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);
442 /* Deduce actual data width in bits from mask value. */
443 off_b = rte_bsf32(mask);
444 size_b = sizeof(uint32_t) * CHAR_BIT -
445 off_b - __builtin_clz(mask);
447 size_b = size_b == sizeof(uint32_t) * CHAR_BIT ? 0 : size_b;
448 actions[i] = (struct mlx5_modification_cmd) {
454 /* Convert entire record to expected big-endian format. */
455 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
456 if (type == MLX5_MODIFICATION_TYPE_COPY) {
458 actions[i].dst_field = dcopy->id;
459 actions[i].dst_offset =
460 (int)dcopy->offset < 0 ? off_b : dcopy->offset;
461 /* Convert entire record to big-endian format. */
462 actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
464 MLX5_ASSERT(item->spec);
465 data = flow_dv_fetch_field((const uint8_t *)item->spec +
466 field->offset, field->size);
467 /* Shift out the trailing masked bits from data. */
468 data = (data & mask) >> off_b;
469 actions[i].data1 = rte_cpu_to_be_32(data);
473 } while (field->size);
474 if (resource->actions_num == i)
475 return rte_flow_error_set(error, EINVAL,
476 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
477 "invalid modification flow item");
478 resource->actions_num = i;
483 * Convert modify-header set IPv4 address action to DV specification.
485 * @param[in,out] resource
486 * Pointer to the modify-header resource.
488 * Pointer to action specification.
490 * Pointer to the error structure.
493 * 0 on success, a negative errno value otherwise and rte_errno is set.
496 flow_dv_convert_action_modify_ipv4
497 (struct mlx5_flow_dv_modify_hdr_resource *resource,
498 const struct rte_flow_action *action,
499 struct rte_flow_error *error)
501 const struct rte_flow_action_set_ipv4 *conf =
502 (const struct rte_flow_action_set_ipv4 *)(action->conf);
503 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
504 struct rte_flow_item_ipv4 ipv4;
505 struct rte_flow_item_ipv4 ipv4_mask;
507 memset(&ipv4, 0, sizeof(ipv4));
508 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
509 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
510 ipv4.hdr.src_addr = conf->ipv4_addr;
511 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
513 ipv4.hdr.dst_addr = conf->ipv4_addr;
514 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
517 item.mask = &ipv4_mask;
518 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
519 MLX5_MODIFICATION_TYPE_SET, error);
523 * Convert modify-header set IPv6 address action to DV specification.
525 * @param[in,out] resource
526 * Pointer to the modify-header resource.
528 * Pointer to action specification.
530 * Pointer to the error structure.
533 * 0 on success, a negative errno value otherwise and rte_errno is set.
536 flow_dv_convert_action_modify_ipv6
537 (struct mlx5_flow_dv_modify_hdr_resource *resource,
538 const struct rte_flow_action *action,
539 struct rte_flow_error *error)
541 const struct rte_flow_action_set_ipv6 *conf =
542 (const struct rte_flow_action_set_ipv6 *)(action->conf);
543 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
544 struct rte_flow_item_ipv6 ipv6;
545 struct rte_flow_item_ipv6 ipv6_mask;
547 memset(&ipv6, 0, sizeof(ipv6));
548 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
549 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
550 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
551 sizeof(ipv6.hdr.src_addr));
552 memcpy(&ipv6_mask.hdr.src_addr,
553 &rte_flow_item_ipv6_mask.hdr.src_addr,
554 sizeof(ipv6.hdr.src_addr));
556 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
557 sizeof(ipv6.hdr.dst_addr));
558 memcpy(&ipv6_mask.hdr.dst_addr,
559 &rte_flow_item_ipv6_mask.hdr.dst_addr,
560 sizeof(ipv6.hdr.dst_addr));
563 item.mask = &ipv6_mask;
564 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
565 MLX5_MODIFICATION_TYPE_SET, error);
569 * Convert modify-header set MAC address action to DV specification.
571 * @param[in,out] resource
572 * Pointer to the modify-header resource.
574 * Pointer to action specification.
576 * Pointer to the error structure.
579 * 0 on success, a negative errno value otherwise and rte_errno is set.
582 flow_dv_convert_action_modify_mac
583 (struct mlx5_flow_dv_modify_hdr_resource *resource,
584 const struct rte_flow_action *action,
585 struct rte_flow_error *error)
587 const struct rte_flow_action_set_mac *conf =
588 (const struct rte_flow_action_set_mac *)(action->conf);
589 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
590 struct rte_flow_item_eth eth;
591 struct rte_flow_item_eth eth_mask;
593 memset(ð, 0, sizeof(eth));
594 memset(ð_mask, 0, sizeof(eth_mask));
595 if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
596 memcpy(ð.src.addr_bytes, &conf->mac_addr,
597 sizeof(eth.src.addr_bytes));
598 memcpy(ð_mask.src.addr_bytes,
599 &rte_flow_item_eth_mask.src.addr_bytes,
600 sizeof(eth_mask.src.addr_bytes));
602 memcpy(ð.dst.addr_bytes, &conf->mac_addr,
603 sizeof(eth.dst.addr_bytes));
604 memcpy(ð_mask.dst.addr_bytes,
605 &rte_flow_item_eth_mask.dst.addr_bytes,
606 sizeof(eth_mask.dst.addr_bytes));
609 item.mask = ð_mask;
610 return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
611 MLX5_MODIFICATION_TYPE_SET, error);
615 * Convert modify-header set VLAN VID action to DV specification.
617 * @param[in,out] resource
618 * Pointer to the modify-header resource.
620 * Pointer to action specification.
622 * Pointer to the error structure.
625 * 0 on success, a negative errno value otherwise and rte_errno is set.
628 flow_dv_convert_action_modify_vlan_vid
629 (struct mlx5_flow_dv_modify_hdr_resource *resource,
630 const struct rte_flow_action *action,
631 struct rte_flow_error *error)
633 const struct rte_flow_action_of_set_vlan_vid *conf =
634 (const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
635 int i = resource->actions_num;
636 struct mlx5_modification_cmd *actions = resource->actions;
637 struct field_modify_info *field = modify_vlan_out_first_vid;
639 if (i >= MLX5_MAX_MODIFY_NUM)
640 return rte_flow_error_set(error, EINVAL,
641 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
642 "too many items to modify");
643 actions[i] = (struct mlx5_modification_cmd) {
644 .action_type = MLX5_MODIFICATION_TYPE_SET,
646 .length = field->size,
647 .offset = field->offset,
649 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
650 actions[i].data1 = conf->vlan_vid;
651 actions[i].data1 = actions[i].data1 << 16;
652 resource->actions_num = ++i;
657 * Convert modify-header set TP action to DV specification.
659 * @param[in,out] resource
660 * Pointer to the modify-header resource.
662 * Pointer to action specification.
664 * Pointer to rte_flow_item objects list.
666 * Pointer to flow attributes structure.
667 * @param[in] dev_flow
668 * Pointer to the sub flow.
669 * @param[in] tunnel_decap
670 * Whether action is after tunnel decapsulation.
672 * Pointer to the error structure.
675 * 0 on success, a negative errno value otherwise and rte_errno is set.
678 flow_dv_convert_action_modify_tp
679 (struct mlx5_flow_dv_modify_hdr_resource *resource,
680 const struct rte_flow_action *action,
681 const struct rte_flow_item *items,
682 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
683 bool tunnel_decap, struct rte_flow_error *error)
685 const struct rte_flow_action_set_tp *conf =
686 (const struct rte_flow_action_set_tp *)(action->conf);
687 struct rte_flow_item item;
688 struct rte_flow_item_udp udp;
689 struct rte_flow_item_udp udp_mask;
690 struct rte_flow_item_tcp tcp;
691 struct rte_flow_item_tcp tcp_mask;
692 struct field_modify_info *field;
695 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
697 memset(&udp, 0, sizeof(udp));
698 memset(&udp_mask, 0, sizeof(udp_mask));
699 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
700 udp.hdr.src_port = conf->port;
701 udp_mask.hdr.src_port =
702 rte_flow_item_udp_mask.hdr.src_port;
704 udp.hdr.dst_port = conf->port;
705 udp_mask.hdr.dst_port =
706 rte_flow_item_udp_mask.hdr.dst_port;
708 item.type = RTE_FLOW_ITEM_TYPE_UDP;
710 item.mask = &udp_mask;
713 MLX5_ASSERT(attr->tcp);
714 memset(&tcp, 0, sizeof(tcp));
715 memset(&tcp_mask, 0, sizeof(tcp_mask));
716 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
717 tcp.hdr.src_port = conf->port;
718 tcp_mask.hdr.src_port =
719 rte_flow_item_tcp_mask.hdr.src_port;
721 tcp.hdr.dst_port = conf->port;
722 tcp_mask.hdr.dst_port =
723 rte_flow_item_tcp_mask.hdr.dst_port;
725 item.type = RTE_FLOW_ITEM_TYPE_TCP;
727 item.mask = &tcp_mask;
730 return flow_dv_convert_modify_action(&item, field, NULL, resource,
731 MLX5_MODIFICATION_TYPE_SET, error);
735 * Convert modify-header set TTL action to DV specification.
737 * @param[in,out] resource
738 * Pointer to the modify-header resource.
740 * Pointer to action specification.
742 * Pointer to rte_flow_item objects list.
744 * Pointer to flow attributes structure.
745 * @param[in] dev_flow
746 * Pointer to the sub flow.
747 * @param[in] tunnel_decap
748 * Whether action is after tunnel decapsulation.
750 * Pointer to the error structure.
753 * 0 on success, a negative errno value otherwise and rte_errno is set.
756 flow_dv_convert_action_modify_ttl
757 (struct mlx5_flow_dv_modify_hdr_resource *resource,
758 const struct rte_flow_action *action,
759 const struct rte_flow_item *items,
760 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
761 bool tunnel_decap, struct rte_flow_error *error)
763 const struct rte_flow_action_set_ttl *conf =
764 (const struct rte_flow_action_set_ttl *)(action->conf);
765 struct rte_flow_item item;
766 struct rte_flow_item_ipv4 ipv4;
767 struct rte_flow_item_ipv4 ipv4_mask;
768 struct rte_flow_item_ipv6 ipv6;
769 struct rte_flow_item_ipv6 ipv6_mask;
770 struct field_modify_info *field;
773 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
775 memset(&ipv4, 0, sizeof(ipv4));
776 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
777 ipv4.hdr.time_to_live = conf->ttl_value;
778 ipv4_mask.hdr.time_to_live = 0xFF;
779 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
781 item.mask = &ipv4_mask;
784 MLX5_ASSERT(attr->ipv6);
785 memset(&ipv6, 0, sizeof(ipv6));
786 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
787 ipv6.hdr.hop_limits = conf->ttl_value;
788 ipv6_mask.hdr.hop_limits = 0xFF;
789 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
791 item.mask = &ipv6_mask;
794 return flow_dv_convert_modify_action(&item, field, NULL, resource,
795 MLX5_MODIFICATION_TYPE_SET, error);
799 * Convert modify-header decrement TTL action to DV specification.
801 * @param[in,out] resource
802 * Pointer to the modify-header resource.
804 * Pointer to action specification.
806 * Pointer to rte_flow_item objects list.
808 * Pointer to flow attributes structure.
809 * @param[in] dev_flow
810 * Pointer to the sub flow.
811 * @param[in] tunnel_decap
812 * Whether action is after tunnel decapsulation.
814 * Pointer to the error structure.
817 * 0 on success, a negative errno value otherwise and rte_errno is set.
820 flow_dv_convert_action_modify_dec_ttl
821 (struct mlx5_flow_dv_modify_hdr_resource *resource,
822 const struct rte_flow_item *items,
823 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
824 bool tunnel_decap, struct rte_flow_error *error)
826 struct rte_flow_item item;
827 struct rte_flow_item_ipv4 ipv4;
828 struct rte_flow_item_ipv4 ipv4_mask;
829 struct rte_flow_item_ipv6 ipv6;
830 struct rte_flow_item_ipv6 ipv6_mask;
831 struct field_modify_info *field;
834 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
836 memset(&ipv4, 0, sizeof(ipv4));
837 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
838 ipv4.hdr.time_to_live = 0xFF;
839 ipv4_mask.hdr.time_to_live = 0xFF;
840 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
842 item.mask = &ipv4_mask;
845 MLX5_ASSERT(attr->ipv6);
846 memset(&ipv6, 0, sizeof(ipv6));
847 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
848 ipv6.hdr.hop_limits = 0xFF;
849 ipv6_mask.hdr.hop_limits = 0xFF;
850 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
852 item.mask = &ipv6_mask;
855 return flow_dv_convert_modify_action(&item, field, NULL, resource,
856 MLX5_MODIFICATION_TYPE_ADD, error);
860 * Convert modify-header increment/decrement TCP Sequence number
861 * to DV specification.
863 * @param[in,out] resource
864 * Pointer to the modify-header resource.
866 * Pointer to action specification.
868 * Pointer to the error structure.
871 * 0 on success, a negative errno value otherwise and rte_errno is set.
874 flow_dv_convert_action_modify_tcp_seq
875 (struct mlx5_flow_dv_modify_hdr_resource *resource,
876 const struct rte_flow_action *action,
877 struct rte_flow_error *error)
879 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
880 uint64_t value = rte_be_to_cpu_32(*conf);
881 struct rte_flow_item item;
882 struct rte_flow_item_tcp tcp;
883 struct rte_flow_item_tcp tcp_mask;
885 memset(&tcp, 0, sizeof(tcp));
886 memset(&tcp_mask, 0, sizeof(tcp_mask));
887 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
889 * The HW has no decrement operation, only increment operation.
890 * To simulate decrement X from Y using increment operation
891 * we need to add UINT32_MAX X times to Y.
892 * Each adding of UINT32_MAX decrements Y by 1.
895 tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
896 tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
897 item.type = RTE_FLOW_ITEM_TYPE_TCP;
899 item.mask = &tcp_mask;
900 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
901 MLX5_MODIFICATION_TYPE_ADD, error);
905 * Convert modify-header increment/decrement TCP Acknowledgment number
906 * to DV specification.
908 * @param[in,out] resource
909 * Pointer to the modify-header resource.
911 * Pointer to action specification.
913 * Pointer to the error structure.
916 * 0 on success, a negative errno value otherwise and rte_errno is set.
919 flow_dv_convert_action_modify_tcp_ack
920 (struct mlx5_flow_dv_modify_hdr_resource *resource,
921 const struct rte_flow_action *action,
922 struct rte_flow_error *error)
924 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
925 uint64_t value = rte_be_to_cpu_32(*conf);
926 struct rte_flow_item item;
927 struct rte_flow_item_tcp tcp;
928 struct rte_flow_item_tcp tcp_mask;
930 memset(&tcp, 0, sizeof(tcp));
931 memset(&tcp_mask, 0, sizeof(tcp_mask));
932 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
934 * The HW has no decrement operation, only increment operation.
935 * To simulate decrement X from Y using increment operation
936 * we need to add UINT32_MAX X times to Y.
937 * Each adding of UINT32_MAX decrements Y by 1.
940 tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
941 tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
942 item.type = RTE_FLOW_ITEM_TYPE_TCP;
944 item.mask = &tcp_mask;
945 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
946 MLX5_MODIFICATION_TYPE_ADD, error);
949 static enum mlx5_modification_field reg_to_field[] = {
950 [REG_NONE] = MLX5_MODI_OUT_NONE,
951 [REG_A] = MLX5_MODI_META_DATA_REG_A,
952 [REG_B] = MLX5_MODI_META_DATA_REG_B,
953 [REG_C_0] = MLX5_MODI_META_REG_C_0,
954 [REG_C_1] = MLX5_MODI_META_REG_C_1,
955 [REG_C_2] = MLX5_MODI_META_REG_C_2,
956 [REG_C_3] = MLX5_MODI_META_REG_C_3,
957 [REG_C_4] = MLX5_MODI_META_REG_C_4,
958 [REG_C_5] = MLX5_MODI_META_REG_C_5,
959 [REG_C_6] = MLX5_MODI_META_REG_C_6,
960 [REG_C_7] = MLX5_MODI_META_REG_C_7,
964 * Convert register set to DV specification.
966 * @param[in,out] resource
967 * Pointer to the modify-header resource.
969 * Pointer to action specification.
971 * Pointer to the error structure.
974 * 0 on success, a negative errno value otherwise and rte_errno is set.
977 flow_dv_convert_action_set_reg
978 (struct mlx5_flow_dv_modify_hdr_resource *resource,
979 const struct rte_flow_action *action,
980 struct rte_flow_error *error)
982 const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
983 struct mlx5_modification_cmd *actions = resource->actions;
984 uint32_t i = resource->actions_num;
986 if (i >= MLX5_MAX_MODIFY_NUM)
987 return rte_flow_error_set(error, EINVAL,
988 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
989 "too many items to modify");
990 MLX5_ASSERT(conf->id != REG_NONE);
991 MLX5_ASSERT(conf->id < RTE_DIM(reg_to_field));
992 actions[i] = (struct mlx5_modification_cmd) {
993 .action_type = MLX5_MODIFICATION_TYPE_SET,
994 .field = reg_to_field[conf->id],
996 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
997 actions[i].data1 = rte_cpu_to_be_32(conf->data);
999 resource->actions_num = i;
1004 * Convert SET_TAG action to DV specification.
1007 * Pointer to the rte_eth_dev structure.
1008 * @param[in,out] resource
1009 * Pointer to the modify-header resource.
1011 * Pointer to action specification.
1013 * Pointer to the error structure.
1016 * 0 on success, a negative errno value otherwise and rte_errno is set.
1019 flow_dv_convert_action_set_tag
1020 (struct rte_eth_dev *dev,
1021 struct mlx5_flow_dv_modify_hdr_resource *resource,
1022 const struct rte_flow_action_set_tag *conf,
1023 struct rte_flow_error *error)
1025 rte_be32_t data = rte_cpu_to_be_32(conf->data);
1026 rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1027 struct rte_flow_item item = {
1031 struct field_modify_info reg_c_x[] = {
1034 enum mlx5_modification_field reg_type;
1037 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1040 MLX5_ASSERT(ret != REG_NONE);
1041 MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1042 reg_type = reg_to_field[ret];
1043 MLX5_ASSERT(reg_type > 0);
1044 reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1045 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1046 MLX5_MODIFICATION_TYPE_SET, error);
1050 * Convert internal COPY_REG action to DV specification.
1053 * Pointer to the rte_eth_dev structure.
1054 * @param[in,out] res
1055 * Pointer to the modify-header resource.
1057 * Pointer to action specification.
1059 * Pointer to the error structure.
1062 * 0 on success, a negative errno value otherwise and rte_errno is set.
1065 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1066 struct mlx5_flow_dv_modify_hdr_resource *res,
1067 const struct rte_flow_action *action,
1068 struct rte_flow_error *error)
1070 const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1071 rte_be32_t mask = RTE_BE32(UINT32_MAX);
1072 struct rte_flow_item item = {
1076 struct field_modify_info reg_src[] = {
1077 {4, 0, reg_to_field[conf->src]},
1080 struct field_modify_info reg_dst = {
1082 .id = reg_to_field[conf->dst],
1084 /* Adjust reg_c[0] usage according to reported mask. */
1085 if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1086 struct mlx5_priv *priv = dev->data->dev_private;
1087 uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1089 MLX5_ASSERT(reg_c0);
1090 MLX5_ASSERT(priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY);
1091 if (conf->dst == REG_C_0) {
1092 /* Copy to reg_c[0], within mask only. */
1093 reg_dst.offset = rte_bsf32(reg_c0);
1095 * Mask is ignoring the enianness, because
1096 * there is no conversion in datapath.
1098 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1099 /* Copy from destination lower bits to reg_c[0]. */
1100 mask = reg_c0 >> reg_dst.offset;
1102 /* Copy from destination upper bits to reg_c[0]. */
1103 mask = reg_c0 << (sizeof(reg_c0) * CHAR_BIT -
1104 rte_fls_u32(reg_c0));
1107 mask = rte_cpu_to_be_32(reg_c0);
1108 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1109 /* Copy from reg_c[0] to destination lower bits. */
1112 /* Copy from reg_c[0] to destination upper bits. */
1113 reg_dst.offset = sizeof(reg_c0) * CHAR_BIT -
1114 (rte_fls_u32(reg_c0) -
1119 return flow_dv_convert_modify_action(&item,
1120 reg_src, ®_dst, res,
1121 MLX5_MODIFICATION_TYPE_COPY,
1126 * Convert MARK action to DV specification. This routine is used
1127 * in extensive metadata only and requires metadata register to be
1128 * handled. In legacy mode hardware tag resource is engaged.
1131 * Pointer to the rte_eth_dev structure.
1133 * Pointer to MARK action specification.
1134 * @param[in,out] resource
1135 * Pointer to the modify-header resource.
1137 * Pointer to the error structure.
1140 * 0 on success, a negative errno value otherwise and rte_errno is set.
1143 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1144 const struct rte_flow_action_mark *conf,
1145 struct mlx5_flow_dv_modify_hdr_resource *resource,
1146 struct rte_flow_error *error)
1148 struct mlx5_priv *priv = dev->data->dev_private;
1149 rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1150 priv->sh->dv_mark_mask);
1151 rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1152 struct rte_flow_item item = {
1156 struct field_modify_info reg_c_x[] = {
1157 {4, 0, 0}, /* dynamic instead of MLX5_MODI_META_REG_C_1. */
1163 return rte_flow_error_set(error, EINVAL,
1164 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1165 NULL, "zero mark action mask");
1166 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1169 MLX5_ASSERT(reg > 0);
1170 if (reg == REG_C_0) {
1171 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1172 uint32_t shl_c0 = rte_bsf32(msk_c0);
1174 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1175 mask = rte_cpu_to_be_32(mask) & msk_c0;
1176 mask = rte_cpu_to_be_32(mask << shl_c0);
1178 reg_c_x[0].id = reg_to_field[reg];
1179 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1180 MLX5_MODIFICATION_TYPE_SET, error);
1184 * Get metadata register index for specified steering domain.
1187 * Pointer to the rte_eth_dev structure.
1189 * Attributes of flow to determine steering domain.
1191 * Pointer to the error structure.
1194 * positive index on success, a negative errno value otherwise
1195 * and rte_errno is set.
1197 static enum modify_reg
1198 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1199 const struct rte_flow_attr *attr,
1200 struct rte_flow_error *error)
1203 mlx5_flow_get_reg_id(dev, attr->transfer ?
1207 MLX5_METADATA_RX, 0, error);
1209 return rte_flow_error_set(error,
1210 ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1211 NULL, "unavailable "
1212 "metadata register");
1217 * Convert SET_META action to DV specification.
1220 * Pointer to the rte_eth_dev structure.
1221 * @param[in,out] resource
1222 * Pointer to the modify-header resource.
1224 * Attributes of flow that includes this item.
1226 * Pointer to action specification.
1228 * Pointer to the error structure.
1231 * 0 on success, a negative errno value otherwise and rte_errno is set.
1234 flow_dv_convert_action_set_meta
1235 (struct rte_eth_dev *dev,
1236 struct mlx5_flow_dv_modify_hdr_resource *resource,
1237 const struct rte_flow_attr *attr,
1238 const struct rte_flow_action_set_meta *conf,
1239 struct rte_flow_error *error)
1241 uint32_t data = conf->data;
1242 uint32_t mask = conf->mask;
1243 struct rte_flow_item item = {
1247 struct field_modify_info reg_c_x[] = {
1250 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1255 * In datapath code there is no endianness
1256 * coversions for perfromance reasons, all
1257 * pattern conversions are done in rte_flow.
1259 if (reg == REG_C_0) {
1260 struct mlx5_priv *priv = dev->data->dev_private;
1261 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1264 MLX5_ASSERT(msk_c0);
1265 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1266 shl_c0 = rte_bsf32(msk_c0);
1268 shl_c0 = sizeof(msk_c0) * CHAR_BIT - rte_fls_u32(msk_c0);
1272 MLX5_ASSERT(!(~msk_c0 & rte_cpu_to_be_32(mask)));
1274 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1275 /* The routine expects parameters in memory as big-endian ones. */
1276 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1277 MLX5_MODIFICATION_TYPE_SET, error);
1281 * Convert modify-header set IPv4 DSCP action to DV specification.
1283 * @param[in,out] resource
1284 * Pointer to the modify-header resource.
1286 * Pointer to action specification.
1288 * Pointer to the error structure.
1291 * 0 on success, a negative errno value otherwise and rte_errno is set.
1294 flow_dv_convert_action_modify_ipv4_dscp
1295 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1296 const struct rte_flow_action *action,
1297 struct rte_flow_error *error)
1299 const struct rte_flow_action_set_dscp *conf =
1300 (const struct rte_flow_action_set_dscp *)(action->conf);
1301 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1302 struct rte_flow_item_ipv4 ipv4;
1303 struct rte_flow_item_ipv4 ipv4_mask;
1305 memset(&ipv4, 0, sizeof(ipv4));
1306 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1307 ipv4.hdr.type_of_service = conf->dscp;
1308 ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1310 item.mask = &ipv4_mask;
1311 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1312 MLX5_MODIFICATION_TYPE_SET, error);
1316 * Convert modify-header set IPv6 DSCP action to DV specification.
1318 * @param[in,out] resource
1319 * Pointer to the modify-header resource.
1321 * Pointer to action specification.
1323 * Pointer to the error structure.
1326 * 0 on success, a negative errno value otherwise and rte_errno is set.
1329 flow_dv_convert_action_modify_ipv6_dscp
1330 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1331 const struct rte_flow_action *action,
1332 struct rte_flow_error *error)
1334 const struct rte_flow_action_set_dscp *conf =
1335 (const struct rte_flow_action_set_dscp *)(action->conf);
1336 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1337 struct rte_flow_item_ipv6 ipv6;
1338 struct rte_flow_item_ipv6 ipv6_mask;
1340 memset(&ipv6, 0, sizeof(ipv6));
1341 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1343 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1344 * rdma-core only accept the DSCP bits byte aligned start from
1345 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1346 * bits in IPv6 case as rdma-core requires byte aligned value.
1348 ipv6.hdr.vtc_flow = conf->dscp;
1349 ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1351 item.mask = &ipv6_mask;
1352 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1353 MLX5_MODIFICATION_TYPE_SET, error);
1357 * Validate MARK item.
1360 * Pointer to the rte_eth_dev structure.
1362 * Item specification.
1364 * Attributes of flow that includes this item.
1366 * Pointer to error structure.
1369 * 0 on success, a negative errno value otherwise and rte_errno is set.
1372 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1373 const struct rte_flow_item *item,
1374 const struct rte_flow_attr *attr __rte_unused,
1375 struct rte_flow_error *error)
1377 struct mlx5_priv *priv = dev->data->dev_private;
1378 struct mlx5_dev_config *config = &priv->config;
1379 const struct rte_flow_item_mark *spec = item->spec;
1380 const struct rte_flow_item_mark *mask = item->mask;
1381 const struct rte_flow_item_mark nic_mask = {
1382 .id = priv->sh->dv_mark_mask,
1386 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1387 return rte_flow_error_set(error, ENOTSUP,
1388 RTE_FLOW_ERROR_TYPE_ITEM, item,
1389 "extended metadata feature"
1391 if (!mlx5_flow_ext_mreg_supported(dev))
1392 return rte_flow_error_set(error, ENOTSUP,
1393 RTE_FLOW_ERROR_TYPE_ITEM, item,
1394 "extended metadata register"
1395 " isn't supported");
1397 return rte_flow_error_set(error, ENOTSUP,
1398 RTE_FLOW_ERROR_TYPE_ITEM, item,
1399 "extended metadata register"
1400 " isn't available");
1401 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1405 return rte_flow_error_set(error, EINVAL,
1406 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1408 "data cannot be empty");
1409 if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1410 return rte_flow_error_set(error, EINVAL,
1411 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1413 "mark id exceeds the limit");
1417 return rte_flow_error_set(error, EINVAL,
1418 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1419 "mask cannot be zero");
1421 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1422 (const uint8_t *)&nic_mask,
1423 sizeof(struct rte_flow_item_mark),
1431 * Validate META item.
1434 * Pointer to the rte_eth_dev structure.
1436 * Item specification.
1438 * Attributes of flow that includes this item.
1440 * Pointer to error structure.
1443 * 0 on success, a negative errno value otherwise and rte_errno is set.
1446 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1447 const struct rte_flow_item *item,
1448 const struct rte_flow_attr *attr,
1449 struct rte_flow_error *error)
1451 struct mlx5_priv *priv = dev->data->dev_private;
1452 struct mlx5_dev_config *config = &priv->config;
1453 const struct rte_flow_item_meta *spec = item->spec;
1454 const struct rte_flow_item_meta *mask = item->mask;
1455 struct rte_flow_item_meta nic_mask = {
1462 return rte_flow_error_set(error, EINVAL,
1463 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1465 "data cannot be empty");
1466 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1467 if (!mlx5_flow_ext_mreg_supported(dev))
1468 return rte_flow_error_set(error, ENOTSUP,
1469 RTE_FLOW_ERROR_TYPE_ITEM, item,
1470 "extended metadata register"
1471 " isn't supported");
1472 reg = flow_dv_get_metadata_reg(dev, attr, error);
1476 return rte_flow_error_set(error, ENOTSUP,
1477 RTE_FLOW_ERROR_TYPE_ITEM, item,
1481 nic_mask.data = priv->sh->dv_meta_mask;
1484 mask = &rte_flow_item_meta_mask;
1486 return rte_flow_error_set(error, EINVAL,
1487 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1488 "mask cannot be zero");
1490 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1491 (const uint8_t *)&nic_mask,
1492 sizeof(struct rte_flow_item_meta),
1498 * Validate TAG item.
1501 * Pointer to the rte_eth_dev structure.
1503 * Item specification.
1505 * Attributes of flow that includes this item.
1507 * Pointer to error structure.
1510 * 0 on success, a negative errno value otherwise and rte_errno is set.
1513 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
1514 const struct rte_flow_item *item,
1515 const struct rte_flow_attr *attr __rte_unused,
1516 struct rte_flow_error *error)
1518 const struct rte_flow_item_tag *spec = item->spec;
1519 const struct rte_flow_item_tag *mask = item->mask;
1520 const struct rte_flow_item_tag nic_mask = {
1521 .data = RTE_BE32(UINT32_MAX),
1526 if (!mlx5_flow_ext_mreg_supported(dev))
1527 return rte_flow_error_set(error, ENOTSUP,
1528 RTE_FLOW_ERROR_TYPE_ITEM, item,
1529 "extensive metadata register"
1530 " isn't supported");
1532 return rte_flow_error_set(error, EINVAL,
1533 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1535 "data cannot be empty");
1537 mask = &rte_flow_item_tag_mask;
1539 return rte_flow_error_set(error, EINVAL,
1540 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1541 "mask cannot be zero");
1543 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1544 (const uint8_t *)&nic_mask,
1545 sizeof(struct rte_flow_item_tag),
1549 if (mask->index != 0xff)
1550 return rte_flow_error_set(error, EINVAL,
1551 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1552 "partial mask for tag index"
1553 " is not supported");
1554 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
1557 MLX5_ASSERT(ret != REG_NONE);
1562 * Validate vport item.
1565 * Pointer to the rte_eth_dev structure.
1567 * Item specification.
1569 * Attributes of flow that includes this item.
1570 * @param[in] item_flags
1571 * Bit-fields that holds the items detected until now.
1573 * Pointer to error structure.
1576 * 0 on success, a negative errno value otherwise and rte_errno is set.
1579 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
1580 const struct rte_flow_item *item,
1581 const struct rte_flow_attr *attr,
1582 uint64_t item_flags,
1583 struct rte_flow_error *error)
1585 const struct rte_flow_item_port_id *spec = item->spec;
1586 const struct rte_flow_item_port_id *mask = item->mask;
1587 const struct rte_flow_item_port_id switch_mask = {
1590 struct mlx5_priv *esw_priv;
1591 struct mlx5_priv *dev_priv;
1594 if (!attr->transfer)
1595 return rte_flow_error_set(error, EINVAL,
1596 RTE_FLOW_ERROR_TYPE_ITEM,
1598 "match on port id is valid only"
1599 " when transfer flag is enabled");
1600 if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
1601 return rte_flow_error_set(error, ENOTSUP,
1602 RTE_FLOW_ERROR_TYPE_ITEM, item,
1603 "multiple source ports are not"
1606 mask = &switch_mask;
1607 if (mask->id != 0xffffffff)
1608 return rte_flow_error_set(error, ENOTSUP,
1609 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1611 "no support for partial mask on"
1613 ret = mlx5_flow_item_acceptable
1614 (item, (const uint8_t *)mask,
1615 (const uint8_t *)&rte_flow_item_port_id_mask,
1616 sizeof(struct rte_flow_item_port_id),
1622 esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
1624 return rte_flow_error_set(error, rte_errno,
1625 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1626 "failed to obtain E-Switch info for"
1628 dev_priv = mlx5_dev_to_eswitch_info(dev);
1630 return rte_flow_error_set(error, rte_errno,
1631 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1633 "failed to obtain E-Switch info");
1634 if (esw_priv->domain_id != dev_priv->domain_id)
1635 return rte_flow_error_set(error, EINVAL,
1636 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1637 "cannot match on a port from a"
1638 " different E-Switch");
1643 * Validate GTP item.
1646 * Pointer to the rte_eth_dev structure.
1648 * Item specification.
1649 * @param[in] item_flags
1650 * Bit-fields that holds the items detected until now.
1652 * Pointer to error structure.
1655 * 0 on success, a negative errno value otherwise and rte_errno is set.
1658 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1659 const struct rte_flow_item *item,
1660 uint64_t item_flags,
1661 struct rte_flow_error *error)
1663 struct mlx5_priv *priv = dev->data->dev_private;
1664 const struct rte_flow_item_gtp *mask = item->mask;
1665 const struct rte_flow_item_gtp nic_mask = {
1667 .teid = RTE_BE32(0xffffffff),
1670 if (!priv->config.hca_attr.tunnel_stateless_gtp)
1671 return rte_flow_error_set(error, ENOTSUP,
1672 RTE_FLOW_ERROR_TYPE_ITEM, item,
1673 "GTP support is not enabled");
1674 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1675 return rte_flow_error_set(error, ENOTSUP,
1676 RTE_FLOW_ERROR_TYPE_ITEM, item,
1677 "multiple tunnel layers not"
1679 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1680 return rte_flow_error_set(error, EINVAL,
1681 RTE_FLOW_ERROR_TYPE_ITEM, item,
1682 "no outer UDP layer found");
1684 mask = &rte_flow_item_gtp_mask;
1685 return mlx5_flow_item_acceptable
1686 (item, (const uint8_t *)mask,
1687 (const uint8_t *)&nic_mask,
1688 sizeof(struct rte_flow_item_gtp),
1693 * Validate the pop VLAN action.
1696 * Pointer to the rte_eth_dev structure.
1697 * @param[in] action_flags
1698 * Holds the actions detected until now.
1700 * Pointer to the pop vlan action.
1701 * @param[in] item_flags
1702 * The items found in this flow rule.
1704 * Pointer to flow attributes.
1706 * Pointer to error structure.
1709 * 0 on success, a negative errno value otherwise and rte_errno is set.
1712 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
1713 uint64_t action_flags,
1714 const struct rte_flow_action *action,
1715 uint64_t item_flags,
1716 const struct rte_flow_attr *attr,
1717 struct rte_flow_error *error)
1719 const struct mlx5_priv *priv = dev->data->dev_private;
1723 if (!priv->sh->pop_vlan_action)
1724 return rte_flow_error_set(error, ENOTSUP,
1725 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1727 "pop vlan action is not supported");
1729 return rte_flow_error_set(error, ENOTSUP,
1730 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
1732 "pop vlan action not supported for "
1734 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1735 return rte_flow_error_set(error, ENOTSUP,
1736 RTE_FLOW_ERROR_TYPE_ACTION, action,
1737 "no support for multiple VLAN "
1739 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
1740 return rte_flow_error_set(error, ENOTSUP,
1741 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1743 "cannot pop vlan without a "
1744 "match on (outer) vlan in the flow");
1745 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1746 return rte_flow_error_set(error, EINVAL,
1747 RTE_FLOW_ERROR_TYPE_ACTION, action,
1748 "wrong action order, port_id should "
1749 "be after pop VLAN action");
1750 if (!attr->transfer && priv->representor)
1751 return rte_flow_error_set(error, ENOTSUP,
1752 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1753 "pop vlan action for VF representor "
1754 "not supported on NIC table");
1759 * Get VLAN default info from vlan match info.
1762 * the list of item specifications.
1764 * pointer VLAN info to fill to.
1767 * 0 on success, a negative errno value otherwise and rte_errno is set.
1770 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
1771 struct rte_vlan_hdr *vlan)
1773 const struct rte_flow_item_vlan nic_mask = {
1774 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
1775 MLX5DV_FLOW_VLAN_VID_MASK),
1776 .inner_type = RTE_BE16(0xffff),
1781 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1782 int type = items->type;
1784 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
1785 type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
1788 if (items->type != RTE_FLOW_ITEM_TYPE_END) {
1789 const struct rte_flow_item_vlan *vlan_m = items->mask;
1790 const struct rte_flow_item_vlan *vlan_v = items->spec;
1794 /* Only full match values are accepted */
1795 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
1796 MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
1797 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
1799 rte_be_to_cpu_16(vlan_v->tci &
1800 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
1802 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
1803 MLX5DV_FLOW_VLAN_VID_MASK_BE) {
1804 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
1806 rte_be_to_cpu_16(vlan_v->tci &
1807 MLX5DV_FLOW_VLAN_VID_MASK_BE);
1809 if (vlan_m->inner_type == nic_mask.inner_type)
1810 vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
1811 vlan_m->inner_type);
1816 * Validate the push VLAN action.
1819 * Pointer to the rte_eth_dev structure.
1820 * @param[in] action_flags
1821 * Holds the actions detected until now.
1822 * @param[in] item_flags
1823 * The items found in this flow rule.
1825 * Pointer to the action structure.
1827 * Pointer to flow attributes
1829 * Pointer to error structure.
1832 * 0 on success, a negative errno value otherwise and rte_errno is set.
1835 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
1836 uint64_t action_flags,
1837 const struct rte_flow_item_vlan *vlan_m,
1838 const struct rte_flow_action *action,
1839 const struct rte_flow_attr *attr,
1840 struct rte_flow_error *error)
1842 const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
1843 const struct mlx5_priv *priv = dev->data->dev_private;
1845 if (!attr->transfer && attr->ingress)
1846 return rte_flow_error_set(error, ENOTSUP,
1847 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1849 "push VLAN action not supported for "
1851 if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
1852 push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
1853 return rte_flow_error_set(error, EINVAL,
1854 RTE_FLOW_ERROR_TYPE_ACTION, action,
1855 "invalid vlan ethertype");
1856 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1857 return rte_flow_error_set(error, ENOTSUP,
1858 RTE_FLOW_ERROR_TYPE_ACTION, action,
1859 "no support for multiple VLAN "
1861 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1862 return rte_flow_error_set(error, EINVAL,
1863 RTE_FLOW_ERROR_TYPE_ACTION, action,
1864 "wrong action order, port_id should "
1865 "be after push VLAN");
1866 if (!attr->transfer && priv->representor)
1867 return rte_flow_error_set(error, ENOTSUP,
1868 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1869 "push vlan action for VF representor "
1870 "not supported on NIC table");
1872 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
1873 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
1874 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
1875 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
1876 !(mlx5_flow_find_action
1877 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
1878 return rte_flow_error_set(error, EINVAL,
1879 RTE_FLOW_ERROR_TYPE_ACTION, action,
1880 "not full match mask on VLAN PCP and "
1881 "there is no of_set_vlan_pcp action, "
1882 "push VLAN action cannot figure out "
1885 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
1886 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
1887 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
1888 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
1889 !(mlx5_flow_find_action
1890 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
1891 return rte_flow_error_set(error, EINVAL,
1892 RTE_FLOW_ERROR_TYPE_ACTION, action,
1893 "not full match mask on VLAN VID and "
1894 "there is no of_set_vlan_vid action, "
1895 "push VLAN action cannot figure out "
1902 * Validate the set VLAN PCP.
1904 * @param[in] action_flags
1905 * Holds the actions detected until now.
1906 * @param[in] actions
1907 * Pointer to the list of actions remaining in the flow rule.
1909 * Pointer to error structure.
1912 * 0 on success, a negative errno value otherwise and rte_errno is set.
1915 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
1916 const struct rte_flow_action actions[],
1917 struct rte_flow_error *error)
1919 const struct rte_flow_action *action = actions;
1920 const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
1922 if (conf->vlan_pcp > 7)
1923 return rte_flow_error_set(error, EINVAL,
1924 RTE_FLOW_ERROR_TYPE_ACTION, action,
1925 "VLAN PCP value is too big");
1926 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
1927 return rte_flow_error_set(error, ENOTSUP,
1928 RTE_FLOW_ERROR_TYPE_ACTION, action,
1929 "set VLAN PCP action must follow "
1930 "the push VLAN action");
1931 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
1932 return rte_flow_error_set(error, ENOTSUP,
1933 RTE_FLOW_ERROR_TYPE_ACTION, action,
1934 "Multiple VLAN PCP modification are "
1936 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1937 return rte_flow_error_set(error, EINVAL,
1938 RTE_FLOW_ERROR_TYPE_ACTION, action,
1939 "wrong action order, port_id should "
1940 "be after set VLAN PCP");
1945 * Validate the set VLAN VID.
1947 * @param[in] item_flags
1948 * Holds the items detected in this rule.
1949 * @param[in] action_flags
1950 * Holds the actions detected until now.
1951 * @param[in] actions
1952 * Pointer to the list of actions remaining in the flow rule.
1954 * Pointer to error structure.
1957 * 0 on success, a negative errno value otherwise and rte_errno is set.
1960 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
1961 uint64_t action_flags,
1962 const struct rte_flow_action actions[],
1963 struct rte_flow_error *error)
1965 const struct rte_flow_action *action = actions;
1966 const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
1968 if (conf->vlan_vid > RTE_BE16(0xFFE))
1969 return rte_flow_error_set(error, EINVAL,
1970 RTE_FLOW_ERROR_TYPE_ACTION, action,
1971 "VLAN VID value is too big");
1972 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
1973 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
1974 return rte_flow_error_set(error, ENOTSUP,
1975 RTE_FLOW_ERROR_TYPE_ACTION, action,
1976 "set VLAN VID action must follow push"
1977 " VLAN action or match on VLAN item");
1978 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
1979 return rte_flow_error_set(error, ENOTSUP,
1980 RTE_FLOW_ERROR_TYPE_ACTION, action,
1981 "Multiple VLAN VID modifications are "
1983 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1984 return rte_flow_error_set(error, EINVAL,
1985 RTE_FLOW_ERROR_TYPE_ACTION, action,
1986 "wrong action order, port_id should "
1987 "be after set VLAN VID");
1992 * Validate the FLAG action.
1995 * Pointer to the rte_eth_dev structure.
1996 * @param[in] action_flags
1997 * Holds the actions detected until now.
1999 * Pointer to flow attributes
2001 * Pointer to error structure.
2004 * 0 on success, a negative errno value otherwise and rte_errno is set.
2007 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2008 uint64_t action_flags,
2009 const struct rte_flow_attr *attr,
2010 struct rte_flow_error *error)
2012 struct mlx5_priv *priv = dev->data->dev_private;
2013 struct mlx5_dev_config *config = &priv->config;
2016 /* Fall back if no extended metadata register support. */
2017 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2018 return mlx5_flow_validate_action_flag(action_flags, attr,
2020 /* Extensive metadata mode requires registers. */
2021 if (!mlx5_flow_ext_mreg_supported(dev))
2022 return rte_flow_error_set(error, ENOTSUP,
2023 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2024 "no metadata registers "
2025 "to support flag action");
2026 if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2027 return rte_flow_error_set(error, ENOTSUP,
2028 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2029 "extended metadata register"
2030 " isn't available");
2031 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2034 MLX5_ASSERT(ret > 0);
2035 if (action_flags & MLX5_FLOW_ACTION_MARK)
2036 return rte_flow_error_set(error, EINVAL,
2037 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2038 "can't mark and flag in same flow");
2039 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2040 return rte_flow_error_set(error, EINVAL,
2041 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2043 " actions in same flow");
2048 * Validate MARK action.
2051 * Pointer to the rte_eth_dev structure.
2053 * Pointer to action.
2054 * @param[in] action_flags
2055 * Holds the actions detected until now.
2057 * Pointer to flow attributes
2059 * Pointer to error structure.
2062 * 0 on success, a negative errno value otherwise and rte_errno is set.
2065 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2066 const struct rte_flow_action *action,
2067 uint64_t action_flags,
2068 const struct rte_flow_attr *attr,
2069 struct rte_flow_error *error)
2071 struct mlx5_priv *priv = dev->data->dev_private;
2072 struct mlx5_dev_config *config = &priv->config;
2073 const struct rte_flow_action_mark *mark = action->conf;
2076 /* Fall back if no extended metadata register support. */
2077 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2078 return mlx5_flow_validate_action_mark(action, action_flags,
2080 /* Extensive metadata mode requires registers. */
2081 if (!mlx5_flow_ext_mreg_supported(dev))
2082 return rte_flow_error_set(error, ENOTSUP,
2083 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2084 "no metadata registers "
2085 "to support mark action");
2086 if (!priv->sh->dv_mark_mask)
2087 return rte_flow_error_set(error, ENOTSUP,
2088 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2089 "extended metadata register"
2090 " isn't available");
2091 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2094 MLX5_ASSERT(ret > 0);
2096 return rte_flow_error_set(error, EINVAL,
2097 RTE_FLOW_ERROR_TYPE_ACTION, action,
2098 "configuration cannot be null");
2099 if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2100 return rte_flow_error_set(error, EINVAL,
2101 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2103 "mark id exceeds the limit");
2104 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2105 return rte_flow_error_set(error, EINVAL,
2106 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2107 "can't flag and mark in same flow");
2108 if (action_flags & MLX5_FLOW_ACTION_MARK)
2109 return rte_flow_error_set(error, EINVAL,
2110 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2111 "can't have 2 mark actions in same"
2117 * Validate SET_META action.
2120 * Pointer to the rte_eth_dev structure.
2122 * Pointer to the action structure.
2123 * @param[in] action_flags
2124 * Holds the actions detected until now.
2126 * Pointer to flow attributes
2128 * Pointer to error structure.
2131 * 0 on success, a negative errno value otherwise and rte_errno is set.
2134 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2135 const struct rte_flow_action *action,
2136 uint64_t action_flags __rte_unused,
2137 const struct rte_flow_attr *attr,
2138 struct rte_flow_error *error)
2140 const struct rte_flow_action_set_meta *conf;
2141 uint32_t nic_mask = UINT32_MAX;
2144 if (!mlx5_flow_ext_mreg_supported(dev))
2145 return rte_flow_error_set(error, ENOTSUP,
2146 RTE_FLOW_ERROR_TYPE_ACTION, action,
2147 "extended metadata register"
2148 " isn't supported");
2149 reg = flow_dv_get_metadata_reg(dev, attr, error);
2152 if (reg != REG_A && reg != REG_B) {
2153 struct mlx5_priv *priv = dev->data->dev_private;
2155 nic_mask = priv->sh->dv_meta_mask;
2157 if (!(action->conf))
2158 return rte_flow_error_set(error, EINVAL,
2159 RTE_FLOW_ERROR_TYPE_ACTION, action,
2160 "configuration cannot be null");
2161 conf = (const struct rte_flow_action_set_meta *)action->conf;
2163 return rte_flow_error_set(error, EINVAL,
2164 RTE_FLOW_ERROR_TYPE_ACTION, action,
2165 "zero mask doesn't have any effect");
2166 if (conf->mask & ~nic_mask)
2167 return rte_flow_error_set(error, EINVAL,
2168 RTE_FLOW_ERROR_TYPE_ACTION, action,
2169 "meta data must be within reg C0");
2174 * Validate SET_TAG action.
2177 * Pointer to the rte_eth_dev structure.
2179 * Pointer to the action structure.
2180 * @param[in] action_flags
2181 * Holds the actions detected until now.
2183 * Pointer to flow attributes
2185 * Pointer to error structure.
2188 * 0 on success, a negative errno value otherwise and rte_errno is set.
2191 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2192 const struct rte_flow_action *action,
2193 uint64_t action_flags,
2194 const struct rte_flow_attr *attr,
2195 struct rte_flow_error *error)
2197 const struct rte_flow_action_set_tag *conf;
2198 const uint64_t terminal_action_flags =
2199 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2200 MLX5_FLOW_ACTION_RSS;
2203 if (!mlx5_flow_ext_mreg_supported(dev))
2204 return rte_flow_error_set(error, ENOTSUP,
2205 RTE_FLOW_ERROR_TYPE_ACTION, action,
2206 "extensive metadata register"
2207 " isn't supported");
2208 if (!(action->conf))
2209 return rte_flow_error_set(error, EINVAL,
2210 RTE_FLOW_ERROR_TYPE_ACTION, action,
2211 "configuration cannot be null");
2212 conf = (const struct rte_flow_action_set_tag *)action->conf;
2214 return rte_flow_error_set(error, EINVAL,
2215 RTE_FLOW_ERROR_TYPE_ACTION, action,
2216 "zero mask doesn't have any effect");
2217 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2220 if (!attr->transfer && attr->ingress &&
2221 (action_flags & terminal_action_flags))
2222 return rte_flow_error_set(error, EINVAL,
2223 RTE_FLOW_ERROR_TYPE_ACTION, action,
2224 "set_tag has no effect"
2225 " with terminal actions");
2230 * Validate count action.
2233 * Pointer to rte_eth_dev structure.
2235 * Pointer to error structure.
2238 * 0 on success, a negative errno value otherwise and rte_errno is set.
2241 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2242 struct rte_flow_error *error)
2244 struct mlx5_priv *priv = dev->data->dev_private;
2246 if (!priv->config.devx)
2248 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2252 return rte_flow_error_set
2254 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2256 "count action not supported");
2260 * Validate the L2 encap action.
2263 * Pointer to the rte_eth_dev structure.
2264 * @param[in] action_flags
2265 * Holds the actions detected until now.
2267 * Pointer to the action structure.
2269 * Pointer to flow attributes.
2271 * Pointer to error structure.
2274 * 0 on success, a negative errno value otherwise and rte_errno is set.
2277 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2278 uint64_t action_flags,
2279 const struct rte_flow_action *action,
2280 const struct rte_flow_attr *attr,
2281 struct rte_flow_error *error)
2283 const struct mlx5_priv *priv = dev->data->dev_private;
2285 if (!(action->conf))
2286 return rte_flow_error_set(error, EINVAL,
2287 RTE_FLOW_ERROR_TYPE_ACTION, action,
2288 "configuration cannot be null");
2289 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2290 return rte_flow_error_set(error, EINVAL,
2291 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2292 "can only have a single encap action "
2294 if (!attr->transfer && priv->representor)
2295 return rte_flow_error_set(error, ENOTSUP,
2296 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2297 "encap action for VF representor "
2298 "not supported on NIC table");
2303 * Validate a decap action.
2306 * Pointer to the rte_eth_dev structure.
2307 * @param[in] action_flags
2308 * Holds the actions detected until now.
2310 * Pointer to flow attributes
2312 * Pointer to error structure.
2315 * 0 on success, a negative errno value otherwise and rte_errno is set.
2318 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2319 uint64_t action_flags,
2320 const struct rte_flow_attr *attr,
2321 struct rte_flow_error *error)
2323 const struct mlx5_priv *priv = dev->data->dev_private;
2325 if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2326 return rte_flow_error_set(error, ENOTSUP,
2327 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2329 MLX5_FLOW_ACTION_DECAP ? "can only "
2330 "have a single decap action" : "decap "
2331 "after encap is not supported");
2332 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2333 return rte_flow_error_set(error, EINVAL,
2334 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2335 "can't have decap action after"
2338 return rte_flow_error_set(error, ENOTSUP,
2339 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2341 "decap action not supported for "
2343 if (!attr->transfer && priv->representor)
2344 return rte_flow_error_set(error, ENOTSUP,
2345 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2346 "decap action for VF representor "
2347 "not supported on NIC table");
2351 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2354 * Validate the raw encap and decap actions.
2357 * Pointer to the rte_eth_dev structure.
2359 * Pointer to the decap action.
2361 * Pointer to the encap action.
2363 * Pointer to flow attributes
2364 * @param[in/out] action_flags
2365 * Holds the actions detected until now.
2366 * @param[out] actions_n
2367 * pointer to the number of actions counter.
2369 * Pointer to error structure.
2372 * 0 on success, a negative errno value otherwise and rte_errno is set.
2375 flow_dv_validate_action_raw_encap_decap
2376 (struct rte_eth_dev *dev,
2377 const struct rte_flow_action_raw_decap *decap,
2378 const struct rte_flow_action_raw_encap *encap,
2379 const struct rte_flow_attr *attr, uint64_t *action_flags,
2380 int *actions_n, struct rte_flow_error *error)
2382 const struct mlx5_priv *priv = dev->data->dev_private;
2385 if (encap && (!encap->size || !encap->data))
2386 return rte_flow_error_set(error, EINVAL,
2387 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2388 "raw encap data cannot be empty");
2389 if (decap && encap) {
2390 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2391 encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2394 else if (encap->size <=
2395 MLX5_ENCAPSULATION_DECISION_SIZE &&
2397 MLX5_ENCAPSULATION_DECISION_SIZE)
2400 else if (encap->size >
2401 MLX5_ENCAPSULATION_DECISION_SIZE &&
2403 MLX5_ENCAPSULATION_DECISION_SIZE)
2404 /* 2 L2 actions: encap and decap. */
2407 return rte_flow_error_set(error,
2409 RTE_FLOW_ERROR_TYPE_ACTION,
2410 NULL, "unsupported too small "
2411 "raw decap and too small raw "
2412 "encap combination");
2415 ret = flow_dv_validate_action_decap(dev, *action_flags, attr,
2419 *action_flags |= MLX5_FLOW_ACTION_DECAP;
2423 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2424 return rte_flow_error_set(error, ENOTSUP,
2425 RTE_FLOW_ERROR_TYPE_ACTION,
2427 "small raw encap size");
2428 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2429 return rte_flow_error_set(error, EINVAL,
2430 RTE_FLOW_ERROR_TYPE_ACTION,
2432 "more than one encap action");
2433 if (!attr->transfer && priv->representor)
2434 return rte_flow_error_set
2436 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2437 "encap action for VF representor "
2438 "not supported on NIC table");
2439 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
2446 * Find existing encap/decap resource or create and register a new one.
2448 * @param[in, out] dev
2449 * Pointer to rte_eth_dev structure.
2450 * @param[in, out] resource
2451 * Pointer to encap/decap resource.
2452 * @parm[in, out] dev_flow
2453 * Pointer to the dev_flow.
2455 * pointer to error structure.
2458 * 0 on success otherwise -errno and errno is set.
2461 flow_dv_encap_decap_resource_register
2462 (struct rte_eth_dev *dev,
2463 struct mlx5_flow_dv_encap_decap_resource *resource,
2464 struct mlx5_flow *dev_flow,
2465 struct rte_flow_error *error)
2467 struct mlx5_priv *priv = dev->data->dev_private;
2468 struct mlx5_ibv_shared *sh = priv->sh;
2469 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2470 struct mlx5dv_dr_domain *domain;
2473 resource->flags = dev_flow->dv.group ? 0 : 1;
2474 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2475 domain = sh->fdb_domain;
2476 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2477 domain = sh->rx_domain;
2479 domain = sh->tx_domain;
2480 /* Lookup a matching resource from cache. */
2481 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], sh->encaps_decaps, idx,
2482 cache_resource, next) {
2483 if (resource->reformat_type == cache_resource->reformat_type &&
2484 resource->ft_type == cache_resource->ft_type &&
2485 resource->flags == cache_resource->flags &&
2486 resource->size == cache_resource->size &&
2487 !memcmp((const void *)resource->buf,
2488 (const void *)cache_resource->buf,
2490 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
2491 (void *)cache_resource,
2492 rte_atomic32_read(&cache_resource->refcnt));
2493 rte_atomic32_inc(&cache_resource->refcnt);
2494 dev_flow->handle->dvh.rix_encap_decap = idx;
2495 dev_flow->dv.encap_decap = cache_resource;
2499 /* Register new encap/decap resource. */
2500 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2501 &dev_flow->handle->dvh.rix_encap_decap);
2502 if (!cache_resource)
2503 return rte_flow_error_set(error, ENOMEM,
2504 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2505 "cannot allocate resource memory");
2506 *cache_resource = *resource;
2507 cache_resource->verbs_action =
2508 mlx5_glue->dv_create_flow_action_packet_reformat
2509 (sh->ctx, cache_resource->reformat_type,
2510 cache_resource->ft_type, domain, cache_resource->flags,
2511 cache_resource->size,
2512 (cache_resource->size ? cache_resource->buf : NULL));
2513 if (!cache_resource->verbs_action) {
2514 rte_free(cache_resource);
2515 return rte_flow_error_set(error, ENOMEM,
2516 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2517 NULL, "cannot create action");
2519 rte_atomic32_init(&cache_resource->refcnt);
2520 rte_atomic32_inc(&cache_resource->refcnt);
2521 ILIST_INSERT(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], &sh->encaps_decaps,
2522 dev_flow->handle->dvh.rix_encap_decap, cache_resource,
2524 dev_flow->dv.encap_decap = cache_resource;
2525 DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
2526 (void *)cache_resource,
2527 rte_atomic32_read(&cache_resource->refcnt));
2532 * Find existing table jump resource or create and register a new one.
2534 * @param[in, out] dev
2535 * Pointer to rte_eth_dev structure.
2536 * @param[in, out] tbl
2537 * Pointer to flow table resource.
2538 * @parm[in, out] dev_flow
2539 * Pointer to the dev_flow.
2541 * pointer to error structure.
2544 * 0 on success otherwise -errno and errno is set.
2547 flow_dv_jump_tbl_resource_register
2548 (struct rte_eth_dev *dev __rte_unused,
2549 struct mlx5_flow_tbl_resource *tbl,
2550 struct mlx5_flow *dev_flow,
2551 struct rte_flow_error *error)
2553 struct mlx5_flow_tbl_data_entry *tbl_data =
2554 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2558 cnt = rte_atomic32_read(&tbl_data->jump.refcnt);
2560 tbl_data->jump.action =
2561 mlx5_glue->dr_create_flow_action_dest_flow_tbl
2563 if (!tbl_data->jump.action)
2564 return rte_flow_error_set(error, ENOMEM,
2565 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2566 NULL, "cannot create jump action");
2567 DRV_LOG(DEBUG, "new jump table resource %p: refcnt %d++",
2568 (void *)&tbl_data->jump, cnt);
2570 /* old jump should not make the table ref++. */
2571 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
2572 MLX5_ASSERT(tbl_data->jump.action);
2573 DRV_LOG(DEBUG, "existed jump table resource %p: refcnt %d++",
2574 (void *)&tbl_data->jump, cnt);
2576 rte_atomic32_inc(&tbl_data->jump.refcnt);
2577 dev_flow->handle->rix_jump = tbl_data->idx;
2578 dev_flow->dv.jump = &tbl_data->jump;
2583 * Find existing table port ID resource or create and register a new one.
2585 * @param[in, out] dev
2586 * Pointer to rte_eth_dev structure.
2587 * @param[in, out] resource
2588 * Pointer to port ID action resource.
2589 * @parm[in, out] dev_flow
2590 * Pointer to the dev_flow.
2592 * pointer to error structure.
2595 * 0 on success otherwise -errno and errno is set.
2598 flow_dv_port_id_action_resource_register
2599 (struct rte_eth_dev *dev,
2600 struct mlx5_flow_dv_port_id_action_resource *resource,
2601 struct mlx5_flow *dev_flow,
2602 struct rte_flow_error *error)
2604 struct mlx5_priv *priv = dev->data->dev_private;
2605 struct mlx5_ibv_shared *sh = priv->sh;
2606 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
2609 /* Lookup a matching resource from cache. */
2610 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
2611 idx, cache_resource, next) {
2612 if (resource->port_id == cache_resource->port_id) {
2613 DRV_LOG(DEBUG, "port id action resource resource %p: "
2615 (void *)cache_resource,
2616 rte_atomic32_read(&cache_resource->refcnt));
2617 rte_atomic32_inc(&cache_resource->refcnt);
2618 dev_flow->handle->rix_port_id_action = idx;
2619 dev_flow->dv.port_id_action = cache_resource;
2623 /* Register new port id action resource. */
2624 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
2625 &dev_flow->handle->rix_port_id_action);
2626 if (!cache_resource)
2627 return rte_flow_error_set(error, ENOMEM,
2628 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2629 "cannot allocate resource memory");
2630 *cache_resource = *resource;
2632 * Depending on rdma_core version the glue routine calls
2633 * either mlx5dv_dr_action_create_dest_ib_port(domain, ibv_port)
2634 * or mlx5dv_dr_action_create_dest_vport(domain, vport_id).
2636 cache_resource->action =
2637 mlx5_glue->dr_create_flow_action_dest_port
2638 (priv->sh->fdb_domain, resource->port_id);
2639 if (!cache_resource->action) {
2640 rte_free(cache_resource);
2641 return rte_flow_error_set(error, ENOMEM,
2642 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2643 NULL, "cannot create action");
2645 rte_atomic32_init(&cache_resource->refcnt);
2646 rte_atomic32_inc(&cache_resource->refcnt);
2647 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
2648 dev_flow->handle->rix_port_id_action, cache_resource,
2650 dev_flow->dv.port_id_action = cache_resource;
2651 DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
2652 (void *)cache_resource,
2653 rte_atomic32_read(&cache_resource->refcnt));
2658 * Find existing push vlan resource or create and register a new one.
2660 * @param [in, out] dev
2661 * Pointer to rte_eth_dev structure.
2662 * @param[in, out] resource
2663 * Pointer to port ID action resource.
2664 * @parm[in, out] dev_flow
2665 * Pointer to the dev_flow.
2667 * pointer to error structure.
2670 * 0 on success otherwise -errno and errno is set.
2673 flow_dv_push_vlan_action_resource_register
2674 (struct rte_eth_dev *dev,
2675 struct mlx5_flow_dv_push_vlan_action_resource *resource,
2676 struct mlx5_flow *dev_flow,
2677 struct rte_flow_error *error)
2679 struct mlx5_priv *priv = dev->data->dev_private;
2680 struct mlx5_ibv_shared *sh = priv->sh;
2681 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
2682 struct mlx5dv_dr_domain *domain;
2685 /* Lookup a matching resource from cache. */
2686 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2687 sh->push_vlan_action_list, idx, cache_resource, next) {
2688 if (resource->vlan_tag == cache_resource->vlan_tag &&
2689 resource->ft_type == cache_resource->ft_type) {
2690 DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
2692 (void *)cache_resource,
2693 rte_atomic32_read(&cache_resource->refcnt));
2694 rte_atomic32_inc(&cache_resource->refcnt);
2695 dev_flow->handle->dvh.rix_push_vlan = idx;
2696 dev_flow->dv.push_vlan_res = cache_resource;
2700 /* Register new push_vlan action resource. */
2701 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2702 &dev_flow->handle->dvh.rix_push_vlan);
2703 if (!cache_resource)
2704 return rte_flow_error_set(error, ENOMEM,
2705 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2706 "cannot allocate resource memory");
2707 *cache_resource = *resource;
2708 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2709 domain = sh->fdb_domain;
2710 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2711 domain = sh->rx_domain;
2713 domain = sh->tx_domain;
2714 cache_resource->action =
2715 mlx5_glue->dr_create_flow_action_push_vlan(domain,
2716 resource->vlan_tag);
2717 if (!cache_resource->action) {
2718 rte_free(cache_resource);
2719 return rte_flow_error_set(error, ENOMEM,
2720 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2721 NULL, "cannot create action");
2723 rte_atomic32_init(&cache_resource->refcnt);
2724 rte_atomic32_inc(&cache_resource->refcnt);
2725 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2726 &sh->push_vlan_action_list,
2727 dev_flow->handle->dvh.rix_push_vlan,
2728 cache_resource, next);
2729 dev_flow->dv.push_vlan_res = cache_resource;
2730 DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
2731 (void *)cache_resource,
2732 rte_atomic32_read(&cache_resource->refcnt));
2736 * Get the size of specific rte_flow_item_type
2738 * @param[in] item_type
2739 * Tested rte_flow_item_type.
2742 * sizeof struct item_type, 0 if void or irrelevant.
2745 flow_dv_get_item_len(const enum rte_flow_item_type item_type)
2749 switch (item_type) {
2750 case RTE_FLOW_ITEM_TYPE_ETH:
2751 retval = sizeof(struct rte_flow_item_eth);
2753 case RTE_FLOW_ITEM_TYPE_VLAN:
2754 retval = sizeof(struct rte_flow_item_vlan);
2756 case RTE_FLOW_ITEM_TYPE_IPV4:
2757 retval = sizeof(struct rte_flow_item_ipv4);
2759 case RTE_FLOW_ITEM_TYPE_IPV6:
2760 retval = sizeof(struct rte_flow_item_ipv6);
2762 case RTE_FLOW_ITEM_TYPE_UDP:
2763 retval = sizeof(struct rte_flow_item_udp);
2765 case RTE_FLOW_ITEM_TYPE_TCP:
2766 retval = sizeof(struct rte_flow_item_tcp);
2768 case RTE_FLOW_ITEM_TYPE_VXLAN:
2769 retval = sizeof(struct rte_flow_item_vxlan);
2771 case RTE_FLOW_ITEM_TYPE_GRE:
2772 retval = sizeof(struct rte_flow_item_gre);
2774 case RTE_FLOW_ITEM_TYPE_NVGRE:
2775 retval = sizeof(struct rte_flow_item_nvgre);
2777 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2778 retval = sizeof(struct rte_flow_item_vxlan_gpe);
2780 case RTE_FLOW_ITEM_TYPE_MPLS:
2781 retval = sizeof(struct rte_flow_item_mpls);
2783 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
2791 #define MLX5_ENCAP_IPV4_VERSION 0x40
2792 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
2793 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
2794 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
2795 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
2796 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
2797 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
2800 * Convert the encap action data from list of rte_flow_item to raw buffer
2803 * Pointer to rte_flow_item objects list.
2805 * Pointer to the output buffer.
2807 * Pointer to the output buffer size.
2809 * Pointer to the error structure.
2812 * 0 on success, a negative errno value otherwise and rte_errno is set.
2815 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
2816 size_t *size, struct rte_flow_error *error)
2818 struct rte_ether_hdr *eth = NULL;
2819 struct rte_vlan_hdr *vlan = NULL;
2820 struct rte_ipv4_hdr *ipv4 = NULL;
2821 struct rte_ipv6_hdr *ipv6 = NULL;
2822 struct rte_udp_hdr *udp = NULL;
2823 struct rte_vxlan_hdr *vxlan = NULL;
2824 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
2825 struct rte_gre_hdr *gre = NULL;
2827 size_t temp_size = 0;
2830 return rte_flow_error_set(error, EINVAL,
2831 RTE_FLOW_ERROR_TYPE_ACTION,
2832 NULL, "invalid empty data");
2833 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2834 len = flow_dv_get_item_len(items->type);
2835 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
2836 return rte_flow_error_set(error, EINVAL,
2837 RTE_FLOW_ERROR_TYPE_ACTION,
2838 (void *)items->type,
2839 "items total size is too big"
2840 " for encap action");
2841 rte_memcpy((void *)&buf[temp_size], items->spec, len);
2842 switch (items->type) {
2843 case RTE_FLOW_ITEM_TYPE_ETH:
2844 eth = (struct rte_ether_hdr *)&buf[temp_size];
2846 case RTE_FLOW_ITEM_TYPE_VLAN:
2847 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
2849 return rte_flow_error_set(error, EINVAL,
2850 RTE_FLOW_ERROR_TYPE_ACTION,
2851 (void *)items->type,
2852 "eth header not found");
2853 if (!eth->ether_type)
2854 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
2856 case RTE_FLOW_ITEM_TYPE_IPV4:
2857 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
2859 return rte_flow_error_set(error, EINVAL,
2860 RTE_FLOW_ERROR_TYPE_ACTION,
2861 (void *)items->type,
2862 "neither eth nor vlan"
2864 if (vlan && !vlan->eth_proto)
2865 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2866 else if (eth && !eth->ether_type)
2867 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2868 if (!ipv4->version_ihl)
2869 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
2870 MLX5_ENCAP_IPV4_IHL_MIN;
2871 if (!ipv4->time_to_live)
2872 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
2874 case RTE_FLOW_ITEM_TYPE_IPV6:
2875 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
2877 return rte_flow_error_set(error, EINVAL,
2878 RTE_FLOW_ERROR_TYPE_ACTION,
2879 (void *)items->type,
2880 "neither eth nor vlan"
2882 if (vlan && !vlan->eth_proto)
2883 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
2884 else if (eth && !eth->ether_type)
2885 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
2886 if (!ipv6->vtc_flow)
2888 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
2889 if (!ipv6->hop_limits)
2890 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
2892 case RTE_FLOW_ITEM_TYPE_UDP:
2893 udp = (struct rte_udp_hdr *)&buf[temp_size];
2895 return rte_flow_error_set(error, EINVAL,
2896 RTE_FLOW_ERROR_TYPE_ACTION,
2897 (void *)items->type,
2898 "ip header not found");
2899 if (ipv4 && !ipv4->next_proto_id)
2900 ipv4->next_proto_id = IPPROTO_UDP;
2901 else if (ipv6 && !ipv6->proto)
2902 ipv6->proto = IPPROTO_UDP;
2904 case RTE_FLOW_ITEM_TYPE_VXLAN:
2905 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
2907 return rte_flow_error_set(error, EINVAL,
2908 RTE_FLOW_ERROR_TYPE_ACTION,
2909 (void *)items->type,
2910 "udp header not found");
2912 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
2913 if (!vxlan->vx_flags)
2915 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
2917 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2918 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
2920 return rte_flow_error_set(error, EINVAL,
2921 RTE_FLOW_ERROR_TYPE_ACTION,
2922 (void *)items->type,
2923 "udp header not found");
2924 if (!vxlan_gpe->proto)
2925 return rte_flow_error_set(error, EINVAL,
2926 RTE_FLOW_ERROR_TYPE_ACTION,
2927 (void *)items->type,
2928 "next protocol not found");
2931 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
2932 if (!vxlan_gpe->vx_flags)
2933 vxlan_gpe->vx_flags =
2934 MLX5_ENCAP_VXLAN_GPE_FLAGS;
2936 case RTE_FLOW_ITEM_TYPE_GRE:
2937 case RTE_FLOW_ITEM_TYPE_NVGRE:
2938 gre = (struct rte_gre_hdr *)&buf[temp_size];
2940 return rte_flow_error_set(error, EINVAL,
2941 RTE_FLOW_ERROR_TYPE_ACTION,
2942 (void *)items->type,
2943 "next protocol not found");
2945 return rte_flow_error_set(error, EINVAL,
2946 RTE_FLOW_ERROR_TYPE_ACTION,
2947 (void *)items->type,
2948 "ip header not found");
2949 if (ipv4 && !ipv4->next_proto_id)
2950 ipv4->next_proto_id = IPPROTO_GRE;
2951 else if (ipv6 && !ipv6->proto)
2952 ipv6->proto = IPPROTO_GRE;
2954 case RTE_FLOW_ITEM_TYPE_VOID:
2957 return rte_flow_error_set(error, EINVAL,
2958 RTE_FLOW_ERROR_TYPE_ACTION,
2959 (void *)items->type,
2960 "unsupported item type");
2970 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
2972 struct rte_ether_hdr *eth = NULL;
2973 struct rte_vlan_hdr *vlan = NULL;
2974 struct rte_ipv6_hdr *ipv6 = NULL;
2975 struct rte_udp_hdr *udp = NULL;
2979 eth = (struct rte_ether_hdr *)data;
2980 next_hdr = (char *)(eth + 1);
2981 proto = RTE_BE16(eth->ether_type);
2984 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
2985 vlan = (struct rte_vlan_hdr *)next_hdr;
2986 proto = RTE_BE16(vlan->eth_proto);
2987 next_hdr += sizeof(struct rte_vlan_hdr);
2990 /* HW calculates IPv4 csum. no need to proceed */
2991 if (proto == RTE_ETHER_TYPE_IPV4)
2994 /* non IPv4/IPv6 header. not supported */
2995 if (proto != RTE_ETHER_TYPE_IPV6) {
2996 return rte_flow_error_set(error, ENOTSUP,
2997 RTE_FLOW_ERROR_TYPE_ACTION,
2998 NULL, "Cannot offload non IPv4/IPv6");
3001 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3003 /* ignore non UDP */
3004 if (ipv6->proto != IPPROTO_UDP)
3007 udp = (struct rte_udp_hdr *)(ipv6 + 1);
3008 udp->dgram_cksum = 0;
3014 * Convert L2 encap action to DV specification.
3017 * Pointer to rte_eth_dev structure.
3019 * Pointer to action structure.
3020 * @param[in, out] dev_flow
3021 * Pointer to the mlx5_flow.
3022 * @param[in] transfer
3023 * Mark if the flow is E-Switch flow.
3025 * Pointer to the error structure.
3028 * 0 on success, a negative errno value otherwise and rte_errno is set.
3031 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3032 const struct rte_flow_action *action,
3033 struct mlx5_flow *dev_flow,
3035 struct rte_flow_error *error)
3037 const struct rte_flow_item *encap_data;
3038 const struct rte_flow_action_raw_encap *raw_encap_data;
3039 struct mlx5_flow_dv_encap_decap_resource res = {
3041 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3042 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3043 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3046 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3048 (const struct rte_flow_action_raw_encap *)action->conf;
3049 res.size = raw_encap_data->size;
3050 memcpy(res.buf, raw_encap_data->data, res.size);
3052 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3054 ((const struct rte_flow_action_vxlan_encap *)
3055 action->conf)->definition;
3058 ((const struct rte_flow_action_nvgre_encap *)
3059 action->conf)->definition;
3060 if (flow_dv_convert_encap_data(encap_data, res.buf,
3064 if (flow_dv_zero_encap_udp_csum(res.buf, error))
3066 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3067 return rte_flow_error_set(error, EINVAL,
3068 RTE_FLOW_ERROR_TYPE_ACTION,
3069 NULL, "can't create L2 encap action");
3074 * Convert L2 decap action to DV specification.
3077 * Pointer to rte_eth_dev structure.
3078 * @param[in, out] dev_flow
3079 * Pointer to the mlx5_flow.
3080 * @param[in] transfer
3081 * Mark if the flow is E-Switch flow.
3083 * Pointer to the error structure.
3086 * 0 on success, a negative errno value otherwise and rte_errno is set.
3089 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3090 struct mlx5_flow *dev_flow,
3092 struct rte_flow_error *error)
3094 struct mlx5_flow_dv_encap_decap_resource res = {
3097 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3098 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3099 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3102 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3103 return rte_flow_error_set(error, EINVAL,
3104 RTE_FLOW_ERROR_TYPE_ACTION,
3105 NULL, "can't create L2 decap action");
3110 * Convert raw decap/encap (L3 tunnel) action to DV specification.
3113 * Pointer to rte_eth_dev structure.
3115 * Pointer to action structure.
3116 * @param[in, out] dev_flow
3117 * Pointer to the mlx5_flow.
3119 * Pointer to the flow attributes.
3121 * Pointer to the error structure.
3124 * 0 on success, a negative errno value otherwise and rte_errno is set.
3127 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3128 const struct rte_flow_action *action,
3129 struct mlx5_flow *dev_flow,
3130 const struct rte_flow_attr *attr,
3131 struct rte_flow_error *error)
3133 const struct rte_flow_action_raw_encap *encap_data;
3134 struct mlx5_flow_dv_encap_decap_resource res;
3136 memset(&res, 0, sizeof(res));
3137 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3138 res.size = encap_data->size;
3139 memcpy(res.buf, encap_data->data, res.size);
3140 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3141 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3142 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3144 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3146 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3147 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3148 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3149 return rte_flow_error_set(error, EINVAL,
3150 RTE_FLOW_ERROR_TYPE_ACTION,
3151 NULL, "can't create encap action");
3156 * Create action push VLAN.
3159 * Pointer to rte_eth_dev structure.
3161 * Pointer to the flow attributes.
3163 * Pointer to the vlan to push to the Ethernet header.
3164 * @param[in, out] dev_flow
3165 * Pointer to the mlx5_flow.
3167 * Pointer to the error structure.
3170 * 0 on success, a negative errno value otherwise and rte_errno is set.
3173 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3174 const struct rte_flow_attr *attr,
3175 const struct rte_vlan_hdr *vlan,
3176 struct mlx5_flow *dev_flow,
3177 struct rte_flow_error *error)
3179 struct mlx5_flow_dv_push_vlan_action_resource res;
3181 memset(&res, 0, sizeof(res));
3183 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3186 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3188 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3189 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3190 return flow_dv_push_vlan_action_resource_register
3191 (dev, &res, dev_flow, error);
3195 * Validate the modify-header actions.
3197 * @param[in] action_flags
3198 * Holds the actions detected until now.
3200 * Pointer to the modify action.
3202 * Pointer to error structure.
3205 * 0 on success, a negative errno value otherwise and rte_errno is set.
3208 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3209 const struct rte_flow_action *action,
3210 struct rte_flow_error *error)
3212 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3213 return rte_flow_error_set(error, EINVAL,
3214 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3215 NULL, "action configuration not set");
3216 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3217 return rte_flow_error_set(error, EINVAL,
3218 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3219 "can't have encap action before"
3225 * Validate the modify-header MAC address actions.
3227 * @param[in] action_flags
3228 * Holds the actions detected until now.
3230 * Pointer to the modify action.
3231 * @param[in] item_flags
3232 * Holds the items detected.
3234 * Pointer to error structure.
3237 * 0 on success, a negative errno value otherwise and rte_errno is set.
3240 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3241 const struct rte_flow_action *action,
3242 const uint64_t item_flags,
3243 struct rte_flow_error *error)
3247 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3249 if (!(item_flags & MLX5_FLOW_LAYER_L2))
3250 return rte_flow_error_set(error, EINVAL,
3251 RTE_FLOW_ERROR_TYPE_ACTION,
3253 "no L2 item in pattern");
3259 * Validate the modify-header IPv4 address actions.
3261 * @param[in] action_flags
3262 * Holds the actions detected until now.
3264 * Pointer to the modify action.
3265 * @param[in] item_flags
3266 * Holds the items detected.
3268 * Pointer to error structure.
3271 * 0 on success, a negative errno value otherwise and rte_errno is set.
3274 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3275 const struct rte_flow_action *action,
3276 const uint64_t item_flags,
3277 struct rte_flow_error *error)
3282 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3284 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3285 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3286 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3287 if (!(item_flags & layer))
3288 return rte_flow_error_set(error, EINVAL,
3289 RTE_FLOW_ERROR_TYPE_ACTION,
3291 "no ipv4 item in pattern");
3297 * Validate the modify-header IPv6 address actions.
3299 * @param[in] action_flags
3300 * Holds the actions detected until now.
3302 * Pointer to the modify action.
3303 * @param[in] item_flags
3304 * Holds the items detected.
3306 * Pointer to error structure.
3309 * 0 on success, a negative errno value otherwise and rte_errno is set.
3312 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3313 const struct rte_flow_action *action,
3314 const uint64_t item_flags,
3315 struct rte_flow_error *error)
3320 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3322 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3323 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3324 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3325 if (!(item_flags & layer))
3326 return rte_flow_error_set(error, EINVAL,
3327 RTE_FLOW_ERROR_TYPE_ACTION,
3329 "no ipv6 item in pattern");
3335 * Validate the modify-header TP actions.
3337 * @param[in] action_flags
3338 * Holds the actions detected until now.
3340 * Pointer to the modify action.
3341 * @param[in] item_flags
3342 * Holds the items detected.
3344 * Pointer to error structure.
3347 * 0 on success, a negative errno value otherwise and rte_errno is set.
3350 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3351 const struct rte_flow_action *action,
3352 const uint64_t item_flags,
3353 struct rte_flow_error *error)
3358 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3360 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3361 MLX5_FLOW_LAYER_INNER_L4 :
3362 MLX5_FLOW_LAYER_OUTER_L4;
3363 if (!(item_flags & layer))
3364 return rte_flow_error_set(error, EINVAL,
3365 RTE_FLOW_ERROR_TYPE_ACTION,
3366 NULL, "no transport layer "
3373 * Validate the modify-header actions of increment/decrement
3374 * TCP Sequence-number.
3376 * @param[in] action_flags
3377 * Holds the actions detected until now.
3379 * Pointer to the modify action.
3380 * @param[in] item_flags
3381 * Holds the items detected.
3383 * Pointer to error structure.
3386 * 0 on success, a negative errno value otherwise and rte_errno is set.
3389 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3390 const struct rte_flow_action *action,
3391 const uint64_t item_flags,
3392 struct rte_flow_error *error)
3397 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3399 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3400 MLX5_FLOW_LAYER_INNER_L4_TCP :
3401 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3402 if (!(item_flags & layer))
3403 return rte_flow_error_set(error, EINVAL,
3404 RTE_FLOW_ERROR_TYPE_ACTION,
3405 NULL, "no TCP item in"
3407 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3408 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3409 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3410 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3411 return rte_flow_error_set(error, EINVAL,
3412 RTE_FLOW_ERROR_TYPE_ACTION,
3414 "cannot decrease and increase"
3415 " TCP sequence number"
3416 " at the same time");
3422 * Validate the modify-header actions of increment/decrement
3423 * TCP Acknowledgment number.
3425 * @param[in] action_flags
3426 * Holds the actions detected until now.
3428 * Pointer to the modify action.
3429 * @param[in] item_flags
3430 * Holds the items detected.
3432 * Pointer to error structure.
3435 * 0 on success, a negative errno value otherwise and rte_errno is set.
3438 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3439 const struct rte_flow_action *action,
3440 const uint64_t item_flags,
3441 struct rte_flow_error *error)
3446 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3448 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3449 MLX5_FLOW_LAYER_INNER_L4_TCP :
3450 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3451 if (!(item_flags & layer))
3452 return rte_flow_error_set(error, EINVAL,
3453 RTE_FLOW_ERROR_TYPE_ACTION,
3454 NULL, "no TCP item in"
3456 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3457 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3458 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3459 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3460 return rte_flow_error_set(error, EINVAL,
3461 RTE_FLOW_ERROR_TYPE_ACTION,
3463 "cannot decrease and increase"
3464 " TCP acknowledgment number"
3465 " at the same time");
3471 * Validate the modify-header TTL actions.
3473 * @param[in] action_flags
3474 * Holds the actions detected until now.
3476 * Pointer to the modify action.
3477 * @param[in] item_flags
3478 * Holds the items detected.
3480 * Pointer to error structure.
3483 * 0 on success, a negative errno value otherwise and rte_errno is set.
3486 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3487 const struct rte_flow_action *action,
3488 const uint64_t item_flags,
3489 struct rte_flow_error *error)
3494 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3496 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3497 MLX5_FLOW_LAYER_INNER_L3 :
3498 MLX5_FLOW_LAYER_OUTER_L3;
3499 if (!(item_flags & layer))
3500 return rte_flow_error_set(error, EINVAL,
3501 RTE_FLOW_ERROR_TYPE_ACTION,
3503 "no IP protocol in pattern");
3509 * Validate jump action.
3512 * Pointer to the jump action.
3513 * @param[in] action_flags
3514 * Holds the actions detected until now.
3515 * @param[in] attributes
3516 * Pointer to flow attributes
3517 * @param[in] external
3518 * Action belongs to flow rule created by request external to PMD.
3520 * Pointer to error structure.
3523 * 0 on success, a negative errno value otherwise and rte_errno is set.
3526 flow_dv_validate_action_jump(const struct rte_flow_action *action,
3527 uint64_t action_flags,
3528 const struct rte_flow_attr *attributes,
3529 bool external, struct rte_flow_error *error)
3531 uint32_t target_group, table;
3534 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3535 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3536 return rte_flow_error_set(error, EINVAL,
3537 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3538 "can't have 2 fate actions in"
3540 if (action_flags & MLX5_FLOW_ACTION_METER)
3541 return rte_flow_error_set(error, ENOTSUP,
3542 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3543 "jump with meter not support");
3545 return rte_flow_error_set(error, EINVAL,
3546 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3547 NULL, "action configuration not set");
3549 ((const struct rte_flow_action_jump *)action->conf)->group;
3550 ret = mlx5_flow_group_to_table(attributes, external, target_group,
3551 true, &table, error);
3554 if (attributes->group == target_group)
3555 return rte_flow_error_set(error, EINVAL,
3556 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3557 "target group must be other than"
3558 " the current flow group");
3563 * Validate the port_id action.
3566 * Pointer to rte_eth_dev structure.
3567 * @param[in] action_flags
3568 * Bit-fields that holds the actions detected until now.
3570 * Port_id RTE action structure.
3572 * Attributes of flow that includes this action.
3574 * Pointer to error structure.
3577 * 0 on success, a negative errno value otherwise and rte_errno is set.
3580 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
3581 uint64_t action_flags,
3582 const struct rte_flow_action *action,
3583 const struct rte_flow_attr *attr,
3584 struct rte_flow_error *error)
3586 const struct rte_flow_action_port_id *port_id;
3587 struct mlx5_priv *act_priv;
3588 struct mlx5_priv *dev_priv;
3591 if (!attr->transfer)
3592 return rte_flow_error_set(error, ENOTSUP,
3593 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3595 "port id action is valid in transfer"
3597 if (!action || !action->conf)
3598 return rte_flow_error_set(error, ENOTSUP,
3599 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3601 "port id action parameters must be"
3603 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3604 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3605 return rte_flow_error_set(error, EINVAL,
3606 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3607 "can have only one fate actions in"
3609 dev_priv = mlx5_dev_to_eswitch_info(dev);
3611 return rte_flow_error_set(error, rte_errno,
3612 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3614 "failed to obtain E-Switch info");
3615 port_id = action->conf;
3616 port = port_id->original ? dev->data->port_id : port_id->id;
3617 act_priv = mlx5_port_to_eswitch_info(port, false);
3619 return rte_flow_error_set
3621 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
3622 "failed to obtain E-Switch port id for port");
3623 if (act_priv->domain_id != dev_priv->domain_id)
3624 return rte_flow_error_set
3626 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3627 "port does not belong to"
3628 " E-Switch being configured");
3633 * Get the maximum number of modify header actions.
3636 * Pointer to rte_eth_dev structure.
3638 * Flags bits to check if root level.
3641 * Max number of modify header actions device can support.
3643 static inline unsigned int
3644 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
3648 * There's no way to directly query the max capacity from FW.
3649 * The maximal value on root table should be assumed to be supported.
3651 if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
3652 return MLX5_MAX_MODIFY_NUM;
3654 return MLX5_ROOT_TBL_MODIFY_NUM;
3658 * Validate the meter action.
3661 * Pointer to rte_eth_dev structure.
3662 * @param[in] action_flags
3663 * Bit-fields that holds the actions detected until now.
3665 * Pointer to the meter action.
3667 * Attributes of flow that includes this action.
3669 * Pointer to error structure.
3672 * 0 on success, a negative errno value otherwise and rte_ernno is set.
3675 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
3676 uint64_t action_flags,
3677 const struct rte_flow_action *action,
3678 const struct rte_flow_attr *attr,
3679 struct rte_flow_error *error)
3681 struct mlx5_priv *priv = dev->data->dev_private;
3682 const struct rte_flow_action_meter *am = action->conf;
3683 struct mlx5_flow_meter *fm;
3686 return rte_flow_error_set(error, EINVAL,
3687 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3688 "meter action conf is NULL");
3690 if (action_flags & MLX5_FLOW_ACTION_METER)
3691 return rte_flow_error_set(error, ENOTSUP,
3692 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3693 "meter chaining not support");
3694 if (action_flags & MLX5_FLOW_ACTION_JUMP)
3695 return rte_flow_error_set(error, ENOTSUP,
3696 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3697 "meter with jump not support");
3699 return rte_flow_error_set(error, ENOTSUP,
3700 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3702 "meter action not supported");
3703 fm = mlx5_flow_meter_find(priv, am->mtr_id);
3705 return rte_flow_error_set(error, EINVAL,
3706 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3708 if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
3709 (!fm->ingress && !attr->ingress && attr->egress) ||
3710 (!fm->egress && !attr->egress && attr->ingress))))
3711 return rte_flow_error_set(error, EINVAL,
3712 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3713 "Flow attributes are either invalid "
3714 "or have a conflict with current "
3715 "meter attributes");
3720 * Validate the modify-header IPv4 DSCP actions.
3722 * @param[in] action_flags
3723 * Holds the actions detected until now.
3725 * Pointer to the modify action.
3726 * @param[in] item_flags
3727 * Holds the items detected.
3729 * Pointer to error structure.
3732 * 0 on success, a negative errno value otherwise and rte_errno is set.
3735 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
3736 const struct rte_flow_action *action,
3737 const uint64_t item_flags,
3738 struct rte_flow_error *error)
3742 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3744 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
3745 return rte_flow_error_set(error, EINVAL,
3746 RTE_FLOW_ERROR_TYPE_ACTION,
3748 "no ipv4 item in pattern");
3754 * Validate the modify-header IPv6 DSCP actions.
3756 * @param[in] action_flags
3757 * Holds the actions detected until now.
3759 * Pointer to the modify action.
3760 * @param[in] item_flags
3761 * Holds the items detected.
3763 * Pointer to error structure.
3766 * 0 on success, a negative errno value otherwise and rte_errno is set.
3769 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
3770 const struct rte_flow_action *action,
3771 const uint64_t item_flags,
3772 struct rte_flow_error *error)
3776 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3778 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
3779 return rte_flow_error_set(error, EINVAL,
3780 RTE_FLOW_ERROR_TYPE_ACTION,
3782 "no ipv6 item in pattern");
3788 * Find existing modify-header resource or create and register a new one.
3790 * @param dev[in, out]
3791 * Pointer to rte_eth_dev structure.
3792 * @param[in, out] resource
3793 * Pointer to modify-header resource.
3794 * @parm[in, out] dev_flow
3795 * Pointer to the dev_flow.
3797 * pointer to error structure.
3800 * 0 on success otherwise -errno and errno is set.
3803 flow_dv_modify_hdr_resource_register
3804 (struct rte_eth_dev *dev,
3805 struct mlx5_flow_dv_modify_hdr_resource *resource,
3806 struct mlx5_flow *dev_flow,
3807 struct rte_flow_error *error)
3809 struct mlx5_priv *priv = dev->data->dev_private;
3810 struct mlx5_ibv_shared *sh = priv->sh;
3811 struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
3812 struct mlx5dv_dr_domain *ns;
3813 uint32_t actions_len;
3815 resource->flags = dev_flow->dv.group ? 0 :
3816 MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
3817 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
3819 return rte_flow_error_set(error, EOVERFLOW,
3820 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3821 "too many modify header items");
3822 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3823 ns = sh->fdb_domain;
3824 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
3828 /* Lookup a matching resource from cache. */
3829 actions_len = resource->actions_num * sizeof(resource->actions[0]);
3830 LIST_FOREACH(cache_resource, &sh->modify_cmds, next) {
3831 if (resource->ft_type == cache_resource->ft_type &&
3832 resource->actions_num == cache_resource->actions_num &&
3833 resource->flags == cache_resource->flags &&
3834 !memcmp((const void *)resource->actions,
3835 (const void *)cache_resource->actions,
3837 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
3838 (void *)cache_resource,
3839 rte_atomic32_read(&cache_resource->refcnt));
3840 rte_atomic32_inc(&cache_resource->refcnt);
3841 dev_flow->handle->dvh.modify_hdr = cache_resource;
3845 /* Register new modify-header resource. */
3846 cache_resource = rte_calloc(__func__, 1,
3847 sizeof(*cache_resource) + actions_len, 0);
3848 if (!cache_resource)
3849 return rte_flow_error_set(error, ENOMEM,
3850 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3851 "cannot allocate resource memory");
3852 *cache_resource = *resource;
3853 rte_memcpy(cache_resource->actions, resource->actions, actions_len);
3854 cache_resource->verbs_action =
3855 mlx5_glue->dv_create_flow_action_modify_header
3856 (sh->ctx, cache_resource->ft_type, ns,
3857 cache_resource->flags, actions_len,
3858 (uint64_t *)cache_resource->actions);
3859 if (!cache_resource->verbs_action) {
3860 rte_free(cache_resource);
3861 return rte_flow_error_set(error, ENOMEM,
3862 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3863 NULL, "cannot create action");
3865 rte_atomic32_init(&cache_resource->refcnt);
3866 rte_atomic32_inc(&cache_resource->refcnt);
3867 LIST_INSERT_HEAD(&sh->modify_cmds, cache_resource, next);
3868 dev_flow->handle->dvh.modify_hdr = cache_resource;
3869 DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
3870 (void *)cache_resource,
3871 rte_atomic32_read(&cache_resource->refcnt));
3876 * Get DV flow counter by index.
3879 * Pointer to the Ethernet device structure.
3881 * mlx5 flow counter index in the container.
3883 * mlx5 flow counter pool in the container,
3886 * Pointer to the counter, NULL otherwise.
3888 static struct mlx5_flow_counter *
3889 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
3891 struct mlx5_flow_counter_pool **ppool)
3893 struct mlx5_priv *priv = dev->data->dev_private;
3894 struct mlx5_pools_container *cont;
3895 struct mlx5_flow_counter_pool *pool;
3899 if (idx >= MLX5_CNT_BATCH_OFFSET) {
3900 idx -= MLX5_CNT_BATCH_OFFSET;
3903 cont = MLX5_CNT_CONTAINER(priv->sh, batch, 0);
3904 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cont->n);
3905 pool = cont->pools[idx / MLX5_COUNTERS_PER_POOL];
3909 return &pool->counters_raw[idx % MLX5_COUNTERS_PER_POOL];
3913 * Get a pool by devx counter ID.
3916 * Pointer to the counter container.
3918 * The counter devx ID.
3921 * The counter pool pointer if exists, NULL otherwise,
3923 static struct mlx5_flow_counter_pool *
3924 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
3927 uint32_t n_valid = rte_atomic16_read(&cont->n_valid);
3929 for (i = 0; i < n_valid; i++) {
3930 struct mlx5_flow_counter_pool *pool = cont->pools[i];
3931 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
3932 MLX5_COUNTERS_PER_POOL;
3934 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL) {
3936 * Move the pool to the head, as counter allocate
3937 * always gets the first pool in the container.
3939 if (pool != TAILQ_FIRST(&cont->pool_list)) {
3940 TAILQ_REMOVE(&cont->pool_list, pool, next);
3941 TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
3950 * Allocate a new memory for the counter values wrapped by all the needed
3954 * Pointer to the Ethernet device structure.
3956 * The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
3959 * The new memory management pointer on success, otherwise NULL and rte_errno
3962 static struct mlx5_counter_stats_mem_mng *
3963 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
3965 struct mlx5_ibv_shared *sh = ((struct mlx5_priv *)
3966 (dev->data->dev_private))->sh;
3967 struct mlx5_devx_mkey_attr mkey_attr;
3968 struct mlx5_counter_stats_mem_mng *mem_mng;
3969 volatile struct flow_counter_stats *raw_data;
3970 int size = (sizeof(struct flow_counter_stats) *
3971 MLX5_COUNTERS_PER_POOL +
3972 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
3973 sizeof(struct mlx5_counter_stats_mem_mng);
3974 uint8_t *mem = rte_calloc(__func__, 1, size, sysconf(_SC_PAGESIZE));
3981 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
3982 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
3983 mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
3984 IBV_ACCESS_LOCAL_WRITE);
3985 if (!mem_mng->umem) {
3990 mkey_attr.addr = (uintptr_t)mem;
3991 mkey_attr.size = size;
3992 mkey_attr.umem_id = mem_mng->umem->umem_id;
3993 mkey_attr.pd = sh->pdn;
3994 mkey_attr.log_entity_size = 0;
3995 mkey_attr.pg_access = 0;
3996 mkey_attr.klm_array = NULL;
3997 mkey_attr.klm_num = 0;
3998 mkey_attr.relaxed_ordering = 1;
3999 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
4001 mlx5_glue->devx_umem_dereg(mem_mng->umem);
4006 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
4007 raw_data = (volatile struct flow_counter_stats *)mem;
4008 for (i = 0; i < raws_n; ++i) {
4009 mem_mng->raws[i].mem_mng = mem_mng;
4010 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
4012 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
4017 * Resize a counter container.
4020 * Pointer to the Ethernet device structure.
4022 * Whether the pool is for counter that was allocated by batch command.
4025 * The new container pointer on success, otherwise NULL and rte_errno is set.
4027 static struct mlx5_pools_container *
4028 flow_dv_container_resize(struct rte_eth_dev *dev, uint32_t batch)
4030 struct mlx5_priv *priv = dev->data->dev_private;
4031 struct mlx5_pools_container *cont =
4032 MLX5_CNT_CONTAINER(priv->sh, batch, 0);
4033 struct mlx5_pools_container *new_cont =
4034 MLX5_CNT_CONTAINER_UNUSED(priv->sh, batch, 0);
4035 struct mlx5_counter_stats_mem_mng *mem_mng = NULL;
4036 uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
4037 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4040 /* Fallback mode has no background thread. Skip the check. */
4041 if (!priv->counter_fallback &&
4042 cont != MLX5_CNT_CONTAINER(priv->sh, batch, 1)) {
4043 /* The last resize still hasn't detected by the host thread. */
4047 new_cont->pools = rte_calloc(__func__, 1, mem_size, 0);
4048 if (!new_cont->pools) {
4053 memcpy(new_cont->pools, cont->pools, cont->n *
4054 sizeof(struct mlx5_flow_counter_pool *));
4056 * Fallback mode query the counter directly, no background query
4057 * resources are needed.
4059 if (!priv->counter_fallback) {
4060 mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
4061 MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES);
4063 rte_free(new_cont->pools);
4066 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
4067 LIST_INSERT_HEAD(&priv->sh->cmng.free_stat_raws,
4069 MLX5_CNT_CONTAINER_RESIZE +
4073 * Release the old container pools directly as no background
4074 * thread helps that.
4076 rte_free(cont->pools);
4078 new_cont->n = resize;
4079 rte_atomic16_set(&new_cont->n_valid, rte_atomic16_read(&cont->n_valid));
4080 TAILQ_INIT(&new_cont->pool_list);
4081 TAILQ_CONCAT(&new_cont->pool_list, &cont->pool_list, next);
4082 new_cont->init_mem_mng = mem_mng;
4084 /* Flip the master container. */
4085 priv->sh->cmng.mhi[batch] ^= (uint8_t)1;
4090 * Query a devx flow counter.
4093 * Pointer to the Ethernet device structure.
4095 * Index to the flow counter.
4097 * The statistics value of packets.
4099 * The statistics value of bytes.
4102 * 0 on success, otherwise a negative errno value and rte_errno is set.
4105 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4108 struct mlx5_priv *priv = dev->data->dev_private;
4109 struct mlx5_flow_counter_pool *pool = NULL;
4110 struct mlx5_flow_counter *cnt;
4111 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4114 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4116 if (counter < MLX5_CNT_BATCH_OFFSET) {
4117 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4118 if (priv->counter_fallback)
4119 return mlx5_devx_cmd_flow_counter_query(cnt_ext->dcs, 0,
4120 0, pkts, bytes, 0, NULL, NULL, 0);
4123 rte_spinlock_lock(&pool->sl);
4125 * The single counters allocation may allocate smaller ID than the
4126 * current allocated in parallel to the host reading.
4127 * In this case the new counter values must be reported as 0.
4129 if (unlikely(cnt_ext && cnt_ext->dcs->id < pool->raw->min_dcs_id)) {
4133 offset = cnt - &pool->counters_raw[0];
4134 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4135 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4137 rte_spinlock_unlock(&pool->sl);
4142 * Create and initialize a new counter pool.
4145 * Pointer to the Ethernet device structure.
4147 * The devX counter handle.
4149 * Whether the pool is for counter that was allocated by batch command.
4150 * @param[in/out] cont_cur
4151 * Pointer to the container pointer, it will be update in pool resize.
4154 * The pool container pointer on success, NULL otherwise and rte_errno is set.
4156 static struct mlx5_pools_container *
4157 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4160 struct mlx5_priv *priv = dev->data->dev_private;
4161 struct mlx5_flow_counter_pool *pool;
4162 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4164 int16_t n_valid = rte_atomic16_read(&cont->n_valid);
4167 if (cont->n == n_valid) {
4168 cont = flow_dv_container_resize(dev, batch);
4172 size = sizeof(*pool);
4174 size += MLX5_COUNTERS_PER_POOL *
4175 sizeof(struct mlx5_flow_counter_ext);
4176 pool = rte_calloc(__func__, 1, size, 0);
4181 pool->min_dcs = dcs;
4182 if (!priv->counter_fallback)
4183 pool->raw = cont->init_mem_mng->raws + n_valid %
4184 MLX5_CNT_CONTAINER_RESIZE;
4185 pool->raw_hw = NULL;
4186 rte_spinlock_init(&pool->sl);
4188 * The generation of the new allocated counters in this pool is 0, 2 in
4189 * the pool generation makes all the counters valid for allocation.
4190 * The start and end query generation protect the counters be released
4191 * between the query and update gap period will not be reallocated
4192 * without the last query finished and stats updated to the memory.
4194 rte_atomic64_set(&pool->start_query_gen, 0x2);
4196 * There's no background query thread for fallback mode, set the
4197 * end_query_gen to the maximum value since no need to wait for
4198 * statistics update.
4200 rte_atomic64_set(&pool->end_query_gen, priv->counter_fallback ?
4202 TAILQ_INIT(&pool->counters);
4203 TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
4204 pool->index = n_valid;
4205 cont->pools[n_valid] = pool;
4206 /* Pool initialization must be updated before host thread access. */
4208 rte_atomic16_add(&cont->n_valid, 1);
4213 * Prepare a new counter and/or a new counter pool.
4216 * Pointer to the Ethernet device structure.
4217 * @param[out] cnt_free
4218 * Where to put the pointer of a new counter.
4220 * Whether the pool is for counter that was allocated by batch command.
4223 * The counter container pointer and @p cnt_free is set on success,
4224 * NULL otherwise and rte_errno is set.
4226 static struct mlx5_pools_container *
4227 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4228 struct mlx5_flow_counter **cnt_free,
4231 struct mlx5_priv *priv = dev->data->dev_private;
4232 struct mlx5_pools_container *cont;
4233 struct mlx5_flow_counter_pool *pool;
4234 struct mlx5_devx_obj *dcs = NULL;
4235 struct mlx5_flow_counter *cnt;
4238 cont = MLX5_CNT_CONTAINER(priv->sh, batch, 0);
4240 /* bulk_bitmap must be 0 for single counter allocation. */
4241 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4244 pool = flow_dv_find_pool_by_id(cont, dcs->id);
4246 cont = flow_dv_pool_create(dev, dcs, batch);
4248 mlx5_devx_cmd_destroy(dcs);
4251 pool = TAILQ_FIRST(&cont->pool_list);
4252 } else if (dcs->id < pool->min_dcs->id) {
4253 rte_atomic64_set(&pool->a64_dcs,
4254 (int64_t)(uintptr_t)dcs);
4256 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4257 cnt = &pool->counters_raw[i];
4258 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
4259 MLX5_GET_POOL_CNT_EXT(pool, i)->dcs = dcs;
4263 /* bulk_bitmap is in 128 counters units. */
4264 if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
4265 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4267 rte_errno = ENODATA;
4270 cont = flow_dv_pool_create(dev, dcs, batch);
4272 mlx5_devx_cmd_destroy(dcs);
4275 pool = TAILQ_FIRST(&cont->pool_list);
4276 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
4277 cnt = &pool->counters_raw[i];
4278 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
4280 *cnt_free = &pool->counters_raw[0];
4285 * Search for existed shared counter.
4288 * Pointer to the relevant counter pool container.
4290 * The shared counter ID to search.
4292 * mlx5 flow counter pool in the container,
4295 * NULL if not existed, otherwise pointer to the shared extend counter.
4297 static struct mlx5_flow_counter_ext *
4298 flow_dv_counter_shared_search(struct mlx5_pools_container *cont, uint32_t id,
4299 struct mlx5_flow_counter_pool **ppool)
4301 static struct mlx5_flow_counter_ext *cnt;
4302 struct mlx5_flow_counter_pool *pool;
4304 uint32_t n_valid = rte_atomic16_read(&cont->n_valid);
4306 for (i = 0; i < n_valid; i++) {
4307 pool = cont->pools[i];
4308 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
4309 cnt = MLX5_GET_POOL_CNT_EXT(pool, i);
4310 if (cnt->ref_cnt && cnt->shared && cnt->id == id) {
4312 *ppool = cont->pools[i];
4321 * Allocate a flow counter.
4324 * Pointer to the Ethernet device structure.
4326 * Indicate if this counter is shared with other flows.
4328 * Counter identifier.
4330 * Counter flow group.
4333 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4336 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
4339 struct mlx5_priv *priv = dev->data->dev_private;
4340 struct mlx5_flow_counter_pool *pool = NULL;
4341 struct mlx5_flow_counter *cnt_free = NULL;
4342 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4344 * Currently group 0 flow counter cannot be assigned to a flow if it is
4345 * not the first one in the batch counter allocation, so it is better
4346 * to allocate counters one by one for these flows in a separate
4348 * A counter can be shared between different groups so need to take
4349 * shared counters from the single container.
4351 uint32_t batch = (group && !shared && !priv->counter_fallback) ? 1 : 0;
4352 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4356 if (!priv->config.devx) {
4357 rte_errno = ENOTSUP;
4361 cnt_ext = flow_dv_counter_shared_search(cont, id, &pool);
4363 if (cnt_ext->ref_cnt + 1 == 0) {
4368 cnt_idx = pool->index * MLX5_COUNTERS_PER_POOL +
4369 (cnt_ext->dcs->id % MLX5_COUNTERS_PER_POOL)
4374 /* Pools which has a free counters are in the start. */
4375 TAILQ_FOREACH(pool, &cont->pool_list, next) {
4377 * The free counter reset values must be updated between the
4378 * counter release to the counter allocation, so, at least one
4379 * query must be done in this time. ensure it by saving the
4380 * query generation in the release time.
4381 * The free list is sorted according to the generation - so if
4382 * the first one is not updated, all the others are not
4385 cnt_free = TAILQ_FIRST(&pool->counters);
4386 if (cnt_free && cnt_free->query_gen <
4387 rte_atomic64_read(&pool->end_query_gen))
4392 cont = flow_dv_counter_pool_prepare(dev, &cnt_free, batch);
4395 pool = TAILQ_FIRST(&cont->pool_list);
4398 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt_free);
4399 /* Create a DV counter action only in the first time usage. */
4400 if (!cnt_free->action) {
4402 struct mlx5_devx_obj *dcs;
4405 offset = cnt_free - &pool->counters_raw[0];
4406 dcs = pool->min_dcs;
4411 cnt_free->action = mlx5_glue->dv_create_flow_action_counter
4413 if (!cnt_free->action) {
4418 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4419 (cnt_free - pool->counters_raw));
4420 cnt_idx += batch * MLX5_CNT_BATCH_OFFSET;
4421 /* Update the counter reset values. */
4422 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4426 cnt_ext->shared = shared;
4427 cnt_ext->ref_cnt = 1;
4430 if (!priv->counter_fallback && !priv->sh->cmng.query_thread_on)
4431 /* Start the asynchronous batch query by the host thread. */
4432 mlx5_set_query_alarm(priv->sh);
4433 TAILQ_REMOVE(&pool->counters, cnt_free, next);
4434 if (TAILQ_EMPTY(&pool->counters)) {
4435 /* Move the pool to the end of the container pool list. */
4436 TAILQ_REMOVE(&cont->pool_list, pool, next);
4437 TAILQ_INSERT_TAIL(&cont->pool_list, pool, next);
4443 * Release a flow counter.
4446 * Pointer to the Ethernet device structure.
4447 * @param[in] counter
4448 * Index to the counter handler.
4451 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
4453 struct mlx5_flow_counter_pool *pool = NULL;
4454 struct mlx5_flow_counter *cnt;
4455 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4459 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4461 if (counter < MLX5_CNT_BATCH_OFFSET) {
4462 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4463 if (cnt_ext && --cnt_ext->ref_cnt)
4466 /* Put the counter in the end - the last updated one. */
4467 TAILQ_INSERT_TAIL(&pool->counters, cnt, next);
4469 * Counters released between query trigger and handler need
4470 * to wait the next round of query. Since the packets arrive
4471 * in the gap period will not be taken into account to the
4474 cnt->query_gen = rte_atomic64_read(&pool->start_query_gen);
4478 * Verify the @p attributes will be correctly understood by the NIC and store
4479 * them in the @p flow if everything is correct.
4482 * Pointer to dev struct.
4483 * @param[in] attributes
4484 * Pointer to flow attributes
4485 * @param[in] external
4486 * This flow rule is created by request external to PMD.
4488 * Pointer to error structure.
4491 * - 0 on success and non root table.
4492 * - 1 on success and root table.
4493 * - 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;
4505 #ifndef HAVE_MLX5DV_DR
4506 if (attributes->group)
4507 return rte_flow_error_set(error, ENOTSUP,
4508 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
4510 "groups are not supported");
4514 ret = mlx5_flow_group_to_table(attributes, external,
4515 attributes->group, !!priv->fdb_def_rule,
4520 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4522 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
4523 attributes->priority >= priority_max)
4524 return rte_flow_error_set(error, ENOTSUP,
4525 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
4527 "priority out of range");
4528 if (attributes->transfer) {
4529 if (!priv->config.dv_esw_en)
4530 return rte_flow_error_set
4532 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4533 "E-Switch dr is not supported");
4534 if (!(priv->representor || priv->master))
4535 return rte_flow_error_set
4536 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4537 NULL, "E-Switch configuration can only be"
4538 " done by a master or a representor device");
4539 if (attributes->egress)
4540 return rte_flow_error_set
4542 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
4543 "egress is not supported");
4545 if (!(attributes->egress ^ attributes->ingress))
4546 return rte_flow_error_set(error, ENOTSUP,
4547 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
4548 "must specify exactly one of "
4549 "ingress or egress");
4554 * Internal validation function. For validating both actions and items.
4557 * Pointer to the rte_eth_dev structure.
4559 * Pointer to the flow attributes.
4561 * Pointer to the list of items.
4562 * @param[in] actions
4563 * Pointer to the list of actions.
4564 * @param[in] external
4565 * This flow rule is created by request external to PMD.
4566 * @param[in] hairpin
4567 * Number of hairpin TX actions, 0 means classic flow.
4569 * Pointer to the error structure.
4572 * 0 on success, a negative errno value otherwise and rte_errno is set.
4575 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
4576 const struct rte_flow_item items[],
4577 const struct rte_flow_action actions[],
4578 bool external, int hairpin, struct rte_flow_error *error)
4581 uint64_t action_flags = 0;
4582 uint64_t item_flags = 0;
4583 uint64_t last_item = 0;
4584 uint8_t next_protocol = 0xff;
4585 uint16_t ether_type = 0;
4587 uint8_t item_ipv6_proto = 0;
4588 const struct rte_flow_item *gre_item = NULL;
4589 const struct rte_flow_action_raw_decap *decap;
4590 const struct rte_flow_action_raw_encap *encap;
4591 const struct rte_flow_action_rss *rss;
4592 const struct rte_flow_item_tcp nic_tcp_mask = {
4595 .src_port = RTE_BE16(UINT16_MAX),
4596 .dst_port = RTE_BE16(UINT16_MAX),
4599 const struct rte_flow_item_ipv4 nic_ipv4_mask = {
4601 .src_addr = RTE_BE32(0xffffffff),
4602 .dst_addr = RTE_BE32(0xffffffff),
4603 .type_of_service = 0xff,
4604 .next_proto_id = 0xff,
4605 .time_to_live = 0xff,
4608 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
4611 "\xff\xff\xff\xff\xff\xff\xff\xff"
4612 "\xff\xff\xff\xff\xff\xff\xff\xff",
4614 "\xff\xff\xff\xff\xff\xff\xff\xff"
4615 "\xff\xff\xff\xff\xff\xff\xff\xff",
4616 .vtc_flow = RTE_BE32(0xffffffff),
4621 struct mlx5_priv *priv = dev->data->dev_private;
4622 struct mlx5_dev_config *dev_conf = &priv->config;
4623 uint16_t queue_index = 0xFFFF;
4624 const struct rte_flow_item_vlan *vlan_m = NULL;
4625 int16_t rw_act_num = 0;
4630 ret = flow_dv_validate_attributes(dev, attr, external, error);
4633 is_root = (uint64_t)ret;
4634 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4635 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
4636 int type = items->type;
4639 case RTE_FLOW_ITEM_TYPE_VOID:
4641 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4642 ret = flow_dv_validate_item_port_id
4643 (dev, items, attr, item_flags, error);
4646 last_item = MLX5_FLOW_ITEM_PORT_ID;
4648 case RTE_FLOW_ITEM_TYPE_ETH:
4649 ret = mlx5_flow_validate_item_eth(items, item_flags,
4653 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
4654 MLX5_FLOW_LAYER_OUTER_L2;
4655 if (items->mask != NULL && items->spec != NULL) {
4657 ((const struct rte_flow_item_eth *)
4660 ((const struct rte_flow_item_eth *)
4662 ether_type = rte_be_to_cpu_16(ether_type);
4667 case RTE_FLOW_ITEM_TYPE_VLAN:
4668 ret = mlx5_flow_validate_item_vlan(items, item_flags,
4672 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
4673 MLX5_FLOW_LAYER_OUTER_VLAN;
4674 if (items->mask != NULL && items->spec != NULL) {
4676 ((const struct rte_flow_item_vlan *)
4677 items->spec)->inner_type;
4679 ((const struct rte_flow_item_vlan *)
4680 items->mask)->inner_type;
4681 ether_type = rte_be_to_cpu_16(ether_type);
4685 /* Store outer VLAN mask for of_push_vlan action. */
4687 vlan_m = items->mask;
4689 case RTE_FLOW_ITEM_TYPE_IPV4:
4690 mlx5_flow_tunnel_ip_check(items, next_protocol,
4691 &item_flags, &tunnel);
4692 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
4699 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
4700 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4701 if (items->mask != NULL &&
4702 ((const struct rte_flow_item_ipv4 *)
4703 items->mask)->hdr.next_proto_id) {
4705 ((const struct rte_flow_item_ipv4 *)
4706 (items->spec))->hdr.next_proto_id;
4708 ((const struct rte_flow_item_ipv4 *)
4709 (items->mask))->hdr.next_proto_id;
4711 /* Reset for inner layer. */
4712 next_protocol = 0xff;
4715 case RTE_FLOW_ITEM_TYPE_IPV6:
4716 mlx5_flow_tunnel_ip_check(items, next_protocol,
4717 &item_flags, &tunnel);
4718 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
4725 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
4726 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4727 if (items->mask != NULL &&
4728 ((const struct rte_flow_item_ipv6 *)
4729 items->mask)->hdr.proto) {
4731 ((const struct rte_flow_item_ipv6 *)
4732 items->spec)->hdr.proto;
4734 ((const struct rte_flow_item_ipv6 *)
4735 items->spec)->hdr.proto;
4737 ((const struct rte_flow_item_ipv6 *)
4738 items->mask)->hdr.proto;
4740 /* Reset for inner layer. */
4741 next_protocol = 0xff;
4744 case RTE_FLOW_ITEM_TYPE_TCP:
4745 ret = mlx5_flow_validate_item_tcp
4752 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
4753 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4755 case RTE_FLOW_ITEM_TYPE_UDP:
4756 ret = mlx5_flow_validate_item_udp(items, item_flags,
4761 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
4762 MLX5_FLOW_LAYER_OUTER_L4_UDP;
4764 case RTE_FLOW_ITEM_TYPE_GRE:
4765 ret = mlx5_flow_validate_item_gre(items, item_flags,
4766 next_protocol, error);
4770 last_item = MLX5_FLOW_LAYER_GRE;
4772 case RTE_FLOW_ITEM_TYPE_NVGRE:
4773 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
4778 last_item = MLX5_FLOW_LAYER_NVGRE;
4780 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
4781 ret = mlx5_flow_validate_item_gre_key
4782 (items, item_flags, gre_item, error);
4785 last_item = MLX5_FLOW_LAYER_GRE_KEY;
4787 case RTE_FLOW_ITEM_TYPE_VXLAN:
4788 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
4792 last_item = MLX5_FLOW_LAYER_VXLAN;
4794 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
4795 ret = mlx5_flow_validate_item_vxlan_gpe(items,
4800 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
4802 case RTE_FLOW_ITEM_TYPE_GENEVE:
4803 ret = mlx5_flow_validate_item_geneve(items,
4808 last_item = MLX5_FLOW_LAYER_GENEVE;
4810 case RTE_FLOW_ITEM_TYPE_MPLS:
4811 ret = mlx5_flow_validate_item_mpls(dev, items,
4816 last_item = MLX5_FLOW_LAYER_MPLS;
4819 case RTE_FLOW_ITEM_TYPE_MARK:
4820 ret = flow_dv_validate_item_mark(dev, items, attr,
4824 last_item = MLX5_FLOW_ITEM_MARK;
4826 case RTE_FLOW_ITEM_TYPE_META:
4827 ret = flow_dv_validate_item_meta(dev, items, attr,
4831 last_item = MLX5_FLOW_ITEM_METADATA;
4833 case RTE_FLOW_ITEM_TYPE_ICMP:
4834 ret = mlx5_flow_validate_item_icmp(items, item_flags,
4839 last_item = MLX5_FLOW_LAYER_ICMP;
4841 case RTE_FLOW_ITEM_TYPE_ICMP6:
4842 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
4847 item_ipv6_proto = IPPROTO_ICMPV6;
4848 last_item = MLX5_FLOW_LAYER_ICMP6;
4850 case RTE_FLOW_ITEM_TYPE_TAG:
4851 ret = flow_dv_validate_item_tag(dev, items,
4855 last_item = MLX5_FLOW_ITEM_TAG;
4857 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
4858 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
4860 case RTE_FLOW_ITEM_TYPE_GTP:
4861 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
4865 last_item = MLX5_FLOW_LAYER_GTP;
4868 return rte_flow_error_set(error, ENOTSUP,
4869 RTE_FLOW_ERROR_TYPE_ITEM,
4870 NULL, "item not supported");
4872 item_flags |= last_item;
4874 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4875 int type = actions->type;
4876 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
4877 return rte_flow_error_set(error, ENOTSUP,
4878 RTE_FLOW_ERROR_TYPE_ACTION,
4879 actions, "too many actions");
4881 case RTE_FLOW_ACTION_TYPE_VOID:
4883 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4884 ret = flow_dv_validate_action_port_id(dev,
4891 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4894 case RTE_FLOW_ACTION_TYPE_FLAG:
4895 ret = flow_dv_validate_action_flag(dev, action_flags,
4899 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
4900 /* Count all modify-header actions as one. */
4901 if (!(action_flags &
4902 MLX5_FLOW_MODIFY_HDR_ACTIONS))
4904 action_flags |= MLX5_FLOW_ACTION_FLAG |
4905 MLX5_FLOW_ACTION_MARK_EXT;
4907 action_flags |= MLX5_FLOW_ACTION_FLAG;
4910 rw_act_num += MLX5_ACT_NUM_SET_MARK;
4912 case RTE_FLOW_ACTION_TYPE_MARK:
4913 ret = flow_dv_validate_action_mark(dev, actions,
4918 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
4919 /* Count all modify-header actions as one. */
4920 if (!(action_flags &
4921 MLX5_FLOW_MODIFY_HDR_ACTIONS))
4923 action_flags |= MLX5_FLOW_ACTION_MARK |
4924 MLX5_FLOW_ACTION_MARK_EXT;
4926 action_flags |= MLX5_FLOW_ACTION_MARK;
4929 rw_act_num += MLX5_ACT_NUM_SET_MARK;
4931 case RTE_FLOW_ACTION_TYPE_SET_META:
4932 ret = flow_dv_validate_action_set_meta(dev, actions,
4937 /* Count all modify-header actions as one action. */
4938 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
4940 action_flags |= MLX5_FLOW_ACTION_SET_META;
4941 rw_act_num += MLX5_ACT_NUM_SET_META;
4943 case RTE_FLOW_ACTION_TYPE_SET_TAG:
4944 ret = flow_dv_validate_action_set_tag(dev, actions,
4949 /* Count all modify-header actions as one action. */
4950 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
4952 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
4953 rw_act_num += MLX5_ACT_NUM_SET_TAG;
4955 case RTE_FLOW_ACTION_TYPE_DROP:
4956 ret = mlx5_flow_validate_action_drop(action_flags,
4960 action_flags |= MLX5_FLOW_ACTION_DROP;
4963 case RTE_FLOW_ACTION_TYPE_QUEUE:
4964 ret = mlx5_flow_validate_action_queue(actions,
4969 queue_index = ((const struct rte_flow_action_queue *)
4970 (actions->conf))->index;
4971 action_flags |= MLX5_FLOW_ACTION_QUEUE;
4974 case RTE_FLOW_ACTION_TYPE_RSS:
4975 rss = actions->conf;
4976 ret = mlx5_flow_validate_action_rss(actions,
4982 if (rss != NULL && rss->queue_num)
4983 queue_index = rss->queue[0];
4984 action_flags |= MLX5_FLOW_ACTION_RSS;
4987 case RTE_FLOW_ACTION_TYPE_COUNT:
4988 ret = flow_dv_validate_action_count(dev, error);
4991 action_flags |= MLX5_FLOW_ACTION_COUNT;
4994 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4995 if (flow_dv_validate_action_pop_vlan(dev,
5001 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5004 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5005 ret = flow_dv_validate_action_push_vlan(dev,
5012 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5015 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5016 ret = flow_dv_validate_action_set_vlan_pcp
5017 (action_flags, actions, error);
5020 /* Count PCP with push_vlan command. */
5021 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5023 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5024 ret = flow_dv_validate_action_set_vlan_vid
5025 (item_flags, action_flags,
5029 /* Count VID with push_vlan command. */
5030 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5031 rw_act_num += MLX5_ACT_NUM_MDF_VID;
5033 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5034 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5035 ret = flow_dv_validate_action_l2_encap(dev,
5041 action_flags |= MLX5_FLOW_ACTION_ENCAP;
5044 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5045 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5046 ret = flow_dv_validate_action_decap(dev, action_flags,
5050 action_flags |= MLX5_FLOW_ACTION_DECAP;
5053 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5054 ret = flow_dv_validate_action_raw_encap_decap
5055 (dev, NULL, actions->conf, attr, &action_flags,
5060 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5061 decap = actions->conf;
5062 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5064 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5068 encap = actions->conf;
5070 ret = flow_dv_validate_action_raw_encap_decap
5072 decap ? decap : &empty_decap, encap,
5073 attr, &action_flags, &actions_n,
5078 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5079 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5080 ret = flow_dv_validate_action_modify_mac(action_flags,
5086 /* Count all modify-header actions as one action. */
5087 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5089 action_flags |= actions->type ==
5090 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5091 MLX5_FLOW_ACTION_SET_MAC_SRC :
5092 MLX5_FLOW_ACTION_SET_MAC_DST;
5094 * Even if the source and destination MAC addresses have
5095 * overlap in the header with 4B alignment, the convert
5096 * function will handle them separately and 4 SW actions
5097 * will be created. And 2 actions will be added each
5098 * time no matter how many bytes of address will be set.
5100 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5102 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5103 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5104 ret = flow_dv_validate_action_modify_ipv4(action_flags,
5110 /* Count all modify-header actions as one action. */
5111 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5113 action_flags |= actions->type ==
5114 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5115 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5116 MLX5_FLOW_ACTION_SET_IPV4_DST;
5117 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5119 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5120 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5121 ret = flow_dv_validate_action_modify_ipv6(action_flags,
5127 if (item_ipv6_proto == IPPROTO_ICMPV6)
5128 return rte_flow_error_set(error, ENOTSUP,
5129 RTE_FLOW_ERROR_TYPE_ACTION,
5131 "Can't change header "
5132 "with ICMPv6 proto");
5133 /* Count all modify-header actions as one action. */
5134 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5136 action_flags |= actions->type ==
5137 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5138 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5139 MLX5_FLOW_ACTION_SET_IPV6_DST;
5140 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5142 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5143 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5144 ret = flow_dv_validate_action_modify_tp(action_flags,
5150 /* Count all modify-header actions as one action. */
5151 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5153 action_flags |= actions->type ==
5154 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5155 MLX5_FLOW_ACTION_SET_TP_SRC :
5156 MLX5_FLOW_ACTION_SET_TP_DST;
5157 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5159 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5160 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5161 ret = flow_dv_validate_action_modify_ttl(action_flags,
5167 /* Count all modify-header actions as one action. */
5168 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5170 action_flags |= actions->type ==
5171 RTE_FLOW_ACTION_TYPE_SET_TTL ?
5172 MLX5_FLOW_ACTION_SET_TTL :
5173 MLX5_FLOW_ACTION_DEC_TTL;
5174 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5176 case RTE_FLOW_ACTION_TYPE_JUMP:
5177 ret = flow_dv_validate_action_jump(actions,
5184 action_flags |= MLX5_FLOW_ACTION_JUMP;
5186 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5187 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5188 ret = flow_dv_validate_action_modify_tcp_seq
5195 /* Count all modify-header actions as one action. */
5196 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5198 action_flags |= actions->type ==
5199 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5200 MLX5_FLOW_ACTION_INC_TCP_SEQ :
5201 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5202 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5204 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5205 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5206 ret = flow_dv_validate_action_modify_tcp_ack
5213 /* Count all modify-header actions as one action. */
5214 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5216 action_flags |= actions->type ==
5217 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5218 MLX5_FLOW_ACTION_INC_TCP_ACK :
5219 MLX5_FLOW_ACTION_DEC_TCP_ACK;
5220 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5222 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5224 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5225 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5226 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5228 case RTE_FLOW_ACTION_TYPE_METER:
5229 ret = mlx5_flow_validate_action_meter(dev,
5235 action_flags |= MLX5_FLOW_ACTION_METER;
5237 /* Meter action will add one more TAG action. */
5238 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5240 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5241 ret = flow_dv_validate_action_modify_ipv4_dscp
5248 /* Count all modify-header actions as one action. */
5249 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5251 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5252 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5254 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5255 ret = flow_dv_validate_action_modify_ipv6_dscp
5262 /* Count all modify-header actions as one action. */
5263 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5265 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5266 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5269 return rte_flow_error_set(error, ENOTSUP,
5270 RTE_FLOW_ERROR_TYPE_ACTION,
5272 "action not supported");
5276 * Validate the drop action mutual exclusion with other actions.
5277 * Drop action is mutually-exclusive with any other action, except for
5280 if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
5281 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
5282 return rte_flow_error_set(error, EINVAL,
5283 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5284 "Drop action is mutually-exclusive "
5285 "with any other action, except for "
5287 /* Eswitch has few restrictions on using items and actions */
5288 if (attr->transfer) {
5289 if (!mlx5_flow_ext_mreg_supported(dev) &&
5290 action_flags & MLX5_FLOW_ACTION_FLAG)
5291 return rte_flow_error_set(error, ENOTSUP,
5292 RTE_FLOW_ERROR_TYPE_ACTION,
5294 "unsupported action FLAG");
5295 if (!mlx5_flow_ext_mreg_supported(dev) &&
5296 action_flags & MLX5_FLOW_ACTION_MARK)
5297 return rte_flow_error_set(error, ENOTSUP,
5298 RTE_FLOW_ERROR_TYPE_ACTION,
5300 "unsupported action MARK");
5301 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
5302 return rte_flow_error_set(error, ENOTSUP,
5303 RTE_FLOW_ERROR_TYPE_ACTION,
5305 "unsupported action QUEUE");
5306 if (action_flags & MLX5_FLOW_ACTION_RSS)
5307 return rte_flow_error_set(error, ENOTSUP,
5308 RTE_FLOW_ERROR_TYPE_ACTION,
5310 "unsupported action RSS");
5311 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
5312 return rte_flow_error_set(error, EINVAL,
5313 RTE_FLOW_ERROR_TYPE_ACTION,
5315 "no fate action is found");
5317 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
5318 return rte_flow_error_set(error, EINVAL,
5319 RTE_FLOW_ERROR_TYPE_ACTION,
5321 "no fate action is found");
5323 /* Continue validation for Xcap actions.*/
5324 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) && (queue_index == 0xFFFF ||
5325 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
5326 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
5327 MLX5_FLOW_XCAP_ACTIONS)
5328 return rte_flow_error_set(error, ENOTSUP,
5329 RTE_FLOW_ERROR_TYPE_ACTION,
5330 NULL, "encap and decap "
5331 "combination aren't supported");
5332 if (!attr->transfer && attr->ingress && (action_flags &
5333 MLX5_FLOW_ACTION_ENCAP))
5334 return rte_flow_error_set(error, ENOTSUP,
5335 RTE_FLOW_ERROR_TYPE_ACTION,
5336 NULL, "encap is not supported"
5337 " for ingress traffic");
5339 /* Hairpin flow will add one more TAG action. */
5341 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5342 /* extra metadata enabled: one more TAG action will be add. */
5343 if (dev_conf->dv_flow_en &&
5344 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
5345 mlx5_flow_ext_mreg_supported(dev))
5346 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5347 if ((uint32_t)rw_act_num >
5348 flow_dv_modify_hdr_action_max(dev, is_root)) {
5349 return rte_flow_error_set(error, ENOTSUP,
5350 RTE_FLOW_ERROR_TYPE_ACTION,
5351 NULL, "too many header modify"
5352 " actions to support");
5358 * Internal preparation function. Allocates the DV flow size,
5359 * this size is constant.
5362 * Pointer to the rte_eth_dev structure.
5364 * Pointer to the flow attributes.
5366 * Pointer to the list of items.
5367 * @param[in] actions
5368 * Pointer to the list of actions.
5370 * Pointer to the error structure.
5373 * Pointer to mlx5_flow object on success,
5374 * otherwise NULL and rte_errno is set.
5376 static struct mlx5_flow *
5377 flow_dv_prepare(struct rte_eth_dev *dev,
5378 const struct rte_flow_attr *attr __rte_unused,
5379 const struct rte_flow_item items[] __rte_unused,
5380 const struct rte_flow_action actions[] __rte_unused,
5381 struct rte_flow_error *error)
5383 uint32_t handle_idx = 0;
5384 struct mlx5_flow *dev_flow;
5385 struct mlx5_flow_handle *dev_handle;
5386 struct mlx5_priv *priv = dev->data->dev_private;
5388 /* In case of corrupting the memory. */
5389 if (priv->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
5390 rte_flow_error_set(error, ENOSPC,
5391 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5392 "not free temporary device flow");
5395 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
5398 rte_flow_error_set(error, ENOMEM,
5399 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5400 "not enough memory to create flow handle");
5403 /* No multi-thread supporting. */
5404 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[priv->flow_idx++];
5405 dev_flow->handle = dev_handle;
5406 dev_flow->handle_idx = handle_idx;
5407 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
5409 * The matching value needs to be cleared to 0 before using. In the
5410 * past, it will be automatically cleared when using rte_*alloc
5411 * API. The time consumption will be almost the same as before.
5413 memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
5414 dev_flow->ingress = attr->ingress;
5415 dev_flow->dv.transfer = attr->transfer;
5419 #ifdef RTE_LIBRTE_MLX5_DEBUG
5421 * Sanity check for match mask and value. Similar to check_valid_spec() in
5422 * kernel driver. If unmasked bit is present in value, it returns failure.
5425 * pointer to match mask buffer.
5426 * @param match_value
5427 * pointer to match value buffer.
5430 * 0 if valid, -EINVAL otherwise.
5433 flow_dv_check_valid_spec(void *match_mask, void *match_value)
5435 uint8_t *m = match_mask;
5436 uint8_t *v = match_value;
5439 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
5442 "match_value differs from match_criteria"
5443 " %p[%u] != %p[%u]",
5444 match_value, i, match_mask, i);
5453 * Add Ethernet item to matcher and to the value.
5455 * @param[in, out] matcher
5457 * @param[in, out] key
5458 * Flow matcher value.
5460 * Flow pattern to translate.
5462 * Item is inner pattern.
5465 flow_dv_translate_item_eth(void *matcher, void *key,
5466 const struct rte_flow_item *item, int inner)
5468 const struct rte_flow_item_eth *eth_m = item->mask;
5469 const struct rte_flow_item_eth *eth_v = item->spec;
5470 const struct rte_flow_item_eth nic_mask = {
5471 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5472 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5473 .type = RTE_BE16(0xffff),
5485 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5487 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5489 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5491 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5493 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
5494 ð_m->dst, sizeof(eth_m->dst));
5495 /* The value must be in the range of the mask. */
5496 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
5497 for (i = 0; i < sizeof(eth_m->dst); ++i)
5498 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
5499 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
5500 ð_m->src, sizeof(eth_m->src));
5501 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
5502 /* The value must be in the range of the mask. */
5503 for (i = 0; i < sizeof(eth_m->dst); ++i)
5504 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
5506 /* When ethertype is present set mask for tagged VLAN. */
5507 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5508 /* Set value for tagged VLAN if ethertype is 802.1Q. */
5509 if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
5510 eth_v->type == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
5511 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag,
5513 /* Return here to avoid setting match on ethertype. */
5518 * HW supports match on one Ethertype, the Ethertype following the last
5519 * VLAN tag of the packet (see PRM).
5520 * Set match on ethertype only if ETH header is not followed by VLAN.
5522 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
5523 rte_be_to_cpu_16(eth_m->type));
5524 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, ethertype);
5525 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
5529 * Add VLAN item to matcher and to the value.
5531 * @param[in, out] dev_flow
5533 * @param[in, out] matcher
5535 * @param[in, out] key
5536 * Flow matcher value.
5538 * Flow pattern to translate.
5540 * Item is inner pattern.
5543 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
5544 void *matcher, void *key,
5545 const struct rte_flow_item *item,
5548 const struct rte_flow_item_vlan *vlan_m = item->mask;
5549 const struct rte_flow_item_vlan *vlan_v = item->spec;
5558 vlan_m = &rte_flow_item_vlan_mask;
5560 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5562 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5564 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5566 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5568 * This is workaround, masks are not supported,
5569 * and pre-validated.
5571 dev_flow->handle->vf_vlan.tag =
5572 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
5574 tci_m = rte_be_to_cpu_16(vlan_m->tci);
5575 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
5576 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5577 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
5578 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
5579 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
5580 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
5581 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
5582 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
5583 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
5584 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
5585 rte_be_to_cpu_16(vlan_m->inner_type));
5586 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
5587 rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
5591 * Add IPV4 item to matcher and to the value.
5593 * @param[in, out] matcher
5595 * @param[in, out] key
5596 * Flow matcher value.
5598 * Flow pattern to translate.
5599 * @param[in] item_flags
5600 * Bit-fields that holds the items detected until now.
5602 * Item is inner pattern.
5604 * The group to insert the rule.
5607 flow_dv_translate_item_ipv4(void *matcher, void *key,
5608 const struct rte_flow_item *item,
5609 const uint64_t item_flags,
5610 int inner, uint32_t group)
5612 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
5613 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
5614 const struct rte_flow_item_ipv4 nic_mask = {
5616 .src_addr = RTE_BE32(0xffffffff),
5617 .dst_addr = RTE_BE32(0xffffffff),
5618 .type_of_service = 0xff,
5619 .next_proto_id = 0xff,
5620 .time_to_live = 0xff,
5630 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5632 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5634 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5636 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5639 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5641 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x4);
5642 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 4);
5644 * On outer header (which must contains L2), or inner header with L2,
5645 * set cvlan_tag mask bit to mark this packet as untagged.
5646 * This should be done even if item->spec is empty.
5648 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
5649 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5654 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5655 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
5656 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5657 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
5658 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
5659 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
5660 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5661 src_ipv4_src_ipv6.ipv4_layout.ipv4);
5662 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5663 src_ipv4_src_ipv6.ipv4_layout.ipv4);
5664 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
5665 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
5666 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
5667 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
5668 ipv4_m->hdr.type_of_service);
5669 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
5670 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
5671 ipv4_m->hdr.type_of_service >> 2);
5672 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
5673 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
5674 ipv4_m->hdr.next_proto_id);
5675 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
5676 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
5677 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
5678 ipv4_m->hdr.time_to_live);
5679 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
5680 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
5684 * Add IPV6 item to matcher and to the value.
5686 * @param[in, out] matcher
5688 * @param[in, out] key
5689 * Flow matcher value.
5691 * Flow pattern to translate.
5692 * @param[in] item_flags
5693 * Bit-fields that holds the items detected until now.
5695 * Item is inner pattern.
5697 * The group to insert the rule.
5700 flow_dv_translate_item_ipv6(void *matcher, void *key,
5701 const struct rte_flow_item *item,
5702 const uint64_t item_flags,
5703 int inner, uint32_t group)
5705 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
5706 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
5707 const struct rte_flow_item_ipv6 nic_mask = {
5710 "\xff\xff\xff\xff\xff\xff\xff\xff"
5711 "\xff\xff\xff\xff\xff\xff\xff\xff",
5713 "\xff\xff\xff\xff\xff\xff\xff\xff"
5714 "\xff\xff\xff\xff\xff\xff\xff\xff",
5715 .vtc_flow = RTE_BE32(0xffffffff),
5722 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5723 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5732 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5734 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5736 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5738 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5741 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5743 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x6);
5744 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 6);
5746 * On outer header (which must contains L2), or inner header with L2,
5747 * set cvlan_tag mask bit to mark this packet as untagged.
5748 * This should be done even if item->spec is empty.
5750 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
5751 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5756 size = sizeof(ipv6_m->hdr.dst_addr);
5757 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5758 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
5759 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5760 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
5761 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
5762 for (i = 0; i < size; ++i)
5763 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
5764 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5765 src_ipv4_src_ipv6.ipv6_layout.ipv6);
5766 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5767 src_ipv4_src_ipv6.ipv6_layout.ipv6);
5768 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
5769 for (i = 0; i < size; ++i)
5770 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
5772 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
5773 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
5774 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
5775 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
5776 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
5777 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
5780 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
5782 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
5785 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
5787 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
5791 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
5793 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
5794 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
5796 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
5797 ipv6_m->hdr.hop_limits);
5798 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
5799 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
5803 * Add TCP item to matcher and to the value.
5805 * @param[in, out] matcher
5807 * @param[in, out] key
5808 * Flow matcher value.
5810 * Flow pattern to translate.
5812 * Item is inner pattern.
5815 flow_dv_translate_item_tcp(void *matcher, void *key,
5816 const struct rte_flow_item *item,
5819 const struct rte_flow_item_tcp *tcp_m = item->mask;
5820 const struct rte_flow_item_tcp *tcp_v = item->spec;
5825 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5827 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5829 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5831 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5833 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5834 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
5838 tcp_m = &rte_flow_item_tcp_mask;
5839 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
5840 rte_be_to_cpu_16(tcp_m->hdr.src_port));
5841 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
5842 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
5843 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
5844 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
5845 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
5846 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
5847 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
5848 tcp_m->hdr.tcp_flags);
5849 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
5850 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
5854 * Add UDP item to matcher and to the value.
5856 * @param[in, out] matcher
5858 * @param[in, out] key
5859 * Flow matcher value.
5861 * Flow pattern to translate.
5863 * Item is inner pattern.
5866 flow_dv_translate_item_udp(void *matcher, void *key,
5867 const struct rte_flow_item *item,
5870 const struct rte_flow_item_udp *udp_m = item->mask;
5871 const struct rte_flow_item_udp *udp_v = item->spec;
5876 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5878 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5880 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5882 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5884 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5885 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
5889 udp_m = &rte_flow_item_udp_mask;
5890 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
5891 rte_be_to_cpu_16(udp_m->hdr.src_port));
5892 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
5893 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
5894 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
5895 rte_be_to_cpu_16(udp_m->hdr.dst_port));
5896 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
5897 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
5901 * Add GRE optional Key item to matcher and to the value.
5903 * @param[in, out] matcher
5905 * @param[in, out] key
5906 * Flow matcher value.
5908 * Flow pattern to translate.
5910 * Item is inner pattern.
5913 flow_dv_translate_item_gre_key(void *matcher, void *key,
5914 const struct rte_flow_item *item)
5916 const rte_be32_t *key_m = item->mask;
5917 const rte_be32_t *key_v = item->spec;
5918 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5919 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5920 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
5922 /* GRE K bit must be on and should already be validated */
5923 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
5924 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
5928 key_m = &gre_key_default_mask;
5929 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
5930 rte_be_to_cpu_32(*key_m) >> 8);
5931 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
5932 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
5933 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
5934 rte_be_to_cpu_32(*key_m) & 0xFF);
5935 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
5936 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
5940 * Add GRE item to matcher and to the value.
5942 * @param[in, out] matcher
5944 * @param[in, out] key
5945 * Flow matcher value.
5947 * Flow pattern to translate.
5949 * Item is inner pattern.
5952 flow_dv_translate_item_gre(void *matcher, void *key,
5953 const struct rte_flow_item *item,
5956 const struct rte_flow_item_gre *gre_m = item->mask;
5957 const struct rte_flow_item_gre *gre_v = item->spec;
5960 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5961 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5968 uint16_t s_present:1;
5969 uint16_t k_present:1;
5970 uint16_t rsvd_bit1:1;
5971 uint16_t c_present:1;
5975 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
5978 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5980 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5982 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5984 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5986 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5987 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
5991 gre_m = &rte_flow_item_gre_mask;
5992 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
5993 rte_be_to_cpu_16(gre_m->protocol));
5994 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
5995 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
5996 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
5997 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
5998 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
5999 gre_crks_rsvd0_ver_m.c_present);
6000 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6001 gre_crks_rsvd0_ver_v.c_present &
6002 gre_crks_rsvd0_ver_m.c_present);
6003 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6004 gre_crks_rsvd0_ver_m.k_present);
6005 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6006 gre_crks_rsvd0_ver_v.k_present &
6007 gre_crks_rsvd0_ver_m.k_present);
6008 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6009 gre_crks_rsvd0_ver_m.s_present);
6010 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6011 gre_crks_rsvd0_ver_v.s_present &
6012 gre_crks_rsvd0_ver_m.s_present);
6016 * Add NVGRE item to matcher and to the value.
6018 * @param[in, out] matcher
6020 * @param[in, out] key
6021 * Flow matcher value.
6023 * Flow pattern to translate.
6025 * Item is inner pattern.
6028 flow_dv_translate_item_nvgre(void *matcher, void *key,
6029 const struct rte_flow_item *item,
6032 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6033 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6034 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6035 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6036 const char *tni_flow_id_m = (const char *)nvgre_m->tni;
6037 const char *tni_flow_id_v = (const char *)nvgre_v->tni;
6043 /* For NVGRE, GRE header fields must be set with defined values. */
6044 const struct rte_flow_item_gre gre_spec = {
6045 .c_rsvd0_ver = RTE_BE16(0x2000),
6046 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6048 const struct rte_flow_item_gre gre_mask = {
6049 .c_rsvd0_ver = RTE_BE16(0xB000),
6050 .protocol = RTE_BE16(UINT16_MAX),
6052 const struct rte_flow_item gre_item = {
6057 flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6061 nvgre_m = &rte_flow_item_nvgre_mask;
6062 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6063 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6064 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6065 memcpy(gre_key_m, tni_flow_id_m, size);
6066 for (i = 0; i < size; ++i)
6067 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6071 * Add VXLAN item to matcher and to the value.
6073 * @param[in, out] matcher
6075 * @param[in, out] key
6076 * Flow matcher value.
6078 * Flow pattern to translate.
6080 * Item is inner pattern.
6083 flow_dv_translate_item_vxlan(void *matcher, void *key,
6084 const struct rte_flow_item *item,
6087 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
6088 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
6091 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6092 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6100 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6102 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6104 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6106 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6108 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6109 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6110 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6111 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6112 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6117 vxlan_m = &rte_flow_item_vxlan_mask;
6118 size = sizeof(vxlan_m->vni);
6119 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
6120 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
6121 memcpy(vni_m, vxlan_m->vni, size);
6122 for (i = 0; i < size; ++i)
6123 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6127 * Add VXLAN-GPE item to matcher and to the value.
6129 * @param[in, out] matcher
6131 * @param[in, out] key
6132 * Flow matcher value.
6134 * Flow pattern to translate.
6136 * Item is inner pattern.
6140 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
6141 const struct rte_flow_item *item, int inner)
6143 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
6144 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
6148 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
6150 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6156 uint8_t flags_m = 0xff;
6157 uint8_t flags_v = 0xc;
6160 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6162 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6164 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6166 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6168 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6169 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6170 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6171 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6172 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6177 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
6178 size = sizeof(vxlan_m->vni);
6179 vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
6180 vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
6181 memcpy(vni_m, vxlan_m->vni, size);
6182 for (i = 0; i < size; ++i)
6183 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6184 if (vxlan_m->flags) {
6185 flags_m = vxlan_m->flags;
6186 flags_v = vxlan_v->flags;
6188 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
6189 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
6190 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
6192 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
6197 * Add Geneve item to matcher and to the value.
6199 * @param[in, out] matcher
6201 * @param[in, out] key
6202 * Flow matcher value.
6204 * Flow pattern to translate.
6206 * Item is inner pattern.
6210 flow_dv_translate_item_geneve(void *matcher, void *key,
6211 const struct rte_flow_item *item, int inner)
6213 const struct rte_flow_item_geneve *geneve_m = item->mask;
6214 const struct rte_flow_item_geneve *geneve_v = item->spec;
6217 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6218 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6227 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6229 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6231 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6233 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6235 dport = MLX5_UDP_PORT_GENEVE;
6236 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6237 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6238 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6243 geneve_m = &rte_flow_item_geneve_mask;
6244 size = sizeof(geneve_m->vni);
6245 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
6246 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
6247 memcpy(vni_m, geneve_m->vni, size);
6248 for (i = 0; i < size; ++i)
6249 vni_v[i] = vni_m[i] & geneve_v->vni[i];
6250 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
6251 rte_be_to_cpu_16(geneve_m->protocol));
6252 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
6253 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
6254 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
6255 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
6256 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
6257 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6258 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
6259 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6260 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
6261 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6262 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
6263 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
6264 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6268 * Add MPLS item to matcher and to the value.
6270 * @param[in, out] matcher
6272 * @param[in, out] key
6273 * Flow matcher value.
6275 * Flow pattern to translate.
6276 * @param[in] prev_layer
6277 * The protocol layer indicated in previous item.
6279 * Item is inner pattern.
6282 flow_dv_translate_item_mpls(void *matcher, void *key,
6283 const struct rte_flow_item *item,
6284 uint64_t prev_layer,
6287 const uint32_t *in_mpls_m = item->mask;
6288 const uint32_t *in_mpls_v = item->spec;
6289 uint32_t *out_mpls_m = 0;
6290 uint32_t *out_mpls_v = 0;
6291 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6292 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6293 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
6295 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6296 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
6297 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6299 switch (prev_layer) {
6300 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6301 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
6302 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6303 MLX5_UDP_PORT_MPLS);
6305 case MLX5_FLOW_LAYER_GRE:
6306 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
6307 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6308 RTE_ETHER_TYPE_MPLS);
6311 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6312 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6319 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
6320 switch (prev_layer) {
6321 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6323 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6324 outer_first_mpls_over_udp);
6326 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6327 outer_first_mpls_over_udp);
6329 case MLX5_FLOW_LAYER_GRE:
6331 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6332 outer_first_mpls_over_gre);
6334 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6335 outer_first_mpls_over_gre);
6338 /* Inner MPLS not over GRE is not supported. */
6341 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6345 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6351 if (out_mpls_m && out_mpls_v) {
6352 *out_mpls_m = *in_mpls_m;
6353 *out_mpls_v = *in_mpls_v & *in_mpls_m;
6358 * Add metadata register item to matcher
6360 * @param[in, out] matcher
6362 * @param[in, out] key
6363 * Flow matcher value.
6364 * @param[in] reg_type
6365 * Type of device metadata register
6372 flow_dv_match_meta_reg(void *matcher, void *key,
6373 enum modify_reg reg_type,
6374 uint32_t data, uint32_t mask)
6377 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
6379 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6385 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
6386 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
6389 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
6390 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
6394 * The metadata register C0 field might be divided into
6395 * source vport index and META item value, we should set
6396 * this field according to specified mask, not as whole one.
6398 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
6400 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
6401 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
6404 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
6407 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
6408 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
6411 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
6412 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
6415 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
6416 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
6419 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
6420 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
6423 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
6424 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
6427 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
6428 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
6431 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
6432 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
6441 * Add MARK item to matcher
6444 * The device to configure through.
6445 * @param[in, out] matcher
6447 * @param[in, out] key
6448 * Flow matcher value.
6450 * Flow pattern to translate.
6453 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
6454 void *matcher, void *key,
6455 const struct rte_flow_item *item)
6457 struct mlx5_priv *priv = dev->data->dev_private;
6458 const struct rte_flow_item_mark *mark;
6462 mark = item->mask ? (const void *)item->mask :
6463 &rte_flow_item_mark_mask;
6464 mask = mark->id & priv->sh->dv_mark_mask;
6465 mark = (const void *)item->spec;
6467 value = mark->id & priv->sh->dv_mark_mask & mask;
6469 enum modify_reg reg;
6471 /* Get the metadata register index for the mark. */
6472 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
6473 MLX5_ASSERT(reg > 0);
6474 if (reg == REG_C_0) {
6475 struct mlx5_priv *priv = dev->data->dev_private;
6476 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6477 uint32_t shl_c0 = rte_bsf32(msk_c0);
6483 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6488 * Add META item to matcher
6491 * The devich to configure through.
6492 * @param[in, out] matcher
6494 * @param[in, out] key
6495 * Flow matcher value.
6497 * Attributes of flow that includes this item.
6499 * Flow pattern to translate.
6502 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
6503 void *matcher, void *key,
6504 const struct rte_flow_attr *attr,
6505 const struct rte_flow_item *item)
6507 const struct rte_flow_item_meta *meta_m;
6508 const struct rte_flow_item_meta *meta_v;
6510 meta_m = (const void *)item->mask;
6512 meta_m = &rte_flow_item_meta_mask;
6513 meta_v = (const void *)item->spec;
6516 uint32_t value = meta_v->data;
6517 uint32_t mask = meta_m->data;
6519 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
6523 * In datapath code there is no endianness
6524 * coversions for perfromance reasons, all
6525 * pattern conversions are done in rte_flow.
6527 value = rte_cpu_to_be_32(value);
6528 mask = rte_cpu_to_be_32(mask);
6529 if (reg == REG_C_0) {
6530 struct mlx5_priv *priv = dev->data->dev_private;
6531 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6532 uint32_t shl_c0 = rte_bsf32(msk_c0);
6533 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
6534 uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
6541 MLX5_ASSERT(msk_c0);
6542 MLX5_ASSERT(!(~msk_c0 & mask));
6544 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6549 * Add vport metadata Reg C0 item to matcher
6551 * @param[in, out] matcher
6553 * @param[in, out] key
6554 * Flow matcher value.
6556 * Flow pattern to translate.
6559 flow_dv_translate_item_meta_vport(void *matcher, void *key,
6560 uint32_t value, uint32_t mask)
6562 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
6566 * Add tag item to matcher
6569 * The devich to configure through.
6570 * @param[in, out] matcher
6572 * @param[in, out] key
6573 * Flow matcher value.
6575 * Flow pattern to translate.
6578 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
6579 void *matcher, void *key,
6580 const struct rte_flow_item *item)
6582 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
6583 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
6584 uint32_t mask, value;
6587 value = tag_v->data;
6588 mask = tag_m ? tag_m->data : UINT32_MAX;
6589 if (tag_v->id == REG_C_0) {
6590 struct mlx5_priv *priv = dev->data->dev_private;
6591 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6592 uint32_t shl_c0 = rte_bsf32(msk_c0);
6598 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
6602 * Add TAG item to matcher
6605 * The devich to configure through.
6606 * @param[in, out] matcher
6608 * @param[in, out] key
6609 * Flow matcher value.
6611 * Flow pattern to translate.
6614 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
6615 void *matcher, void *key,
6616 const struct rte_flow_item *item)
6618 const struct rte_flow_item_tag *tag_v = item->spec;
6619 const struct rte_flow_item_tag *tag_m = item->mask;
6620 enum modify_reg reg;
6623 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
6624 /* Get the metadata register index for the tag. */
6625 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
6626 MLX5_ASSERT(reg > 0);
6627 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
6631 * Add source vport match to the specified matcher.
6633 * @param[in, out] matcher
6635 * @param[in, out] key
6636 * Flow matcher value.
6638 * Source vport value to match
6643 flow_dv_translate_item_source_vport(void *matcher, void *key,
6644 int16_t port, uint16_t mask)
6646 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6647 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6649 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
6650 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
6654 * Translate port-id item to eswitch match on port-id.
6657 * The devich to configure through.
6658 * @param[in, out] matcher
6660 * @param[in, out] key
6661 * Flow matcher value.
6663 * Flow pattern to translate.
6666 * 0 on success, a negative errno value otherwise.
6669 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
6670 void *key, const struct rte_flow_item *item)
6672 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
6673 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
6674 struct mlx5_priv *priv;
6677 mask = pid_m ? pid_m->id : 0xffff;
6678 id = pid_v ? pid_v->id : dev->data->port_id;
6679 priv = mlx5_port_to_eswitch_info(id, item == NULL);
6682 /* Translate to vport field or to metadata, depending on mode. */
6683 if (priv->vport_meta_mask)
6684 flow_dv_translate_item_meta_vport(matcher, key,
6685 priv->vport_meta_tag,
6686 priv->vport_meta_mask);
6688 flow_dv_translate_item_source_vport(matcher, key,
6689 priv->vport_id, mask);
6694 * Add ICMP6 item to matcher and to the value.
6696 * @param[in, out] matcher
6698 * @param[in, out] key
6699 * Flow matcher value.
6701 * Flow pattern to translate.
6703 * Item is inner pattern.
6706 flow_dv_translate_item_icmp6(void *matcher, void *key,
6707 const struct rte_flow_item *item,
6710 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
6711 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
6714 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6716 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6718 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6720 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6722 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6724 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6726 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
6727 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
6731 icmp6_m = &rte_flow_item_icmp6_mask;
6733 * Force flow only to match the non-fragmented IPv6 ICMPv6 packets.
6734 * If only the protocol is specified, no need to match the frag.
6736 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6737 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
6738 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
6739 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
6740 icmp6_v->type & icmp6_m->type);
6741 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
6742 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
6743 icmp6_v->code & icmp6_m->code);
6747 * Add ICMP item to matcher and to the value.
6749 * @param[in, out] matcher
6751 * @param[in, out] key
6752 * Flow matcher value.
6754 * Flow pattern to translate.
6756 * Item is inner pattern.
6759 flow_dv_translate_item_icmp(void *matcher, void *key,
6760 const struct rte_flow_item *item,
6763 const struct rte_flow_item_icmp *icmp_m = item->mask;
6764 const struct rte_flow_item_icmp *icmp_v = item->spec;
6767 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6769 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6771 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6773 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6775 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6777 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6779 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
6780 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
6784 icmp_m = &rte_flow_item_icmp_mask;
6786 * Force flow only to match the non-fragmented IPv4 ICMP packets.
6787 * If only the protocol is specified, no need to match the frag.
6789 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6790 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
6791 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
6792 icmp_m->hdr.icmp_type);
6793 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
6794 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
6795 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
6796 icmp_m->hdr.icmp_code);
6797 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
6798 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
6802 * Add GTP item to matcher and to the value.
6804 * @param[in, out] matcher
6806 * @param[in, out] key
6807 * Flow matcher value.
6809 * Flow pattern to translate.
6811 * Item is inner pattern.
6814 flow_dv_translate_item_gtp(void *matcher, void *key,
6815 const struct rte_flow_item *item, int inner)
6817 const struct rte_flow_item_gtp *gtp_m = item->mask;
6818 const struct rte_flow_item_gtp *gtp_v = item->spec;
6821 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6823 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6824 uint16_t dport = RTE_GTPU_UDP_PORT;
6827 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6829 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6831 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6833 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6835 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6836 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6837 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6842 gtp_m = &rte_flow_item_gtp_mask;
6843 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
6844 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
6845 gtp_v->msg_type & gtp_m->msg_type);
6846 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
6847 rte_be_to_cpu_32(gtp_m->teid));
6848 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
6849 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
6852 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
6854 #define HEADER_IS_ZERO(match_criteria, headers) \
6855 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
6856 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
6859 * Calculate flow matcher enable bitmap.
6861 * @param match_criteria
6862 * Pointer to flow matcher criteria.
6865 * Bitmap of enabled fields.
6868 flow_dv_matcher_enable(uint32_t *match_criteria)
6870 uint8_t match_criteria_enable;
6872 match_criteria_enable =
6873 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
6874 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
6875 match_criteria_enable |=
6876 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
6877 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
6878 match_criteria_enable |=
6879 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
6880 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
6881 match_criteria_enable |=
6882 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
6883 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
6884 match_criteria_enable |=
6885 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
6886 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
6887 return match_criteria_enable;
6894 * @param[in, out] dev
6895 * Pointer to rte_eth_dev structure.
6896 * @param[in] table_id
6899 * Direction of the table.
6900 * @param[in] transfer
6901 * E-Switch or NIC flow.
6903 * pointer to error structure.
6906 * Returns tables resource based on the index, NULL in case of failed.
6908 static struct mlx5_flow_tbl_resource *
6909 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
6910 uint32_t table_id, uint8_t egress,
6912 struct rte_flow_error *error)
6914 struct mlx5_priv *priv = dev->data->dev_private;
6915 struct mlx5_ibv_shared *sh = priv->sh;
6916 struct mlx5_flow_tbl_resource *tbl;
6917 union mlx5_flow_tbl_key table_key = {
6919 .table_id = table_id,
6921 .domain = !!transfer,
6922 .direction = !!egress,
6925 struct mlx5_hlist_entry *pos = mlx5_hlist_lookup(sh->flow_tbls,
6927 struct mlx5_flow_tbl_data_entry *tbl_data;
6933 tbl_data = container_of(pos, struct mlx5_flow_tbl_data_entry,
6935 tbl = &tbl_data->tbl;
6936 rte_atomic32_inc(&tbl->refcnt);
6939 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
6941 rte_flow_error_set(error, ENOMEM,
6942 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6944 "cannot allocate flow table data entry");
6947 tbl_data->idx = idx;
6948 tbl = &tbl_data->tbl;
6949 pos = &tbl_data->entry;
6951 domain = sh->fdb_domain;
6953 domain = sh->tx_domain;
6955 domain = sh->rx_domain;
6956 tbl->obj = mlx5_glue->dr_create_flow_tbl(domain, table_id);
6958 rte_flow_error_set(error, ENOMEM,
6959 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6960 NULL, "cannot create flow table object");
6961 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
6965 * No multi-threads now, but still better to initialize the reference
6966 * count before insert it into the hash list.
6968 rte_atomic32_init(&tbl->refcnt);
6969 /* Jump action reference count is initialized here. */
6970 rte_atomic32_init(&tbl_data->jump.refcnt);
6971 pos->key = table_key.v64;
6972 ret = mlx5_hlist_insert(sh->flow_tbls, pos);
6974 rte_flow_error_set(error, -ret,
6975 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6976 "cannot insert flow table data entry");
6977 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
6978 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
6980 rte_atomic32_inc(&tbl->refcnt);
6985 * Release a flow table.
6988 * Pointer to rte_eth_dev structure.
6990 * Table resource to be released.
6993 * Returns 0 if table was released, else return 1;
6996 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
6997 struct mlx5_flow_tbl_resource *tbl)
6999 struct mlx5_priv *priv = dev->data->dev_private;
7000 struct mlx5_ibv_shared *sh = priv->sh;
7001 struct mlx5_flow_tbl_data_entry *tbl_data =
7002 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7006 if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
7007 struct mlx5_hlist_entry *pos = &tbl_data->entry;
7009 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
7011 /* remove the entry from the hash list and free memory. */
7012 mlx5_hlist_remove(sh->flow_tbls, pos);
7013 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_JUMP],
7021 * Register the flow matcher.
7023 * @param[in, out] dev
7024 * Pointer to rte_eth_dev structure.
7025 * @param[in, out] matcher
7026 * Pointer to flow matcher.
7027 * @param[in, out] key
7028 * Pointer to flow table key.
7029 * @parm[in, out] dev_flow
7030 * Pointer to the dev_flow.
7032 * pointer to error structure.
7035 * 0 on success otherwise -errno and errno is set.
7038 flow_dv_matcher_register(struct rte_eth_dev *dev,
7039 struct mlx5_flow_dv_matcher *matcher,
7040 union mlx5_flow_tbl_key *key,
7041 struct mlx5_flow *dev_flow,
7042 struct rte_flow_error *error)
7044 struct mlx5_priv *priv = dev->data->dev_private;
7045 struct mlx5_ibv_shared *sh = priv->sh;
7046 struct mlx5_flow_dv_matcher *cache_matcher;
7047 struct mlx5dv_flow_matcher_attr dv_attr = {
7048 .type = IBV_FLOW_ATTR_NORMAL,
7049 .match_mask = (void *)&matcher->mask,
7051 struct mlx5_flow_tbl_resource *tbl;
7052 struct mlx5_flow_tbl_data_entry *tbl_data;
7054 tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
7055 key->domain, error);
7057 return -rte_errno; /* No need to refill the error info */
7058 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7059 /* Lookup from cache. */
7060 LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
7061 if (matcher->crc == cache_matcher->crc &&
7062 matcher->priority == cache_matcher->priority &&
7063 !memcmp((const void *)matcher->mask.buf,
7064 (const void *)cache_matcher->mask.buf,
7065 cache_matcher->mask.size)) {
7067 "%s group %u priority %hd use %s "
7068 "matcher %p: refcnt %d++",
7069 key->domain ? "FDB" : "NIC", key->table_id,
7070 cache_matcher->priority,
7071 key->direction ? "tx" : "rx",
7072 (void *)cache_matcher,
7073 rte_atomic32_read(&cache_matcher->refcnt));
7074 rte_atomic32_inc(&cache_matcher->refcnt);
7075 dev_flow->handle->dvh.matcher = cache_matcher;
7076 /* old matcher should not make the table ref++. */
7077 flow_dv_tbl_resource_release(dev, tbl);
7081 /* Register new matcher. */
7082 cache_matcher = rte_calloc(__func__, 1, sizeof(*cache_matcher), 0);
7083 if (!cache_matcher) {
7084 flow_dv_tbl_resource_release(dev, tbl);
7085 return rte_flow_error_set(error, ENOMEM,
7086 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7087 "cannot allocate matcher memory");
7089 *cache_matcher = *matcher;
7090 dv_attr.match_criteria_enable =
7091 flow_dv_matcher_enable(cache_matcher->mask.buf);
7092 dv_attr.priority = matcher->priority;
7094 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
7095 cache_matcher->matcher_object =
7096 mlx5_glue->dv_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj);
7097 if (!cache_matcher->matcher_object) {
7098 rte_free(cache_matcher);
7099 #ifdef HAVE_MLX5DV_DR
7100 flow_dv_tbl_resource_release(dev, tbl);
7102 return rte_flow_error_set(error, ENOMEM,
7103 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7104 NULL, "cannot create matcher");
7106 /* Save the table information */
7107 cache_matcher->tbl = tbl;
7108 rte_atomic32_init(&cache_matcher->refcnt);
7109 /* only matcher ref++, table ref++ already done above in get API. */
7110 rte_atomic32_inc(&cache_matcher->refcnt);
7111 LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
7112 dev_flow->handle->dvh.matcher = cache_matcher;
7113 DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
7114 key->domain ? "FDB" : "NIC", key->table_id,
7115 cache_matcher->priority,
7116 key->direction ? "tx" : "rx", (void *)cache_matcher,
7117 rte_atomic32_read(&cache_matcher->refcnt));
7122 * Find existing tag resource or create and register a new one.
7124 * @param dev[in, out]
7125 * Pointer to rte_eth_dev structure.
7126 * @param[in, out] tag_be24
7127 * Tag value in big endian then R-shift 8.
7128 * @parm[in, out] dev_flow
7129 * Pointer to the dev_flow.
7131 * pointer to error structure.
7134 * 0 on success otherwise -errno and errno is set.
7137 flow_dv_tag_resource_register
7138 (struct rte_eth_dev *dev,
7140 struct mlx5_flow *dev_flow,
7141 struct rte_flow_error *error)
7143 struct mlx5_priv *priv = dev->data->dev_private;
7144 struct mlx5_ibv_shared *sh = priv->sh;
7145 struct mlx5_flow_dv_tag_resource *cache_resource;
7146 struct mlx5_hlist_entry *entry;
7148 /* Lookup a matching resource from cache. */
7149 entry = mlx5_hlist_lookup(sh->tag_table, (uint64_t)tag_be24);
7151 cache_resource = container_of
7152 (entry, struct mlx5_flow_dv_tag_resource, entry);
7153 rte_atomic32_inc(&cache_resource->refcnt);
7154 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
7155 dev_flow->dv.tag_resource = cache_resource;
7156 DRV_LOG(DEBUG, "cached tag resource %p: refcnt now %d++",
7157 (void *)cache_resource,
7158 rte_atomic32_read(&cache_resource->refcnt));
7161 /* Register new resource. */
7162 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG],
7163 &dev_flow->handle->dvh.rix_tag);
7164 if (!cache_resource)
7165 return rte_flow_error_set(error, ENOMEM,
7166 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7167 "cannot allocate resource memory");
7168 cache_resource->entry.key = (uint64_t)tag_be24;
7169 cache_resource->action = mlx5_glue->dv_create_flow_action_tag(tag_be24);
7170 if (!cache_resource->action) {
7171 rte_free(cache_resource);
7172 return rte_flow_error_set(error, ENOMEM,
7173 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7174 NULL, "cannot create action");
7176 rte_atomic32_init(&cache_resource->refcnt);
7177 rte_atomic32_inc(&cache_resource->refcnt);
7178 if (mlx5_hlist_insert(sh->tag_table, &cache_resource->entry)) {
7179 mlx5_glue->destroy_flow_action(cache_resource->action);
7180 rte_free(cache_resource);
7181 return rte_flow_error_set(error, EEXIST,
7182 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7183 NULL, "cannot insert tag");
7185 dev_flow->dv.tag_resource = cache_resource;
7186 DRV_LOG(DEBUG, "new tag resource %p: refcnt now %d++",
7187 (void *)cache_resource,
7188 rte_atomic32_read(&cache_resource->refcnt));
7196 * Pointer to Ethernet device.
7201 * 1 while a reference on it exists, 0 when freed.
7204 flow_dv_tag_release(struct rte_eth_dev *dev,
7207 struct mlx5_priv *priv = dev->data->dev_private;
7208 struct mlx5_ibv_shared *sh = priv->sh;
7209 struct mlx5_flow_dv_tag_resource *tag;
7211 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7214 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
7215 dev->data->port_id, (void *)tag,
7216 rte_atomic32_read(&tag->refcnt));
7217 if (rte_atomic32_dec_and_test(&tag->refcnt)) {
7218 claim_zero(mlx5_glue->destroy_flow_action(tag->action));
7219 mlx5_hlist_remove(sh->tag_table, &tag->entry);
7220 DRV_LOG(DEBUG, "port %u tag %p: removed",
7221 dev->data->port_id, (void *)tag);
7222 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7229 * Translate port ID action to vport.
7232 * Pointer to rte_eth_dev structure.
7234 * Pointer to the port ID action.
7235 * @param[out] dst_port_id
7236 * The target port ID.
7238 * Pointer to the error structure.
7241 * 0 on success, a negative errno value otherwise and rte_errno is set.
7244 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
7245 const struct rte_flow_action *action,
7246 uint32_t *dst_port_id,
7247 struct rte_flow_error *error)
7250 struct mlx5_priv *priv;
7251 const struct rte_flow_action_port_id *conf =
7252 (const struct rte_flow_action_port_id *)action->conf;
7254 port = conf->original ? dev->data->port_id : conf->id;
7255 priv = mlx5_port_to_eswitch_info(port, false);
7257 return rte_flow_error_set(error, -rte_errno,
7258 RTE_FLOW_ERROR_TYPE_ACTION,
7260 "No eswitch info was found for port");
7261 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
7263 * This parameter is transferred to
7264 * mlx5dv_dr_action_create_dest_ib_port().
7266 *dst_port_id = priv->ibv_port;
7269 * Legacy mode, no LAG configurations is supported.
7270 * This parameter is transferred to
7271 * mlx5dv_dr_action_create_dest_vport().
7273 *dst_port_id = priv->vport_id;
7279 * Add Tx queue matcher
7282 * Pointer to the dev struct.
7283 * @param[in, out] matcher
7285 * @param[in, out] key
7286 * Flow matcher value.
7288 * Flow pattern to translate.
7290 * Item is inner pattern.
7293 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
7294 void *matcher, void *key,
7295 const struct rte_flow_item *item)
7297 const struct mlx5_rte_flow_item_tx_queue *queue_m;
7298 const struct mlx5_rte_flow_item_tx_queue *queue_v;
7300 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7302 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7303 struct mlx5_txq_ctrl *txq;
7307 queue_m = (const void *)item->mask;
7310 queue_v = (const void *)item->spec;
7313 txq = mlx5_txq_get(dev, queue_v->queue);
7316 queue = txq->obj->sq->id;
7317 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
7318 MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
7319 queue & queue_m->queue);
7320 mlx5_txq_release(dev, queue_v->queue);
7324 * Set the hash fields according to the @p flow information.
7326 * @param[in] dev_flow
7327 * Pointer to the mlx5_flow.
7328 * @param[in] rss_desc
7329 * Pointer to the mlx5_flow_rss_desc.
7332 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
7333 struct mlx5_flow_rss_desc *rss_desc)
7335 uint64_t items = dev_flow->handle->layers;
7337 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
7339 dev_flow->hash_fields = 0;
7340 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
7341 if (rss_desc->level >= 2) {
7342 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
7346 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
7347 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
7348 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
7349 if (rss_types & ETH_RSS_L3_SRC_ONLY)
7350 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
7351 else if (rss_types & ETH_RSS_L3_DST_ONLY)
7352 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
7354 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
7356 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
7357 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
7358 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
7359 if (rss_types & ETH_RSS_L3_SRC_ONLY)
7360 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
7361 else if (rss_types & ETH_RSS_L3_DST_ONLY)
7362 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
7364 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
7367 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
7368 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
7369 if (rss_types & ETH_RSS_UDP) {
7370 if (rss_types & ETH_RSS_L4_SRC_ONLY)
7371 dev_flow->hash_fields |=
7372 IBV_RX_HASH_SRC_PORT_UDP;
7373 else if (rss_types & ETH_RSS_L4_DST_ONLY)
7374 dev_flow->hash_fields |=
7375 IBV_RX_HASH_DST_PORT_UDP;
7377 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
7379 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
7380 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
7381 if (rss_types & ETH_RSS_TCP) {
7382 if (rss_types & ETH_RSS_L4_SRC_ONLY)
7383 dev_flow->hash_fields |=
7384 IBV_RX_HASH_SRC_PORT_TCP;
7385 else if (rss_types & ETH_RSS_L4_DST_ONLY)
7386 dev_flow->hash_fields |=
7387 IBV_RX_HASH_DST_PORT_TCP;
7389 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
7395 * Fill the flow with DV spec, lock free
7396 * (mutex should be acquired by caller).
7399 * Pointer to rte_eth_dev structure.
7400 * @param[in, out] dev_flow
7401 * Pointer to the sub flow.
7403 * Pointer to the flow attributes.
7405 * Pointer to the list of items.
7406 * @param[in] actions
7407 * Pointer to the list of actions.
7409 * Pointer to the error structure.
7412 * 0 on success, a negative errno value otherwise and rte_errno is set.
7415 __flow_dv_translate(struct rte_eth_dev *dev,
7416 struct mlx5_flow *dev_flow,
7417 const struct rte_flow_attr *attr,
7418 const struct rte_flow_item items[],
7419 const struct rte_flow_action actions[],
7420 struct rte_flow_error *error)
7422 struct mlx5_priv *priv = dev->data->dev_private;
7423 struct mlx5_dev_config *dev_conf = &priv->config;
7424 struct rte_flow *flow = dev_flow->flow;
7425 struct mlx5_flow_handle *handle = dev_flow->handle;
7426 struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
7428 [!!priv->flow_nested_idx];
7429 uint64_t item_flags = 0;
7430 uint64_t last_item = 0;
7431 uint64_t action_flags = 0;
7432 uint64_t priority = attr->priority;
7433 struct mlx5_flow_dv_matcher matcher = {
7435 .size = sizeof(matcher.mask.buf),
7439 bool actions_end = false;
7441 struct mlx5_flow_dv_modify_hdr_resource res;
7442 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
7443 sizeof(struct mlx5_modification_cmd) *
7444 (MLX5_MAX_MODIFY_NUM + 1)];
7446 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
7447 union flow_dv_attr flow_attr = { .attr = 0 };
7449 union mlx5_flow_tbl_key tbl_key;
7450 uint32_t modify_action_position = UINT32_MAX;
7451 void *match_mask = matcher.mask.buf;
7452 void *match_value = dev_flow->dv.value.buf;
7453 uint8_t next_protocol = 0xff;
7454 struct rte_vlan_hdr vlan = { 0 };
7458 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
7459 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
7460 ret = mlx5_flow_group_to_table(attr, dev_flow->external, attr->group,
7461 !!priv->fdb_def_rule, &table, error);
7464 dev_flow->dv.group = table;
7466 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
7467 if (priority == MLX5_FLOW_PRIO_RSVD)
7468 priority = dev_conf->flow_prio - 1;
7469 /* number of actions must be set to 0 in case of dirty stack. */
7470 mhdr_res->actions_num = 0;
7471 for (; !actions_end ; actions++) {
7472 const struct rte_flow_action_queue *queue;
7473 const struct rte_flow_action_rss *rss;
7474 const struct rte_flow_action *action = actions;
7475 const struct rte_flow_action_count *count = action->conf;
7476 const uint8_t *rss_key;
7477 const struct rte_flow_action_jump *jump_data;
7478 const struct rte_flow_action_meter *mtr;
7479 struct mlx5_flow_tbl_resource *tbl;
7480 uint32_t port_id = 0;
7481 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
7482 int action_type = actions->type;
7483 const struct rte_flow_action *found_action = NULL;
7484 struct mlx5_flow_meter *fm = NULL;
7486 switch (action_type) {
7487 case RTE_FLOW_ACTION_TYPE_VOID:
7489 case RTE_FLOW_ACTION_TYPE_PORT_ID:
7490 if (flow_dv_translate_action_port_id(dev, action,
7493 memset(&port_id_resource, 0, sizeof(port_id_resource));
7494 port_id_resource.port_id = port_id;
7495 if (flow_dv_port_id_action_resource_register
7496 (dev, &port_id_resource, dev_flow, error))
7498 MLX5_ASSERT(!handle->rix_port_id_action);
7499 dev_flow->dv.actions[actions_n++] =
7500 dev_flow->dv.port_id_action->action;
7501 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
7502 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
7504 case RTE_FLOW_ACTION_TYPE_FLAG:
7505 action_flags |= MLX5_FLOW_ACTION_FLAG;
7506 dev_flow->handle->mark = 1;
7507 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7508 struct rte_flow_action_mark mark = {
7509 .id = MLX5_FLOW_MARK_DEFAULT,
7512 if (flow_dv_convert_action_mark(dev, &mark,
7516 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
7519 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
7521 * Only one FLAG or MARK is supported per device flow
7522 * right now. So the pointer to the tag resource must be
7523 * zero before the register process.
7525 MLX5_ASSERT(!handle->dvh.rix_tag);
7526 if (flow_dv_tag_resource_register(dev, tag_be,
7529 MLX5_ASSERT(dev_flow->dv.tag_resource);
7530 dev_flow->dv.actions[actions_n++] =
7531 dev_flow->dv.tag_resource->action;
7533 case RTE_FLOW_ACTION_TYPE_MARK:
7534 action_flags |= MLX5_FLOW_ACTION_MARK;
7535 dev_flow->handle->mark = 1;
7536 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7537 const struct rte_flow_action_mark *mark =
7538 (const struct rte_flow_action_mark *)
7541 if (flow_dv_convert_action_mark(dev, mark,
7545 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
7549 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
7550 /* Legacy (non-extensive) MARK action. */
7551 tag_be = mlx5_flow_mark_set
7552 (((const struct rte_flow_action_mark *)
7553 (actions->conf))->id);
7554 MLX5_ASSERT(!handle->dvh.rix_tag);
7555 if (flow_dv_tag_resource_register(dev, tag_be,
7558 MLX5_ASSERT(dev_flow->dv.tag_resource);
7559 dev_flow->dv.actions[actions_n++] =
7560 dev_flow->dv.tag_resource->action;
7562 case RTE_FLOW_ACTION_TYPE_SET_META:
7563 if (flow_dv_convert_action_set_meta
7564 (dev, mhdr_res, attr,
7565 (const struct rte_flow_action_set_meta *)
7566 actions->conf, error))
7568 action_flags |= MLX5_FLOW_ACTION_SET_META;
7570 case RTE_FLOW_ACTION_TYPE_SET_TAG:
7571 if (flow_dv_convert_action_set_tag
7573 (const struct rte_flow_action_set_tag *)
7574 actions->conf, error))
7576 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7578 case RTE_FLOW_ACTION_TYPE_DROP:
7579 action_flags |= MLX5_FLOW_ACTION_DROP;
7580 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
7582 case RTE_FLOW_ACTION_TYPE_QUEUE:
7583 queue = actions->conf;
7584 rss_desc->queue_num = 1;
7585 rss_desc->queue[0] = queue->index;
7586 action_flags |= MLX5_FLOW_ACTION_QUEUE;
7587 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
7589 case RTE_FLOW_ACTION_TYPE_RSS:
7590 rss = actions->conf;
7591 memcpy(rss_desc->queue, rss->queue,
7592 rss->queue_num * sizeof(uint16_t));
7593 rss_desc->queue_num = rss->queue_num;
7594 /* NULL RSS key indicates default RSS key. */
7595 rss_key = !rss->key ? rss_hash_default_key : rss->key;
7596 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
7598 * rss->level and rss.types should be set in advance
7599 * when expanding items for RSS.
7601 action_flags |= MLX5_FLOW_ACTION_RSS;
7602 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
7604 case RTE_FLOW_ACTION_TYPE_COUNT:
7605 if (!dev_conf->devx) {
7606 rte_errno = ENOTSUP;
7609 flow->counter = flow_dv_counter_alloc(dev,
7612 dev_flow->dv.group);
7615 dev_flow->dv.actions[actions_n++] =
7616 (flow_dv_counter_get_by_idx(dev,
7617 flow->counter, NULL))->action;
7618 action_flags |= MLX5_FLOW_ACTION_COUNT;
7621 if (rte_errno == ENOTSUP)
7622 return rte_flow_error_set
7624 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7626 "count action not supported");
7628 return rte_flow_error_set
7630 RTE_FLOW_ERROR_TYPE_ACTION,
7632 "cannot create counter"
7635 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
7636 dev_flow->dv.actions[actions_n++] =
7637 priv->sh->pop_vlan_action;
7638 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
7640 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
7641 if (!(action_flags &
7642 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
7643 flow_dev_get_vlan_info_from_items(items, &vlan);
7644 vlan.eth_proto = rte_be_to_cpu_16
7645 ((((const struct rte_flow_action_of_push_vlan *)
7646 actions->conf)->ethertype));
7647 found_action = mlx5_flow_find_action
7649 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
7651 mlx5_update_vlan_vid_pcp(found_action, &vlan);
7652 found_action = mlx5_flow_find_action
7654 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
7656 mlx5_update_vlan_vid_pcp(found_action, &vlan);
7657 if (flow_dv_create_action_push_vlan
7658 (dev, attr, &vlan, dev_flow, error))
7660 dev_flow->dv.actions[actions_n++] =
7661 dev_flow->dv.push_vlan_res->action;
7662 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
7664 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
7665 /* of_vlan_push action handled this action */
7666 MLX5_ASSERT(action_flags &
7667 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
7669 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
7670 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
7672 flow_dev_get_vlan_info_from_items(items, &vlan);
7673 mlx5_update_vlan_vid_pcp(actions, &vlan);
7674 /* If no VLAN push - this is a modify header action */
7675 if (flow_dv_convert_action_modify_vlan_vid
7676 (mhdr_res, actions, error))
7678 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
7680 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
7681 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
7682 if (flow_dv_create_action_l2_encap(dev, actions,
7687 dev_flow->dv.actions[actions_n++] =
7688 dev_flow->dv.encap_decap->verbs_action;
7689 action_flags |= MLX5_FLOW_ACTION_ENCAP;
7691 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
7692 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
7693 if (flow_dv_create_action_l2_decap(dev, dev_flow,
7697 dev_flow->dv.actions[actions_n++] =
7698 dev_flow->dv.encap_decap->verbs_action;
7699 action_flags |= MLX5_FLOW_ACTION_DECAP;
7701 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
7702 /* Handle encap with preceding decap. */
7703 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
7704 if (flow_dv_create_action_raw_encap
7705 (dev, actions, dev_flow, attr, error))
7707 dev_flow->dv.actions[actions_n++] =
7708 dev_flow->dv.encap_decap->verbs_action;
7710 /* Handle encap without preceding decap. */
7711 if (flow_dv_create_action_l2_encap
7712 (dev, actions, dev_flow, attr->transfer,
7715 dev_flow->dv.actions[actions_n++] =
7716 dev_flow->dv.encap_decap->verbs_action;
7718 action_flags |= MLX5_FLOW_ACTION_ENCAP;
7720 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
7721 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
7723 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
7724 if (flow_dv_create_action_l2_decap
7725 (dev, dev_flow, attr->transfer, error))
7727 dev_flow->dv.actions[actions_n++] =
7728 dev_flow->dv.encap_decap->verbs_action;
7730 /* If decap is followed by encap, handle it at encap. */
7731 action_flags |= MLX5_FLOW_ACTION_DECAP;
7733 case RTE_FLOW_ACTION_TYPE_JUMP:
7734 jump_data = action->conf;
7735 ret = mlx5_flow_group_to_table(attr, dev_flow->external,
7737 !!priv->fdb_def_rule,
7741 tbl = flow_dv_tbl_resource_get(dev, table,
7743 attr->transfer, error);
7745 return rte_flow_error_set
7747 RTE_FLOW_ERROR_TYPE_ACTION,
7749 "cannot create jump action.");
7750 if (flow_dv_jump_tbl_resource_register
7751 (dev, tbl, dev_flow, error)) {
7752 flow_dv_tbl_resource_release(dev, tbl);
7753 return rte_flow_error_set
7755 RTE_FLOW_ERROR_TYPE_ACTION,
7757 "cannot create jump action.");
7759 dev_flow->dv.actions[actions_n++] =
7760 dev_flow->dv.jump->action;
7761 action_flags |= MLX5_FLOW_ACTION_JUMP;
7762 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
7764 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
7765 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
7766 if (flow_dv_convert_action_modify_mac
7767 (mhdr_res, actions, error))
7769 action_flags |= actions->type ==
7770 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
7771 MLX5_FLOW_ACTION_SET_MAC_SRC :
7772 MLX5_FLOW_ACTION_SET_MAC_DST;
7774 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
7775 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
7776 if (flow_dv_convert_action_modify_ipv4
7777 (mhdr_res, actions, error))
7779 action_flags |= actions->type ==
7780 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
7781 MLX5_FLOW_ACTION_SET_IPV4_SRC :
7782 MLX5_FLOW_ACTION_SET_IPV4_DST;
7784 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
7785 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
7786 if (flow_dv_convert_action_modify_ipv6
7787 (mhdr_res, actions, error))
7789 action_flags |= actions->type ==
7790 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
7791 MLX5_FLOW_ACTION_SET_IPV6_SRC :
7792 MLX5_FLOW_ACTION_SET_IPV6_DST;
7794 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
7795 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
7796 if (flow_dv_convert_action_modify_tp
7797 (mhdr_res, actions, items,
7798 &flow_attr, dev_flow, !!(action_flags &
7799 MLX5_FLOW_ACTION_DECAP), error))
7801 action_flags |= actions->type ==
7802 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
7803 MLX5_FLOW_ACTION_SET_TP_SRC :
7804 MLX5_FLOW_ACTION_SET_TP_DST;
7806 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
7807 if (flow_dv_convert_action_modify_dec_ttl
7808 (mhdr_res, items, &flow_attr, dev_flow,
7810 MLX5_FLOW_ACTION_DECAP), error))
7812 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
7814 case RTE_FLOW_ACTION_TYPE_SET_TTL:
7815 if (flow_dv_convert_action_modify_ttl
7816 (mhdr_res, actions, items, &flow_attr,
7817 dev_flow, !!(action_flags &
7818 MLX5_FLOW_ACTION_DECAP), error))
7820 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
7822 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
7823 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
7824 if (flow_dv_convert_action_modify_tcp_seq
7825 (mhdr_res, actions, error))
7827 action_flags |= actions->type ==
7828 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
7829 MLX5_FLOW_ACTION_INC_TCP_SEQ :
7830 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
7833 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
7834 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
7835 if (flow_dv_convert_action_modify_tcp_ack
7836 (mhdr_res, actions, error))
7838 action_flags |= actions->type ==
7839 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
7840 MLX5_FLOW_ACTION_INC_TCP_ACK :
7841 MLX5_FLOW_ACTION_DEC_TCP_ACK;
7843 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
7844 if (flow_dv_convert_action_set_reg
7845 (mhdr_res, actions, error))
7847 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7849 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
7850 if (flow_dv_convert_action_copy_mreg
7851 (dev, mhdr_res, actions, error))
7853 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7855 case RTE_FLOW_ACTION_TYPE_METER:
7856 mtr = actions->conf;
7858 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
7861 return rte_flow_error_set(error,
7863 RTE_FLOW_ERROR_TYPE_ACTION,
7866 "or invalid parameters");
7867 flow->meter = fm->idx;
7869 /* Set the meter action. */
7871 fm = mlx5_ipool_get(priv->sh->ipool
7872 [MLX5_IPOOL_MTR], flow->meter);
7874 return rte_flow_error_set(error,
7876 RTE_FLOW_ERROR_TYPE_ACTION,
7879 "or invalid parameters");
7881 dev_flow->dv.actions[actions_n++] =
7882 fm->mfts->meter_action;
7883 action_flags |= MLX5_FLOW_ACTION_METER;
7885 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
7886 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
7889 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
7891 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
7892 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
7895 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
7897 case RTE_FLOW_ACTION_TYPE_END:
7899 if (mhdr_res->actions_num) {
7900 /* create modify action if needed. */
7901 if (flow_dv_modify_hdr_resource_register
7902 (dev, mhdr_res, dev_flow, error))
7904 dev_flow->dv.actions[modify_action_position] =
7905 handle->dvh.modify_hdr->verbs_action;
7911 if (mhdr_res->actions_num &&
7912 modify_action_position == UINT32_MAX)
7913 modify_action_position = actions_n++;
7915 dev_flow->dv.actions_n = actions_n;
7916 dev_flow->act_flags = action_flags;
7917 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
7918 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
7919 int item_type = items->type;
7921 switch (item_type) {
7922 case RTE_FLOW_ITEM_TYPE_PORT_ID:
7923 flow_dv_translate_item_port_id(dev, match_mask,
7924 match_value, items);
7925 last_item = MLX5_FLOW_ITEM_PORT_ID;
7927 case RTE_FLOW_ITEM_TYPE_ETH:
7928 flow_dv_translate_item_eth(match_mask, match_value,
7930 matcher.priority = MLX5_PRIORITY_MAP_L2;
7931 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
7932 MLX5_FLOW_LAYER_OUTER_L2;
7934 case RTE_FLOW_ITEM_TYPE_VLAN:
7935 flow_dv_translate_item_vlan(dev_flow,
7936 match_mask, match_value,
7938 matcher.priority = MLX5_PRIORITY_MAP_L2;
7939 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
7940 MLX5_FLOW_LAYER_INNER_VLAN) :
7941 (MLX5_FLOW_LAYER_OUTER_L2 |
7942 MLX5_FLOW_LAYER_OUTER_VLAN);
7944 case RTE_FLOW_ITEM_TYPE_IPV4:
7945 mlx5_flow_tunnel_ip_check(items, next_protocol,
7946 &item_flags, &tunnel);
7947 flow_dv_translate_item_ipv4(match_mask, match_value,
7948 items, item_flags, tunnel,
7949 dev_flow->dv.group);
7950 matcher.priority = MLX5_PRIORITY_MAP_L3;
7951 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
7952 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
7953 if (items->mask != NULL &&
7954 ((const struct rte_flow_item_ipv4 *)
7955 items->mask)->hdr.next_proto_id) {
7957 ((const struct rte_flow_item_ipv4 *)
7958 (items->spec))->hdr.next_proto_id;
7960 ((const struct rte_flow_item_ipv4 *)
7961 (items->mask))->hdr.next_proto_id;
7963 /* Reset for inner layer. */
7964 next_protocol = 0xff;
7967 case RTE_FLOW_ITEM_TYPE_IPV6:
7968 mlx5_flow_tunnel_ip_check(items, next_protocol,
7969 &item_flags, &tunnel);
7970 flow_dv_translate_item_ipv6(match_mask, match_value,
7971 items, item_flags, tunnel,
7972 dev_flow->dv.group);
7973 matcher.priority = MLX5_PRIORITY_MAP_L3;
7974 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
7975 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
7976 if (items->mask != NULL &&
7977 ((const struct rte_flow_item_ipv6 *)
7978 items->mask)->hdr.proto) {
7980 ((const struct rte_flow_item_ipv6 *)
7981 items->spec)->hdr.proto;
7983 ((const struct rte_flow_item_ipv6 *)
7984 items->mask)->hdr.proto;
7986 /* Reset for inner layer. */
7987 next_protocol = 0xff;
7990 case RTE_FLOW_ITEM_TYPE_TCP:
7991 flow_dv_translate_item_tcp(match_mask, match_value,
7993 matcher.priority = MLX5_PRIORITY_MAP_L4;
7994 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
7995 MLX5_FLOW_LAYER_OUTER_L4_TCP;
7997 case RTE_FLOW_ITEM_TYPE_UDP:
7998 flow_dv_translate_item_udp(match_mask, match_value,
8000 matcher.priority = MLX5_PRIORITY_MAP_L4;
8001 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
8002 MLX5_FLOW_LAYER_OUTER_L4_UDP;
8004 case RTE_FLOW_ITEM_TYPE_GRE:
8005 flow_dv_translate_item_gre(match_mask, match_value,
8007 matcher.priority = rss_desc->level >= 2 ?
8008 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8009 last_item = MLX5_FLOW_LAYER_GRE;
8011 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
8012 flow_dv_translate_item_gre_key(match_mask,
8013 match_value, items);
8014 last_item = MLX5_FLOW_LAYER_GRE_KEY;
8016 case RTE_FLOW_ITEM_TYPE_NVGRE:
8017 flow_dv_translate_item_nvgre(match_mask, match_value,
8019 matcher.priority = rss_desc->level >= 2 ?
8020 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8021 last_item = MLX5_FLOW_LAYER_GRE;
8023 case RTE_FLOW_ITEM_TYPE_VXLAN:
8024 flow_dv_translate_item_vxlan(match_mask, match_value,
8026 matcher.priority = rss_desc->level >= 2 ?
8027 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8028 last_item = MLX5_FLOW_LAYER_VXLAN;
8030 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
8031 flow_dv_translate_item_vxlan_gpe(match_mask,
8034 matcher.priority = rss_desc->level >= 2 ?
8035 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8036 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
8038 case RTE_FLOW_ITEM_TYPE_GENEVE:
8039 flow_dv_translate_item_geneve(match_mask, match_value,
8041 matcher.priority = rss_desc->level >= 2 ?
8042 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8043 last_item = MLX5_FLOW_LAYER_GENEVE;
8045 case RTE_FLOW_ITEM_TYPE_MPLS:
8046 flow_dv_translate_item_mpls(match_mask, match_value,
8047 items, last_item, tunnel);
8048 matcher.priority = rss_desc->level >= 2 ?
8049 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8050 last_item = MLX5_FLOW_LAYER_MPLS;
8052 case RTE_FLOW_ITEM_TYPE_MARK:
8053 flow_dv_translate_item_mark(dev, match_mask,
8054 match_value, items);
8055 last_item = MLX5_FLOW_ITEM_MARK;
8057 case RTE_FLOW_ITEM_TYPE_META:
8058 flow_dv_translate_item_meta(dev, match_mask,
8059 match_value, attr, items);
8060 last_item = MLX5_FLOW_ITEM_METADATA;
8062 case RTE_FLOW_ITEM_TYPE_ICMP:
8063 flow_dv_translate_item_icmp(match_mask, match_value,
8065 last_item = MLX5_FLOW_LAYER_ICMP;
8067 case RTE_FLOW_ITEM_TYPE_ICMP6:
8068 flow_dv_translate_item_icmp6(match_mask, match_value,
8070 last_item = MLX5_FLOW_LAYER_ICMP6;
8072 case RTE_FLOW_ITEM_TYPE_TAG:
8073 flow_dv_translate_item_tag(dev, match_mask,
8074 match_value, items);
8075 last_item = MLX5_FLOW_ITEM_TAG;
8077 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
8078 flow_dv_translate_mlx5_item_tag(dev, match_mask,
8079 match_value, items);
8080 last_item = MLX5_FLOW_ITEM_TAG;
8082 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
8083 flow_dv_translate_item_tx_queue(dev, match_mask,
8086 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
8088 case RTE_FLOW_ITEM_TYPE_GTP:
8089 flow_dv_translate_item_gtp(match_mask, match_value,
8091 matcher.priority = rss_desc->level >= 2 ?
8092 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8093 last_item = MLX5_FLOW_LAYER_GTP;
8098 item_flags |= last_item;
8101 * When E-Switch mode is enabled, we have two cases where we need to
8102 * set the source port manually.
8103 * The first one, is in case of Nic steering rule, and the second is
8104 * E-Switch rule where no port_id item was found. In both cases
8105 * the source port is set according the current port in use.
8107 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
8108 (priv->representor || priv->master)) {
8109 if (flow_dv_translate_item_port_id(dev, match_mask,
8113 #ifdef RTE_LIBRTE_MLX5_DEBUG
8114 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
8115 dev_flow->dv.value.buf));
8118 * Layers may be already initialized from prefix flow if this dev_flow
8119 * is the suffix flow.
8121 handle->layers |= item_flags;
8122 if (action_flags & MLX5_FLOW_ACTION_RSS)
8123 flow_dv_hashfields_set(dev_flow, rss_desc);
8124 /* Register matcher. */
8125 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
8127 matcher.priority = mlx5_flow_adjust_priority(dev, priority,
8129 /* reserved field no needs to be set to 0 here. */
8130 tbl_key.domain = attr->transfer;
8131 tbl_key.direction = attr->egress;
8132 tbl_key.table_id = dev_flow->dv.group;
8133 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
8139 * Apply the flow to the NIC, lock free,
8140 * (mutex should be acquired by caller).
8143 * Pointer to the Ethernet device structure.
8144 * @param[in, out] flow
8145 * Pointer to flow structure.
8147 * Pointer to error structure.
8150 * 0 on success, a negative errno value otherwise and rte_errno is set.
8153 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
8154 struct rte_flow_error *error)
8156 struct mlx5_flow_dv_workspace *dv;
8157 struct mlx5_flow_handle *dh;
8158 struct mlx5_flow_handle_dv *dv_h;
8159 struct mlx5_flow *dev_flow;
8160 struct mlx5_priv *priv = dev->data->dev_private;
8161 uint32_t handle_idx;
8166 for (idx = priv->flow_idx - 1; idx >= priv->flow_nested_idx; idx--) {
8167 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[idx];
8169 dh = dev_flow->handle;
8172 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8174 dv->actions[n++] = priv->sh->esw_drop_action;
8176 struct mlx5_hrxq *drop_hrxq;
8177 drop_hrxq = mlx5_hrxq_drop_new(dev);
8181 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8183 "cannot get drop hash queue");
8187 * Drop queues will be released by the specify
8188 * mlx5_hrxq_drop_release() function. Assign
8189 * the special index to hrxq to mark the queue
8190 * has been allocated.
8192 dh->rix_hrxq = UINT32_MAX;
8193 dv->actions[n++] = drop_hrxq->action;
8195 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8196 struct mlx5_hrxq *hrxq;
8198 struct mlx5_flow_rss_desc *rss_desc =
8199 &((struct mlx5_flow_rss_desc *)priv->rss_desc)
8200 [!!priv->flow_nested_idx];
8202 MLX5_ASSERT(rss_desc->queue_num);
8203 hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
8204 MLX5_RSS_HASH_KEY_LEN,
8205 dev_flow->hash_fields,
8207 rss_desc->queue_num);
8209 hrxq_idx = mlx5_hrxq_new
8210 (dev, rss_desc->key,
8211 MLX5_RSS_HASH_KEY_LEN,
8212 dev_flow->hash_fields,
8214 rss_desc->queue_num,
8216 MLX5_FLOW_LAYER_TUNNEL));
8218 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8223 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8224 "cannot get hash queue");
8227 dh->rix_hrxq = hrxq_idx;
8228 dv->actions[n++] = hrxq->action;
8231 mlx5_glue->dv_create_flow(dv_h->matcher->matcher_object,
8232 (void *)&dv->value, n,
8235 rte_flow_error_set(error, errno,
8236 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8238 "hardware refuses to create flow");
8241 if (priv->vmwa_context &&
8242 dh->vf_vlan.tag && !dh->vf_vlan.created) {
8244 * The rule contains the VLAN pattern.
8245 * For VF we are going to create VLAN
8246 * interface to make hypervisor set correct
8247 * e-Switch vport context.
8249 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
8254 err = rte_errno; /* Save rte_errno before cleanup. */
8255 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
8256 handle_idx, dh, next) {
8257 /* hrxq is union, don't clear it if the flag is not set. */
8259 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8260 mlx5_hrxq_drop_release(dev);
8262 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8263 mlx5_hrxq_release(dev, dh->rix_hrxq);
8267 if (dh->vf_vlan.tag && dh->vf_vlan.created)
8268 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8270 rte_errno = err; /* Restore rte_errno. */
8275 * Release the flow matcher.
8278 * Pointer to Ethernet device.
8280 * Pointer to mlx5_flow_handle.
8283 * 1 while a reference on it exists, 0 when freed.
8286 flow_dv_matcher_release(struct rte_eth_dev *dev,
8287 struct mlx5_flow_handle *handle)
8289 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
8291 MLX5_ASSERT(matcher->matcher_object);
8292 DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
8293 dev->data->port_id, (void *)matcher,
8294 rte_atomic32_read(&matcher->refcnt));
8295 if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
8296 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8297 (matcher->matcher_object));
8298 LIST_REMOVE(matcher, next);
8299 /* table ref-- in release interface. */
8300 flow_dv_tbl_resource_release(dev, matcher->tbl);
8302 DRV_LOG(DEBUG, "port %u matcher %p: removed",
8303 dev->data->port_id, (void *)matcher);
8310 * Release an encap/decap resource.
8313 * Pointer to Ethernet device.
8315 * Pointer to mlx5_flow_handle.
8318 * 1 while a reference on it exists, 0 when freed.
8321 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
8322 struct mlx5_flow_handle *handle)
8324 struct mlx5_priv *priv = dev->data->dev_private;
8325 uint32_t idx = handle->dvh.rix_encap_decap;
8326 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
8328 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
8330 if (!cache_resource)
8332 MLX5_ASSERT(cache_resource->verbs_action);
8333 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
8334 (void *)cache_resource,
8335 rte_atomic32_read(&cache_resource->refcnt));
8336 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8337 claim_zero(mlx5_glue->destroy_flow_action
8338 (cache_resource->verbs_action));
8339 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
8340 &priv->sh->encaps_decaps, idx,
8341 cache_resource, next);
8342 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
8343 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
8344 (void *)cache_resource);
8351 * Release an jump to table action resource.
8354 * Pointer to Ethernet device.
8356 * Pointer to mlx5_flow_handle.
8359 * 1 while a reference on it exists, 0 when freed.
8362 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
8363 struct mlx5_flow_handle *handle)
8365 struct mlx5_priv *priv = dev->data->dev_private;
8366 struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
8367 struct mlx5_flow_tbl_data_entry *tbl_data;
8369 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
8373 cache_resource = &tbl_data->jump;
8374 MLX5_ASSERT(cache_resource->action);
8375 DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
8376 (void *)cache_resource,
8377 rte_atomic32_read(&cache_resource->refcnt));
8378 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8379 claim_zero(mlx5_glue->destroy_flow_action
8380 (cache_resource->action));
8381 /* jump action memory free is inside the table release. */
8382 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
8383 DRV_LOG(DEBUG, "jump table resource %p: removed",
8384 (void *)cache_resource);
8391 * Release a modify-header resource.
8394 * Pointer to mlx5_flow_handle.
8397 * 1 while a reference on it exists, 0 when freed.
8400 flow_dv_modify_hdr_resource_release(struct mlx5_flow_handle *handle)
8402 struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
8403 handle->dvh.modify_hdr;
8405 MLX5_ASSERT(cache_resource->verbs_action);
8406 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
8407 (void *)cache_resource,
8408 rte_atomic32_read(&cache_resource->refcnt));
8409 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8410 claim_zero(mlx5_glue->destroy_flow_action
8411 (cache_resource->verbs_action));
8412 LIST_REMOVE(cache_resource, next);
8413 rte_free(cache_resource);
8414 DRV_LOG(DEBUG, "modify-header resource %p: removed",
8415 (void *)cache_resource);
8422 * Release port ID action resource.
8425 * Pointer to Ethernet device.
8427 * Pointer to mlx5_flow_handle.
8430 * 1 while a reference on it exists, 0 when freed.
8433 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
8434 struct mlx5_flow_handle *handle)
8436 struct mlx5_priv *priv = dev->data->dev_private;
8437 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
8438 uint32_t idx = handle->rix_port_id_action;
8440 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
8442 if (!cache_resource)
8444 MLX5_ASSERT(cache_resource->action);
8445 DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
8446 (void *)cache_resource,
8447 rte_atomic32_read(&cache_resource->refcnt));
8448 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8449 claim_zero(mlx5_glue->destroy_flow_action
8450 (cache_resource->action));
8451 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
8452 &priv->sh->port_id_action_list, idx,
8453 cache_resource, next);
8454 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
8455 DRV_LOG(DEBUG, "port id action resource %p: removed",
8456 (void *)cache_resource);
8463 * Release push vlan action resource.
8466 * Pointer to Ethernet device.
8468 * Pointer to mlx5_flow_handle.
8471 * 1 while a reference on it exists, 0 when freed.
8474 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
8475 struct mlx5_flow_handle *handle)
8477 struct mlx5_priv *priv = dev->data->dev_private;
8478 uint32_t idx = handle->dvh.rix_push_vlan;
8479 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
8481 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
8483 if (!cache_resource)
8485 MLX5_ASSERT(cache_resource->action);
8486 DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
8487 (void *)cache_resource,
8488 rte_atomic32_read(&cache_resource->refcnt));
8489 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8490 claim_zero(mlx5_glue->destroy_flow_action
8491 (cache_resource->action));
8492 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
8493 &priv->sh->push_vlan_action_list, idx,
8494 cache_resource, next);
8495 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
8496 DRV_LOG(DEBUG, "push vlan action resource %p: removed",
8497 (void *)cache_resource);
8504 * Release the fate resource.
8507 * Pointer to Ethernet device.
8509 * Pointer to mlx5_flow_handle.
8512 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
8513 struct mlx5_flow_handle *handle)
8515 if (!handle->rix_fate)
8517 if (handle->fate_action == MLX5_FLOW_FATE_DROP)
8518 mlx5_hrxq_drop_release(dev);
8519 else if (handle->fate_action == MLX5_FLOW_FATE_QUEUE)
8520 mlx5_hrxq_release(dev, handle->rix_hrxq);
8521 else if (handle->fate_action == MLX5_FLOW_FATE_JUMP)
8522 flow_dv_jump_tbl_resource_release(dev, handle);
8523 else if (handle->fate_action == MLX5_FLOW_FATE_PORT_ID)
8524 flow_dv_port_id_action_resource_release(dev, handle);
8526 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
8527 handle->rix_fate = 0;
8531 * Remove the flow from the NIC but keeps it in memory.
8532 * Lock free, (mutex should be acquired by caller).
8535 * Pointer to Ethernet device.
8536 * @param[in, out] flow
8537 * Pointer to flow structure.
8540 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
8542 struct mlx5_flow_handle *dh;
8543 uint32_t handle_idx;
8544 struct mlx5_priv *priv = dev->data->dev_private;
8548 handle_idx = flow->dev_handles;
8549 while (handle_idx) {
8550 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8555 claim_zero(mlx5_glue->dv_destroy_flow(dh->ib_flow));
8558 if (dh->fate_action == MLX5_FLOW_FATE_DROP ||
8559 dh->fate_action == MLX5_FLOW_FATE_QUEUE)
8560 flow_dv_fate_resource_release(dev, dh);
8561 if (dh->vf_vlan.tag && dh->vf_vlan.created)
8562 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8563 handle_idx = dh->next.next;
8568 * Remove the flow from the NIC and the memory.
8569 * Lock free, (mutex should be acquired by caller).
8572 * Pointer to the Ethernet device structure.
8573 * @param[in, out] flow
8574 * Pointer to flow structure.
8577 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
8579 struct mlx5_flow_handle *dev_handle;
8580 struct mlx5_priv *priv = dev->data->dev_private;
8584 __flow_dv_remove(dev, flow);
8585 if (flow->counter) {
8586 flow_dv_counter_release(dev, flow->counter);
8590 struct mlx5_flow_meter *fm;
8592 fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
8595 mlx5_flow_meter_detach(fm);
8598 while (flow->dev_handles) {
8599 uint32_t tmp_idx = flow->dev_handles;
8601 dev_handle = mlx5_ipool_get(priv->sh->ipool
8602 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
8605 flow->dev_handles = dev_handle->next.next;
8606 if (dev_handle->dvh.matcher)
8607 flow_dv_matcher_release(dev, dev_handle);
8608 if (dev_handle->dvh.rix_encap_decap)
8609 flow_dv_encap_decap_resource_release(dev, dev_handle);
8610 if (dev_handle->dvh.modify_hdr)
8611 flow_dv_modify_hdr_resource_release(dev_handle);
8612 if (dev_handle->dvh.rix_push_vlan)
8613 flow_dv_push_vlan_action_resource_release(dev,
8615 if (dev_handle->dvh.rix_tag)
8616 flow_dv_tag_release(dev,
8617 dev_handle->dvh.rix_tag);
8618 flow_dv_fate_resource_release(dev, dev_handle);
8619 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8625 * Query a dv flow rule for its statistics via devx.
8628 * Pointer to Ethernet device.
8630 * Pointer to the sub flow.
8632 * data retrieved by the query.
8634 * Perform verbose error reporting if not NULL.
8637 * 0 on success, a negative errno value otherwise and rte_errno is set.
8640 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
8641 void *data, struct rte_flow_error *error)
8643 struct mlx5_priv *priv = dev->data->dev_private;
8644 struct rte_flow_query_count *qc = data;
8646 if (!priv->config.devx)
8647 return rte_flow_error_set(error, ENOTSUP,
8648 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8650 "counters are not supported");
8651 if (flow->counter) {
8652 uint64_t pkts, bytes;
8653 struct mlx5_flow_counter *cnt;
8655 cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
8657 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
8661 return rte_flow_error_set(error, -err,
8662 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8663 NULL, "cannot read counters");
8666 qc->hits = pkts - cnt->hits;
8667 qc->bytes = bytes - cnt->bytes;
8674 return rte_flow_error_set(error, EINVAL,
8675 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8677 "counters are not available");
8683 * @see rte_flow_query()
8687 flow_dv_query(struct rte_eth_dev *dev,
8688 struct rte_flow *flow __rte_unused,
8689 const struct rte_flow_action *actions __rte_unused,
8690 void *data __rte_unused,
8691 struct rte_flow_error *error __rte_unused)
8695 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
8696 switch (actions->type) {
8697 case RTE_FLOW_ACTION_TYPE_VOID:
8699 case RTE_FLOW_ACTION_TYPE_COUNT:
8700 ret = flow_dv_query_count(dev, flow, data, error);
8703 return rte_flow_error_set(error, ENOTSUP,
8704 RTE_FLOW_ERROR_TYPE_ACTION,
8706 "action not supported");
8713 * Destroy the meter table set.
8714 * Lock free, (mutex should be acquired by caller).
8717 * Pointer to Ethernet device.
8719 * Pointer to the meter table set.
8725 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
8726 struct mlx5_meter_domains_infos *tbl)
8728 struct mlx5_priv *priv = dev->data->dev_private;
8729 struct mlx5_meter_domains_infos *mtd =
8730 (struct mlx5_meter_domains_infos *)tbl;
8732 if (!mtd || !priv->config.dv_flow_en)
8734 if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
8735 claim_zero(mlx5_glue->dv_destroy_flow
8736 (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
8737 if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
8738 claim_zero(mlx5_glue->dv_destroy_flow
8739 (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
8740 if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
8741 claim_zero(mlx5_glue->dv_destroy_flow
8742 (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
8743 if (mtd->egress.color_matcher)
8744 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8745 (mtd->egress.color_matcher));
8746 if (mtd->egress.any_matcher)
8747 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8748 (mtd->egress.any_matcher));
8749 if (mtd->egress.tbl)
8750 flow_dv_tbl_resource_release(dev, mtd->egress.tbl);
8751 if (mtd->egress.sfx_tbl)
8752 flow_dv_tbl_resource_release(dev, mtd->egress.sfx_tbl);
8753 if (mtd->ingress.color_matcher)
8754 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8755 (mtd->ingress.color_matcher));
8756 if (mtd->ingress.any_matcher)
8757 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8758 (mtd->ingress.any_matcher));
8759 if (mtd->ingress.tbl)
8760 flow_dv_tbl_resource_release(dev, mtd->ingress.tbl);
8761 if (mtd->ingress.sfx_tbl)
8762 flow_dv_tbl_resource_release(dev, mtd->ingress.sfx_tbl);
8763 if (mtd->transfer.color_matcher)
8764 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8765 (mtd->transfer.color_matcher));
8766 if (mtd->transfer.any_matcher)
8767 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8768 (mtd->transfer.any_matcher));
8769 if (mtd->transfer.tbl)
8770 flow_dv_tbl_resource_release(dev, mtd->transfer.tbl);
8771 if (mtd->transfer.sfx_tbl)
8772 flow_dv_tbl_resource_release(dev, mtd->transfer.sfx_tbl);
8774 claim_zero(mlx5_glue->destroy_flow_action(mtd->drop_actn));
8779 /* Number of meter flow actions, count and jump or count and drop. */
8780 #define METER_ACTIONS 2
8783 * Create specify domain meter table and suffix table.
8786 * Pointer to Ethernet device.
8787 * @param[in,out] mtb
8788 * Pointer to DV meter table set.
8791 * @param[in] transfer
8793 * @param[in] color_reg_c_idx
8794 * Reg C index for color match.
8797 * 0 on success, -1 otherwise and rte_errno is set.
8800 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
8801 struct mlx5_meter_domains_infos *mtb,
8802 uint8_t egress, uint8_t transfer,
8803 uint32_t color_reg_c_idx)
8805 struct mlx5_priv *priv = dev->data->dev_private;
8806 struct mlx5_ibv_shared *sh = priv->sh;
8807 struct mlx5_flow_dv_match_params mask = {
8808 .size = sizeof(mask.buf),
8810 struct mlx5_flow_dv_match_params value = {
8811 .size = sizeof(value.buf),
8813 struct mlx5dv_flow_matcher_attr dv_attr = {
8814 .type = IBV_FLOW_ATTR_NORMAL,
8816 .match_criteria_enable = 0,
8817 .match_mask = (void *)&mask,
8819 void *actions[METER_ACTIONS];
8820 struct mlx5_meter_domain_info *dtb;
8821 struct rte_flow_error error;
8825 dtb = &mtb->transfer;
8829 dtb = &mtb->ingress;
8830 /* Create the meter table with METER level. */
8831 dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
8832 egress, transfer, &error);
8834 DRV_LOG(ERR, "Failed to create meter policer table.");
8837 /* Create the meter suffix table with SUFFIX level. */
8838 dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
8839 MLX5_FLOW_TABLE_LEVEL_SUFFIX,
8840 egress, transfer, &error);
8841 if (!dtb->sfx_tbl) {
8842 DRV_LOG(ERR, "Failed to create meter suffix table.");
8845 /* Create matchers, Any and Color. */
8846 dv_attr.priority = 3;
8847 dv_attr.match_criteria_enable = 0;
8848 dtb->any_matcher = mlx5_glue->dv_create_flow_matcher(sh->ctx,
8851 if (!dtb->any_matcher) {
8852 DRV_LOG(ERR, "Failed to create meter"
8853 " policer default matcher.");
8856 dv_attr.priority = 0;
8857 dv_attr.match_criteria_enable =
8858 1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
8859 flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
8860 rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
8861 dtb->color_matcher = mlx5_glue->dv_create_flow_matcher(sh->ctx,
8864 if (!dtb->color_matcher) {
8865 DRV_LOG(ERR, "Failed to create meter policer color matcher.");
8868 if (mtb->count_actns[RTE_MTR_DROPPED])
8869 actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
8870 actions[i++] = mtb->drop_actn;
8871 /* Default rule: lowest priority, match any, actions: drop. */
8872 dtb->policer_rules[RTE_MTR_DROPPED] =
8873 mlx5_glue->dv_create_flow(dtb->any_matcher,
8874 (void *)&value, i, actions);
8875 if (!dtb->policer_rules[RTE_MTR_DROPPED]) {
8876 DRV_LOG(ERR, "Failed to create meter policer drop rule.");
8885 * Create the needed meter and suffix tables.
8886 * Lock free, (mutex should be acquired by caller).
8889 * Pointer to Ethernet device.
8891 * Pointer to the flow meter.
8894 * Pointer to table set on success, NULL otherwise and rte_errno is set.
8896 static struct mlx5_meter_domains_infos *
8897 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
8898 const struct mlx5_flow_meter *fm)
8900 struct mlx5_priv *priv = dev->data->dev_private;
8901 struct mlx5_meter_domains_infos *mtb;
8905 if (!priv->mtr_en) {
8906 rte_errno = ENOTSUP;
8909 mtb = rte_calloc(__func__, 1, sizeof(*mtb), 0);
8911 DRV_LOG(ERR, "Failed to allocate memory for meter.");
8914 /* Create meter count actions */
8915 for (i = 0; i <= RTE_MTR_DROPPED; i++) {
8916 struct mlx5_flow_counter *cnt;
8917 if (!fm->policer_stats.cnt[i])
8919 cnt = flow_dv_counter_get_by_idx(dev,
8920 fm->policer_stats.cnt[i], NULL);
8921 mtb->count_actns[i] = cnt->action;
8923 /* Create drop action. */
8924 mtb->drop_actn = mlx5_glue->dr_create_flow_action_drop();
8925 if (!mtb->drop_actn) {
8926 DRV_LOG(ERR, "Failed to create drop action.");
8929 /* Egress meter table. */
8930 ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
8932 DRV_LOG(ERR, "Failed to prepare egress meter table.");
8935 /* Ingress meter table. */
8936 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
8938 DRV_LOG(ERR, "Failed to prepare ingress meter table.");
8941 /* FDB meter table. */
8942 if (priv->config.dv_esw_en) {
8943 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
8944 priv->mtr_color_reg);
8946 DRV_LOG(ERR, "Failed to prepare fdb meter table.");
8952 flow_dv_destroy_mtr_tbl(dev, mtb);
8957 * Destroy domain policer rule.
8960 * Pointer to domain table.
8963 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
8967 for (i = 0; i < RTE_MTR_DROPPED; i++) {
8968 if (dt->policer_rules[i]) {
8969 claim_zero(mlx5_glue->dv_destroy_flow
8970 (dt->policer_rules[i]));
8971 dt->policer_rules[i] = NULL;
8974 if (dt->jump_actn) {
8975 claim_zero(mlx5_glue->destroy_flow_action(dt->jump_actn));
8976 dt->jump_actn = NULL;
8981 * Destroy policer rules.
8984 * Pointer to Ethernet device.
8986 * Pointer to flow meter structure.
8988 * Pointer to flow attributes.
8994 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
8995 const struct mlx5_flow_meter *fm,
8996 const struct rte_flow_attr *attr)
8998 struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
9003 flow_dv_destroy_domain_policer_rule(&mtb->egress);
9005 flow_dv_destroy_domain_policer_rule(&mtb->ingress);
9007 flow_dv_destroy_domain_policer_rule(&mtb->transfer);
9012 * Create specify domain meter policer rule.
9015 * Pointer to flow meter structure.
9017 * Pointer to DV meter table set.
9018 * @param[in] mtr_reg_c
9019 * Color match REG_C.
9022 * 0 on success, -1 otherwise.
9025 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
9026 struct mlx5_meter_domain_info *dtb,
9029 struct mlx5_flow_dv_match_params matcher = {
9030 .size = sizeof(matcher.buf),
9032 struct mlx5_flow_dv_match_params value = {
9033 .size = sizeof(value.buf),
9035 struct mlx5_meter_domains_infos *mtb = fm->mfts;
9036 void *actions[METER_ACTIONS];
9039 /* Create jump action. */
9040 if (!dtb->jump_actn)
9042 mlx5_glue->dr_create_flow_action_dest_flow_tbl
9043 (dtb->sfx_tbl->obj);
9044 if (!dtb->jump_actn) {
9045 DRV_LOG(ERR, "Failed to create policer jump action.");
9048 for (i = 0; i < RTE_MTR_DROPPED; i++) {
9051 flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
9052 rte_col_2_mlx5_col(i), UINT8_MAX);
9053 if (mtb->count_actns[i])
9054 actions[j++] = mtb->count_actns[i];
9055 if (fm->action[i] == MTR_POLICER_ACTION_DROP)
9056 actions[j++] = mtb->drop_actn;
9058 actions[j++] = dtb->jump_actn;
9059 dtb->policer_rules[i] =
9060 mlx5_glue->dv_create_flow(dtb->color_matcher,
9063 if (!dtb->policer_rules[i]) {
9064 DRV_LOG(ERR, "Failed to create policer rule.");
9075 * Create policer rules.
9078 * Pointer to Ethernet device.
9080 * Pointer to flow meter structure.
9082 * Pointer to flow attributes.
9085 * 0 on success, -1 otherwise.
9088 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
9089 struct mlx5_flow_meter *fm,
9090 const struct rte_flow_attr *attr)
9092 struct mlx5_priv *priv = dev->data->dev_private;
9093 struct mlx5_meter_domains_infos *mtb = fm->mfts;
9097 ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
9098 priv->mtr_color_reg);
9100 DRV_LOG(ERR, "Failed to create egress policer.");
9104 if (attr->ingress) {
9105 ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
9106 priv->mtr_color_reg);
9108 DRV_LOG(ERR, "Failed to create ingress policer.");
9112 if (attr->transfer) {
9113 ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
9114 priv->mtr_color_reg);
9116 DRV_LOG(ERR, "Failed to create transfer policer.");
9122 flow_dv_destroy_policer_rules(dev, fm, attr);
9127 * Query a devx counter.
9130 * Pointer to the Ethernet device structure.
9132 * Index to the flow counter.
9134 * Set to clear the counter statistics.
9136 * The statistics value of packets.
9138 * The statistics value of bytes.
9141 * 0 on success, otherwise return -1.
9144 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
9145 uint64_t *pkts, uint64_t *bytes)
9147 struct mlx5_priv *priv = dev->data->dev_private;
9148 struct mlx5_flow_counter *cnt;
9149 uint64_t inn_pkts, inn_bytes;
9152 if (!priv->config.devx)
9155 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
9158 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
9159 *pkts = inn_pkts - cnt->hits;
9160 *bytes = inn_bytes - cnt->bytes;
9162 cnt->hits = inn_pkts;
9163 cnt->bytes = inn_bytes;
9169 * Mutex-protected thunk to lock-free __flow_dv_translate().
9172 flow_dv_translate(struct rte_eth_dev *dev,
9173 struct mlx5_flow *dev_flow,
9174 const struct rte_flow_attr *attr,
9175 const struct rte_flow_item items[],
9176 const struct rte_flow_action actions[],
9177 struct rte_flow_error *error)
9181 flow_dv_shared_lock(dev);
9182 ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
9183 flow_dv_shared_unlock(dev);
9188 * Mutex-protected thunk to lock-free __flow_dv_apply().
9191 flow_dv_apply(struct rte_eth_dev *dev,
9192 struct rte_flow *flow,
9193 struct rte_flow_error *error)
9197 flow_dv_shared_lock(dev);
9198 ret = __flow_dv_apply(dev, flow, error);
9199 flow_dv_shared_unlock(dev);
9204 * Mutex-protected thunk to lock-free __flow_dv_remove().
9207 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
9209 flow_dv_shared_lock(dev);
9210 __flow_dv_remove(dev, flow);
9211 flow_dv_shared_unlock(dev);
9215 * Mutex-protected thunk to lock-free __flow_dv_destroy().
9218 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
9220 flow_dv_shared_lock(dev);
9221 __flow_dv_destroy(dev, flow);
9222 flow_dv_shared_unlock(dev);
9226 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
9229 flow_dv_counter_allocate(struct rte_eth_dev *dev)
9233 flow_dv_shared_lock(dev);
9234 cnt = flow_dv_counter_alloc(dev, 0, 0, 1);
9235 flow_dv_shared_unlock(dev);
9240 * Mutex-protected thunk to lock-free flow_dv_counter_release().
9243 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
9245 flow_dv_shared_lock(dev);
9246 flow_dv_counter_release(dev, cnt);
9247 flow_dv_shared_unlock(dev);
9250 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
9251 .validate = flow_dv_validate,
9252 .prepare = flow_dv_prepare,
9253 .translate = flow_dv_translate,
9254 .apply = flow_dv_apply,
9255 .remove = flow_dv_remove,
9256 .destroy = flow_dv_destroy,
9257 .query = flow_dv_query,
9258 .create_mtr_tbls = flow_dv_create_mtr_tbl,
9259 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
9260 .create_policer_rules = flow_dv_create_policer_rules,
9261 .destroy_policer_rules = flow_dv_destroy_policer_rules,
9262 .counter_alloc = flow_dv_counter_allocate,
9263 .counter_free = flow_dv_counter_free,
9264 .counter_query = flow_dv_counter_query,
9267 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */