1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018 Mellanox Technologies, Ltd
12 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #pragma GCC diagnostic ignored "-Wpedantic"
16 #include <infiniband/verbs.h>
18 #pragma GCC diagnostic error "-Wpedantic"
21 #include <rte_common.h>
22 #include <rte_ether.h>
23 #include <rte_ethdev_driver.h>
25 #include <rte_flow_driver.h>
26 #include <rte_malloc.h>
29 #include <rte_vxlan.h>
32 #include <mlx5_glue.h>
33 #include <mlx5_devx_cmds.h>
36 #include "mlx5_defs.h"
38 #include "mlx5_flow.h"
39 #include "mlx5_rxtx.h"
41 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 #ifndef HAVE_IBV_FLOW_DEVX_COUNTERS
44 #define MLX5DV_FLOW_ACTION_COUNTERS_DEVX 0
47 #ifndef HAVE_MLX5DV_DR_ESWITCH
48 #ifndef MLX5DV_FLOW_TABLE_TYPE_FDB
49 #define MLX5DV_FLOW_TABLE_TYPE_FDB 0
53 #ifndef HAVE_MLX5DV_DR
54 #define MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL 1
57 /* VLAN header definitions */
58 #define MLX5DV_FLOW_VLAN_PCP_SHIFT 13
59 #define MLX5DV_FLOW_VLAN_PCP_MASK (0x7 << MLX5DV_FLOW_VLAN_PCP_SHIFT)
60 #define MLX5DV_FLOW_VLAN_VID_MASK 0x0fff
61 #define MLX5DV_FLOW_VLAN_PCP_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK)
62 #define MLX5DV_FLOW_VLAN_VID_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_VID_MASK)
77 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
78 struct mlx5_flow_tbl_resource *tbl);
81 * Initialize flow attributes structure according to flow items' types.
83 * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
84 * mode. For tunnel mode, the items to be modified are the outermost ones.
87 * Pointer to item specification.
89 * Pointer to flow attributes structure.
91 * Pointer to the sub flow.
92 * @param[in] tunnel_decap
93 * Whether action is after tunnel decapsulation.
96 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
97 struct mlx5_flow *dev_flow, bool tunnel_decap)
99 uint64_t layers = dev_flow->handle->layers;
102 * If layers is already initialized, it means this dev_flow is the
103 * suffix flow, the layers flags is set by the prefix flow. Need to
104 * use the layer flags from prefix flow as the suffix flow may not
105 * have the user defined items as the flow is split.
108 if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
110 else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
112 if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
114 else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
119 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
120 uint8_t next_protocol = 0xff;
121 switch (item->type) {
122 case RTE_FLOW_ITEM_TYPE_GRE:
123 case RTE_FLOW_ITEM_TYPE_NVGRE:
124 case RTE_FLOW_ITEM_TYPE_VXLAN:
125 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
126 case RTE_FLOW_ITEM_TYPE_GENEVE:
127 case RTE_FLOW_ITEM_TYPE_MPLS:
131 case RTE_FLOW_ITEM_TYPE_IPV4:
134 if (item->mask != NULL &&
135 ((const struct rte_flow_item_ipv4 *)
136 item->mask)->hdr.next_proto_id)
138 ((const struct rte_flow_item_ipv4 *)
139 (item->spec))->hdr.next_proto_id &
140 ((const struct rte_flow_item_ipv4 *)
141 (item->mask))->hdr.next_proto_id;
142 if ((next_protocol == IPPROTO_IPIP ||
143 next_protocol == IPPROTO_IPV6) && tunnel_decap)
146 case RTE_FLOW_ITEM_TYPE_IPV6:
149 if (item->mask != NULL &&
150 ((const struct rte_flow_item_ipv6 *)
151 item->mask)->hdr.proto)
153 ((const struct rte_flow_item_ipv6 *)
154 (item->spec))->hdr.proto &
155 ((const struct rte_flow_item_ipv6 *)
156 (item->mask))->hdr.proto;
157 if ((next_protocol == IPPROTO_IPIP ||
158 next_protocol == IPPROTO_IPV6) && tunnel_decap)
161 case RTE_FLOW_ITEM_TYPE_UDP:
165 case RTE_FLOW_ITEM_TYPE_TCP:
177 * Convert rte_mtr_color to mlx5 color.
186 rte_col_2_mlx5_col(enum rte_color rcol)
189 case RTE_COLOR_GREEN:
190 return MLX5_FLOW_COLOR_GREEN;
191 case RTE_COLOR_YELLOW:
192 return MLX5_FLOW_COLOR_YELLOW;
194 return MLX5_FLOW_COLOR_RED;
198 return MLX5_FLOW_COLOR_UNDEFINED;
201 struct field_modify_info {
202 uint32_t size; /* Size of field in protocol header, in bytes. */
203 uint32_t offset; /* Offset of field in protocol header, in bytes. */
204 enum mlx5_modification_field id;
207 struct field_modify_info modify_eth[] = {
208 {4, 0, MLX5_MODI_OUT_DMAC_47_16},
209 {2, 4, MLX5_MODI_OUT_DMAC_15_0},
210 {4, 6, MLX5_MODI_OUT_SMAC_47_16},
211 {2, 10, MLX5_MODI_OUT_SMAC_15_0},
215 struct field_modify_info modify_vlan_out_first_vid[] = {
216 /* Size in bits !!! */
217 {12, 0, MLX5_MODI_OUT_FIRST_VID},
221 struct field_modify_info modify_ipv4[] = {
222 {1, 1, MLX5_MODI_OUT_IP_DSCP},
223 {1, 8, MLX5_MODI_OUT_IPV4_TTL},
224 {4, 12, MLX5_MODI_OUT_SIPV4},
225 {4, 16, MLX5_MODI_OUT_DIPV4},
229 struct field_modify_info modify_ipv6[] = {
230 {1, 0, MLX5_MODI_OUT_IP_DSCP},
231 {1, 7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
232 {4, 8, MLX5_MODI_OUT_SIPV6_127_96},
233 {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
234 {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
235 {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
236 {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
237 {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
238 {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
239 {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
243 struct field_modify_info modify_udp[] = {
244 {2, 0, MLX5_MODI_OUT_UDP_SPORT},
245 {2, 2, MLX5_MODI_OUT_UDP_DPORT},
249 struct field_modify_info modify_tcp[] = {
250 {2, 0, MLX5_MODI_OUT_TCP_SPORT},
251 {2, 2, MLX5_MODI_OUT_TCP_DPORT},
252 {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
253 {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
258 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
259 uint8_t next_protocol, uint64_t *item_flags,
262 MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
263 item->type == RTE_FLOW_ITEM_TYPE_IPV6);
264 if (next_protocol == IPPROTO_IPIP) {
265 *item_flags |= MLX5_FLOW_LAYER_IPIP;
268 if (next_protocol == IPPROTO_IPV6) {
269 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
275 * Acquire the synchronizing object to protect multithreaded access
276 * to shared dv context. Lock occurs only if context is actually
277 * shared, i.e. we have multiport IB device and representors are
281 * Pointer to the rte_eth_dev structure.
284 flow_dv_shared_lock(struct rte_eth_dev *dev)
286 struct mlx5_priv *priv = dev->data->dev_private;
287 struct mlx5_ibv_shared *sh = priv->sh;
289 if (sh->dv_refcnt > 1) {
292 ret = pthread_mutex_lock(&sh->dv_mutex);
299 flow_dv_shared_unlock(struct rte_eth_dev *dev)
301 struct mlx5_priv *priv = dev->data->dev_private;
302 struct mlx5_ibv_shared *sh = priv->sh;
304 if (sh->dv_refcnt > 1) {
307 ret = pthread_mutex_unlock(&sh->dv_mutex);
313 /* Update VLAN's VID/PCP based on input rte_flow_action.
316 * Pointer to struct rte_flow_action.
318 * Pointer to struct rte_vlan_hdr.
321 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
322 struct rte_vlan_hdr *vlan)
325 if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
327 ((const struct rte_flow_action_of_set_vlan_pcp *)
328 action->conf)->vlan_pcp;
329 vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
330 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
331 vlan->vlan_tci |= vlan_tci;
332 } else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
333 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
334 vlan->vlan_tci |= rte_be_to_cpu_16
335 (((const struct rte_flow_action_of_set_vlan_vid *)
336 action->conf)->vlan_vid);
341 * Fetch 1, 2, 3 or 4 byte field from the byte array
342 * and return as unsigned integer in host-endian format.
345 * Pointer to data array.
347 * Size of field to extract.
350 * converted field in host endian format.
352 static inline uint32_t
353 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
362 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
365 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
366 ret = (ret << 8) | *(data + sizeof(uint16_t));
369 ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
380 * Convert modify-header action to DV specification.
382 * Data length of each action is determined by provided field description
383 * and the item mask. Data bit offset and width of each action is determined
384 * by provided item mask.
387 * Pointer to item specification.
389 * Pointer to field modification information.
390 * For MLX5_MODIFICATION_TYPE_SET specifies destination field.
391 * For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
392 * For MLX5_MODIFICATION_TYPE_COPY specifies source field.
394 * Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
395 * Negative offset value sets the same offset as source offset.
396 * size field is ignored, value is taken from source field.
397 * @param[in,out] resource
398 * Pointer to the modify-header resource.
400 * Type of modification.
402 * Pointer to the error structure.
405 * 0 on success, a negative errno value otherwise and rte_errno is set.
408 flow_dv_convert_modify_action(struct rte_flow_item *item,
409 struct field_modify_info *field,
410 struct field_modify_info *dcopy,
411 struct mlx5_flow_dv_modify_hdr_resource *resource,
412 uint32_t type, struct rte_flow_error *error)
414 uint32_t i = resource->actions_num;
415 struct mlx5_modification_cmd *actions = resource->actions;
418 * The item and mask are provided in big-endian format.
419 * The fields should be presented as in big-endian format either.
420 * Mask must be always present, it defines the actual field width.
422 MLX5_ASSERT(item->mask);
423 MLX5_ASSERT(field->size);
430 if (i >= MLX5_MAX_MODIFY_NUM)
431 return rte_flow_error_set(error, EINVAL,
432 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
433 "too many items to modify");
434 /* Fetch variable byte size mask from the array. */
435 mask = flow_dv_fetch_field((const uint8_t *)item->mask +
436 field->offset, field->size);
441 /* Deduce actual data width in bits from mask value. */
442 off_b = rte_bsf32(mask);
443 size_b = sizeof(uint32_t) * CHAR_BIT -
444 off_b - __builtin_clz(mask);
446 size_b = size_b == sizeof(uint32_t) * CHAR_BIT ? 0 : size_b;
447 actions[i] = (struct mlx5_modification_cmd) {
453 /* Convert entire record to expected big-endian format. */
454 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
455 if (type == MLX5_MODIFICATION_TYPE_COPY) {
457 actions[i].dst_field = dcopy->id;
458 actions[i].dst_offset =
459 (int)dcopy->offset < 0 ? off_b : dcopy->offset;
460 /* Convert entire record to big-endian format. */
461 actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
463 MLX5_ASSERT(item->spec);
464 data = flow_dv_fetch_field((const uint8_t *)item->spec +
465 field->offset, field->size);
466 /* Shift out the trailing masked bits from data. */
467 data = (data & mask) >> off_b;
468 actions[i].data1 = rte_cpu_to_be_32(data);
472 } while (field->size);
473 if (resource->actions_num == i)
474 return rte_flow_error_set(error, EINVAL,
475 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
476 "invalid modification flow item");
477 resource->actions_num = i;
482 * Convert modify-header set IPv4 address action to DV specification.
484 * @param[in,out] resource
485 * Pointer to the modify-header resource.
487 * Pointer to action specification.
489 * Pointer to the error structure.
492 * 0 on success, a negative errno value otherwise and rte_errno is set.
495 flow_dv_convert_action_modify_ipv4
496 (struct mlx5_flow_dv_modify_hdr_resource *resource,
497 const struct rte_flow_action *action,
498 struct rte_flow_error *error)
500 const struct rte_flow_action_set_ipv4 *conf =
501 (const struct rte_flow_action_set_ipv4 *)(action->conf);
502 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
503 struct rte_flow_item_ipv4 ipv4;
504 struct rte_flow_item_ipv4 ipv4_mask;
506 memset(&ipv4, 0, sizeof(ipv4));
507 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
508 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
509 ipv4.hdr.src_addr = conf->ipv4_addr;
510 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
512 ipv4.hdr.dst_addr = conf->ipv4_addr;
513 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
516 item.mask = &ipv4_mask;
517 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
518 MLX5_MODIFICATION_TYPE_SET, error);
522 * Convert modify-header set IPv6 address action to DV specification.
524 * @param[in,out] resource
525 * Pointer to the modify-header resource.
527 * Pointer to action specification.
529 * Pointer to the error structure.
532 * 0 on success, a negative errno value otherwise and rte_errno is set.
535 flow_dv_convert_action_modify_ipv6
536 (struct mlx5_flow_dv_modify_hdr_resource *resource,
537 const struct rte_flow_action *action,
538 struct rte_flow_error *error)
540 const struct rte_flow_action_set_ipv6 *conf =
541 (const struct rte_flow_action_set_ipv6 *)(action->conf);
542 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
543 struct rte_flow_item_ipv6 ipv6;
544 struct rte_flow_item_ipv6 ipv6_mask;
546 memset(&ipv6, 0, sizeof(ipv6));
547 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
548 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
549 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
550 sizeof(ipv6.hdr.src_addr));
551 memcpy(&ipv6_mask.hdr.src_addr,
552 &rte_flow_item_ipv6_mask.hdr.src_addr,
553 sizeof(ipv6.hdr.src_addr));
555 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
556 sizeof(ipv6.hdr.dst_addr));
557 memcpy(&ipv6_mask.hdr.dst_addr,
558 &rte_flow_item_ipv6_mask.hdr.dst_addr,
559 sizeof(ipv6.hdr.dst_addr));
562 item.mask = &ipv6_mask;
563 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
564 MLX5_MODIFICATION_TYPE_SET, error);
568 * Convert modify-header set MAC address action to DV specification.
570 * @param[in,out] resource
571 * Pointer to the modify-header resource.
573 * Pointer to action specification.
575 * Pointer to the error structure.
578 * 0 on success, a negative errno value otherwise and rte_errno is set.
581 flow_dv_convert_action_modify_mac
582 (struct mlx5_flow_dv_modify_hdr_resource *resource,
583 const struct rte_flow_action *action,
584 struct rte_flow_error *error)
586 const struct rte_flow_action_set_mac *conf =
587 (const struct rte_flow_action_set_mac *)(action->conf);
588 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
589 struct rte_flow_item_eth eth;
590 struct rte_flow_item_eth eth_mask;
592 memset(ð, 0, sizeof(eth));
593 memset(ð_mask, 0, sizeof(eth_mask));
594 if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
595 memcpy(ð.src.addr_bytes, &conf->mac_addr,
596 sizeof(eth.src.addr_bytes));
597 memcpy(ð_mask.src.addr_bytes,
598 &rte_flow_item_eth_mask.src.addr_bytes,
599 sizeof(eth_mask.src.addr_bytes));
601 memcpy(ð.dst.addr_bytes, &conf->mac_addr,
602 sizeof(eth.dst.addr_bytes));
603 memcpy(ð_mask.dst.addr_bytes,
604 &rte_flow_item_eth_mask.dst.addr_bytes,
605 sizeof(eth_mask.dst.addr_bytes));
608 item.mask = ð_mask;
609 return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
610 MLX5_MODIFICATION_TYPE_SET, error);
614 * Convert modify-header set VLAN VID action to DV specification.
616 * @param[in,out] resource
617 * Pointer to the modify-header resource.
619 * Pointer to action specification.
621 * Pointer to the error structure.
624 * 0 on success, a negative errno value otherwise and rte_errno is set.
627 flow_dv_convert_action_modify_vlan_vid
628 (struct mlx5_flow_dv_modify_hdr_resource *resource,
629 const struct rte_flow_action *action,
630 struct rte_flow_error *error)
632 const struct rte_flow_action_of_set_vlan_vid *conf =
633 (const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
634 int i = resource->actions_num;
635 struct mlx5_modification_cmd *actions = resource->actions;
636 struct field_modify_info *field = modify_vlan_out_first_vid;
638 if (i >= MLX5_MAX_MODIFY_NUM)
639 return rte_flow_error_set(error, EINVAL,
640 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
641 "too many items to modify");
642 actions[i] = (struct mlx5_modification_cmd) {
643 .action_type = MLX5_MODIFICATION_TYPE_SET,
645 .length = field->size,
646 .offset = field->offset,
648 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
649 actions[i].data1 = conf->vlan_vid;
650 actions[i].data1 = actions[i].data1 << 16;
651 resource->actions_num = ++i;
656 * Convert modify-header set TP action to DV specification.
658 * @param[in,out] resource
659 * Pointer to the modify-header resource.
661 * Pointer to action specification.
663 * Pointer to rte_flow_item objects list.
665 * Pointer to flow attributes structure.
666 * @param[in] dev_flow
667 * Pointer to the sub flow.
668 * @param[in] tunnel_decap
669 * Whether action is after tunnel decapsulation.
671 * Pointer to the error structure.
674 * 0 on success, a negative errno value otherwise and rte_errno is set.
677 flow_dv_convert_action_modify_tp
678 (struct mlx5_flow_dv_modify_hdr_resource *resource,
679 const struct rte_flow_action *action,
680 const struct rte_flow_item *items,
681 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
682 bool tunnel_decap, struct rte_flow_error *error)
684 const struct rte_flow_action_set_tp *conf =
685 (const struct rte_flow_action_set_tp *)(action->conf);
686 struct rte_flow_item item;
687 struct rte_flow_item_udp udp;
688 struct rte_flow_item_udp udp_mask;
689 struct rte_flow_item_tcp tcp;
690 struct rte_flow_item_tcp tcp_mask;
691 struct field_modify_info *field;
694 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
696 memset(&udp, 0, sizeof(udp));
697 memset(&udp_mask, 0, sizeof(udp_mask));
698 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
699 udp.hdr.src_port = conf->port;
700 udp_mask.hdr.src_port =
701 rte_flow_item_udp_mask.hdr.src_port;
703 udp.hdr.dst_port = conf->port;
704 udp_mask.hdr.dst_port =
705 rte_flow_item_udp_mask.hdr.dst_port;
707 item.type = RTE_FLOW_ITEM_TYPE_UDP;
709 item.mask = &udp_mask;
712 MLX5_ASSERT(attr->tcp);
713 memset(&tcp, 0, sizeof(tcp));
714 memset(&tcp_mask, 0, sizeof(tcp_mask));
715 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
716 tcp.hdr.src_port = conf->port;
717 tcp_mask.hdr.src_port =
718 rte_flow_item_tcp_mask.hdr.src_port;
720 tcp.hdr.dst_port = conf->port;
721 tcp_mask.hdr.dst_port =
722 rte_flow_item_tcp_mask.hdr.dst_port;
724 item.type = RTE_FLOW_ITEM_TYPE_TCP;
726 item.mask = &tcp_mask;
729 return flow_dv_convert_modify_action(&item, field, NULL, resource,
730 MLX5_MODIFICATION_TYPE_SET, error);
734 * Convert modify-header set TTL action to DV specification.
736 * @param[in,out] resource
737 * Pointer to the modify-header resource.
739 * Pointer to action specification.
741 * Pointer to rte_flow_item objects list.
743 * Pointer to flow attributes structure.
744 * @param[in] dev_flow
745 * Pointer to the sub flow.
746 * @param[in] tunnel_decap
747 * Whether action is after tunnel decapsulation.
749 * Pointer to the error structure.
752 * 0 on success, a negative errno value otherwise and rte_errno is set.
755 flow_dv_convert_action_modify_ttl
756 (struct mlx5_flow_dv_modify_hdr_resource *resource,
757 const struct rte_flow_action *action,
758 const struct rte_flow_item *items,
759 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
760 bool tunnel_decap, struct rte_flow_error *error)
762 const struct rte_flow_action_set_ttl *conf =
763 (const struct rte_flow_action_set_ttl *)(action->conf);
764 struct rte_flow_item item;
765 struct rte_flow_item_ipv4 ipv4;
766 struct rte_flow_item_ipv4 ipv4_mask;
767 struct rte_flow_item_ipv6 ipv6;
768 struct rte_flow_item_ipv6 ipv6_mask;
769 struct field_modify_info *field;
772 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
774 memset(&ipv4, 0, sizeof(ipv4));
775 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
776 ipv4.hdr.time_to_live = conf->ttl_value;
777 ipv4_mask.hdr.time_to_live = 0xFF;
778 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
780 item.mask = &ipv4_mask;
783 MLX5_ASSERT(attr->ipv6);
784 memset(&ipv6, 0, sizeof(ipv6));
785 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
786 ipv6.hdr.hop_limits = conf->ttl_value;
787 ipv6_mask.hdr.hop_limits = 0xFF;
788 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
790 item.mask = &ipv6_mask;
793 return flow_dv_convert_modify_action(&item, field, NULL, resource,
794 MLX5_MODIFICATION_TYPE_SET, error);
798 * Convert modify-header decrement TTL action to DV specification.
800 * @param[in,out] resource
801 * Pointer to the modify-header resource.
803 * Pointer to action specification.
805 * Pointer to rte_flow_item objects list.
807 * Pointer to flow attributes structure.
808 * @param[in] dev_flow
809 * Pointer to the sub flow.
810 * @param[in] tunnel_decap
811 * Whether action is after tunnel decapsulation.
813 * Pointer to the error structure.
816 * 0 on success, a negative errno value otherwise and rte_errno is set.
819 flow_dv_convert_action_modify_dec_ttl
820 (struct mlx5_flow_dv_modify_hdr_resource *resource,
821 const struct rte_flow_item *items,
822 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
823 bool tunnel_decap, struct rte_flow_error *error)
825 struct rte_flow_item item;
826 struct rte_flow_item_ipv4 ipv4;
827 struct rte_flow_item_ipv4 ipv4_mask;
828 struct rte_flow_item_ipv6 ipv6;
829 struct rte_flow_item_ipv6 ipv6_mask;
830 struct field_modify_info *field;
833 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
835 memset(&ipv4, 0, sizeof(ipv4));
836 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
837 ipv4.hdr.time_to_live = 0xFF;
838 ipv4_mask.hdr.time_to_live = 0xFF;
839 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
841 item.mask = &ipv4_mask;
844 MLX5_ASSERT(attr->ipv6);
845 memset(&ipv6, 0, sizeof(ipv6));
846 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
847 ipv6.hdr.hop_limits = 0xFF;
848 ipv6_mask.hdr.hop_limits = 0xFF;
849 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
851 item.mask = &ipv6_mask;
854 return flow_dv_convert_modify_action(&item, field, NULL, resource,
855 MLX5_MODIFICATION_TYPE_ADD, error);
859 * Convert modify-header increment/decrement TCP Sequence number
860 * to DV specification.
862 * @param[in,out] resource
863 * Pointer to the modify-header resource.
865 * Pointer to action specification.
867 * Pointer to the error structure.
870 * 0 on success, a negative errno value otherwise and rte_errno is set.
873 flow_dv_convert_action_modify_tcp_seq
874 (struct mlx5_flow_dv_modify_hdr_resource *resource,
875 const struct rte_flow_action *action,
876 struct rte_flow_error *error)
878 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
879 uint64_t value = rte_be_to_cpu_32(*conf);
880 struct rte_flow_item item;
881 struct rte_flow_item_tcp tcp;
882 struct rte_flow_item_tcp tcp_mask;
884 memset(&tcp, 0, sizeof(tcp));
885 memset(&tcp_mask, 0, sizeof(tcp_mask));
886 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
888 * The HW has no decrement operation, only increment operation.
889 * To simulate decrement X from Y using increment operation
890 * we need to add UINT32_MAX X times to Y.
891 * Each adding of UINT32_MAX decrements Y by 1.
894 tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
895 tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
896 item.type = RTE_FLOW_ITEM_TYPE_TCP;
898 item.mask = &tcp_mask;
899 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
900 MLX5_MODIFICATION_TYPE_ADD, error);
904 * Convert modify-header increment/decrement TCP Acknowledgment number
905 * to DV specification.
907 * @param[in,out] resource
908 * Pointer to the modify-header resource.
910 * Pointer to action specification.
912 * Pointer to the error structure.
915 * 0 on success, a negative errno value otherwise and rte_errno is set.
918 flow_dv_convert_action_modify_tcp_ack
919 (struct mlx5_flow_dv_modify_hdr_resource *resource,
920 const struct rte_flow_action *action,
921 struct rte_flow_error *error)
923 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
924 uint64_t value = rte_be_to_cpu_32(*conf);
925 struct rte_flow_item item;
926 struct rte_flow_item_tcp tcp;
927 struct rte_flow_item_tcp tcp_mask;
929 memset(&tcp, 0, sizeof(tcp));
930 memset(&tcp_mask, 0, sizeof(tcp_mask));
931 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
933 * The HW has no decrement operation, only increment operation.
934 * To simulate decrement X from Y using increment operation
935 * we need to add UINT32_MAX X times to Y.
936 * Each adding of UINT32_MAX decrements Y by 1.
939 tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
940 tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
941 item.type = RTE_FLOW_ITEM_TYPE_TCP;
943 item.mask = &tcp_mask;
944 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
945 MLX5_MODIFICATION_TYPE_ADD, error);
948 static enum mlx5_modification_field reg_to_field[] = {
949 [REG_NONE] = MLX5_MODI_OUT_NONE,
950 [REG_A] = MLX5_MODI_META_DATA_REG_A,
951 [REG_B] = MLX5_MODI_META_DATA_REG_B,
952 [REG_C_0] = MLX5_MODI_META_REG_C_0,
953 [REG_C_1] = MLX5_MODI_META_REG_C_1,
954 [REG_C_2] = MLX5_MODI_META_REG_C_2,
955 [REG_C_3] = MLX5_MODI_META_REG_C_3,
956 [REG_C_4] = MLX5_MODI_META_REG_C_4,
957 [REG_C_5] = MLX5_MODI_META_REG_C_5,
958 [REG_C_6] = MLX5_MODI_META_REG_C_6,
959 [REG_C_7] = MLX5_MODI_META_REG_C_7,
963 * Convert register set to DV specification.
965 * @param[in,out] resource
966 * Pointer to the modify-header resource.
968 * Pointer to action specification.
970 * Pointer to the error structure.
973 * 0 on success, a negative errno value otherwise and rte_errno is set.
976 flow_dv_convert_action_set_reg
977 (struct mlx5_flow_dv_modify_hdr_resource *resource,
978 const struct rte_flow_action *action,
979 struct rte_flow_error *error)
981 const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
982 struct mlx5_modification_cmd *actions = resource->actions;
983 uint32_t i = resource->actions_num;
985 if (i >= MLX5_MAX_MODIFY_NUM)
986 return rte_flow_error_set(error, EINVAL,
987 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
988 "too many items to modify");
989 MLX5_ASSERT(conf->id != REG_NONE);
990 MLX5_ASSERT(conf->id < RTE_DIM(reg_to_field));
991 actions[i] = (struct mlx5_modification_cmd) {
992 .action_type = MLX5_MODIFICATION_TYPE_SET,
993 .field = reg_to_field[conf->id],
995 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
996 actions[i].data1 = rte_cpu_to_be_32(conf->data);
998 resource->actions_num = i;
1003 * Convert SET_TAG action to DV specification.
1006 * Pointer to the rte_eth_dev structure.
1007 * @param[in,out] resource
1008 * Pointer to the modify-header resource.
1010 * Pointer to action specification.
1012 * Pointer to the error structure.
1015 * 0 on success, a negative errno value otherwise and rte_errno is set.
1018 flow_dv_convert_action_set_tag
1019 (struct rte_eth_dev *dev,
1020 struct mlx5_flow_dv_modify_hdr_resource *resource,
1021 const struct rte_flow_action_set_tag *conf,
1022 struct rte_flow_error *error)
1024 rte_be32_t data = rte_cpu_to_be_32(conf->data);
1025 rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1026 struct rte_flow_item item = {
1030 struct field_modify_info reg_c_x[] = {
1033 enum mlx5_modification_field reg_type;
1036 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1039 MLX5_ASSERT(ret != REG_NONE);
1040 MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1041 reg_type = reg_to_field[ret];
1042 MLX5_ASSERT(reg_type > 0);
1043 reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1044 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1045 MLX5_MODIFICATION_TYPE_SET, error);
1049 * Convert internal COPY_REG action to DV specification.
1052 * Pointer to the rte_eth_dev structure.
1053 * @param[in,out] res
1054 * Pointer to the modify-header resource.
1056 * Pointer to action specification.
1058 * Pointer to the error structure.
1061 * 0 on success, a negative errno value otherwise and rte_errno is set.
1064 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1065 struct mlx5_flow_dv_modify_hdr_resource *res,
1066 const struct rte_flow_action *action,
1067 struct rte_flow_error *error)
1069 const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1070 rte_be32_t mask = RTE_BE32(UINT32_MAX);
1071 struct rte_flow_item item = {
1075 struct field_modify_info reg_src[] = {
1076 {4, 0, reg_to_field[conf->src]},
1079 struct field_modify_info reg_dst = {
1081 .id = reg_to_field[conf->dst],
1083 /* Adjust reg_c[0] usage according to reported mask. */
1084 if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1085 struct mlx5_priv *priv = dev->data->dev_private;
1086 uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1088 MLX5_ASSERT(reg_c0);
1089 MLX5_ASSERT(priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY);
1090 if (conf->dst == REG_C_0) {
1091 /* Copy to reg_c[0], within mask only. */
1092 reg_dst.offset = rte_bsf32(reg_c0);
1094 * Mask is ignoring the enianness, because
1095 * there is no conversion in datapath.
1097 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1098 /* Copy from destination lower bits to reg_c[0]. */
1099 mask = reg_c0 >> reg_dst.offset;
1101 /* Copy from destination upper bits to reg_c[0]. */
1102 mask = reg_c0 << (sizeof(reg_c0) * CHAR_BIT -
1103 rte_fls_u32(reg_c0));
1106 mask = rte_cpu_to_be_32(reg_c0);
1107 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1108 /* Copy from reg_c[0] to destination lower bits. */
1111 /* Copy from reg_c[0] to destination upper bits. */
1112 reg_dst.offset = sizeof(reg_c0) * CHAR_BIT -
1113 (rte_fls_u32(reg_c0) -
1118 return flow_dv_convert_modify_action(&item,
1119 reg_src, ®_dst, res,
1120 MLX5_MODIFICATION_TYPE_COPY,
1125 * Convert MARK action to DV specification. This routine is used
1126 * in extensive metadata only and requires metadata register to be
1127 * handled. In legacy mode hardware tag resource is engaged.
1130 * Pointer to the rte_eth_dev structure.
1132 * Pointer to MARK action specification.
1133 * @param[in,out] resource
1134 * Pointer to the modify-header resource.
1136 * Pointer to the error structure.
1139 * 0 on success, a negative errno value otherwise and rte_errno is set.
1142 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1143 const struct rte_flow_action_mark *conf,
1144 struct mlx5_flow_dv_modify_hdr_resource *resource,
1145 struct rte_flow_error *error)
1147 struct mlx5_priv *priv = dev->data->dev_private;
1148 rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1149 priv->sh->dv_mark_mask);
1150 rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1151 struct rte_flow_item item = {
1155 struct field_modify_info reg_c_x[] = {
1156 {4, 0, 0}, /* dynamic instead of MLX5_MODI_META_REG_C_1. */
1162 return rte_flow_error_set(error, EINVAL,
1163 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1164 NULL, "zero mark action mask");
1165 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1168 MLX5_ASSERT(reg > 0);
1169 if (reg == REG_C_0) {
1170 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1171 uint32_t shl_c0 = rte_bsf32(msk_c0);
1173 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1174 mask = rte_cpu_to_be_32(mask) & msk_c0;
1175 mask = rte_cpu_to_be_32(mask << shl_c0);
1177 reg_c_x[0].id = reg_to_field[reg];
1178 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1179 MLX5_MODIFICATION_TYPE_SET, error);
1183 * Get metadata register index for specified steering domain.
1186 * Pointer to the rte_eth_dev structure.
1188 * Attributes of flow to determine steering domain.
1190 * Pointer to the error structure.
1193 * positive index on success, a negative errno value otherwise
1194 * and rte_errno is set.
1196 static enum modify_reg
1197 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1198 const struct rte_flow_attr *attr,
1199 struct rte_flow_error *error)
1202 mlx5_flow_get_reg_id(dev, attr->transfer ?
1206 MLX5_METADATA_RX, 0, error);
1208 return rte_flow_error_set(error,
1209 ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1210 NULL, "unavailable "
1211 "metadata register");
1216 * Convert SET_META action to DV specification.
1219 * Pointer to the rte_eth_dev structure.
1220 * @param[in,out] resource
1221 * Pointer to the modify-header resource.
1223 * Attributes of flow that includes this item.
1225 * Pointer to action specification.
1227 * Pointer to the error structure.
1230 * 0 on success, a negative errno value otherwise and rte_errno is set.
1233 flow_dv_convert_action_set_meta
1234 (struct rte_eth_dev *dev,
1235 struct mlx5_flow_dv_modify_hdr_resource *resource,
1236 const struct rte_flow_attr *attr,
1237 const struct rte_flow_action_set_meta *conf,
1238 struct rte_flow_error *error)
1240 uint32_t data = conf->data;
1241 uint32_t mask = conf->mask;
1242 struct rte_flow_item item = {
1246 struct field_modify_info reg_c_x[] = {
1249 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1254 * In datapath code there is no endianness
1255 * coversions for perfromance reasons, all
1256 * pattern conversions are done in rte_flow.
1258 if (reg == REG_C_0) {
1259 struct mlx5_priv *priv = dev->data->dev_private;
1260 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1263 MLX5_ASSERT(msk_c0);
1264 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1265 shl_c0 = rte_bsf32(msk_c0);
1267 shl_c0 = sizeof(msk_c0) * CHAR_BIT - rte_fls_u32(msk_c0);
1271 MLX5_ASSERT(!(~msk_c0 & rte_cpu_to_be_32(mask)));
1273 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1274 /* The routine expects parameters in memory as big-endian ones. */
1275 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1276 MLX5_MODIFICATION_TYPE_SET, error);
1280 * Convert modify-header set IPv4 DSCP action to DV specification.
1282 * @param[in,out] resource
1283 * Pointer to the modify-header resource.
1285 * Pointer to action specification.
1287 * Pointer to the error structure.
1290 * 0 on success, a negative errno value otherwise and rte_errno is set.
1293 flow_dv_convert_action_modify_ipv4_dscp
1294 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1295 const struct rte_flow_action *action,
1296 struct rte_flow_error *error)
1298 const struct rte_flow_action_set_dscp *conf =
1299 (const struct rte_flow_action_set_dscp *)(action->conf);
1300 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1301 struct rte_flow_item_ipv4 ipv4;
1302 struct rte_flow_item_ipv4 ipv4_mask;
1304 memset(&ipv4, 0, sizeof(ipv4));
1305 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1306 ipv4.hdr.type_of_service = conf->dscp;
1307 ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1309 item.mask = &ipv4_mask;
1310 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1311 MLX5_MODIFICATION_TYPE_SET, error);
1315 * Convert modify-header set IPv6 DSCP action to DV specification.
1317 * @param[in,out] resource
1318 * Pointer to the modify-header resource.
1320 * Pointer to action specification.
1322 * Pointer to the error structure.
1325 * 0 on success, a negative errno value otherwise and rte_errno is set.
1328 flow_dv_convert_action_modify_ipv6_dscp
1329 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1330 const struct rte_flow_action *action,
1331 struct rte_flow_error *error)
1333 const struct rte_flow_action_set_dscp *conf =
1334 (const struct rte_flow_action_set_dscp *)(action->conf);
1335 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1336 struct rte_flow_item_ipv6 ipv6;
1337 struct rte_flow_item_ipv6 ipv6_mask;
1339 memset(&ipv6, 0, sizeof(ipv6));
1340 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1342 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1343 * rdma-core only accept the DSCP bits byte aligned start from
1344 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1345 * bits in IPv6 case as rdma-core requires byte aligned value.
1347 ipv6.hdr.vtc_flow = conf->dscp;
1348 ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1350 item.mask = &ipv6_mask;
1351 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1352 MLX5_MODIFICATION_TYPE_SET, error);
1356 * Validate MARK item.
1359 * Pointer to the rte_eth_dev structure.
1361 * Item specification.
1363 * Attributes of flow that includes this item.
1365 * Pointer to error structure.
1368 * 0 on success, a negative errno value otherwise and rte_errno is set.
1371 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1372 const struct rte_flow_item *item,
1373 const struct rte_flow_attr *attr __rte_unused,
1374 struct rte_flow_error *error)
1376 struct mlx5_priv *priv = dev->data->dev_private;
1377 struct mlx5_dev_config *config = &priv->config;
1378 const struct rte_flow_item_mark *spec = item->spec;
1379 const struct rte_flow_item_mark *mask = item->mask;
1380 const struct rte_flow_item_mark nic_mask = {
1381 .id = priv->sh->dv_mark_mask,
1385 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1386 return rte_flow_error_set(error, ENOTSUP,
1387 RTE_FLOW_ERROR_TYPE_ITEM, item,
1388 "extended metadata feature"
1390 if (!mlx5_flow_ext_mreg_supported(dev))
1391 return rte_flow_error_set(error, ENOTSUP,
1392 RTE_FLOW_ERROR_TYPE_ITEM, item,
1393 "extended metadata register"
1394 " isn't supported");
1396 return rte_flow_error_set(error, ENOTSUP,
1397 RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 "extended metadata register"
1399 " isn't available");
1400 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1404 return rte_flow_error_set(error, EINVAL,
1405 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1407 "data cannot be empty");
1408 if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1409 return rte_flow_error_set(error, EINVAL,
1410 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1412 "mark id exceeds the limit");
1416 return rte_flow_error_set(error, EINVAL,
1417 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1418 "mask cannot be zero");
1420 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1421 (const uint8_t *)&nic_mask,
1422 sizeof(struct rte_flow_item_mark),
1430 * Validate META item.
1433 * Pointer to the rte_eth_dev structure.
1435 * Item specification.
1437 * Attributes of flow that includes this item.
1439 * Pointer to error structure.
1442 * 0 on success, a negative errno value otherwise and rte_errno is set.
1445 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1446 const struct rte_flow_item *item,
1447 const struct rte_flow_attr *attr,
1448 struct rte_flow_error *error)
1450 struct mlx5_priv *priv = dev->data->dev_private;
1451 struct mlx5_dev_config *config = &priv->config;
1452 const struct rte_flow_item_meta *spec = item->spec;
1453 const struct rte_flow_item_meta *mask = item->mask;
1454 struct rte_flow_item_meta nic_mask = {
1461 return rte_flow_error_set(error, EINVAL,
1462 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1464 "data cannot be empty");
1465 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1466 if (!mlx5_flow_ext_mreg_supported(dev))
1467 return rte_flow_error_set(error, ENOTSUP,
1468 RTE_FLOW_ERROR_TYPE_ITEM, item,
1469 "extended metadata register"
1470 " isn't supported");
1471 reg = flow_dv_get_metadata_reg(dev, attr, error);
1475 return rte_flow_error_set(error, ENOTSUP,
1476 RTE_FLOW_ERROR_TYPE_ITEM, item,
1480 nic_mask.data = priv->sh->dv_meta_mask;
1483 mask = &rte_flow_item_meta_mask;
1485 return rte_flow_error_set(error, EINVAL,
1486 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1487 "mask cannot be zero");
1489 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1490 (const uint8_t *)&nic_mask,
1491 sizeof(struct rte_flow_item_meta),
1497 * Validate TAG item.
1500 * Pointer to the rte_eth_dev structure.
1502 * Item specification.
1504 * Attributes of flow that includes this item.
1506 * Pointer to error structure.
1509 * 0 on success, a negative errno value otherwise and rte_errno is set.
1512 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
1513 const struct rte_flow_item *item,
1514 const struct rte_flow_attr *attr __rte_unused,
1515 struct rte_flow_error *error)
1517 const struct rte_flow_item_tag *spec = item->spec;
1518 const struct rte_flow_item_tag *mask = item->mask;
1519 const struct rte_flow_item_tag nic_mask = {
1520 .data = RTE_BE32(UINT32_MAX),
1525 if (!mlx5_flow_ext_mreg_supported(dev))
1526 return rte_flow_error_set(error, ENOTSUP,
1527 RTE_FLOW_ERROR_TYPE_ITEM, item,
1528 "extensive metadata register"
1529 " isn't supported");
1531 return rte_flow_error_set(error, EINVAL,
1532 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1534 "data cannot be empty");
1536 mask = &rte_flow_item_tag_mask;
1538 return rte_flow_error_set(error, EINVAL,
1539 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1540 "mask cannot be zero");
1542 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1543 (const uint8_t *)&nic_mask,
1544 sizeof(struct rte_flow_item_tag),
1548 if (mask->index != 0xff)
1549 return rte_flow_error_set(error, EINVAL,
1550 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1551 "partial mask for tag index"
1552 " is not supported");
1553 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
1556 MLX5_ASSERT(ret != REG_NONE);
1561 * Validate vport item.
1564 * Pointer to the rte_eth_dev structure.
1566 * Item specification.
1568 * Attributes of flow that includes this item.
1569 * @param[in] item_flags
1570 * Bit-fields that holds the items detected until now.
1572 * Pointer to error structure.
1575 * 0 on success, a negative errno value otherwise and rte_errno is set.
1578 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
1579 const struct rte_flow_item *item,
1580 const struct rte_flow_attr *attr,
1581 uint64_t item_flags,
1582 struct rte_flow_error *error)
1584 const struct rte_flow_item_port_id *spec = item->spec;
1585 const struct rte_flow_item_port_id *mask = item->mask;
1586 const struct rte_flow_item_port_id switch_mask = {
1589 struct mlx5_priv *esw_priv;
1590 struct mlx5_priv *dev_priv;
1593 if (!attr->transfer)
1594 return rte_flow_error_set(error, EINVAL,
1595 RTE_FLOW_ERROR_TYPE_ITEM,
1597 "match on port id is valid only"
1598 " when transfer flag is enabled");
1599 if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
1600 return rte_flow_error_set(error, ENOTSUP,
1601 RTE_FLOW_ERROR_TYPE_ITEM, item,
1602 "multiple source ports are not"
1605 mask = &switch_mask;
1606 if (mask->id != 0xffffffff)
1607 return rte_flow_error_set(error, ENOTSUP,
1608 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1610 "no support for partial mask on"
1612 ret = mlx5_flow_item_acceptable
1613 (item, (const uint8_t *)mask,
1614 (const uint8_t *)&rte_flow_item_port_id_mask,
1615 sizeof(struct rte_flow_item_port_id),
1621 esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
1623 return rte_flow_error_set(error, rte_errno,
1624 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1625 "failed to obtain E-Switch info for"
1627 dev_priv = mlx5_dev_to_eswitch_info(dev);
1629 return rte_flow_error_set(error, rte_errno,
1630 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1632 "failed to obtain E-Switch info");
1633 if (esw_priv->domain_id != dev_priv->domain_id)
1634 return rte_flow_error_set(error, EINVAL,
1635 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1636 "cannot match on a port from a"
1637 " different E-Switch");
1642 * Validate GTP item.
1645 * Pointer to the rte_eth_dev structure.
1647 * Item specification.
1648 * @param[in] item_flags
1649 * Bit-fields that holds the items detected until now.
1651 * Pointer to error structure.
1654 * 0 on success, a negative errno value otherwise and rte_errno is set.
1657 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1658 const struct rte_flow_item *item,
1659 uint64_t item_flags,
1660 struct rte_flow_error *error)
1662 struct mlx5_priv *priv = dev->data->dev_private;
1663 const struct rte_flow_item_gtp *mask = item->mask;
1664 const struct rte_flow_item_gtp nic_mask = {
1666 .teid = RTE_BE32(0xffffffff),
1669 if (!priv->config.hca_attr.tunnel_stateless_gtp)
1670 return rte_flow_error_set(error, ENOTSUP,
1671 RTE_FLOW_ERROR_TYPE_ITEM, item,
1672 "GTP support is not enabled");
1673 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1674 return rte_flow_error_set(error, ENOTSUP,
1675 RTE_FLOW_ERROR_TYPE_ITEM, item,
1676 "multiple tunnel layers not"
1678 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1679 return rte_flow_error_set(error, EINVAL,
1680 RTE_FLOW_ERROR_TYPE_ITEM, item,
1681 "no outer UDP layer found");
1683 mask = &rte_flow_item_gtp_mask;
1684 return mlx5_flow_item_acceptable
1685 (item, (const uint8_t *)mask,
1686 (const uint8_t *)&nic_mask,
1687 sizeof(struct rte_flow_item_gtp),
1692 * Validate the pop VLAN action.
1695 * Pointer to the rte_eth_dev structure.
1696 * @param[in] action_flags
1697 * Holds the actions detected until now.
1699 * Pointer to the pop vlan action.
1700 * @param[in] item_flags
1701 * The items found in this flow rule.
1703 * Pointer to flow attributes.
1705 * Pointer to error structure.
1708 * 0 on success, a negative errno value otherwise and rte_errno is set.
1711 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
1712 uint64_t action_flags,
1713 const struct rte_flow_action *action,
1714 uint64_t item_flags,
1715 const struct rte_flow_attr *attr,
1716 struct rte_flow_error *error)
1718 const struct mlx5_priv *priv = dev->data->dev_private;
1722 if (!priv->sh->pop_vlan_action)
1723 return rte_flow_error_set(error, ENOTSUP,
1724 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1726 "pop vlan action is not supported");
1728 return rte_flow_error_set(error, ENOTSUP,
1729 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
1731 "pop vlan action not supported for "
1733 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1734 return rte_flow_error_set(error, ENOTSUP,
1735 RTE_FLOW_ERROR_TYPE_ACTION, action,
1736 "no support for multiple VLAN "
1738 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
1739 return rte_flow_error_set(error, ENOTSUP,
1740 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1742 "cannot pop vlan without a "
1743 "match on (outer) vlan in the flow");
1744 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1745 return rte_flow_error_set(error, EINVAL,
1746 RTE_FLOW_ERROR_TYPE_ACTION, action,
1747 "wrong action order, port_id should "
1748 "be after pop VLAN action");
1749 if (!attr->transfer && priv->representor)
1750 return rte_flow_error_set(error, ENOTSUP,
1751 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1752 "pop vlan action for VF representor "
1753 "not supported on NIC table");
1758 * Get VLAN default info from vlan match info.
1761 * the list of item specifications.
1763 * pointer VLAN info to fill to.
1766 * 0 on success, a negative errno value otherwise and rte_errno is set.
1769 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
1770 struct rte_vlan_hdr *vlan)
1772 const struct rte_flow_item_vlan nic_mask = {
1773 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
1774 MLX5DV_FLOW_VLAN_VID_MASK),
1775 .inner_type = RTE_BE16(0xffff),
1780 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1781 int type = items->type;
1783 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
1784 type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
1787 if (items->type != RTE_FLOW_ITEM_TYPE_END) {
1788 const struct rte_flow_item_vlan *vlan_m = items->mask;
1789 const struct rte_flow_item_vlan *vlan_v = items->spec;
1793 /* Only full match values are accepted */
1794 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
1795 MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
1796 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
1798 rte_be_to_cpu_16(vlan_v->tci &
1799 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
1801 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
1802 MLX5DV_FLOW_VLAN_VID_MASK_BE) {
1803 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
1805 rte_be_to_cpu_16(vlan_v->tci &
1806 MLX5DV_FLOW_VLAN_VID_MASK_BE);
1808 if (vlan_m->inner_type == nic_mask.inner_type)
1809 vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
1810 vlan_m->inner_type);
1815 * Validate the push VLAN action.
1818 * Pointer to the rte_eth_dev structure.
1819 * @param[in] action_flags
1820 * Holds the actions detected until now.
1821 * @param[in] item_flags
1822 * The items found in this flow rule.
1824 * Pointer to the action structure.
1826 * Pointer to flow attributes
1828 * Pointer to error structure.
1831 * 0 on success, a negative errno value otherwise and rte_errno is set.
1834 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
1835 uint64_t action_flags,
1836 const struct rte_flow_item_vlan *vlan_m,
1837 const struct rte_flow_action *action,
1838 const struct rte_flow_attr *attr,
1839 struct rte_flow_error *error)
1841 const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
1842 const struct mlx5_priv *priv = dev->data->dev_private;
1844 if (!attr->transfer && attr->ingress)
1845 return rte_flow_error_set(error, ENOTSUP,
1846 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1848 "push VLAN action not supported for "
1850 if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
1851 push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
1852 return rte_flow_error_set(error, EINVAL,
1853 RTE_FLOW_ERROR_TYPE_ACTION, action,
1854 "invalid vlan ethertype");
1855 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1856 return rte_flow_error_set(error, ENOTSUP,
1857 RTE_FLOW_ERROR_TYPE_ACTION, action,
1858 "no support for multiple VLAN "
1860 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1861 return rte_flow_error_set(error, EINVAL,
1862 RTE_FLOW_ERROR_TYPE_ACTION, action,
1863 "wrong action order, port_id should "
1864 "be after push VLAN");
1865 if (!attr->transfer && priv->representor)
1866 return rte_flow_error_set(error, ENOTSUP,
1867 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1868 "push vlan action for VF representor "
1869 "not supported on NIC table");
1871 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
1872 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
1873 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
1874 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
1875 !(mlx5_flow_find_action
1876 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
1877 return rte_flow_error_set(error, EINVAL,
1878 RTE_FLOW_ERROR_TYPE_ACTION, action,
1879 "not full match mask on VLAN PCP and "
1880 "there is no of_set_vlan_pcp action, "
1881 "push VLAN action cannot figure out "
1884 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
1885 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
1886 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
1887 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
1888 !(mlx5_flow_find_action
1889 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
1890 return rte_flow_error_set(error, EINVAL,
1891 RTE_FLOW_ERROR_TYPE_ACTION, action,
1892 "not full match mask on VLAN VID and "
1893 "there is no of_set_vlan_vid action, "
1894 "push VLAN action cannot figure out "
1901 * Validate the set VLAN PCP.
1903 * @param[in] action_flags
1904 * Holds the actions detected until now.
1905 * @param[in] actions
1906 * Pointer to the list of actions remaining in the flow rule.
1908 * Pointer to error structure.
1911 * 0 on success, a negative errno value otherwise and rte_errno is set.
1914 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
1915 const struct rte_flow_action actions[],
1916 struct rte_flow_error *error)
1918 const struct rte_flow_action *action = actions;
1919 const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
1921 if (conf->vlan_pcp > 7)
1922 return rte_flow_error_set(error, EINVAL,
1923 RTE_FLOW_ERROR_TYPE_ACTION, action,
1924 "VLAN PCP value is too big");
1925 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
1926 return rte_flow_error_set(error, ENOTSUP,
1927 RTE_FLOW_ERROR_TYPE_ACTION, action,
1928 "set VLAN PCP action must follow "
1929 "the push VLAN action");
1930 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
1931 return rte_flow_error_set(error, ENOTSUP,
1932 RTE_FLOW_ERROR_TYPE_ACTION, action,
1933 "Multiple VLAN PCP modification are "
1935 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1936 return rte_flow_error_set(error, EINVAL,
1937 RTE_FLOW_ERROR_TYPE_ACTION, action,
1938 "wrong action order, port_id should "
1939 "be after set VLAN PCP");
1944 * Validate the set VLAN VID.
1946 * @param[in] item_flags
1947 * Holds the items detected in this rule.
1948 * @param[in] action_flags
1949 * Holds the actions detected until now.
1950 * @param[in] actions
1951 * Pointer to the list of actions remaining in the flow rule.
1953 * Pointer to error structure.
1956 * 0 on success, a negative errno value otherwise and rte_errno is set.
1959 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
1960 uint64_t action_flags,
1961 const struct rte_flow_action actions[],
1962 struct rte_flow_error *error)
1964 const struct rte_flow_action *action = actions;
1965 const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
1967 if (conf->vlan_vid > RTE_BE16(0xFFE))
1968 return rte_flow_error_set(error, EINVAL,
1969 RTE_FLOW_ERROR_TYPE_ACTION, action,
1970 "VLAN VID value is too big");
1971 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
1972 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
1973 return rte_flow_error_set(error, ENOTSUP,
1974 RTE_FLOW_ERROR_TYPE_ACTION, action,
1975 "set VLAN VID action must follow push"
1976 " VLAN action or match on VLAN item");
1977 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
1978 return rte_flow_error_set(error, ENOTSUP,
1979 RTE_FLOW_ERROR_TYPE_ACTION, action,
1980 "Multiple VLAN VID modifications are "
1982 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1983 return rte_flow_error_set(error, EINVAL,
1984 RTE_FLOW_ERROR_TYPE_ACTION, action,
1985 "wrong action order, port_id should "
1986 "be after set VLAN VID");
1991 * Validate the FLAG action.
1994 * Pointer to the rte_eth_dev structure.
1995 * @param[in] action_flags
1996 * Holds the actions detected until now.
1998 * Pointer to flow attributes
2000 * Pointer to error structure.
2003 * 0 on success, a negative errno value otherwise and rte_errno is set.
2006 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2007 uint64_t action_flags,
2008 const struct rte_flow_attr *attr,
2009 struct rte_flow_error *error)
2011 struct mlx5_priv *priv = dev->data->dev_private;
2012 struct mlx5_dev_config *config = &priv->config;
2015 /* Fall back if no extended metadata register support. */
2016 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2017 return mlx5_flow_validate_action_flag(action_flags, attr,
2019 /* Extensive metadata mode requires registers. */
2020 if (!mlx5_flow_ext_mreg_supported(dev))
2021 return rte_flow_error_set(error, ENOTSUP,
2022 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2023 "no metadata registers "
2024 "to support flag action");
2025 if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2026 return rte_flow_error_set(error, ENOTSUP,
2027 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2028 "extended metadata register"
2029 " isn't available");
2030 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2033 MLX5_ASSERT(ret > 0);
2034 if (action_flags & MLX5_FLOW_ACTION_MARK)
2035 return rte_flow_error_set(error, EINVAL,
2036 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2037 "can't mark and flag in same flow");
2038 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2039 return rte_flow_error_set(error, EINVAL,
2040 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2042 " actions in same flow");
2047 * Validate MARK action.
2050 * Pointer to the rte_eth_dev structure.
2052 * Pointer to action.
2053 * @param[in] action_flags
2054 * Holds the actions detected until now.
2056 * Pointer to flow attributes
2058 * Pointer to error structure.
2061 * 0 on success, a negative errno value otherwise and rte_errno is set.
2064 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2065 const struct rte_flow_action *action,
2066 uint64_t action_flags,
2067 const struct rte_flow_attr *attr,
2068 struct rte_flow_error *error)
2070 struct mlx5_priv *priv = dev->data->dev_private;
2071 struct mlx5_dev_config *config = &priv->config;
2072 const struct rte_flow_action_mark *mark = action->conf;
2075 /* Fall back if no extended metadata register support. */
2076 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2077 return mlx5_flow_validate_action_mark(action, action_flags,
2079 /* Extensive metadata mode requires registers. */
2080 if (!mlx5_flow_ext_mreg_supported(dev))
2081 return rte_flow_error_set(error, ENOTSUP,
2082 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2083 "no metadata registers "
2084 "to support mark action");
2085 if (!priv->sh->dv_mark_mask)
2086 return rte_flow_error_set(error, ENOTSUP,
2087 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2088 "extended metadata register"
2089 " isn't available");
2090 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2093 MLX5_ASSERT(ret > 0);
2095 return rte_flow_error_set(error, EINVAL,
2096 RTE_FLOW_ERROR_TYPE_ACTION, action,
2097 "configuration cannot be null");
2098 if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2099 return rte_flow_error_set(error, EINVAL,
2100 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2102 "mark id exceeds the limit");
2103 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2104 return rte_flow_error_set(error, EINVAL,
2105 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2106 "can't flag and mark in same flow");
2107 if (action_flags & MLX5_FLOW_ACTION_MARK)
2108 return rte_flow_error_set(error, EINVAL,
2109 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2110 "can't have 2 mark actions in same"
2116 * Validate SET_META action.
2119 * Pointer to the rte_eth_dev structure.
2121 * Pointer to the action structure.
2122 * @param[in] action_flags
2123 * Holds the actions detected until now.
2125 * Pointer to flow attributes
2127 * Pointer to error structure.
2130 * 0 on success, a negative errno value otherwise and rte_errno is set.
2133 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2134 const struct rte_flow_action *action,
2135 uint64_t action_flags __rte_unused,
2136 const struct rte_flow_attr *attr,
2137 struct rte_flow_error *error)
2139 const struct rte_flow_action_set_meta *conf;
2140 uint32_t nic_mask = UINT32_MAX;
2143 if (!mlx5_flow_ext_mreg_supported(dev))
2144 return rte_flow_error_set(error, ENOTSUP,
2145 RTE_FLOW_ERROR_TYPE_ACTION, action,
2146 "extended metadata register"
2147 " isn't supported");
2148 reg = flow_dv_get_metadata_reg(dev, attr, error);
2151 if (reg != REG_A && reg != REG_B) {
2152 struct mlx5_priv *priv = dev->data->dev_private;
2154 nic_mask = priv->sh->dv_meta_mask;
2156 if (!(action->conf))
2157 return rte_flow_error_set(error, EINVAL,
2158 RTE_FLOW_ERROR_TYPE_ACTION, action,
2159 "configuration cannot be null");
2160 conf = (const struct rte_flow_action_set_meta *)action->conf;
2162 return rte_flow_error_set(error, EINVAL,
2163 RTE_FLOW_ERROR_TYPE_ACTION, action,
2164 "zero mask doesn't have any effect");
2165 if (conf->mask & ~nic_mask)
2166 return rte_flow_error_set(error, EINVAL,
2167 RTE_FLOW_ERROR_TYPE_ACTION, action,
2168 "meta data must be within reg C0");
2173 * Validate SET_TAG action.
2176 * Pointer to the rte_eth_dev structure.
2178 * Pointer to the action structure.
2179 * @param[in] action_flags
2180 * Holds the actions detected until now.
2182 * Pointer to flow attributes
2184 * Pointer to error structure.
2187 * 0 on success, a negative errno value otherwise and rte_errno is set.
2190 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2191 const struct rte_flow_action *action,
2192 uint64_t action_flags,
2193 const struct rte_flow_attr *attr,
2194 struct rte_flow_error *error)
2196 const struct rte_flow_action_set_tag *conf;
2197 const uint64_t terminal_action_flags =
2198 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2199 MLX5_FLOW_ACTION_RSS;
2202 if (!mlx5_flow_ext_mreg_supported(dev))
2203 return rte_flow_error_set(error, ENOTSUP,
2204 RTE_FLOW_ERROR_TYPE_ACTION, action,
2205 "extensive metadata register"
2206 " isn't supported");
2207 if (!(action->conf))
2208 return rte_flow_error_set(error, EINVAL,
2209 RTE_FLOW_ERROR_TYPE_ACTION, action,
2210 "configuration cannot be null");
2211 conf = (const struct rte_flow_action_set_tag *)action->conf;
2213 return rte_flow_error_set(error, EINVAL,
2214 RTE_FLOW_ERROR_TYPE_ACTION, action,
2215 "zero mask doesn't have any effect");
2216 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2219 if (!attr->transfer && attr->ingress &&
2220 (action_flags & terminal_action_flags))
2221 return rte_flow_error_set(error, EINVAL,
2222 RTE_FLOW_ERROR_TYPE_ACTION, action,
2223 "set_tag has no effect"
2224 " with terminal actions");
2229 * Validate count action.
2232 * Pointer to rte_eth_dev structure.
2234 * Pointer to error structure.
2237 * 0 on success, a negative errno value otherwise and rte_errno is set.
2240 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2241 struct rte_flow_error *error)
2243 struct mlx5_priv *priv = dev->data->dev_private;
2245 if (!priv->config.devx)
2247 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2251 return rte_flow_error_set
2253 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2255 "count action not supported");
2259 * Validate the L2 encap action.
2262 * Pointer to the rte_eth_dev structure.
2263 * @param[in] action_flags
2264 * Holds the actions detected until now.
2266 * Pointer to the action structure.
2268 * Pointer to flow attributes.
2270 * Pointer to error structure.
2273 * 0 on success, a negative errno value otherwise and rte_errno is set.
2276 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2277 uint64_t action_flags,
2278 const struct rte_flow_action *action,
2279 const struct rte_flow_attr *attr,
2280 struct rte_flow_error *error)
2282 const struct mlx5_priv *priv = dev->data->dev_private;
2284 if (!(action->conf))
2285 return rte_flow_error_set(error, EINVAL,
2286 RTE_FLOW_ERROR_TYPE_ACTION, action,
2287 "configuration cannot be null");
2288 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2289 return rte_flow_error_set(error, EINVAL,
2290 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2291 "can only have a single encap action "
2293 if (!attr->transfer && priv->representor)
2294 return rte_flow_error_set(error, ENOTSUP,
2295 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2296 "encap action for VF representor "
2297 "not supported on NIC table");
2302 * Validate a decap action.
2305 * Pointer to the rte_eth_dev structure.
2306 * @param[in] action_flags
2307 * Holds the actions detected until now.
2309 * Pointer to flow attributes
2311 * Pointer to error structure.
2314 * 0 on success, a negative errno value otherwise and rte_errno is set.
2317 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2318 uint64_t action_flags,
2319 const struct rte_flow_attr *attr,
2320 struct rte_flow_error *error)
2322 const struct mlx5_priv *priv = dev->data->dev_private;
2324 if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2325 return rte_flow_error_set(error, ENOTSUP,
2326 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2328 MLX5_FLOW_ACTION_DECAP ? "can only "
2329 "have a single decap action" : "decap "
2330 "after encap is not supported");
2331 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2332 return rte_flow_error_set(error, EINVAL,
2333 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2334 "can't have decap action after"
2337 return rte_flow_error_set(error, ENOTSUP,
2338 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2340 "decap action not supported for "
2342 if (!attr->transfer && priv->representor)
2343 return rte_flow_error_set(error, ENOTSUP,
2344 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2345 "decap action for VF representor "
2346 "not supported on NIC table");
2350 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2353 * Validate the raw encap and decap actions.
2356 * Pointer to the rte_eth_dev structure.
2358 * Pointer to the decap action.
2360 * Pointer to the encap action.
2362 * Pointer to flow attributes
2363 * @param[in/out] action_flags
2364 * Holds the actions detected until now.
2365 * @param[out] actions_n
2366 * pointer to the number of actions counter.
2368 * Pointer to error structure.
2371 * 0 on success, a negative errno value otherwise and rte_errno is set.
2374 flow_dv_validate_action_raw_encap_decap
2375 (struct rte_eth_dev *dev,
2376 const struct rte_flow_action_raw_decap *decap,
2377 const struct rte_flow_action_raw_encap *encap,
2378 const struct rte_flow_attr *attr, uint64_t *action_flags,
2379 int *actions_n, struct rte_flow_error *error)
2381 const struct mlx5_priv *priv = dev->data->dev_private;
2384 if (encap && (!encap->size || !encap->data))
2385 return rte_flow_error_set(error, EINVAL,
2386 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2387 "raw encap data cannot be empty");
2388 if (decap && encap) {
2389 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2390 encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2393 else if (encap->size <=
2394 MLX5_ENCAPSULATION_DECISION_SIZE &&
2396 MLX5_ENCAPSULATION_DECISION_SIZE)
2399 else if (encap->size >
2400 MLX5_ENCAPSULATION_DECISION_SIZE &&
2402 MLX5_ENCAPSULATION_DECISION_SIZE)
2403 /* 2 L2 actions: encap and decap. */
2406 return rte_flow_error_set(error,
2408 RTE_FLOW_ERROR_TYPE_ACTION,
2409 NULL, "unsupported too small "
2410 "raw decap and too small raw "
2411 "encap combination");
2414 ret = flow_dv_validate_action_decap(dev, *action_flags, attr,
2418 *action_flags |= MLX5_FLOW_ACTION_DECAP;
2422 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2423 return rte_flow_error_set(error, ENOTSUP,
2424 RTE_FLOW_ERROR_TYPE_ACTION,
2426 "small raw encap size");
2427 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2428 return rte_flow_error_set(error, EINVAL,
2429 RTE_FLOW_ERROR_TYPE_ACTION,
2431 "more than one encap action");
2432 if (!attr->transfer && priv->representor)
2433 return rte_flow_error_set
2435 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2436 "encap action for VF representor "
2437 "not supported on NIC table");
2438 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
2445 * Find existing encap/decap resource or create and register a new one.
2447 * @param[in, out] dev
2448 * Pointer to rte_eth_dev structure.
2449 * @param[in, out] resource
2450 * Pointer to encap/decap resource.
2451 * @parm[in, out] dev_flow
2452 * Pointer to the dev_flow.
2454 * pointer to error structure.
2457 * 0 on success otherwise -errno and errno is set.
2460 flow_dv_encap_decap_resource_register
2461 (struct rte_eth_dev *dev,
2462 struct mlx5_flow_dv_encap_decap_resource *resource,
2463 struct mlx5_flow *dev_flow,
2464 struct rte_flow_error *error)
2466 struct mlx5_priv *priv = dev->data->dev_private;
2467 struct mlx5_ibv_shared *sh = priv->sh;
2468 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2469 struct mlx5dv_dr_domain *domain;
2472 resource->flags = dev_flow->dv.group ? 0 : 1;
2473 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2474 domain = sh->fdb_domain;
2475 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2476 domain = sh->rx_domain;
2478 domain = sh->tx_domain;
2479 /* Lookup a matching resource from cache. */
2480 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], sh->encaps_decaps, idx,
2481 cache_resource, next) {
2482 if (resource->reformat_type == cache_resource->reformat_type &&
2483 resource->ft_type == cache_resource->ft_type &&
2484 resource->flags == cache_resource->flags &&
2485 resource->size == cache_resource->size &&
2486 !memcmp((const void *)resource->buf,
2487 (const void *)cache_resource->buf,
2489 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
2490 (void *)cache_resource,
2491 rte_atomic32_read(&cache_resource->refcnt));
2492 rte_atomic32_inc(&cache_resource->refcnt);
2493 dev_flow->handle->dvh.rix_encap_decap = idx;
2494 dev_flow->dv.encap_decap = cache_resource;
2498 /* Register new encap/decap resource. */
2499 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2500 &dev_flow->handle->dvh.rix_encap_decap);
2501 if (!cache_resource)
2502 return rte_flow_error_set(error, ENOMEM,
2503 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2504 "cannot allocate resource memory");
2505 *cache_resource = *resource;
2506 cache_resource->verbs_action =
2507 mlx5_glue->dv_create_flow_action_packet_reformat
2508 (sh->ctx, cache_resource->reformat_type,
2509 cache_resource->ft_type, domain, cache_resource->flags,
2510 cache_resource->size,
2511 (cache_resource->size ? cache_resource->buf : NULL));
2512 if (!cache_resource->verbs_action) {
2513 rte_free(cache_resource);
2514 return rte_flow_error_set(error, ENOMEM,
2515 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2516 NULL, "cannot create action");
2518 rte_atomic32_init(&cache_resource->refcnt);
2519 rte_atomic32_inc(&cache_resource->refcnt);
2520 ILIST_INSERT(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], &sh->encaps_decaps,
2521 dev_flow->handle->dvh.rix_encap_decap, cache_resource,
2523 dev_flow->dv.encap_decap = cache_resource;
2524 DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
2525 (void *)cache_resource,
2526 rte_atomic32_read(&cache_resource->refcnt));
2531 * Find existing table jump resource or create and register a new one.
2533 * @param[in, out] dev
2534 * Pointer to rte_eth_dev structure.
2535 * @param[in, out] tbl
2536 * Pointer to flow table resource.
2537 * @parm[in, out] dev_flow
2538 * Pointer to the dev_flow.
2540 * pointer to error structure.
2543 * 0 on success otherwise -errno and errno is set.
2546 flow_dv_jump_tbl_resource_register
2547 (struct rte_eth_dev *dev __rte_unused,
2548 struct mlx5_flow_tbl_resource *tbl,
2549 struct mlx5_flow *dev_flow,
2550 struct rte_flow_error *error)
2552 struct mlx5_flow_tbl_data_entry *tbl_data =
2553 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2557 cnt = rte_atomic32_read(&tbl_data->jump.refcnt);
2559 tbl_data->jump.action =
2560 mlx5_glue->dr_create_flow_action_dest_flow_tbl
2562 if (!tbl_data->jump.action)
2563 return rte_flow_error_set(error, ENOMEM,
2564 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2565 NULL, "cannot create jump action");
2566 DRV_LOG(DEBUG, "new jump table resource %p: refcnt %d++",
2567 (void *)&tbl_data->jump, cnt);
2569 /* old jump should not make the table ref++. */
2570 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
2571 MLX5_ASSERT(tbl_data->jump.action);
2572 DRV_LOG(DEBUG, "existed jump table resource %p: refcnt %d++",
2573 (void *)&tbl_data->jump, cnt);
2575 rte_atomic32_inc(&tbl_data->jump.refcnt);
2576 dev_flow->handle->rix_jump = tbl_data->idx;
2577 dev_flow->dv.jump = &tbl_data->jump;
2582 * Find existing table port ID resource or create and register a new one.
2584 * @param[in, out] dev
2585 * Pointer to rte_eth_dev structure.
2586 * @param[in, out] resource
2587 * Pointer to port ID action resource.
2588 * @parm[in, out] dev_flow
2589 * Pointer to the dev_flow.
2591 * pointer to error structure.
2594 * 0 on success otherwise -errno and errno is set.
2597 flow_dv_port_id_action_resource_register
2598 (struct rte_eth_dev *dev,
2599 struct mlx5_flow_dv_port_id_action_resource *resource,
2600 struct mlx5_flow *dev_flow,
2601 struct rte_flow_error *error)
2603 struct mlx5_priv *priv = dev->data->dev_private;
2604 struct mlx5_ibv_shared *sh = priv->sh;
2605 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
2608 /* Lookup a matching resource from cache. */
2609 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
2610 idx, cache_resource, next) {
2611 if (resource->port_id == cache_resource->port_id) {
2612 DRV_LOG(DEBUG, "port id action resource resource %p: "
2614 (void *)cache_resource,
2615 rte_atomic32_read(&cache_resource->refcnt));
2616 rte_atomic32_inc(&cache_resource->refcnt);
2617 dev_flow->handle->rix_port_id_action = idx;
2618 dev_flow->dv.port_id_action = cache_resource;
2622 /* Register new port id action resource. */
2623 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
2624 &dev_flow->handle->rix_port_id_action);
2625 if (!cache_resource)
2626 return rte_flow_error_set(error, ENOMEM,
2627 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2628 "cannot allocate resource memory");
2629 *cache_resource = *resource;
2631 * Depending on rdma_core version the glue routine calls
2632 * either mlx5dv_dr_action_create_dest_ib_port(domain, ibv_port)
2633 * or mlx5dv_dr_action_create_dest_vport(domain, vport_id).
2635 cache_resource->action =
2636 mlx5_glue->dr_create_flow_action_dest_port
2637 (priv->sh->fdb_domain, resource->port_id);
2638 if (!cache_resource->action) {
2639 rte_free(cache_resource);
2640 return rte_flow_error_set(error, ENOMEM,
2641 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2642 NULL, "cannot create action");
2644 rte_atomic32_init(&cache_resource->refcnt);
2645 rte_atomic32_inc(&cache_resource->refcnt);
2646 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
2647 dev_flow->handle->rix_port_id_action, cache_resource,
2649 dev_flow->dv.port_id_action = cache_resource;
2650 DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
2651 (void *)cache_resource,
2652 rte_atomic32_read(&cache_resource->refcnt));
2657 * Find existing push vlan resource or create and register a new one.
2659 * @param [in, out] dev
2660 * Pointer to rte_eth_dev structure.
2661 * @param[in, out] resource
2662 * Pointer to port ID action resource.
2663 * @parm[in, out] dev_flow
2664 * Pointer to the dev_flow.
2666 * pointer to error structure.
2669 * 0 on success otherwise -errno and errno is set.
2672 flow_dv_push_vlan_action_resource_register
2673 (struct rte_eth_dev *dev,
2674 struct mlx5_flow_dv_push_vlan_action_resource *resource,
2675 struct mlx5_flow *dev_flow,
2676 struct rte_flow_error *error)
2678 struct mlx5_priv *priv = dev->data->dev_private;
2679 struct mlx5_ibv_shared *sh = priv->sh;
2680 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
2681 struct mlx5dv_dr_domain *domain;
2684 /* Lookup a matching resource from cache. */
2685 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2686 sh->push_vlan_action_list, idx, cache_resource, next) {
2687 if (resource->vlan_tag == cache_resource->vlan_tag &&
2688 resource->ft_type == cache_resource->ft_type) {
2689 DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
2691 (void *)cache_resource,
2692 rte_atomic32_read(&cache_resource->refcnt));
2693 rte_atomic32_inc(&cache_resource->refcnt);
2694 dev_flow->handle->dvh.rix_push_vlan = idx;
2695 dev_flow->dv.push_vlan_res = cache_resource;
2699 /* Register new push_vlan action resource. */
2700 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2701 &dev_flow->handle->dvh.rix_push_vlan);
2702 if (!cache_resource)
2703 return rte_flow_error_set(error, ENOMEM,
2704 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2705 "cannot allocate resource memory");
2706 *cache_resource = *resource;
2707 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2708 domain = sh->fdb_domain;
2709 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2710 domain = sh->rx_domain;
2712 domain = sh->tx_domain;
2713 cache_resource->action =
2714 mlx5_glue->dr_create_flow_action_push_vlan(domain,
2715 resource->vlan_tag);
2716 if (!cache_resource->action) {
2717 rte_free(cache_resource);
2718 return rte_flow_error_set(error, ENOMEM,
2719 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2720 NULL, "cannot create action");
2722 rte_atomic32_init(&cache_resource->refcnt);
2723 rte_atomic32_inc(&cache_resource->refcnt);
2724 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2725 &sh->push_vlan_action_list,
2726 dev_flow->handle->dvh.rix_push_vlan,
2727 cache_resource, next);
2728 dev_flow->dv.push_vlan_res = cache_resource;
2729 DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
2730 (void *)cache_resource,
2731 rte_atomic32_read(&cache_resource->refcnt));
2735 * Get the size of specific rte_flow_item_type
2737 * @param[in] item_type
2738 * Tested rte_flow_item_type.
2741 * sizeof struct item_type, 0 if void or irrelevant.
2744 flow_dv_get_item_len(const enum rte_flow_item_type item_type)
2748 switch (item_type) {
2749 case RTE_FLOW_ITEM_TYPE_ETH:
2750 retval = sizeof(struct rte_flow_item_eth);
2752 case RTE_FLOW_ITEM_TYPE_VLAN:
2753 retval = sizeof(struct rte_flow_item_vlan);
2755 case RTE_FLOW_ITEM_TYPE_IPV4:
2756 retval = sizeof(struct rte_flow_item_ipv4);
2758 case RTE_FLOW_ITEM_TYPE_IPV6:
2759 retval = sizeof(struct rte_flow_item_ipv6);
2761 case RTE_FLOW_ITEM_TYPE_UDP:
2762 retval = sizeof(struct rte_flow_item_udp);
2764 case RTE_FLOW_ITEM_TYPE_TCP:
2765 retval = sizeof(struct rte_flow_item_tcp);
2767 case RTE_FLOW_ITEM_TYPE_VXLAN:
2768 retval = sizeof(struct rte_flow_item_vxlan);
2770 case RTE_FLOW_ITEM_TYPE_GRE:
2771 retval = sizeof(struct rte_flow_item_gre);
2773 case RTE_FLOW_ITEM_TYPE_NVGRE:
2774 retval = sizeof(struct rte_flow_item_nvgre);
2776 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2777 retval = sizeof(struct rte_flow_item_vxlan_gpe);
2779 case RTE_FLOW_ITEM_TYPE_MPLS:
2780 retval = sizeof(struct rte_flow_item_mpls);
2782 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
2790 #define MLX5_ENCAP_IPV4_VERSION 0x40
2791 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
2792 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
2793 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
2794 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
2795 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
2796 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
2799 * Convert the encap action data from list of rte_flow_item to raw buffer
2802 * Pointer to rte_flow_item objects list.
2804 * Pointer to the output buffer.
2806 * Pointer to the output buffer size.
2808 * Pointer to the error structure.
2811 * 0 on success, a negative errno value otherwise and rte_errno is set.
2814 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
2815 size_t *size, struct rte_flow_error *error)
2817 struct rte_ether_hdr *eth = NULL;
2818 struct rte_vlan_hdr *vlan = NULL;
2819 struct rte_ipv4_hdr *ipv4 = NULL;
2820 struct rte_ipv6_hdr *ipv6 = NULL;
2821 struct rte_udp_hdr *udp = NULL;
2822 struct rte_vxlan_hdr *vxlan = NULL;
2823 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
2824 struct rte_gre_hdr *gre = NULL;
2826 size_t temp_size = 0;
2829 return rte_flow_error_set(error, EINVAL,
2830 RTE_FLOW_ERROR_TYPE_ACTION,
2831 NULL, "invalid empty data");
2832 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2833 len = flow_dv_get_item_len(items->type);
2834 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
2835 return rte_flow_error_set(error, EINVAL,
2836 RTE_FLOW_ERROR_TYPE_ACTION,
2837 (void *)items->type,
2838 "items total size is too big"
2839 " for encap action");
2840 rte_memcpy((void *)&buf[temp_size], items->spec, len);
2841 switch (items->type) {
2842 case RTE_FLOW_ITEM_TYPE_ETH:
2843 eth = (struct rte_ether_hdr *)&buf[temp_size];
2845 case RTE_FLOW_ITEM_TYPE_VLAN:
2846 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
2848 return rte_flow_error_set(error, EINVAL,
2849 RTE_FLOW_ERROR_TYPE_ACTION,
2850 (void *)items->type,
2851 "eth header not found");
2852 if (!eth->ether_type)
2853 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
2855 case RTE_FLOW_ITEM_TYPE_IPV4:
2856 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
2858 return rte_flow_error_set(error, EINVAL,
2859 RTE_FLOW_ERROR_TYPE_ACTION,
2860 (void *)items->type,
2861 "neither eth nor vlan"
2863 if (vlan && !vlan->eth_proto)
2864 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2865 else if (eth && !eth->ether_type)
2866 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2867 if (!ipv4->version_ihl)
2868 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
2869 MLX5_ENCAP_IPV4_IHL_MIN;
2870 if (!ipv4->time_to_live)
2871 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
2873 case RTE_FLOW_ITEM_TYPE_IPV6:
2874 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
2876 return rte_flow_error_set(error, EINVAL,
2877 RTE_FLOW_ERROR_TYPE_ACTION,
2878 (void *)items->type,
2879 "neither eth nor vlan"
2881 if (vlan && !vlan->eth_proto)
2882 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
2883 else if (eth && !eth->ether_type)
2884 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
2885 if (!ipv6->vtc_flow)
2887 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
2888 if (!ipv6->hop_limits)
2889 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
2891 case RTE_FLOW_ITEM_TYPE_UDP:
2892 udp = (struct rte_udp_hdr *)&buf[temp_size];
2894 return rte_flow_error_set(error, EINVAL,
2895 RTE_FLOW_ERROR_TYPE_ACTION,
2896 (void *)items->type,
2897 "ip header not found");
2898 if (ipv4 && !ipv4->next_proto_id)
2899 ipv4->next_proto_id = IPPROTO_UDP;
2900 else if (ipv6 && !ipv6->proto)
2901 ipv6->proto = IPPROTO_UDP;
2903 case RTE_FLOW_ITEM_TYPE_VXLAN:
2904 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
2906 return rte_flow_error_set(error, EINVAL,
2907 RTE_FLOW_ERROR_TYPE_ACTION,
2908 (void *)items->type,
2909 "udp header not found");
2911 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
2912 if (!vxlan->vx_flags)
2914 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
2916 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2917 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
2919 return rte_flow_error_set(error, EINVAL,
2920 RTE_FLOW_ERROR_TYPE_ACTION,
2921 (void *)items->type,
2922 "udp header not found");
2923 if (!vxlan_gpe->proto)
2924 return rte_flow_error_set(error, EINVAL,
2925 RTE_FLOW_ERROR_TYPE_ACTION,
2926 (void *)items->type,
2927 "next protocol not found");
2930 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
2931 if (!vxlan_gpe->vx_flags)
2932 vxlan_gpe->vx_flags =
2933 MLX5_ENCAP_VXLAN_GPE_FLAGS;
2935 case RTE_FLOW_ITEM_TYPE_GRE:
2936 case RTE_FLOW_ITEM_TYPE_NVGRE:
2937 gre = (struct rte_gre_hdr *)&buf[temp_size];
2939 return rte_flow_error_set(error, EINVAL,
2940 RTE_FLOW_ERROR_TYPE_ACTION,
2941 (void *)items->type,
2942 "next protocol not found");
2944 return rte_flow_error_set(error, EINVAL,
2945 RTE_FLOW_ERROR_TYPE_ACTION,
2946 (void *)items->type,
2947 "ip header not found");
2948 if (ipv4 && !ipv4->next_proto_id)
2949 ipv4->next_proto_id = IPPROTO_GRE;
2950 else if (ipv6 && !ipv6->proto)
2951 ipv6->proto = IPPROTO_GRE;
2953 case RTE_FLOW_ITEM_TYPE_VOID:
2956 return rte_flow_error_set(error, EINVAL,
2957 RTE_FLOW_ERROR_TYPE_ACTION,
2958 (void *)items->type,
2959 "unsupported item type");
2969 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
2971 struct rte_ether_hdr *eth = NULL;
2972 struct rte_vlan_hdr *vlan = NULL;
2973 struct rte_ipv6_hdr *ipv6 = NULL;
2974 struct rte_udp_hdr *udp = NULL;
2978 eth = (struct rte_ether_hdr *)data;
2979 next_hdr = (char *)(eth + 1);
2980 proto = RTE_BE16(eth->ether_type);
2983 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
2984 vlan = (struct rte_vlan_hdr *)next_hdr;
2985 proto = RTE_BE16(vlan->eth_proto);
2986 next_hdr += sizeof(struct rte_vlan_hdr);
2989 /* HW calculates IPv4 csum. no need to proceed */
2990 if (proto == RTE_ETHER_TYPE_IPV4)
2993 /* non IPv4/IPv6 header. not supported */
2994 if (proto != RTE_ETHER_TYPE_IPV6) {
2995 return rte_flow_error_set(error, ENOTSUP,
2996 RTE_FLOW_ERROR_TYPE_ACTION,
2997 NULL, "Cannot offload non IPv4/IPv6");
3000 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3002 /* ignore non UDP */
3003 if (ipv6->proto != IPPROTO_UDP)
3006 udp = (struct rte_udp_hdr *)(ipv6 + 1);
3007 udp->dgram_cksum = 0;
3013 * Convert L2 encap action to DV specification.
3016 * Pointer to rte_eth_dev structure.
3018 * Pointer to action structure.
3019 * @param[in, out] dev_flow
3020 * Pointer to the mlx5_flow.
3021 * @param[in] transfer
3022 * Mark if the flow is E-Switch flow.
3024 * Pointer to the error structure.
3027 * 0 on success, a negative errno value otherwise and rte_errno is set.
3030 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3031 const struct rte_flow_action *action,
3032 struct mlx5_flow *dev_flow,
3034 struct rte_flow_error *error)
3036 const struct rte_flow_item *encap_data;
3037 const struct rte_flow_action_raw_encap *raw_encap_data;
3038 struct mlx5_flow_dv_encap_decap_resource res = {
3040 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3041 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3042 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3045 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3047 (const struct rte_flow_action_raw_encap *)action->conf;
3048 res.size = raw_encap_data->size;
3049 memcpy(res.buf, raw_encap_data->data, res.size);
3051 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3053 ((const struct rte_flow_action_vxlan_encap *)
3054 action->conf)->definition;
3057 ((const struct rte_flow_action_nvgre_encap *)
3058 action->conf)->definition;
3059 if (flow_dv_convert_encap_data(encap_data, res.buf,
3063 if (flow_dv_zero_encap_udp_csum(res.buf, error))
3065 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3066 return rte_flow_error_set(error, EINVAL,
3067 RTE_FLOW_ERROR_TYPE_ACTION,
3068 NULL, "can't create L2 encap action");
3073 * Convert L2 decap action to DV specification.
3076 * Pointer to rte_eth_dev structure.
3077 * @param[in, out] dev_flow
3078 * Pointer to the mlx5_flow.
3079 * @param[in] transfer
3080 * Mark if the flow is E-Switch flow.
3082 * Pointer to the error structure.
3085 * 0 on success, a negative errno value otherwise and rte_errno is set.
3088 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3089 struct mlx5_flow *dev_flow,
3091 struct rte_flow_error *error)
3093 struct mlx5_flow_dv_encap_decap_resource res = {
3096 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3097 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3098 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3101 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3102 return rte_flow_error_set(error, EINVAL,
3103 RTE_FLOW_ERROR_TYPE_ACTION,
3104 NULL, "can't create L2 decap action");
3109 * Convert raw decap/encap (L3 tunnel) action to DV specification.
3112 * Pointer to rte_eth_dev structure.
3114 * Pointer to action structure.
3115 * @param[in, out] dev_flow
3116 * Pointer to the mlx5_flow.
3118 * Pointer to the flow attributes.
3120 * Pointer to the error structure.
3123 * 0 on success, a negative errno value otherwise and rte_errno is set.
3126 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3127 const struct rte_flow_action *action,
3128 struct mlx5_flow *dev_flow,
3129 const struct rte_flow_attr *attr,
3130 struct rte_flow_error *error)
3132 const struct rte_flow_action_raw_encap *encap_data;
3133 struct mlx5_flow_dv_encap_decap_resource res;
3135 memset(&res, 0, sizeof(res));
3136 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3137 res.size = encap_data->size;
3138 memcpy(res.buf, encap_data->data, res.size);
3139 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3140 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3141 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3143 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3145 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3146 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3147 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3148 return rte_flow_error_set(error, EINVAL,
3149 RTE_FLOW_ERROR_TYPE_ACTION,
3150 NULL, "can't create encap action");
3155 * Create action push VLAN.
3158 * Pointer to rte_eth_dev structure.
3160 * Pointer to the flow attributes.
3162 * Pointer to the vlan to push to the Ethernet header.
3163 * @param[in, out] dev_flow
3164 * Pointer to the mlx5_flow.
3166 * Pointer to the error structure.
3169 * 0 on success, a negative errno value otherwise and rte_errno is set.
3172 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3173 const struct rte_flow_attr *attr,
3174 const struct rte_vlan_hdr *vlan,
3175 struct mlx5_flow *dev_flow,
3176 struct rte_flow_error *error)
3178 struct mlx5_flow_dv_push_vlan_action_resource res;
3180 memset(&res, 0, sizeof(res));
3182 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3185 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3187 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3188 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3189 return flow_dv_push_vlan_action_resource_register
3190 (dev, &res, dev_flow, error);
3194 * Validate the modify-header actions.
3196 * @param[in] action_flags
3197 * Holds the actions detected until now.
3199 * Pointer to the modify action.
3201 * Pointer to error structure.
3204 * 0 on success, a negative errno value otherwise and rte_errno is set.
3207 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3208 const struct rte_flow_action *action,
3209 struct rte_flow_error *error)
3211 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3212 return rte_flow_error_set(error, EINVAL,
3213 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3214 NULL, "action configuration not set");
3215 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3216 return rte_flow_error_set(error, EINVAL,
3217 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3218 "can't have encap action before"
3224 * Validate the modify-header MAC address actions.
3226 * @param[in] action_flags
3227 * Holds the actions detected until now.
3229 * Pointer to the modify action.
3230 * @param[in] item_flags
3231 * Holds the items detected.
3233 * Pointer to error structure.
3236 * 0 on success, a negative errno value otherwise and rte_errno is set.
3239 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3240 const struct rte_flow_action *action,
3241 const uint64_t item_flags,
3242 struct rte_flow_error *error)
3246 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3248 if (!(item_flags & MLX5_FLOW_LAYER_L2))
3249 return rte_flow_error_set(error, EINVAL,
3250 RTE_FLOW_ERROR_TYPE_ACTION,
3252 "no L2 item in pattern");
3258 * Validate the modify-header IPv4 address actions.
3260 * @param[in] action_flags
3261 * Holds the actions detected until now.
3263 * Pointer to the modify action.
3264 * @param[in] item_flags
3265 * Holds the items detected.
3267 * Pointer to error structure.
3270 * 0 on success, a negative errno value otherwise and rte_errno is set.
3273 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3274 const struct rte_flow_action *action,
3275 const uint64_t item_flags,
3276 struct rte_flow_error *error)
3281 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3283 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3284 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3285 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3286 if (!(item_flags & layer))
3287 return rte_flow_error_set(error, EINVAL,
3288 RTE_FLOW_ERROR_TYPE_ACTION,
3290 "no ipv4 item in pattern");
3296 * Validate the modify-header IPv6 address actions.
3298 * @param[in] action_flags
3299 * Holds the actions detected until now.
3301 * Pointer to the modify action.
3302 * @param[in] item_flags
3303 * Holds the items detected.
3305 * Pointer to error structure.
3308 * 0 on success, a negative errno value otherwise and rte_errno is set.
3311 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3312 const struct rte_flow_action *action,
3313 const uint64_t item_flags,
3314 struct rte_flow_error *error)
3319 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3321 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3322 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3323 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3324 if (!(item_flags & layer))
3325 return rte_flow_error_set(error, EINVAL,
3326 RTE_FLOW_ERROR_TYPE_ACTION,
3328 "no ipv6 item in pattern");
3334 * Validate the modify-header TP actions.
3336 * @param[in] action_flags
3337 * Holds the actions detected until now.
3339 * Pointer to the modify action.
3340 * @param[in] item_flags
3341 * Holds the items detected.
3343 * Pointer to error structure.
3346 * 0 on success, a negative errno value otherwise and rte_errno is set.
3349 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3350 const struct rte_flow_action *action,
3351 const uint64_t item_flags,
3352 struct rte_flow_error *error)
3357 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3359 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3360 MLX5_FLOW_LAYER_INNER_L4 :
3361 MLX5_FLOW_LAYER_OUTER_L4;
3362 if (!(item_flags & layer))
3363 return rte_flow_error_set(error, EINVAL,
3364 RTE_FLOW_ERROR_TYPE_ACTION,
3365 NULL, "no transport layer "
3372 * Validate the modify-header actions of increment/decrement
3373 * TCP Sequence-number.
3375 * @param[in] action_flags
3376 * Holds the actions detected until now.
3378 * Pointer to the modify action.
3379 * @param[in] item_flags
3380 * Holds the items detected.
3382 * Pointer to error structure.
3385 * 0 on success, a negative errno value otherwise and rte_errno is set.
3388 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3389 const struct rte_flow_action *action,
3390 const uint64_t item_flags,
3391 struct rte_flow_error *error)
3396 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3398 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3399 MLX5_FLOW_LAYER_INNER_L4_TCP :
3400 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3401 if (!(item_flags & layer))
3402 return rte_flow_error_set(error, EINVAL,
3403 RTE_FLOW_ERROR_TYPE_ACTION,
3404 NULL, "no TCP item in"
3406 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3407 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3408 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3409 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3410 return rte_flow_error_set(error, EINVAL,
3411 RTE_FLOW_ERROR_TYPE_ACTION,
3413 "cannot decrease and increase"
3414 " TCP sequence number"
3415 " at the same time");
3421 * Validate the modify-header actions of increment/decrement
3422 * TCP Acknowledgment number.
3424 * @param[in] action_flags
3425 * Holds the actions detected until now.
3427 * Pointer to the modify action.
3428 * @param[in] item_flags
3429 * Holds the items detected.
3431 * Pointer to error structure.
3434 * 0 on success, a negative errno value otherwise and rte_errno is set.
3437 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3438 const struct rte_flow_action *action,
3439 const uint64_t item_flags,
3440 struct rte_flow_error *error)
3445 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3447 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3448 MLX5_FLOW_LAYER_INNER_L4_TCP :
3449 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3450 if (!(item_flags & layer))
3451 return rte_flow_error_set(error, EINVAL,
3452 RTE_FLOW_ERROR_TYPE_ACTION,
3453 NULL, "no TCP item in"
3455 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3456 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3457 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3458 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3459 return rte_flow_error_set(error, EINVAL,
3460 RTE_FLOW_ERROR_TYPE_ACTION,
3462 "cannot decrease and increase"
3463 " TCP acknowledgment number"
3464 " at the same time");
3470 * Validate the modify-header TTL actions.
3472 * @param[in] action_flags
3473 * Holds the actions detected until now.
3475 * Pointer to the modify action.
3476 * @param[in] item_flags
3477 * Holds the items detected.
3479 * Pointer to error structure.
3482 * 0 on success, a negative errno value otherwise and rte_errno is set.
3485 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3486 const struct rte_flow_action *action,
3487 const uint64_t item_flags,
3488 struct rte_flow_error *error)
3493 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3495 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3496 MLX5_FLOW_LAYER_INNER_L3 :
3497 MLX5_FLOW_LAYER_OUTER_L3;
3498 if (!(item_flags & layer))
3499 return rte_flow_error_set(error, EINVAL,
3500 RTE_FLOW_ERROR_TYPE_ACTION,
3502 "no IP protocol in pattern");
3508 * Validate jump action.
3511 * Pointer to the jump action.
3512 * @param[in] action_flags
3513 * Holds the actions detected until now.
3514 * @param[in] attributes
3515 * Pointer to flow attributes
3516 * @param[in] external
3517 * Action belongs to flow rule created by request external to PMD.
3519 * Pointer to error structure.
3522 * 0 on success, a negative errno value otherwise and rte_errno is set.
3525 flow_dv_validate_action_jump(const struct rte_flow_action *action,
3526 uint64_t action_flags,
3527 const struct rte_flow_attr *attributes,
3528 bool external, struct rte_flow_error *error)
3530 uint32_t target_group, table;
3533 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3534 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3535 return rte_flow_error_set(error, EINVAL,
3536 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3537 "can't have 2 fate actions in"
3539 if (action_flags & MLX5_FLOW_ACTION_METER)
3540 return rte_flow_error_set(error, ENOTSUP,
3541 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3542 "jump with meter not support");
3544 return rte_flow_error_set(error, EINVAL,
3545 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3546 NULL, "action configuration not set");
3548 ((const struct rte_flow_action_jump *)action->conf)->group;
3549 ret = mlx5_flow_group_to_table(attributes, external, target_group,
3550 true, &table, error);
3553 if (attributes->group == target_group)
3554 return rte_flow_error_set(error, EINVAL,
3555 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3556 "target group must be other than"
3557 " the current flow group");
3562 * Validate the port_id action.
3565 * Pointer to rte_eth_dev structure.
3566 * @param[in] action_flags
3567 * Bit-fields that holds the actions detected until now.
3569 * Port_id RTE action structure.
3571 * Attributes of flow that includes this action.
3573 * Pointer to error structure.
3576 * 0 on success, a negative errno value otherwise and rte_errno is set.
3579 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
3580 uint64_t action_flags,
3581 const struct rte_flow_action *action,
3582 const struct rte_flow_attr *attr,
3583 struct rte_flow_error *error)
3585 const struct rte_flow_action_port_id *port_id;
3586 struct mlx5_priv *act_priv;
3587 struct mlx5_priv *dev_priv;
3590 if (!attr->transfer)
3591 return rte_flow_error_set(error, ENOTSUP,
3592 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3594 "port id action is valid in transfer"
3596 if (!action || !action->conf)
3597 return rte_flow_error_set(error, ENOTSUP,
3598 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3600 "port id action parameters must be"
3602 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3603 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3604 return rte_flow_error_set(error, EINVAL,
3605 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3606 "can have only one fate actions in"
3608 dev_priv = mlx5_dev_to_eswitch_info(dev);
3610 return rte_flow_error_set(error, rte_errno,
3611 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3613 "failed to obtain E-Switch info");
3614 port_id = action->conf;
3615 port = port_id->original ? dev->data->port_id : port_id->id;
3616 act_priv = mlx5_port_to_eswitch_info(port, false);
3618 return rte_flow_error_set
3620 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
3621 "failed to obtain E-Switch port id for port");
3622 if (act_priv->domain_id != dev_priv->domain_id)
3623 return rte_flow_error_set
3625 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3626 "port does not belong to"
3627 " E-Switch being configured");
3632 * Get the maximum number of modify header actions.
3635 * Pointer to rte_eth_dev structure.
3637 * Flags bits to check if root level.
3640 * Max number of modify header actions device can support.
3643 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev, uint64_t flags)
3646 * There's no way to directly query the max cap. Although it has to be
3647 * acquried by iterative trial, it is a safe assumption that more
3648 * actions are supported by FW if extensive metadata register is
3649 * supported. (Only in the root table)
3651 if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
3652 return MLX5_MAX_MODIFY_NUM;
3654 return mlx5_flow_ext_mreg_supported(dev) ?
3655 MLX5_ROOT_TBL_MODIFY_NUM :
3656 MLX5_ROOT_TBL_MODIFY_NUM_NO_MREG;
3660 * Validate the meter action.
3663 * Pointer to rte_eth_dev structure.
3664 * @param[in] action_flags
3665 * Bit-fields that holds the actions detected until now.
3667 * Pointer to the meter action.
3669 * Attributes of flow that includes this action.
3671 * Pointer to error structure.
3674 * 0 on success, a negative errno value otherwise and rte_ernno is set.
3677 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
3678 uint64_t action_flags,
3679 const struct rte_flow_action *action,
3680 const struct rte_flow_attr *attr,
3681 struct rte_flow_error *error)
3683 struct mlx5_priv *priv = dev->data->dev_private;
3684 const struct rte_flow_action_meter *am = action->conf;
3685 struct mlx5_flow_meter *fm;
3688 return rte_flow_error_set(error, EINVAL,
3689 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3690 "meter action conf is NULL");
3692 if (action_flags & MLX5_FLOW_ACTION_METER)
3693 return rte_flow_error_set(error, ENOTSUP,
3694 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3695 "meter chaining not support");
3696 if (action_flags & MLX5_FLOW_ACTION_JUMP)
3697 return rte_flow_error_set(error, ENOTSUP,
3698 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3699 "meter with jump not support");
3701 return rte_flow_error_set(error, ENOTSUP,
3702 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3704 "meter action not supported");
3705 fm = mlx5_flow_meter_find(priv, am->mtr_id);
3707 return rte_flow_error_set(error, EINVAL,
3708 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3710 if (fm->ref_cnt && (!(fm->attr.transfer == attr->transfer ||
3711 (!fm->attr.ingress && !attr->ingress && attr->egress) ||
3712 (!fm->attr.egress && !attr->egress && attr->ingress))))
3713 return rte_flow_error_set(error, EINVAL,
3714 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3715 "Flow attributes are either invalid "
3716 "or have a conflict with current "
3717 "meter attributes");
3722 * Validate the modify-header IPv4 DSCP actions.
3724 * @param[in] action_flags
3725 * Holds the actions detected until now.
3727 * Pointer to the modify action.
3728 * @param[in] item_flags
3729 * Holds the items detected.
3731 * Pointer to error structure.
3734 * 0 on success, a negative errno value otherwise and rte_errno is set.
3737 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
3738 const struct rte_flow_action *action,
3739 const uint64_t item_flags,
3740 struct rte_flow_error *error)
3744 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3746 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
3747 return rte_flow_error_set(error, EINVAL,
3748 RTE_FLOW_ERROR_TYPE_ACTION,
3750 "no ipv4 item in pattern");
3756 * Validate the modify-header IPv6 DSCP actions.
3758 * @param[in] action_flags
3759 * Holds the actions detected until now.
3761 * Pointer to the modify action.
3762 * @param[in] item_flags
3763 * Holds the items detected.
3765 * Pointer to error structure.
3768 * 0 on success, a negative errno value otherwise and rte_errno is set.
3771 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
3772 const struct rte_flow_action *action,
3773 const uint64_t item_flags,
3774 struct rte_flow_error *error)
3778 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3780 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
3781 return rte_flow_error_set(error, EINVAL,
3782 RTE_FLOW_ERROR_TYPE_ACTION,
3784 "no ipv6 item in pattern");
3790 * Find existing modify-header resource or create and register a new one.
3792 * @param dev[in, out]
3793 * Pointer to rte_eth_dev structure.
3794 * @param[in, out] resource
3795 * Pointer to modify-header resource.
3796 * @parm[in, out] dev_flow
3797 * Pointer to the dev_flow.
3799 * pointer to error structure.
3802 * 0 on success otherwise -errno and errno is set.
3805 flow_dv_modify_hdr_resource_register
3806 (struct rte_eth_dev *dev,
3807 struct mlx5_flow_dv_modify_hdr_resource *resource,
3808 struct mlx5_flow *dev_flow,
3809 struct rte_flow_error *error)
3811 struct mlx5_priv *priv = dev->data->dev_private;
3812 struct mlx5_ibv_shared *sh = priv->sh;
3813 struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
3814 struct mlx5dv_dr_domain *ns;
3815 uint32_t actions_len;
3817 resource->flags = dev_flow->dv.group ? 0 :
3818 MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
3819 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
3821 return rte_flow_error_set(error, EOVERFLOW,
3822 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3823 "too many modify header items");
3824 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3825 ns = sh->fdb_domain;
3826 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
3830 /* Lookup a matching resource from cache. */
3831 actions_len = resource->actions_num * sizeof(resource->actions[0]);
3832 LIST_FOREACH(cache_resource, &sh->modify_cmds, next) {
3833 if (resource->ft_type == cache_resource->ft_type &&
3834 resource->actions_num == cache_resource->actions_num &&
3835 resource->flags == cache_resource->flags &&
3836 !memcmp((const void *)resource->actions,
3837 (const void *)cache_resource->actions,
3839 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
3840 (void *)cache_resource,
3841 rte_atomic32_read(&cache_resource->refcnt));
3842 rte_atomic32_inc(&cache_resource->refcnt);
3843 dev_flow->handle->dvh.modify_hdr = cache_resource;
3847 /* Register new modify-header resource. */
3848 cache_resource = rte_calloc(__func__, 1,
3849 sizeof(*cache_resource) + actions_len, 0);
3850 if (!cache_resource)
3851 return rte_flow_error_set(error, ENOMEM,
3852 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3853 "cannot allocate resource memory");
3854 *cache_resource = *resource;
3855 rte_memcpy(cache_resource->actions, resource->actions, actions_len);
3856 cache_resource->verbs_action =
3857 mlx5_glue->dv_create_flow_action_modify_header
3858 (sh->ctx, cache_resource->ft_type, ns,
3859 cache_resource->flags, actions_len,
3860 (uint64_t *)cache_resource->actions);
3861 if (!cache_resource->verbs_action) {
3862 rte_free(cache_resource);
3863 return rte_flow_error_set(error, ENOMEM,
3864 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3865 NULL, "cannot create action");
3867 rte_atomic32_init(&cache_resource->refcnt);
3868 rte_atomic32_inc(&cache_resource->refcnt);
3869 LIST_INSERT_HEAD(&sh->modify_cmds, cache_resource, next);
3870 dev_flow->handle->dvh.modify_hdr = cache_resource;
3871 DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
3872 (void *)cache_resource,
3873 rte_atomic32_read(&cache_resource->refcnt));
3878 * Get DV flow counter by index.
3881 * Pointer to the Ethernet device structure.
3883 * mlx5 flow counter index in the container.
3885 * mlx5 flow counter pool in the container,
3888 * Pointer to the counter, NULL otherwise.
3890 static struct mlx5_flow_counter *
3891 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
3893 struct mlx5_flow_counter_pool **ppool)
3895 struct mlx5_priv *priv = dev->data->dev_private;
3896 struct mlx5_pools_container *cont;
3897 struct mlx5_flow_counter_pool *pool;
3901 if (idx >= MLX5_CNT_BATCH_OFFSET) {
3902 idx -= MLX5_CNT_BATCH_OFFSET;
3905 cont = MLX5_CNT_CONTAINER(priv->sh, batch, 0);
3906 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cont->n);
3907 pool = cont->pools[idx / MLX5_COUNTERS_PER_POOL];
3911 return &pool->counters_raw[idx % MLX5_COUNTERS_PER_POOL];
3915 * Get a pool by devx counter ID.
3918 * Pointer to the counter container.
3920 * The counter devx ID.
3923 * The counter pool pointer if exists, NULL otherwise,
3925 static struct mlx5_flow_counter_pool *
3926 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
3929 uint32_t n_valid = rte_atomic16_read(&cont->n_valid);
3931 for (i = 0; i < n_valid; i++) {
3932 struct mlx5_flow_counter_pool *pool = cont->pools[i];
3933 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
3934 MLX5_COUNTERS_PER_POOL;
3936 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL) {
3938 * Move the pool to the head, as counter allocate
3939 * always gets the first pool in the container.
3941 if (pool != TAILQ_FIRST(&cont->pool_list)) {
3942 TAILQ_REMOVE(&cont->pool_list, pool, next);
3943 TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
3952 * Allocate a new memory for the counter values wrapped by all the needed
3956 * Pointer to the Ethernet device structure.
3958 * The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
3961 * The new memory management pointer on success, otherwise NULL and rte_errno
3964 static struct mlx5_counter_stats_mem_mng *
3965 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
3967 struct mlx5_ibv_shared *sh = ((struct mlx5_priv *)
3968 (dev->data->dev_private))->sh;
3969 struct mlx5_devx_mkey_attr mkey_attr;
3970 struct mlx5_counter_stats_mem_mng *mem_mng;
3971 volatile struct flow_counter_stats *raw_data;
3972 int size = (sizeof(struct flow_counter_stats) *
3973 MLX5_COUNTERS_PER_POOL +
3974 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
3975 sizeof(struct mlx5_counter_stats_mem_mng);
3976 uint8_t *mem = rte_calloc(__func__, 1, size, sysconf(_SC_PAGESIZE));
3983 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
3984 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
3985 mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
3986 IBV_ACCESS_LOCAL_WRITE);
3987 if (!mem_mng->umem) {
3992 mkey_attr.addr = (uintptr_t)mem;
3993 mkey_attr.size = size;
3994 mkey_attr.umem_id = mem_mng->umem->umem_id;
3995 mkey_attr.pd = sh->pdn;
3996 mkey_attr.log_entity_size = 0;
3997 mkey_attr.pg_access = 0;
3998 mkey_attr.klm_array = NULL;
3999 mkey_attr.klm_num = 0;
4000 mkey_attr.relaxed_ordering = 1;
4001 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
4003 mlx5_glue->devx_umem_dereg(mem_mng->umem);
4008 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
4009 raw_data = (volatile struct flow_counter_stats *)mem;
4010 for (i = 0; i < raws_n; ++i) {
4011 mem_mng->raws[i].mem_mng = mem_mng;
4012 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
4014 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
4019 * Resize a counter container.
4022 * Pointer to the Ethernet device structure.
4024 * Whether the pool is for counter that was allocated by batch command.
4027 * The new container pointer on success, otherwise NULL and rte_errno is set.
4029 static struct mlx5_pools_container *
4030 flow_dv_container_resize(struct rte_eth_dev *dev, uint32_t batch)
4032 struct mlx5_priv *priv = dev->data->dev_private;
4033 struct mlx5_pools_container *cont =
4034 MLX5_CNT_CONTAINER(priv->sh, batch, 0);
4035 struct mlx5_pools_container *new_cont =
4036 MLX5_CNT_CONTAINER_UNUSED(priv->sh, batch, 0);
4037 struct mlx5_counter_stats_mem_mng *mem_mng = NULL;
4038 uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
4039 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4042 /* Fallback mode has no background thread. Skip the check. */
4043 if (!priv->counter_fallback &&
4044 cont != MLX5_CNT_CONTAINER(priv->sh, batch, 1)) {
4045 /* The last resize still hasn't detected by the host thread. */
4049 new_cont->pools = rte_calloc(__func__, 1, mem_size, 0);
4050 if (!new_cont->pools) {
4055 memcpy(new_cont->pools, cont->pools, cont->n *
4056 sizeof(struct mlx5_flow_counter_pool *));
4058 * Fallback mode query the counter directly, no background query
4059 * resources are needed.
4061 if (!priv->counter_fallback) {
4062 mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
4063 MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES);
4065 rte_free(new_cont->pools);
4068 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
4069 LIST_INSERT_HEAD(&priv->sh->cmng.free_stat_raws,
4071 MLX5_CNT_CONTAINER_RESIZE +
4075 * Release the old container pools directly as no background
4076 * thread helps that.
4078 rte_free(cont->pools);
4080 new_cont->n = resize;
4081 rte_atomic16_set(&new_cont->n_valid, rte_atomic16_read(&cont->n_valid));
4082 TAILQ_INIT(&new_cont->pool_list);
4083 TAILQ_CONCAT(&new_cont->pool_list, &cont->pool_list, next);
4084 new_cont->init_mem_mng = mem_mng;
4086 /* Flip the master container. */
4087 priv->sh->cmng.mhi[batch] ^= (uint8_t)1;
4092 * Query a devx flow counter.
4095 * Pointer to the Ethernet device structure.
4097 * Index to the flow counter.
4099 * The statistics value of packets.
4101 * The statistics value of bytes.
4104 * 0 on success, otherwise a negative errno value and rte_errno is set.
4107 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4110 struct mlx5_priv *priv = dev->data->dev_private;
4111 struct mlx5_flow_counter_pool *pool = NULL;
4112 struct mlx5_flow_counter *cnt;
4113 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4116 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4118 if (counter < MLX5_CNT_BATCH_OFFSET) {
4119 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4120 if (priv->counter_fallback)
4121 return mlx5_devx_cmd_flow_counter_query(cnt_ext->dcs, 0,
4122 0, pkts, bytes, 0, NULL, NULL, 0);
4125 rte_spinlock_lock(&pool->sl);
4127 * The single counters allocation may allocate smaller ID than the
4128 * current allocated in parallel to the host reading.
4129 * In this case the new counter values must be reported as 0.
4131 if (unlikely(cnt_ext && cnt_ext->dcs->id < pool->raw->min_dcs_id)) {
4135 offset = cnt - &pool->counters_raw[0];
4136 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4137 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4139 rte_spinlock_unlock(&pool->sl);
4144 * Create and initialize a new counter pool.
4147 * Pointer to the Ethernet device structure.
4149 * The devX counter handle.
4151 * Whether the pool is for counter that was allocated by batch command.
4152 * @param[in/out] cont_cur
4153 * Pointer to the container pointer, it will be update in pool resize.
4156 * The pool container pointer on success, NULL otherwise and rte_errno is set.
4158 static struct mlx5_pools_container *
4159 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4162 struct mlx5_priv *priv = dev->data->dev_private;
4163 struct mlx5_flow_counter_pool *pool;
4164 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4166 int16_t n_valid = rte_atomic16_read(&cont->n_valid);
4169 if (cont->n == n_valid) {
4170 cont = flow_dv_container_resize(dev, batch);
4174 size = sizeof(*pool);
4176 size += MLX5_COUNTERS_PER_POOL *
4177 sizeof(struct mlx5_flow_counter_ext);
4178 pool = rte_calloc(__func__, 1, size, 0);
4183 pool->min_dcs = dcs;
4184 if (!priv->counter_fallback)
4185 pool->raw = cont->init_mem_mng->raws + n_valid %
4186 MLX5_CNT_CONTAINER_RESIZE;
4187 pool->raw_hw = NULL;
4188 rte_spinlock_init(&pool->sl);
4190 * The generation of the new allocated counters in this pool is 0, 2 in
4191 * the pool generation makes all the counters valid for allocation.
4192 * The start and end query generation protect the counters be released
4193 * between the query and update gap period will not be reallocated
4194 * without the last query finished and stats updated to the memory.
4196 rte_atomic64_set(&pool->start_query_gen, 0x2);
4198 * There's no background query thread for fallback mode, set the
4199 * end_query_gen to the maximum value since no need to wait for
4200 * statistics update.
4202 rte_atomic64_set(&pool->end_query_gen, priv->counter_fallback ?
4204 TAILQ_INIT(&pool->counters);
4205 TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
4206 pool->index = n_valid;
4207 cont->pools[n_valid] = pool;
4208 /* Pool initialization must be updated before host thread access. */
4210 rte_atomic16_add(&cont->n_valid, 1);
4215 * Prepare a new counter and/or a new counter pool.
4218 * Pointer to the Ethernet device structure.
4219 * @param[out] cnt_free
4220 * Where to put the pointer of a new counter.
4222 * Whether the pool is for counter that was allocated by batch command.
4225 * The counter container pointer and @p cnt_free is set on success,
4226 * NULL otherwise and rte_errno is set.
4228 static struct mlx5_pools_container *
4229 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4230 struct mlx5_flow_counter **cnt_free,
4233 struct mlx5_priv *priv = dev->data->dev_private;
4234 struct mlx5_pools_container *cont;
4235 struct mlx5_flow_counter_pool *pool;
4236 struct mlx5_devx_obj *dcs = NULL;
4237 struct mlx5_flow_counter *cnt;
4240 cont = MLX5_CNT_CONTAINER(priv->sh, batch, 0);
4242 /* bulk_bitmap must be 0 for single counter allocation. */
4243 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4246 pool = flow_dv_find_pool_by_id(cont, dcs->id);
4248 cont = flow_dv_pool_create(dev, dcs, batch);
4250 mlx5_devx_cmd_destroy(dcs);
4253 pool = TAILQ_FIRST(&cont->pool_list);
4254 } else if (dcs->id < pool->min_dcs->id) {
4255 rte_atomic64_set(&pool->a64_dcs,
4256 (int64_t)(uintptr_t)dcs);
4258 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4259 cnt = &pool->counters_raw[i];
4260 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
4261 MLX5_GET_POOL_CNT_EXT(pool, i)->dcs = dcs;
4265 /* bulk_bitmap is in 128 counters units. */
4266 if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
4267 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4269 rte_errno = ENODATA;
4272 cont = flow_dv_pool_create(dev, dcs, batch);
4274 mlx5_devx_cmd_destroy(dcs);
4277 pool = TAILQ_FIRST(&cont->pool_list);
4278 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
4279 cnt = &pool->counters_raw[i];
4280 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
4282 *cnt_free = &pool->counters_raw[0];
4287 * Search for existed shared counter.
4290 * Pointer to the relevant counter pool container.
4292 * The shared counter ID to search.
4294 * mlx5 flow counter pool in the container,
4297 * NULL if not existed, otherwise pointer to the shared extend counter.
4299 static struct mlx5_flow_counter_ext *
4300 flow_dv_counter_shared_search(struct mlx5_pools_container *cont, uint32_t id,
4301 struct mlx5_flow_counter_pool **ppool)
4303 static struct mlx5_flow_counter_ext *cnt;
4304 struct mlx5_flow_counter_pool *pool;
4306 uint32_t n_valid = rte_atomic16_read(&cont->n_valid);
4308 for (i = 0; i < n_valid; i++) {
4309 pool = cont->pools[i];
4310 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
4311 cnt = MLX5_GET_POOL_CNT_EXT(pool, i);
4312 if (cnt->ref_cnt && cnt->shared && cnt->id == id) {
4314 *ppool = cont->pools[i];
4323 * Allocate a flow counter.
4326 * Pointer to the Ethernet device structure.
4328 * Indicate if this counter is shared with other flows.
4330 * Counter identifier.
4332 * Counter flow group.
4335 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4338 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
4341 struct mlx5_priv *priv = dev->data->dev_private;
4342 struct mlx5_flow_counter_pool *pool = NULL;
4343 struct mlx5_flow_counter *cnt_free = NULL;
4344 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4346 * Currently group 0 flow counter cannot be assigned to a flow if it is
4347 * not the first one in the batch counter allocation, so it is better
4348 * to allocate counters one by one for these flows in a separate
4350 * A counter can be shared between different groups so need to take
4351 * shared counters from the single container.
4353 uint32_t batch = (group && !shared && !priv->counter_fallback) ? 1 : 0;
4354 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4358 if (!priv->config.devx) {
4359 rte_errno = ENOTSUP;
4363 cnt_ext = flow_dv_counter_shared_search(cont, id, &pool);
4365 if (cnt_ext->ref_cnt + 1 == 0) {
4370 cnt_idx = pool->index * MLX5_COUNTERS_PER_POOL +
4371 (cnt_ext->dcs->id % MLX5_COUNTERS_PER_POOL)
4376 /* Pools which has a free counters are in the start. */
4377 TAILQ_FOREACH(pool, &cont->pool_list, next) {
4379 * The free counter reset values must be updated between the
4380 * counter release to the counter allocation, so, at least one
4381 * query must be done in this time. ensure it by saving the
4382 * query generation in the release time.
4383 * The free list is sorted according to the generation - so if
4384 * the first one is not updated, all the others are not
4387 cnt_free = TAILQ_FIRST(&pool->counters);
4388 if (cnt_free && cnt_free->query_gen <
4389 rte_atomic64_read(&pool->end_query_gen))
4394 cont = flow_dv_counter_pool_prepare(dev, &cnt_free, batch);
4397 pool = TAILQ_FIRST(&cont->pool_list);
4400 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt_free);
4401 /* Create a DV counter action only in the first time usage. */
4402 if (!cnt_free->action) {
4404 struct mlx5_devx_obj *dcs;
4407 offset = cnt_free - &pool->counters_raw[0];
4408 dcs = pool->min_dcs;
4413 cnt_free->action = mlx5_glue->dv_create_flow_action_counter
4415 if (!cnt_free->action) {
4420 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4421 (cnt_free - pool->counters_raw));
4422 cnt_idx += batch * MLX5_CNT_BATCH_OFFSET;
4423 /* Update the counter reset values. */
4424 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4428 cnt_ext->shared = shared;
4429 cnt_ext->ref_cnt = 1;
4432 if (!priv->counter_fallback && !priv->sh->cmng.query_thread_on)
4433 /* Start the asynchronous batch query by the host thread. */
4434 mlx5_set_query_alarm(priv->sh);
4435 TAILQ_REMOVE(&pool->counters, cnt_free, next);
4436 if (TAILQ_EMPTY(&pool->counters)) {
4437 /* Move the pool to the end of the container pool list. */
4438 TAILQ_REMOVE(&cont->pool_list, pool, next);
4439 TAILQ_INSERT_TAIL(&cont->pool_list, pool, next);
4445 * Release a flow counter.
4448 * Pointer to the Ethernet device structure.
4449 * @param[in] counter
4450 * Index to the counter handler.
4453 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
4455 struct mlx5_flow_counter_pool *pool = NULL;
4456 struct mlx5_flow_counter *cnt;
4457 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4461 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4463 if (counter < MLX5_CNT_BATCH_OFFSET) {
4464 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4465 if (cnt_ext && --cnt_ext->ref_cnt)
4468 /* Put the counter in the end - the last updated one. */
4469 TAILQ_INSERT_TAIL(&pool->counters, cnt, next);
4471 * Counters released between query trigger and handler need
4472 * to wait the next round of query. Since the packets arrive
4473 * in the gap period will not be taken into account to the
4476 cnt->query_gen = rte_atomic64_read(&pool->start_query_gen);
4480 * Verify the @p attributes will be correctly understood by the NIC and store
4481 * them in the @p flow if everything is correct.
4484 * Pointer to dev struct.
4485 * @param[in] attributes
4486 * Pointer to flow attributes
4487 * @param[in] external
4488 * This flow rule is created by request external to PMD.
4490 * Pointer to error structure.
4493 * 0 on success, a negative errno value otherwise and rte_errno is set.
4496 flow_dv_validate_attributes(struct rte_eth_dev *dev,
4497 const struct rte_flow_attr *attributes,
4498 bool external __rte_unused,
4499 struct rte_flow_error *error)
4501 struct mlx5_priv *priv = dev->data->dev_private;
4502 uint32_t priority_max = priv->config.flow_prio - 1;
4504 #ifndef HAVE_MLX5DV_DR
4505 if (attributes->group)
4506 return rte_flow_error_set(error, ENOTSUP,
4507 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
4509 "groups are not supported");
4514 ret = mlx5_flow_group_to_table(attributes, external,
4515 attributes->group, !!priv->fdb_def_rule,
4520 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
4521 attributes->priority >= priority_max)
4522 return rte_flow_error_set(error, ENOTSUP,
4523 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
4525 "priority out of range");
4526 if (attributes->transfer) {
4527 if (!priv->config.dv_esw_en)
4528 return rte_flow_error_set
4530 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4531 "E-Switch dr is not supported");
4532 if (!(priv->representor || priv->master))
4533 return rte_flow_error_set
4534 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4535 NULL, "E-Switch configuration can only be"
4536 " done by a master or a representor device");
4537 if (attributes->egress)
4538 return rte_flow_error_set
4540 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
4541 "egress is not supported");
4543 if (!(attributes->egress ^ attributes->ingress))
4544 return rte_flow_error_set(error, ENOTSUP,
4545 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
4546 "must specify exactly one of "
4547 "ingress or egress");
4552 * Internal validation function. For validating both actions and items.
4555 * Pointer to the rte_eth_dev structure.
4557 * Pointer to the flow attributes.
4559 * Pointer to the list of items.
4560 * @param[in] actions
4561 * Pointer to the list of actions.
4562 * @param[in] external
4563 * This flow rule is created by request external to PMD.
4565 * Pointer to the error structure.
4568 * 0 on success, a negative errno value otherwise and rte_errno is set.
4571 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
4572 const struct rte_flow_item items[],
4573 const struct rte_flow_action actions[],
4574 bool external, struct rte_flow_error *error)
4577 uint64_t action_flags = 0;
4578 uint64_t item_flags = 0;
4579 uint64_t last_item = 0;
4580 uint8_t next_protocol = 0xff;
4581 uint16_t ether_type = 0;
4583 uint8_t item_ipv6_proto = 0;
4584 const struct rte_flow_item *gre_item = NULL;
4585 const struct rte_flow_action_raw_decap *decap;
4586 const struct rte_flow_action_raw_encap *encap;
4587 const struct rte_flow_action_rss *rss;
4588 const struct rte_flow_item_tcp nic_tcp_mask = {
4591 .src_port = RTE_BE16(UINT16_MAX),
4592 .dst_port = RTE_BE16(UINT16_MAX),
4595 const struct rte_flow_item_ipv4 nic_ipv4_mask = {
4597 .src_addr = RTE_BE32(0xffffffff),
4598 .dst_addr = RTE_BE32(0xffffffff),
4599 .type_of_service = 0xff,
4600 .next_proto_id = 0xff,
4601 .time_to_live = 0xff,
4604 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
4607 "\xff\xff\xff\xff\xff\xff\xff\xff"
4608 "\xff\xff\xff\xff\xff\xff\xff\xff",
4610 "\xff\xff\xff\xff\xff\xff\xff\xff"
4611 "\xff\xff\xff\xff\xff\xff\xff\xff",
4612 .vtc_flow = RTE_BE32(0xffffffff),
4617 struct mlx5_priv *priv = dev->data->dev_private;
4618 struct mlx5_dev_config *dev_conf = &priv->config;
4619 uint16_t queue_index = 0xFFFF;
4620 const struct rte_flow_item_vlan *vlan_m = NULL;
4624 ret = flow_dv_validate_attributes(dev, attr, external, error);
4627 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4628 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
4629 int type = items->type;
4632 case RTE_FLOW_ITEM_TYPE_VOID:
4634 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4635 ret = flow_dv_validate_item_port_id
4636 (dev, items, attr, item_flags, error);
4639 last_item = MLX5_FLOW_ITEM_PORT_ID;
4641 case RTE_FLOW_ITEM_TYPE_ETH:
4642 ret = mlx5_flow_validate_item_eth(items, item_flags,
4646 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
4647 MLX5_FLOW_LAYER_OUTER_L2;
4648 if (items->mask != NULL && items->spec != NULL) {
4650 ((const struct rte_flow_item_eth *)
4653 ((const struct rte_flow_item_eth *)
4655 ether_type = rte_be_to_cpu_16(ether_type);
4660 case RTE_FLOW_ITEM_TYPE_VLAN:
4661 ret = mlx5_flow_validate_item_vlan(items, item_flags,
4665 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
4666 MLX5_FLOW_LAYER_OUTER_VLAN;
4667 if (items->mask != NULL && items->spec != NULL) {
4669 ((const struct rte_flow_item_vlan *)
4670 items->spec)->inner_type;
4672 ((const struct rte_flow_item_vlan *)
4673 items->mask)->inner_type;
4674 ether_type = rte_be_to_cpu_16(ether_type);
4678 /* Store outer VLAN mask for of_push_vlan action. */
4680 vlan_m = items->mask;
4682 case RTE_FLOW_ITEM_TYPE_IPV4:
4683 mlx5_flow_tunnel_ip_check(items, next_protocol,
4684 &item_flags, &tunnel);
4685 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
4692 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
4693 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4694 if (items->mask != NULL &&
4695 ((const struct rte_flow_item_ipv4 *)
4696 items->mask)->hdr.next_proto_id) {
4698 ((const struct rte_flow_item_ipv4 *)
4699 (items->spec))->hdr.next_proto_id;
4701 ((const struct rte_flow_item_ipv4 *)
4702 (items->mask))->hdr.next_proto_id;
4704 /* Reset for inner layer. */
4705 next_protocol = 0xff;
4708 case RTE_FLOW_ITEM_TYPE_IPV6:
4709 mlx5_flow_tunnel_ip_check(items, next_protocol,
4710 &item_flags, &tunnel);
4711 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
4718 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
4719 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4720 if (items->mask != NULL &&
4721 ((const struct rte_flow_item_ipv6 *)
4722 items->mask)->hdr.proto) {
4724 ((const struct rte_flow_item_ipv6 *)
4725 items->spec)->hdr.proto;
4727 ((const struct rte_flow_item_ipv6 *)
4728 items->spec)->hdr.proto;
4730 ((const struct rte_flow_item_ipv6 *)
4731 items->mask)->hdr.proto;
4733 /* Reset for inner layer. */
4734 next_protocol = 0xff;
4737 case RTE_FLOW_ITEM_TYPE_TCP:
4738 ret = mlx5_flow_validate_item_tcp
4745 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
4746 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4748 case RTE_FLOW_ITEM_TYPE_UDP:
4749 ret = mlx5_flow_validate_item_udp(items, item_flags,
4754 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
4755 MLX5_FLOW_LAYER_OUTER_L4_UDP;
4757 case RTE_FLOW_ITEM_TYPE_GRE:
4758 ret = mlx5_flow_validate_item_gre(items, item_flags,
4759 next_protocol, error);
4763 last_item = MLX5_FLOW_LAYER_GRE;
4765 case RTE_FLOW_ITEM_TYPE_NVGRE:
4766 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
4771 last_item = MLX5_FLOW_LAYER_NVGRE;
4773 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
4774 ret = mlx5_flow_validate_item_gre_key
4775 (items, item_flags, gre_item, error);
4778 last_item = MLX5_FLOW_LAYER_GRE_KEY;
4780 case RTE_FLOW_ITEM_TYPE_VXLAN:
4781 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
4785 last_item = MLX5_FLOW_LAYER_VXLAN;
4787 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
4788 ret = mlx5_flow_validate_item_vxlan_gpe(items,
4793 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
4795 case RTE_FLOW_ITEM_TYPE_GENEVE:
4796 ret = mlx5_flow_validate_item_geneve(items,
4801 last_item = MLX5_FLOW_LAYER_GENEVE;
4803 case RTE_FLOW_ITEM_TYPE_MPLS:
4804 ret = mlx5_flow_validate_item_mpls(dev, items,
4809 last_item = MLX5_FLOW_LAYER_MPLS;
4812 case RTE_FLOW_ITEM_TYPE_MARK:
4813 ret = flow_dv_validate_item_mark(dev, items, attr,
4817 last_item = MLX5_FLOW_ITEM_MARK;
4819 case RTE_FLOW_ITEM_TYPE_META:
4820 ret = flow_dv_validate_item_meta(dev, items, attr,
4824 last_item = MLX5_FLOW_ITEM_METADATA;
4826 case RTE_FLOW_ITEM_TYPE_ICMP:
4827 ret = mlx5_flow_validate_item_icmp(items, item_flags,
4832 last_item = MLX5_FLOW_LAYER_ICMP;
4834 case RTE_FLOW_ITEM_TYPE_ICMP6:
4835 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
4840 item_ipv6_proto = IPPROTO_ICMPV6;
4841 last_item = MLX5_FLOW_LAYER_ICMP6;
4843 case RTE_FLOW_ITEM_TYPE_TAG:
4844 ret = flow_dv_validate_item_tag(dev, items,
4848 last_item = MLX5_FLOW_ITEM_TAG;
4850 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
4851 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
4853 case RTE_FLOW_ITEM_TYPE_GTP:
4854 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
4858 last_item = MLX5_FLOW_LAYER_GTP;
4861 return rte_flow_error_set(error, ENOTSUP,
4862 RTE_FLOW_ERROR_TYPE_ITEM,
4863 NULL, "item not supported");
4865 item_flags |= last_item;
4867 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4868 int type = actions->type;
4869 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
4870 return rte_flow_error_set(error, ENOTSUP,
4871 RTE_FLOW_ERROR_TYPE_ACTION,
4872 actions, "too many actions");
4874 case RTE_FLOW_ACTION_TYPE_VOID:
4876 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4877 ret = flow_dv_validate_action_port_id(dev,
4884 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4887 case RTE_FLOW_ACTION_TYPE_FLAG:
4888 ret = flow_dv_validate_action_flag(dev, action_flags,
4892 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
4893 /* Count all modify-header actions as one. */
4894 if (!(action_flags &
4895 MLX5_FLOW_MODIFY_HDR_ACTIONS))
4897 action_flags |= MLX5_FLOW_ACTION_FLAG |
4898 MLX5_FLOW_ACTION_MARK_EXT;
4900 action_flags |= MLX5_FLOW_ACTION_FLAG;
4904 case RTE_FLOW_ACTION_TYPE_MARK:
4905 ret = flow_dv_validate_action_mark(dev, actions,
4910 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
4911 /* Count all modify-header actions as one. */
4912 if (!(action_flags &
4913 MLX5_FLOW_MODIFY_HDR_ACTIONS))
4915 action_flags |= MLX5_FLOW_ACTION_MARK |
4916 MLX5_FLOW_ACTION_MARK_EXT;
4918 action_flags |= MLX5_FLOW_ACTION_MARK;
4922 case RTE_FLOW_ACTION_TYPE_SET_META:
4923 ret = flow_dv_validate_action_set_meta(dev, actions,
4928 /* Count all modify-header actions as one action. */
4929 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
4931 action_flags |= MLX5_FLOW_ACTION_SET_META;
4933 case RTE_FLOW_ACTION_TYPE_SET_TAG:
4934 ret = flow_dv_validate_action_set_tag(dev, actions,
4939 /* Count all modify-header actions as one action. */
4940 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
4942 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
4944 case RTE_FLOW_ACTION_TYPE_DROP:
4945 ret = mlx5_flow_validate_action_drop(action_flags,
4949 action_flags |= MLX5_FLOW_ACTION_DROP;
4952 case RTE_FLOW_ACTION_TYPE_QUEUE:
4953 ret = mlx5_flow_validate_action_queue(actions,
4958 queue_index = ((const struct rte_flow_action_queue *)
4959 (actions->conf))->index;
4960 action_flags |= MLX5_FLOW_ACTION_QUEUE;
4963 case RTE_FLOW_ACTION_TYPE_RSS:
4964 rss = actions->conf;
4965 ret = mlx5_flow_validate_action_rss(actions,
4971 if (rss != NULL && rss->queue_num)
4972 queue_index = rss->queue[0];
4973 action_flags |= MLX5_FLOW_ACTION_RSS;
4976 case RTE_FLOW_ACTION_TYPE_COUNT:
4977 ret = flow_dv_validate_action_count(dev, error);
4980 action_flags |= MLX5_FLOW_ACTION_COUNT;
4983 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4984 if (flow_dv_validate_action_pop_vlan(dev,
4990 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
4993 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4994 ret = flow_dv_validate_action_push_vlan(dev,
5001 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5004 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5005 ret = flow_dv_validate_action_set_vlan_pcp
5006 (action_flags, actions, error);
5009 /* Count PCP with push_vlan command. */
5010 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5012 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5013 ret = flow_dv_validate_action_set_vlan_vid
5014 (item_flags, action_flags,
5018 /* Count VID with push_vlan command. */
5019 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5021 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5022 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5023 ret = flow_dv_validate_action_l2_encap(dev,
5029 action_flags |= MLX5_FLOW_ACTION_ENCAP;
5032 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5033 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5034 ret = flow_dv_validate_action_decap(dev, action_flags,
5038 action_flags |= MLX5_FLOW_ACTION_DECAP;
5041 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5042 ret = flow_dv_validate_action_raw_encap_decap
5043 (dev, NULL, actions->conf, attr, &action_flags,
5048 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5049 decap = actions->conf;
5050 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5052 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5056 encap = actions->conf;
5058 ret = flow_dv_validate_action_raw_encap_decap
5060 decap ? decap : &empty_decap, encap,
5061 attr, &action_flags, &actions_n,
5066 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5067 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5068 ret = flow_dv_validate_action_modify_mac(action_flags,
5074 /* Count all modify-header actions as one action. */
5075 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5077 action_flags |= actions->type ==
5078 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5079 MLX5_FLOW_ACTION_SET_MAC_SRC :
5080 MLX5_FLOW_ACTION_SET_MAC_DST;
5083 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5084 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5085 ret = flow_dv_validate_action_modify_ipv4(action_flags,
5091 /* Count all modify-header actions as one action. */
5092 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5094 action_flags |= actions->type ==
5095 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5096 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5097 MLX5_FLOW_ACTION_SET_IPV4_DST;
5099 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5100 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5101 ret = flow_dv_validate_action_modify_ipv6(action_flags,
5107 if (item_ipv6_proto == IPPROTO_ICMPV6)
5108 return rte_flow_error_set(error, ENOTSUP,
5109 RTE_FLOW_ERROR_TYPE_ACTION,
5111 "Can't change header "
5112 "with ICMPv6 proto");
5113 /* Count all modify-header actions as one action. */
5114 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5116 action_flags |= actions->type ==
5117 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5118 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5119 MLX5_FLOW_ACTION_SET_IPV6_DST;
5121 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5122 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5123 ret = flow_dv_validate_action_modify_tp(action_flags,
5129 /* Count all modify-header actions as one action. */
5130 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5132 action_flags |= actions->type ==
5133 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5134 MLX5_FLOW_ACTION_SET_TP_SRC :
5135 MLX5_FLOW_ACTION_SET_TP_DST;
5137 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5138 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5139 ret = flow_dv_validate_action_modify_ttl(action_flags,
5145 /* Count all modify-header actions as one action. */
5146 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5148 action_flags |= actions->type ==
5149 RTE_FLOW_ACTION_TYPE_SET_TTL ?
5150 MLX5_FLOW_ACTION_SET_TTL :
5151 MLX5_FLOW_ACTION_DEC_TTL;
5153 case RTE_FLOW_ACTION_TYPE_JUMP:
5154 ret = flow_dv_validate_action_jump(actions,
5161 action_flags |= MLX5_FLOW_ACTION_JUMP;
5163 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5164 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5165 ret = flow_dv_validate_action_modify_tcp_seq
5172 /* Count all modify-header actions as one action. */
5173 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5175 action_flags |= actions->type ==
5176 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5177 MLX5_FLOW_ACTION_INC_TCP_SEQ :
5178 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5180 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5181 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5182 ret = flow_dv_validate_action_modify_tcp_ack
5189 /* Count all modify-header actions as one action. */
5190 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5192 action_flags |= actions->type ==
5193 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5194 MLX5_FLOW_ACTION_INC_TCP_ACK :
5195 MLX5_FLOW_ACTION_DEC_TCP_ACK;
5197 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5198 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5199 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5201 case RTE_FLOW_ACTION_TYPE_METER:
5202 ret = mlx5_flow_validate_action_meter(dev,
5208 action_flags |= MLX5_FLOW_ACTION_METER;
5211 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5212 ret = flow_dv_validate_action_modify_ipv4_dscp
5219 /* Count all modify-header actions as one action. */
5220 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5222 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5224 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5225 ret = flow_dv_validate_action_modify_ipv6_dscp
5232 /* Count all modify-header actions as one action. */
5233 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5235 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5238 return rte_flow_error_set(error, ENOTSUP,
5239 RTE_FLOW_ERROR_TYPE_ACTION,
5241 "action not supported");
5245 * Validate the drop action mutual exclusion with other actions.
5246 * Drop action is mutually-exclusive with any other action, except for
5249 if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
5250 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
5251 return rte_flow_error_set(error, EINVAL,
5252 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5253 "Drop action is mutually-exclusive "
5254 "with any other action, except for "
5256 /* Eswitch has few restrictions on using items and actions */
5257 if (attr->transfer) {
5258 if (!mlx5_flow_ext_mreg_supported(dev) &&
5259 action_flags & MLX5_FLOW_ACTION_FLAG)
5260 return rte_flow_error_set(error, ENOTSUP,
5261 RTE_FLOW_ERROR_TYPE_ACTION,
5263 "unsupported action FLAG");
5264 if (!mlx5_flow_ext_mreg_supported(dev) &&
5265 action_flags & MLX5_FLOW_ACTION_MARK)
5266 return rte_flow_error_set(error, ENOTSUP,
5267 RTE_FLOW_ERROR_TYPE_ACTION,
5269 "unsupported action MARK");
5270 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
5271 return rte_flow_error_set(error, ENOTSUP,
5272 RTE_FLOW_ERROR_TYPE_ACTION,
5274 "unsupported action QUEUE");
5275 if (action_flags & MLX5_FLOW_ACTION_RSS)
5276 return rte_flow_error_set(error, ENOTSUP,
5277 RTE_FLOW_ERROR_TYPE_ACTION,
5279 "unsupported action RSS");
5280 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
5281 return rte_flow_error_set(error, EINVAL,
5282 RTE_FLOW_ERROR_TYPE_ACTION,
5284 "no fate action is found");
5286 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
5287 return rte_flow_error_set(error, EINVAL,
5288 RTE_FLOW_ERROR_TYPE_ACTION,
5290 "no fate action is found");
5292 /* Continue validation for Xcap actions.*/
5293 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) && (queue_index == 0xFFFF ||
5294 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
5295 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
5296 MLX5_FLOW_XCAP_ACTIONS)
5297 return rte_flow_error_set(error, ENOTSUP,
5298 RTE_FLOW_ERROR_TYPE_ACTION,
5299 NULL, "encap and decap "
5300 "combination aren't supported");
5301 if (!attr->transfer && attr->ingress && (action_flags &
5302 MLX5_FLOW_ACTION_ENCAP))
5303 return rte_flow_error_set(error, ENOTSUP,
5304 RTE_FLOW_ERROR_TYPE_ACTION,
5305 NULL, "encap is not supported"
5306 " for ingress traffic");
5312 * Internal preparation function. Allocates the DV flow size,
5313 * this size is constant.
5316 * Pointer to the rte_eth_dev structure.
5318 * Pointer to the flow attributes.
5320 * Pointer to the list of items.
5321 * @param[in] actions
5322 * Pointer to the list of actions.
5324 * Pointer to the error structure.
5327 * Pointer to mlx5_flow object on success,
5328 * otherwise NULL and rte_errno is set.
5330 static struct mlx5_flow *
5331 flow_dv_prepare(struct rte_eth_dev *dev,
5332 const struct rte_flow_attr *attr __rte_unused,
5333 const struct rte_flow_item items[] __rte_unused,
5334 const struct rte_flow_action actions[] __rte_unused,
5335 struct rte_flow_error *error)
5337 uint32_t handle_idx = 0;
5338 struct mlx5_flow *dev_flow;
5339 struct mlx5_flow_handle *dev_handle;
5340 struct mlx5_priv *priv = dev->data->dev_private;
5342 /* In case of corrupting the memory. */
5343 if (priv->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
5344 rte_flow_error_set(error, ENOSPC,
5345 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5346 "not free temporary device flow");
5349 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
5352 rte_flow_error_set(error, ENOMEM,
5353 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5354 "not enough memory to create flow handle");
5357 /* No multi-thread supporting. */
5358 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[priv->flow_idx++];
5359 dev_flow->handle = dev_handle;
5360 dev_flow->handle_idx = handle_idx;
5361 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
5363 * The matching value needs to be cleared to 0 before using. In the
5364 * past, it will be automatically cleared when using rte_*alloc
5365 * API. The time consumption will be almost the same as before.
5367 memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
5368 dev_flow->ingress = attr->ingress;
5369 dev_flow->dv.transfer = attr->transfer;
5373 #ifdef RTE_LIBRTE_MLX5_DEBUG
5375 * Sanity check for match mask and value. Similar to check_valid_spec() in
5376 * kernel driver. If unmasked bit is present in value, it returns failure.
5379 * pointer to match mask buffer.
5380 * @param match_value
5381 * pointer to match value buffer.
5384 * 0 if valid, -EINVAL otherwise.
5387 flow_dv_check_valid_spec(void *match_mask, void *match_value)
5389 uint8_t *m = match_mask;
5390 uint8_t *v = match_value;
5393 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
5396 "match_value differs from match_criteria"
5397 " %p[%u] != %p[%u]",
5398 match_value, i, match_mask, i);
5407 * Add Ethernet item to matcher and to the value.
5409 * @param[in, out] matcher
5411 * @param[in, out] key
5412 * Flow matcher value.
5414 * Flow pattern to translate.
5416 * Item is inner pattern.
5419 flow_dv_translate_item_eth(void *matcher, void *key,
5420 const struct rte_flow_item *item, int inner)
5422 const struct rte_flow_item_eth *eth_m = item->mask;
5423 const struct rte_flow_item_eth *eth_v = item->spec;
5424 const struct rte_flow_item_eth nic_mask = {
5425 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5426 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5427 .type = RTE_BE16(0xffff),
5439 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5441 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5443 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5445 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5447 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
5448 ð_m->dst, sizeof(eth_m->dst));
5449 /* The value must be in the range of the mask. */
5450 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
5451 for (i = 0; i < sizeof(eth_m->dst); ++i)
5452 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
5453 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
5454 ð_m->src, sizeof(eth_m->src));
5455 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
5456 /* The value must be in the range of the mask. */
5457 for (i = 0; i < sizeof(eth_m->dst); ++i)
5458 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
5460 /* When ethertype is present set mask for tagged VLAN. */
5461 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5462 /* Set value for tagged VLAN if ethertype is 802.1Q. */
5463 if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
5464 eth_v->type == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
5465 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag,
5467 /* Return here to avoid setting match on ethertype. */
5472 * HW supports match on one Ethertype, the Ethertype following the last
5473 * VLAN tag of the packet (see PRM).
5474 * Set match on ethertype only if ETH header is not followed by VLAN.
5476 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
5477 rte_be_to_cpu_16(eth_m->type));
5478 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, ethertype);
5479 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
5483 * Add VLAN item to matcher and to the value.
5485 * @param[in, out] dev_flow
5487 * @param[in, out] matcher
5489 * @param[in, out] key
5490 * Flow matcher value.
5492 * Flow pattern to translate.
5494 * Item is inner pattern.
5497 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
5498 void *matcher, void *key,
5499 const struct rte_flow_item *item,
5502 const struct rte_flow_item_vlan *vlan_m = item->mask;
5503 const struct rte_flow_item_vlan *vlan_v = item->spec;
5512 vlan_m = &rte_flow_item_vlan_mask;
5514 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5516 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5518 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5520 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5522 * This is workaround, masks are not supported,
5523 * and pre-validated.
5525 dev_flow->handle->vf_vlan.tag =
5526 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
5528 tci_m = rte_be_to_cpu_16(vlan_m->tci);
5529 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
5530 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5531 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
5532 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
5533 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
5534 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
5535 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
5536 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
5537 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
5538 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
5539 rte_be_to_cpu_16(vlan_m->inner_type));
5540 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
5541 rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
5545 * Add IPV4 item to matcher and to the value.
5547 * @param[in, out] matcher
5549 * @param[in, out] key
5550 * Flow matcher value.
5552 * Flow pattern to translate.
5553 * @param[in] item_flags
5554 * Bit-fields that holds the items detected until now.
5556 * Item is inner pattern.
5558 * The group to insert the rule.
5561 flow_dv_translate_item_ipv4(void *matcher, void *key,
5562 const struct rte_flow_item *item,
5563 const uint64_t item_flags,
5564 int inner, uint32_t group)
5566 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
5567 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
5568 const struct rte_flow_item_ipv4 nic_mask = {
5570 .src_addr = RTE_BE32(0xffffffff),
5571 .dst_addr = RTE_BE32(0xffffffff),
5572 .type_of_service = 0xff,
5573 .next_proto_id = 0xff,
5574 .time_to_live = 0xff,
5584 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5586 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5588 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5590 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5593 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5595 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x4);
5596 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 4);
5598 * On outer header (which must contains L2), or inner header with L2,
5599 * set cvlan_tag mask bit to mark this packet as untagged.
5600 * This should be done even if item->spec is empty.
5602 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
5603 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5608 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5609 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
5610 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5611 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
5612 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
5613 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
5614 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5615 src_ipv4_src_ipv6.ipv4_layout.ipv4);
5616 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5617 src_ipv4_src_ipv6.ipv4_layout.ipv4);
5618 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
5619 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
5620 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
5621 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
5622 ipv4_m->hdr.type_of_service);
5623 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
5624 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
5625 ipv4_m->hdr.type_of_service >> 2);
5626 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
5627 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
5628 ipv4_m->hdr.next_proto_id);
5629 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
5630 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
5631 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
5632 ipv4_m->hdr.time_to_live);
5633 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
5634 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
5638 * Add IPV6 item to matcher and to the value.
5640 * @param[in, out] matcher
5642 * @param[in, out] key
5643 * Flow matcher value.
5645 * Flow pattern to translate.
5646 * @param[in] item_flags
5647 * Bit-fields that holds the items detected until now.
5649 * Item is inner pattern.
5651 * The group to insert the rule.
5654 flow_dv_translate_item_ipv6(void *matcher, void *key,
5655 const struct rte_flow_item *item,
5656 const uint64_t item_flags,
5657 int inner, uint32_t group)
5659 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
5660 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
5661 const struct rte_flow_item_ipv6 nic_mask = {
5664 "\xff\xff\xff\xff\xff\xff\xff\xff"
5665 "\xff\xff\xff\xff\xff\xff\xff\xff",
5667 "\xff\xff\xff\xff\xff\xff\xff\xff"
5668 "\xff\xff\xff\xff\xff\xff\xff\xff",
5669 .vtc_flow = RTE_BE32(0xffffffff),
5676 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5677 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5686 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5688 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5690 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5692 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5695 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5697 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x6);
5698 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 6);
5700 * On outer header (which must contains L2), or inner header with L2,
5701 * set cvlan_tag mask bit to mark this packet as untagged.
5702 * This should be done even if item->spec is empty.
5704 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
5705 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5710 size = sizeof(ipv6_m->hdr.dst_addr);
5711 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5712 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
5713 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5714 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
5715 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
5716 for (i = 0; i < size; ++i)
5717 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
5718 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5719 src_ipv4_src_ipv6.ipv6_layout.ipv6);
5720 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5721 src_ipv4_src_ipv6.ipv6_layout.ipv6);
5722 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
5723 for (i = 0; i < size; ++i)
5724 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
5726 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
5727 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
5728 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
5729 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
5730 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
5731 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
5734 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
5736 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
5739 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
5741 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
5745 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
5747 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
5748 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
5750 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
5751 ipv6_m->hdr.hop_limits);
5752 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
5753 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
5757 * Add TCP item to matcher and to the value.
5759 * @param[in, out] matcher
5761 * @param[in, out] key
5762 * Flow matcher value.
5764 * Flow pattern to translate.
5766 * Item is inner pattern.
5769 flow_dv_translate_item_tcp(void *matcher, void *key,
5770 const struct rte_flow_item *item,
5773 const struct rte_flow_item_tcp *tcp_m = item->mask;
5774 const struct rte_flow_item_tcp *tcp_v = item->spec;
5779 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5781 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5783 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5785 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5787 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5788 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
5792 tcp_m = &rte_flow_item_tcp_mask;
5793 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
5794 rte_be_to_cpu_16(tcp_m->hdr.src_port));
5795 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
5796 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
5797 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
5798 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
5799 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
5800 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
5801 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
5802 tcp_m->hdr.tcp_flags);
5803 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
5804 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
5808 * Add UDP item to matcher and to the value.
5810 * @param[in, out] matcher
5812 * @param[in, out] key
5813 * Flow matcher value.
5815 * Flow pattern to translate.
5817 * Item is inner pattern.
5820 flow_dv_translate_item_udp(void *matcher, void *key,
5821 const struct rte_flow_item *item,
5824 const struct rte_flow_item_udp *udp_m = item->mask;
5825 const struct rte_flow_item_udp *udp_v = item->spec;
5830 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5832 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5834 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5836 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5838 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5839 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
5843 udp_m = &rte_flow_item_udp_mask;
5844 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
5845 rte_be_to_cpu_16(udp_m->hdr.src_port));
5846 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
5847 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
5848 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
5849 rte_be_to_cpu_16(udp_m->hdr.dst_port));
5850 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
5851 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
5855 * Add GRE optional Key item to matcher and to the value.
5857 * @param[in, out] matcher
5859 * @param[in, out] key
5860 * Flow matcher value.
5862 * Flow pattern to translate.
5864 * Item is inner pattern.
5867 flow_dv_translate_item_gre_key(void *matcher, void *key,
5868 const struct rte_flow_item *item)
5870 const rte_be32_t *key_m = item->mask;
5871 const rte_be32_t *key_v = item->spec;
5872 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5873 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5874 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
5876 /* GRE K bit must be on and should already be validated */
5877 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
5878 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
5882 key_m = &gre_key_default_mask;
5883 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
5884 rte_be_to_cpu_32(*key_m) >> 8);
5885 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
5886 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
5887 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
5888 rte_be_to_cpu_32(*key_m) & 0xFF);
5889 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
5890 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
5894 * Add GRE item to matcher and to the value.
5896 * @param[in, out] matcher
5898 * @param[in, out] key
5899 * Flow matcher value.
5901 * Flow pattern to translate.
5903 * Item is inner pattern.
5906 flow_dv_translate_item_gre(void *matcher, void *key,
5907 const struct rte_flow_item *item,
5910 const struct rte_flow_item_gre *gre_m = item->mask;
5911 const struct rte_flow_item_gre *gre_v = item->spec;
5914 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5915 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5922 uint16_t s_present:1;
5923 uint16_t k_present:1;
5924 uint16_t rsvd_bit1:1;
5925 uint16_t c_present:1;
5929 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
5932 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5934 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5936 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5938 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5940 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5941 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
5945 gre_m = &rte_flow_item_gre_mask;
5946 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
5947 rte_be_to_cpu_16(gre_m->protocol));
5948 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
5949 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
5950 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
5951 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
5952 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
5953 gre_crks_rsvd0_ver_m.c_present);
5954 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
5955 gre_crks_rsvd0_ver_v.c_present &
5956 gre_crks_rsvd0_ver_m.c_present);
5957 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
5958 gre_crks_rsvd0_ver_m.k_present);
5959 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
5960 gre_crks_rsvd0_ver_v.k_present &
5961 gre_crks_rsvd0_ver_m.k_present);
5962 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
5963 gre_crks_rsvd0_ver_m.s_present);
5964 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
5965 gre_crks_rsvd0_ver_v.s_present &
5966 gre_crks_rsvd0_ver_m.s_present);
5970 * Add NVGRE item to matcher and to the value.
5972 * @param[in, out] matcher
5974 * @param[in, out] key
5975 * Flow matcher value.
5977 * Flow pattern to translate.
5979 * Item is inner pattern.
5982 flow_dv_translate_item_nvgre(void *matcher, void *key,
5983 const struct rte_flow_item *item,
5986 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
5987 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
5988 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5989 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5990 const char *tni_flow_id_m = (const char *)nvgre_m->tni;
5991 const char *tni_flow_id_v = (const char *)nvgre_v->tni;
5997 /* For NVGRE, GRE header fields must be set with defined values. */
5998 const struct rte_flow_item_gre gre_spec = {
5999 .c_rsvd0_ver = RTE_BE16(0x2000),
6000 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6002 const struct rte_flow_item_gre gre_mask = {
6003 .c_rsvd0_ver = RTE_BE16(0xB000),
6004 .protocol = RTE_BE16(UINT16_MAX),
6006 const struct rte_flow_item gre_item = {
6011 flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6015 nvgre_m = &rte_flow_item_nvgre_mask;
6016 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6017 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6018 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6019 memcpy(gre_key_m, tni_flow_id_m, size);
6020 for (i = 0; i < size; ++i)
6021 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6025 * Add VXLAN item to matcher and to the value.
6027 * @param[in, out] matcher
6029 * @param[in, out] key
6030 * Flow matcher value.
6032 * Flow pattern to translate.
6034 * Item is inner pattern.
6037 flow_dv_translate_item_vxlan(void *matcher, void *key,
6038 const struct rte_flow_item *item,
6041 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
6042 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
6045 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6046 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6054 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6056 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6058 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6060 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6062 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6063 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6064 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6065 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6066 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6071 vxlan_m = &rte_flow_item_vxlan_mask;
6072 size = sizeof(vxlan_m->vni);
6073 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
6074 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
6075 memcpy(vni_m, vxlan_m->vni, size);
6076 for (i = 0; i < size; ++i)
6077 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6081 * Add VXLAN-GPE item to matcher and to the value.
6083 * @param[in, out] matcher
6085 * @param[in, out] key
6086 * Flow matcher value.
6088 * Flow pattern to translate.
6090 * Item is inner pattern.
6094 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
6095 const struct rte_flow_item *item, int inner)
6097 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
6098 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
6102 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
6104 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6110 uint8_t flags_m = 0xff;
6111 uint8_t flags_v = 0xc;
6114 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6116 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6118 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6120 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6122 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6123 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6124 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6125 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6126 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6131 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
6132 size = sizeof(vxlan_m->vni);
6133 vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
6134 vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
6135 memcpy(vni_m, vxlan_m->vni, size);
6136 for (i = 0; i < size; ++i)
6137 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6138 if (vxlan_m->flags) {
6139 flags_m = vxlan_m->flags;
6140 flags_v = vxlan_v->flags;
6142 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
6143 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
6144 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
6146 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
6151 * Add Geneve item to matcher and to the value.
6153 * @param[in, out] matcher
6155 * @param[in, out] key
6156 * Flow matcher value.
6158 * Flow pattern to translate.
6160 * Item is inner pattern.
6164 flow_dv_translate_item_geneve(void *matcher, void *key,
6165 const struct rte_flow_item *item, int inner)
6167 const struct rte_flow_item_geneve *geneve_m = item->mask;
6168 const struct rte_flow_item_geneve *geneve_v = item->spec;
6171 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6172 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6181 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6183 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6185 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6187 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6189 dport = MLX5_UDP_PORT_GENEVE;
6190 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6191 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6192 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6197 geneve_m = &rte_flow_item_geneve_mask;
6198 size = sizeof(geneve_m->vni);
6199 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
6200 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
6201 memcpy(vni_m, geneve_m->vni, size);
6202 for (i = 0; i < size; ++i)
6203 vni_v[i] = vni_m[i] & geneve_v->vni[i];
6204 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
6205 rte_be_to_cpu_16(geneve_m->protocol));
6206 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
6207 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
6208 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
6209 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
6210 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
6211 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6212 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
6213 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6214 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
6215 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6216 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
6217 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
6218 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6222 * Add MPLS item to matcher and to the value.
6224 * @param[in, out] matcher
6226 * @param[in, out] key
6227 * Flow matcher value.
6229 * Flow pattern to translate.
6230 * @param[in] prev_layer
6231 * The protocol layer indicated in previous item.
6233 * Item is inner pattern.
6236 flow_dv_translate_item_mpls(void *matcher, void *key,
6237 const struct rte_flow_item *item,
6238 uint64_t prev_layer,
6241 const uint32_t *in_mpls_m = item->mask;
6242 const uint32_t *in_mpls_v = item->spec;
6243 uint32_t *out_mpls_m = 0;
6244 uint32_t *out_mpls_v = 0;
6245 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6246 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6247 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
6249 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6250 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
6251 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6253 switch (prev_layer) {
6254 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6255 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
6256 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6257 MLX5_UDP_PORT_MPLS);
6259 case MLX5_FLOW_LAYER_GRE:
6260 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
6261 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6262 RTE_ETHER_TYPE_MPLS);
6265 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6266 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6273 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
6274 switch (prev_layer) {
6275 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6277 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6278 outer_first_mpls_over_udp);
6280 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6281 outer_first_mpls_over_udp);
6283 case MLX5_FLOW_LAYER_GRE:
6285 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6286 outer_first_mpls_over_gre);
6288 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6289 outer_first_mpls_over_gre);
6292 /* Inner MPLS not over GRE is not supported. */
6295 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6299 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6305 if (out_mpls_m && out_mpls_v) {
6306 *out_mpls_m = *in_mpls_m;
6307 *out_mpls_v = *in_mpls_v & *in_mpls_m;
6312 * Add metadata register item to matcher
6314 * @param[in, out] matcher
6316 * @param[in, out] key
6317 * Flow matcher value.
6318 * @param[in] reg_type
6319 * Type of device metadata register
6326 flow_dv_match_meta_reg(void *matcher, void *key,
6327 enum modify_reg reg_type,
6328 uint32_t data, uint32_t mask)
6331 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
6333 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6339 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
6340 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
6343 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
6344 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
6348 * The metadata register C0 field might be divided into
6349 * source vport index and META item value, we should set
6350 * this field according to specified mask, not as whole one.
6352 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
6354 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
6355 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
6358 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
6361 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
6362 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
6365 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
6366 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
6369 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
6370 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
6373 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
6374 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
6377 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
6378 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
6381 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
6382 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
6385 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
6386 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
6395 * Add MARK item to matcher
6398 * The device to configure through.
6399 * @param[in, out] matcher
6401 * @param[in, out] key
6402 * Flow matcher value.
6404 * Flow pattern to translate.
6407 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
6408 void *matcher, void *key,
6409 const struct rte_flow_item *item)
6411 struct mlx5_priv *priv = dev->data->dev_private;
6412 const struct rte_flow_item_mark *mark;
6416 mark = item->mask ? (const void *)item->mask :
6417 &rte_flow_item_mark_mask;
6418 mask = mark->id & priv->sh->dv_mark_mask;
6419 mark = (const void *)item->spec;
6421 value = mark->id & priv->sh->dv_mark_mask & mask;
6423 enum modify_reg reg;
6425 /* Get the metadata register index for the mark. */
6426 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
6427 MLX5_ASSERT(reg > 0);
6428 if (reg == REG_C_0) {
6429 struct mlx5_priv *priv = dev->data->dev_private;
6430 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6431 uint32_t shl_c0 = rte_bsf32(msk_c0);
6437 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6442 * Add META item to matcher
6445 * The devich to configure through.
6446 * @param[in, out] matcher
6448 * @param[in, out] key
6449 * Flow matcher value.
6451 * Attributes of flow that includes this item.
6453 * Flow pattern to translate.
6456 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
6457 void *matcher, void *key,
6458 const struct rte_flow_attr *attr,
6459 const struct rte_flow_item *item)
6461 const struct rte_flow_item_meta *meta_m;
6462 const struct rte_flow_item_meta *meta_v;
6464 meta_m = (const void *)item->mask;
6466 meta_m = &rte_flow_item_meta_mask;
6467 meta_v = (const void *)item->spec;
6470 uint32_t value = meta_v->data;
6471 uint32_t mask = meta_m->data;
6473 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
6477 * In datapath code there is no endianness
6478 * coversions for perfromance reasons, all
6479 * pattern conversions are done in rte_flow.
6481 value = rte_cpu_to_be_32(value);
6482 mask = rte_cpu_to_be_32(mask);
6483 if (reg == REG_C_0) {
6484 struct mlx5_priv *priv = dev->data->dev_private;
6485 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6486 uint32_t shl_c0 = rte_bsf32(msk_c0);
6487 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
6488 uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
6495 MLX5_ASSERT(msk_c0);
6496 MLX5_ASSERT(!(~msk_c0 & mask));
6498 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6503 * Add vport metadata Reg C0 item to matcher
6505 * @param[in, out] matcher
6507 * @param[in, out] key
6508 * Flow matcher value.
6510 * Flow pattern to translate.
6513 flow_dv_translate_item_meta_vport(void *matcher, void *key,
6514 uint32_t value, uint32_t mask)
6516 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
6520 * Add tag item to matcher
6523 * The devich to configure through.
6524 * @param[in, out] matcher
6526 * @param[in, out] key
6527 * Flow matcher value.
6529 * Flow pattern to translate.
6532 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
6533 void *matcher, void *key,
6534 const struct rte_flow_item *item)
6536 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
6537 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
6538 uint32_t mask, value;
6541 value = tag_v->data;
6542 mask = tag_m ? tag_m->data : UINT32_MAX;
6543 if (tag_v->id == REG_C_0) {
6544 struct mlx5_priv *priv = dev->data->dev_private;
6545 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6546 uint32_t shl_c0 = rte_bsf32(msk_c0);
6552 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
6556 * Add TAG item to matcher
6559 * The devich to configure through.
6560 * @param[in, out] matcher
6562 * @param[in, out] key
6563 * Flow matcher value.
6565 * Flow pattern to translate.
6568 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
6569 void *matcher, void *key,
6570 const struct rte_flow_item *item)
6572 const struct rte_flow_item_tag *tag_v = item->spec;
6573 const struct rte_flow_item_tag *tag_m = item->mask;
6574 enum modify_reg reg;
6577 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
6578 /* Get the metadata register index for the tag. */
6579 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
6580 MLX5_ASSERT(reg > 0);
6581 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
6585 * Add source vport match to the specified matcher.
6587 * @param[in, out] matcher
6589 * @param[in, out] key
6590 * Flow matcher value.
6592 * Source vport value to match
6597 flow_dv_translate_item_source_vport(void *matcher, void *key,
6598 int16_t port, uint16_t mask)
6600 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6601 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6603 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
6604 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
6608 * Translate port-id item to eswitch match on port-id.
6611 * The devich to configure through.
6612 * @param[in, out] matcher
6614 * @param[in, out] key
6615 * Flow matcher value.
6617 * Flow pattern to translate.
6620 * 0 on success, a negative errno value otherwise.
6623 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
6624 void *key, const struct rte_flow_item *item)
6626 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
6627 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
6628 struct mlx5_priv *priv;
6631 mask = pid_m ? pid_m->id : 0xffff;
6632 id = pid_v ? pid_v->id : dev->data->port_id;
6633 priv = mlx5_port_to_eswitch_info(id, item == NULL);
6636 /* Translate to vport field or to metadata, depending on mode. */
6637 if (priv->vport_meta_mask)
6638 flow_dv_translate_item_meta_vport(matcher, key,
6639 priv->vport_meta_tag,
6640 priv->vport_meta_mask);
6642 flow_dv_translate_item_source_vport(matcher, key,
6643 priv->vport_id, mask);
6648 * Add ICMP6 item to matcher and to the value.
6650 * @param[in, out] matcher
6652 * @param[in, out] key
6653 * Flow matcher value.
6655 * Flow pattern to translate.
6657 * Item is inner pattern.
6660 flow_dv_translate_item_icmp6(void *matcher, void *key,
6661 const struct rte_flow_item *item,
6664 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
6665 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
6668 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6670 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6672 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6674 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6676 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6678 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6680 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
6681 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
6685 icmp6_m = &rte_flow_item_icmp6_mask;
6687 * Force flow only to match the non-fragmented IPv6 ICMPv6 packets.
6688 * If only the protocol is specified, no need to match the frag.
6690 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6691 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
6692 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
6693 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
6694 icmp6_v->type & icmp6_m->type);
6695 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
6696 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
6697 icmp6_v->code & icmp6_m->code);
6701 * Add ICMP item to matcher and to the value.
6703 * @param[in, out] matcher
6705 * @param[in, out] key
6706 * Flow matcher value.
6708 * Flow pattern to translate.
6710 * Item is inner pattern.
6713 flow_dv_translate_item_icmp(void *matcher, void *key,
6714 const struct rte_flow_item *item,
6717 const struct rte_flow_item_icmp *icmp_m = item->mask;
6718 const struct rte_flow_item_icmp *icmp_v = item->spec;
6721 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6723 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6725 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6727 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6729 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6731 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6733 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
6734 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
6738 icmp_m = &rte_flow_item_icmp_mask;
6740 * Force flow only to match the non-fragmented IPv4 ICMP packets.
6741 * If only the protocol is specified, no need to match the frag.
6743 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6744 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
6745 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
6746 icmp_m->hdr.icmp_type);
6747 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
6748 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
6749 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
6750 icmp_m->hdr.icmp_code);
6751 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
6752 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
6756 * Add GTP item to matcher and to the value.
6758 * @param[in, out] matcher
6760 * @param[in, out] key
6761 * Flow matcher value.
6763 * Flow pattern to translate.
6765 * Item is inner pattern.
6768 flow_dv_translate_item_gtp(void *matcher, void *key,
6769 const struct rte_flow_item *item, int inner)
6771 const struct rte_flow_item_gtp *gtp_m = item->mask;
6772 const struct rte_flow_item_gtp *gtp_v = item->spec;
6775 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6777 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6778 uint16_t dport = RTE_GTPU_UDP_PORT;
6781 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6783 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6785 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6787 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6789 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6790 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6791 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6796 gtp_m = &rte_flow_item_gtp_mask;
6797 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
6798 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
6799 gtp_v->msg_type & gtp_m->msg_type);
6800 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
6801 rte_be_to_cpu_32(gtp_m->teid));
6802 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
6803 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
6806 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
6808 #define HEADER_IS_ZERO(match_criteria, headers) \
6809 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
6810 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
6813 * Calculate flow matcher enable bitmap.
6815 * @param match_criteria
6816 * Pointer to flow matcher criteria.
6819 * Bitmap of enabled fields.
6822 flow_dv_matcher_enable(uint32_t *match_criteria)
6824 uint8_t match_criteria_enable;
6826 match_criteria_enable =
6827 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
6828 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
6829 match_criteria_enable |=
6830 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
6831 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
6832 match_criteria_enable |=
6833 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
6834 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
6835 match_criteria_enable |=
6836 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
6837 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
6838 match_criteria_enable |=
6839 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
6840 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
6841 return match_criteria_enable;
6848 * @param[in, out] dev
6849 * Pointer to rte_eth_dev structure.
6850 * @param[in] table_id
6853 * Direction of the table.
6854 * @param[in] transfer
6855 * E-Switch or NIC flow.
6857 * pointer to error structure.
6860 * Returns tables resource based on the index, NULL in case of failed.
6862 static struct mlx5_flow_tbl_resource *
6863 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
6864 uint32_t table_id, uint8_t egress,
6866 struct rte_flow_error *error)
6868 struct mlx5_priv *priv = dev->data->dev_private;
6869 struct mlx5_ibv_shared *sh = priv->sh;
6870 struct mlx5_flow_tbl_resource *tbl;
6871 union mlx5_flow_tbl_key table_key = {
6873 .table_id = table_id,
6875 .domain = !!transfer,
6876 .direction = !!egress,
6879 struct mlx5_hlist_entry *pos = mlx5_hlist_lookup(sh->flow_tbls,
6881 struct mlx5_flow_tbl_data_entry *tbl_data;
6887 tbl_data = container_of(pos, struct mlx5_flow_tbl_data_entry,
6889 tbl = &tbl_data->tbl;
6890 rte_atomic32_inc(&tbl->refcnt);
6893 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
6895 rte_flow_error_set(error, ENOMEM,
6896 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6898 "cannot allocate flow table data entry");
6901 tbl_data->idx = idx;
6902 tbl = &tbl_data->tbl;
6903 pos = &tbl_data->entry;
6905 domain = sh->fdb_domain;
6907 domain = sh->tx_domain;
6909 domain = sh->rx_domain;
6910 tbl->obj = mlx5_glue->dr_create_flow_tbl(domain, table_id);
6912 rte_flow_error_set(error, ENOMEM,
6913 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6914 NULL, "cannot create flow table object");
6915 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
6919 * No multi-threads now, but still better to initialize the reference
6920 * count before insert it into the hash list.
6922 rte_atomic32_init(&tbl->refcnt);
6923 /* Jump action reference count is initialized here. */
6924 rte_atomic32_init(&tbl_data->jump.refcnt);
6925 pos->key = table_key.v64;
6926 ret = mlx5_hlist_insert(sh->flow_tbls, pos);
6928 rte_flow_error_set(error, -ret,
6929 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6930 "cannot insert flow table data entry");
6931 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
6932 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
6934 rte_atomic32_inc(&tbl->refcnt);
6939 * Release a flow table.
6942 * Pointer to rte_eth_dev structure.
6944 * Table resource to be released.
6947 * Returns 0 if table was released, else return 1;
6950 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
6951 struct mlx5_flow_tbl_resource *tbl)
6953 struct mlx5_priv *priv = dev->data->dev_private;
6954 struct mlx5_ibv_shared *sh = priv->sh;
6955 struct mlx5_flow_tbl_data_entry *tbl_data =
6956 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
6960 if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
6961 struct mlx5_hlist_entry *pos = &tbl_data->entry;
6963 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
6965 /* remove the entry from the hash list and free memory. */
6966 mlx5_hlist_remove(sh->flow_tbls, pos);
6967 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_JUMP],
6975 * Register the flow matcher.
6977 * @param[in, out] dev
6978 * Pointer to rte_eth_dev structure.
6979 * @param[in, out] matcher
6980 * Pointer to flow matcher.
6981 * @param[in, out] key
6982 * Pointer to flow table key.
6983 * @parm[in, out] dev_flow
6984 * Pointer to the dev_flow.
6986 * pointer to error structure.
6989 * 0 on success otherwise -errno and errno is set.
6992 flow_dv_matcher_register(struct rte_eth_dev *dev,
6993 struct mlx5_flow_dv_matcher *matcher,
6994 union mlx5_flow_tbl_key *key,
6995 struct mlx5_flow *dev_flow,
6996 struct rte_flow_error *error)
6998 struct mlx5_priv *priv = dev->data->dev_private;
6999 struct mlx5_ibv_shared *sh = priv->sh;
7000 struct mlx5_flow_dv_matcher *cache_matcher;
7001 struct mlx5dv_flow_matcher_attr dv_attr = {
7002 .type = IBV_FLOW_ATTR_NORMAL,
7003 .match_mask = (void *)&matcher->mask,
7005 struct mlx5_flow_tbl_resource *tbl;
7006 struct mlx5_flow_tbl_data_entry *tbl_data;
7008 tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
7009 key->domain, error);
7011 return -rte_errno; /* No need to refill the error info */
7012 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7013 /* Lookup from cache. */
7014 LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
7015 if (matcher->crc == cache_matcher->crc &&
7016 matcher->priority == cache_matcher->priority &&
7017 !memcmp((const void *)matcher->mask.buf,
7018 (const void *)cache_matcher->mask.buf,
7019 cache_matcher->mask.size)) {
7021 "%s group %u priority %hd use %s "
7022 "matcher %p: refcnt %d++",
7023 key->domain ? "FDB" : "NIC", key->table_id,
7024 cache_matcher->priority,
7025 key->direction ? "tx" : "rx",
7026 (void *)cache_matcher,
7027 rte_atomic32_read(&cache_matcher->refcnt));
7028 rte_atomic32_inc(&cache_matcher->refcnt);
7029 dev_flow->handle->dvh.matcher = cache_matcher;
7030 /* old matcher should not make the table ref++. */
7031 flow_dv_tbl_resource_release(dev, tbl);
7035 /* Register new matcher. */
7036 cache_matcher = rte_calloc(__func__, 1, sizeof(*cache_matcher), 0);
7037 if (!cache_matcher) {
7038 flow_dv_tbl_resource_release(dev, tbl);
7039 return rte_flow_error_set(error, ENOMEM,
7040 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7041 "cannot allocate matcher memory");
7043 *cache_matcher = *matcher;
7044 dv_attr.match_criteria_enable =
7045 flow_dv_matcher_enable(cache_matcher->mask.buf);
7046 dv_attr.priority = matcher->priority;
7048 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
7049 cache_matcher->matcher_object =
7050 mlx5_glue->dv_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj);
7051 if (!cache_matcher->matcher_object) {
7052 rte_free(cache_matcher);
7053 #ifdef HAVE_MLX5DV_DR
7054 flow_dv_tbl_resource_release(dev, tbl);
7056 return rte_flow_error_set(error, ENOMEM,
7057 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7058 NULL, "cannot create matcher");
7060 /* Save the table information */
7061 cache_matcher->tbl = tbl;
7062 rte_atomic32_init(&cache_matcher->refcnt);
7063 /* only matcher ref++, table ref++ already done above in get API. */
7064 rte_atomic32_inc(&cache_matcher->refcnt);
7065 LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
7066 dev_flow->handle->dvh.matcher = cache_matcher;
7067 DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
7068 key->domain ? "FDB" : "NIC", key->table_id,
7069 cache_matcher->priority,
7070 key->direction ? "tx" : "rx", (void *)cache_matcher,
7071 rte_atomic32_read(&cache_matcher->refcnt));
7076 * Find existing tag resource or create and register a new one.
7078 * @param dev[in, out]
7079 * Pointer to rte_eth_dev structure.
7080 * @param[in, out] tag_be24
7081 * Tag value in big endian then R-shift 8.
7082 * @parm[in, out] dev_flow
7083 * Pointer to the dev_flow.
7085 * pointer to error structure.
7088 * 0 on success otherwise -errno and errno is set.
7091 flow_dv_tag_resource_register
7092 (struct rte_eth_dev *dev,
7094 struct mlx5_flow *dev_flow,
7095 struct rte_flow_error *error)
7097 struct mlx5_priv *priv = dev->data->dev_private;
7098 struct mlx5_ibv_shared *sh = priv->sh;
7099 struct mlx5_flow_dv_tag_resource *cache_resource;
7100 struct mlx5_hlist_entry *entry;
7102 /* Lookup a matching resource from cache. */
7103 entry = mlx5_hlist_lookup(sh->tag_table, (uint64_t)tag_be24);
7105 cache_resource = container_of
7106 (entry, struct mlx5_flow_dv_tag_resource, entry);
7107 rte_atomic32_inc(&cache_resource->refcnt);
7108 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
7109 dev_flow->dv.tag_resource = cache_resource;
7110 DRV_LOG(DEBUG, "cached tag resource %p: refcnt now %d++",
7111 (void *)cache_resource,
7112 rte_atomic32_read(&cache_resource->refcnt));
7115 /* Register new resource. */
7116 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG],
7117 &dev_flow->handle->dvh.rix_tag);
7118 if (!cache_resource)
7119 return rte_flow_error_set(error, ENOMEM,
7120 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7121 "cannot allocate resource memory");
7122 cache_resource->entry.key = (uint64_t)tag_be24;
7123 cache_resource->action = mlx5_glue->dv_create_flow_action_tag(tag_be24);
7124 if (!cache_resource->action) {
7125 rte_free(cache_resource);
7126 return rte_flow_error_set(error, ENOMEM,
7127 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7128 NULL, "cannot create action");
7130 rte_atomic32_init(&cache_resource->refcnt);
7131 rte_atomic32_inc(&cache_resource->refcnt);
7132 if (mlx5_hlist_insert(sh->tag_table, &cache_resource->entry)) {
7133 mlx5_glue->destroy_flow_action(cache_resource->action);
7134 rte_free(cache_resource);
7135 return rte_flow_error_set(error, EEXIST,
7136 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7137 NULL, "cannot insert tag");
7139 dev_flow->dv.tag_resource = cache_resource;
7140 DRV_LOG(DEBUG, "new tag resource %p: refcnt now %d++",
7141 (void *)cache_resource,
7142 rte_atomic32_read(&cache_resource->refcnt));
7150 * Pointer to Ethernet device.
7155 * 1 while a reference on it exists, 0 when freed.
7158 flow_dv_tag_release(struct rte_eth_dev *dev,
7161 struct mlx5_priv *priv = dev->data->dev_private;
7162 struct mlx5_ibv_shared *sh = priv->sh;
7163 struct mlx5_flow_dv_tag_resource *tag;
7165 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7168 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
7169 dev->data->port_id, (void *)tag,
7170 rte_atomic32_read(&tag->refcnt));
7171 if (rte_atomic32_dec_and_test(&tag->refcnt)) {
7172 claim_zero(mlx5_glue->destroy_flow_action(tag->action));
7173 mlx5_hlist_remove(sh->tag_table, &tag->entry);
7174 DRV_LOG(DEBUG, "port %u tag %p: removed",
7175 dev->data->port_id, (void *)tag);
7176 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7183 * Translate port ID action to vport.
7186 * Pointer to rte_eth_dev structure.
7188 * Pointer to the port ID action.
7189 * @param[out] dst_port_id
7190 * The target port ID.
7192 * Pointer to the error structure.
7195 * 0 on success, a negative errno value otherwise and rte_errno is set.
7198 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
7199 const struct rte_flow_action *action,
7200 uint32_t *dst_port_id,
7201 struct rte_flow_error *error)
7204 struct mlx5_priv *priv;
7205 const struct rte_flow_action_port_id *conf =
7206 (const struct rte_flow_action_port_id *)action->conf;
7208 port = conf->original ? dev->data->port_id : conf->id;
7209 priv = mlx5_port_to_eswitch_info(port, false);
7211 return rte_flow_error_set(error, -rte_errno,
7212 RTE_FLOW_ERROR_TYPE_ACTION,
7214 "No eswitch info was found for port");
7215 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
7217 * This parameter is transferred to
7218 * mlx5dv_dr_action_create_dest_ib_port().
7220 *dst_port_id = priv->ibv_port;
7223 * Legacy mode, no LAG configurations is supported.
7224 * This parameter is transferred to
7225 * mlx5dv_dr_action_create_dest_vport().
7227 *dst_port_id = priv->vport_id;
7233 * Add Tx queue matcher
7236 * Pointer to the dev struct.
7237 * @param[in, out] matcher
7239 * @param[in, out] key
7240 * Flow matcher value.
7242 * Flow pattern to translate.
7244 * Item is inner pattern.
7247 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
7248 void *matcher, void *key,
7249 const struct rte_flow_item *item)
7251 const struct mlx5_rte_flow_item_tx_queue *queue_m;
7252 const struct mlx5_rte_flow_item_tx_queue *queue_v;
7254 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7256 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7257 struct mlx5_txq_ctrl *txq;
7261 queue_m = (const void *)item->mask;
7264 queue_v = (const void *)item->spec;
7267 txq = mlx5_txq_get(dev, queue_v->queue);
7270 queue = txq->obj->sq->id;
7271 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
7272 MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
7273 queue & queue_m->queue);
7274 mlx5_txq_release(dev, queue_v->queue);
7278 * Set the hash fields according to the @p flow information.
7280 * @param[in] dev_flow
7281 * Pointer to the mlx5_flow.
7284 flow_dv_hashfields_set(struct mlx5_flow *dev_flow)
7286 struct rte_flow *flow = dev_flow->flow;
7287 uint64_t items = dev_flow->handle->layers;
7289 uint64_t rss_types = rte_eth_rss_hf_refine(flow->rss.types);
7291 dev_flow->hash_fields = 0;
7292 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
7293 if (flow->rss.level >= 2) {
7294 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
7298 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
7299 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
7300 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
7301 if (rss_types & ETH_RSS_L3_SRC_ONLY)
7302 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
7303 else if (rss_types & ETH_RSS_L3_DST_ONLY)
7304 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
7306 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
7308 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
7309 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
7310 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
7311 if (rss_types & ETH_RSS_L3_SRC_ONLY)
7312 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
7313 else if (rss_types & ETH_RSS_L3_DST_ONLY)
7314 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
7316 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
7319 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
7320 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
7321 if (rss_types & ETH_RSS_UDP) {
7322 if (rss_types & ETH_RSS_L4_SRC_ONLY)
7323 dev_flow->hash_fields |=
7324 IBV_RX_HASH_SRC_PORT_UDP;
7325 else if (rss_types & ETH_RSS_L4_DST_ONLY)
7326 dev_flow->hash_fields |=
7327 IBV_RX_HASH_DST_PORT_UDP;
7329 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
7331 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
7332 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
7333 if (rss_types & ETH_RSS_TCP) {
7334 if (rss_types & ETH_RSS_L4_SRC_ONLY)
7335 dev_flow->hash_fields |=
7336 IBV_RX_HASH_SRC_PORT_TCP;
7337 else if (rss_types & ETH_RSS_L4_DST_ONLY)
7338 dev_flow->hash_fields |=
7339 IBV_RX_HASH_DST_PORT_TCP;
7341 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
7347 * Fill the flow with DV spec, lock free
7348 * (mutex should be acquired by caller).
7351 * Pointer to rte_eth_dev structure.
7352 * @param[in, out] dev_flow
7353 * Pointer to the sub flow.
7355 * Pointer to the flow attributes.
7357 * Pointer to the list of items.
7358 * @param[in] actions
7359 * Pointer to the list of actions.
7361 * Pointer to the error structure.
7364 * 0 on success, a negative errno value otherwise and rte_errno is set.
7367 __flow_dv_translate(struct rte_eth_dev *dev,
7368 struct mlx5_flow *dev_flow,
7369 const struct rte_flow_attr *attr,
7370 const struct rte_flow_item items[],
7371 const struct rte_flow_action actions[],
7372 struct rte_flow_error *error)
7374 struct mlx5_priv *priv = dev->data->dev_private;
7375 struct mlx5_dev_config *dev_conf = &priv->config;
7376 struct rte_flow *flow = dev_flow->flow;
7377 struct mlx5_flow_handle *handle = dev_flow->handle;
7378 uint64_t item_flags = 0;
7379 uint64_t last_item = 0;
7380 uint64_t action_flags = 0;
7381 uint64_t priority = attr->priority;
7382 struct mlx5_flow_dv_matcher matcher = {
7384 .size = sizeof(matcher.mask.buf),
7388 bool actions_end = false;
7390 struct mlx5_flow_dv_modify_hdr_resource res;
7391 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
7392 sizeof(struct mlx5_modification_cmd) *
7393 (MLX5_MAX_MODIFY_NUM + 1)];
7395 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
7396 union flow_dv_attr flow_attr = { .attr = 0 };
7398 union mlx5_flow_tbl_key tbl_key;
7399 uint32_t modify_action_position = UINT32_MAX;
7400 void *match_mask = matcher.mask.buf;
7401 void *match_value = dev_flow->dv.value.buf;
7402 uint8_t next_protocol = 0xff;
7403 struct rte_vlan_hdr vlan = { 0 };
7407 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
7408 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
7409 ret = mlx5_flow_group_to_table(attr, dev_flow->external, attr->group,
7410 !!priv->fdb_def_rule, &table, error);
7413 dev_flow->dv.group = table;
7415 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
7416 if (priority == MLX5_FLOW_PRIO_RSVD)
7417 priority = dev_conf->flow_prio - 1;
7418 /* number of actions must be set to 0 in case of dirty stack. */
7419 mhdr_res->actions_num = 0;
7420 for (; !actions_end ; actions++) {
7421 const struct rte_flow_action_queue *queue;
7422 const struct rte_flow_action_rss *rss;
7423 const struct rte_flow_action *action = actions;
7424 const struct rte_flow_action_count *count = action->conf;
7425 const uint8_t *rss_key;
7426 const struct rte_flow_action_jump *jump_data;
7427 const struct rte_flow_action_meter *mtr;
7428 struct mlx5_flow_tbl_resource *tbl;
7429 uint32_t port_id = 0;
7430 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
7431 int action_type = actions->type;
7432 const struct rte_flow_action *found_action = NULL;
7433 struct mlx5_flow_meter *fm = NULL;
7435 switch (action_type) {
7436 case RTE_FLOW_ACTION_TYPE_VOID:
7438 case RTE_FLOW_ACTION_TYPE_PORT_ID:
7439 if (flow_dv_translate_action_port_id(dev, action,
7442 memset(&port_id_resource, 0, sizeof(port_id_resource));
7443 port_id_resource.port_id = port_id;
7444 if (flow_dv_port_id_action_resource_register
7445 (dev, &port_id_resource, dev_flow, error))
7447 MLX5_ASSERT(!handle->rix_port_id_action);
7448 dev_flow->dv.actions[actions_n++] =
7449 dev_flow->dv.port_id_action->action;
7450 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
7451 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
7453 case RTE_FLOW_ACTION_TYPE_FLAG:
7454 action_flags |= MLX5_FLOW_ACTION_FLAG;
7455 dev_flow->handle->mark = 1;
7456 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7457 struct rte_flow_action_mark mark = {
7458 .id = MLX5_FLOW_MARK_DEFAULT,
7461 if (flow_dv_convert_action_mark(dev, &mark,
7465 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
7468 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
7470 * Only one FLAG or MARK is supported per device flow
7471 * right now. So the pointer to the tag resource must be
7472 * zero before the register process.
7474 MLX5_ASSERT(!handle->dvh.rix_tag);
7475 if (flow_dv_tag_resource_register(dev, tag_be,
7478 MLX5_ASSERT(dev_flow->dv.tag_resource);
7479 dev_flow->dv.actions[actions_n++] =
7480 dev_flow->dv.tag_resource->action;
7482 case RTE_FLOW_ACTION_TYPE_MARK:
7483 action_flags |= MLX5_FLOW_ACTION_MARK;
7484 dev_flow->handle->mark = 1;
7485 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7486 const struct rte_flow_action_mark *mark =
7487 (const struct rte_flow_action_mark *)
7490 if (flow_dv_convert_action_mark(dev, mark,
7494 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
7498 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
7499 /* Legacy (non-extensive) MARK action. */
7500 tag_be = mlx5_flow_mark_set
7501 (((const struct rte_flow_action_mark *)
7502 (actions->conf))->id);
7503 MLX5_ASSERT(!handle->dvh.rix_tag);
7504 if (flow_dv_tag_resource_register(dev, tag_be,
7507 MLX5_ASSERT(dev_flow->dv.tag_resource);
7508 dev_flow->dv.actions[actions_n++] =
7509 dev_flow->dv.tag_resource->action;
7511 case RTE_FLOW_ACTION_TYPE_SET_META:
7512 if (flow_dv_convert_action_set_meta
7513 (dev, mhdr_res, attr,
7514 (const struct rte_flow_action_set_meta *)
7515 actions->conf, error))
7517 action_flags |= MLX5_FLOW_ACTION_SET_META;
7519 case RTE_FLOW_ACTION_TYPE_SET_TAG:
7520 if (flow_dv_convert_action_set_tag
7522 (const struct rte_flow_action_set_tag *)
7523 actions->conf, error))
7525 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7527 case RTE_FLOW_ACTION_TYPE_DROP:
7528 action_flags |= MLX5_FLOW_ACTION_DROP;
7529 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
7531 case RTE_FLOW_ACTION_TYPE_QUEUE:
7532 MLX5_ASSERT(flow->rss.queue);
7533 queue = actions->conf;
7534 flow->rss.queue_num = 1;
7535 (*flow->rss.queue)[0] = queue->index;
7536 action_flags |= MLX5_FLOW_ACTION_QUEUE;
7537 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
7539 case RTE_FLOW_ACTION_TYPE_RSS:
7540 MLX5_ASSERT(flow->rss.queue);
7541 rss = actions->conf;
7542 if (flow->rss.queue)
7543 memcpy((*flow->rss.queue), rss->queue,
7544 rss->queue_num * sizeof(uint16_t));
7545 flow->rss.queue_num = rss->queue_num;
7546 /* NULL RSS key indicates default RSS key. */
7547 rss_key = !rss->key ? rss_hash_default_key : rss->key;
7548 memcpy(flow->rss.key, rss_key, MLX5_RSS_HASH_KEY_LEN);
7550 * rss->level and rss.types should be set in advance
7551 * when expanding items for RSS.
7553 action_flags |= MLX5_FLOW_ACTION_RSS;
7554 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
7556 case RTE_FLOW_ACTION_TYPE_COUNT:
7557 if (!dev_conf->devx) {
7558 rte_errno = ENOTSUP;
7561 flow->counter = flow_dv_counter_alloc(dev,
7564 dev_flow->dv.group);
7567 dev_flow->dv.actions[actions_n++] =
7568 (flow_dv_counter_get_by_idx(dev,
7569 flow->counter, NULL))->action;
7570 action_flags |= MLX5_FLOW_ACTION_COUNT;
7573 if (rte_errno == ENOTSUP)
7574 return rte_flow_error_set
7576 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7578 "count action not supported");
7580 return rte_flow_error_set
7582 RTE_FLOW_ERROR_TYPE_ACTION,
7584 "cannot create counter"
7587 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
7588 dev_flow->dv.actions[actions_n++] =
7589 priv->sh->pop_vlan_action;
7590 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
7592 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
7593 if (!(action_flags &
7594 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
7595 flow_dev_get_vlan_info_from_items(items, &vlan);
7596 vlan.eth_proto = rte_be_to_cpu_16
7597 ((((const struct rte_flow_action_of_push_vlan *)
7598 actions->conf)->ethertype));
7599 found_action = mlx5_flow_find_action
7601 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
7603 mlx5_update_vlan_vid_pcp(found_action, &vlan);
7604 found_action = mlx5_flow_find_action
7606 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
7608 mlx5_update_vlan_vid_pcp(found_action, &vlan);
7609 if (flow_dv_create_action_push_vlan
7610 (dev, attr, &vlan, dev_flow, error))
7612 dev_flow->dv.actions[actions_n++] =
7613 dev_flow->dv.push_vlan_res->action;
7614 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
7616 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
7617 /* of_vlan_push action handled this action */
7618 MLX5_ASSERT(action_flags &
7619 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
7621 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
7622 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
7624 flow_dev_get_vlan_info_from_items(items, &vlan);
7625 mlx5_update_vlan_vid_pcp(actions, &vlan);
7626 /* If no VLAN push - this is a modify header action */
7627 if (flow_dv_convert_action_modify_vlan_vid
7628 (mhdr_res, actions, error))
7630 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
7632 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
7633 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
7634 if (flow_dv_create_action_l2_encap(dev, actions,
7639 dev_flow->dv.actions[actions_n++] =
7640 dev_flow->dv.encap_decap->verbs_action;
7641 action_flags |= MLX5_FLOW_ACTION_ENCAP;
7643 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
7644 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
7645 if (flow_dv_create_action_l2_decap(dev, dev_flow,
7649 dev_flow->dv.actions[actions_n++] =
7650 dev_flow->dv.encap_decap->verbs_action;
7651 action_flags |= MLX5_FLOW_ACTION_DECAP;
7653 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
7654 /* Handle encap with preceding decap. */
7655 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
7656 if (flow_dv_create_action_raw_encap
7657 (dev, actions, dev_flow, attr, error))
7659 dev_flow->dv.actions[actions_n++] =
7660 dev_flow->dv.encap_decap->verbs_action;
7662 /* Handle encap without preceding decap. */
7663 if (flow_dv_create_action_l2_encap
7664 (dev, actions, dev_flow, attr->transfer,
7667 dev_flow->dv.actions[actions_n++] =
7668 dev_flow->dv.encap_decap->verbs_action;
7670 action_flags |= MLX5_FLOW_ACTION_ENCAP;
7672 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
7673 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
7675 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
7676 if (flow_dv_create_action_l2_decap
7677 (dev, dev_flow, attr->transfer, error))
7679 dev_flow->dv.actions[actions_n++] =
7680 dev_flow->dv.encap_decap->verbs_action;
7682 /* If decap is followed by encap, handle it at encap. */
7683 action_flags |= MLX5_FLOW_ACTION_DECAP;
7685 case RTE_FLOW_ACTION_TYPE_JUMP:
7686 jump_data = action->conf;
7687 ret = mlx5_flow_group_to_table(attr, dev_flow->external,
7689 !!priv->fdb_def_rule,
7693 tbl = flow_dv_tbl_resource_get(dev, table,
7695 attr->transfer, error);
7697 return rte_flow_error_set
7699 RTE_FLOW_ERROR_TYPE_ACTION,
7701 "cannot create jump action.");
7702 if (flow_dv_jump_tbl_resource_register
7703 (dev, tbl, dev_flow, error)) {
7704 flow_dv_tbl_resource_release(dev, tbl);
7705 return rte_flow_error_set
7707 RTE_FLOW_ERROR_TYPE_ACTION,
7709 "cannot create jump action.");
7711 dev_flow->dv.actions[actions_n++] =
7712 dev_flow->dv.jump->action;
7713 action_flags |= MLX5_FLOW_ACTION_JUMP;
7714 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
7716 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
7717 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
7718 if (flow_dv_convert_action_modify_mac
7719 (mhdr_res, actions, error))
7721 action_flags |= actions->type ==
7722 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
7723 MLX5_FLOW_ACTION_SET_MAC_SRC :
7724 MLX5_FLOW_ACTION_SET_MAC_DST;
7726 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
7727 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
7728 if (flow_dv_convert_action_modify_ipv4
7729 (mhdr_res, actions, error))
7731 action_flags |= actions->type ==
7732 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
7733 MLX5_FLOW_ACTION_SET_IPV4_SRC :
7734 MLX5_FLOW_ACTION_SET_IPV4_DST;
7736 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
7737 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
7738 if (flow_dv_convert_action_modify_ipv6
7739 (mhdr_res, actions, error))
7741 action_flags |= actions->type ==
7742 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
7743 MLX5_FLOW_ACTION_SET_IPV6_SRC :
7744 MLX5_FLOW_ACTION_SET_IPV6_DST;
7746 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
7747 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
7748 if (flow_dv_convert_action_modify_tp
7749 (mhdr_res, actions, items,
7750 &flow_attr, dev_flow, !!(action_flags &
7751 MLX5_FLOW_ACTION_DECAP), error))
7753 action_flags |= actions->type ==
7754 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
7755 MLX5_FLOW_ACTION_SET_TP_SRC :
7756 MLX5_FLOW_ACTION_SET_TP_DST;
7758 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
7759 if (flow_dv_convert_action_modify_dec_ttl
7760 (mhdr_res, items, &flow_attr, dev_flow,
7762 MLX5_FLOW_ACTION_DECAP), error))
7764 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
7766 case RTE_FLOW_ACTION_TYPE_SET_TTL:
7767 if (flow_dv_convert_action_modify_ttl
7768 (mhdr_res, actions, items, &flow_attr,
7769 dev_flow, !!(action_flags &
7770 MLX5_FLOW_ACTION_DECAP), error))
7772 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
7774 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
7775 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
7776 if (flow_dv_convert_action_modify_tcp_seq
7777 (mhdr_res, actions, error))
7779 action_flags |= actions->type ==
7780 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
7781 MLX5_FLOW_ACTION_INC_TCP_SEQ :
7782 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
7785 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
7786 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
7787 if (flow_dv_convert_action_modify_tcp_ack
7788 (mhdr_res, actions, error))
7790 action_flags |= actions->type ==
7791 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
7792 MLX5_FLOW_ACTION_INC_TCP_ACK :
7793 MLX5_FLOW_ACTION_DEC_TCP_ACK;
7795 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
7796 if (flow_dv_convert_action_set_reg
7797 (mhdr_res, actions, error))
7799 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7801 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
7802 if (flow_dv_convert_action_copy_mreg
7803 (dev, mhdr_res, actions, error))
7805 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7807 case RTE_FLOW_ACTION_TYPE_METER:
7808 mtr = actions->conf;
7810 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
7813 return rte_flow_error_set(error,
7815 RTE_FLOW_ERROR_TYPE_ACTION,
7818 "or invalid parameters");
7819 flow->meter = fm->meter_id;
7821 /* Set the meter action. */
7823 fm = mlx5_flow_meter_find(priv, flow->meter);
7825 return rte_flow_error_set(error,
7827 RTE_FLOW_ERROR_TYPE_ACTION,
7830 "or invalid parameters");
7832 dev_flow->dv.actions[actions_n++] =
7833 fm->mfts->meter_action;
7834 action_flags |= MLX5_FLOW_ACTION_METER;
7836 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
7837 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
7840 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
7842 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
7843 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
7846 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
7848 case RTE_FLOW_ACTION_TYPE_END:
7850 if (mhdr_res->actions_num) {
7851 /* create modify action if needed. */
7852 if (flow_dv_modify_hdr_resource_register
7853 (dev, mhdr_res, dev_flow, error))
7855 dev_flow->dv.actions[modify_action_position] =
7856 handle->dvh.modify_hdr->verbs_action;
7862 if (mhdr_res->actions_num &&
7863 modify_action_position == UINT32_MAX)
7864 modify_action_position = actions_n++;
7866 dev_flow->dv.actions_n = actions_n;
7867 dev_flow->act_flags = action_flags;
7868 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
7869 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
7870 int item_type = items->type;
7872 switch (item_type) {
7873 case RTE_FLOW_ITEM_TYPE_PORT_ID:
7874 flow_dv_translate_item_port_id(dev, match_mask,
7875 match_value, items);
7876 last_item = MLX5_FLOW_ITEM_PORT_ID;
7878 case RTE_FLOW_ITEM_TYPE_ETH:
7879 flow_dv_translate_item_eth(match_mask, match_value,
7881 matcher.priority = MLX5_PRIORITY_MAP_L2;
7882 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
7883 MLX5_FLOW_LAYER_OUTER_L2;
7885 case RTE_FLOW_ITEM_TYPE_VLAN:
7886 flow_dv_translate_item_vlan(dev_flow,
7887 match_mask, match_value,
7889 matcher.priority = MLX5_PRIORITY_MAP_L2;
7890 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
7891 MLX5_FLOW_LAYER_INNER_VLAN) :
7892 (MLX5_FLOW_LAYER_OUTER_L2 |
7893 MLX5_FLOW_LAYER_OUTER_VLAN);
7895 case RTE_FLOW_ITEM_TYPE_IPV4:
7896 mlx5_flow_tunnel_ip_check(items, next_protocol,
7897 &item_flags, &tunnel);
7898 flow_dv_translate_item_ipv4(match_mask, match_value,
7899 items, item_flags, tunnel,
7900 dev_flow->dv.group);
7901 matcher.priority = MLX5_PRIORITY_MAP_L3;
7902 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
7903 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
7904 if (items->mask != NULL &&
7905 ((const struct rte_flow_item_ipv4 *)
7906 items->mask)->hdr.next_proto_id) {
7908 ((const struct rte_flow_item_ipv4 *)
7909 (items->spec))->hdr.next_proto_id;
7911 ((const struct rte_flow_item_ipv4 *)
7912 (items->mask))->hdr.next_proto_id;
7914 /* Reset for inner layer. */
7915 next_protocol = 0xff;
7918 case RTE_FLOW_ITEM_TYPE_IPV6:
7919 mlx5_flow_tunnel_ip_check(items, next_protocol,
7920 &item_flags, &tunnel);
7921 flow_dv_translate_item_ipv6(match_mask, match_value,
7922 items, item_flags, tunnel,
7923 dev_flow->dv.group);
7924 matcher.priority = MLX5_PRIORITY_MAP_L3;
7925 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
7926 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
7927 if (items->mask != NULL &&
7928 ((const struct rte_flow_item_ipv6 *)
7929 items->mask)->hdr.proto) {
7931 ((const struct rte_flow_item_ipv6 *)
7932 items->spec)->hdr.proto;
7934 ((const struct rte_flow_item_ipv6 *)
7935 items->mask)->hdr.proto;
7937 /* Reset for inner layer. */
7938 next_protocol = 0xff;
7941 case RTE_FLOW_ITEM_TYPE_TCP:
7942 flow_dv_translate_item_tcp(match_mask, match_value,
7944 matcher.priority = MLX5_PRIORITY_MAP_L4;
7945 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
7946 MLX5_FLOW_LAYER_OUTER_L4_TCP;
7948 case RTE_FLOW_ITEM_TYPE_UDP:
7949 flow_dv_translate_item_udp(match_mask, match_value,
7951 matcher.priority = MLX5_PRIORITY_MAP_L4;
7952 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
7953 MLX5_FLOW_LAYER_OUTER_L4_UDP;
7955 case RTE_FLOW_ITEM_TYPE_GRE:
7956 flow_dv_translate_item_gre(match_mask, match_value,
7958 matcher.priority = flow->rss.level >= 2 ?
7959 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
7960 last_item = MLX5_FLOW_LAYER_GRE;
7962 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
7963 flow_dv_translate_item_gre_key(match_mask,
7964 match_value, items);
7965 last_item = MLX5_FLOW_LAYER_GRE_KEY;
7967 case RTE_FLOW_ITEM_TYPE_NVGRE:
7968 flow_dv_translate_item_nvgre(match_mask, match_value,
7970 matcher.priority = flow->rss.level >= 2 ?
7971 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
7972 last_item = MLX5_FLOW_LAYER_GRE;
7974 case RTE_FLOW_ITEM_TYPE_VXLAN:
7975 flow_dv_translate_item_vxlan(match_mask, match_value,
7977 matcher.priority = flow->rss.level >= 2 ?
7978 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
7979 last_item = MLX5_FLOW_LAYER_VXLAN;
7981 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
7982 flow_dv_translate_item_vxlan_gpe(match_mask,
7985 matcher.priority = flow->rss.level >= 2 ?
7986 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
7987 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
7989 case RTE_FLOW_ITEM_TYPE_GENEVE:
7990 flow_dv_translate_item_geneve(match_mask, match_value,
7992 matcher.priority = flow->rss.level >= 2 ?
7993 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
7994 last_item = MLX5_FLOW_LAYER_GENEVE;
7996 case RTE_FLOW_ITEM_TYPE_MPLS:
7997 flow_dv_translate_item_mpls(match_mask, match_value,
7998 items, last_item, tunnel);
7999 matcher.priority = flow->rss.level >= 2 ?
8000 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8001 last_item = MLX5_FLOW_LAYER_MPLS;
8003 case RTE_FLOW_ITEM_TYPE_MARK:
8004 flow_dv_translate_item_mark(dev, match_mask,
8005 match_value, items);
8006 last_item = MLX5_FLOW_ITEM_MARK;
8008 case RTE_FLOW_ITEM_TYPE_META:
8009 flow_dv_translate_item_meta(dev, match_mask,
8010 match_value, attr, items);
8011 last_item = MLX5_FLOW_ITEM_METADATA;
8013 case RTE_FLOW_ITEM_TYPE_ICMP:
8014 flow_dv_translate_item_icmp(match_mask, match_value,
8016 last_item = MLX5_FLOW_LAYER_ICMP;
8018 case RTE_FLOW_ITEM_TYPE_ICMP6:
8019 flow_dv_translate_item_icmp6(match_mask, match_value,
8021 last_item = MLX5_FLOW_LAYER_ICMP6;
8023 case RTE_FLOW_ITEM_TYPE_TAG:
8024 flow_dv_translate_item_tag(dev, match_mask,
8025 match_value, items);
8026 last_item = MLX5_FLOW_ITEM_TAG;
8028 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
8029 flow_dv_translate_mlx5_item_tag(dev, match_mask,
8030 match_value, items);
8031 last_item = MLX5_FLOW_ITEM_TAG;
8033 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
8034 flow_dv_translate_item_tx_queue(dev, match_mask,
8037 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
8039 case RTE_FLOW_ITEM_TYPE_GTP:
8040 flow_dv_translate_item_gtp(match_mask, match_value,
8042 matcher.priority = flow->rss.level >= 2 ?
8043 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8044 last_item = MLX5_FLOW_LAYER_GTP;
8049 item_flags |= last_item;
8052 * When E-Switch mode is enabled, we have two cases where we need to
8053 * set the source port manually.
8054 * The first one, is in case of Nic steering rule, and the second is
8055 * E-Switch rule where no port_id item was found. In both cases
8056 * the source port is set according the current port in use.
8058 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
8059 (priv->representor || priv->master)) {
8060 if (flow_dv_translate_item_port_id(dev, match_mask,
8064 #ifdef RTE_LIBRTE_MLX5_DEBUG
8065 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
8066 dev_flow->dv.value.buf));
8069 * Layers may be already initialized from prefix flow if this dev_flow
8070 * is the suffix flow.
8072 handle->layers |= item_flags;
8073 if (action_flags & MLX5_FLOW_ACTION_RSS)
8074 flow_dv_hashfields_set(dev_flow);
8075 /* Register matcher. */
8076 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
8078 matcher.priority = mlx5_flow_adjust_priority(dev, priority,
8080 /* reserved field no needs to be set to 0 here. */
8081 tbl_key.domain = attr->transfer;
8082 tbl_key.direction = attr->egress;
8083 tbl_key.table_id = dev_flow->dv.group;
8084 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
8090 * Apply the flow to the NIC, lock free,
8091 * (mutex should be acquired by caller).
8094 * Pointer to the Ethernet device structure.
8095 * @param[in, out] flow
8096 * Pointer to flow structure.
8098 * Pointer to error structure.
8101 * 0 on success, a negative errno value otherwise and rte_errno is set.
8104 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
8105 struct rte_flow_error *error)
8107 struct mlx5_flow_dv_workspace *dv;
8108 struct mlx5_flow_handle *dh;
8109 struct mlx5_flow_handle_dv *dv_h;
8110 struct mlx5_flow *dev_flow;
8111 struct mlx5_priv *priv = dev->data->dev_private;
8112 uint32_t handle_idx;
8117 for (idx = priv->flow_idx - 1; idx >= priv->flow_nested_idx; idx--) {
8118 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[idx];
8120 dh = dev_flow->handle;
8123 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8125 dv->actions[n++] = priv->sh->esw_drop_action;
8127 struct mlx5_hrxq *drop_hrxq;
8128 drop_hrxq = mlx5_hrxq_drop_new(dev);
8132 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8134 "cannot get drop hash queue");
8138 * Drop queues will be released by the specify
8139 * mlx5_hrxq_drop_release() function. Assign
8140 * the special index to hrxq to mark the queue
8141 * has been allocated.
8143 dh->rix_hrxq = UINT32_MAX;
8144 dv->actions[n++] = drop_hrxq->action;
8146 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8147 struct mlx5_hrxq *hrxq;
8150 MLX5_ASSERT(flow->rss.queue);
8151 hrxq_idx = mlx5_hrxq_get(dev, flow->rss.key,
8152 MLX5_RSS_HASH_KEY_LEN,
8153 dev_flow->hash_fields,
8155 flow->rss.queue_num);
8157 hrxq_idx = mlx5_hrxq_new
8158 (dev, flow->rss.key,
8159 MLX5_RSS_HASH_KEY_LEN,
8160 dev_flow->hash_fields,
8162 flow->rss.queue_num,
8164 MLX5_FLOW_LAYER_TUNNEL));
8166 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8171 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8172 "cannot get hash queue");
8175 dh->rix_hrxq = hrxq_idx;
8176 dv->actions[n++] = hrxq->action;
8179 mlx5_glue->dv_create_flow(dv_h->matcher->matcher_object,
8180 (void *)&dv->value, n,
8183 rte_flow_error_set(error, errno,
8184 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8186 "hardware refuses to create flow");
8189 if (priv->vmwa_context &&
8190 dh->vf_vlan.tag && !dh->vf_vlan.created) {
8192 * The rule contains the VLAN pattern.
8193 * For VF we are going to create VLAN
8194 * interface to make hypervisor set correct
8195 * e-Switch vport context.
8197 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
8202 err = rte_errno; /* Save rte_errno before cleanup. */
8203 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
8204 handle_idx, dh, next) {
8205 /* hrxq is union, don't clear it if the flag is not set. */
8207 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8208 mlx5_hrxq_drop_release(dev);
8210 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8211 mlx5_hrxq_release(dev, dh->rix_hrxq);
8215 if (dh->vf_vlan.tag && dh->vf_vlan.created)
8216 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8218 rte_errno = err; /* Restore rte_errno. */
8223 * Release the flow matcher.
8226 * Pointer to Ethernet device.
8228 * Pointer to mlx5_flow_handle.
8231 * 1 while a reference on it exists, 0 when freed.
8234 flow_dv_matcher_release(struct rte_eth_dev *dev,
8235 struct mlx5_flow_handle *handle)
8237 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
8239 MLX5_ASSERT(matcher->matcher_object);
8240 DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
8241 dev->data->port_id, (void *)matcher,
8242 rte_atomic32_read(&matcher->refcnt));
8243 if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
8244 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8245 (matcher->matcher_object));
8246 LIST_REMOVE(matcher, next);
8247 /* table ref-- in release interface. */
8248 flow_dv_tbl_resource_release(dev, matcher->tbl);
8250 DRV_LOG(DEBUG, "port %u matcher %p: removed",
8251 dev->data->port_id, (void *)matcher);
8258 * Release an encap/decap resource.
8261 * Pointer to Ethernet device.
8263 * Pointer to mlx5_flow_handle.
8266 * 1 while a reference on it exists, 0 when freed.
8269 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
8270 struct mlx5_flow_handle *handle)
8272 struct mlx5_priv *priv = dev->data->dev_private;
8273 uint32_t idx = handle->dvh.rix_encap_decap;
8274 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
8276 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
8278 if (!cache_resource)
8280 MLX5_ASSERT(cache_resource->verbs_action);
8281 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
8282 (void *)cache_resource,
8283 rte_atomic32_read(&cache_resource->refcnt));
8284 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8285 claim_zero(mlx5_glue->destroy_flow_action
8286 (cache_resource->verbs_action));
8287 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
8288 &priv->sh->encaps_decaps, idx,
8289 cache_resource, next);
8290 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
8291 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
8292 (void *)cache_resource);
8299 * Release an jump to table action resource.
8302 * Pointer to Ethernet device.
8304 * Pointer to mlx5_flow_handle.
8307 * 1 while a reference on it exists, 0 when freed.
8310 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
8311 struct mlx5_flow_handle *handle)
8313 struct mlx5_priv *priv = dev->data->dev_private;
8314 struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
8315 struct mlx5_flow_tbl_data_entry *tbl_data;
8317 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
8321 cache_resource = &tbl_data->jump;
8322 MLX5_ASSERT(cache_resource->action);
8323 DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
8324 (void *)cache_resource,
8325 rte_atomic32_read(&cache_resource->refcnt));
8326 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8327 claim_zero(mlx5_glue->destroy_flow_action
8328 (cache_resource->action));
8329 /* jump action memory free is inside the table release. */
8330 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
8331 DRV_LOG(DEBUG, "jump table resource %p: removed",
8332 (void *)cache_resource);
8339 * Release a modify-header resource.
8342 * Pointer to mlx5_flow_handle.
8345 * 1 while a reference on it exists, 0 when freed.
8348 flow_dv_modify_hdr_resource_release(struct mlx5_flow_handle *handle)
8350 struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
8351 handle->dvh.modify_hdr;
8353 MLX5_ASSERT(cache_resource->verbs_action);
8354 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
8355 (void *)cache_resource,
8356 rte_atomic32_read(&cache_resource->refcnt));
8357 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8358 claim_zero(mlx5_glue->destroy_flow_action
8359 (cache_resource->verbs_action));
8360 LIST_REMOVE(cache_resource, next);
8361 rte_free(cache_resource);
8362 DRV_LOG(DEBUG, "modify-header resource %p: removed",
8363 (void *)cache_resource);
8370 * Release port ID action resource.
8373 * Pointer to Ethernet device.
8375 * Pointer to mlx5_flow_handle.
8378 * 1 while a reference on it exists, 0 when freed.
8381 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
8382 struct mlx5_flow_handle *handle)
8384 struct mlx5_priv *priv = dev->data->dev_private;
8385 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
8386 uint32_t idx = handle->rix_port_id_action;
8388 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
8390 if (!cache_resource)
8392 MLX5_ASSERT(cache_resource->action);
8393 DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
8394 (void *)cache_resource,
8395 rte_atomic32_read(&cache_resource->refcnt));
8396 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8397 claim_zero(mlx5_glue->destroy_flow_action
8398 (cache_resource->action));
8399 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
8400 &priv->sh->port_id_action_list, idx,
8401 cache_resource, next);
8402 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
8403 DRV_LOG(DEBUG, "port id action resource %p: removed",
8404 (void *)cache_resource);
8411 * Release push vlan action resource.
8414 * Pointer to Ethernet device.
8416 * Pointer to mlx5_flow_handle.
8419 * 1 while a reference on it exists, 0 when freed.
8422 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
8423 struct mlx5_flow_handle *handle)
8425 struct mlx5_priv *priv = dev->data->dev_private;
8426 uint32_t idx = handle->dvh.rix_push_vlan;
8427 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
8429 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
8431 if (!cache_resource)
8433 MLX5_ASSERT(cache_resource->action);
8434 DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
8435 (void *)cache_resource,
8436 rte_atomic32_read(&cache_resource->refcnt));
8437 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8438 claim_zero(mlx5_glue->destroy_flow_action
8439 (cache_resource->action));
8440 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
8441 &priv->sh->push_vlan_action_list, idx,
8442 cache_resource, next);
8443 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
8444 DRV_LOG(DEBUG, "push vlan action resource %p: removed",
8445 (void *)cache_resource);
8452 * Release the fate resource.
8455 * Pointer to Ethernet device.
8457 * Pointer to mlx5_flow_handle.
8460 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
8461 struct mlx5_flow_handle *handle)
8463 if (!handle->rix_fate)
8465 if (handle->fate_action == MLX5_FLOW_FATE_DROP)
8466 mlx5_hrxq_drop_release(dev);
8467 else if (handle->fate_action == MLX5_FLOW_FATE_QUEUE)
8468 mlx5_hrxq_release(dev, handle->rix_hrxq);
8469 else if (handle->fate_action == MLX5_FLOW_FATE_JUMP)
8470 flow_dv_jump_tbl_resource_release(dev, handle);
8471 else if (handle->fate_action == MLX5_FLOW_FATE_PORT_ID)
8472 flow_dv_port_id_action_resource_release(dev, handle);
8474 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
8475 handle->rix_fate = 0;
8479 * Remove the flow from the NIC but keeps it in memory.
8480 * Lock free, (mutex should be acquired by caller).
8483 * Pointer to Ethernet device.
8484 * @param[in, out] flow
8485 * Pointer to flow structure.
8488 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
8490 struct mlx5_flow_handle *dh;
8491 uint32_t handle_idx;
8492 struct mlx5_priv *priv = dev->data->dev_private;
8496 handle_idx = flow->dev_handles;
8497 while (handle_idx) {
8498 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8503 claim_zero(mlx5_glue->dv_destroy_flow(dh->ib_flow));
8506 if (dh->fate_action == MLX5_FLOW_FATE_DROP ||
8507 dh->fate_action == MLX5_FLOW_FATE_QUEUE)
8508 flow_dv_fate_resource_release(dev, dh);
8509 if (dh->vf_vlan.tag && dh->vf_vlan.created)
8510 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8511 handle_idx = dh->next.next;
8516 * Remove the flow from the NIC and the memory.
8517 * Lock free, (mutex should be acquired by caller).
8520 * Pointer to the Ethernet device structure.
8521 * @param[in, out] flow
8522 * Pointer to flow structure.
8525 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
8527 struct mlx5_flow_handle *dev_handle;
8528 struct mlx5_priv *priv = dev->data->dev_private;
8532 __flow_dv_remove(dev, flow);
8533 if (flow->counter) {
8534 flow_dv_counter_release(dev, flow->counter);
8538 struct mlx5_flow_meter *fm;
8540 fm = mlx5_flow_meter_find(priv, flow->meter);
8542 mlx5_flow_meter_detach(fm);
8545 while (flow->dev_handles) {
8546 uint32_t tmp_idx = flow->dev_handles;
8548 dev_handle = mlx5_ipool_get(priv->sh->ipool
8549 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
8552 flow->dev_handles = dev_handle->next.next;
8553 if (dev_handle->dvh.matcher)
8554 flow_dv_matcher_release(dev, dev_handle);
8555 if (dev_handle->dvh.rix_encap_decap)
8556 flow_dv_encap_decap_resource_release(dev, dev_handle);
8557 if (dev_handle->dvh.modify_hdr)
8558 flow_dv_modify_hdr_resource_release(dev_handle);
8559 if (dev_handle->dvh.rix_push_vlan)
8560 flow_dv_push_vlan_action_resource_release(dev,
8562 if (dev_handle->dvh.rix_tag)
8563 flow_dv_tag_release(dev,
8564 dev_handle->dvh.rix_tag);
8565 flow_dv_fate_resource_release(dev, dev_handle);
8566 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8572 * Query a dv flow rule for its statistics via devx.
8575 * Pointer to Ethernet device.
8577 * Pointer to the sub flow.
8579 * data retrieved by the query.
8581 * Perform verbose error reporting if not NULL.
8584 * 0 on success, a negative errno value otherwise and rte_errno is set.
8587 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
8588 void *data, struct rte_flow_error *error)
8590 struct mlx5_priv *priv = dev->data->dev_private;
8591 struct rte_flow_query_count *qc = data;
8593 if (!priv->config.devx)
8594 return rte_flow_error_set(error, ENOTSUP,
8595 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8597 "counters are not supported");
8598 if (flow->counter) {
8599 uint64_t pkts, bytes;
8600 struct mlx5_flow_counter *cnt;
8602 cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
8604 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
8608 return rte_flow_error_set(error, -err,
8609 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8610 NULL, "cannot read counters");
8613 qc->hits = pkts - cnt->hits;
8614 qc->bytes = bytes - cnt->bytes;
8621 return rte_flow_error_set(error, EINVAL,
8622 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8624 "counters are not available");
8630 * @see rte_flow_query()
8634 flow_dv_query(struct rte_eth_dev *dev,
8635 struct rte_flow *flow __rte_unused,
8636 const struct rte_flow_action *actions __rte_unused,
8637 void *data __rte_unused,
8638 struct rte_flow_error *error __rte_unused)
8642 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
8643 switch (actions->type) {
8644 case RTE_FLOW_ACTION_TYPE_VOID:
8646 case RTE_FLOW_ACTION_TYPE_COUNT:
8647 ret = flow_dv_query_count(dev, flow, data, error);
8650 return rte_flow_error_set(error, ENOTSUP,
8651 RTE_FLOW_ERROR_TYPE_ACTION,
8653 "action not supported");
8660 * Destroy the meter table set.
8661 * Lock free, (mutex should be acquired by caller).
8664 * Pointer to Ethernet device.
8666 * Pointer to the meter table set.
8672 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
8673 struct mlx5_meter_domains_infos *tbl)
8675 struct mlx5_priv *priv = dev->data->dev_private;
8676 struct mlx5_meter_domains_infos *mtd =
8677 (struct mlx5_meter_domains_infos *)tbl;
8679 if (!mtd || !priv->config.dv_flow_en)
8681 if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
8682 claim_zero(mlx5_glue->dv_destroy_flow
8683 (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
8684 if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
8685 claim_zero(mlx5_glue->dv_destroy_flow
8686 (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
8687 if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
8688 claim_zero(mlx5_glue->dv_destroy_flow
8689 (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
8690 if (mtd->egress.color_matcher)
8691 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8692 (mtd->egress.color_matcher));
8693 if (mtd->egress.any_matcher)
8694 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8695 (mtd->egress.any_matcher));
8696 if (mtd->egress.tbl)
8697 claim_zero(flow_dv_tbl_resource_release(dev,
8699 if (mtd->egress.sfx_tbl)
8700 claim_zero(flow_dv_tbl_resource_release(dev,
8701 mtd->egress.sfx_tbl));
8702 if (mtd->ingress.color_matcher)
8703 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8704 (mtd->ingress.color_matcher));
8705 if (mtd->ingress.any_matcher)
8706 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8707 (mtd->ingress.any_matcher));
8708 if (mtd->ingress.tbl)
8709 claim_zero(flow_dv_tbl_resource_release(dev,
8711 if (mtd->ingress.sfx_tbl)
8712 claim_zero(flow_dv_tbl_resource_release(dev,
8713 mtd->ingress.sfx_tbl));
8714 if (mtd->transfer.color_matcher)
8715 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8716 (mtd->transfer.color_matcher));
8717 if (mtd->transfer.any_matcher)
8718 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8719 (mtd->transfer.any_matcher));
8720 if (mtd->transfer.tbl)
8721 claim_zero(flow_dv_tbl_resource_release(dev,
8722 mtd->transfer.tbl));
8723 if (mtd->transfer.sfx_tbl)
8724 claim_zero(flow_dv_tbl_resource_release(dev,
8725 mtd->transfer.sfx_tbl));
8727 claim_zero(mlx5_glue->destroy_flow_action(mtd->drop_actn));
8732 /* Number of meter flow actions, count and jump or count and drop. */
8733 #define METER_ACTIONS 2
8736 * Create specify domain meter table and suffix table.
8739 * Pointer to Ethernet device.
8740 * @param[in,out] mtb
8741 * Pointer to DV meter table set.
8744 * @param[in] transfer
8746 * @param[in] color_reg_c_idx
8747 * Reg C index for color match.
8750 * 0 on success, -1 otherwise and rte_errno is set.
8753 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
8754 struct mlx5_meter_domains_infos *mtb,
8755 uint8_t egress, uint8_t transfer,
8756 uint32_t color_reg_c_idx)
8758 struct mlx5_priv *priv = dev->data->dev_private;
8759 struct mlx5_ibv_shared *sh = priv->sh;
8760 struct mlx5_flow_dv_match_params mask = {
8761 .size = sizeof(mask.buf),
8763 struct mlx5_flow_dv_match_params value = {
8764 .size = sizeof(value.buf),
8766 struct mlx5dv_flow_matcher_attr dv_attr = {
8767 .type = IBV_FLOW_ATTR_NORMAL,
8769 .match_criteria_enable = 0,
8770 .match_mask = (void *)&mask,
8772 void *actions[METER_ACTIONS];
8773 struct mlx5_meter_domain_info *dtb;
8774 struct rte_flow_error error;
8778 dtb = &mtb->transfer;
8782 dtb = &mtb->ingress;
8783 /* Create the meter table with METER level. */
8784 dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
8785 egress, transfer, &error);
8787 DRV_LOG(ERR, "Failed to create meter policer table.");
8790 /* Create the meter suffix table with SUFFIX level. */
8791 dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
8792 MLX5_FLOW_TABLE_LEVEL_SUFFIX,
8793 egress, transfer, &error);
8794 if (!dtb->sfx_tbl) {
8795 DRV_LOG(ERR, "Failed to create meter suffix table.");
8798 /* Create matchers, Any and Color. */
8799 dv_attr.priority = 3;
8800 dv_attr.match_criteria_enable = 0;
8801 dtb->any_matcher = mlx5_glue->dv_create_flow_matcher(sh->ctx,
8804 if (!dtb->any_matcher) {
8805 DRV_LOG(ERR, "Failed to create meter"
8806 " policer default matcher.");
8809 dv_attr.priority = 0;
8810 dv_attr.match_criteria_enable =
8811 1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
8812 flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
8813 rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
8814 dtb->color_matcher = mlx5_glue->dv_create_flow_matcher(sh->ctx,
8817 if (!dtb->color_matcher) {
8818 DRV_LOG(ERR, "Failed to create meter policer color matcher.");
8821 if (mtb->count_actns[RTE_MTR_DROPPED])
8822 actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
8823 actions[i++] = mtb->drop_actn;
8824 /* Default rule: lowest priority, match any, actions: drop. */
8825 dtb->policer_rules[RTE_MTR_DROPPED] =
8826 mlx5_glue->dv_create_flow(dtb->any_matcher,
8827 (void *)&value, i, actions);
8828 if (!dtb->policer_rules[RTE_MTR_DROPPED]) {
8829 DRV_LOG(ERR, "Failed to create meter policer drop rule.");
8838 * Create the needed meter and suffix tables.
8839 * Lock free, (mutex should be acquired by caller).
8842 * Pointer to Ethernet device.
8844 * Pointer to the flow meter.
8847 * Pointer to table set on success, NULL otherwise and rte_errno is set.
8849 static struct mlx5_meter_domains_infos *
8850 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
8851 const struct mlx5_flow_meter *fm)
8853 struct mlx5_priv *priv = dev->data->dev_private;
8854 struct mlx5_meter_domains_infos *mtb;
8858 if (!priv->mtr_en) {
8859 rte_errno = ENOTSUP;
8862 mtb = rte_calloc(__func__, 1, sizeof(*mtb), 0);
8864 DRV_LOG(ERR, "Failed to allocate memory for meter.");
8867 /* Create meter count actions */
8868 for (i = 0; i <= RTE_MTR_DROPPED; i++) {
8869 struct mlx5_flow_counter *cnt;
8870 if (!fm->policer_stats.cnt[i])
8872 cnt = flow_dv_counter_get_by_idx(dev,
8873 fm->policer_stats.cnt[i], NULL);
8874 mtb->count_actns[i] = cnt->action;
8876 /* Create drop action. */
8877 mtb->drop_actn = mlx5_glue->dr_create_flow_action_drop();
8878 if (!mtb->drop_actn) {
8879 DRV_LOG(ERR, "Failed to create drop action.");
8882 /* Egress meter table. */
8883 ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
8885 DRV_LOG(ERR, "Failed to prepare egress meter table.");
8888 /* Ingress meter table. */
8889 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
8891 DRV_LOG(ERR, "Failed to prepare ingress meter table.");
8894 /* FDB meter table. */
8895 if (priv->config.dv_esw_en) {
8896 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
8897 priv->mtr_color_reg);
8899 DRV_LOG(ERR, "Failed to prepare fdb meter table.");
8905 flow_dv_destroy_mtr_tbl(dev, mtb);
8910 * Destroy domain policer rule.
8913 * Pointer to domain table.
8916 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
8920 for (i = 0; i < RTE_MTR_DROPPED; i++) {
8921 if (dt->policer_rules[i]) {
8922 claim_zero(mlx5_glue->dv_destroy_flow
8923 (dt->policer_rules[i]));
8924 dt->policer_rules[i] = NULL;
8927 if (dt->jump_actn) {
8928 claim_zero(mlx5_glue->destroy_flow_action(dt->jump_actn));
8929 dt->jump_actn = NULL;
8934 * Destroy policer rules.
8937 * Pointer to Ethernet device.
8939 * Pointer to flow meter structure.
8941 * Pointer to flow attributes.
8947 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
8948 const struct mlx5_flow_meter *fm,
8949 const struct rte_flow_attr *attr)
8951 struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
8956 flow_dv_destroy_domain_policer_rule(&mtb->egress);
8958 flow_dv_destroy_domain_policer_rule(&mtb->ingress);
8960 flow_dv_destroy_domain_policer_rule(&mtb->transfer);
8965 * Create specify domain meter policer rule.
8968 * Pointer to flow meter structure.
8970 * Pointer to DV meter table set.
8971 * @param[in] mtr_reg_c
8972 * Color match REG_C.
8975 * 0 on success, -1 otherwise.
8978 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
8979 struct mlx5_meter_domain_info *dtb,
8982 struct mlx5_flow_dv_match_params matcher = {
8983 .size = sizeof(matcher.buf),
8985 struct mlx5_flow_dv_match_params value = {
8986 .size = sizeof(value.buf),
8988 struct mlx5_meter_domains_infos *mtb = fm->mfts;
8989 void *actions[METER_ACTIONS];
8992 /* Create jump action. */
8993 if (!dtb->jump_actn)
8995 mlx5_glue->dr_create_flow_action_dest_flow_tbl
8996 (dtb->sfx_tbl->obj);
8997 if (!dtb->jump_actn) {
8998 DRV_LOG(ERR, "Failed to create policer jump action.");
9001 for (i = 0; i < RTE_MTR_DROPPED; i++) {
9004 flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
9005 rte_col_2_mlx5_col(i), UINT8_MAX);
9006 if (mtb->count_actns[i])
9007 actions[j++] = mtb->count_actns[i];
9008 if (fm->params.action[i] == MTR_POLICER_ACTION_DROP)
9009 actions[j++] = mtb->drop_actn;
9011 actions[j++] = dtb->jump_actn;
9012 dtb->policer_rules[i] =
9013 mlx5_glue->dv_create_flow(dtb->color_matcher,
9016 if (!dtb->policer_rules[i]) {
9017 DRV_LOG(ERR, "Failed to create policer rule.");
9028 * Create policer rules.
9031 * Pointer to Ethernet device.
9033 * Pointer to flow meter structure.
9035 * Pointer to flow attributes.
9038 * 0 on success, -1 otherwise.
9041 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
9042 struct mlx5_flow_meter *fm,
9043 const struct rte_flow_attr *attr)
9045 struct mlx5_priv *priv = dev->data->dev_private;
9046 struct mlx5_meter_domains_infos *mtb = fm->mfts;
9050 ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
9051 priv->mtr_color_reg);
9053 DRV_LOG(ERR, "Failed to create egress policer.");
9057 if (attr->ingress) {
9058 ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
9059 priv->mtr_color_reg);
9061 DRV_LOG(ERR, "Failed to create ingress policer.");
9065 if (attr->transfer) {
9066 ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
9067 priv->mtr_color_reg);
9069 DRV_LOG(ERR, "Failed to create transfer policer.");
9075 flow_dv_destroy_policer_rules(dev, fm, attr);
9080 * Query a devx counter.
9083 * Pointer to the Ethernet device structure.
9085 * Index to the flow counter.
9087 * Set to clear the counter statistics.
9089 * The statistics value of packets.
9091 * The statistics value of bytes.
9094 * 0 on success, otherwise return -1.
9097 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
9098 uint64_t *pkts, uint64_t *bytes)
9100 struct mlx5_priv *priv = dev->data->dev_private;
9101 struct mlx5_flow_counter *cnt;
9102 uint64_t inn_pkts, inn_bytes;
9105 if (!priv->config.devx)
9108 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
9111 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
9112 *pkts = inn_pkts - cnt->hits;
9113 *bytes = inn_bytes - cnt->bytes;
9115 cnt->hits = inn_pkts;
9116 cnt->bytes = inn_bytes;
9122 * Mutex-protected thunk to lock-free __flow_dv_translate().
9125 flow_dv_translate(struct rte_eth_dev *dev,
9126 struct mlx5_flow *dev_flow,
9127 const struct rte_flow_attr *attr,
9128 const struct rte_flow_item items[],
9129 const struct rte_flow_action actions[],
9130 struct rte_flow_error *error)
9134 flow_dv_shared_lock(dev);
9135 ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
9136 flow_dv_shared_unlock(dev);
9141 * Mutex-protected thunk to lock-free __flow_dv_apply().
9144 flow_dv_apply(struct rte_eth_dev *dev,
9145 struct rte_flow *flow,
9146 struct rte_flow_error *error)
9150 flow_dv_shared_lock(dev);
9151 ret = __flow_dv_apply(dev, flow, error);
9152 flow_dv_shared_unlock(dev);
9157 * Mutex-protected thunk to lock-free __flow_dv_remove().
9160 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
9162 flow_dv_shared_lock(dev);
9163 __flow_dv_remove(dev, flow);
9164 flow_dv_shared_unlock(dev);
9168 * Mutex-protected thunk to lock-free __flow_dv_destroy().
9171 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
9173 flow_dv_shared_lock(dev);
9174 __flow_dv_destroy(dev, flow);
9175 flow_dv_shared_unlock(dev);
9179 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
9182 flow_dv_counter_allocate(struct rte_eth_dev *dev)
9186 flow_dv_shared_lock(dev);
9187 cnt = flow_dv_counter_alloc(dev, 0, 0, 1);
9188 flow_dv_shared_unlock(dev);
9193 * Mutex-protected thunk to lock-free flow_dv_counter_release().
9196 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
9198 flow_dv_shared_lock(dev);
9199 flow_dv_counter_release(dev, cnt);
9200 flow_dv_shared_unlock(dev);
9203 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
9204 .validate = flow_dv_validate,
9205 .prepare = flow_dv_prepare,
9206 .translate = flow_dv_translate,
9207 .apply = flow_dv_apply,
9208 .remove = flow_dv_remove,
9209 .destroy = flow_dv_destroy,
9210 .query = flow_dv_query,
9211 .create_mtr_tbls = flow_dv_create_mtr_tbl,
9212 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
9213 .create_policer_rules = flow_dv_create_policer_rules,
9214 .destroy_policer_rules = flow_dv_destroy_policer_rules,
9215 .counter_alloc = flow_dv_counter_allocate,
9216 .counter_free = flow_dv_counter_free,
9217 .counter_query = flow_dv_counter_query,
9220 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */