher
6629 * @param[in, out] key
6630 * Flow matcher value.
6632 * Flow pattern to translate.
6634 * Item is inner pattern.
6637 flow_dv_translate_item_ipv6_frag_ext(void *matcher, void *key,
6638 const struct rte_flow_item *item,
6641 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_m = item->mask;
6642 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_v = item->spec;
6643 const struct rte_flow_item_ipv6_frag_ext nic_mask = {
6645 .next_header = 0xff,
6646 .frag_data = RTE_BE16(0xffff),
6653 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6655 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6657 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6659 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6661 /* IPv6 fragment extension item exists, so packet is IP fragment. */
6662 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6663 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 1);
6664 if (!ipv6_frag_ext_v)
6666 if (!ipv6_frag_ext_m)
6667 ipv6_frag_ext_m = &nic_mask;
6668 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6669 ipv6_frag_ext_m->hdr.next_header);
6670 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6671 ipv6_frag_ext_v->hdr.next_header &
6672 ipv6_frag_ext_m->hdr.next_header);
6676 * Add TCP item to matcher and to the value.
6678 * @param[in, out] matcher
6680 * @param[in, out] key
6681 * Flow matcher value.
6683 * Flow pattern to translate.
6685 * Item is inner pattern.
6688 flow_dv_translate_item_tcp(void *matcher, void *key,
6689 const struct rte_flow_item *item,
6692 const struct rte_flow_item_tcp *tcp_m = item->mask;
6693 const struct rte_flow_item_tcp *tcp_v = item->spec;
6698 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6700 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6702 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6704 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6706 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6707 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
6711 tcp_m = &rte_flow_item_tcp_mask;
6712 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
6713 rte_be_to_cpu_16(tcp_m->hdr.src_port));
6714 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
6715 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
6716 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
6717 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
6718 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
6719 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
6720 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
6721 tcp_m->hdr.tcp_flags);
6722 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
6723 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
6727 * Add UDP item to matcher and to the value.
6729 * @param[in, out] matcher
6731 * @param[in, out] key
6732 * Flow matcher value.
6734 * Flow pattern to translate.
6736 * Item is inner pattern.
6739 flow_dv_translate_item_udp(void *matcher, void *key,
6740 const struct rte_flow_item *item,
6743 const struct rte_flow_item_udp *udp_m = item->mask;
6744 const struct rte_flow_item_udp *udp_v = item->spec;
6749 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6751 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6753 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6755 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6757 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6758 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
6762 udp_m = &rte_flow_item_udp_mask;
6763 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
6764 rte_be_to_cpu_16(udp_m->hdr.src_port));
6765 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
6766 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
6767 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
6768 rte_be_to_cpu_16(udp_m->hdr.dst_port));
6769 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6770 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
6774 * Add GRE optional Key item to matcher and to the value.
6776 * @param[in, out] matcher
6778 * @param[in, out] key
6779 * Flow matcher value.
6781 * Flow pattern to translate.
6783 * Item is inner pattern.
6786 flow_dv_translate_item_gre_key(void *matcher, void *key,
6787 const struct rte_flow_item *item)
6789 const rte_be32_t *key_m = item->mask;
6790 const rte_be32_t *key_v = item->spec;
6791 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6792 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6793 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
6795 /* GRE K bit must be on and should already be validated */
6796 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
6797 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
6801 key_m = &gre_key_default_mask;
6802 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
6803 rte_be_to_cpu_32(*key_m) >> 8);
6804 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
6805 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
6806 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
6807 rte_be_to_cpu_32(*key_m) & 0xFF);
6808 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
6809 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
6813 * Add GRE item to matcher and to the value.
6815 * @param[in, out] matcher
6817 * @param[in, out] key
6818 * Flow matcher value.
6820 * Flow pattern to translate.
6822 * Item is inner pattern.
6825 flow_dv_translate_item_gre(void *matcher, void *key,
6826 const struct rte_flow_item *item,
6829 const struct rte_flow_item_gre *gre_m = item->mask;
6830 const struct rte_flow_item_gre *gre_v = item->spec;
6833 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6834 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6841 uint16_t s_present:1;
6842 uint16_t k_present:1;
6843 uint16_t rsvd_bit1:1;
6844 uint16_t c_present:1;
6848 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
6851 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6853 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6855 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6857 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6859 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6860 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
6864 gre_m = &rte_flow_item_gre_mask;
6865 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
6866 rte_be_to_cpu_16(gre_m->protocol));
6867 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6868 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
6869 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
6870 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
6871 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
6872 gre_crks_rsvd0_ver_m.c_present);
6873 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6874 gre_crks_rsvd0_ver_v.c_present &
6875 gre_crks_rsvd0_ver_m.c_present);
6876 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6877 gre_crks_rsvd0_ver_m.k_present);
6878 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6879 gre_crks_rsvd0_ver_v.k_present &
6880 gre_crks_rsvd0_ver_m.k_present);
6881 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6882 gre_crks_rsvd0_ver_m.s_present);
6883 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6884 gre_crks_rsvd0_ver_v.s_present &
6885 gre_crks_rsvd0_ver_m.s_present);
6889 * Add NVGRE item to matcher and to the value.
6891 * @param[in, out] matcher
6893 * @param[in, out] key
6894 * Flow matcher value.
6896 * Flow pattern to translate.
6898 * Item is inner pattern.
6901 flow_dv_translate_item_nvgre(void *matcher, void *key,
6902 const struct rte_flow_item *item,
6905 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6906 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6907 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6908 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6909 const char *tni_flow_id_m;
6910 const char *tni_flow_id_v;
6916 /* For NVGRE, GRE header fields must be set with defined values. */
6917 const struct rte_flow_item_gre gre_spec = {
6918 .c_rsvd0_ver = RTE_BE16(0x2000),
6919 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6921 const struct rte_flow_item_gre gre_mask = {
6922 .c_rsvd0_ver = RTE_BE16(0xB000),
6923 .protocol = RTE_BE16(UINT16_MAX),
6925 const struct rte_flow_item gre_item = {
6930 flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6934 nvgre_m = &rte_flow_item_nvgre_mask;
6935 tni_flow_id_m = (const char *)nvgre_m->tni;
6936 tni_flow_id_v = (const char *)nvgre_v->tni;
6937 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6938 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6939 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6940 memcpy(gre_key_m, tni_flow_id_m, size);
6941 for (i = 0; i < size; ++i)
6942 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6946 * Add VXLAN item to matcher and to the value.
6948 * @param[in, out] matcher
6950 * @param[in, out] key
6951 * Flow matcher value.
6953 * Flow pattern to translate.
6955 * Item is inner pattern.
6958 flow_dv_translate_item_vxlan(void *matcher, void *key,
6959 const struct rte_flow_item *item,
6962 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
6963 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
6966 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6967 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6975 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6977 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6979 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6981 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6983 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6984 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6985 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6986 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6987 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6992 vxlan_m = &rte_flow_item_vxlan_mask;
6993 size = sizeof(vxlan_m->vni);
6994 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
6995 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
6996 memcpy(vni_m, vxlan_m->vni, size);
6997 for (i = 0; i < size; ++i)
6998 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7002 * Add VXLAN-GPE item to matcher and to the value.
7004 * @param[in, out] matcher
7006 * @param[in, out] key
7007 * Flow matcher value.
7009 * Flow pattern to translate.
7011 * Item is inner pattern.
7015 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
7016 const struct rte_flow_item *item, int inner)
7018 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
7019 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
7023 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
7025 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7031 uint8_t flags_m = 0xff;
7032 uint8_t flags_v = 0xc;
7035 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7037 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7039 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7041 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7043 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
7044 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
7045 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7046 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7047 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7052 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
7053 size = sizeof(vxlan_m->vni);
7054 vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
7055 vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
7056 memcpy(vni_m, vxlan_m->vni, size);
7057 for (i = 0; i < size; ++i)
7058 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7059 if (vxlan_m->flags) {
7060 flags_m = vxlan_m->flags;
7061 flags_v = vxlan_v->flags;
7063 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
7064 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
7065 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
7067 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
7072 * Add Geneve item to matcher and to the value.
7074 * @param[in, out] matcher
7076 * @param[in, out] key
7077 * Flow matcher value.
7079 * Flow pattern to translate.
7081 * Item is inner pattern.
7085 flow_dv_translate_item_geneve(void *matcher, void *key,
7086 const struct rte_flow_item *item, int inner)
7088 const struct rte_flow_item_geneve *geneve_m = item->mask;
7089 const struct rte_flow_item_geneve *geneve_v = item->spec;
7092 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7093 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7102 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7104 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7106 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7108 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7110 dport = MLX5_UDP_PORT_GENEVE;
7111 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7112 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7113 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7118 geneve_m = &rte_flow_item_geneve_mask;
7119 size = sizeof(geneve_m->vni);
7120 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
7121 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
7122 memcpy(vni_m, geneve_m->vni, size);
7123 for (i = 0; i < size; ++i)
7124 vni_v[i] = vni_m[i] & geneve_v->vni[i];
7125 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
7126 rte_be_to_cpu_16(geneve_m->protocol));
7127 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
7128 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
7129 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
7130 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
7131 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
7132 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7133 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
7134 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7135 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
7136 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7137 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
7138 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
7139 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7143 * Add MPLS item to matcher and to the value.
7145 * @param[in, out] matcher
7147 * @param[in, out] key
7148 * Flow matcher value.
7150 * Flow pattern to translate.
7151 * @param[in] prev_layer
7152 * The protocol layer indicated in previous item.
7154 * Item is inner pattern.
7157 flow_dv_translate_item_mpls(void *matcher, void *key,
7158 const struct rte_flow_item *item,
7159 uint64_t prev_layer,
7162 const uint32_t *in_mpls_m = item->mask;
7163 const uint32_t *in_mpls_v = item->spec;
7164 uint32_t *out_mpls_m = 0;
7165 uint32_t *out_mpls_v = 0;
7166 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7167 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7168 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
7170 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7171 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
7172 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7174 switch (prev_layer) {
7175 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7176 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
7177 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
7178 MLX5_UDP_PORT_MPLS);
7180 case MLX5_FLOW_LAYER_GRE:
7181 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
7182 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
7183 RTE_ETHER_TYPE_MPLS);
7186 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
7187 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
7194 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
7195 switch (prev_layer) {
7196 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7198 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7199 outer_first_mpls_over_udp);
7201 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7202 outer_first_mpls_over_udp);
7204 case MLX5_FLOW_LAYER_GRE:
7206 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7207 outer_first_mpls_over_gre);
7209 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7210 outer_first_mpls_over_gre);
7213 /* Inner MPLS not over GRE is not supported. */
7216 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7220 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7226 if (out_mpls_m && out_mpls_v) {
7227 *out_mpls_m = *in_mpls_m;
7228 *out_mpls_v = *in_mpls_v & *in_mpls_m;
7233 * Add metadata register item to matcher
7235 * @param[in, out] matcher
7237 * @param[in, out] key
7238 * Flow matcher value.
7239 * @param[in] reg_type
7240 * Type of device metadata register
7247 flow_dv_match_meta_reg(void *matcher, void *key,
7248 enum modify_reg reg_type,
7249 uint32_t data, uint32_t mask)
7252 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
7254 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7260 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
7261 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
7264 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
7265 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
7269 * The metadata register C0 field might be divided into
7270 * source vport index and META item value, we should set
7271 * this field according to specified mask, not as whole one.
7273 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
7275 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
7276 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
7279 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
7282 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
7283 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
7286 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
7287 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
7290 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
7291 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
7294 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
7295 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
7298 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
7299 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
7302 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
7303 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
7306 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
7307 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
7316 * Add MARK item to matcher
7319 * The device to configure through.
7320 * @param[in, out] matcher
7322 * @param[in, out] key
7323 * Flow matcher value.
7325 * Flow pattern to translate.
7328 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
7329 void *matcher, void *key,
7330 const struct rte_flow_item *item)
7332 struct mlx5_priv *priv = dev->data->dev_private;
7333 const struct rte_flow_item_mark *mark;
7337 mark = item->mask ? (const void *)item->mask :
7338 &rte_flow_item_mark_mask;
7339 mask = mark->id & priv->sh->dv_mark_mask;
7340 mark = (const void *)item->spec;
7342 value = mark->id & priv->sh->dv_mark_mask & mask;
7344 enum modify_reg reg;
7346 /* Get the metadata register index for the mark. */
7347 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
7348 MLX5_ASSERT(reg > 0);
7349 if (reg == REG_C_0) {
7350 struct mlx5_priv *priv = dev->data->dev_private;
7351 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7352 uint32_t shl_c0 = rte_bsf32(msk_c0);
7358 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7363 * Add META item to matcher
7366 * The devich to configure through.
7367 * @param[in, out] matcher
7369 * @param[in, out] key
7370 * Flow matcher value.
7372 * Attributes of flow that includes this item.
7374 * Flow pattern to translate.
7377 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
7378 void *matcher, void *key,
7379 const struct rte_flow_attr *attr,
7380 const struct rte_flow_item *item)
7382 const struct rte_flow_item_meta *meta_m;
7383 const struct rte_flow_item_meta *meta_v;
7385 meta_m = (const void *)item->mask;
7387 meta_m = &rte_flow_item_meta_mask;
7388 meta_v = (const void *)item->spec;
7391 uint32_t value = meta_v->data;
7392 uint32_t mask = meta_m->data;
7394 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
7398 * In datapath code there is no endianness
7399 * coversions for perfromance reasons, all
7400 * pattern conversions are done in rte_flow.
7402 value = rte_cpu_to_be_32(value);
7403 mask = rte_cpu_to_be_32(mask);
7404 if (reg == REG_C_0) {
7405 struct mlx5_priv *priv = dev->data->dev_private;
7406 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7407 uint32_t shl_c0 = rte_bsf32(msk_c0);
7408 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
7409 uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
7416 MLX5_ASSERT(msk_c0);
7417 MLX5_ASSERT(!(~msk_c0 & mask));
7419 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7424 * Add vport metadata Reg C0 item to matcher
7426 * @param[in, out] matcher
7428 * @param[in, out] key
7429 * Flow matcher value.
7431 * Flow pattern to translate.
7434 flow_dv_translate_item_meta_vport(void *matcher, void *key,
7435 uint32_t value, uint32_t mask)
7437 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
7441 * Add tag item to matcher
7444 * The devich to configure through.
7445 * @param[in, out] matcher
7447 * @param[in, out] key
7448 * Flow matcher value.
7450 * Flow pattern to translate.
7453 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
7454 void *matcher, void *key,
7455 const struct rte_flow_item *item)
7457 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
7458 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
7459 uint32_t mask, value;
7462 value = tag_v->data;
7463 mask = tag_m ? tag_m->data : UINT32_MAX;
7464 if (tag_v->id == REG_C_0) {
7465 struct mlx5_priv *priv = dev->data->dev_private;
7466 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7467 uint32_t shl_c0 = rte_bsf32(msk_c0);
7473 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
7477 * Add TAG item to matcher
7480 * The devich to configure through.
7481 * @param[in, out] matcher
7483 * @param[in, out] key
7484 * Flow matcher value.
7486 * Flow pattern to translate.
7489 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
7490 void *matcher, void *key,
7491 const struct rte_flow_item *item)
7493 const struct rte_flow_item_tag *tag_v = item->spec;
7494 const struct rte_flow_item_tag *tag_m = item->mask;
7495 enum modify_reg reg;
7498 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
7499 /* Get the metadata register index for the tag. */
7500 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
7501 MLX5_ASSERT(reg > 0);
7502 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
7506 * Add source vport match to the specified matcher.
7508 * @param[in, out] matcher
7510 * @param[in, out] key
7511 * Flow matcher value.
7513 * Source vport value to match
7518 flow_dv_translate_item_source_vport(void *matcher, void *key,
7519 int16_t port, uint16_t mask)
7521 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7522 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7524 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
7525 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
7529 * Translate port-id item to eswitch match on port-id.
7532 * The devich to configure through.
7533 * @param[in, out] matcher
7535 * @param[in, out] key
7536 * Flow matcher value.
7538 * Flow pattern to translate.
7541 * 0 on success, a negative errno value otherwise.
7544 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
7545 void *key, const struct rte_flow_item *item)
7547 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
7548 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
7549 struct mlx5_priv *priv;
7552 mask = pid_m ? pid_m->id : 0xffff;
7553 id = pid_v ? pid_v->id : dev->data->port_id;
7554 priv = mlx5_port_to_eswitch_info(id, item == NULL);
7557 /* Translate to vport field or to metadata, depending on mode. */
7558 if (priv->vport_meta_mask)
7559 flow_dv_translate_item_meta_vport(matcher, key,
7560 priv->vport_meta_tag,
7561 priv->vport_meta_mask);
7563 flow_dv_translate_item_source_vport(matcher, key,
7564 priv->vport_id, mask);
7569 * Add ICMP6 item to matcher and to the value.
7571 * @param[in, out] matcher
7573 * @param[in, out] key
7574 * Flow matcher value.
7576 * Flow pattern to translate.
7578 * Item is inner pattern.
7581 flow_dv_translate_item_icmp6(void *matcher, void *key,
7582 const struct rte_flow_item *item,
7585 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
7586 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
7589 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7591 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7593 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7595 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7597 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7599 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7601 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7602 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
7606 icmp6_m = &rte_flow_item_icmp6_mask;
7607 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
7608 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
7609 icmp6_v->type & icmp6_m->type);
7610 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
7611 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
7612 icmp6_v->code & icmp6_m->code);
7616 * Add ICMP item to matcher and to the value.
7618 * @param[in, out] matcher
7620 * @param[in, out] key
7621 * Flow matcher value.
7623 * Flow pattern to translate.
7625 * Item is inner pattern.
7628 flow_dv_translate_item_icmp(void *matcher, void *key,
7629 const struct rte_flow_item *item,
7632 const struct rte_flow_item_icmp *icmp_m = item->mask;
7633 const struct rte_flow_item_icmp *icmp_v = item->spec;
7634 uint32_t icmp_header_data_m = 0;
7635 uint32_t icmp_header_data_v = 0;
7638 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7640 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7642 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7644 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7646 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7648 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7650 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7651 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
7655 icmp_m = &rte_flow_item_icmp_mask;
7656 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
7657 icmp_m->hdr.icmp_type);
7658 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
7659 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
7660 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
7661 icmp_m->hdr.icmp_code);
7662 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
7663 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
7664 icmp_header_data_m = rte_be_to_cpu_16(icmp_m->hdr.icmp_seq_nb);
7665 icmp_header_data_m |= rte_be_to_cpu_16(icmp_m->hdr.icmp_ident) << 16;
7666 if (icmp_header_data_m) {
7667 icmp_header_data_v = rte_be_to_cpu_16(icmp_v->hdr.icmp_seq_nb);
7668 icmp_header_data_v |=
7669 rte_be_to_cpu_16(icmp_v->hdr.icmp_ident) << 16;
7670 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_header_data,
7671 icmp_header_data_m);
7672 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_header_data,
7673 icmp_header_data_v & icmp_header_data_m);
7678 * Add GTP item to matcher and to the value.
7680 * @param[in, out] matcher
7682 * @param[in, out] key
7683 * Flow matcher value.
7685 * Flow pattern to translate.
7687 * Item is inner pattern.
7690 flow_dv_translate_item_gtp(void *matcher, void *key,
7691 const struct rte_flow_item *item, int inner)
7693 const struct rte_flow_item_gtp *gtp_m = item->mask;
7694 const struct rte_flow_item_gtp *gtp_v = item->spec;
7697 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7699 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7700 uint16_t dport = RTE_GTPU_UDP_PORT;
7703 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7705 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7707 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7709 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7711 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7712 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7713 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7718 gtp_m = &rte_flow_item_gtp_mask;
7719 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
7720 gtp_m->v_pt_rsv_flags);
7721 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
7722 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
7723 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
7724 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
7725 gtp_v->msg_type & gtp_m->msg_type);
7726 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
7727 rte_be_to_cpu_32(gtp_m->teid));
7728 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
7729 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
7733 * Add eCPRI item to matcher and to the value.
7736 * The devich to configure through.
7737 * @param[in, out] matcher
7739 * @param[in, out] key
7740 * Flow matcher value.
7742 * Flow pattern to translate.
7743 * @param[in] samples
7744 * Sample IDs to be used in the matching.
7747 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
7748 void *key, const struct rte_flow_item *item)
7750 struct mlx5_priv *priv = dev->data->dev_private;
7751 const struct rte_flow_item_ecpri *ecpri_m = item->mask;
7752 const struct rte_flow_item_ecpri *ecpri_v = item->spec;
7753 void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
7755 void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
7763 ecpri_m = &rte_flow_item_ecpri_mask;
7765 * Maximal four DW samples are supported in a single matching now.
7766 * Two are used now for a eCPRI matching:
7767 * 1. Type: one byte, mask should be 0x00ff0000 in network order
7768 * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
7771 if (!ecpri_m->hdr.common.u32)
7773 samples = priv->sh->fp[MLX5_FLEX_PARSER_ECPRI_0].ids;
7774 /* Need to take the whole DW as the mask to fill the entry. */
7775 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7776 prog_sample_field_value_0);
7777 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7778 prog_sample_field_value_0);
7779 /* Already big endian (network order) in the header. */
7780 *(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
7781 *(uint32_t *)dw_v = ecpri_v->hdr.common.u32;
7782 /* Sample#0, used for matching type, offset 0. */
7783 MLX5_SET(fte_match_set_misc4, misc4_m,
7784 prog_sample_field_id_0, samples[0]);
7785 /* It makes no sense to set the sample ID in the mask field. */
7786 MLX5_SET(fte_match_set_misc4, misc4_v,
7787 prog_sample_field_id_0, samples[0]);
7789 * Checking if message body part needs to be matched.
7790 * Some wildcard rules only matching type field should be supported.
7792 if (ecpri_m->hdr.dummy[0]) {
7793 switch (ecpri_v->hdr.common.type) {
7794 case RTE_ECPRI_MSG_TYPE_IQ_DATA:
7795 case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
7796 case RTE_ECPRI_MSG_TYPE_DLY_MSR:
7797 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7798 prog_sample_field_value_1);
7799 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7800 prog_sample_field_value_1);
7801 *(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
7802 *(uint32_t *)dw_v = ecpri_v->hdr.dummy[0];
7803 /* Sample#1, to match message body, offset 4. */
7804 MLX5_SET(fte_match_set_misc4, misc4_m,
7805 prog_sample_field_id_1, samples[1]);
7806 MLX5_SET(fte_match_set_misc4, misc4_v,
7807 prog_sample_field_id_1, samples[1]);
7810 /* Others, do not match any sample ID. */
7816 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
7818 #define HEADER_IS_ZERO(match_criteria, headers) \
7819 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
7820 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
7823 * Calculate flow matcher enable bitmap.
7825 * @param match_criteria
7826 * Pointer to flow matcher criteria.
7829 * Bitmap of enabled fields.
7832 flow_dv_matcher_enable(uint32_t *match_criteria)
7834 uint8_t match_criteria_enable;
7836 match_criteria_enable =
7837 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
7838 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
7839 match_criteria_enable |=
7840 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
7841 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
7842 match_criteria_enable |=
7843 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
7844 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
7845 match_criteria_enable |=
7846 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
7847 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
7848 match_criteria_enable |=
7849 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
7850 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
7851 match_criteria_enable |=
7852 (!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
7853 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
7854 return match_criteria_enable;
7861 * @param[in, out] dev
7862 * Pointer to rte_eth_dev structure.
7863 * @param[in] table_id
7866 * Direction of the table.
7867 * @param[in] transfer
7868 * E-Switch or NIC flow.
7870 * pointer to error structure.
7873 * Returns tables resource based on the index, NULL in case of failed.
7875 static struct mlx5_flow_tbl_resource *
7876 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
7877 uint32_t table_id, uint8_t egress,
7879 struct rte_flow_error *error)
7881 struct mlx5_priv *priv = dev->data->dev_private;
7882 struct mlx5_dev_ctx_shared *sh = priv->sh;
7883 struct mlx5_flow_tbl_resource *tbl;
7884 union mlx5_flow_tbl_key table_key = {
7886 .table_id = table_id,
7888 .domain = !!transfer,
7889 .direction = !!egress,
7892 struct mlx5_hlist_entry *pos = mlx5_hlist_lookup(sh->flow_tbls,
7894 struct mlx5_flow_tbl_data_entry *tbl_data;
7900 tbl_data = container_of(pos, struct mlx5_flow_tbl_data_entry,
7902 tbl = &tbl_data->tbl;
7903 rte_atomic32_inc(&tbl->refcnt);
7906 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
7908 rte_flow_error_set(error, ENOMEM,
7909 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7911 "cannot allocate flow table data entry");
7914 tbl_data->idx = idx;
7915 tbl = &tbl_data->tbl;
7916 pos = &tbl_data->entry;
7918 domain = sh->fdb_domain;
7920 domain = sh->tx_domain;
7922 domain = sh->rx_domain;
7923 ret = mlx5_flow_os_create_flow_tbl(domain, table_id, &tbl->obj);
7925 rte_flow_error_set(error, ENOMEM,
7926 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7927 NULL, "cannot create flow table object");
7928 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7932 * No multi-threads now, but still better to initialize the reference
7933 * count before insert it into the hash list.
7935 rte_atomic32_init(&tbl->refcnt);
7936 /* Jump action reference count is initialized here. */
7937 rte_atomic32_init(&tbl_data->jump.refcnt);
7938 pos->key = table_key.v64;
7939 ret = mlx5_hlist_insert(sh->flow_tbls, pos);
7941 rte_flow_error_set(error, -ret,
7942 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7943 "cannot insert flow table data entry");
7944 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7945 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7947 rte_atomic32_inc(&tbl->refcnt);
7952 * Release a flow table.
7955 * Pointer to rte_eth_dev structure.
7957 * Table resource to be released.
7960 * Returns 0 if table was released, else return 1;
7963 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
7964 struct mlx5_flow_tbl_resource *tbl)
7966 struct mlx5_priv *priv = dev->data->dev_private;
7967 struct mlx5_dev_ctx_shared *sh = priv->sh;
7968 struct mlx5_flow_tbl_data_entry *tbl_data =
7969 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7973 if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
7974 struct mlx5_hlist_entry *pos = &tbl_data->entry;
7976 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7978 /* remove the entry from the hash list and free memory. */
7979 mlx5_hlist_remove(sh->flow_tbls, pos);
7980 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_JUMP],
7988 * Register the flow matcher.
7990 * @param[in, out] dev
7991 * Pointer to rte_eth_dev structure.
7992 * @param[in, out] matcher
7993 * Pointer to flow matcher.
7994 * @param[in, out] key
7995 * Pointer to flow table key.
7996 * @parm[in, out] dev_flow
7997 * Pointer to the dev_flow.
7999 * pointer to error structure.
8002 * 0 on success otherwise -errno and errno is set.
8005 flow_dv_matcher_register(struct rte_eth_dev *dev,
8006 struct mlx5_flow_dv_matcher *matcher,
8007 union mlx5_flow_tbl_key *key,
8008 struct mlx5_flow *dev_flow,
8009 struct rte_flow_error *error)
8011 struct mlx5_priv *priv = dev->data->dev_private;
8012 struct mlx5_dev_ctx_shared *sh = priv->sh;
8013 struct mlx5_flow_dv_matcher *cache_matcher;
8014 struct mlx5dv_flow_matcher_attr dv_attr = {
8015 .type = IBV_FLOW_ATTR_NORMAL,
8016 .match_mask = (void *)&matcher->mask,
8018 struct mlx5_flow_tbl_resource *tbl;
8019 struct mlx5_flow_tbl_data_entry *tbl_data;
8022 tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
8023 key->domain, error);
8025 return -rte_errno; /* No need to refill the error info */
8026 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
8027 /* Lookup from cache. */
8028 LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
8029 if (matcher->crc == cache_matcher->crc &&
8030 matcher->priority == cache_matcher->priority &&
8031 !memcmp((const void *)matcher->mask.buf,
8032 (const void *)cache_matcher->mask.buf,
8033 cache_matcher->mask.size)) {
8035 "%s group %u priority %hd use %s "
8036 "matcher %p: refcnt %d++",
8037 key->domain ? "FDB" : "NIC", key->table_id,
8038 cache_matcher->priority,
8039 key->direction ? "tx" : "rx",
8040 (void *)cache_matcher,
8041 rte_atomic32_read(&cache_matcher->refcnt));
8042 rte_atomic32_inc(&cache_matcher->refcnt);
8043 dev_flow->handle->dvh.matcher = cache_matcher;
8044 /* old matcher should not make the table ref++. */
8045 flow_dv_tbl_resource_release(dev, tbl);
8049 /* Register new matcher. */
8050 cache_matcher = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*cache_matcher), 0,
8052 if (!cache_matcher) {
8053 flow_dv_tbl_resource_release(dev, tbl);
8054 return rte_flow_error_set(error, ENOMEM,
8055 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8056 "cannot allocate matcher memory");
8058 *cache_matcher = *matcher;
8059 dv_attr.match_criteria_enable =
8060 flow_dv_matcher_enable(cache_matcher->mask.buf);
8061 dv_attr.priority = matcher->priority;
8063 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
8064 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
8065 &cache_matcher->matcher_object);
8067 mlx5_free(cache_matcher);
8068 #ifdef HAVE_MLX5DV_DR
8069 flow_dv_tbl_resource_release(dev, tbl);
8071 return rte_flow_error_set(error, ENOMEM,
8072 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8073 NULL, "cannot create matcher");
8075 /* Save the table information */
8076 cache_matcher->tbl = tbl;
8077 rte_atomic32_init(&cache_matcher->refcnt);
8078 /* only matcher ref++, table ref++ already done above in get API. */
8079 rte_atomic32_inc(&cache_matcher->refcnt);
8080 LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
8081 dev_flow->handle->dvh.matcher = cache_matcher;
8082 DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
8083 key->domain ? "FDB" : "NIC", key->table_id,
8084 cache_matcher->priority,
8085 key->direction ? "tx" : "rx", (void *)cache_matcher,
8086 rte_atomic32_read(&cache_matcher->refcnt));
8091 * Find existing tag resource or create and register a new one.
8093 * @param dev[in, out]
8094 * Pointer to rte_eth_dev structure.
8095 * @param[in, out] tag_be24
8096 * Tag value in big endian then R-shift 8.
8097 * @parm[in, out] dev_flow
8098 * Pointer to the dev_flow.
8100 * pointer to error structure.
8103 * 0 on success otherwise -errno and errno is set.
8106 flow_dv_tag_resource_register
8107 (struct rte_eth_dev *dev,
8109 struct mlx5_flow *dev_flow,
8110 struct rte_flow_error *error)
8112 struct mlx5_priv *priv = dev->data->dev_private;
8113 struct mlx5_dev_ctx_shared *sh = priv->sh;
8114 struct mlx5_flow_dv_tag_resource *cache_resource;
8115 struct mlx5_hlist_entry *entry;
8118 /* Lookup a matching resource from cache. */
8119 entry = mlx5_hlist_lookup(sh->tag_table, (uint64_t)tag_be24);
8121 cache_resource = container_of
8122 (entry, struct mlx5_flow_dv_tag_resource, entry);
8123 rte_atomic32_inc(&cache_resource->refcnt);
8124 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
8125 dev_flow->dv.tag_resource = cache_resource;
8126 DRV_LOG(DEBUG, "cached tag resource %p: refcnt now %d++",
8127 (void *)cache_resource,
8128 rte_atomic32_read(&cache_resource->refcnt));
8131 /* Register new resource. */
8132 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG],
8133 &dev_flow->handle->dvh.rix_tag);
8134 if (!cache_resource)
8135 return rte_flow_error_set(error, ENOMEM,
8136 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8137 "cannot allocate resource memory");
8138 cache_resource->entry.key = (uint64_t)tag_be24;
8139 ret = mlx5_flow_os_create_flow_action_tag(tag_be24,
8140 &cache_resource->action);
8142 mlx5_free(cache_resource);
8143 return rte_flow_error_set(error, ENOMEM,
8144 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8145 NULL, "cannot create action");
8147 rte_atomic32_init(&cache_resource->refcnt);
8148 rte_atomic32_inc(&cache_resource->refcnt);
8149 if (mlx5_hlist_insert(sh->tag_table, &cache_resource->entry)) {
8150 mlx5_flow_os_destroy_flow_action(cache_resource->action);
8151 mlx5_free(cache_resource);
8152 return rte_flow_error_set(error, EEXIST,
8153 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8154 NULL, "cannot insert tag");
8156 dev_flow->dv.tag_resource = cache_resource;
8157 DRV_LOG(DEBUG, "new tag resource %p: refcnt now %d++",
8158 (void *)cache_resource,
8159 rte_atomic32_read(&cache_resource->refcnt));
8167 * Pointer to Ethernet device.
8172 * 1 while a reference on it exists, 0 when freed.
8175 flow_dv_tag_release(struct rte_eth_dev *dev,
8178 struct mlx5_priv *priv = dev->data->dev_private;
8179 struct mlx5_dev_ctx_shared *sh = priv->sh;
8180 struct mlx5_flow_dv_tag_resource *tag;
8182 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
8185 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
8186 dev->data->port_id, (void *)tag,
8187 rte_atomic32_read(&tag->refcnt));
8188 if (rte_atomic32_dec_and_test(&tag->refcnt)) {
8189 claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
8190 mlx5_hlist_remove(sh->tag_table, &tag->entry);
8191 DRV_LOG(DEBUG, "port %u tag %p: removed",
8192 dev->data->port_id, (void *)tag);
8193 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
8200 * Translate port ID action to vport.
8203 * Pointer to rte_eth_dev structure.
8205 * Pointer to the port ID action.
8206 * @param[out] dst_port_id
8207 * The target port ID.
8209 * Pointer to the error structure.
8212 * 0 on success, a negative errno value otherwise and rte_errno is set.
8215 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
8216 const struct rte_flow_action *action,
8217 uint32_t *dst_port_id,
8218 struct rte_flow_error *error)
8221 struct mlx5_priv *priv;
8222 const struct rte_flow_action_port_id *conf =
8223 (const struct rte_flow_action_port_id *)action->conf;
8225 port = conf->original ? dev->data->port_id : conf->id;
8226 priv = mlx5_port_to_eswitch_info(port, false);
8228 return rte_flow_error_set(error, -rte_errno,
8229 RTE_FLOW_ERROR_TYPE_ACTION,
8231 "No eswitch info was found for port");
8232 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
8234 * This parameter is transferred to
8235 * mlx5dv_dr_action_create_dest_ib_port().
8237 *dst_port_id = priv->dev_port;
8240 * Legacy mode, no LAG configurations is supported.
8241 * This parameter is transferred to
8242 * mlx5dv_dr_action_create_dest_vport().
8244 *dst_port_id = priv->vport_id;
8250 * Create a counter with aging configuration.
8253 * Pointer to rte_eth_dev structure.
8255 * Pointer to the counter action configuration.
8257 * Pointer to the aging action configuration.
8260 * Index to flow counter on success, 0 otherwise.
8263 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
8264 struct mlx5_flow *dev_flow,
8265 const struct rte_flow_action_count *count,
8266 const struct rte_flow_action_age *age)
8269 struct mlx5_age_param *age_param;
8271 if (count && count->shared)
8272 counter = flow_dv_counter_get_shared(dev, count->id);
8274 counter = flow_dv_counter_alloc(dev, !!age);
8275 if (!counter || age == NULL)
8277 age_param = flow_dv_counter_idx_get_age(dev, counter);
8278 age_param->context = age->context ? age->context :
8279 (void *)(uintptr_t)(dev_flow->flow_idx);
8280 age_param->timeout = age->timeout;
8281 age_param->port_id = dev->data->port_id;
8282 __atomic_store_n(&age_param->sec_since_last_hit, 0, __ATOMIC_RELAXED);
8283 __atomic_store_n(&age_param->state, AGE_CANDIDATE, __ATOMIC_RELAXED);
8287 * Add Tx queue matcher
8290 * Pointer to the dev struct.
8291 * @param[in, out] matcher
8293 * @param[in, out] key
8294 * Flow matcher value.
8296 * Flow pattern to translate.
8298 * Item is inner pattern.
8301 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
8302 void *matcher, void *key,
8303 const struct rte_flow_item *item)
8305 const struct mlx5_rte_flow_item_tx_queue *queue_m;
8306 const struct mlx5_rte_flow_item_tx_queue *queue_v;
8308 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8310 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8311 struct mlx5_txq_ctrl *txq;
8315 queue_m = (const void *)item->mask;
8318 queue_v = (const void *)item->spec;
8321 txq = mlx5_txq_get(dev, queue_v->queue);
8324 queue = txq->obj->sq->id;
8325 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
8326 MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
8327 queue & queue_m->queue);
8328 mlx5_txq_release(dev, queue_v->queue);
8332 * Set the hash fields according to the @p flow information.
8334 * @param[in] dev_flow
8335 * Pointer to the mlx5_flow.
8336 * @param[in] rss_desc
8337 * Pointer to the mlx5_flow_rss_desc.
8340 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
8341 struct mlx5_flow_rss_desc *rss_desc)
8343 uint64_t items = dev_flow->handle->layers;
8345 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
8347 dev_flow->hash_fields = 0;
8348 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
8349 if (rss_desc->level >= 2) {
8350 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
8354 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
8355 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
8356 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
8357 if (rss_types & ETH_RSS_L3_SRC_ONLY)
8358 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
8359 else if (rss_types & ETH_RSS_L3_DST_ONLY)
8360 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
8362 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
8364 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
8365 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
8366 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
8367 if (rss_types & ETH_RSS_L3_SRC_ONLY)
8368 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
8369 else if (rss_types & ETH_RSS_L3_DST_ONLY)
8370 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
8372 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
8375 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
8376 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
8377 if (rss_types & ETH_RSS_UDP) {
8378 if (rss_types & ETH_RSS_L4_SRC_ONLY)
8379 dev_flow->hash_fields |=
8380 IBV_RX_HASH_SRC_PORT_UDP;
8381 else if (rss_types & ETH_RSS_L4_DST_ONLY)
8382 dev_flow->hash_fields |=
8383 IBV_RX_HASH_DST_PORT_UDP;
8385 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
8387 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
8388 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
8389 if (rss_types & ETH_RSS_TCP) {
8390 if (rss_types & ETH_RSS_L4_SRC_ONLY)
8391 dev_flow->hash_fields |=
8392 IBV_RX_HASH_SRC_PORT_TCP;
8393 else if (rss_types & ETH_RSS_L4_DST_ONLY)
8394 dev_flow->hash_fields |=
8395 IBV_RX_HASH_DST_PORT_TCP;
8397 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
8403 * Create an Rx Hash queue.
8406 * Pointer to Ethernet device.
8407 * @param[in] dev_flow
8408 * Pointer to the mlx5_flow.
8409 * @param[in] rss_desc
8410 * Pointer to the mlx5_flow_rss_desc.
8411 * @param[out] hrxq_idx
8412 * Hash Rx queue index.
8415 * The Verbs/DevX object initialised, NULL otherwise and rte_errno is set.
8417 static struct mlx5_hrxq *
8418 flow_dv_handle_rx_queue(struct rte_eth_dev *dev,
8419 struct mlx5_flow *dev_flow,
8420 struct mlx5_flow_rss_desc *rss_desc,
8423 struct mlx5_priv *priv = dev->data->dev_private;
8424 struct mlx5_flow_handle *dh = dev_flow->handle;
8425 struct mlx5_hrxq *hrxq;
8427 MLX5_ASSERT(rss_desc->queue_num);
8428 *hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
8429 MLX5_RSS_HASH_KEY_LEN,
8430 dev_flow->hash_fields,
8432 rss_desc->queue_num);
8434 *hrxq_idx = mlx5_hrxq_new
8435 (dev, rss_desc->key,
8436 MLX5_RSS_HASH_KEY_LEN,
8437 dev_flow->hash_fields,
8439 rss_desc->queue_num,
8441 MLX5_FLOW_LAYER_TUNNEL));
8445 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8451 * Find existing sample resource or create and register a new one.
8453 * @param[in, out] dev
8454 * Pointer to rte_eth_dev structure.
8456 * Attributes of flow that includes this item.
8457 * @param[in] resource
8458 * Pointer to sample resource.
8459 * @parm[in, out] dev_flow
8460 * Pointer to the dev_flow.
8461 * @param[in, out] sample_dv_actions
8462 * Pointer to sample actions list.
8464 * pointer to error structure.
8467 * 0 on success otherwise -errno and errno is set.
8470 flow_dv_sample_resource_register(struct rte_eth_dev *dev,
8471 const struct rte_flow_attr *attr,
8472 struct mlx5_flow_dv_sample_resource *resource,
8473 struct mlx5_flow *dev_flow,
8474 void **sample_dv_actions,
8475 struct rte_flow_error *error)
8477 struct mlx5_flow_dv_sample_resource *cache_resource;
8478 struct mlx5dv_dr_flow_sampler_attr sampler_attr;
8479 struct mlx5_priv *priv = dev->data->dev_private;
8480 struct mlx5_dev_ctx_shared *sh = priv->sh;
8481 struct mlx5_flow_tbl_resource *tbl;
8483 const uint32_t next_ft_step = 1;
8484 uint32_t next_ft_id = resource->ft_id + next_ft_step;
8486 /* Lookup a matching resource from cache. */
8487 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_SAMPLE], sh->sample_action_list,
8488 idx, cache_resource, next) {
8489 if (resource->ratio == cache_resource->ratio &&
8490 resource->ft_type == cache_resource->ft_type &&
8491 resource->ft_id == cache_resource->ft_id &&
8492 resource->set_action == cache_resource->set_action &&
8493 !memcmp((void *)&resource->sample_act,
8494 (void *)&cache_resource->sample_act,
8495 sizeof(struct mlx5_flow_sub_actions_list))) {
8496 DRV_LOG(DEBUG, "sample resource %p: refcnt %d++",
8497 (void *)cache_resource,
8498 __atomic_load_n(&cache_resource->refcnt,
8500 __atomic_fetch_add(&cache_resource->refcnt, 1,
8502 dev_flow->handle->dvh.rix_sample = idx;
8503 dev_flow->dv.sample_res = cache_resource;
8507 /* Register new sample resource. */
8508 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE],
8509 &dev_flow->handle->dvh.rix_sample);
8510 if (!cache_resource)
8511 return rte_flow_error_set(error, ENOMEM,
8512 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8514 "cannot allocate resource memory");
8515 *cache_resource = *resource;
8516 /* Create normal path table level */
8517 tbl = flow_dv_tbl_resource_get(dev, next_ft_id,
8518 attr->egress, attr->transfer, error);
8520 rte_flow_error_set(error, ENOMEM,
8521 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8523 "fail to create normal path table "
8527 cache_resource->normal_path_tbl = tbl;
8528 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
8529 cache_resource->default_miss =
8530 mlx5_glue->dr_create_flow_action_default_miss();
8531 if (!cache_resource->default_miss) {
8532 rte_flow_error_set(error, ENOMEM,
8533 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8535 "cannot create default miss "
8539 sample_dv_actions[resource->sample_act.actions_num++] =
8540 cache_resource->default_miss;
8542 /* Create a DR sample action */
8543 sampler_attr.sample_ratio = cache_resource->ratio;
8544 sampler_attr.default_next_table = tbl->obj;
8545 sampler_attr.num_sample_actions = resource->sample_act.actions_num;
8546 sampler_attr.sample_actions = (struct mlx5dv_dr_action **)
8547 &sample_dv_actions[0];
8548 sampler_attr.action = cache_resource->set_action;
8549 cache_resource->verbs_action =
8550 mlx5_glue->dr_create_flow_action_sampler(&sampler_attr);
8551 if (!cache_resource->verbs_action) {
8552 rte_flow_error_set(error, ENOMEM,
8553 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8554 NULL, "cannot create sample action");
8557 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
8558 ILIST_INSERT(sh->ipool[MLX5_IPOOL_SAMPLE], &sh->sample_action_list,
8559 dev_flow->handle->dvh.rix_sample, cache_resource,
8561 dev_flow->dv.sample_res = cache_resource;
8562 DRV_LOG(DEBUG, "new sample resource %p: refcnt %d++",
8563 (void *)cache_resource,
8564 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
8567 if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
8568 if (cache_resource->default_miss)
8569 claim_zero(mlx5_glue->destroy_flow_action
8570 (cache_resource->default_miss));
8572 if (cache_resource->sample_idx.rix_hrxq &&
8573 !mlx5_hrxq_release(dev,
8574 cache_resource->sample_idx.rix_hrxq))
8575 cache_resource->sample_idx.rix_hrxq = 0;
8576 if (cache_resource->sample_idx.rix_tag &&
8577 !flow_dv_tag_release(dev,
8578 cache_resource->sample_idx.rix_tag))
8579 cache_resource->sample_idx.rix_tag = 0;
8580 if (cache_resource->sample_idx.cnt) {
8581 flow_dv_counter_release(dev,
8582 cache_resource->sample_idx.cnt);
8583 cache_resource->sample_idx.cnt = 0;
8586 if (cache_resource->normal_path_tbl)
8587 flow_dv_tbl_resource_release(dev,
8588 cache_resource->normal_path_tbl);
8589 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_SAMPLE],
8590 dev_flow->handle->dvh.rix_sample);
8591 dev_flow->handle->dvh.rix_sample = 0;
8596 * Find existing destination array resource or create and register a new one.
8598 * @param[in, out] dev
8599 * Pointer to rte_eth_dev structure.
8601 * Attributes of flow that includes this item.
8602 * @param[in] resource
8603 * Pointer to destination array resource.
8604 * @parm[in, out] dev_flow
8605 * Pointer to the dev_flow.
8607 * pointer to error structure.
8610 * 0 on success otherwise -errno and errno is set.
8613 flow_dv_dest_array_resource_register(struct rte_eth_dev *dev,
8614 const struct rte_flow_attr *attr,
8615 struct mlx5_flow_dv_dest_array_resource *resource,
8616 struct mlx5_flow *dev_flow,
8617 struct rte_flow_error *error)
8619 struct mlx5_flow_dv_dest_array_resource *cache_resource;
8620 struct mlx5dv_dr_action_dest_attr *dest_attr[MLX5_MAX_DEST_NUM] = { 0 };
8621 struct mlx5dv_dr_action_dest_reformat dest_reformat[MLX5_MAX_DEST_NUM];
8622 struct mlx5_priv *priv = dev->data->dev_private;
8623 struct mlx5_dev_ctx_shared *sh = priv->sh;
8624 struct mlx5_flow_sub_actions_list *sample_act;
8625 struct mlx5dv_dr_domain *domain;
8628 /* Lookup a matching resource from cache. */
8629 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8630 sh->dest_array_list,
8631 idx, cache_resource, next) {
8632 if (resource->num_of_dest == cache_resource->num_of_dest &&
8633 resource->ft_type == cache_resource->ft_type &&
8634 !memcmp((void *)cache_resource->sample_act,
8635 (void *)resource->sample_act,
8636 (resource->num_of_dest *
8637 sizeof(struct mlx5_flow_sub_actions_list)))) {
8638 DRV_LOG(DEBUG, "dest array resource %p: refcnt %d++",
8639 (void *)cache_resource,
8640 __atomic_load_n(&cache_resource->refcnt,
8642 __atomic_fetch_add(&cache_resource->refcnt, 1,
8644 dev_flow->handle->dvh.rix_dest_array = idx;
8645 dev_flow->dv.dest_array_res = cache_resource;
8649 /* Register new destination array resource. */
8650 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8651 &dev_flow->handle->dvh.rix_dest_array);
8652 if (!cache_resource)
8653 return rte_flow_error_set(error, ENOMEM,
8654 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8656 "cannot allocate resource memory");
8657 *cache_resource = *resource;
8659 domain = sh->fdb_domain;
8660 else if (attr->ingress)
8661 domain = sh->rx_domain;
8663 domain = sh->tx_domain;
8664 for (idx = 0; idx < resource->num_of_dest; idx++) {
8665 dest_attr[idx] = (struct mlx5dv_dr_action_dest_attr *)
8666 mlx5_malloc(MLX5_MEM_ZERO,
8667 sizeof(struct mlx5dv_dr_action_dest_attr),
8669 if (!dest_attr[idx]) {
8670 rte_flow_error_set(error, ENOMEM,
8671 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8673 "cannot allocate resource memory");
8676 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST;
8677 sample_act = &resource->sample_act[idx];
8678 if (sample_act->action_flags == MLX5_FLOW_ACTION_QUEUE) {
8679 dest_attr[idx]->dest = sample_act->dr_queue_action;
8680 } else if (sample_act->action_flags ==
8681 (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_ENCAP)) {
8682 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST_REFORMAT;
8683 dest_attr[idx]->dest_reformat = &dest_reformat[idx];
8684 dest_attr[idx]->dest_reformat->reformat =
8685 sample_act->dr_encap_action;
8686 dest_attr[idx]->dest_reformat->dest =
8687 sample_act->dr_port_id_action;
8688 } else if (sample_act->action_flags ==
8689 MLX5_FLOW_ACTION_PORT_ID) {
8690 dest_attr[idx]->dest = sample_act->dr_port_id_action;
8693 /* create a dest array actioin */
8694 cache_resource->action = mlx5_glue->dr_create_flow_action_dest_array
8696 cache_resource->num_of_dest,
8698 if (!cache_resource->action) {
8699 rte_flow_error_set(error, ENOMEM,
8700 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8702 "cannot create destination array action");
8705 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
8706 ILIST_INSERT(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8707 &sh->dest_array_list,
8708 dev_flow->handle->dvh.rix_dest_array, cache_resource,
8710 dev_flow->dv.dest_array_res = cache_resource;
8711 DRV_LOG(DEBUG, "new destination array resource %p: refcnt %d++",
8712 (void *)cache_resource,
8713 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
8714 for (idx = 0; idx < resource->num_of_dest; idx++)
8715 mlx5_free(dest_attr[idx]);
8718 for (idx = 0; idx < resource->num_of_dest; idx++) {
8719 struct mlx5_flow_sub_actions_idx *act_res =
8720 &cache_resource->sample_idx[idx];
8721 if (act_res->rix_hrxq &&
8722 !mlx5_hrxq_release(dev,
8724 act_res->rix_hrxq = 0;
8725 if (act_res->rix_encap_decap &&
8726 !flow_dv_encap_decap_resource_release(dev,
8727 act_res->rix_encap_decap))
8728 act_res->rix_encap_decap = 0;
8729 if (act_res->rix_port_id_action &&
8730 !flow_dv_port_id_action_resource_release(dev,
8731 act_res->rix_port_id_action))
8732 act_res->rix_port_id_action = 0;
8734 mlx5_free(dest_attr[idx]);
8737 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8738 dev_flow->handle->dvh.rix_dest_array);
8739 dev_flow->handle->dvh.rix_dest_array = 0;
8744 * Convert Sample action to DV specification.
8747 * Pointer to rte_eth_dev structure.
8749 * Pointer to action structure.
8750 * @param[in, out] dev_flow
8751 * Pointer to the mlx5_flow.
8753 * Pointer to the flow attributes.
8754 * @param[in, out] num_of_dest
8755 * Pointer to the num of destination.
8756 * @param[in, out] sample_actions
8757 * Pointer to sample actions list.
8758 * @param[in, out] res
8759 * Pointer to sample resource.
8761 * Pointer to the error structure.
8764 * 0 on success, a negative errno value otherwise and rte_errno is set.
8767 flow_dv_translate_action_sample(struct rte_eth_dev *dev,
8768 const struct rte_flow_action *action,
8769 struct mlx5_flow *dev_flow,
8770 const struct rte_flow_attr *attr,
8771 uint32_t *num_of_dest,
8772 void **sample_actions,
8773 struct mlx5_flow_dv_sample_resource *res,
8774 struct rte_flow_error *error)
8776 struct mlx5_priv *priv = dev->data->dev_private;
8777 const struct rte_flow_action_sample *sample_action;
8778 const struct rte_flow_action *sub_actions;
8779 const struct rte_flow_action_queue *queue;
8780 struct mlx5_flow_sub_actions_list *sample_act;
8781 struct mlx5_flow_sub_actions_idx *sample_idx;
8782 struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
8784 [!!priv->flow_nested_idx];
8785 uint64_t action_flags = 0;
8787 sample_act = &res->sample_act;
8788 sample_idx = &res->sample_idx;
8789 sample_action = (const struct rte_flow_action_sample *)action->conf;
8790 res->ratio = sample_action->ratio;
8791 sub_actions = sample_action->actions;
8792 for (; sub_actions->type != RTE_FLOW_ACTION_TYPE_END; sub_actions++) {
8793 int type = sub_actions->type;
8794 uint32_t pre_rix = 0;
8797 case RTE_FLOW_ACTION_TYPE_QUEUE:
8799 struct mlx5_hrxq *hrxq;
8802 queue = sub_actions->conf;
8803 rss_desc->queue_num = 1;
8804 rss_desc->queue[0] = queue->index;
8805 hrxq = flow_dv_handle_rx_queue(dev, dev_flow,
8806 rss_desc, &hrxq_idx);
8808 return rte_flow_error_set
8810 RTE_FLOW_ERROR_TYPE_ACTION,
8812 "cannot create fate queue");
8813 sample_act->dr_queue_action = hrxq->action;
8814 sample_idx->rix_hrxq = hrxq_idx;
8815 sample_actions[sample_act->actions_num++] =
8818 action_flags |= MLX5_FLOW_ACTION_QUEUE;
8819 if (action_flags & MLX5_FLOW_ACTION_MARK)
8820 dev_flow->handle->rix_hrxq = hrxq_idx;
8821 dev_flow->handle->fate_action =
8822 MLX5_FLOW_FATE_QUEUE;
8825 case RTE_FLOW_ACTION_TYPE_MARK:
8827 uint32_t tag_be = mlx5_flow_mark_set
8828 (((const struct rte_flow_action_mark *)
8829 (sub_actions->conf))->id);
8831 dev_flow->handle->mark = 1;
8832 pre_rix = dev_flow->handle->dvh.rix_tag;
8833 /* Save the mark resource before sample */
8834 pre_r = dev_flow->dv.tag_resource;
8835 if (flow_dv_tag_resource_register(dev, tag_be,
8838 MLX5_ASSERT(dev_flow->dv.tag_resource);
8839 sample_act->dr_tag_action =
8840 dev_flow->dv.tag_resource->action;
8841 sample_idx->rix_tag =
8842 dev_flow->handle->dvh.rix_tag;
8843 sample_actions[sample_act->actions_num++] =
8844 sample_act->dr_tag_action;
8845 /* Recover the mark resource after sample */
8846 dev_flow->dv.tag_resource = pre_r;
8847 dev_flow->handle->dvh.rix_tag = pre_rix;
8848 action_flags |= MLX5_FLOW_ACTION_MARK;
8851 case RTE_FLOW_ACTION_TYPE_COUNT:
8855 counter = flow_dv_translate_create_counter(dev,
8856 dev_flow, sub_actions->conf, 0);
8858 return rte_flow_error_set
8860 RTE_FLOW_ERROR_TYPE_ACTION,
8862 "cannot create counter"
8864 sample_idx->cnt = counter;
8865 sample_act->dr_cnt_action =
8866 (flow_dv_counter_get_by_idx(dev,
8867 counter, NULL))->action;
8868 sample_actions[sample_act->actions_num++] =
8869 sample_act->dr_cnt_action;
8870 action_flags |= MLX5_FLOW_ACTION_COUNT;
8873 case RTE_FLOW_ACTION_TYPE_PORT_ID:
8875 struct mlx5_flow_dv_port_id_action_resource
8877 uint32_t port_id = 0;
8879 memset(&port_id_resource, 0, sizeof(port_id_resource));
8880 /* Save the port id resource before sample */
8881 pre_rix = dev_flow->handle->rix_port_id_action;
8882 pre_r = dev_flow->dv.port_id_action;
8883 if (flow_dv_translate_action_port_id(dev, sub_actions,
8886 port_id_resource.port_id = port_id;
8887 if (flow_dv_port_id_action_resource_register
8888 (dev, &port_id_resource, dev_flow, error))
8890 sample_act->dr_port_id_action =
8891 dev_flow->dv.port_id_action->action;
8892 sample_idx->rix_port_id_action =
8893 dev_flow->handle->rix_port_id_action;
8894 sample_actions[sample_act->actions_num++] =
8895 sample_act->dr_port_id_action;
8896 /* Recover the port id resource after sample */
8897 dev_flow->dv.port_id_action = pre_r;
8898 dev_flow->handle->rix_port_id_action = pre_rix;
8900 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
8903 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
8904 /* Save the encap resource before sample */
8905 pre_rix = dev_flow->handle->dvh.rix_encap_decap;
8906 pre_r = dev_flow->dv.encap_decap;
8907 if (flow_dv_create_action_l2_encap(dev, sub_actions,
8912 sample_act->dr_encap_action =
8913 dev_flow->dv.encap_decap->action;
8914 sample_idx->rix_encap_decap =
8915 dev_flow->handle->dvh.rix_encap_decap;
8916 sample_actions[sample_act->actions_num++] =
8917 sample_act->dr_encap_action;
8918 /* Recover the encap resource after sample */
8919 dev_flow->dv.encap_decap = pre_r;
8920 dev_flow->handle->dvh.rix_encap_decap = pre_rix;
8921 action_flags |= MLX5_FLOW_ACTION_ENCAP;
8924 return rte_flow_error_set(error, EINVAL,
8925 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8927 "Not support for sampler action");
8930 sample_act->action_flags = action_flags;
8931 res->ft_id = dev_flow->dv.group;
8932 if (attr->transfer) {
8934 uint32_t action_in[MLX5_ST_SZ_DW(set_action_in)];
8935 uint64_t set_action;
8936 } action_ctx = { .set_action = 0 };
8938 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
8939 MLX5_SET(set_action_in, action_ctx.action_in, action_type,
8940 MLX5_MODIFICATION_TYPE_SET);
8941 MLX5_SET(set_action_in, action_ctx.action_in, field,
8942 MLX5_MODI_META_REG_C_0);
8943 MLX5_SET(set_action_in, action_ctx.action_in, data,
8944 priv->vport_meta_tag);
8945 res->set_action = action_ctx.set_action;
8946 } else if (attr->ingress) {
8947 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
8953 * Convert Sample action to DV specification.
8956 * Pointer to rte_eth_dev structure.
8957 * @param[in, out] dev_flow
8958 * Pointer to the mlx5_flow.
8960 * Pointer to the flow attributes.
8961 * @param[in] num_of_dest
8962 * The num of destination.
8963 * @param[in, out] res
8964 * Pointer to sample resource.
8965 * @param[in, out] mdest_res
8966 * Pointer to destination array resource.
8967 * @param[in] sample_actions
8968 * Pointer to sample path actions list.
8969 * @param[in] action_flags
8970 * Holds the actions detected until now.
8972 * Pointer to the error structure.
8975 * 0 on success, a negative errno value otherwise and rte_errno is set.
8978 flow_dv_create_action_sample(struct rte_eth_dev *dev,
8979 struct mlx5_flow *dev_flow,
8980 const struct rte_flow_attr *attr,
8981 uint32_t num_of_dest,
8982 struct mlx5_flow_dv_sample_resource *res,
8983 struct mlx5_flow_dv_dest_array_resource *mdest_res,
8984 void **sample_actions,
8985 uint64_t action_flags,
8986 struct rte_flow_error *error)
8988 struct mlx5_priv *priv = dev->data->dev_private;
8989 /* update normal path action resource into last index of array */
8990 uint32_t dest_index = MLX5_MAX_DEST_NUM - 1;
8991 struct mlx5_flow_sub_actions_list *sample_act =
8992 &mdest_res->sample_act[dest_index];
8993 struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
8995 [!!priv->flow_nested_idx];
8996 uint32_t normal_idx = 0;
8997 struct mlx5_hrxq *hrxq;
9000 if (num_of_dest > 1) {
9001 if (sample_act->action_flags & MLX5_FLOW_ACTION_QUEUE) {
9002 /* Handle QP action for mirroring */
9003 hrxq = flow_dv_handle_rx_queue(dev, dev_flow,
9004 rss_desc, &hrxq_idx);
9006 return rte_flow_error_set
9008 RTE_FLOW_ERROR_TYPE_ACTION,
9010 "cannot create rx queue");
9012 mdest_res->sample_idx[dest_index].rix_hrxq = hrxq_idx;
9013 sample_act->dr_queue_action = hrxq->action;
9014 if (action_flags & MLX5_FLOW_ACTION_MARK)
9015 dev_flow->handle->rix_hrxq = hrxq_idx;
9016 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9018 if (sample_act->action_flags & MLX5_FLOW_ACTION_ENCAP) {
9020 mdest_res->sample_idx[dest_index].rix_encap_decap =
9021 dev_flow->handle->dvh.rix_encap_decap;
9022 sample_act->dr_encap_action =
9023 dev_flow->dv.encap_decap->action;
9025 if (sample_act->action_flags & MLX5_FLOW_ACTION_PORT_ID) {
9027 mdest_res->sample_idx[dest_index].rix_port_id_action =
9028 dev_flow->handle->rix_port_id_action;
9029 sample_act->dr_port_id_action =
9030 dev_flow->dv.port_id_action->action;
9032 sample_act->actions_num = normal_idx;
9033 /* update sample action resource into first index of array */
9034 mdest_res->ft_type = res->ft_type;
9035 memcpy(&mdest_res->sample_idx[0], &res->sample_idx,
9036 sizeof(struct mlx5_flow_sub_actions_idx));
9037 memcpy(&mdest_res->sample_act[0], &res->sample_act,
9038 sizeof(struct mlx5_flow_sub_actions_list));
9039 mdest_res->num_of_dest = num_of_dest;
9040 if (flow_dv_dest_array_resource_register(dev, attr, mdest_res,
9042 return rte_flow_error_set(error, EINVAL,
9043 RTE_FLOW_ERROR_TYPE_ACTION,
9044 NULL, "can't create sample "
9047 if (flow_dv_sample_resource_register(dev, attr, res, dev_flow,
9048 sample_actions, error))
9049 return rte_flow_error_set(error, EINVAL,
9050 RTE_FLOW_ERROR_TYPE_ACTION,
9052 "can't create sample action");
9058 * Fill the flow with DV spec, lock free
9059 * (mutex should be acquired by caller).
9062 * Pointer to rte_eth_dev structure.
9063 * @param[in, out] dev_flow
9064 * Pointer to the sub flow.
9066 * Pointer to the flow attributes.
9068 * Pointer to the list of items.
9069 * @param[in] actions
9070 * Pointer to the list of actions.
9072 * Pointer to the error structure.
9075 * 0 on success, a negative errno value otherwise and rte_errno is set.
9078 __flow_dv_translate(struct rte_eth_dev *dev,
9079 struct mlx5_flow *dev_flow,
9080 const struct rte_flow_attr *attr,
9081 const struct rte_flow_item items[],
9082 const struct rte_flow_action actions[],
9083 struct rte_flow_error *error)
9085 struct mlx5_priv *priv = dev->data->dev_private;
9086 struct mlx5_dev_config *dev_conf = &priv->config;
9087 struct rte_flow *flow = dev_flow->flow;
9088 struct mlx5_flow_handle *handle = dev_flow->handle;
9089 struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
9091 [!!priv->flow_nested_idx];
9092 uint64_t item_flags = 0;
9093 uint64_t last_item = 0;
9094 uint64_t action_flags = 0;
9095 uint64_t priority = attr->priority;
9096 struct mlx5_flow_dv_matcher matcher = {
9098 .size = sizeof(matcher.mask.buf) -
9099 MLX5_ST_SZ_BYTES(fte_match_set_misc4),
9103 bool actions_end = false;
9105 struct mlx5_flow_dv_modify_hdr_resource res;
9106 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
9107 sizeof(struct mlx5_modification_cmd) *
9108 (MLX5_MAX_MODIFY_NUM + 1)];
9110 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
9111 const struct rte_flow_action_count *count = NULL;
9112 const struct rte_flow_action_age *age = NULL;
9113 union flow_dv_attr flow_attr = { .attr = 0 };
9115 union mlx5_flow_tbl_key tbl_key;
9116 uint32_t modify_action_position = UINT32_MAX;
9117 void *match_mask = matcher.mask.buf;
9118 void *match_value = dev_flow->dv.value.buf;
9119 uint8_t next_protocol = 0xff;
9120 struct rte_vlan_hdr vlan = { 0 };
9121 struct mlx5_flow_dv_dest_array_resource mdest_res;
9122 struct mlx5_flow_dv_sample_resource sample_res;
9123 void *sample_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9124 struct mlx5_flow_sub_actions_list *sample_act;
9125 uint32_t sample_act_pos = UINT32_MAX;
9126 uint32_t num_of_dest = 0;
9127 int tmp_actions_n = 0;
9131 memset(&mdest_res, 0, sizeof(struct mlx5_flow_dv_dest_array_resource));
9132 memset(&sample_res, 0, sizeof(struct mlx5_flow_dv_sample_resource));
9133 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
9134 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9135 /* update normal path action resource into last index of array */
9136 sample_act = &mdest_res.sample_act[MLX5_MAX_DEST_NUM - 1];
9137 ret = mlx5_flow_group_to_table(attr, dev_flow->external, attr->group,
9138 !!priv->fdb_def_rule, &table, error);
9141 dev_flow->dv.group = table;
9143 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
9144 if (priority == MLX5_FLOW_PRIO_RSVD)
9145 priority = dev_conf->flow_prio - 1;
9146 /* number of actions must be set to 0 in case of dirty stack. */
9147 mhdr_res->actions_num = 0;
9148 for (; !actions_end ; actions++) {
9149 const struct rte_flow_action_queue *queue;
9150 const struct rte_flow_action_rss *rss;
9151 const struct rte_flow_action *action = actions;
9152 const uint8_t *rss_key;
9153 const struct rte_flow_action_meter *mtr;
9154 struct mlx5_flow_tbl_resource *tbl;
9155 uint32_t port_id = 0;
9156 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
9157 int action_type = actions->type;
9158 const struct rte_flow_action *found_action = NULL;
9159 struct mlx5_flow_meter *fm = NULL;
9160 uint32_t jump_group = 0;
9162 if (!mlx5_flow_os_action_supported(action_type))
9163 return rte_flow_error_set(error, ENOTSUP,
9164 RTE_FLOW_ERROR_TYPE_ACTION,
9166 "action not supported");
9167 switch (action_type) {
9168 case RTE_FLOW_ACTION_TYPE_VOID:
9170 case RTE_FLOW_ACTION_TYPE_PORT_ID:
9171 if (flow_dv_translate_action_port_id(dev, action,
9174 port_id_resource.port_id = port_id;
9175 MLX5_ASSERT(!handle->rix_port_id_action);
9176 if (flow_dv_port_id_action_resource_register
9177 (dev, &port_id_resource, dev_flow, error))
9179 dev_flow->dv.actions[actions_n++] =
9180 dev_flow->dv.port_id_action->action;
9181 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9182 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
9183 sample_act->action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9186 case RTE_FLOW_ACTION_TYPE_FLAG:
9187 action_flags |= MLX5_FLOW_ACTION_FLAG;
9188 dev_flow->handle->mark = 1;
9189 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
9190 struct rte_flow_action_mark mark = {
9191 .id = MLX5_FLOW_MARK_DEFAULT,
9194 if (flow_dv_convert_action_mark(dev, &mark,
9198 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
9201 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
9203 * Only one FLAG or MARK is supported per device flow
9204 * right now. So the pointer to the tag resource must be
9205 * zero before the register process.
9207 MLX5_ASSERT(!handle->dvh.rix_tag);
9208 if (flow_dv_tag_resource_register(dev, tag_be,
9211 MLX5_ASSERT(dev_flow->dv.tag_resource);
9212 dev_flow->dv.actions[actions_n++] =
9213 dev_flow->dv.tag_resource->action;
9215 case RTE_FLOW_ACTION_TYPE_MARK:
9216 action_flags |= MLX5_FLOW_ACTION_MARK;
9217 dev_flow->handle->mark = 1;
9218 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
9219 const struct rte_flow_action_mark *mark =
9220 (const struct rte_flow_action_mark *)
9223 if (flow_dv_convert_action_mark(dev, mark,
9227 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
9231 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
9232 /* Legacy (non-extensive) MARK action. */
9233 tag_be = mlx5_flow_mark_set
9234 (((const struct rte_flow_action_mark *)
9235 (actions->conf))->id);
9236 MLX5_ASSERT(!handle->dvh.rix_tag);
9237 if (flow_dv_tag_resource_register(dev, tag_be,
9240 MLX5_ASSERT(dev_flow->dv.tag_resource);
9241 dev_flow->dv.actions[actions_n++] =
9242 dev_flow->dv.tag_resource->action;
9244 case RTE_FLOW_ACTION_TYPE_SET_META:
9245 if (flow_dv_convert_action_set_meta
9246 (dev, mhdr_res, attr,
9247 (const struct rte_flow_action_set_meta *)
9248 actions->conf, error))
9250 action_flags |= MLX5_FLOW_ACTION_SET_META;
9252 case RTE_FLOW_ACTION_TYPE_SET_TAG:
9253 if (flow_dv_convert_action_set_tag
9255 (const struct rte_flow_action_set_tag *)
9256 actions->conf, error))
9258 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9260 case RTE_FLOW_ACTION_TYPE_DROP:
9261 action_flags |= MLX5_FLOW_ACTION_DROP;
9262 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
9264 case RTE_FLOW_ACTION_TYPE_QUEUE:
9265 queue = actions->conf;
9266 rss_desc->queue_num = 1;
9267 rss_desc->queue[0] = queue->index;
9268 action_flags |= MLX5_FLOW_ACTION_QUEUE;
9269 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9270 sample_act->action_flags |= MLX5_FLOW_ACTION_QUEUE;
9273 case RTE_FLOW_ACTION_TYPE_RSS:
9274 rss = actions->conf;
9275 memcpy(rss_desc->queue, rss->queue,
9276 rss->queue_num * sizeof(uint16_t));
9277 rss_desc->queue_num = rss->queue_num;
9278 /* NULL RSS key indicates default RSS key. */
9279 rss_key = !rss->key ? rss_hash_default_key : rss->key;
9280 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
9282 * rss->level and rss.types should be set in advance
9283 * when expanding items for RSS.
9285 action_flags |= MLX5_FLOW_ACTION_RSS;
9286 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9288 case RTE_FLOW_ACTION_TYPE_AGE:
9289 case RTE_FLOW_ACTION_TYPE_COUNT:
9290 if (!dev_conf->devx) {
9291 return rte_flow_error_set
9293 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9295 "count action not supported");
9297 /* Save information first, will apply later. */
9298 if (actions->type == RTE_FLOW_ACTION_TYPE_COUNT)
9299 count = action->conf;
9302 action_flags |= MLX5_FLOW_ACTION_COUNT;
9304 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
9305 dev_flow->dv.actions[actions_n++] =
9306 priv->sh->pop_vlan_action;
9307 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
9309 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
9310 if (!(action_flags &
9311 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
9312 flow_dev_get_vlan_info_from_items(items, &vlan);
9313 vlan.eth_proto = rte_be_to_cpu_16
9314 ((((const struct rte_flow_action_of_push_vlan *)
9315 actions->conf)->ethertype));
9316 found_action = mlx5_flow_find_action
9318 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
9320 mlx5_update_vlan_vid_pcp(found_action, &vlan);
9321 found_action = mlx5_flow_find_action
9323 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
9325 mlx5_update_vlan_vid_pcp(found_action, &vlan);
9326 if (flow_dv_create_action_push_vlan
9327 (dev, attr, &vlan, dev_flow, error))
9329 dev_flow->dv.actions[actions_n++] =
9330 dev_flow->dv.push_vlan_res->action;
9331 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
9333 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
9334 /* of_vlan_push action handled this action */
9335 MLX5_ASSERT(action_flags &
9336 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
9338 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
9339 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
9341 flow_dev_get_vlan_info_from_items(items, &vlan);
9342 mlx5_update_vlan_vid_pcp(actions, &vlan);
9343 /* If no VLAN push - this is a modify header action */
9344 if (flow_dv_convert_action_modify_vlan_vid
9345 (mhdr_res, actions, error))
9347 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
9349 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
9350 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
9351 if (flow_dv_create_action_l2_encap(dev, actions,
9356 dev_flow->dv.actions[actions_n++] =
9357 dev_flow->dv.encap_decap->action;
9358 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9359 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
9360 sample_act->action_flags |=
9361 MLX5_FLOW_ACTION_ENCAP;
9363 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
9364 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
9365 if (flow_dv_create_action_l2_decap(dev, dev_flow,
9369 dev_flow->dv.actions[actions_n++] =
9370 dev_flow->dv.encap_decap->action;
9371 action_flags |= MLX5_FLOW_ACTION_DECAP;
9373 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
9374 /* Handle encap with preceding decap. */
9375 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
9376 if (flow_dv_create_action_raw_encap
9377 (dev, actions, dev_flow, attr, error))
9379 dev_flow->dv.actions[actions_n++] =
9380 dev_flow->dv.encap_decap->action;
9382 /* Handle encap without preceding decap. */
9383 if (flow_dv_create_action_l2_encap
9384 (dev, actions, dev_flow, attr->transfer,
9387 dev_flow->dv.actions[actions_n++] =
9388 dev_flow->dv.encap_decap->action;
9390 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9391 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
9392 sample_act->action_flags |=
9393 MLX5_FLOW_ACTION_ENCAP;
9395 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
9396 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
9398 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
9399 if (flow_dv_create_action_l2_decap
9400 (dev, dev_flow, attr->transfer, error))
9402 dev_flow->dv.actions[actions_n++] =
9403 dev_flow->dv.encap_decap->action;
9405 /* If decap is followed by encap, handle it at encap. */
9406 action_flags |= MLX5_FLOW_ACTION_DECAP;
9408 case RTE_FLOW_ACTION_TYPE_JUMP:
9409 jump_group = ((const struct rte_flow_action_jump *)
9410 action->conf)->group;
9411 if (dev_flow->external && jump_group <
9412 MLX5_MAX_TABLES_EXTERNAL)
9413 jump_group *= MLX5_FLOW_TABLE_FACTOR;
9414 ret = mlx5_flow_group_to_table(attr, dev_flow->external,
9416 !!priv->fdb_def_rule,
9420 tbl = flow_dv_tbl_resource_get(dev, table,
9422 attr->transfer, error);
9424 return rte_flow_error_set
9426 RTE_FLOW_ERROR_TYPE_ACTION,
9428 "cannot create jump action.");
9429 if (flow_dv_jump_tbl_resource_register
9430 (dev, tbl, dev_flow, error)) {
9431 flow_dv_tbl_resource_release(dev, tbl);
9432 return rte_flow_error_set
9434 RTE_FLOW_ERROR_TYPE_ACTION,
9436 "cannot create jump action.");
9438 dev_flow->dv.actions[actions_n++] =
9439 dev_flow->dv.jump->action;
9440 action_flags |= MLX5_FLOW_ACTION_JUMP;
9441 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
9443 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
9444 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
9445 if (flow_dv_convert_action_modify_mac
9446 (mhdr_res, actions, error))
9448 action_flags |= actions->type ==
9449 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
9450 MLX5_FLOW_ACTION_SET_MAC_SRC :
9451 MLX5_FLOW_ACTION_SET_MAC_DST;
9453 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
9454 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
9455 if (flow_dv_convert_action_modify_ipv4
9456 (mhdr_res, actions, error))
9458 action_flags |= actions->type ==
9459 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
9460 MLX5_FLOW_ACTION_SET_IPV4_SRC :
9461 MLX5_FLOW_ACTION_SET_IPV4_DST;
9463 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
9464 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
9465 if (flow_dv_convert_action_modify_ipv6
9466 (mhdr_res, actions, error))
9468 action_flags |= actions->type ==
9469 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
9470 MLX5_FLOW_ACTION_SET_IPV6_SRC :
9471 MLX5_FLOW_ACTION_SET_IPV6_DST;
9473 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
9474 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
9475 if (flow_dv_convert_action_modify_tp
9476 (mhdr_res, actions, items,
9477 &flow_attr, dev_flow, !!(action_flags &
9478 MLX5_FLOW_ACTION_DECAP), error))
9480 action_flags |= actions->type ==
9481 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
9482 MLX5_FLOW_ACTION_SET_TP_SRC :
9483 MLX5_FLOW_ACTION_SET_TP_DST;
9485 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
9486 if (flow_dv_convert_action_modify_dec_ttl
9487 (mhdr_res, items, &flow_attr, dev_flow,
9489 MLX5_FLOW_ACTION_DECAP), error))
9491 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
9493 case RTE_FLOW_ACTION_TYPE_SET_TTL:
9494 if (flow_dv_convert_action_modify_ttl
9495 (mhdr_res, actions, items, &flow_attr,
9496 dev_flow, !!(action_flags &
9497 MLX5_FLOW_ACTION_DECAP), error))
9499 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
9501 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
9502 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
9503 if (flow_dv_convert_action_modify_tcp_seq
9504 (mhdr_res, actions, error))
9506 action_flags |= actions->type ==
9507 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
9508 MLX5_FLOW_ACTION_INC_TCP_SEQ :
9509 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
9512 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
9513 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
9514 if (flow_dv_convert_action_modify_tcp_ack
9515 (mhdr_res, actions, error))
9517 action_flags |= actions->type ==
9518 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
9519 MLX5_FLOW_ACTION_INC_TCP_ACK :
9520 MLX5_FLOW_ACTION_DEC_TCP_ACK;
9522 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
9523 if (flow_dv_convert_action_set_reg
9524 (mhdr_res, actions, error))
9526 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9528 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
9529 if (flow_dv_convert_action_copy_mreg
9530 (dev, mhdr_res, actions, error))
9532 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9534 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
9535 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
9536 dev_flow->handle->fate_action =
9537 MLX5_FLOW_FATE_DEFAULT_MISS;
9539 case RTE_FLOW_ACTION_TYPE_METER:
9540 mtr = actions->conf;
9542 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
9545 return rte_flow_error_set(error,
9547 RTE_FLOW_ERROR_TYPE_ACTION,
9550 "or invalid parameters");
9551 flow->meter = fm->idx;
9553 /* Set the meter action. */
9555 fm = mlx5_ipool_get(priv->sh->ipool
9556 [MLX5_IPOOL_MTR], flow->meter);
9558 return rte_flow_error_set(error,
9560 RTE_FLOW_ERROR_TYPE_ACTION,
9563 "or invalid parameters");
9565 dev_flow->dv.actions[actions_n++] =
9566 fm->mfts->meter_action;
9567 action_flags |= MLX5_FLOW_ACTION_METER;
9569 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
9570 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
9573 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
9575 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
9576 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
9579 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
9581 case RTE_FLOW_ACTION_TYPE_SAMPLE:
9582 sample_act_pos = actions_n;
9583 ret = flow_dv_translate_action_sample(dev,
9593 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
9594 /* put encap action into group if work with port id */
9595 if ((action_flags & MLX5_FLOW_ACTION_ENCAP) &&
9596 (action_flags & MLX5_FLOW_ACTION_PORT_ID))
9597 sample_act->action_flags |=
9598 MLX5_FLOW_ACTION_ENCAP;
9600 case RTE_FLOW_ACTION_TYPE_END:
9602 if (mhdr_res->actions_num) {
9603 /* create modify action if needed. */
9604 if (flow_dv_modify_hdr_resource_register
9605 (dev, mhdr_res, dev_flow, error))
9607 dev_flow->dv.actions[modify_action_position] =
9608 handle->dvh.modify_hdr->action;
9610 if (action_flags & MLX5_FLOW_ACTION_COUNT) {
9612 flow_dv_translate_create_counter(dev,
9613 dev_flow, count, age);
9616 return rte_flow_error_set
9618 RTE_FLOW_ERROR_TYPE_ACTION,
9620 "cannot create counter"
9622 dev_flow->dv.actions[actions_n] =
9623 (flow_dv_counter_get_by_idx(dev,
9624 flow->counter, NULL))->action;
9627 if (action_flags & MLX5_FLOW_ACTION_SAMPLE) {
9628 ret = flow_dv_create_action_sample(dev,
9637 return rte_flow_error_set
9639 RTE_FLOW_ERROR_TYPE_ACTION,
9641 "cannot create sample action");
9642 if (num_of_dest > 1) {
9643 dev_flow->dv.actions[sample_act_pos] =
9644 dev_flow->dv.dest_array_res->action;
9646 dev_flow->dv.actions[sample_act_pos] =
9647 dev_flow->dv.sample_res->verbs_action;
9654 if (mhdr_res->actions_num &&
9655 modify_action_position == UINT32_MAX)
9656 modify_action_position = actions_n++;
9659 * For multiple destination (sample action with ratio=1), the encap
9660 * action and port id action will be combined into group action.
9661 * So need remove the original these actions in the flow and only
9662 * use the sample action instead of.
9664 if (num_of_dest > 1 && sample_act->dr_port_id_action) {
9666 void *temp_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9668 for (i = 0; i < actions_n; i++) {
9669 if ((sample_act->dr_encap_action &&
9670 sample_act->dr_encap_action ==
9671 dev_flow->dv.actions[i]) ||
9672 (sample_act->dr_port_id_action &&
9673 sample_act->dr_port_id_action ==
9674 dev_flow->dv.actions[i]))
9676 temp_actions[tmp_actions_n++] = dev_flow->dv.actions[i];
9678 memcpy((void *)dev_flow->dv.actions,
9679 (void *)temp_actions,
9680 tmp_actions_n * sizeof(void *));
9681 actions_n = tmp_actions_n;
9683 dev_flow->dv.actions_n = actions_n;
9684 dev_flow->act_flags = action_flags;
9685 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
9686 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
9687 int item_type = items->type;
9689 if (!mlx5_flow_os_item_supported(item_type))
9690 return rte_flow_error_set(error, ENOTSUP,
9691 RTE_FLOW_ERROR_TYPE_ITEM,
9692 NULL, "item not supported");
9693 switch (item_type) {
9694 case RTE_FLOW_ITEM_TYPE_PORT_ID:
9695 flow_dv_translate_item_port_id(dev, match_mask,
9696 match_value, items);
9697 last_item = MLX5_FLOW_ITEM_PORT_ID;
9699 case RTE_FLOW_ITEM_TYPE_ETH:
9700 flow_dv_translate_item_eth(match_mask, match_value,
9702 dev_flow->dv.group);
9703 matcher.priority = action_flags &
9704 MLX5_FLOW_ACTION_DEFAULT_MISS &&
9705 !dev_flow->external ?
9706 MLX5_PRIORITY_MAP_L3 :
9707 MLX5_PRIORITY_MAP_L2;
9708 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
9709 MLX5_FLOW_LAYER_OUTER_L2;
9711 case RTE_FLOW_ITEM_TYPE_VLAN:
9712 flow_dv_translate_item_vlan(dev_flow,
9713 match_mask, match_value,
9715 dev_flow->dv.group);
9716 matcher.priority = MLX5_PRIORITY_MAP_L2;
9717 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
9718 MLX5_FLOW_LAYER_INNER_VLAN) :
9719 (MLX5_FLOW_LAYER_OUTER_L2 |
9720 MLX5_FLOW_LAYER_OUTER_VLAN);
9722 case RTE_FLOW_ITEM_TYPE_IPV4:
9723 mlx5_flow_tunnel_ip_check(items, next_protocol,
9724 &item_flags, &tunnel);
9725 flow_dv_translate_item_ipv4(match_mask, match_value,
9726 items, item_flags, tunnel,
9727 dev_flow->dv.group);
9728 matcher.priority = MLX5_PRIORITY_MAP_L3;
9729 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
9730 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
9731 if (items->mask != NULL &&
9732 ((const struct rte_flow_item_ipv4 *)
9733 items->mask)->hdr.next_proto_id) {
9735 ((const struct rte_flow_item_ipv4 *)
9736 (items->spec))->hdr.next_proto_id;
9738 ((const struct rte_flow_item_ipv4 *)
9739 (items->mask))->hdr.next_proto_id;
9741 /* Reset for inner layer. */
9742 next_protocol = 0xff;
9745 case RTE_FLOW_ITEM_TYPE_IPV6:
9746 mlx5_flow_tunnel_ip_check(items, next_protocol,
9747 &item_flags, &tunnel);
9748 flow_dv_translate_item_ipv6(match_mask, match_value,
9749 items, item_flags, tunnel,
9750 dev_flow->dv.group);
9751 matcher.priority = MLX5_PRIORITY_MAP_L3;
9752 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
9753 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
9754 if (items->mask != NULL &&
9755 ((const struct rte_flow_item_ipv6 *)
9756 items->mask)->hdr.proto) {
9758 ((const struct rte_flow_item_ipv6 *)
9759 items->spec)->hdr.proto;
9761 ((const struct rte_flow_item_ipv6 *)
9762 items->mask)->hdr.proto;
9764 /* Reset for inner layer. */
9765 next_protocol = 0xff;
9768 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
9769 flow_dv_translate_item_ipv6_frag_ext(match_mask,
9772 last_item = tunnel ?
9773 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
9774 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
9775 if (items->mask != NULL &&
9776 ((const struct rte_flow_item_ipv6_frag_ext *)
9777 items->mask)->hdr.next_header) {
9779 ((const struct rte_flow_item_ipv6_frag_ext *)
9780 items->spec)->hdr.next_header;
9782 ((const struct rte_flow_item_ipv6_frag_ext *)
9783 items->mask)->hdr.next_header;
9785 /* Reset for inner layer. */
9786 next_protocol = 0xff;
9789 case RTE_FLOW_ITEM_TYPE_TCP:
9790 flow_dv_translate_item_tcp(match_mask, match_value,
9792 matcher.priority = MLX5_PRIORITY_MAP_L4;
9793 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
9794 MLX5_FLOW_LAYER_OUTER_L4_TCP;
9796 case RTE_FLOW_ITEM_TYPE_UDP:
9797 flow_dv_translate_item_udp(match_mask, match_value,
9799 matcher.priority = MLX5_PRIORITY_MAP_L4;
9800 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
9801 MLX5_FLOW_LAYER_OUTER_L4_UDP;
9803 case RTE_FLOW_ITEM_TYPE_GRE:
9804 flow_dv_translate_item_gre(match_mask, match_value,
9806 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9807 last_item = MLX5_FLOW_LAYER_GRE;
9809 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
9810 flow_dv_translate_item_gre_key(match_mask,
9811 match_value, items);
9812 last_item = MLX5_FLOW_LAYER_GRE_KEY;
9814 case RTE_FLOW_ITEM_TYPE_NVGRE:
9815 flow_dv_translate_item_nvgre(match_mask, match_value,
9817 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9818 last_item = MLX5_FLOW_LAYER_GRE;
9820 case RTE_FLOW_ITEM_TYPE_VXLAN:
9821 flow_dv_translate_item_vxlan(match_mask, match_value,
9823 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9824 last_item = MLX5_FLOW_LAYER_VXLAN;
9826 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
9827 flow_dv_translate_item_vxlan_gpe(match_mask,
9830 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9831 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
9833 case RTE_FLOW_ITEM_TYPE_GENEVE:
9834 flow_dv_translate_item_geneve(match_mask, match_value,
9836 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9837 last_item = MLX5_FLOW_LAYER_GENEVE;
9839 case RTE_FLOW_ITEM_TYPE_MPLS:
9840 flow_dv_translate_item_mpls(match_mask, match_value,
9841 items, last_item, tunnel);
9842 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9843 last_item = MLX5_FLOW_LAYER_MPLS;
9845 case RTE_FLOW_ITEM_TYPE_MARK:
9846 flow_dv_translate_item_mark(dev, match_mask,
9847 match_value, items);
9848 last_item = MLX5_FLOW_ITEM_MARK;
9850 case RTE_FLOW_ITEM_TYPE_META:
9851 flow_dv_translate_item_meta(dev, match_mask,
9852 match_value, attr, items);
9853 last_item = MLX5_FLOW_ITEM_METADATA;
9855 case RTE_FLOW_ITEM_TYPE_ICMP:
9856 flow_dv_translate_item_icmp(match_mask, match_value,
9858 last_item = MLX5_FLOW_LAYER_ICMP;
9860 case RTE_FLOW_ITEM_TYPE_ICMP6:
9861 flow_dv_translate_item_icmp6(match_mask, match_value,
9863 last_item = MLX5_FLOW_LAYER_ICMP6;
9865 case RTE_FLOW_ITEM_TYPE_TAG:
9866 flow_dv_translate_item_tag(dev, match_mask,
9867 match_value, items);
9868 last_item = MLX5_FLOW_ITEM_TAG;
9870 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
9871 flow_dv_translate_mlx5_item_tag(dev, match_mask,
9872 match_value, items);
9873 last_item = MLX5_FLOW_ITEM_TAG;
9875 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
9876 flow_dv_translate_item_tx_queue(dev, match_mask,
9879 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
9881 case RTE_FLOW_ITEM_TYPE_GTP:
9882 flow_dv_translate_item_gtp(match_mask, match_value,
9884 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9885 last_item = MLX5_FLOW_LAYER_GTP;
9887 case RTE_FLOW_ITEM_TYPE_ECPRI:
9888 if (!mlx5_flex_parser_ecpri_exist(dev)) {
9889 /* Create it only the first time to be used. */
9890 ret = mlx5_flex_parser_ecpri_alloc(dev);
9892 return rte_flow_error_set
9894 RTE_FLOW_ERROR_TYPE_ITEM,
9896 "cannot create eCPRI parser");
9898 /* Adjust the length matcher and device flow value. */
9899 matcher.mask.size = MLX5_ST_SZ_BYTES(fte_match_param);
9900 dev_flow->dv.value.size =
9901 MLX5_ST_SZ_BYTES(fte_match_param);
9902 flow_dv_translate_item_ecpri(dev, match_mask,
9903 match_value, items);
9904 /* No other protocol should follow eCPRI layer. */
9905 last_item = MLX5_FLOW_LAYER_ECPRI;
9910 item_flags |= last_item;
9913 * When E-Switch mode is enabled, we have two cases where we need to
9914 * set the source port manually.
9915 * The first one, is in case of Nic steering rule, and the second is
9916 * E-Switch rule where no port_id item was found. In both cases
9917 * the source port is set according the current port in use.
9919 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
9920 (priv->representor || priv->master)) {
9921 if (flow_dv_translate_item_port_id(dev, match_mask,
9925 #ifdef RTE_LIBRTE_MLX5_DEBUG
9926 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
9927 dev_flow->dv.value.buf));
9930 * Layers may be already initialized from prefix flow if this dev_flow
9931 * is the suffix flow.
9933 handle->layers |= item_flags;
9934 if (action_flags & MLX5_FLOW_ACTION_RSS)
9935 flow_dv_hashfields_set(dev_flow, rss_desc);
9936 /* Register matcher. */
9937 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
9939 matcher.priority = mlx5_flow_adjust_priority(dev, priority,
9941 /* reserved field no needs to be set to 0 here. */
9942 tbl_key.domain = attr->transfer;
9943 tbl_key.direction = attr->egress;
9944 tbl_key.table_id = dev_flow->dv.group;
9945 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
9951 * Apply the flow to the NIC, lock free,
9952 * (mutex should be acquired by caller).
9955 * Pointer to the Ethernet device structure.
9956 * @param[in, out] flow
9957 * Pointer to flow structure.
9959 * Pointer to error structure.
9962 * 0 on success, a negative errno value otherwise and rte_errno is set.
9965 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
9966 struct rte_flow_error *error)
9968 struct mlx5_flow_dv_workspace *dv;
9969 struct mlx5_flow_handle *dh;
9970 struct mlx5_flow_handle_dv *dv_h;
9971 struct mlx5_flow *dev_flow;
9972 struct mlx5_priv *priv = dev->data->dev_private;
9973 uint32_t handle_idx;
9978 for (idx = priv->flow_idx - 1; idx >= priv->flow_nested_idx; idx--) {
9979 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[idx];
9981 dh = dev_flow->handle;
9984 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
9986 dv->actions[n++] = priv->sh->esw_drop_action;
9988 struct mlx5_hrxq *drop_hrxq;
9989 drop_hrxq = mlx5_drop_action_create(dev);
9993 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9995 "cannot get drop hash queue");
9999 * Drop queues will be released by the specify
10000 * mlx5_drop_action_destroy() function. Assign
10001 * the special index to hrxq to mark the queue
10002 * has been allocated.
10004 dh->rix_hrxq = UINT32_MAX;
10005 dv->actions[n++] = drop_hrxq->action;
10007 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
10008 !dv_h->rix_sample && !dv_h->rix_dest_array) {
10009 struct mlx5_hrxq *hrxq;
10011 struct mlx5_flow_rss_desc *rss_desc =
10012 &((struct mlx5_flow_rss_desc *)priv->rss_desc)
10013 [!!priv->flow_nested_idx];
10015 MLX5_ASSERT(rss_desc->queue_num);
10016 hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
10017 MLX5_RSS_HASH_KEY_LEN,
10018 dev_flow->hash_fields,
10020 rss_desc->queue_num);
10022 hrxq_idx = mlx5_hrxq_new
10023 (dev, rss_desc->key,
10024 MLX5_RSS_HASH_KEY_LEN,
10025 dev_flow->hash_fields,
10027 rss_desc->queue_num,
10029 MLX5_FLOW_LAYER_TUNNEL));
10031 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
10036 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10037 "cannot get hash queue");
10040 dh->rix_hrxq = hrxq_idx;
10041 dv->actions[n++] = hrxq->action;
10042 } else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
10043 if (flow_dv_default_miss_resource_register
10047 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10048 "cannot create default miss resource");
10049 goto error_default_miss;
10051 dh->rix_default_fate = MLX5_FLOW_FATE_DEFAULT_MISS;
10052 dv->actions[n++] = priv->sh->default_miss.action;
10054 err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
10055 (void *)&dv->value, n,
10056 dv->actions, &dh->drv_flow);
10058 rte_flow_error_set(error, errno,
10059 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10061 "hardware refuses to create flow");
10064 if (priv->vmwa_context &&
10065 dh->vf_vlan.tag && !dh->vf_vlan.created) {
10067 * The rule contains the VLAN pattern.
10068 * For VF we are going to create VLAN
10069 * interface to make hypervisor set correct
10070 * e-Switch vport context.
10072 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
10077 if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS)
10078 flow_dv_default_miss_resource_release(dev);
10079 error_default_miss:
10080 err = rte_errno; /* Save rte_errno before cleanup. */
10081 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
10082 handle_idx, dh, next) {
10083 /* hrxq is union, don't clear it if the flag is not set. */
10084 if (dh->rix_hrxq) {
10085 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
10086 mlx5_drop_action_destroy(dev);
10088 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
10089 mlx5_hrxq_release(dev, dh->rix_hrxq);
10093 if (dh->vf_vlan.tag && dh->vf_vlan.created)
10094 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
10096 rte_errno = err; /* Restore rte_errno. */
10101 * Release the flow matcher.
10104 * Pointer to Ethernet device.
10106 * Pointer to mlx5_flow_handle.
10109 * 1 while a reference on it exists, 0 when freed.
10112 flow_dv_matcher_release(struct rte_eth_dev *dev,
10113 struct mlx5_flow_handle *handle)
10115 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
10117 MLX5_ASSERT(matcher->matcher_object);
10118 DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
10119 dev->data->port_id, (void *)matcher,
10120 rte_atomic32_read(&matcher->refcnt));
10121 if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
10122 claim_zero(mlx5_flow_os_destroy_flow_matcher
10123 (matcher->matcher_object));
10124 LIST_REMOVE(matcher, next);
10125 /* table ref-- in release interface. */
10126 flow_dv_tbl_resource_release(dev, matcher->tbl);
10127 mlx5_free(matcher);
10128 DRV_LOG(DEBUG, "port %u matcher %p: removed",
10129 dev->data->port_id, (void *)matcher);
10136 * Release an encap/decap resource.
10139 * Pointer to Ethernet device.
10140 * @param encap_decap_idx
10141 * Index of encap decap resource.
10144 * 1 while a reference on it exists, 0 when freed.
10147 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
10148 uint32_t encap_decap_idx)
10150 struct mlx5_priv *priv = dev->data->dev_private;
10151 uint32_t idx = encap_decap_idx;
10152 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
10154 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
10156 if (!cache_resource)
10158 MLX5_ASSERT(cache_resource->action);
10159 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
10160 (void *)cache_resource,
10161 rte_atomic32_read(&cache_resource->refcnt));
10162 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10163 claim_zero(mlx5_flow_os_destroy_flow_action
10164 (cache_resource->action));
10165 mlx5_hlist_remove(priv->sh->encaps_decaps,
10166 &cache_resource->entry);
10167 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
10168 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
10169 (void *)cache_resource);
10176 * Release an jump to table action resource.
10179 * Pointer to Ethernet device.
10181 * Pointer to mlx5_flow_handle.
10184 * 1 while a reference on it exists, 0 when freed.
10187 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
10188 struct mlx5_flow_handle *handle)
10190 struct mlx5_priv *priv = dev->data->dev_private;
10191 struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
10192 struct mlx5_flow_tbl_data_entry *tbl_data;
10194 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
10198 cache_resource = &tbl_data->jump;
10199 MLX5_ASSERT(cache_resource->action);
10200 DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
10201 (void *)cache_resource,
10202 rte_atomic32_read(&cache_resource->refcnt));
10203 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10204 claim_zero(mlx5_flow_os_destroy_flow_action
10205 (cache_resource->action));
10206 /* jump action memory free is inside the table release. */
10207 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
10208 DRV_LOG(DEBUG, "jump table resource %p: removed",
10209 (void *)cache_resource);
10216 * Release a default miss resource.
10219 * Pointer to Ethernet device.
10221 * 1 while a reference on it exists, 0 when freed.
10224 flow_dv_default_miss_resource_release(struct rte_eth_dev *dev)
10226 struct mlx5_priv *priv = dev->data->dev_private;
10227 struct mlx5_dev_ctx_shared *sh = priv->sh;
10228 struct mlx5_flow_default_miss_resource *cache_resource =
10231 MLX5_ASSERT(cache_resource->action);
10232 DRV_LOG(DEBUG, "default miss resource %p: refcnt %d--",
10233 (void *)cache_resource->action,
10234 rte_atomic32_read(&cache_resource->refcnt));
10235 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10236 claim_zero(mlx5_glue->destroy_flow_action
10237 (cache_resource->action));
10238 DRV_LOG(DEBUG, "default miss resource %p: removed",
10239 (void *)cache_resource->action);
10246 * Release a modify-header resource.
10249 * Pointer to Ethernet device.
10251 * Pointer to mlx5_flow_handle.
10254 * 1 while a reference on it exists, 0 when freed.
10257 flow_dv_modify_hdr_resource_release(struct rte_eth_dev *dev,
10258 struct mlx5_flow_handle *handle)
10260 struct mlx5_priv *priv = dev->data->dev_private;
10261 struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
10262 handle->dvh.modify_hdr;
10264 MLX5_ASSERT(cache_resource->action);
10265 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
10266 (void *)cache_resource,
10267 rte_atomic32_read(&cache_resource->refcnt));
10268 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10269 claim_zero(mlx5_flow_os_destroy_flow_action
10270 (cache_resource->action));
10271 mlx5_hlist_remove(priv->sh->modify_cmds,
10272 &cache_resource->entry);
10273 mlx5_free(cache_resource);
10274 DRV_LOG(DEBUG, "modify-header resource %p: removed",
10275 (void *)cache_resource);
10282 * Release port ID action resource.
10285 * Pointer to Ethernet device.
10287 * Pointer to mlx5_flow_handle.
10290 * 1 while a reference on it exists, 0 when freed.
10293 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
10296 struct mlx5_priv *priv = dev->data->dev_private;
10297 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
10298 uint32_t idx = port_id;
10300 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
10302 if (!cache_resource)
10304 MLX5_ASSERT(cache_resource->action);
10305 DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
10306 (void *)cache_resource,
10307 rte_atomic32_read(&cache_resource->refcnt));
10308 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10309 claim_zero(mlx5_flow_os_destroy_flow_action
10310 (cache_resource->action));
10311 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
10312 &priv->sh->port_id_action_list, idx,
10313 cache_resource, next);
10314 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
10315 DRV_LOG(DEBUG, "port id action resource %p: removed",
10316 (void *)cache_resource);
10323 * Release push vlan action resource.
10326 * Pointer to Ethernet device.
10328 * Pointer to mlx5_flow_handle.
10331 * 1 while a reference on it exists, 0 when freed.
10334 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
10335 struct mlx5_flow_handle *handle)
10337 struct mlx5_priv *priv = dev->data->dev_private;
10338 uint32_t idx = handle->dvh.rix_push_vlan;
10339 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
10341 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
10343 if (!cache_resource)
10345 MLX5_ASSERT(cache_resource->action);
10346 DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
10347 (void *)cache_resource,
10348 rte_atomic32_read(&cache_resource->refcnt));
10349 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10350 claim_zero(mlx5_flow_os_destroy_flow_action
10351 (cache_resource->action));
10352 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
10353 &priv->sh->push_vlan_action_list, idx,
10354 cache_resource, next);
10355 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
10356 DRV_LOG(DEBUG, "push vlan action resource %p: removed",
10357 (void *)cache_resource);
10364 * Release the fate resource.
10367 * Pointer to Ethernet device.
10369 * Pointer to mlx5_flow_handle.
10372 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
10373 struct mlx5_flow_handle *handle)
10375 if (!handle->rix_fate)
10377 switch (handle->fate_action) {
10378 case MLX5_FLOW_FATE_DROP:
10379 mlx5_drop_action_destroy(dev);
10381 case MLX5_FLOW_FATE_QUEUE:
10382 mlx5_hrxq_release(dev, handle->rix_hrxq);
10384 case MLX5_FLOW_FATE_JUMP:
10385 flow_dv_jump_tbl_resource_release(dev, handle);
10387 case MLX5_FLOW_FATE_PORT_ID:
10388 flow_dv_port_id_action_resource_release(dev,
10389 handle->rix_port_id_action);
10391 case MLX5_FLOW_FATE_DEFAULT_MISS:
10392 flow_dv_default_miss_resource_release(dev);
10395 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
10398 handle->rix_fate = 0;
10402 * Release an sample resource.
10405 * Pointer to Ethernet device.
10407 * Pointer to mlx5_flow_handle.
10410 * 1 while a reference on it exists, 0 when freed.
10413 flow_dv_sample_resource_release(struct rte_eth_dev *dev,
10414 struct mlx5_flow_handle *handle)
10416 struct mlx5_priv *priv = dev->data->dev_private;
10417 uint32_t idx = handle->dvh.rix_sample;
10418 struct mlx5_flow_dv_sample_resource *cache_resource;
10420 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
10422 if (!cache_resource)
10424 MLX5_ASSERT(cache_resource->verbs_action);
10425 DRV_LOG(DEBUG, "sample resource %p: refcnt %d--",
10426 (void *)cache_resource,
10427 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10428 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10429 __ATOMIC_RELAXED) == 0) {
10430 if (cache_resource->verbs_action)
10431 claim_zero(mlx5_glue->destroy_flow_action
10432 (cache_resource->verbs_action));
10433 if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
10434 if (cache_resource->default_miss)
10435 claim_zero(mlx5_glue->destroy_flow_action
10436 (cache_resource->default_miss));
10438 if (cache_resource->normal_path_tbl)
10439 flow_dv_tbl_resource_release(dev,
10440 cache_resource->normal_path_tbl);
10442 if (cache_resource->sample_idx.rix_hrxq &&
10443 !mlx5_hrxq_release(dev,
10444 cache_resource->sample_idx.rix_hrxq))
10445 cache_resource->sample_idx.rix_hrxq = 0;
10446 if (cache_resource->sample_idx.rix_tag &&
10447 !flow_dv_tag_release(dev,
10448 cache_resource->sample_idx.rix_tag))
10449 cache_resource->sample_idx.rix_tag = 0;
10450 if (cache_resource->sample_idx.cnt) {
10451 flow_dv_counter_release(dev,
10452 cache_resource->sample_idx.cnt);
10453 cache_resource->sample_idx.cnt = 0;
10455 if (!__atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED)) {
10456 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
10457 &priv->sh->sample_action_list, idx,
10458 cache_resource, next);
10459 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE], idx);
10460 DRV_LOG(DEBUG, "sample resource %p: removed",
10461 (void *)cache_resource);
10468 * Release an destination array resource.
10471 * Pointer to Ethernet device.
10473 * Pointer to mlx5_flow_handle.
10476 * 1 while a reference on it exists, 0 when freed.
10479 flow_dv_dest_array_resource_release(struct rte_eth_dev *dev,
10480 struct mlx5_flow_handle *handle)
10482 struct mlx5_priv *priv = dev->data->dev_private;
10483 struct mlx5_flow_dv_dest_array_resource *cache_resource;
10484 struct mlx5_flow_sub_actions_idx *mdest_act_res;
10485 uint32_t idx = handle->dvh.rix_dest_array;
10488 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
10490 if (!cache_resource)
10492 MLX5_ASSERT(cache_resource->action);
10493 DRV_LOG(DEBUG, "destination array resource %p: refcnt %d--",
10494 (void *)cache_resource,
10495 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10496 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10497 __ATOMIC_RELAXED) == 0) {
10498 if (cache_resource->action)
10499 claim_zero(mlx5_glue->destroy_flow_action
10500 (cache_resource->action));
10501 for (; i < cache_resource->num_of_dest; i++) {
10502 mdest_act_res = &cache_resource->sample_idx[i];
10503 if (mdest_act_res->rix_hrxq) {
10504 mlx5_hrxq_release(dev,
10505 mdest_act_res->rix_hrxq);
10506 mdest_act_res->rix_hrxq = 0;
10508 if (mdest_act_res->rix_encap_decap) {
10509 flow_dv_encap_decap_resource_release(dev,
10510 mdest_act_res->rix_encap_decap);
10511 mdest_act_res->rix_encap_decap = 0;
10513 if (mdest_act_res->rix_port_id_action) {
10514 flow_dv_port_id_action_resource_release(dev,
10515 mdest_act_res->rix_port_id_action);
10516 mdest_act_res->rix_port_id_action = 0;
10518 if (mdest_act_res->rix_tag) {
10519 flow_dv_tag_release(dev,
10520 mdest_act_res->rix_tag);
10521 mdest_act_res->rix_tag = 0;
10524 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
10525 &priv->sh->dest_array_list, idx,
10526 cache_resource, next);
10527 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY], idx);
10528 DRV_LOG(DEBUG, "destination array resource %p: removed",
10529 (void *)cache_resource);
10536 * Remove the flow from the NIC but keeps it in memory.
10537 * Lock free, (mutex should be acquired by caller).
10540 * Pointer to Ethernet device.
10541 * @param[in, out] flow
10542 * Pointer to flow structure.
10545 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
10547 struct mlx5_flow_handle *dh;
10548 uint32_t handle_idx;
10549 struct mlx5_priv *priv = dev->data->dev_private;
10553 handle_idx = flow->dev_handles;
10554 while (handle_idx) {
10555 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
10559 if (dh->drv_flow) {
10560 claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
10561 dh->drv_flow = NULL;
10563 if (dh->fate_action == MLX5_FLOW_FATE_DROP ||
10564 dh->fate_action == MLX5_FLOW_FATE_QUEUE ||
10565 dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS)
10566 flow_dv_fate_resource_release(dev, dh);
10567 if (dh->vf_vlan.tag && dh->vf_vlan.created)
10568 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
10569 handle_idx = dh->next.next;
10574 * Remove the flow from the NIC and the memory.
10575 * Lock free, (mutex should be acquired by caller).
10578 * Pointer to the Ethernet device structure.
10579 * @param[in, out] flow
10580 * Pointer to flow structure.
10583 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
10585 struct mlx5_flow_handle *dev_handle;
10586 struct mlx5_priv *priv = dev->data->dev_private;
10590 __flow_dv_remove(dev, flow);
10591 if (flow->counter) {
10592 flow_dv_counter_release(dev, flow->counter);
10596 struct mlx5_flow_meter *fm;
10598 fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
10601 mlx5_flow_meter_detach(fm);
10604 while (flow->dev_handles) {
10605 uint32_t tmp_idx = flow->dev_handles;
10607 dev_handle = mlx5_ipool_get(priv->sh->ipool
10608 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
10611 flow->dev_handles = dev_handle->next.next;
10612 if (dev_handle->dvh.matcher)
10613 flow_dv_matcher_release(dev, dev_handle);
10614 if (dev_handle->dvh.rix_sample)
10615 flow_dv_sample_resource_release(dev, dev_handle);
10616 if (dev_handle->dvh.rix_dest_array)
10617 flow_dv_dest_array_resource_release(dev, dev_handle);
10618 if (dev_handle->dvh.rix_encap_decap)
10619 flow_dv_encap_decap_resource_release(dev,
10620 dev_handle->dvh.rix_encap_decap);
10621 if (dev_handle->dvh.modify_hdr)
10622 flow_dv_modify_hdr_resource_release(dev, dev_handle);
10623 if (dev_handle->dvh.rix_push_vlan)
10624 flow_dv_push_vlan_action_resource_release(dev,
10626 if (dev_handle->dvh.rix_tag)
10627 flow_dv_tag_release(dev,
10628 dev_handle->dvh.rix_tag);
10629 flow_dv_fate_resource_release(dev, dev_handle);
10630 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
10636 * Query a dv flow rule for its statistics via devx.
10639 * Pointer to Ethernet device.
10641 * Pointer to the sub flow.
10643 * data retrieved by the query.
10644 * @param[out] error
10645 * Perform verbose error reporting if not NULL.
10648 * 0 on success, a negative errno value otherwise and rte_errno is set.
10651 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
10652 void *data, struct rte_flow_error *error)
10654 struct mlx5_priv *priv = dev->data->dev_private;
10655 struct rte_flow_query_count *qc = data;
10657 if (!priv->config.devx)
10658 return rte_flow_error_set(error, ENOTSUP,
10659 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10661 "counters are not supported");
10662 if (flow->counter) {
10663 uint64_t pkts, bytes;
10664 struct mlx5_flow_counter *cnt;
10666 cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
10668 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
10672 return rte_flow_error_set(error, -err,
10673 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10674 NULL, "cannot read counters");
10677 qc->hits = pkts - cnt->hits;
10678 qc->bytes = bytes - cnt->bytes;
10681 cnt->bytes = bytes;
10685 return rte_flow_error_set(error, EINVAL,
10686 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10688 "counters are not available");
10692 * Query a flow rule AGE action for aging information.
10695 * Pointer to Ethernet device.
10697 * Pointer to the sub flow.
10699 * data retrieved by the query.
10700 * @param[out] error
10701 * Perform verbose error reporting if not NULL.
10704 * 0 on success, a negative errno value otherwise and rte_errno is set.
10707 flow_dv_query_age(struct rte_eth_dev *dev, struct rte_flow *flow,
10708 void *data, struct rte_flow_error *error)
10710 struct rte_flow_query_age *resp = data;
10712 if (flow->counter) {
10713 struct mlx5_age_param *age_param =
10714 flow_dv_counter_idx_get_age(dev, flow->counter);
10716 if (!age_param || !age_param->timeout)
10717 return rte_flow_error_set
10719 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10720 NULL, "cannot read age data");
10721 resp->aged = __atomic_load_n(&age_param->state,
10722 __ATOMIC_RELAXED) ==
10724 resp->sec_since_last_hit_valid = !resp->aged;
10725 if (resp->sec_since_last_hit_valid)
10726 resp->sec_since_last_hit =
10727 __atomic_load_n(&age_param->sec_since_last_hit,
10731 return rte_flow_error_set(error, EINVAL,
10732 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10734 "age data not available");
10740 * @see rte_flow_query()
10741 * @see rte_flow_ops
10744 flow_dv_query(struct rte_eth_dev *dev,
10745 struct rte_flow *flow __rte_unused,
10746 const struct rte_flow_action *actions __rte_unused,
10747 void *data __rte_unused,
10748 struct rte_flow_error *error __rte_unused)
10752 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
10753 switch (actions->type) {
10754 case RTE_FLOW_ACTION_TYPE_VOID:
10756 case RTE_FLOW_ACTION_TYPE_COUNT:
10757 ret = flow_dv_query_count(dev, flow, data, error);
10759 case RTE_FLOW_ACTION_TYPE_AGE:
10760 ret = flow_dv_query_age(dev, flow, data, error);
10763 return rte_flow_error_set(error, ENOTSUP,
10764 RTE_FLOW_ERROR_TYPE_ACTION,
10766 "action not supported");
10773 * Destroy the meter table set.
10774 * Lock free, (mutex should be acquired by caller).
10777 * Pointer to Ethernet device.
10779 * Pointer to the meter table set.
10785 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
10786 struct mlx5_meter_domains_infos *tbl)
10788 struct mlx5_priv *priv = dev->data->dev_private;
10789 struct mlx5_meter_domains_infos *mtd =
10790 (struct mlx5_meter_domains_infos *)tbl;
10792 if (!mtd || !priv->config.dv_flow_en)
10794 if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
10795 claim_zero(mlx5_flow_os_destroy_flow
10796 (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
10797 if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
10798 claim_zero(mlx5_flow_os_destroy_flow
10799 (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
10800 if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
10801 claim_zero(mlx5_flow_os_destroy_flow
10802 (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
10803 if (mtd->egress.color_matcher)
10804 claim_zero(mlx5_flow_os_destroy_flow_matcher
10805 (mtd->egress.color_matcher));
10806 if (mtd->egress.any_matcher)
10807 claim_zero(mlx5_flow_os_destroy_flow_matcher
10808 (mtd->egress.any_matcher));
10809 if (mtd->egress.tbl)
10810 flow_dv_tbl_resource_release(dev, mtd->egress.tbl);
10811 if (mtd->egress.sfx_tbl)
10812 flow_dv_tbl_resource_release(dev, mtd->egress.sfx_tbl);
10813 if (mtd->ingress.color_matcher)
10814 claim_zero(mlx5_flow_os_destroy_flow_matcher
10815 (mtd->ingress.color_matcher));
10816 if (mtd->ingress.any_matcher)
10817 claim_zero(mlx5_flow_os_destroy_flow_matcher
10818 (mtd->ingress.any_matcher));
10819 if (mtd->ingress.tbl)
10820 flow_dv_tbl_resource_release(dev, mtd->ingress.tbl);
10821 if (mtd->ingress.sfx_tbl)
10822 flow_dv_tbl_resource_release(dev, mtd->ingress.sfx_tbl);
10823 if (mtd->transfer.color_matcher)
10824 claim_zero(mlx5_flow_os_destroy_flow_matcher
10825 (mtd->transfer.color_matcher));
10826 if (mtd->transfer.any_matcher)
10827 claim_zero(mlx5_flow_os_destroy_flow_matcher
10828 (mtd->transfer.any_matcher));
10829 if (mtd->transfer.tbl)
10830 flow_dv_tbl_resource_release(dev, mtd->transfer.tbl);
10831 if (mtd->transfer.sfx_tbl)
10832 flow_dv_tbl_resource_release(dev, mtd->transfer.sfx_tbl);
10833 if (mtd->drop_actn)
10834 claim_zero(mlx5_flow_os_destroy_flow_action(mtd->drop_actn));
10839 /* Number of meter flow actions, count and jump or count and drop. */
10840 #define METER_ACTIONS 2
10843 * Create specify domain meter table and suffix table.
10846 * Pointer to Ethernet device.
10847 * @param[in,out] mtb
10848 * Pointer to DV meter table set.
10849 * @param[in] egress
10851 * @param[in] transfer
10853 * @param[in] color_reg_c_idx
10854 * Reg C index for color match.
10857 * 0 on success, -1 otherwise and rte_errno is set.
10860 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
10861 struct mlx5_meter_domains_infos *mtb,
10862 uint8_t egress, uint8_t transfer,
10863 uint32_t color_reg_c_idx)
10865 struct mlx5_priv *priv = dev->data->dev_private;
10866 struct mlx5_dev_ctx_shared *sh = priv->sh;
10867 struct mlx5_flow_dv_match_params mask = {
10868 .size = sizeof(mask.buf),
10870 struct mlx5_flow_dv_match_params value = {
10871 .size = sizeof(value.buf),
10873 struct mlx5dv_flow_matcher_attr dv_attr = {
10874 .type = IBV_FLOW_ATTR_NORMAL,
10876 .match_criteria_enable = 0,
10877 .match_mask = (void *)&mask,
10879 void *actions[METER_ACTIONS];
10880 struct mlx5_meter_domain_info *dtb;
10881 struct rte_flow_error error;
10886 dtb = &mtb->transfer;
10888 dtb = &mtb->egress;
10890 dtb = &mtb->ingress;
10891 /* Create the meter table with METER level. */
10892 dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
10893 egress, transfer, &error);
10895 DRV_LOG(ERR, "Failed to create meter policer table.");
10898 /* Create the meter suffix table with SUFFIX level. */
10899 dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
10900 MLX5_FLOW_TABLE_LEVEL_SUFFIX,
10901 egress, transfer, &error);
10902 if (!dtb->sfx_tbl) {
10903 DRV_LOG(ERR, "Failed to create meter suffix table.");
10906 /* Create matchers, Any and Color. */
10907 dv_attr.priority = 3;
10908 dv_attr.match_criteria_enable = 0;
10909 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
10910 &dtb->any_matcher);
10912 DRV_LOG(ERR, "Failed to create meter"
10913 " policer default matcher.");
10916 dv_attr.priority = 0;
10917 dv_attr.match_criteria_enable =
10918 1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
10919 flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
10920 rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
10921 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
10922 &dtb->color_matcher);
10924 DRV_LOG(ERR, "Failed to create meter policer color matcher.");
10927 if (mtb->count_actns[RTE_MTR_DROPPED])
10928 actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
10929 actions[i++] = mtb->drop_actn;
10930 /* Default rule: lowest priority, match any, actions: drop. */
10931 ret = mlx5_flow_os_create_flow(dtb->any_matcher, (void *)&value, i,
10933 &dtb->policer_rules[RTE_MTR_DROPPED]);
10935 DRV_LOG(ERR, "Failed to create meter policer drop rule.");
10944 * Create the needed meter and suffix tables.
10945 * Lock free, (mutex should be acquired by caller).
10948 * Pointer to Ethernet device.
10950 * Pointer to the flow meter.
10953 * Pointer to table set on success, NULL otherwise and rte_errno is set.
10955 static struct mlx5_meter_domains_infos *
10956 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
10957 const struct mlx5_flow_meter *fm)
10959 struct mlx5_priv *priv = dev->data->dev_private;
10960 struct mlx5_meter_domains_infos *mtb;
10964 if (!priv->mtr_en) {
10965 rte_errno = ENOTSUP;
10968 mtb = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*mtb), 0, SOCKET_ID_ANY);
10970 DRV_LOG(ERR, "Failed to allocate memory for meter.");
10973 /* Create meter count actions */
10974 for (i = 0; i <= RTE_MTR_DROPPED; i++) {
10975 struct mlx5_flow_counter *cnt;
10976 if (!fm->policer_stats.cnt[i])
10978 cnt = flow_dv_counter_get_by_idx(dev,
10979 fm->policer_stats.cnt[i], NULL);
10980 mtb->count_actns[i] = cnt->action;
10982 /* Create drop action. */
10983 ret = mlx5_flow_os_create_flow_action_drop(&mtb->drop_actn);
10985 DRV_LOG(ERR, "Failed to create drop action.");
10988 /* Egress meter table. */
10989 ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
10991 DRV_LOG(ERR, "Failed to prepare egress meter table.");
10994 /* Ingress meter table. */
10995 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
10997 DRV_LOG(ERR, "Failed to prepare ingress meter table.");
11000 /* FDB meter table. */
11001 if (priv->config.dv_esw_en) {
11002 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
11003 priv->mtr_color_reg);
11005 DRV_LOG(ERR, "Failed to prepare fdb meter table.");
11011 flow_dv_destroy_mtr_tbl(dev, mtb);
11016 * Destroy domain policer rule.
11019 * Pointer to domain table.
11022 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
11026 for (i = 0; i < RTE_MTR_DROPPED; i++) {
11027 if (dt->policer_rules[i]) {
11028 claim_zero(mlx5_flow_os_destroy_flow
11029 (dt->policer_rules[i]));
11030 dt->policer_rules[i] = NULL;
11033 if (dt->jump_actn) {
11034 claim_zero(mlx5_flow_os_destroy_flow_action(dt->jump_actn));
11035 dt->jump_actn = NULL;
11040 * Destroy policer rules.
11043 * Pointer to Ethernet device.
11045 * Pointer to flow meter structure.
11047 * Pointer to flow attributes.
11053 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
11054 const struct mlx5_flow_meter *fm,
11055 const struct rte_flow_attr *attr)
11057 struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
11062 flow_dv_destroy_domain_policer_rule(&mtb->egress);
11064 flow_dv_destroy_domain_policer_rule(&mtb->ingress);
11065 if (attr->transfer)
11066 flow_dv_destroy_domain_policer_rule(&mtb->transfer);
11071 * Create specify domain meter policer rule.
11074 * Pointer to flow meter structure.
11076 * Pointer to DV meter table set.
11077 * @param[in] mtr_reg_c
11078 * Color match REG_C.
11081 * 0 on success, -1 otherwise.
11084 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
11085 struct mlx5_meter_domain_info *dtb,
11088 struct mlx5_flow_dv_match_params matcher = {
11089 .size = sizeof(matcher.buf),
11091 struct mlx5_flow_dv_match_params value = {
11092 .size = sizeof(value.buf),
11094 struct mlx5_meter_domains_infos *mtb = fm->mfts;
11095 void *actions[METER_ACTIONS];
11099 /* Create jump action. */
11100 if (!dtb->jump_actn)
11101 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
11102 (dtb->sfx_tbl->obj, &dtb->jump_actn);
11104 DRV_LOG(ERR, "Failed to create policer jump action.");
11107 for (i = 0; i < RTE_MTR_DROPPED; i++) {
11110 flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
11111 rte_col_2_mlx5_col(i), UINT8_MAX);
11112 if (mtb->count_actns[i])
11113 actions[j++] = mtb->count_actns[i];
11114 if (fm->action[i] == MTR_POLICER_ACTION_DROP)
11115 actions[j++] = mtb->drop_actn;
11117 actions[j++] = dtb->jump_actn;
11118 ret = mlx5_flow_os_create_flow(dtb->color_matcher,
11119 (void *)&value, j, actions,
11120 &dtb->policer_rules[i]);
11122 DRV_LOG(ERR, "Failed to create policer rule.");
11133 * Create policer rules.
11136 * Pointer to Ethernet device.
11138 * Pointer to flow meter structure.
11140 * Pointer to flow attributes.
11143 * 0 on success, -1 otherwise.
11146 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
11147 struct mlx5_flow_meter *fm,
11148 const struct rte_flow_attr *attr)
11150 struct mlx5_priv *priv = dev->data->dev_private;
11151 struct mlx5_meter_domains_infos *mtb = fm->mfts;
11154 if (attr->egress) {
11155 ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
11156 priv->mtr_color_reg);
11158 DRV_LOG(ERR, "Failed to create egress policer.");
11162 if (attr->ingress) {
11163 ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
11164 priv->mtr_color_reg);
11166 DRV_LOG(ERR, "Failed to create ingress policer.");
11170 if (attr->transfer) {
11171 ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
11172 priv->mtr_color_reg);
11174 DRV_LOG(ERR, "Failed to create transfer policer.");
11180 flow_dv_destroy_policer_rules(dev, fm, attr);
11185 * Validate the batch counter support in root table.
11187 * Create a simple flow with invalid counter and drop action on root table to
11188 * validate if batch counter with offset on root table is supported or not.
11191 * Pointer to rte_eth_dev structure.
11194 * 0 on success, a negative errno value otherwise and rte_errno is set.
11197 mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev)
11199 struct mlx5_priv *priv = dev->data->dev_private;
11200 struct mlx5_dev_ctx_shared *sh = priv->sh;
11201 struct mlx5_flow_dv_match_params mask = {
11202 .size = sizeof(mask.buf),
11204 struct mlx5_flow_dv_match_params value = {
11205 .size = sizeof(value.buf),
11207 struct mlx5dv_flow_matcher_attr dv_attr = {
11208 .type = IBV_FLOW_ATTR_NORMAL,
11210 .match_criteria_enable = 0,
11211 .match_mask = (void *)&mask,
11213 void *actions[2] = { 0 };
11214 struct mlx5_flow_tbl_resource *tbl = NULL, *dest_tbl = NULL;
11215 struct mlx5_devx_obj *dcs = NULL;
11216 void *matcher = NULL;
11220 tbl = flow_dv_tbl_resource_get(dev, 0, 0, 0, NULL);
11223 dest_tbl = flow_dv_tbl_resource_get(dev, 1, 0, 0, NULL);
11226 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
11229 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, UINT16_MAX,
11233 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
11234 (dest_tbl->obj, &actions[1]);
11237 dv_attr.match_criteria_enable = flow_dv_matcher_enable(mask.buf);
11238 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
11242 ret = mlx5_flow_os_create_flow(matcher, (void *)&value, 2,
11246 * If batch counter with offset is not supported, the driver will not
11247 * validate the invalid offset value, flow create should success.
11248 * In this case, it means batch counter is not supported in root table.
11250 * Otherwise, if flow create is failed, counter offset is supported.
11253 DRV_LOG(INFO, "Batch counter is not supported in root "
11254 "table. Switch to fallback mode.");
11255 rte_errno = ENOTSUP;
11257 claim_zero(mlx5_flow_os_destroy_flow(flow));
11259 /* Check matcher to make sure validate fail at flow create. */
11260 if (!matcher || (matcher && errno != EINVAL))
11261 DRV_LOG(ERR, "Unexpected error in counter offset "
11262 "support detection");
11265 for (i = 0; i < 2; i++) {
11267 claim_zero(mlx5_flow_os_destroy_flow_action
11271 claim_zero(mlx5_flow_os_destroy_flow_matcher(matcher));
11273 flow_dv_tbl_resource_release(dev, tbl);
11275 flow_dv_tbl_resource_release(dev, dest_tbl);
11277 claim_zero(mlx5_devx_cmd_destroy(dcs));
11282 * Query a devx counter.
11285 * Pointer to the Ethernet device structure.
11287 * Index to the flow counter.
11289 * Set to clear the counter statistics.
11291 * The statistics value of packets.
11292 * @param[out] bytes
11293 * The statistics value of bytes.
11296 * 0 on success, otherwise return -1.
11299 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
11300 uint64_t *pkts, uint64_t *bytes)
11302 struct mlx5_priv *priv = dev->data->dev_private;
11303 struct mlx5_flow_counter *cnt;
11304 uint64_t inn_pkts, inn_bytes;
11307 if (!priv->config.devx)
11310 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
11313 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
11314 *pkts = inn_pkts - cnt->hits;
11315 *bytes = inn_bytes - cnt->bytes;
11317 cnt->hits = inn_pkts;
11318 cnt->bytes = inn_bytes;
11324 * Get aged-out flows.
11327 * Pointer to the Ethernet device structure.
11328 * @param[in] context
11329 * The address of an array of pointers to the aged-out flows contexts.
11330 * @param[in] nb_contexts
11331 * The length of context array pointers.
11332 * @param[out] error
11333 * Perform verbose error reporting if not NULL. Initialized in case of
11337 * how many contexts get in success, otherwise negative errno value.
11338 * if nb_contexts is 0, return the amount of all aged contexts.
11339 * if nb_contexts is not 0 , return the amount of aged flows reported
11340 * in the context array.
11341 * @note: only stub for now
11344 flow_get_aged_flows(struct rte_eth_dev *dev,
11346 uint32_t nb_contexts,
11347 struct rte_flow_error *error)
11349 struct mlx5_priv *priv = dev->data->dev_private;
11350 struct mlx5_age_info *age_info;
11351 struct mlx5_age_param *age_param;
11352 struct mlx5_flow_counter *counter;
11355 if (nb_contexts && !context)
11356 return rte_flow_error_set(error, EINVAL,
11357 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11359 "Should assign at least one flow or"
11360 " context to get if nb_contexts != 0");
11361 age_info = GET_PORT_AGE_INFO(priv);
11362 rte_spinlock_lock(&age_info->aged_sl);
11363 TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
11366 age_param = MLX5_CNT_TO_AGE(counter);
11367 context[nb_flows - 1] = age_param->context;
11368 if (!(--nb_contexts))
11372 rte_spinlock_unlock(&age_info->aged_sl);
11373 MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
11378 * Mutex-protected thunk to lock-free __flow_dv_translate().
11381 flow_dv_translate(struct rte_eth_dev *dev,
11382 struct mlx5_flow *dev_flow,
11383 const struct rte_flow_attr *attr,
11384 const struct rte_flow_item items[],
11385 const struct rte_flow_action actions[],
11386 struct rte_flow_error *error)
11390 flow_dv_shared_lock(dev);
11391 ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
11392 flow_dv_shared_unlock(dev);
11397 * Mutex-protected thunk to lock-free __flow_dv_apply().
11400 flow_dv_apply(struct rte_eth_dev *dev,
11401 struct rte_flow *flow,
11402 struct rte_flow_error *error)
11406 flow_dv_shared_lock(dev);
11407 ret = __flow_dv_apply(dev, flow, error);
11408 flow_dv_shared_unlock(dev);
11413 * Mutex-protected thunk to lock-free __flow_dv_remove().
11416 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
11418 flow_dv_shared_lock(dev);
11419 __flow_dv_remove(dev, flow);
11420 flow_dv_shared_unlock(dev);
11424 * Mutex-protected thunk to lock-free __flow_dv_destroy().
11427 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
11429 flow_dv_shared_lock(dev);
11430 __flow_dv_destroy(dev, flow);
11431 flow_dv_shared_unlock(dev);
11435 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
11438 flow_dv_counter_allocate(struct rte_eth_dev *dev)
11442 flow_dv_shared_lock(dev);
11443 cnt = flow_dv_counter_alloc(dev, 0);
11444 flow_dv_shared_unlock(dev);
11449 * Mutex-protected thunk to lock-free flow_dv_counter_release().
11452 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
11454 flow_dv_shared_lock(dev);
11455 flow_dv_counter_release(dev, cnt);
11456 flow_dv_shared_unlock(dev);
11459 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
11460 .validate = flow_dv_validate,
11461 .prepare = flow_dv_prepare,
11462 .translate = flow_dv_translate,
11463 .apply = flow_dv_apply,
11464 .remove = flow_dv_remove,
11465 .destroy = flow_dv_destroy,
11466 .query = flow_dv_query,
11467 .create_mtr_tbls = flow_dv_create_mtr_tbl,
11468 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
11469 .create_policer_rules = flow_dv_create_policer_rules,
11470 .destroy_policer_rules = flow_dv_destroy_policer_rules,
11471 .counter_alloc = flow_dv_counter_allocate,
11472 .counter_free = flow_dv_counter_free,
11473 .counter_query = flow_dv_counter_query,
11474 .get_aged_flows = flow_get_aged_flows,
11477 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */