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.
3644 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev, uint64_t flags)
3647 * There's no way to directly query the max cap. Although it has to be
3648 * acquried by iterative trial, it is a safe assumption that more
3649 * actions are supported by FW if extensive metadata register is
3650 * supported. (Only in the root table)
3652 if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
3653 return MLX5_MAX_MODIFY_NUM;
3655 return mlx5_flow_ext_mreg_supported(dev) ?
3656 MLX5_ROOT_TBL_MODIFY_NUM :
3657 MLX5_ROOT_TBL_MODIFY_NUM_NO_MREG;
3661 * Validate the meter action.
3664 * Pointer to rte_eth_dev structure.
3665 * @param[in] action_flags
3666 * Bit-fields that holds the actions detected until now.
3668 * Pointer to the meter action.
3670 * Attributes of flow that includes this action.
3672 * Pointer to error structure.
3675 * 0 on success, a negative errno value otherwise and rte_ernno is set.
3678 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
3679 uint64_t action_flags,
3680 const struct rte_flow_action *action,
3681 const struct rte_flow_attr *attr,
3682 struct rte_flow_error *error)
3684 struct mlx5_priv *priv = dev->data->dev_private;
3685 const struct rte_flow_action_meter *am = action->conf;
3686 struct mlx5_flow_meter *fm;
3689 return rte_flow_error_set(error, EINVAL,
3690 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3691 "meter action conf is NULL");
3693 if (action_flags & MLX5_FLOW_ACTION_METER)
3694 return rte_flow_error_set(error, ENOTSUP,
3695 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3696 "meter chaining not support");
3697 if (action_flags & MLX5_FLOW_ACTION_JUMP)
3698 return rte_flow_error_set(error, ENOTSUP,
3699 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3700 "meter with jump not support");
3702 return rte_flow_error_set(error, ENOTSUP,
3703 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3705 "meter action not supported");
3706 fm = mlx5_flow_meter_find(priv, am->mtr_id);
3708 return rte_flow_error_set(error, EINVAL,
3709 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3711 if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
3712 (!fm->ingress && !attr->ingress && attr->egress) ||
3713 (!fm->egress && !attr->egress && attr->ingress))))
3714 return rte_flow_error_set(error, EINVAL,
3715 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3716 "Flow attributes are either invalid "
3717 "or have a conflict with current "
3718 "meter attributes");
3723 * Validate the modify-header IPv4 DSCP actions.
3725 * @param[in] action_flags
3726 * Holds the actions detected until now.
3728 * Pointer to the modify action.
3729 * @param[in] item_flags
3730 * Holds the items detected.
3732 * Pointer to error structure.
3735 * 0 on success, a negative errno value otherwise and rte_errno is set.
3738 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
3739 const struct rte_flow_action *action,
3740 const uint64_t item_flags,
3741 struct rte_flow_error *error)
3745 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3747 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
3748 return rte_flow_error_set(error, EINVAL,
3749 RTE_FLOW_ERROR_TYPE_ACTION,
3751 "no ipv4 item in pattern");
3757 * Validate the modify-header IPv6 DSCP actions.
3759 * @param[in] action_flags
3760 * Holds the actions detected until now.
3762 * Pointer to the modify action.
3763 * @param[in] item_flags
3764 * Holds the items detected.
3766 * Pointer to error structure.
3769 * 0 on success, a negative errno value otherwise and rte_errno is set.
3772 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
3773 const struct rte_flow_action *action,
3774 const uint64_t item_flags,
3775 struct rte_flow_error *error)
3779 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3781 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
3782 return rte_flow_error_set(error, EINVAL,
3783 RTE_FLOW_ERROR_TYPE_ACTION,
3785 "no ipv6 item in pattern");
3791 * Find existing modify-header resource or create and register a new one.
3793 * @param dev[in, out]
3794 * Pointer to rte_eth_dev structure.
3795 * @param[in, out] resource
3796 * Pointer to modify-header resource.
3797 * @parm[in, out] dev_flow
3798 * Pointer to the dev_flow.
3800 * pointer to error structure.
3803 * 0 on success otherwise -errno and errno is set.
3806 flow_dv_modify_hdr_resource_register
3807 (struct rte_eth_dev *dev,
3808 struct mlx5_flow_dv_modify_hdr_resource *resource,
3809 struct mlx5_flow *dev_flow,
3810 struct rte_flow_error *error)
3812 struct mlx5_priv *priv = dev->data->dev_private;
3813 struct mlx5_ibv_shared *sh = priv->sh;
3814 struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
3815 struct mlx5dv_dr_domain *ns;
3816 uint32_t actions_len;
3818 resource->flags = dev_flow->dv.group ? 0 :
3819 MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
3820 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
3822 return rte_flow_error_set(error, EOVERFLOW,
3823 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3824 "too many modify header items");
3825 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3826 ns = sh->fdb_domain;
3827 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
3831 /* Lookup a matching resource from cache. */
3832 actions_len = resource->actions_num * sizeof(resource->actions[0]);
3833 LIST_FOREACH(cache_resource, &sh->modify_cmds, next) {
3834 if (resource->ft_type == cache_resource->ft_type &&
3835 resource->actions_num == cache_resource->actions_num &&
3836 resource->flags == cache_resource->flags &&
3837 !memcmp((const void *)resource->actions,
3838 (const void *)cache_resource->actions,
3840 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
3841 (void *)cache_resource,
3842 rte_atomic32_read(&cache_resource->refcnt));
3843 rte_atomic32_inc(&cache_resource->refcnt);
3844 dev_flow->handle->dvh.modify_hdr = cache_resource;
3848 /* Register new modify-header resource. */
3849 cache_resource = rte_calloc(__func__, 1,
3850 sizeof(*cache_resource) + actions_len, 0);
3851 if (!cache_resource)
3852 return rte_flow_error_set(error, ENOMEM,
3853 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3854 "cannot allocate resource memory");
3855 *cache_resource = *resource;
3856 rte_memcpy(cache_resource->actions, resource->actions, actions_len);
3857 cache_resource->verbs_action =
3858 mlx5_glue->dv_create_flow_action_modify_header
3859 (sh->ctx, cache_resource->ft_type, ns,
3860 cache_resource->flags, actions_len,
3861 (uint64_t *)cache_resource->actions);
3862 if (!cache_resource->verbs_action) {
3863 rte_free(cache_resource);
3864 return rte_flow_error_set(error, ENOMEM,
3865 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3866 NULL, "cannot create action");
3868 rte_atomic32_init(&cache_resource->refcnt);
3869 rte_atomic32_inc(&cache_resource->refcnt);
3870 LIST_INSERT_HEAD(&sh->modify_cmds, cache_resource, next);
3871 dev_flow->handle->dvh.modify_hdr = cache_resource;
3872 DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
3873 (void *)cache_resource,
3874 rte_atomic32_read(&cache_resource->refcnt));
3879 * Get DV flow counter by index.
3882 * Pointer to the Ethernet device structure.
3884 * mlx5 flow counter index in the container.
3886 * mlx5 flow counter pool in the container,
3889 * Pointer to the counter, NULL otherwise.
3891 static struct mlx5_flow_counter *
3892 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
3894 struct mlx5_flow_counter_pool **ppool)
3896 struct mlx5_priv *priv = dev->data->dev_private;
3897 struct mlx5_pools_container *cont;
3898 struct mlx5_flow_counter_pool *pool;
3902 if (idx >= MLX5_CNT_BATCH_OFFSET) {
3903 idx -= MLX5_CNT_BATCH_OFFSET;
3906 cont = MLX5_CNT_CONTAINER(priv->sh, batch, 0);
3907 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cont->n);
3908 pool = cont->pools[idx / MLX5_COUNTERS_PER_POOL];
3912 return &pool->counters_raw[idx % MLX5_COUNTERS_PER_POOL];
3916 * Get a pool by devx counter ID.
3919 * Pointer to the counter container.
3921 * The counter devx ID.
3924 * The counter pool pointer if exists, NULL otherwise,
3926 static struct mlx5_flow_counter_pool *
3927 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
3930 uint32_t n_valid = rte_atomic16_read(&cont->n_valid);
3932 for (i = 0; i < n_valid; i++) {
3933 struct mlx5_flow_counter_pool *pool = cont->pools[i];
3934 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
3935 MLX5_COUNTERS_PER_POOL;
3937 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL) {
3939 * Move the pool to the head, as counter allocate
3940 * always gets the first pool in the container.
3942 if (pool != TAILQ_FIRST(&cont->pool_list)) {
3943 TAILQ_REMOVE(&cont->pool_list, pool, next);
3944 TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
3953 * Allocate a new memory for the counter values wrapped by all the needed
3957 * Pointer to the Ethernet device structure.
3959 * The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
3962 * The new memory management pointer on success, otherwise NULL and rte_errno
3965 static struct mlx5_counter_stats_mem_mng *
3966 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
3968 struct mlx5_ibv_shared *sh = ((struct mlx5_priv *)
3969 (dev->data->dev_private))->sh;
3970 struct mlx5_devx_mkey_attr mkey_attr;
3971 struct mlx5_counter_stats_mem_mng *mem_mng;
3972 volatile struct flow_counter_stats *raw_data;
3973 int size = (sizeof(struct flow_counter_stats) *
3974 MLX5_COUNTERS_PER_POOL +
3975 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
3976 sizeof(struct mlx5_counter_stats_mem_mng);
3977 uint8_t *mem = rte_calloc(__func__, 1, size, sysconf(_SC_PAGESIZE));
3984 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
3985 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
3986 mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
3987 IBV_ACCESS_LOCAL_WRITE);
3988 if (!mem_mng->umem) {
3993 mkey_attr.addr = (uintptr_t)mem;
3994 mkey_attr.size = size;
3995 mkey_attr.umem_id = mem_mng->umem->umem_id;
3996 mkey_attr.pd = sh->pdn;
3997 mkey_attr.log_entity_size = 0;
3998 mkey_attr.pg_access = 0;
3999 mkey_attr.klm_array = NULL;
4000 mkey_attr.klm_num = 0;
4001 mkey_attr.relaxed_ordering = 1;
4002 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
4004 mlx5_glue->devx_umem_dereg(mem_mng->umem);
4009 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
4010 raw_data = (volatile struct flow_counter_stats *)mem;
4011 for (i = 0; i < raws_n; ++i) {
4012 mem_mng->raws[i].mem_mng = mem_mng;
4013 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
4015 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
4020 * Resize a counter container.
4023 * Pointer to the Ethernet device structure.
4025 * Whether the pool is for counter that was allocated by batch command.
4028 * The new container pointer on success, otherwise NULL and rte_errno is set.
4030 static struct mlx5_pools_container *
4031 flow_dv_container_resize(struct rte_eth_dev *dev, uint32_t batch)
4033 struct mlx5_priv *priv = dev->data->dev_private;
4034 struct mlx5_pools_container *cont =
4035 MLX5_CNT_CONTAINER(priv->sh, batch, 0);
4036 struct mlx5_pools_container *new_cont =
4037 MLX5_CNT_CONTAINER_UNUSED(priv->sh, batch, 0);
4038 struct mlx5_counter_stats_mem_mng *mem_mng = NULL;
4039 uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
4040 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4043 /* Fallback mode has no background thread. Skip the check. */
4044 if (!priv->counter_fallback &&
4045 cont != MLX5_CNT_CONTAINER(priv->sh, batch, 1)) {
4046 /* The last resize still hasn't detected by the host thread. */
4050 new_cont->pools = rte_calloc(__func__, 1, mem_size, 0);
4051 if (!new_cont->pools) {
4056 memcpy(new_cont->pools, cont->pools, cont->n *
4057 sizeof(struct mlx5_flow_counter_pool *));
4059 * Fallback mode query the counter directly, no background query
4060 * resources are needed.
4062 if (!priv->counter_fallback) {
4063 mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
4064 MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES);
4066 rte_free(new_cont->pools);
4069 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
4070 LIST_INSERT_HEAD(&priv->sh->cmng.free_stat_raws,
4072 MLX5_CNT_CONTAINER_RESIZE +
4076 * Release the old container pools directly as no background
4077 * thread helps that.
4079 rte_free(cont->pools);
4081 new_cont->n = resize;
4082 rte_atomic16_set(&new_cont->n_valid, rte_atomic16_read(&cont->n_valid));
4083 TAILQ_INIT(&new_cont->pool_list);
4084 TAILQ_CONCAT(&new_cont->pool_list, &cont->pool_list, next);
4085 new_cont->init_mem_mng = mem_mng;
4087 /* Flip the master container. */
4088 priv->sh->cmng.mhi[batch] ^= (uint8_t)1;
4093 * Query a devx flow counter.
4096 * Pointer to the Ethernet device structure.
4098 * Index to the flow counter.
4100 * The statistics value of packets.
4102 * The statistics value of bytes.
4105 * 0 on success, otherwise a negative errno value and rte_errno is set.
4108 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4111 struct mlx5_priv *priv = dev->data->dev_private;
4112 struct mlx5_flow_counter_pool *pool = NULL;
4113 struct mlx5_flow_counter *cnt;
4114 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4117 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4119 if (counter < MLX5_CNT_BATCH_OFFSET) {
4120 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4121 if (priv->counter_fallback)
4122 return mlx5_devx_cmd_flow_counter_query(cnt_ext->dcs, 0,
4123 0, pkts, bytes, 0, NULL, NULL, 0);
4126 rte_spinlock_lock(&pool->sl);
4128 * The single counters allocation may allocate smaller ID than the
4129 * current allocated in parallel to the host reading.
4130 * In this case the new counter values must be reported as 0.
4132 if (unlikely(cnt_ext && cnt_ext->dcs->id < pool->raw->min_dcs_id)) {
4136 offset = cnt - &pool->counters_raw[0];
4137 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4138 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4140 rte_spinlock_unlock(&pool->sl);
4145 * Create and initialize a new counter pool.
4148 * Pointer to the Ethernet device structure.
4150 * The devX counter handle.
4152 * Whether the pool is for counter that was allocated by batch command.
4153 * @param[in/out] cont_cur
4154 * Pointer to the container pointer, it will be update in pool resize.
4157 * The pool container pointer on success, NULL otherwise and rte_errno is set.
4159 static struct mlx5_pools_container *
4160 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4163 struct mlx5_priv *priv = dev->data->dev_private;
4164 struct mlx5_flow_counter_pool *pool;
4165 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4167 int16_t n_valid = rte_atomic16_read(&cont->n_valid);
4170 if (cont->n == n_valid) {
4171 cont = flow_dv_container_resize(dev, batch);
4175 size = sizeof(*pool);
4177 size += MLX5_COUNTERS_PER_POOL *
4178 sizeof(struct mlx5_flow_counter_ext);
4179 pool = rte_calloc(__func__, 1, size, 0);
4184 pool->min_dcs = dcs;
4185 if (!priv->counter_fallback)
4186 pool->raw = cont->init_mem_mng->raws + n_valid %
4187 MLX5_CNT_CONTAINER_RESIZE;
4188 pool->raw_hw = NULL;
4189 rte_spinlock_init(&pool->sl);
4191 * The generation of the new allocated counters in this pool is 0, 2 in
4192 * the pool generation makes all the counters valid for allocation.
4193 * The start and end query generation protect the counters be released
4194 * between the query and update gap period will not be reallocated
4195 * without the last query finished and stats updated to the memory.
4197 rte_atomic64_set(&pool->start_query_gen, 0x2);
4199 * There's no background query thread for fallback mode, set the
4200 * end_query_gen to the maximum value since no need to wait for
4201 * statistics update.
4203 rte_atomic64_set(&pool->end_query_gen, priv->counter_fallback ?
4205 TAILQ_INIT(&pool->counters);
4206 TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
4207 pool->index = n_valid;
4208 cont->pools[n_valid] = pool;
4209 /* Pool initialization must be updated before host thread access. */
4211 rte_atomic16_add(&cont->n_valid, 1);
4216 * Prepare a new counter and/or a new counter pool.
4219 * Pointer to the Ethernet device structure.
4220 * @param[out] cnt_free
4221 * Where to put the pointer of a new counter.
4223 * Whether the pool is for counter that was allocated by batch command.
4226 * The counter container pointer and @p cnt_free is set on success,
4227 * NULL otherwise and rte_errno is set.
4229 static struct mlx5_pools_container *
4230 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4231 struct mlx5_flow_counter **cnt_free,
4234 struct mlx5_priv *priv = dev->data->dev_private;
4235 struct mlx5_pools_container *cont;
4236 struct mlx5_flow_counter_pool *pool;
4237 struct mlx5_devx_obj *dcs = NULL;
4238 struct mlx5_flow_counter *cnt;
4241 cont = MLX5_CNT_CONTAINER(priv->sh, batch, 0);
4243 /* bulk_bitmap must be 0 for single counter allocation. */
4244 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4247 pool = flow_dv_find_pool_by_id(cont, dcs->id);
4249 cont = flow_dv_pool_create(dev, dcs, batch);
4251 mlx5_devx_cmd_destroy(dcs);
4254 pool = TAILQ_FIRST(&cont->pool_list);
4255 } else if (dcs->id < pool->min_dcs->id) {
4256 rte_atomic64_set(&pool->a64_dcs,
4257 (int64_t)(uintptr_t)dcs);
4259 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4260 cnt = &pool->counters_raw[i];
4261 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
4262 MLX5_GET_POOL_CNT_EXT(pool, i)->dcs = dcs;
4266 /* bulk_bitmap is in 128 counters units. */
4267 if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
4268 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4270 rte_errno = ENODATA;
4273 cont = flow_dv_pool_create(dev, dcs, batch);
4275 mlx5_devx_cmd_destroy(dcs);
4278 pool = TAILQ_FIRST(&cont->pool_list);
4279 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
4280 cnt = &pool->counters_raw[i];
4281 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
4283 *cnt_free = &pool->counters_raw[0];
4288 * Search for existed shared counter.
4291 * Pointer to the relevant counter pool container.
4293 * The shared counter ID to search.
4295 * mlx5 flow counter pool in the container,
4298 * NULL if not existed, otherwise pointer to the shared extend counter.
4300 static struct mlx5_flow_counter_ext *
4301 flow_dv_counter_shared_search(struct mlx5_pools_container *cont, uint32_t id,
4302 struct mlx5_flow_counter_pool **ppool)
4304 static struct mlx5_flow_counter_ext *cnt;
4305 struct mlx5_flow_counter_pool *pool;
4307 uint32_t n_valid = rte_atomic16_read(&cont->n_valid);
4309 for (i = 0; i < n_valid; i++) {
4310 pool = cont->pools[i];
4311 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
4312 cnt = MLX5_GET_POOL_CNT_EXT(pool, i);
4313 if (cnt->ref_cnt && cnt->shared && cnt->id == id) {
4315 *ppool = cont->pools[i];
4324 * Allocate a flow counter.
4327 * Pointer to the Ethernet device structure.
4329 * Indicate if this counter is shared with other flows.
4331 * Counter identifier.
4333 * Counter flow group.
4336 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4339 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
4342 struct mlx5_priv *priv = dev->data->dev_private;
4343 struct mlx5_flow_counter_pool *pool = NULL;
4344 struct mlx5_flow_counter *cnt_free = NULL;
4345 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4347 * Currently group 0 flow counter cannot be assigned to a flow if it is
4348 * not the first one in the batch counter allocation, so it is better
4349 * to allocate counters one by one for these flows in a separate
4351 * A counter can be shared between different groups so need to take
4352 * shared counters from the single container.
4354 uint32_t batch = (group && !shared && !priv->counter_fallback) ? 1 : 0;
4355 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4359 if (!priv->config.devx) {
4360 rte_errno = ENOTSUP;
4364 cnt_ext = flow_dv_counter_shared_search(cont, id, &pool);
4366 if (cnt_ext->ref_cnt + 1 == 0) {
4371 cnt_idx = pool->index * MLX5_COUNTERS_PER_POOL +
4372 (cnt_ext->dcs->id % MLX5_COUNTERS_PER_POOL)
4377 /* Pools which has a free counters are in the start. */
4378 TAILQ_FOREACH(pool, &cont->pool_list, next) {
4380 * The free counter reset values must be updated between the
4381 * counter release to the counter allocation, so, at least one
4382 * query must be done in this time. ensure it by saving the
4383 * query generation in the release time.
4384 * The free list is sorted according to the generation - so if
4385 * the first one is not updated, all the others are not
4388 cnt_free = TAILQ_FIRST(&pool->counters);
4389 if (cnt_free && cnt_free->query_gen <
4390 rte_atomic64_read(&pool->end_query_gen))
4395 cont = flow_dv_counter_pool_prepare(dev, &cnt_free, batch);
4398 pool = TAILQ_FIRST(&cont->pool_list);
4401 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt_free);
4402 /* Create a DV counter action only in the first time usage. */
4403 if (!cnt_free->action) {
4405 struct mlx5_devx_obj *dcs;
4408 offset = cnt_free - &pool->counters_raw[0];
4409 dcs = pool->min_dcs;
4414 cnt_free->action = mlx5_glue->dv_create_flow_action_counter
4416 if (!cnt_free->action) {
4421 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4422 (cnt_free - pool->counters_raw));
4423 cnt_idx += batch * MLX5_CNT_BATCH_OFFSET;
4424 /* Update the counter reset values. */
4425 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4429 cnt_ext->shared = shared;
4430 cnt_ext->ref_cnt = 1;
4433 if (!priv->counter_fallback && !priv->sh->cmng.query_thread_on)
4434 /* Start the asynchronous batch query by the host thread. */
4435 mlx5_set_query_alarm(priv->sh);
4436 TAILQ_REMOVE(&pool->counters, cnt_free, next);
4437 if (TAILQ_EMPTY(&pool->counters)) {
4438 /* Move the pool to the end of the container pool list. */
4439 TAILQ_REMOVE(&cont->pool_list, pool, next);
4440 TAILQ_INSERT_TAIL(&cont->pool_list, pool, next);
4446 * Release a flow counter.
4449 * Pointer to the Ethernet device structure.
4450 * @param[in] counter
4451 * Index to the counter handler.
4454 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
4456 struct mlx5_flow_counter_pool *pool = NULL;
4457 struct mlx5_flow_counter *cnt;
4458 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4462 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4464 if (counter < MLX5_CNT_BATCH_OFFSET) {
4465 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4466 if (cnt_ext && --cnt_ext->ref_cnt)
4469 /* Put the counter in the end - the last updated one. */
4470 TAILQ_INSERT_TAIL(&pool->counters, cnt, next);
4472 * Counters released between query trigger and handler need
4473 * to wait the next round of query. Since the packets arrive
4474 * in the gap period will not be taken into account to the
4477 cnt->query_gen = rte_atomic64_read(&pool->start_query_gen);
4481 * Verify the @p attributes will be correctly understood by the NIC and store
4482 * them in the @p flow if everything is correct.
4485 * Pointer to dev struct.
4486 * @param[in] attributes
4487 * Pointer to flow attributes
4488 * @param[in] external
4489 * This flow rule is created by request external to PMD.
4491 * Pointer to error structure.
4494 * - 0 on success and non root table.
4495 * - 1 on success and root table.
4496 * - a negative errno value otherwise and rte_errno is set.
4499 flow_dv_validate_attributes(struct rte_eth_dev *dev,
4500 const struct rte_flow_attr *attributes,
4501 bool external __rte_unused,
4502 struct rte_flow_error *error)
4504 struct mlx5_priv *priv = dev->data->dev_private;
4505 uint32_t priority_max = priv->config.flow_prio - 1;
4508 #ifndef HAVE_MLX5DV_DR
4509 if (attributes->group)
4510 return rte_flow_error_set(error, ENOTSUP,
4511 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
4513 "groups are not supported");
4517 ret = mlx5_flow_group_to_table(attributes, external,
4518 attributes->group, !!priv->fdb_def_rule,
4523 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4525 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
4526 attributes->priority >= priority_max)
4527 return rte_flow_error_set(error, ENOTSUP,
4528 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
4530 "priority out of range");
4531 if (attributes->transfer) {
4532 if (!priv->config.dv_esw_en)
4533 return rte_flow_error_set
4535 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4536 "E-Switch dr is not supported");
4537 if (!(priv->representor || priv->master))
4538 return rte_flow_error_set
4539 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4540 NULL, "E-Switch configuration can only be"
4541 " done by a master or a representor device");
4542 if (attributes->egress)
4543 return rte_flow_error_set
4545 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
4546 "egress is not supported");
4548 if (!(attributes->egress ^ attributes->ingress))
4549 return rte_flow_error_set(error, ENOTSUP,
4550 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
4551 "must specify exactly one of "
4552 "ingress or egress");
4557 * Internal validation function. For validating both actions and items.
4560 * Pointer to the rte_eth_dev structure.
4562 * Pointer to the flow attributes.
4564 * Pointer to the list of items.
4565 * @param[in] actions
4566 * Pointer to the list of actions.
4567 * @param[in] external
4568 * This flow rule is created by request external to PMD.
4569 * @param[in] hairpin
4570 * Number of hairpin TX actions, 0 means classic flow.
4572 * Pointer to the error structure.
4575 * 0 on success, a negative errno value otherwise and rte_errno is set.
4578 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
4579 const struct rte_flow_item items[],
4580 const struct rte_flow_action actions[],
4581 bool external, int hairpin, struct rte_flow_error *error)
4584 uint64_t action_flags = 0;
4585 uint64_t item_flags = 0;
4586 uint64_t last_item = 0;
4587 uint8_t next_protocol = 0xff;
4588 uint16_t ether_type = 0;
4590 uint8_t item_ipv6_proto = 0;
4591 const struct rte_flow_item *gre_item = NULL;
4592 const struct rte_flow_action_raw_decap *decap;
4593 const struct rte_flow_action_raw_encap *encap;
4594 const struct rte_flow_action_rss *rss;
4595 const struct rte_flow_item_tcp nic_tcp_mask = {
4598 .src_port = RTE_BE16(UINT16_MAX),
4599 .dst_port = RTE_BE16(UINT16_MAX),
4602 const struct rte_flow_item_ipv4 nic_ipv4_mask = {
4604 .src_addr = RTE_BE32(0xffffffff),
4605 .dst_addr = RTE_BE32(0xffffffff),
4606 .type_of_service = 0xff,
4607 .next_proto_id = 0xff,
4608 .time_to_live = 0xff,
4611 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
4614 "\xff\xff\xff\xff\xff\xff\xff\xff"
4615 "\xff\xff\xff\xff\xff\xff\xff\xff",
4617 "\xff\xff\xff\xff\xff\xff\xff\xff"
4618 "\xff\xff\xff\xff\xff\xff\xff\xff",
4619 .vtc_flow = RTE_BE32(0xffffffff),
4624 struct mlx5_priv *priv = dev->data->dev_private;
4625 struct mlx5_dev_config *dev_conf = &priv->config;
4626 uint16_t queue_index = 0xFFFF;
4627 const struct rte_flow_item_vlan *vlan_m = NULL;
4628 int16_t rw_act_num = 0;
4633 ret = flow_dv_validate_attributes(dev, attr, external, error);
4636 is_root = (uint64_t)ret;
4637 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4638 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
4639 int type = items->type;
4642 case RTE_FLOW_ITEM_TYPE_VOID:
4644 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4645 ret = flow_dv_validate_item_port_id
4646 (dev, items, attr, item_flags, error);
4649 last_item = MLX5_FLOW_ITEM_PORT_ID;
4651 case RTE_FLOW_ITEM_TYPE_ETH:
4652 ret = mlx5_flow_validate_item_eth(items, item_flags,
4656 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
4657 MLX5_FLOW_LAYER_OUTER_L2;
4658 if (items->mask != NULL && items->spec != NULL) {
4660 ((const struct rte_flow_item_eth *)
4663 ((const struct rte_flow_item_eth *)
4665 ether_type = rte_be_to_cpu_16(ether_type);
4670 case RTE_FLOW_ITEM_TYPE_VLAN:
4671 ret = mlx5_flow_validate_item_vlan(items, item_flags,
4675 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
4676 MLX5_FLOW_LAYER_OUTER_VLAN;
4677 if (items->mask != NULL && items->spec != NULL) {
4679 ((const struct rte_flow_item_vlan *)
4680 items->spec)->inner_type;
4682 ((const struct rte_flow_item_vlan *)
4683 items->mask)->inner_type;
4684 ether_type = rte_be_to_cpu_16(ether_type);
4688 /* Store outer VLAN mask for of_push_vlan action. */
4690 vlan_m = items->mask;
4692 case RTE_FLOW_ITEM_TYPE_IPV4:
4693 mlx5_flow_tunnel_ip_check(items, next_protocol,
4694 &item_flags, &tunnel);
4695 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
4702 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
4703 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4704 if (items->mask != NULL &&
4705 ((const struct rte_flow_item_ipv4 *)
4706 items->mask)->hdr.next_proto_id) {
4708 ((const struct rte_flow_item_ipv4 *)
4709 (items->spec))->hdr.next_proto_id;
4711 ((const struct rte_flow_item_ipv4 *)
4712 (items->mask))->hdr.next_proto_id;
4714 /* Reset for inner layer. */
4715 next_protocol = 0xff;
4718 case RTE_FLOW_ITEM_TYPE_IPV6:
4719 mlx5_flow_tunnel_ip_check(items, next_protocol,
4720 &item_flags, &tunnel);
4721 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
4728 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
4729 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4730 if (items->mask != NULL &&
4731 ((const struct rte_flow_item_ipv6 *)
4732 items->mask)->hdr.proto) {
4734 ((const struct rte_flow_item_ipv6 *)
4735 items->spec)->hdr.proto;
4737 ((const struct rte_flow_item_ipv6 *)
4738 items->spec)->hdr.proto;
4740 ((const struct rte_flow_item_ipv6 *)
4741 items->mask)->hdr.proto;
4743 /* Reset for inner layer. */
4744 next_protocol = 0xff;
4747 case RTE_FLOW_ITEM_TYPE_TCP:
4748 ret = mlx5_flow_validate_item_tcp
4755 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
4756 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4758 case RTE_FLOW_ITEM_TYPE_UDP:
4759 ret = mlx5_flow_validate_item_udp(items, item_flags,
4764 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
4765 MLX5_FLOW_LAYER_OUTER_L4_UDP;
4767 case RTE_FLOW_ITEM_TYPE_GRE:
4768 ret = mlx5_flow_validate_item_gre(items, item_flags,
4769 next_protocol, error);
4773 last_item = MLX5_FLOW_LAYER_GRE;
4775 case RTE_FLOW_ITEM_TYPE_NVGRE:
4776 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
4781 last_item = MLX5_FLOW_LAYER_NVGRE;
4783 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
4784 ret = mlx5_flow_validate_item_gre_key
4785 (items, item_flags, gre_item, error);
4788 last_item = MLX5_FLOW_LAYER_GRE_KEY;
4790 case RTE_FLOW_ITEM_TYPE_VXLAN:
4791 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
4795 last_item = MLX5_FLOW_LAYER_VXLAN;
4797 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
4798 ret = mlx5_flow_validate_item_vxlan_gpe(items,
4803 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
4805 case RTE_FLOW_ITEM_TYPE_GENEVE:
4806 ret = mlx5_flow_validate_item_geneve(items,
4811 last_item = MLX5_FLOW_LAYER_GENEVE;
4813 case RTE_FLOW_ITEM_TYPE_MPLS:
4814 ret = mlx5_flow_validate_item_mpls(dev, items,
4819 last_item = MLX5_FLOW_LAYER_MPLS;
4822 case RTE_FLOW_ITEM_TYPE_MARK:
4823 ret = flow_dv_validate_item_mark(dev, items, attr,
4827 last_item = MLX5_FLOW_ITEM_MARK;
4829 case RTE_FLOW_ITEM_TYPE_META:
4830 ret = flow_dv_validate_item_meta(dev, items, attr,
4834 last_item = MLX5_FLOW_ITEM_METADATA;
4836 case RTE_FLOW_ITEM_TYPE_ICMP:
4837 ret = mlx5_flow_validate_item_icmp(items, item_flags,
4842 last_item = MLX5_FLOW_LAYER_ICMP;
4844 case RTE_FLOW_ITEM_TYPE_ICMP6:
4845 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
4850 item_ipv6_proto = IPPROTO_ICMPV6;
4851 last_item = MLX5_FLOW_LAYER_ICMP6;
4853 case RTE_FLOW_ITEM_TYPE_TAG:
4854 ret = flow_dv_validate_item_tag(dev, items,
4858 last_item = MLX5_FLOW_ITEM_TAG;
4860 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
4861 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
4863 case RTE_FLOW_ITEM_TYPE_GTP:
4864 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
4868 last_item = MLX5_FLOW_LAYER_GTP;
4871 return rte_flow_error_set(error, ENOTSUP,
4872 RTE_FLOW_ERROR_TYPE_ITEM,
4873 NULL, "item not supported");
4875 item_flags |= last_item;
4877 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4878 int type = actions->type;
4879 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
4880 return rte_flow_error_set(error, ENOTSUP,
4881 RTE_FLOW_ERROR_TYPE_ACTION,
4882 actions, "too many actions");
4884 case RTE_FLOW_ACTION_TYPE_VOID:
4886 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4887 ret = flow_dv_validate_action_port_id(dev,
4894 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4897 case RTE_FLOW_ACTION_TYPE_FLAG:
4898 ret = flow_dv_validate_action_flag(dev, action_flags,
4902 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
4903 /* Count all modify-header actions as one. */
4904 if (!(action_flags &
4905 MLX5_FLOW_MODIFY_HDR_ACTIONS))
4907 action_flags |= MLX5_FLOW_ACTION_FLAG |
4908 MLX5_FLOW_ACTION_MARK_EXT;
4910 action_flags |= MLX5_FLOW_ACTION_FLAG;
4913 rw_act_num += MLX5_ACT_NUM_SET_MARK;
4915 case RTE_FLOW_ACTION_TYPE_MARK:
4916 ret = flow_dv_validate_action_mark(dev, actions,
4921 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
4922 /* Count all modify-header actions as one. */
4923 if (!(action_flags &
4924 MLX5_FLOW_MODIFY_HDR_ACTIONS))
4926 action_flags |= MLX5_FLOW_ACTION_MARK |
4927 MLX5_FLOW_ACTION_MARK_EXT;
4929 action_flags |= MLX5_FLOW_ACTION_MARK;
4932 rw_act_num += MLX5_ACT_NUM_SET_MARK;
4934 case RTE_FLOW_ACTION_TYPE_SET_META:
4935 ret = flow_dv_validate_action_set_meta(dev, actions,
4940 /* Count all modify-header actions as one action. */
4941 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
4943 action_flags |= MLX5_FLOW_ACTION_SET_META;
4944 rw_act_num += MLX5_ACT_NUM_SET_META;
4946 case RTE_FLOW_ACTION_TYPE_SET_TAG:
4947 ret = flow_dv_validate_action_set_tag(dev, actions,
4952 /* Count all modify-header actions as one action. */
4953 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
4955 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
4956 rw_act_num += MLX5_ACT_NUM_SET_TAG;
4958 case RTE_FLOW_ACTION_TYPE_DROP:
4959 ret = mlx5_flow_validate_action_drop(action_flags,
4963 action_flags |= MLX5_FLOW_ACTION_DROP;
4966 case RTE_FLOW_ACTION_TYPE_QUEUE:
4967 ret = mlx5_flow_validate_action_queue(actions,
4972 queue_index = ((const struct rte_flow_action_queue *)
4973 (actions->conf))->index;
4974 action_flags |= MLX5_FLOW_ACTION_QUEUE;
4977 case RTE_FLOW_ACTION_TYPE_RSS:
4978 rss = actions->conf;
4979 ret = mlx5_flow_validate_action_rss(actions,
4985 if (rss != NULL && rss->queue_num)
4986 queue_index = rss->queue[0];
4987 action_flags |= MLX5_FLOW_ACTION_RSS;
4990 case RTE_FLOW_ACTION_TYPE_COUNT:
4991 ret = flow_dv_validate_action_count(dev, error);
4994 action_flags |= MLX5_FLOW_ACTION_COUNT;
4997 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4998 if (flow_dv_validate_action_pop_vlan(dev,
5004 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5007 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5008 ret = flow_dv_validate_action_push_vlan(dev,
5015 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5018 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5019 ret = flow_dv_validate_action_set_vlan_pcp
5020 (action_flags, actions, error);
5023 /* Count PCP with push_vlan command. */
5024 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5026 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5027 ret = flow_dv_validate_action_set_vlan_vid
5028 (item_flags, action_flags,
5032 /* Count VID with push_vlan command. */
5033 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5034 rw_act_num += MLX5_ACT_NUM_MDF_VID;
5036 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5037 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5038 ret = flow_dv_validate_action_l2_encap(dev,
5044 action_flags |= MLX5_FLOW_ACTION_ENCAP;
5047 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5048 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5049 ret = flow_dv_validate_action_decap(dev, action_flags,
5053 action_flags |= MLX5_FLOW_ACTION_DECAP;
5056 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5057 ret = flow_dv_validate_action_raw_encap_decap
5058 (dev, NULL, actions->conf, attr, &action_flags,
5063 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5064 decap = actions->conf;
5065 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5067 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5071 encap = actions->conf;
5073 ret = flow_dv_validate_action_raw_encap_decap
5075 decap ? decap : &empty_decap, encap,
5076 attr, &action_flags, &actions_n,
5081 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5082 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5083 ret = flow_dv_validate_action_modify_mac(action_flags,
5089 /* Count all modify-header actions as one action. */
5090 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5092 action_flags |= actions->type ==
5093 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5094 MLX5_FLOW_ACTION_SET_MAC_SRC :
5095 MLX5_FLOW_ACTION_SET_MAC_DST;
5097 * Even if the source and destination MAC addresses have
5098 * overlap in the header with 4B alignment, the convert
5099 * function will handle them separately and 4 SW actions
5100 * will be created. And 2 actions will be added each
5101 * time no matter how many bytes of address will be set.
5103 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5105 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5106 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5107 ret = flow_dv_validate_action_modify_ipv4(action_flags,
5113 /* Count all modify-header actions as one action. */
5114 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5116 action_flags |= actions->type ==
5117 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5118 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5119 MLX5_FLOW_ACTION_SET_IPV4_DST;
5120 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5122 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5123 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5124 ret = flow_dv_validate_action_modify_ipv6(action_flags,
5130 if (item_ipv6_proto == IPPROTO_ICMPV6)
5131 return rte_flow_error_set(error, ENOTSUP,
5132 RTE_FLOW_ERROR_TYPE_ACTION,
5134 "Can't change header "
5135 "with ICMPv6 proto");
5136 /* Count all modify-header actions as one action. */
5137 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5139 action_flags |= actions->type ==
5140 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5141 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5142 MLX5_FLOW_ACTION_SET_IPV6_DST;
5143 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5145 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5146 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5147 ret = flow_dv_validate_action_modify_tp(action_flags,
5153 /* Count all modify-header actions as one action. */
5154 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5156 action_flags |= actions->type ==
5157 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5158 MLX5_FLOW_ACTION_SET_TP_SRC :
5159 MLX5_FLOW_ACTION_SET_TP_DST;
5160 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5162 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5163 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5164 ret = flow_dv_validate_action_modify_ttl(action_flags,
5170 /* Count all modify-header actions as one action. */
5171 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5173 action_flags |= actions->type ==
5174 RTE_FLOW_ACTION_TYPE_SET_TTL ?
5175 MLX5_FLOW_ACTION_SET_TTL :
5176 MLX5_FLOW_ACTION_DEC_TTL;
5177 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5179 case RTE_FLOW_ACTION_TYPE_JUMP:
5180 ret = flow_dv_validate_action_jump(actions,
5187 action_flags |= MLX5_FLOW_ACTION_JUMP;
5189 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5190 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5191 ret = flow_dv_validate_action_modify_tcp_seq
5198 /* Count all modify-header actions as one action. */
5199 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5201 action_flags |= actions->type ==
5202 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5203 MLX5_FLOW_ACTION_INC_TCP_SEQ :
5204 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5205 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5207 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5208 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5209 ret = flow_dv_validate_action_modify_tcp_ack
5216 /* Count all modify-header actions as one action. */
5217 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5219 action_flags |= actions->type ==
5220 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5221 MLX5_FLOW_ACTION_INC_TCP_ACK :
5222 MLX5_FLOW_ACTION_DEC_TCP_ACK;
5223 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5225 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5227 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5228 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5229 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5231 case RTE_FLOW_ACTION_TYPE_METER:
5232 ret = mlx5_flow_validate_action_meter(dev,
5238 action_flags |= MLX5_FLOW_ACTION_METER;
5240 /* Meter action will add one more TAG action. */
5241 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5243 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5244 ret = flow_dv_validate_action_modify_ipv4_dscp
5251 /* Count all modify-header actions as one action. */
5252 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5254 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5255 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5257 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5258 ret = flow_dv_validate_action_modify_ipv6_dscp
5265 /* Count all modify-header actions as one action. */
5266 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5268 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5269 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5272 return rte_flow_error_set(error, ENOTSUP,
5273 RTE_FLOW_ERROR_TYPE_ACTION,
5275 "action not supported");
5279 * Validate the drop action mutual exclusion with other actions.
5280 * Drop action is mutually-exclusive with any other action, except for
5283 if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
5284 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
5285 return rte_flow_error_set(error, EINVAL,
5286 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5287 "Drop action is mutually-exclusive "
5288 "with any other action, except for "
5290 /* Eswitch has few restrictions on using items and actions */
5291 if (attr->transfer) {
5292 if (!mlx5_flow_ext_mreg_supported(dev) &&
5293 action_flags & MLX5_FLOW_ACTION_FLAG)
5294 return rte_flow_error_set(error, ENOTSUP,
5295 RTE_FLOW_ERROR_TYPE_ACTION,
5297 "unsupported action FLAG");
5298 if (!mlx5_flow_ext_mreg_supported(dev) &&
5299 action_flags & MLX5_FLOW_ACTION_MARK)
5300 return rte_flow_error_set(error, ENOTSUP,
5301 RTE_FLOW_ERROR_TYPE_ACTION,
5303 "unsupported action MARK");
5304 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
5305 return rte_flow_error_set(error, ENOTSUP,
5306 RTE_FLOW_ERROR_TYPE_ACTION,
5308 "unsupported action QUEUE");
5309 if (action_flags & MLX5_FLOW_ACTION_RSS)
5310 return rte_flow_error_set(error, ENOTSUP,
5311 RTE_FLOW_ERROR_TYPE_ACTION,
5313 "unsupported action RSS");
5314 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
5315 return rte_flow_error_set(error, EINVAL,
5316 RTE_FLOW_ERROR_TYPE_ACTION,
5318 "no fate action is found");
5320 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
5321 return rte_flow_error_set(error, EINVAL,
5322 RTE_FLOW_ERROR_TYPE_ACTION,
5324 "no fate action is found");
5326 /* Continue validation for Xcap actions.*/
5327 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) && (queue_index == 0xFFFF ||
5328 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
5329 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
5330 MLX5_FLOW_XCAP_ACTIONS)
5331 return rte_flow_error_set(error, ENOTSUP,
5332 RTE_FLOW_ERROR_TYPE_ACTION,
5333 NULL, "encap and decap "
5334 "combination aren't supported");
5335 if (!attr->transfer && attr->ingress && (action_flags &
5336 MLX5_FLOW_ACTION_ENCAP))
5337 return rte_flow_error_set(error, ENOTSUP,
5338 RTE_FLOW_ERROR_TYPE_ACTION,
5339 NULL, "encap is not supported"
5340 " for ingress traffic");
5342 /* Hairpin flow will add one more TAG action. */
5344 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5345 /* extra metadata enabled: one more TAG action will be add. */
5346 if (dev_conf->dv_flow_en &&
5347 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
5348 mlx5_flow_ext_mreg_supported(dev))
5349 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5350 if ((uint32_t)rw_act_num >=
5351 flow_dv_modify_hdr_action_max(dev, is_root)) {
5352 return rte_flow_error_set(error, ENOTSUP,
5353 RTE_FLOW_ERROR_TYPE_ACTION,
5354 NULL, "too many header modify"
5355 " actions to support");
5361 * Internal preparation function. Allocates the DV flow size,
5362 * this size is constant.
5365 * Pointer to the rte_eth_dev structure.
5367 * Pointer to the flow attributes.
5369 * Pointer to the list of items.
5370 * @param[in] actions
5371 * Pointer to the list of actions.
5373 * Pointer to the error structure.
5376 * Pointer to mlx5_flow object on success,
5377 * otherwise NULL and rte_errno is set.
5379 static struct mlx5_flow *
5380 flow_dv_prepare(struct rte_eth_dev *dev,
5381 const struct rte_flow_attr *attr __rte_unused,
5382 const struct rte_flow_item items[] __rte_unused,
5383 const struct rte_flow_action actions[] __rte_unused,
5384 struct rte_flow_error *error)
5386 uint32_t handle_idx = 0;
5387 struct mlx5_flow *dev_flow;
5388 struct mlx5_flow_handle *dev_handle;
5389 struct mlx5_priv *priv = dev->data->dev_private;
5391 /* In case of corrupting the memory. */
5392 if (priv->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
5393 rte_flow_error_set(error, ENOSPC,
5394 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5395 "not free temporary device flow");
5398 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
5401 rte_flow_error_set(error, ENOMEM,
5402 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5403 "not enough memory to create flow handle");
5406 /* No multi-thread supporting. */
5407 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[priv->flow_idx++];
5408 dev_flow->handle = dev_handle;
5409 dev_flow->handle_idx = handle_idx;
5410 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
5412 * The matching value needs to be cleared to 0 before using. In the
5413 * past, it will be automatically cleared when using rte_*alloc
5414 * API. The time consumption will be almost the same as before.
5416 memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
5417 dev_flow->ingress = attr->ingress;
5418 dev_flow->dv.transfer = attr->transfer;
5422 #ifdef RTE_LIBRTE_MLX5_DEBUG
5424 * Sanity check for match mask and value. Similar to check_valid_spec() in
5425 * kernel driver. If unmasked bit is present in value, it returns failure.
5428 * pointer to match mask buffer.
5429 * @param match_value
5430 * pointer to match value buffer.
5433 * 0 if valid, -EINVAL otherwise.
5436 flow_dv_check_valid_spec(void *match_mask, void *match_value)
5438 uint8_t *m = match_mask;
5439 uint8_t *v = match_value;
5442 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
5445 "match_value differs from match_criteria"
5446 " %p[%u] != %p[%u]",
5447 match_value, i, match_mask, i);
5456 * Add Ethernet item to matcher and to the value.
5458 * @param[in, out] matcher
5460 * @param[in, out] key
5461 * Flow matcher value.
5463 * Flow pattern to translate.
5465 * Item is inner pattern.
5468 flow_dv_translate_item_eth(void *matcher, void *key,
5469 const struct rte_flow_item *item, int inner)
5471 const struct rte_flow_item_eth *eth_m = item->mask;
5472 const struct rte_flow_item_eth *eth_v = item->spec;
5473 const struct rte_flow_item_eth nic_mask = {
5474 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5475 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5476 .type = RTE_BE16(0xffff),
5488 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5490 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5492 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5494 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5496 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
5497 ð_m->dst, sizeof(eth_m->dst));
5498 /* The value must be in the range of the mask. */
5499 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
5500 for (i = 0; i < sizeof(eth_m->dst); ++i)
5501 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
5502 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
5503 ð_m->src, sizeof(eth_m->src));
5504 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
5505 /* The value must be in the range of the mask. */
5506 for (i = 0; i < sizeof(eth_m->dst); ++i)
5507 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
5509 /* When ethertype is present set mask for tagged VLAN. */
5510 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5511 /* Set value for tagged VLAN if ethertype is 802.1Q. */
5512 if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
5513 eth_v->type == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
5514 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag,
5516 /* Return here to avoid setting match on ethertype. */
5521 * HW supports match on one Ethertype, the Ethertype following the last
5522 * VLAN tag of the packet (see PRM).
5523 * Set match on ethertype only if ETH header is not followed by VLAN.
5525 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
5526 rte_be_to_cpu_16(eth_m->type));
5527 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, ethertype);
5528 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
5532 * Add VLAN item to matcher and to the value.
5534 * @param[in, out] dev_flow
5536 * @param[in, out] matcher
5538 * @param[in, out] key
5539 * Flow matcher value.
5541 * Flow pattern to translate.
5543 * Item is inner pattern.
5546 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
5547 void *matcher, void *key,
5548 const struct rte_flow_item *item,
5551 const struct rte_flow_item_vlan *vlan_m = item->mask;
5552 const struct rte_flow_item_vlan *vlan_v = item->spec;
5561 vlan_m = &rte_flow_item_vlan_mask;
5563 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5565 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5567 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5569 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5571 * This is workaround, masks are not supported,
5572 * and pre-validated.
5574 dev_flow->handle->vf_vlan.tag =
5575 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
5577 tci_m = rte_be_to_cpu_16(vlan_m->tci);
5578 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
5579 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5580 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
5581 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
5582 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
5583 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
5584 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
5585 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
5586 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
5587 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
5588 rte_be_to_cpu_16(vlan_m->inner_type));
5589 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
5590 rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
5594 * Add IPV4 item to matcher and to the value.
5596 * @param[in, out] matcher
5598 * @param[in, out] key
5599 * Flow matcher value.
5601 * Flow pattern to translate.
5602 * @param[in] item_flags
5603 * Bit-fields that holds the items detected until now.
5605 * Item is inner pattern.
5607 * The group to insert the rule.
5610 flow_dv_translate_item_ipv4(void *matcher, void *key,
5611 const struct rte_flow_item *item,
5612 const uint64_t item_flags,
5613 int inner, uint32_t group)
5615 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
5616 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
5617 const struct rte_flow_item_ipv4 nic_mask = {
5619 .src_addr = RTE_BE32(0xffffffff),
5620 .dst_addr = RTE_BE32(0xffffffff),
5621 .type_of_service = 0xff,
5622 .next_proto_id = 0xff,
5623 .time_to_live = 0xff,
5633 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5635 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5637 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5639 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5642 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5644 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x4);
5645 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 4);
5647 * On outer header (which must contains L2), or inner header with L2,
5648 * set cvlan_tag mask bit to mark this packet as untagged.
5649 * This should be done even if item->spec is empty.
5651 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
5652 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5657 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5658 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
5659 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5660 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
5661 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
5662 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
5663 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5664 src_ipv4_src_ipv6.ipv4_layout.ipv4);
5665 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5666 src_ipv4_src_ipv6.ipv4_layout.ipv4);
5667 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
5668 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
5669 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
5670 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
5671 ipv4_m->hdr.type_of_service);
5672 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
5673 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
5674 ipv4_m->hdr.type_of_service >> 2);
5675 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
5676 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
5677 ipv4_m->hdr.next_proto_id);
5678 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
5679 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
5680 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
5681 ipv4_m->hdr.time_to_live);
5682 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
5683 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
5687 * Add IPV6 item to matcher and to the value.
5689 * @param[in, out] matcher
5691 * @param[in, out] key
5692 * Flow matcher value.
5694 * Flow pattern to translate.
5695 * @param[in] item_flags
5696 * Bit-fields that holds the items detected until now.
5698 * Item is inner pattern.
5700 * The group to insert the rule.
5703 flow_dv_translate_item_ipv6(void *matcher, void *key,
5704 const struct rte_flow_item *item,
5705 const uint64_t item_flags,
5706 int inner, uint32_t group)
5708 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
5709 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
5710 const struct rte_flow_item_ipv6 nic_mask = {
5713 "\xff\xff\xff\xff\xff\xff\xff\xff"
5714 "\xff\xff\xff\xff\xff\xff\xff\xff",
5716 "\xff\xff\xff\xff\xff\xff\xff\xff"
5717 "\xff\xff\xff\xff\xff\xff\xff\xff",
5718 .vtc_flow = RTE_BE32(0xffffffff),
5725 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5726 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5735 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5737 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5739 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5741 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5744 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5746 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0x6);
5747 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, 6);
5749 * On outer header (which must contains L2), or inner header with L2,
5750 * set cvlan_tag mask bit to mark this packet as untagged.
5751 * This should be done even if item->spec is empty.
5753 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
5754 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5759 size = sizeof(ipv6_m->hdr.dst_addr);
5760 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5761 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
5762 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5763 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
5764 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
5765 for (i = 0; i < size; ++i)
5766 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
5767 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5768 src_ipv4_src_ipv6.ipv6_layout.ipv6);
5769 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5770 src_ipv4_src_ipv6.ipv6_layout.ipv6);
5771 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
5772 for (i = 0; i < size; ++i)
5773 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
5775 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
5776 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
5777 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
5778 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
5779 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
5780 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
5783 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
5785 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
5788 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
5790 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
5794 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
5796 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
5797 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
5799 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
5800 ipv6_m->hdr.hop_limits);
5801 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
5802 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
5806 * Add TCP item to matcher and to the value.
5808 * @param[in, out] matcher
5810 * @param[in, out] key
5811 * Flow matcher value.
5813 * Flow pattern to translate.
5815 * Item is inner pattern.
5818 flow_dv_translate_item_tcp(void *matcher, void *key,
5819 const struct rte_flow_item *item,
5822 const struct rte_flow_item_tcp *tcp_m = item->mask;
5823 const struct rte_flow_item_tcp *tcp_v = item->spec;
5828 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5830 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5832 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5834 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5836 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5837 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
5841 tcp_m = &rte_flow_item_tcp_mask;
5842 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
5843 rte_be_to_cpu_16(tcp_m->hdr.src_port));
5844 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
5845 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
5846 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
5847 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
5848 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
5849 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
5850 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
5851 tcp_m->hdr.tcp_flags);
5852 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
5853 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
5857 * Add UDP item to matcher and to the value.
5859 * @param[in, out] matcher
5861 * @param[in, out] key
5862 * Flow matcher value.
5864 * Flow pattern to translate.
5866 * Item is inner pattern.
5869 flow_dv_translate_item_udp(void *matcher, void *key,
5870 const struct rte_flow_item *item,
5873 const struct rte_flow_item_udp *udp_m = item->mask;
5874 const struct rte_flow_item_udp *udp_v = item->spec;
5879 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5881 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5883 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5885 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5887 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5888 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
5892 udp_m = &rte_flow_item_udp_mask;
5893 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
5894 rte_be_to_cpu_16(udp_m->hdr.src_port));
5895 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
5896 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
5897 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
5898 rte_be_to_cpu_16(udp_m->hdr.dst_port));
5899 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
5900 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
5904 * Add GRE optional Key item to matcher and to the value.
5906 * @param[in, out] matcher
5908 * @param[in, out] key
5909 * Flow matcher value.
5911 * Flow pattern to translate.
5913 * Item is inner pattern.
5916 flow_dv_translate_item_gre_key(void *matcher, void *key,
5917 const struct rte_flow_item *item)
5919 const rte_be32_t *key_m = item->mask;
5920 const rte_be32_t *key_v = item->spec;
5921 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5922 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5923 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
5925 /* GRE K bit must be on and should already be validated */
5926 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
5927 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
5931 key_m = &gre_key_default_mask;
5932 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
5933 rte_be_to_cpu_32(*key_m) >> 8);
5934 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
5935 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
5936 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
5937 rte_be_to_cpu_32(*key_m) & 0xFF);
5938 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
5939 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
5943 * Add GRE item to matcher and to the value.
5945 * @param[in, out] matcher
5947 * @param[in, out] key
5948 * Flow matcher value.
5950 * Flow pattern to translate.
5952 * Item is inner pattern.
5955 flow_dv_translate_item_gre(void *matcher, void *key,
5956 const struct rte_flow_item *item,
5959 const struct rte_flow_item_gre *gre_m = item->mask;
5960 const struct rte_flow_item_gre *gre_v = item->spec;
5963 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
5964 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
5971 uint16_t s_present:1;
5972 uint16_t k_present:1;
5973 uint16_t rsvd_bit1:1;
5974 uint16_t c_present:1;
5978 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
5981 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5983 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5985 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5987 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5989 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
5990 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
5994 gre_m = &rte_flow_item_gre_mask;
5995 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
5996 rte_be_to_cpu_16(gre_m->protocol));
5997 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
5998 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
5999 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
6000 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
6001 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
6002 gre_crks_rsvd0_ver_m.c_present);
6003 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6004 gre_crks_rsvd0_ver_v.c_present &
6005 gre_crks_rsvd0_ver_m.c_present);
6006 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6007 gre_crks_rsvd0_ver_m.k_present);
6008 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6009 gre_crks_rsvd0_ver_v.k_present &
6010 gre_crks_rsvd0_ver_m.k_present);
6011 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6012 gre_crks_rsvd0_ver_m.s_present);
6013 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6014 gre_crks_rsvd0_ver_v.s_present &
6015 gre_crks_rsvd0_ver_m.s_present);
6019 * Add NVGRE item to matcher and to the value.
6021 * @param[in, out] matcher
6023 * @param[in, out] key
6024 * Flow matcher value.
6026 * Flow pattern to translate.
6028 * Item is inner pattern.
6031 flow_dv_translate_item_nvgre(void *matcher, void *key,
6032 const struct rte_flow_item *item,
6035 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6036 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6037 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6038 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6039 const char *tni_flow_id_m = (const char *)nvgre_m->tni;
6040 const char *tni_flow_id_v = (const char *)nvgre_v->tni;
6046 /* For NVGRE, GRE header fields must be set with defined values. */
6047 const struct rte_flow_item_gre gre_spec = {
6048 .c_rsvd0_ver = RTE_BE16(0x2000),
6049 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6051 const struct rte_flow_item_gre gre_mask = {
6052 .c_rsvd0_ver = RTE_BE16(0xB000),
6053 .protocol = RTE_BE16(UINT16_MAX),
6055 const struct rte_flow_item gre_item = {
6060 flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6064 nvgre_m = &rte_flow_item_nvgre_mask;
6065 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6066 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6067 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6068 memcpy(gre_key_m, tni_flow_id_m, size);
6069 for (i = 0; i < size; ++i)
6070 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6074 * Add VXLAN item to matcher and to the value.
6076 * @param[in, out] matcher
6078 * @param[in, out] key
6079 * Flow matcher value.
6081 * Flow pattern to translate.
6083 * Item is inner pattern.
6086 flow_dv_translate_item_vxlan(void *matcher, void *key,
6087 const struct rte_flow_item *item,
6090 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
6091 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
6094 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6095 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6103 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6105 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6107 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6109 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6111 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6112 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6113 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6114 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6115 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6120 vxlan_m = &rte_flow_item_vxlan_mask;
6121 size = sizeof(vxlan_m->vni);
6122 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
6123 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
6124 memcpy(vni_m, vxlan_m->vni, size);
6125 for (i = 0; i < size; ++i)
6126 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6130 * Add VXLAN-GPE item to matcher and to the value.
6132 * @param[in, out] matcher
6134 * @param[in, out] key
6135 * Flow matcher value.
6137 * Flow pattern to translate.
6139 * Item is inner pattern.
6143 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
6144 const struct rte_flow_item *item, int inner)
6146 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
6147 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
6151 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
6153 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6159 uint8_t flags_m = 0xff;
6160 uint8_t flags_v = 0xc;
6163 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6165 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6167 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6169 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6171 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6172 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6173 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6174 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6175 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6180 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
6181 size = sizeof(vxlan_m->vni);
6182 vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
6183 vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
6184 memcpy(vni_m, vxlan_m->vni, size);
6185 for (i = 0; i < size; ++i)
6186 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6187 if (vxlan_m->flags) {
6188 flags_m = vxlan_m->flags;
6189 flags_v = vxlan_v->flags;
6191 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
6192 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
6193 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
6195 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
6200 * Add Geneve item to matcher and to the value.
6202 * @param[in, out] matcher
6204 * @param[in, out] key
6205 * Flow matcher value.
6207 * Flow pattern to translate.
6209 * Item is inner pattern.
6213 flow_dv_translate_item_geneve(void *matcher, void *key,
6214 const struct rte_flow_item *item, int inner)
6216 const struct rte_flow_item_geneve *geneve_m = item->mask;
6217 const struct rte_flow_item_geneve *geneve_v = item->spec;
6220 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6221 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6230 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6232 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6234 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6236 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6238 dport = MLX5_UDP_PORT_GENEVE;
6239 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6240 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6241 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6246 geneve_m = &rte_flow_item_geneve_mask;
6247 size = sizeof(geneve_m->vni);
6248 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
6249 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
6250 memcpy(vni_m, geneve_m->vni, size);
6251 for (i = 0; i < size; ++i)
6252 vni_v[i] = vni_m[i] & geneve_v->vni[i];
6253 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
6254 rte_be_to_cpu_16(geneve_m->protocol));
6255 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
6256 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
6257 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
6258 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
6259 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
6260 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6261 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
6262 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6263 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
6264 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6265 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
6266 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
6267 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6271 * Add MPLS item to matcher and to the value.
6273 * @param[in, out] matcher
6275 * @param[in, out] key
6276 * Flow matcher value.
6278 * Flow pattern to translate.
6279 * @param[in] prev_layer
6280 * The protocol layer indicated in previous item.
6282 * Item is inner pattern.
6285 flow_dv_translate_item_mpls(void *matcher, void *key,
6286 const struct rte_flow_item *item,
6287 uint64_t prev_layer,
6290 const uint32_t *in_mpls_m = item->mask;
6291 const uint32_t *in_mpls_v = item->spec;
6292 uint32_t *out_mpls_m = 0;
6293 uint32_t *out_mpls_v = 0;
6294 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6295 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6296 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
6298 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6299 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
6300 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6302 switch (prev_layer) {
6303 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6304 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
6305 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6306 MLX5_UDP_PORT_MPLS);
6308 case MLX5_FLOW_LAYER_GRE:
6309 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
6310 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6311 RTE_ETHER_TYPE_MPLS);
6314 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6315 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6322 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
6323 switch (prev_layer) {
6324 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6326 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6327 outer_first_mpls_over_udp);
6329 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6330 outer_first_mpls_over_udp);
6332 case MLX5_FLOW_LAYER_GRE:
6334 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6335 outer_first_mpls_over_gre);
6337 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6338 outer_first_mpls_over_gre);
6341 /* Inner MPLS not over GRE is not supported. */
6344 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6348 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6354 if (out_mpls_m && out_mpls_v) {
6355 *out_mpls_m = *in_mpls_m;
6356 *out_mpls_v = *in_mpls_v & *in_mpls_m;
6361 * Add metadata register item to matcher
6363 * @param[in, out] matcher
6365 * @param[in, out] key
6366 * Flow matcher value.
6367 * @param[in] reg_type
6368 * Type of device metadata register
6375 flow_dv_match_meta_reg(void *matcher, void *key,
6376 enum modify_reg reg_type,
6377 uint32_t data, uint32_t mask)
6380 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
6382 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6388 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
6389 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
6392 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
6393 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
6397 * The metadata register C0 field might be divided into
6398 * source vport index and META item value, we should set
6399 * this field according to specified mask, not as whole one.
6401 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
6403 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
6404 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
6407 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
6410 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
6411 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
6414 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
6415 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
6418 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
6419 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
6422 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
6423 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
6426 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
6427 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
6430 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
6431 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
6434 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
6435 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
6444 * Add MARK item to matcher
6447 * The device to configure through.
6448 * @param[in, out] matcher
6450 * @param[in, out] key
6451 * Flow matcher value.
6453 * Flow pattern to translate.
6456 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
6457 void *matcher, void *key,
6458 const struct rte_flow_item *item)
6460 struct mlx5_priv *priv = dev->data->dev_private;
6461 const struct rte_flow_item_mark *mark;
6465 mark = item->mask ? (const void *)item->mask :
6466 &rte_flow_item_mark_mask;
6467 mask = mark->id & priv->sh->dv_mark_mask;
6468 mark = (const void *)item->spec;
6470 value = mark->id & priv->sh->dv_mark_mask & mask;
6472 enum modify_reg reg;
6474 /* Get the metadata register index for the mark. */
6475 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
6476 MLX5_ASSERT(reg > 0);
6477 if (reg == REG_C_0) {
6478 struct mlx5_priv *priv = dev->data->dev_private;
6479 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6480 uint32_t shl_c0 = rte_bsf32(msk_c0);
6486 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6491 * Add META item to matcher
6494 * The devich to configure through.
6495 * @param[in, out] matcher
6497 * @param[in, out] key
6498 * Flow matcher value.
6500 * Attributes of flow that includes this item.
6502 * Flow pattern to translate.
6505 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
6506 void *matcher, void *key,
6507 const struct rte_flow_attr *attr,
6508 const struct rte_flow_item *item)
6510 const struct rte_flow_item_meta *meta_m;
6511 const struct rte_flow_item_meta *meta_v;
6513 meta_m = (const void *)item->mask;
6515 meta_m = &rte_flow_item_meta_mask;
6516 meta_v = (const void *)item->spec;
6519 uint32_t value = meta_v->data;
6520 uint32_t mask = meta_m->data;
6522 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
6526 * In datapath code there is no endianness
6527 * coversions for perfromance reasons, all
6528 * pattern conversions are done in rte_flow.
6530 value = rte_cpu_to_be_32(value);
6531 mask = rte_cpu_to_be_32(mask);
6532 if (reg == REG_C_0) {
6533 struct mlx5_priv *priv = dev->data->dev_private;
6534 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6535 uint32_t shl_c0 = rte_bsf32(msk_c0);
6536 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
6537 uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
6544 MLX5_ASSERT(msk_c0);
6545 MLX5_ASSERT(!(~msk_c0 & mask));
6547 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6552 * Add vport metadata Reg C0 item to matcher
6554 * @param[in, out] matcher
6556 * @param[in, out] key
6557 * Flow matcher value.
6559 * Flow pattern to translate.
6562 flow_dv_translate_item_meta_vport(void *matcher, void *key,
6563 uint32_t value, uint32_t mask)
6565 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
6569 * Add tag item to matcher
6572 * The devich to configure through.
6573 * @param[in, out] matcher
6575 * @param[in, out] key
6576 * Flow matcher value.
6578 * Flow pattern to translate.
6581 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
6582 void *matcher, void *key,
6583 const struct rte_flow_item *item)
6585 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
6586 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
6587 uint32_t mask, value;
6590 value = tag_v->data;
6591 mask = tag_m ? tag_m->data : UINT32_MAX;
6592 if (tag_v->id == REG_C_0) {
6593 struct mlx5_priv *priv = dev->data->dev_private;
6594 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6595 uint32_t shl_c0 = rte_bsf32(msk_c0);
6601 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
6605 * Add TAG item to matcher
6608 * The devich to configure through.
6609 * @param[in, out] matcher
6611 * @param[in, out] key
6612 * Flow matcher value.
6614 * Flow pattern to translate.
6617 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
6618 void *matcher, void *key,
6619 const struct rte_flow_item *item)
6621 const struct rte_flow_item_tag *tag_v = item->spec;
6622 const struct rte_flow_item_tag *tag_m = item->mask;
6623 enum modify_reg reg;
6626 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
6627 /* Get the metadata register index for the tag. */
6628 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
6629 MLX5_ASSERT(reg > 0);
6630 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
6634 * Add source vport match to the specified matcher.
6636 * @param[in, out] matcher
6638 * @param[in, out] key
6639 * Flow matcher value.
6641 * Source vport value to match
6646 flow_dv_translate_item_source_vport(void *matcher, void *key,
6647 int16_t port, uint16_t mask)
6649 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6650 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6652 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
6653 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
6657 * Translate port-id item to eswitch match on port-id.
6660 * The devich to configure through.
6661 * @param[in, out] matcher
6663 * @param[in, out] key
6664 * Flow matcher value.
6666 * Flow pattern to translate.
6669 * 0 on success, a negative errno value otherwise.
6672 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
6673 void *key, const struct rte_flow_item *item)
6675 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
6676 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
6677 struct mlx5_priv *priv;
6680 mask = pid_m ? pid_m->id : 0xffff;
6681 id = pid_v ? pid_v->id : dev->data->port_id;
6682 priv = mlx5_port_to_eswitch_info(id, item == NULL);
6685 /* Translate to vport field or to metadata, depending on mode. */
6686 if (priv->vport_meta_mask)
6687 flow_dv_translate_item_meta_vport(matcher, key,
6688 priv->vport_meta_tag,
6689 priv->vport_meta_mask);
6691 flow_dv_translate_item_source_vport(matcher, key,
6692 priv->vport_id, mask);
6697 * Add ICMP6 item to matcher and to the value.
6699 * @param[in, out] matcher
6701 * @param[in, out] key
6702 * Flow matcher value.
6704 * Flow pattern to translate.
6706 * Item is inner pattern.
6709 flow_dv_translate_item_icmp6(void *matcher, void *key,
6710 const struct rte_flow_item *item,
6713 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
6714 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
6717 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6719 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6721 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6723 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6725 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6727 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6729 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
6730 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
6734 icmp6_m = &rte_flow_item_icmp6_mask;
6736 * Force flow only to match the non-fragmented IPv6 ICMPv6 packets.
6737 * If only the protocol is specified, no need to match the frag.
6739 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6740 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
6741 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
6742 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
6743 icmp6_v->type & icmp6_m->type);
6744 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
6745 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
6746 icmp6_v->code & icmp6_m->code);
6750 * Add ICMP item to matcher and to the value.
6752 * @param[in, out] matcher
6754 * @param[in, out] key
6755 * Flow matcher value.
6757 * Flow pattern to translate.
6759 * Item is inner pattern.
6762 flow_dv_translate_item_icmp(void *matcher, void *key,
6763 const struct rte_flow_item *item,
6766 const struct rte_flow_item_icmp *icmp_m = item->mask;
6767 const struct rte_flow_item_icmp *icmp_v = item->spec;
6770 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6772 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6774 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6776 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6778 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6780 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6782 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
6783 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
6787 icmp_m = &rte_flow_item_icmp_mask;
6789 * Force flow only to match the non-fragmented IPv4 ICMP packets.
6790 * If only the protocol is specified, no need to match the frag.
6792 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6793 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
6794 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
6795 icmp_m->hdr.icmp_type);
6796 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
6797 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
6798 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
6799 icmp_m->hdr.icmp_code);
6800 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
6801 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
6805 * Add GTP item to matcher and to the value.
6807 * @param[in, out] matcher
6809 * @param[in, out] key
6810 * Flow matcher value.
6812 * Flow pattern to translate.
6814 * Item is inner pattern.
6817 flow_dv_translate_item_gtp(void *matcher, void *key,
6818 const struct rte_flow_item *item, int inner)
6820 const struct rte_flow_item_gtp *gtp_m = item->mask;
6821 const struct rte_flow_item_gtp *gtp_v = item->spec;
6824 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
6826 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6827 uint16_t dport = RTE_GTPU_UDP_PORT;
6830 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6832 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6834 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6836 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6838 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6839 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6840 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6845 gtp_m = &rte_flow_item_gtp_mask;
6846 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
6847 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
6848 gtp_v->msg_type & gtp_m->msg_type);
6849 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
6850 rte_be_to_cpu_32(gtp_m->teid));
6851 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
6852 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
6855 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
6857 #define HEADER_IS_ZERO(match_criteria, headers) \
6858 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
6859 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
6862 * Calculate flow matcher enable bitmap.
6864 * @param match_criteria
6865 * Pointer to flow matcher criteria.
6868 * Bitmap of enabled fields.
6871 flow_dv_matcher_enable(uint32_t *match_criteria)
6873 uint8_t match_criteria_enable;
6875 match_criteria_enable =
6876 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
6877 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
6878 match_criteria_enable |=
6879 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
6880 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
6881 match_criteria_enable |=
6882 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
6883 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
6884 match_criteria_enable |=
6885 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
6886 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
6887 match_criteria_enable |=
6888 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
6889 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
6890 return match_criteria_enable;
6897 * @param[in, out] dev
6898 * Pointer to rte_eth_dev structure.
6899 * @param[in] table_id
6902 * Direction of the table.
6903 * @param[in] transfer
6904 * E-Switch or NIC flow.
6906 * pointer to error structure.
6909 * Returns tables resource based on the index, NULL in case of failed.
6911 static struct mlx5_flow_tbl_resource *
6912 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
6913 uint32_t table_id, uint8_t egress,
6915 struct rte_flow_error *error)
6917 struct mlx5_priv *priv = dev->data->dev_private;
6918 struct mlx5_ibv_shared *sh = priv->sh;
6919 struct mlx5_flow_tbl_resource *tbl;
6920 union mlx5_flow_tbl_key table_key = {
6922 .table_id = table_id,
6924 .domain = !!transfer,
6925 .direction = !!egress,
6928 struct mlx5_hlist_entry *pos = mlx5_hlist_lookup(sh->flow_tbls,
6930 struct mlx5_flow_tbl_data_entry *tbl_data;
6936 tbl_data = container_of(pos, struct mlx5_flow_tbl_data_entry,
6938 tbl = &tbl_data->tbl;
6939 rte_atomic32_inc(&tbl->refcnt);
6942 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
6944 rte_flow_error_set(error, ENOMEM,
6945 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6947 "cannot allocate flow table data entry");
6950 tbl_data->idx = idx;
6951 tbl = &tbl_data->tbl;
6952 pos = &tbl_data->entry;
6954 domain = sh->fdb_domain;
6956 domain = sh->tx_domain;
6958 domain = sh->rx_domain;
6959 tbl->obj = mlx5_glue->dr_create_flow_tbl(domain, table_id);
6961 rte_flow_error_set(error, ENOMEM,
6962 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6963 NULL, "cannot create flow table object");
6964 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
6968 * No multi-threads now, but still better to initialize the reference
6969 * count before insert it into the hash list.
6971 rte_atomic32_init(&tbl->refcnt);
6972 /* Jump action reference count is initialized here. */
6973 rte_atomic32_init(&tbl_data->jump.refcnt);
6974 pos->key = table_key.v64;
6975 ret = mlx5_hlist_insert(sh->flow_tbls, pos);
6977 rte_flow_error_set(error, -ret,
6978 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6979 "cannot insert flow table data entry");
6980 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
6981 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
6983 rte_atomic32_inc(&tbl->refcnt);
6988 * Release a flow table.
6991 * Pointer to rte_eth_dev structure.
6993 * Table resource to be released.
6996 * Returns 0 if table was released, else return 1;
6999 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
7000 struct mlx5_flow_tbl_resource *tbl)
7002 struct mlx5_priv *priv = dev->data->dev_private;
7003 struct mlx5_ibv_shared *sh = priv->sh;
7004 struct mlx5_flow_tbl_data_entry *tbl_data =
7005 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7009 if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
7010 struct mlx5_hlist_entry *pos = &tbl_data->entry;
7012 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
7014 /* remove the entry from the hash list and free memory. */
7015 mlx5_hlist_remove(sh->flow_tbls, pos);
7016 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_JUMP],
7024 * Register the flow matcher.
7026 * @param[in, out] dev
7027 * Pointer to rte_eth_dev structure.
7028 * @param[in, out] matcher
7029 * Pointer to flow matcher.
7030 * @param[in, out] key
7031 * Pointer to flow table key.
7032 * @parm[in, out] dev_flow
7033 * Pointer to the dev_flow.
7035 * pointer to error structure.
7038 * 0 on success otherwise -errno and errno is set.
7041 flow_dv_matcher_register(struct rte_eth_dev *dev,
7042 struct mlx5_flow_dv_matcher *matcher,
7043 union mlx5_flow_tbl_key *key,
7044 struct mlx5_flow *dev_flow,
7045 struct rte_flow_error *error)
7047 struct mlx5_priv *priv = dev->data->dev_private;
7048 struct mlx5_ibv_shared *sh = priv->sh;
7049 struct mlx5_flow_dv_matcher *cache_matcher;
7050 struct mlx5dv_flow_matcher_attr dv_attr = {
7051 .type = IBV_FLOW_ATTR_NORMAL,
7052 .match_mask = (void *)&matcher->mask,
7054 struct mlx5_flow_tbl_resource *tbl;
7055 struct mlx5_flow_tbl_data_entry *tbl_data;
7057 tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
7058 key->domain, error);
7060 return -rte_errno; /* No need to refill the error info */
7061 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7062 /* Lookup from cache. */
7063 LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
7064 if (matcher->crc == cache_matcher->crc &&
7065 matcher->priority == cache_matcher->priority &&
7066 !memcmp((const void *)matcher->mask.buf,
7067 (const void *)cache_matcher->mask.buf,
7068 cache_matcher->mask.size)) {
7070 "%s group %u priority %hd use %s "
7071 "matcher %p: refcnt %d++",
7072 key->domain ? "FDB" : "NIC", key->table_id,
7073 cache_matcher->priority,
7074 key->direction ? "tx" : "rx",
7075 (void *)cache_matcher,
7076 rte_atomic32_read(&cache_matcher->refcnt));
7077 rte_atomic32_inc(&cache_matcher->refcnt);
7078 dev_flow->handle->dvh.matcher = cache_matcher;
7079 /* old matcher should not make the table ref++. */
7080 flow_dv_tbl_resource_release(dev, tbl);
7084 /* Register new matcher. */
7085 cache_matcher = rte_calloc(__func__, 1, sizeof(*cache_matcher), 0);
7086 if (!cache_matcher) {
7087 flow_dv_tbl_resource_release(dev, tbl);
7088 return rte_flow_error_set(error, ENOMEM,
7089 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7090 "cannot allocate matcher memory");
7092 *cache_matcher = *matcher;
7093 dv_attr.match_criteria_enable =
7094 flow_dv_matcher_enable(cache_matcher->mask.buf);
7095 dv_attr.priority = matcher->priority;
7097 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
7098 cache_matcher->matcher_object =
7099 mlx5_glue->dv_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj);
7100 if (!cache_matcher->matcher_object) {
7101 rte_free(cache_matcher);
7102 #ifdef HAVE_MLX5DV_DR
7103 flow_dv_tbl_resource_release(dev, tbl);
7105 return rte_flow_error_set(error, ENOMEM,
7106 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7107 NULL, "cannot create matcher");
7109 /* Save the table information */
7110 cache_matcher->tbl = tbl;
7111 rte_atomic32_init(&cache_matcher->refcnt);
7112 /* only matcher ref++, table ref++ already done above in get API. */
7113 rte_atomic32_inc(&cache_matcher->refcnt);
7114 LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
7115 dev_flow->handle->dvh.matcher = cache_matcher;
7116 DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
7117 key->domain ? "FDB" : "NIC", key->table_id,
7118 cache_matcher->priority,
7119 key->direction ? "tx" : "rx", (void *)cache_matcher,
7120 rte_atomic32_read(&cache_matcher->refcnt));
7125 * Find existing tag resource or create and register a new one.
7127 * @param dev[in, out]
7128 * Pointer to rte_eth_dev structure.
7129 * @param[in, out] tag_be24
7130 * Tag value in big endian then R-shift 8.
7131 * @parm[in, out] dev_flow
7132 * Pointer to the dev_flow.
7134 * pointer to error structure.
7137 * 0 on success otherwise -errno and errno is set.
7140 flow_dv_tag_resource_register
7141 (struct rte_eth_dev *dev,
7143 struct mlx5_flow *dev_flow,
7144 struct rte_flow_error *error)
7146 struct mlx5_priv *priv = dev->data->dev_private;
7147 struct mlx5_ibv_shared *sh = priv->sh;
7148 struct mlx5_flow_dv_tag_resource *cache_resource;
7149 struct mlx5_hlist_entry *entry;
7151 /* Lookup a matching resource from cache. */
7152 entry = mlx5_hlist_lookup(sh->tag_table, (uint64_t)tag_be24);
7154 cache_resource = container_of
7155 (entry, struct mlx5_flow_dv_tag_resource, entry);
7156 rte_atomic32_inc(&cache_resource->refcnt);
7157 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
7158 dev_flow->dv.tag_resource = cache_resource;
7159 DRV_LOG(DEBUG, "cached tag resource %p: refcnt now %d++",
7160 (void *)cache_resource,
7161 rte_atomic32_read(&cache_resource->refcnt));
7164 /* Register new resource. */
7165 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG],
7166 &dev_flow->handle->dvh.rix_tag);
7167 if (!cache_resource)
7168 return rte_flow_error_set(error, ENOMEM,
7169 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7170 "cannot allocate resource memory");
7171 cache_resource->entry.key = (uint64_t)tag_be24;
7172 cache_resource->action = mlx5_glue->dv_create_flow_action_tag(tag_be24);
7173 if (!cache_resource->action) {
7174 rte_free(cache_resource);
7175 return rte_flow_error_set(error, ENOMEM,
7176 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7177 NULL, "cannot create action");
7179 rte_atomic32_init(&cache_resource->refcnt);
7180 rte_atomic32_inc(&cache_resource->refcnt);
7181 if (mlx5_hlist_insert(sh->tag_table, &cache_resource->entry)) {
7182 mlx5_glue->destroy_flow_action(cache_resource->action);
7183 rte_free(cache_resource);
7184 return rte_flow_error_set(error, EEXIST,
7185 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7186 NULL, "cannot insert tag");
7188 dev_flow->dv.tag_resource = cache_resource;
7189 DRV_LOG(DEBUG, "new tag resource %p: refcnt now %d++",
7190 (void *)cache_resource,
7191 rte_atomic32_read(&cache_resource->refcnt));
7199 * Pointer to Ethernet device.
7204 * 1 while a reference on it exists, 0 when freed.
7207 flow_dv_tag_release(struct rte_eth_dev *dev,
7210 struct mlx5_priv *priv = dev->data->dev_private;
7211 struct mlx5_ibv_shared *sh = priv->sh;
7212 struct mlx5_flow_dv_tag_resource *tag;
7214 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7217 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
7218 dev->data->port_id, (void *)tag,
7219 rte_atomic32_read(&tag->refcnt));
7220 if (rte_atomic32_dec_and_test(&tag->refcnt)) {
7221 claim_zero(mlx5_glue->destroy_flow_action(tag->action));
7222 mlx5_hlist_remove(sh->tag_table, &tag->entry);
7223 DRV_LOG(DEBUG, "port %u tag %p: removed",
7224 dev->data->port_id, (void *)tag);
7225 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7232 * Translate port ID action to vport.
7235 * Pointer to rte_eth_dev structure.
7237 * Pointer to the port ID action.
7238 * @param[out] dst_port_id
7239 * The target port ID.
7241 * Pointer to the error structure.
7244 * 0 on success, a negative errno value otherwise and rte_errno is set.
7247 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
7248 const struct rte_flow_action *action,
7249 uint32_t *dst_port_id,
7250 struct rte_flow_error *error)
7253 struct mlx5_priv *priv;
7254 const struct rte_flow_action_port_id *conf =
7255 (const struct rte_flow_action_port_id *)action->conf;
7257 port = conf->original ? dev->data->port_id : conf->id;
7258 priv = mlx5_port_to_eswitch_info(port, false);
7260 return rte_flow_error_set(error, -rte_errno,
7261 RTE_FLOW_ERROR_TYPE_ACTION,
7263 "No eswitch info was found for port");
7264 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
7266 * This parameter is transferred to
7267 * mlx5dv_dr_action_create_dest_ib_port().
7269 *dst_port_id = priv->ibv_port;
7272 * Legacy mode, no LAG configurations is supported.
7273 * This parameter is transferred to
7274 * mlx5dv_dr_action_create_dest_vport().
7276 *dst_port_id = priv->vport_id;
7282 * Add Tx queue matcher
7285 * Pointer to the dev struct.
7286 * @param[in, out] matcher
7288 * @param[in, out] key
7289 * Flow matcher value.
7291 * Flow pattern to translate.
7293 * Item is inner pattern.
7296 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
7297 void *matcher, void *key,
7298 const struct rte_flow_item *item)
7300 const struct mlx5_rte_flow_item_tx_queue *queue_m;
7301 const struct mlx5_rte_flow_item_tx_queue *queue_v;
7303 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7305 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7306 struct mlx5_txq_ctrl *txq;
7310 queue_m = (const void *)item->mask;
7313 queue_v = (const void *)item->spec;
7316 txq = mlx5_txq_get(dev, queue_v->queue);
7319 queue = txq->obj->sq->id;
7320 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
7321 MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
7322 queue & queue_m->queue);
7323 mlx5_txq_release(dev, queue_v->queue);
7327 * Set the hash fields according to the @p flow information.
7329 * @param[in] dev_flow
7330 * Pointer to the mlx5_flow.
7331 * @param[in] rss_desc
7332 * Pointer to the mlx5_flow_rss_desc.
7335 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
7336 struct mlx5_flow_rss_desc *rss_desc)
7338 uint64_t items = dev_flow->handle->layers;
7340 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
7342 dev_flow->hash_fields = 0;
7343 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
7344 if (rss_desc->level >= 2) {
7345 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
7349 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
7350 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
7351 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
7352 if (rss_types & ETH_RSS_L3_SRC_ONLY)
7353 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
7354 else if (rss_types & ETH_RSS_L3_DST_ONLY)
7355 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
7357 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
7359 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
7360 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
7361 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
7362 if (rss_types & ETH_RSS_L3_SRC_ONLY)
7363 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
7364 else if (rss_types & ETH_RSS_L3_DST_ONLY)
7365 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
7367 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
7370 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
7371 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
7372 if (rss_types & ETH_RSS_UDP) {
7373 if (rss_types & ETH_RSS_L4_SRC_ONLY)
7374 dev_flow->hash_fields |=
7375 IBV_RX_HASH_SRC_PORT_UDP;
7376 else if (rss_types & ETH_RSS_L4_DST_ONLY)
7377 dev_flow->hash_fields |=
7378 IBV_RX_HASH_DST_PORT_UDP;
7380 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
7382 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
7383 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
7384 if (rss_types & ETH_RSS_TCP) {
7385 if (rss_types & ETH_RSS_L4_SRC_ONLY)
7386 dev_flow->hash_fields |=
7387 IBV_RX_HASH_SRC_PORT_TCP;
7388 else if (rss_types & ETH_RSS_L4_DST_ONLY)
7389 dev_flow->hash_fields |=
7390 IBV_RX_HASH_DST_PORT_TCP;
7392 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
7398 * Fill the flow with DV spec, lock free
7399 * (mutex should be acquired by caller).
7402 * Pointer to rte_eth_dev structure.
7403 * @param[in, out] dev_flow
7404 * Pointer to the sub flow.
7406 * Pointer to the flow attributes.
7408 * Pointer to the list of items.
7409 * @param[in] actions
7410 * Pointer to the list of actions.
7412 * Pointer to the error structure.
7415 * 0 on success, a negative errno value otherwise and rte_errno is set.
7418 __flow_dv_translate(struct rte_eth_dev *dev,
7419 struct mlx5_flow *dev_flow,
7420 const struct rte_flow_attr *attr,
7421 const struct rte_flow_item items[],
7422 const struct rte_flow_action actions[],
7423 struct rte_flow_error *error)
7425 struct mlx5_priv *priv = dev->data->dev_private;
7426 struct mlx5_dev_config *dev_conf = &priv->config;
7427 struct rte_flow *flow = dev_flow->flow;
7428 struct mlx5_flow_handle *handle = dev_flow->handle;
7429 struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
7431 [!!priv->flow_nested_idx];
7432 uint64_t item_flags = 0;
7433 uint64_t last_item = 0;
7434 uint64_t action_flags = 0;
7435 uint64_t priority = attr->priority;
7436 struct mlx5_flow_dv_matcher matcher = {
7438 .size = sizeof(matcher.mask.buf),
7442 bool actions_end = false;
7444 struct mlx5_flow_dv_modify_hdr_resource res;
7445 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
7446 sizeof(struct mlx5_modification_cmd) *
7447 (MLX5_MAX_MODIFY_NUM + 1)];
7449 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
7450 union flow_dv_attr flow_attr = { .attr = 0 };
7452 union mlx5_flow_tbl_key tbl_key;
7453 uint32_t modify_action_position = UINT32_MAX;
7454 void *match_mask = matcher.mask.buf;
7455 void *match_value = dev_flow->dv.value.buf;
7456 uint8_t next_protocol = 0xff;
7457 struct rte_vlan_hdr vlan = { 0 };
7461 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
7462 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
7463 ret = mlx5_flow_group_to_table(attr, dev_flow->external, attr->group,
7464 !!priv->fdb_def_rule, &table, error);
7467 dev_flow->dv.group = table;
7469 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
7470 if (priority == MLX5_FLOW_PRIO_RSVD)
7471 priority = dev_conf->flow_prio - 1;
7472 /* number of actions must be set to 0 in case of dirty stack. */
7473 mhdr_res->actions_num = 0;
7474 for (; !actions_end ; actions++) {
7475 const struct rte_flow_action_queue *queue;
7476 const struct rte_flow_action_rss *rss;
7477 const struct rte_flow_action *action = actions;
7478 const struct rte_flow_action_count *count = action->conf;
7479 const uint8_t *rss_key;
7480 const struct rte_flow_action_jump *jump_data;
7481 const struct rte_flow_action_meter *mtr;
7482 struct mlx5_flow_tbl_resource *tbl;
7483 uint32_t port_id = 0;
7484 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
7485 int action_type = actions->type;
7486 const struct rte_flow_action *found_action = NULL;
7487 struct mlx5_flow_meter *fm = NULL;
7489 switch (action_type) {
7490 case RTE_FLOW_ACTION_TYPE_VOID:
7492 case RTE_FLOW_ACTION_TYPE_PORT_ID:
7493 if (flow_dv_translate_action_port_id(dev, action,
7496 memset(&port_id_resource, 0, sizeof(port_id_resource));
7497 port_id_resource.port_id = port_id;
7498 if (flow_dv_port_id_action_resource_register
7499 (dev, &port_id_resource, dev_flow, error))
7501 MLX5_ASSERT(!handle->rix_port_id_action);
7502 dev_flow->dv.actions[actions_n++] =
7503 dev_flow->dv.port_id_action->action;
7504 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
7505 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
7507 case RTE_FLOW_ACTION_TYPE_FLAG:
7508 action_flags |= MLX5_FLOW_ACTION_FLAG;
7509 dev_flow->handle->mark = 1;
7510 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7511 struct rte_flow_action_mark mark = {
7512 .id = MLX5_FLOW_MARK_DEFAULT,
7515 if (flow_dv_convert_action_mark(dev, &mark,
7519 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
7522 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
7524 * Only one FLAG or MARK is supported per device flow
7525 * right now. So the pointer to the tag resource must be
7526 * zero before the register process.
7528 MLX5_ASSERT(!handle->dvh.rix_tag);
7529 if (flow_dv_tag_resource_register(dev, tag_be,
7532 MLX5_ASSERT(dev_flow->dv.tag_resource);
7533 dev_flow->dv.actions[actions_n++] =
7534 dev_flow->dv.tag_resource->action;
7536 case RTE_FLOW_ACTION_TYPE_MARK:
7537 action_flags |= MLX5_FLOW_ACTION_MARK;
7538 dev_flow->handle->mark = 1;
7539 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7540 const struct rte_flow_action_mark *mark =
7541 (const struct rte_flow_action_mark *)
7544 if (flow_dv_convert_action_mark(dev, mark,
7548 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
7552 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
7553 /* Legacy (non-extensive) MARK action. */
7554 tag_be = mlx5_flow_mark_set
7555 (((const struct rte_flow_action_mark *)
7556 (actions->conf))->id);
7557 MLX5_ASSERT(!handle->dvh.rix_tag);
7558 if (flow_dv_tag_resource_register(dev, tag_be,
7561 MLX5_ASSERT(dev_flow->dv.tag_resource);
7562 dev_flow->dv.actions[actions_n++] =
7563 dev_flow->dv.tag_resource->action;
7565 case RTE_FLOW_ACTION_TYPE_SET_META:
7566 if (flow_dv_convert_action_set_meta
7567 (dev, mhdr_res, attr,
7568 (const struct rte_flow_action_set_meta *)
7569 actions->conf, error))
7571 action_flags |= MLX5_FLOW_ACTION_SET_META;
7573 case RTE_FLOW_ACTION_TYPE_SET_TAG:
7574 if (flow_dv_convert_action_set_tag
7576 (const struct rte_flow_action_set_tag *)
7577 actions->conf, error))
7579 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7581 case RTE_FLOW_ACTION_TYPE_DROP:
7582 action_flags |= MLX5_FLOW_ACTION_DROP;
7583 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
7585 case RTE_FLOW_ACTION_TYPE_QUEUE:
7586 queue = actions->conf;
7587 rss_desc->queue_num = 1;
7588 rss_desc->queue[0] = queue->index;
7589 action_flags |= MLX5_FLOW_ACTION_QUEUE;
7590 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
7592 case RTE_FLOW_ACTION_TYPE_RSS:
7593 rss = actions->conf;
7594 memcpy(rss_desc->queue, rss->queue,
7595 rss->queue_num * sizeof(uint16_t));
7596 rss_desc->queue_num = rss->queue_num;
7597 /* NULL RSS key indicates default RSS key. */
7598 rss_key = !rss->key ? rss_hash_default_key : rss->key;
7599 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
7601 * rss->level and rss.types should be set in advance
7602 * when expanding items for RSS.
7604 action_flags |= MLX5_FLOW_ACTION_RSS;
7605 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
7607 case RTE_FLOW_ACTION_TYPE_COUNT:
7608 if (!dev_conf->devx) {
7609 rte_errno = ENOTSUP;
7612 flow->counter = flow_dv_counter_alloc(dev,
7615 dev_flow->dv.group);
7618 dev_flow->dv.actions[actions_n++] =
7619 (flow_dv_counter_get_by_idx(dev,
7620 flow->counter, NULL))->action;
7621 action_flags |= MLX5_FLOW_ACTION_COUNT;
7624 if (rte_errno == ENOTSUP)
7625 return rte_flow_error_set
7627 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7629 "count action not supported");
7631 return rte_flow_error_set
7633 RTE_FLOW_ERROR_TYPE_ACTION,
7635 "cannot create counter"
7638 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
7639 dev_flow->dv.actions[actions_n++] =
7640 priv->sh->pop_vlan_action;
7641 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
7643 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
7644 if (!(action_flags &
7645 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
7646 flow_dev_get_vlan_info_from_items(items, &vlan);
7647 vlan.eth_proto = rte_be_to_cpu_16
7648 ((((const struct rte_flow_action_of_push_vlan *)
7649 actions->conf)->ethertype));
7650 found_action = mlx5_flow_find_action
7652 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
7654 mlx5_update_vlan_vid_pcp(found_action, &vlan);
7655 found_action = mlx5_flow_find_action
7657 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
7659 mlx5_update_vlan_vid_pcp(found_action, &vlan);
7660 if (flow_dv_create_action_push_vlan
7661 (dev, attr, &vlan, dev_flow, error))
7663 dev_flow->dv.actions[actions_n++] =
7664 dev_flow->dv.push_vlan_res->action;
7665 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
7667 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
7668 /* of_vlan_push action handled this action */
7669 MLX5_ASSERT(action_flags &
7670 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
7672 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
7673 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
7675 flow_dev_get_vlan_info_from_items(items, &vlan);
7676 mlx5_update_vlan_vid_pcp(actions, &vlan);
7677 /* If no VLAN push - this is a modify header action */
7678 if (flow_dv_convert_action_modify_vlan_vid
7679 (mhdr_res, actions, error))
7681 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
7683 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
7684 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
7685 if (flow_dv_create_action_l2_encap(dev, actions,
7690 dev_flow->dv.actions[actions_n++] =
7691 dev_flow->dv.encap_decap->verbs_action;
7692 action_flags |= MLX5_FLOW_ACTION_ENCAP;
7694 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
7695 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
7696 if (flow_dv_create_action_l2_decap(dev, dev_flow,
7700 dev_flow->dv.actions[actions_n++] =
7701 dev_flow->dv.encap_decap->verbs_action;
7702 action_flags |= MLX5_FLOW_ACTION_DECAP;
7704 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
7705 /* Handle encap with preceding decap. */
7706 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
7707 if (flow_dv_create_action_raw_encap
7708 (dev, actions, dev_flow, attr, error))
7710 dev_flow->dv.actions[actions_n++] =
7711 dev_flow->dv.encap_decap->verbs_action;
7713 /* Handle encap without preceding decap. */
7714 if (flow_dv_create_action_l2_encap
7715 (dev, actions, dev_flow, attr->transfer,
7718 dev_flow->dv.actions[actions_n++] =
7719 dev_flow->dv.encap_decap->verbs_action;
7721 action_flags |= MLX5_FLOW_ACTION_ENCAP;
7723 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
7724 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
7726 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
7727 if (flow_dv_create_action_l2_decap
7728 (dev, dev_flow, attr->transfer, error))
7730 dev_flow->dv.actions[actions_n++] =
7731 dev_flow->dv.encap_decap->verbs_action;
7733 /* If decap is followed by encap, handle it at encap. */
7734 action_flags |= MLX5_FLOW_ACTION_DECAP;
7736 case RTE_FLOW_ACTION_TYPE_JUMP:
7737 jump_data = action->conf;
7738 ret = mlx5_flow_group_to_table(attr, dev_flow->external,
7740 !!priv->fdb_def_rule,
7744 tbl = flow_dv_tbl_resource_get(dev, table,
7746 attr->transfer, error);
7748 return rte_flow_error_set
7750 RTE_FLOW_ERROR_TYPE_ACTION,
7752 "cannot create jump action.");
7753 if (flow_dv_jump_tbl_resource_register
7754 (dev, tbl, dev_flow, error)) {
7755 flow_dv_tbl_resource_release(dev, tbl);
7756 return rte_flow_error_set
7758 RTE_FLOW_ERROR_TYPE_ACTION,
7760 "cannot create jump action.");
7762 dev_flow->dv.actions[actions_n++] =
7763 dev_flow->dv.jump->action;
7764 action_flags |= MLX5_FLOW_ACTION_JUMP;
7765 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
7767 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
7768 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
7769 if (flow_dv_convert_action_modify_mac
7770 (mhdr_res, actions, error))
7772 action_flags |= actions->type ==
7773 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
7774 MLX5_FLOW_ACTION_SET_MAC_SRC :
7775 MLX5_FLOW_ACTION_SET_MAC_DST;
7777 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
7778 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
7779 if (flow_dv_convert_action_modify_ipv4
7780 (mhdr_res, actions, error))
7782 action_flags |= actions->type ==
7783 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
7784 MLX5_FLOW_ACTION_SET_IPV4_SRC :
7785 MLX5_FLOW_ACTION_SET_IPV4_DST;
7787 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
7788 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
7789 if (flow_dv_convert_action_modify_ipv6
7790 (mhdr_res, actions, error))
7792 action_flags |= actions->type ==
7793 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
7794 MLX5_FLOW_ACTION_SET_IPV6_SRC :
7795 MLX5_FLOW_ACTION_SET_IPV6_DST;
7797 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
7798 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
7799 if (flow_dv_convert_action_modify_tp
7800 (mhdr_res, actions, items,
7801 &flow_attr, dev_flow, !!(action_flags &
7802 MLX5_FLOW_ACTION_DECAP), error))
7804 action_flags |= actions->type ==
7805 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
7806 MLX5_FLOW_ACTION_SET_TP_SRC :
7807 MLX5_FLOW_ACTION_SET_TP_DST;
7809 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
7810 if (flow_dv_convert_action_modify_dec_ttl
7811 (mhdr_res, items, &flow_attr, dev_flow,
7813 MLX5_FLOW_ACTION_DECAP), error))
7815 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
7817 case RTE_FLOW_ACTION_TYPE_SET_TTL:
7818 if (flow_dv_convert_action_modify_ttl
7819 (mhdr_res, actions, items, &flow_attr,
7820 dev_flow, !!(action_flags &
7821 MLX5_FLOW_ACTION_DECAP), error))
7823 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
7825 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
7826 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
7827 if (flow_dv_convert_action_modify_tcp_seq
7828 (mhdr_res, actions, error))
7830 action_flags |= actions->type ==
7831 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
7832 MLX5_FLOW_ACTION_INC_TCP_SEQ :
7833 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
7836 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
7837 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
7838 if (flow_dv_convert_action_modify_tcp_ack
7839 (mhdr_res, actions, error))
7841 action_flags |= actions->type ==
7842 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
7843 MLX5_FLOW_ACTION_INC_TCP_ACK :
7844 MLX5_FLOW_ACTION_DEC_TCP_ACK;
7846 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
7847 if (flow_dv_convert_action_set_reg
7848 (mhdr_res, actions, error))
7850 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7852 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
7853 if (flow_dv_convert_action_copy_mreg
7854 (dev, mhdr_res, actions, error))
7856 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7858 case RTE_FLOW_ACTION_TYPE_METER:
7859 mtr = actions->conf;
7861 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
7864 return rte_flow_error_set(error,
7866 RTE_FLOW_ERROR_TYPE_ACTION,
7869 "or invalid parameters");
7870 flow->meter = fm->idx;
7872 /* Set the meter action. */
7874 fm = mlx5_ipool_get(priv->sh->ipool
7875 [MLX5_IPOOL_MTR], flow->meter);
7877 return rte_flow_error_set(error,
7879 RTE_FLOW_ERROR_TYPE_ACTION,
7882 "or invalid parameters");
7884 dev_flow->dv.actions[actions_n++] =
7885 fm->mfts->meter_action;
7886 action_flags |= MLX5_FLOW_ACTION_METER;
7888 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
7889 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
7892 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
7894 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
7895 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
7898 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
7900 case RTE_FLOW_ACTION_TYPE_END:
7902 if (mhdr_res->actions_num) {
7903 /* create modify action if needed. */
7904 if (flow_dv_modify_hdr_resource_register
7905 (dev, mhdr_res, dev_flow, error))
7907 dev_flow->dv.actions[modify_action_position] =
7908 handle->dvh.modify_hdr->verbs_action;
7914 if (mhdr_res->actions_num &&
7915 modify_action_position == UINT32_MAX)
7916 modify_action_position = actions_n++;
7918 dev_flow->dv.actions_n = actions_n;
7919 dev_flow->act_flags = action_flags;
7920 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
7921 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
7922 int item_type = items->type;
7924 switch (item_type) {
7925 case RTE_FLOW_ITEM_TYPE_PORT_ID:
7926 flow_dv_translate_item_port_id(dev, match_mask,
7927 match_value, items);
7928 last_item = MLX5_FLOW_ITEM_PORT_ID;
7930 case RTE_FLOW_ITEM_TYPE_ETH:
7931 flow_dv_translate_item_eth(match_mask, match_value,
7933 matcher.priority = MLX5_PRIORITY_MAP_L2;
7934 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
7935 MLX5_FLOW_LAYER_OUTER_L2;
7937 case RTE_FLOW_ITEM_TYPE_VLAN:
7938 flow_dv_translate_item_vlan(dev_flow,
7939 match_mask, match_value,
7941 matcher.priority = MLX5_PRIORITY_MAP_L2;
7942 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
7943 MLX5_FLOW_LAYER_INNER_VLAN) :
7944 (MLX5_FLOW_LAYER_OUTER_L2 |
7945 MLX5_FLOW_LAYER_OUTER_VLAN);
7947 case RTE_FLOW_ITEM_TYPE_IPV4:
7948 mlx5_flow_tunnel_ip_check(items, next_protocol,
7949 &item_flags, &tunnel);
7950 flow_dv_translate_item_ipv4(match_mask, match_value,
7951 items, item_flags, tunnel,
7952 dev_flow->dv.group);
7953 matcher.priority = MLX5_PRIORITY_MAP_L3;
7954 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
7955 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
7956 if (items->mask != NULL &&
7957 ((const struct rte_flow_item_ipv4 *)
7958 items->mask)->hdr.next_proto_id) {
7960 ((const struct rte_flow_item_ipv4 *)
7961 (items->spec))->hdr.next_proto_id;
7963 ((const struct rte_flow_item_ipv4 *)
7964 (items->mask))->hdr.next_proto_id;
7966 /* Reset for inner layer. */
7967 next_protocol = 0xff;
7970 case RTE_FLOW_ITEM_TYPE_IPV6:
7971 mlx5_flow_tunnel_ip_check(items, next_protocol,
7972 &item_flags, &tunnel);
7973 flow_dv_translate_item_ipv6(match_mask, match_value,
7974 items, item_flags, tunnel,
7975 dev_flow->dv.group);
7976 matcher.priority = MLX5_PRIORITY_MAP_L3;
7977 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
7978 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
7979 if (items->mask != NULL &&
7980 ((const struct rte_flow_item_ipv6 *)
7981 items->mask)->hdr.proto) {
7983 ((const struct rte_flow_item_ipv6 *)
7984 items->spec)->hdr.proto;
7986 ((const struct rte_flow_item_ipv6 *)
7987 items->mask)->hdr.proto;
7989 /* Reset for inner layer. */
7990 next_protocol = 0xff;
7993 case RTE_FLOW_ITEM_TYPE_TCP:
7994 flow_dv_translate_item_tcp(match_mask, match_value,
7996 matcher.priority = MLX5_PRIORITY_MAP_L4;
7997 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
7998 MLX5_FLOW_LAYER_OUTER_L4_TCP;
8000 case RTE_FLOW_ITEM_TYPE_UDP:
8001 flow_dv_translate_item_udp(match_mask, match_value,
8003 matcher.priority = MLX5_PRIORITY_MAP_L4;
8004 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
8005 MLX5_FLOW_LAYER_OUTER_L4_UDP;
8007 case RTE_FLOW_ITEM_TYPE_GRE:
8008 flow_dv_translate_item_gre(match_mask, match_value,
8010 matcher.priority = rss_desc->level >= 2 ?
8011 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8012 last_item = MLX5_FLOW_LAYER_GRE;
8014 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
8015 flow_dv_translate_item_gre_key(match_mask,
8016 match_value, items);
8017 last_item = MLX5_FLOW_LAYER_GRE_KEY;
8019 case RTE_FLOW_ITEM_TYPE_NVGRE:
8020 flow_dv_translate_item_nvgre(match_mask, match_value,
8022 matcher.priority = rss_desc->level >= 2 ?
8023 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8024 last_item = MLX5_FLOW_LAYER_GRE;
8026 case RTE_FLOW_ITEM_TYPE_VXLAN:
8027 flow_dv_translate_item_vxlan(match_mask, match_value,
8029 matcher.priority = rss_desc->level >= 2 ?
8030 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8031 last_item = MLX5_FLOW_LAYER_VXLAN;
8033 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
8034 flow_dv_translate_item_vxlan_gpe(match_mask,
8037 matcher.priority = rss_desc->level >= 2 ?
8038 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8039 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
8041 case RTE_FLOW_ITEM_TYPE_GENEVE:
8042 flow_dv_translate_item_geneve(match_mask, match_value,
8044 matcher.priority = rss_desc->level >= 2 ?
8045 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8046 last_item = MLX5_FLOW_LAYER_GENEVE;
8048 case RTE_FLOW_ITEM_TYPE_MPLS:
8049 flow_dv_translate_item_mpls(match_mask, match_value,
8050 items, last_item, tunnel);
8051 matcher.priority = rss_desc->level >= 2 ?
8052 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8053 last_item = MLX5_FLOW_LAYER_MPLS;
8055 case RTE_FLOW_ITEM_TYPE_MARK:
8056 flow_dv_translate_item_mark(dev, match_mask,
8057 match_value, items);
8058 last_item = MLX5_FLOW_ITEM_MARK;
8060 case RTE_FLOW_ITEM_TYPE_META:
8061 flow_dv_translate_item_meta(dev, match_mask,
8062 match_value, attr, items);
8063 last_item = MLX5_FLOW_ITEM_METADATA;
8065 case RTE_FLOW_ITEM_TYPE_ICMP:
8066 flow_dv_translate_item_icmp(match_mask, match_value,
8068 last_item = MLX5_FLOW_LAYER_ICMP;
8070 case RTE_FLOW_ITEM_TYPE_ICMP6:
8071 flow_dv_translate_item_icmp6(match_mask, match_value,
8073 last_item = MLX5_FLOW_LAYER_ICMP6;
8075 case RTE_FLOW_ITEM_TYPE_TAG:
8076 flow_dv_translate_item_tag(dev, match_mask,
8077 match_value, items);
8078 last_item = MLX5_FLOW_ITEM_TAG;
8080 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
8081 flow_dv_translate_mlx5_item_tag(dev, match_mask,
8082 match_value, items);
8083 last_item = MLX5_FLOW_ITEM_TAG;
8085 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
8086 flow_dv_translate_item_tx_queue(dev, match_mask,
8089 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
8091 case RTE_FLOW_ITEM_TYPE_GTP:
8092 flow_dv_translate_item_gtp(match_mask, match_value,
8094 matcher.priority = rss_desc->level >= 2 ?
8095 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8096 last_item = MLX5_FLOW_LAYER_GTP;
8101 item_flags |= last_item;
8104 * When E-Switch mode is enabled, we have two cases where we need to
8105 * set the source port manually.
8106 * The first one, is in case of Nic steering rule, and the second is
8107 * E-Switch rule where no port_id item was found. In both cases
8108 * the source port is set according the current port in use.
8110 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
8111 (priv->representor || priv->master)) {
8112 if (flow_dv_translate_item_port_id(dev, match_mask,
8116 #ifdef RTE_LIBRTE_MLX5_DEBUG
8117 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
8118 dev_flow->dv.value.buf));
8121 * Layers may be already initialized from prefix flow if this dev_flow
8122 * is the suffix flow.
8124 handle->layers |= item_flags;
8125 if (action_flags & MLX5_FLOW_ACTION_RSS)
8126 flow_dv_hashfields_set(dev_flow, rss_desc);
8127 /* Register matcher. */
8128 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
8130 matcher.priority = mlx5_flow_adjust_priority(dev, priority,
8132 /* reserved field no needs to be set to 0 here. */
8133 tbl_key.domain = attr->transfer;
8134 tbl_key.direction = attr->egress;
8135 tbl_key.table_id = dev_flow->dv.group;
8136 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
8142 * Apply the flow to the NIC, lock free,
8143 * (mutex should be acquired by caller).
8146 * Pointer to the Ethernet device structure.
8147 * @param[in, out] flow
8148 * Pointer to flow structure.
8150 * Pointer to error structure.
8153 * 0 on success, a negative errno value otherwise and rte_errno is set.
8156 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
8157 struct rte_flow_error *error)
8159 struct mlx5_flow_dv_workspace *dv;
8160 struct mlx5_flow_handle *dh;
8161 struct mlx5_flow_handle_dv *dv_h;
8162 struct mlx5_flow *dev_flow;
8163 struct mlx5_priv *priv = dev->data->dev_private;
8164 uint32_t handle_idx;
8169 for (idx = priv->flow_idx - 1; idx >= priv->flow_nested_idx; idx--) {
8170 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[idx];
8172 dh = dev_flow->handle;
8175 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8177 dv->actions[n++] = priv->sh->esw_drop_action;
8179 struct mlx5_hrxq *drop_hrxq;
8180 drop_hrxq = mlx5_hrxq_drop_new(dev);
8184 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8186 "cannot get drop hash queue");
8190 * Drop queues will be released by the specify
8191 * mlx5_hrxq_drop_release() function. Assign
8192 * the special index to hrxq to mark the queue
8193 * has been allocated.
8195 dh->rix_hrxq = UINT32_MAX;
8196 dv->actions[n++] = drop_hrxq->action;
8198 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8199 struct mlx5_hrxq *hrxq;
8201 struct mlx5_flow_rss_desc *rss_desc =
8202 &((struct mlx5_flow_rss_desc *)priv->rss_desc)
8203 [!!priv->flow_nested_idx];
8205 MLX5_ASSERT(rss_desc->queue_num);
8206 hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
8207 MLX5_RSS_HASH_KEY_LEN,
8208 dev_flow->hash_fields,
8210 rss_desc->queue_num);
8212 hrxq_idx = mlx5_hrxq_new
8213 (dev, rss_desc->key,
8214 MLX5_RSS_HASH_KEY_LEN,
8215 dev_flow->hash_fields,
8217 rss_desc->queue_num,
8219 MLX5_FLOW_LAYER_TUNNEL));
8221 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8226 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8227 "cannot get hash queue");
8230 dh->rix_hrxq = hrxq_idx;
8231 dv->actions[n++] = hrxq->action;
8234 mlx5_glue->dv_create_flow(dv_h->matcher->matcher_object,
8235 (void *)&dv->value, n,
8238 rte_flow_error_set(error, errno,
8239 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8241 "hardware refuses to create flow");
8244 if (priv->vmwa_context &&
8245 dh->vf_vlan.tag && !dh->vf_vlan.created) {
8247 * The rule contains the VLAN pattern.
8248 * For VF we are going to create VLAN
8249 * interface to make hypervisor set correct
8250 * e-Switch vport context.
8252 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
8257 err = rte_errno; /* Save rte_errno before cleanup. */
8258 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
8259 handle_idx, dh, next) {
8260 /* hrxq is union, don't clear it if the flag is not set. */
8262 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8263 mlx5_hrxq_drop_release(dev);
8265 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8266 mlx5_hrxq_release(dev, dh->rix_hrxq);
8270 if (dh->vf_vlan.tag && dh->vf_vlan.created)
8271 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8273 rte_errno = err; /* Restore rte_errno. */
8278 * Release the flow matcher.
8281 * Pointer to Ethernet device.
8283 * Pointer to mlx5_flow_handle.
8286 * 1 while a reference on it exists, 0 when freed.
8289 flow_dv_matcher_release(struct rte_eth_dev *dev,
8290 struct mlx5_flow_handle *handle)
8292 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
8294 MLX5_ASSERT(matcher->matcher_object);
8295 DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
8296 dev->data->port_id, (void *)matcher,
8297 rte_atomic32_read(&matcher->refcnt));
8298 if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
8299 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8300 (matcher->matcher_object));
8301 LIST_REMOVE(matcher, next);
8302 /* table ref-- in release interface. */
8303 flow_dv_tbl_resource_release(dev, matcher->tbl);
8305 DRV_LOG(DEBUG, "port %u matcher %p: removed",
8306 dev->data->port_id, (void *)matcher);
8313 * Release an encap/decap resource.
8316 * Pointer to Ethernet device.
8318 * Pointer to mlx5_flow_handle.
8321 * 1 while a reference on it exists, 0 when freed.
8324 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
8325 struct mlx5_flow_handle *handle)
8327 struct mlx5_priv *priv = dev->data->dev_private;
8328 uint32_t idx = handle->dvh.rix_encap_decap;
8329 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
8331 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
8333 if (!cache_resource)
8335 MLX5_ASSERT(cache_resource->verbs_action);
8336 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
8337 (void *)cache_resource,
8338 rte_atomic32_read(&cache_resource->refcnt));
8339 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8340 claim_zero(mlx5_glue->destroy_flow_action
8341 (cache_resource->verbs_action));
8342 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
8343 &priv->sh->encaps_decaps, idx,
8344 cache_resource, next);
8345 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
8346 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
8347 (void *)cache_resource);
8354 * Release an jump to table action resource.
8357 * Pointer to Ethernet device.
8359 * Pointer to mlx5_flow_handle.
8362 * 1 while a reference on it exists, 0 when freed.
8365 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
8366 struct mlx5_flow_handle *handle)
8368 struct mlx5_priv *priv = dev->data->dev_private;
8369 struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
8370 struct mlx5_flow_tbl_data_entry *tbl_data;
8372 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
8376 cache_resource = &tbl_data->jump;
8377 MLX5_ASSERT(cache_resource->action);
8378 DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
8379 (void *)cache_resource,
8380 rte_atomic32_read(&cache_resource->refcnt));
8381 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8382 claim_zero(mlx5_glue->destroy_flow_action
8383 (cache_resource->action));
8384 /* jump action memory free is inside the table release. */
8385 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
8386 DRV_LOG(DEBUG, "jump table resource %p: removed",
8387 (void *)cache_resource);
8394 * Release a modify-header resource.
8397 * Pointer to mlx5_flow_handle.
8400 * 1 while a reference on it exists, 0 when freed.
8403 flow_dv_modify_hdr_resource_release(struct mlx5_flow_handle *handle)
8405 struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
8406 handle->dvh.modify_hdr;
8408 MLX5_ASSERT(cache_resource->verbs_action);
8409 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
8410 (void *)cache_resource,
8411 rte_atomic32_read(&cache_resource->refcnt));
8412 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8413 claim_zero(mlx5_glue->destroy_flow_action
8414 (cache_resource->verbs_action));
8415 LIST_REMOVE(cache_resource, next);
8416 rte_free(cache_resource);
8417 DRV_LOG(DEBUG, "modify-header resource %p: removed",
8418 (void *)cache_resource);
8425 * Release port ID action resource.
8428 * Pointer to Ethernet device.
8430 * Pointer to mlx5_flow_handle.
8433 * 1 while a reference on it exists, 0 when freed.
8436 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
8437 struct mlx5_flow_handle *handle)
8439 struct mlx5_priv *priv = dev->data->dev_private;
8440 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
8441 uint32_t idx = handle->rix_port_id_action;
8443 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
8445 if (!cache_resource)
8447 MLX5_ASSERT(cache_resource->action);
8448 DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
8449 (void *)cache_resource,
8450 rte_atomic32_read(&cache_resource->refcnt));
8451 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8452 claim_zero(mlx5_glue->destroy_flow_action
8453 (cache_resource->action));
8454 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
8455 &priv->sh->port_id_action_list, idx,
8456 cache_resource, next);
8457 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
8458 DRV_LOG(DEBUG, "port id action resource %p: removed",
8459 (void *)cache_resource);
8466 * Release push vlan action resource.
8469 * Pointer to Ethernet device.
8471 * Pointer to mlx5_flow_handle.
8474 * 1 while a reference on it exists, 0 when freed.
8477 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
8478 struct mlx5_flow_handle *handle)
8480 struct mlx5_priv *priv = dev->data->dev_private;
8481 uint32_t idx = handle->dvh.rix_push_vlan;
8482 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
8484 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
8486 if (!cache_resource)
8488 MLX5_ASSERT(cache_resource->action);
8489 DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
8490 (void *)cache_resource,
8491 rte_atomic32_read(&cache_resource->refcnt));
8492 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8493 claim_zero(mlx5_glue->destroy_flow_action
8494 (cache_resource->action));
8495 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
8496 &priv->sh->push_vlan_action_list, idx,
8497 cache_resource, next);
8498 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
8499 DRV_LOG(DEBUG, "push vlan action resource %p: removed",
8500 (void *)cache_resource);
8507 * Release the fate resource.
8510 * Pointer to Ethernet device.
8512 * Pointer to mlx5_flow_handle.
8515 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
8516 struct mlx5_flow_handle *handle)
8518 if (!handle->rix_fate)
8520 if (handle->fate_action == MLX5_FLOW_FATE_DROP)
8521 mlx5_hrxq_drop_release(dev);
8522 else if (handle->fate_action == MLX5_FLOW_FATE_QUEUE)
8523 mlx5_hrxq_release(dev, handle->rix_hrxq);
8524 else if (handle->fate_action == MLX5_FLOW_FATE_JUMP)
8525 flow_dv_jump_tbl_resource_release(dev, handle);
8526 else if (handle->fate_action == MLX5_FLOW_FATE_PORT_ID)
8527 flow_dv_port_id_action_resource_release(dev, handle);
8529 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
8530 handle->rix_fate = 0;
8534 * Remove the flow from the NIC but keeps it in memory.
8535 * Lock free, (mutex should be acquired by caller).
8538 * Pointer to Ethernet device.
8539 * @param[in, out] flow
8540 * Pointer to flow structure.
8543 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
8545 struct mlx5_flow_handle *dh;
8546 uint32_t handle_idx;
8547 struct mlx5_priv *priv = dev->data->dev_private;
8551 handle_idx = flow->dev_handles;
8552 while (handle_idx) {
8553 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8558 claim_zero(mlx5_glue->dv_destroy_flow(dh->ib_flow));
8561 if (dh->fate_action == MLX5_FLOW_FATE_DROP ||
8562 dh->fate_action == MLX5_FLOW_FATE_QUEUE)
8563 flow_dv_fate_resource_release(dev, dh);
8564 if (dh->vf_vlan.tag && dh->vf_vlan.created)
8565 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8566 handle_idx = dh->next.next;
8571 * Remove the flow from the NIC and the memory.
8572 * Lock free, (mutex should be acquired by caller).
8575 * Pointer to the Ethernet device structure.
8576 * @param[in, out] flow
8577 * Pointer to flow structure.
8580 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
8582 struct mlx5_flow_handle *dev_handle;
8583 struct mlx5_priv *priv = dev->data->dev_private;
8587 __flow_dv_remove(dev, flow);
8588 if (flow->counter) {
8589 flow_dv_counter_release(dev, flow->counter);
8593 struct mlx5_flow_meter *fm;
8595 fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
8598 mlx5_flow_meter_detach(fm);
8601 while (flow->dev_handles) {
8602 uint32_t tmp_idx = flow->dev_handles;
8604 dev_handle = mlx5_ipool_get(priv->sh->ipool
8605 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
8608 flow->dev_handles = dev_handle->next.next;
8609 if (dev_handle->dvh.matcher)
8610 flow_dv_matcher_release(dev, dev_handle);
8611 if (dev_handle->dvh.rix_encap_decap)
8612 flow_dv_encap_decap_resource_release(dev, dev_handle);
8613 if (dev_handle->dvh.modify_hdr)
8614 flow_dv_modify_hdr_resource_release(dev_handle);
8615 if (dev_handle->dvh.rix_push_vlan)
8616 flow_dv_push_vlan_action_resource_release(dev,
8618 if (dev_handle->dvh.rix_tag)
8619 flow_dv_tag_release(dev,
8620 dev_handle->dvh.rix_tag);
8621 flow_dv_fate_resource_release(dev, dev_handle);
8622 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8628 * Query a dv flow rule for its statistics via devx.
8631 * Pointer to Ethernet device.
8633 * Pointer to the sub flow.
8635 * data retrieved by the query.
8637 * Perform verbose error reporting if not NULL.
8640 * 0 on success, a negative errno value otherwise and rte_errno is set.
8643 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
8644 void *data, struct rte_flow_error *error)
8646 struct mlx5_priv *priv = dev->data->dev_private;
8647 struct rte_flow_query_count *qc = data;
8649 if (!priv->config.devx)
8650 return rte_flow_error_set(error, ENOTSUP,
8651 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8653 "counters are not supported");
8654 if (flow->counter) {
8655 uint64_t pkts, bytes;
8656 struct mlx5_flow_counter *cnt;
8658 cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
8660 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
8664 return rte_flow_error_set(error, -err,
8665 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8666 NULL, "cannot read counters");
8669 qc->hits = pkts - cnt->hits;
8670 qc->bytes = bytes - cnt->bytes;
8677 return rte_flow_error_set(error, EINVAL,
8678 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8680 "counters are not available");
8686 * @see rte_flow_query()
8690 flow_dv_query(struct rte_eth_dev *dev,
8691 struct rte_flow *flow __rte_unused,
8692 const struct rte_flow_action *actions __rte_unused,
8693 void *data __rte_unused,
8694 struct rte_flow_error *error __rte_unused)
8698 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
8699 switch (actions->type) {
8700 case RTE_FLOW_ACTION_TYPE_VOID:
8702 case RTE_FLOW_ACTION_TYPE_COUNT:
8703 ret = flow_dv_query_count(dev, flow, data, error);
8706 return rte_flow_error_set(error, ENOTSUP,
8707 RTE_FLOW_ERROR_TYPE_ACTION,
8709 "action not supported");
8716 * Destroy the meter table set.
8717 * Lock free, (mutex should be acquired by caller).
8720 * Pointer to Ethernet device.
8722 * Pointer to the meter table set.
8728 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
8729 struct mlx5_meter_domains_infos *tbl)
8731 struct mlx5_priv *priv = dev->data->dev_private;
8732 struct mlx5_meter_domains_infos *mtd =
8733 (struct mlx5_meter_domains_infos *)tbl;
8735 if (!mtd || !priv->config.dv_flow_en)
8737 if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
8738 claim_zero(mlx5_glue->dv_destroy_flow
8739 (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
8740 if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
8741 claim_zero(mlx5_glue->dv_destroy_flow
8742 (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
8743 if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
8744 claim_zero(mlx5_glue->dv_destroy_flow
8745 (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
8746 if (mtd->egress.color_matcher)
8747 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8748 (mtd->egress.color_matcher));
8749 if (mtd->egress.any_matcher)
8750 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8751 (mtd->egress.any_matcher));
8752 if (mtd->egress.tbl)
8753 flow_dv_tbl_resource_release(dev, mtd->egress.tbl);
8754 if (mtd->egress.sfx_tbl)
8755 flow_dv_tbl_resource_release(dev, mtd->egress.sfx_tbl);
8756 if (mtd->ingress.color_matcher)
8757 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8758 (mtd->ingress.color_matcher));
8759 if (mtd->ingress.any_matcher)
8760 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8761 (mtd->ingress.any_matcher));
8762 if (mtd->ingress.tbl)
8763 flow_dv_tbl_resource_release(dev, mtd->ingress.tbl);
8764 if (mtd->ingress.sfx_tbl)
8765 flow_dv_tbl_resource_release(dev, mtd->ingress.sfx_tbl);
8766 if (mtd->transfer.color_matcher)
8767 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8768 (mtd->transfer.color_matcher));
8769 if (mtd->transfer.any_matcher)
8770 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8771 (mtd->transfer.any_matcher));
8772 if (mtd->transfer.tbl)
8773 flow_dv_tbl_resource_release(dev, mtd->transfer.tbl);
8774 if (mtd->transfer.sfx_tbl)
8775 flow_dv_tbl_resource_release(dev, mtd->transfer.sfx_tbl);
8777 claim_zero(mlx5_glue->destroy_flow_action(mtd->drop_actn));
8782 /* Number of meter flow actions, count and jump or count and drop. */
8783 #define METER_ACTIONS 2
8786 * Create specify domain meter table and suffix table.
8789 * Pointer to Ethernet device.
8790 * @param[in,out] mtb
8791 * Pointer to DV meter table set.
8794 * @param[in] transfer
8796 * @param[in] color_reg_c_idx
8797 * Reg C index for color match.
8800 * 0 on success, -1 otherwise and rte_errno is set.
8803 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
8804 struct mlx5_meter_domains_infos *mtb,
8805 uint8_t egress, uint8_t transfer,
8806 uint32_t color_reg_c_idx)
8808 struct mlx5_priv *priv = dev->data->dev_private;
8809 struct mlx5_ibv_shared *sh = priv->sh;
8810 struct mlx5_flow_dv_match_params mask = {
8811 .size = sizeof(mask.buf),
8813 struct mlx5_flow_dv_match_params value = {
8814 .size = sizeof(value.buf),
8816 struct mlx5dv_flow_matcher_attr dv_attr = {
8817 .type = IBV_FLOW_ATTR_NORMAL,
8819 .match_criteria_enable = 0,
8820 .match_mask = (void *)&mask,
8822 void *actions[METER_ACTIONS];
8823 struct mlx5_meter_domain_info *dtb;
8824 struct rte_flow_error error;
8828 dtb = &mtb->transfer;
8832 dtb = &mtb->ingress;
8833 /* Create the meter table with METER level. */
8834 dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
8835 egress, transfer, &error);
8837 DRV_LOG(ERR, "Failed to create meter policer table.");
8840 /* Create the meter suffix table with SUFFIX level. */
8841 dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
8842 MLX5_FLOW_TABLE_LEVEL_SUFFIX,
8843 egress, transfer, &error);
8844 if (!dtb->sfx_tbl) {
8845 DRV_LOG(ERR, "Failed to create meter suffix table.");
8848 /* Create matchers, Any and Color. */
8849 dv_attr.priority = 3;
8850 dv_attr.match_criteria_enable = 0;
8851 dtb->any_matcher = mlx5_glue->dv_create_flow_matcher(sh->ctx,
8854 if (!dtb->any_matcher) {
8855 DRV_LOG(ERR, "Failed to create meter"
8856 " policer default matcher.");
8859 dv_attr.priority = 0;
8860 dv_attr.match_criteria_enable =
8861 1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
8862 flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
8863 rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
8864 dtb->color_matcher = mlx5_glue->dv_create_flow_matcher(sh->ctx,
8867 if (!dtb->color_matcher) {
8868 DRV_LOG(ERR, "Failed to create meter policer color matcher.");
8871 if (mtb->count_actns[RTE_MTR_DROPPED])
8872 actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
8873 actions[i++] = mtb->drop_actn;
8874 /* Default rule: lowest priority, match any, actions: drop. */
8875 dtb->policer_rules[RTE_MTR_DROPPED] =
8876 mlx5_glue->dv_create_flow(dtb->any_matcher,
8877 (void *)&value, i, actions);
8878 if (!dtb->policer_rules[RTE_MTR_DROPPED]) {
8879 DRV_LOG(ERR, "Failed to create meter policer drop rule.");
8888 * Create the needed meter and suffix tables.
8889 * Lock free, (mutex should be acquired by caller).
8892 * Pointer to Ethernet device.
8894 * Pointer to the flow meter.
8897 * Pointer to table set on success, NULL otherwise and rte_errno is set.
8899 static struct mlx5_meter_domains_infos *
8900 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
8901 const struct mlx5_flow_meter *fm)
8903 struct mlx5_priv *priv = dev->data->dev_private;
8904 struct mlx5_meter_domains_infos *mtb;
8908 if (!priv->mtr_en) {
8909 rte_errno = ENOTSUP;
8912 mtb = rte_calloc(__func__, 1, sizeof(*mtb), 0);
8914 DRV_LOG(ERR, "Failed to allocate memory for meter.");
8917 /* Create meter count actions */
8918 for (i = 0; i <= RTE_MTR_DROPPED; i++) {
8919 struct mlx5_flow_counter *cnt;
8920 if (!fm->policer_stats.cnt[i])
8922 cnt = flow_dv_counter_get_by_idx(dev,
8923 fm->policer_stats.cnt[i], NULL);
8924 mtb->count_actns[i] = cnt->action;
8926 /* Create drop action. */
8927 mtb->drop_actn = mlx5_glue->dr_create_flow_action_drop();
8928 if (!mtb->drop_actn) {
8929 DRV_LOG(ERR, "Failed to create drop action.");
8932 /* Egress meter table. */
8933 ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
8935 DRV_LOG(ERR, "Failed to prepare egress meter table.");
8938 /* Ingress meter table. */
8939 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
8941 DRV_LOG(ERR, "Failed to prepare ingress meter table.");
8944 /* FDB meter table. */
8945 if (priv->config.dv_esw_en) {
8946 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
8947 priv->mtr_color_reg);
8949 DRV_LOG(ERR, "Failed to prepare fdb meter table.");
8955 flow_dv_destroy_mtr_tbl(dev, mtb);
8960 * Destroy domain policer rule.
8963 * Pointer to domain table.
8966 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
8970 for (i = 0; i < RTE_MTR_DROPPED; i++) {
8971 if (dt->policer_rules[i]) {
8972 claim_zero(mlx5_glue->dv_destroy_flow
8973 (dt->policer_rules[i]));
8974 dt->policer_rules[i] = NULL;
8977 if (dt->jump_actn) {
8978 claim_zero(mlx5_glue->destroy_flow_action(dt->jump_actn));
8979 dt->jump_actn = NULL;
8984 * Destroy policer rules.
8987 * Pointer to Ethernet device.
8989 * Pointer to flow meter structure.
8991 * Pointer to flow attributes.
8997 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
8998 const struct mlx5_flow_meter *fm,
8999 const struct rte_flow_attr *attr)
9001 struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
9006 flow_dv_destroy_domain_policer_rule(&mtb->egress);
9008 flow_dv_destroy_domain_policer_rule(&mtb->ingress);
9010 flow_dv_destroy_domain_policer_rule(&mtb->transfer);
9015 * Create specify domain meter policer rule.
9018 * Pointer to flow meter structure.
9020 * Pointer to DV meter table set.
9021 * @param[in] mtr_reg_c
9022 * Color match REG_C.
9025 * 0 on success, -1 otherwise.
9028 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
9029 struct mlx5_meter_domain_info *dtb,
9032 struct mlx5_flow_dv_match_params matcher = {
9033 .size = sizeof(matcher.buf),
9035 struct mlx5_flow_dv_match_params value = {
9036 .size = sizeof(value.buf),
9038 struct mlx5_meter_domains_infos *mtb = fm->mfts;
9039 void *actions[METER_ACTIONS];
9042 /* Create jump action. */
9043 if (!dtb->jump_actn)
9045 mlx5_glue->dr_create_flow_action_dest_flow_tbl
9046 (dtb->sfx_tbl->obj);
9047 if (!dtb->jump_actn) {
9048 DRV_LOG(ERR, "Failed to create policer jump action.");
9051 for (i = 0; i < RTE_MTR_DROPPED; i++) {
9054 flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
9055 rte_col_2_mlx5_col(i), UINT8_MAX);
9056 if (mtb->count_actns[i])
9057 actions[j++] = mtb->count_actns[i];
9058 if (fm->action[i] == MTR_POLICER_ACTION_DROP)
9059 actions[j++] = mtb->drop_actn;
9061 actions[j++] = dtb->jump_actn;
9062 dtb->policer_rules[i] =
9063 mlx5_glue->dv_create_flow(dtb->color_matcher,
9066 if (!dtb->policer_rules[i]) {
9067 DRV_LOG(ERR, "Failed to create policer rule.");
9078 * Create policer rules.
9081 * Pointer to Ethernet device.
9083 * Pointer to flow meter structure.
9085 * Pointer to flow attributes.
9088 * 0 on success, -1 otherwise.
9091 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
9092 struct mlx5_flow_meter *fm,
9093 const struct rte_flow_attr *attr)
9095 struct mlx5_priv *priv = dev->data->dev_private;
9096 struct mlx5_meter_domains_infos *mtb = fm->mfts;
9100 ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
9101 priv->mtr_color_reg);
9103 DRV_LOG(ERR, "Failed to create egress policer.");
9107 if (attr->ingress) {
9108 ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
9109 priv->mtr_color_reg);
9111 DRV_LOG(ERR, "Failed to create ingress policer.");
9115 if (attr->transfer) {
9116 ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
9117 priv->mtr_color_reg);
9119 DRV_LOG(ERR, "Failed to create transfer policer.");
9125 flow_dv_destroy_policer_rules(dev, fm, attr);
9130 * Query a devx counter.
9133 * Pointer to the Ethernet device structure.
9135 * Index to the flow counter.
9137 * Set to clear the counter statistics.
9139 * The statistics value of packets.
9141 * The statistics value of bytes.
9144 * 0 on success, otherwise return -1.
9147 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
9148 uint64_t *pkts, uint64_t *bytes)
9150 struct mlx5_priv *priv = dev->data->dev_private;
9151 struct mlx5_flow_counter *cnt;
9152 uint64_t inn_pkts, inn_bytes;
9155 if (!priv->config.devx)
9158 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
9161 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
9162 *pkts = inn_pkts - cnt->hits;
9163 *bytes = inn_bytes - cnt->bytes;
9165 cnt->hits = inn_pkts;
9166 cnt->bytes = inn_bytes;
9172 * Mutex-protected thunk to lock-free __flow_dv_translate().
9175 flow_dv_translate(struct rte_eth_dev *dev,
9176 struct mlx5_flow *dev_flow,
9177 const struct rte_flow_attr *attr,
9178 const struct rte_flow_item items[],
9179 const struct rte_flow_action actions[],
9180 struct rte_flow_error *error)
9184 flow_dv_shared_lock(dev);
9185 ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
9186 flow_dv_shared_unlock(dev);
9191 * Mutex-protected thunk to lock-free __flow_dv_apply().
9194 flow_dv_apply(struct rte_eth_dev *dev,
9195 struct rte_flow *flow,
9196 struct rte_flow_error *error)
9200 flow_dv_shared_lock(dev);
9201 ret = __flow_dv_apply(dev, flow, error);
9202 flow_dv_shared_unlock(dev);
9207 * Mutex-protected thunk to lock-free __flow_dv_remove().
9210 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
9212 flow_dv_shared_lock(dev);
9213 __flow_dv_remove(dev, flow);
9214 flow_dv_shared_unlock(dev);
9218 * Mutex-protected thunk to lock-free __flow_dv_destroy().
9221 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
9223 flow_dv_shared_lock(dev);
9224 __flow_dv_destroy(dev, flow);
9225 flow_dv_shared_unlock(dev);
9229 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
9232 flow_dv_counter_allocate(struct rte_eth_dev *dev)
9236 flow_dv_shared_lock(dev);
9237 cnt = flow_dv_counter_alloc(dev, 0, 0, 1);
9238 flow_dv_shared_unlock(dev);
9243 * Mutex-protected thunk to lock-free flow_dv_counter_release().
9246 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
9248 flow_dv_shared_lock(dev);
9249 flow_dv_counter_release(dev, cnt);
9250 flow_dv_shared_unlock(dev);
9253 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
9254 .validate = flow_dv_validate,
9255 .prepare = flow_dv_prepare,
9256 .translate = flow_dv_translate,
9257 .apply = flow_dv_apply,
9258 .remove = flow_dv_remove,
9259 .destroy = flow_dv_destroy,
9260 .query = flow_dv_query,
9261 .create_mtr_tbls = flow_dv_create_mtr_tbl,
9262 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
9263 .create_policer_rules = flow_dv_create_policer_rules,
9264 .destroy_policer_rules = flow_dv_destroy_policer_rules,
9265 .counter_alloc = flow_dv_counter_allocate,
9266 .counter_free = flow_dv_counter_free,
9267 .counter_query = flow_dv_counter_query,
9270 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */