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>
27 #include <rte_cycles.h>
30 #include <rte_vxlan.h>
33 #include <mlx5_glue.h>
34 #include <mlx5_devx_cmds.h>
37 #include "mlx5_defs.h"
39 #include "mlx5_flow.h"
40 #include "mlx5_rxtx.h"
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
44 #ifndef HAVE_IBV_FLOW_DEVX_COUNTERS
45 #define MLX5DV_FLOW_ACTION_COUNTERS_DEVX 0
48 #ifndef HAVE_MLX5DV_DR_ESWITCH
49 #ifndef MLX5DV_FLOW_TABLE_TYPE_FDB
50 #define MLX5DV_FLOW_TABLE_TYPE_FDB 0
54 #ifndef HAVE_MLX5DV_DR
55 #define MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL 1
58 /* VLAN header definitions */
59 #define MLX5DV_FLOW_VLAN_PCP_SHIFT 13
60 #define MLX5DV_FLOW_VLAN_PCP_MASK (0x7 << MLX5DV_FLOW_VLAN_PCP_SHIFT)
61 #define MLX5DV_FLOW_VLAN_VID_MASK 0x0fff
62 #define MLX5DV_FLOW_VLAN_PCP_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK)
63 #define MLX5DV_FLOW_VLAN_VID_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_VID_MASK)
78 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
79 struct mlx5_flow_tbl_resource *tbl);
82 * Initialize flow attributes structure according to flow items' types.
84 * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
85 * mode. For tunnel mode, the items to be modified are the outermost ones.
88 * Pointer to item specification.
90 * Pointer to flow attributes structure.
92 * Pointer to the sub flow.
93 * @param[in] tunnel_decap
94 * Whether action is after tunnel decapsulation.
97 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
98 struct mlx5_flow *dev_flow, bool tunnel_decap)
100 uint64_t layers = dev_flow->handle->layers;
103 * If layers is already initialized, it means this dev_flow is the
104 * suffix flow, the layers flags is set by the prefix flow. Need to
105 * use the layer flags from prefix flow as the suffix flow may not
106 * have the user defined items as the flow is split.
109 if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
111 else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
113 if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
115 else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
120 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
121 uint8_t next_protocol = 0xff;
122 switch (item->type) {
123 case RTE_FLOW_ITEM_TYPE_GRE:
124 case RTE_FLOW_ITEM_TYPE_NVGRE:
125 case RTE_FLOW_ITEM_TYPE_VXLAN:
126 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
127 case RTE_FLOW_ITEM_TYPE_GENEVE:
128 case RTE_FLOW_ITEM_TYPE_MPLS:
132 case RTE_FLOW_ITEM_TYPE_IPV4:
135 if (item->mask != NULL &&
136 ((const struct rte_flow_item_ipv4 *)
137 item->mask)->hdr.next_proto_id)
139 ((const struct rte_flow_item_ipv4 *)
140 (item->spec))->hdr.next_proto_id &
141 ((const struct rte_flow_item_ipv4 *)
142 (item->mask))->hdr.next_proto_id;
143 if ((next_protocol == IPPROTO_IPIP ||
144 next_protocol == IPPROTO_IPV6) && tunnel_decap)
147 case RTE_FLOW_ITEM_TYPE_IPV6:
150 if (item->mask != NULL &&
151 ((const struct rte_flow_item_ipv6 *)
152 item->mask)->hdr.proto)
154 ((const struct rte_flow_item_ipv6 *)
155 (item->spec))->hdr.proto &
156 ((const struct rte_flow_item_ipv6 *)
157 (item->mask))->hdr.proto;
158 if ((next_protocol == IPPROTO_IPIP ||
159 next_protocol == IPPROTO_IPV6) && tunnel_decap)
162 case RTE_FLOW_ITEM_TYPE_UDP:
166 case RTE_FLOW_ITEM_TYPE_TCP:
178 * Convert rte_mtr_color to mlx5 color.
187 rte_col_2_mlx5_col(enum rte_color rcol)
190 case RTE_COLOR_GREEN:
191 return MLX5_FLOW_COLOR_GREEN;
192 case RTE_COLOR_YELLOW:
193 return MLX5_FLOW_COLOR_YELLOW;
195 return MLX5_FLOW_COLOR_RED;
199 return MLX5_FLOW_COLOR_UNDEFINED;
202 struct field_modify_info {
203 uint32_t size; /* Size of field in protocol header, in bytes. */
204 uint32_t offset; /* Offset of field in protocol header, in bytes. */
205 enum mlx5_modification_field id;
208 struct field_modify_info modify_eth[] = {
209 {4, 0, MLX5_MODI_OUT_DMAC_47_16},
210 {2, 4, MLX5_MODI_OUT_DMAC_15_0},
211 {4, 6, MLX5_MODI_OUT_SMAC_47_16},
212 {2, 10, MLX5_MODI_OUT_SMAC_15_0},
216 struct field_modify_info modify_vlan_out_first_vid[] = {
217 /* Size in bits !!! */
218 {12, 0, MLX5_MODI_OUT_FIRST_VID},
222 struct field_modify_info modify_ipv4[] = {
223 {1, 1, MLX5_MODI_OUT_IP_DSCP},
224 {1, 8, MLX5_MODI_OUT_IPV4_TTL},
225 {4, 12, MLX5_MODI_OUT_SIPV4},
226 {4, 16, MLX5_MODI_OUT_DIPV4},
230 struct field_modify_info modify_ipv6[] = {
231 {1, 0, MLX5_MODI_OUT_IP_DSCP},
232 {1, 7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
233 {4, 8, MLX5_MODI_OUT_SIPV6_127_96},
234 {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
235 {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
236 {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
237 {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
238 {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
239 {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
240 {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
244 struct field_modify_info modify_udp[] = {
245 {2, 0, MLX5_MODI_OUT_UDP_SPORT},
246 {2, 2, MLX5_MODI_OUT_UDP_DPORT},
250 struct field_modify_info modify_tcp[] = {
251 {2, 0, MLX5_MODI_OUT_TCP_SPORT},
252 {2, 2, MLX5_MODI_OUT_TCP_DPORT},
253 {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
254 {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
259 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
260 uint8_t next_protocol, uint64_t *item_flags,
263 MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
264 item->type == RTE_FLOW_ITEM_TYPE_IPV6);
265 if (next_protocol == IPPROTO_IPIP) {
266 *item_flags |= MLX5_FLOW_LAYER_IPIP;
269 if (next_protocol == IPPROTO_IPV6) {
270 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
276 * Acquire the synchronizing object to protect multithreaded access
277 * to shared dv context. Lock occurs only if context is actually
278 * shared, i.e. we have multiport IB device and representors are
282 * Pointer to the rte_eth_dev structure.
285 flow_dv_shared_lock(struct rte_eth_dev *dev)
287 struct mlx5_priv *priv = dev->data->dev_private;
288 struct mlx5_ibv_shared *sh = priv->sh;
290 if (sh->dv_refcnt > 1) {
293 ret = pthread_mutex_lock(&sh->dv_mutex);
300 flow_dv_shared_unlock(struct rte_eth_dev *dev)
302 struct mlx5_priv *priv = dev->data->dev_private;
303 struct mlx5_ibv_shared *sh = priv->sh;
305 if (sh->dv_refcnt > 1) {
308 ret = pthread_mutex_unlock(&sh->dv_mutex);
314 /* Update VLAN's VID/PCP based on input rte_flow_action.
317 * Pointer to struct rte_flow_action.
319 * Pointer to struct rte_vlan_hdr.
322 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
323 struct rte_vlan_hdr *vlan)
326 if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
328 ((const struct rte_flow_action_of_set_vlan_pcp *)
329 action->conf)->vlan_pcp;
330 vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
331 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
332 vlan->vlan_tci |= vlan_tci;
333 } else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
334 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
335 vlan->vlan_tci |= rte_be_to_cpu_16
336 (((const struct rte_flow_action_of_set_vlan_vid *)
337 action->conf)->vlan_vid);
342 * Fetch 1, 2, 3 or 4 byte field from the byte array
343 * and return as unsigned integer in host-endian format.
346 * Pointer to data array.
348 * Size of field to extract.
351 * converted field in host endian format.
353 static inline uint32_t
354 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
363 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
366 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
367 ret = (ret << 8) | *(data + sizeof(uint16_t));
370 ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
381 * Convert modify-header action to DV specification.
383 * Data length of each action is determined by provided field description
384 * and the item mask. Data bit offset and width of each action is determined
385 * by provided item mask.
388 * Pointer to item specification.
390 * Pointer to field modification information.
391 * For MLX5_MODIFICATION_TYPE_SET specifies destination field.
392 * For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
393 * For MLX5_MODIFICATION_TYPE_COPY specifies source field.
395 * Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
396 * Negative offset value sets the same offset as source offset.
397 * size field is ignored, value is taken from source field.
398 * @param[in,out] resource
399 * Pointer to the modify-header resource.
401 * Type of modification.
403 * Pointer to the error structure.
406 * 0 on success, a negative errno value otherwise and rte_errno is set.
409 flow_dv_convert_modify_action(struct rte_flow_item *item,
410 struct field_modify_info *field,
411 struct field_modify_info *dcopy,
412 struct mlx5_flow_dv_modify_hdr_resource *resource,
413 uint32_t type, struct rte_flow_error *error)
415 uint32_t i = resource->actions_num;
416 struct mlx5_modification_cmd *actions = resource->actions;
419 * The item and mask are provided in big-endian format.
420 * The fields should be presented as in big-endian format either.
421 * Mask must be always present, it defines the actual field width.
423 MLX5_ASSERT(item->mask);
424 MLX5_ASSERT(field->size);
431 if (i >= MLX5_MAX_MODIFY_NUM)
432 return rte_flow_error_set(error, EINVAL,
433 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
434 "too many items to modify");
435 /* Fetch variable byte size mask from the array. */
436 mask = flow_dv_fetch_field((const uint8_t *)item->mask +
437 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 * GTP flags are contained in 1 byte of the format:
1644 * -------------------------------------------
1645 * | bit | 0 - 2 | 3 | 4 | 5 | 6 | 7 |
1646 * |-----------------------------------------|
1647 * | value | Version | PT | Res | E | S | PN |
1648 * -------------------------------------------
1650 * Matching is supported only for GTP flags E, S, PN.
1652 #define MLX5_GTP_FLAGS_MASK 0x07
1655 * Validate VLAN item.
1658 * Item specification.
1659 * @param[in] item_flags
1660 * Bit-fields that holds the items detected until now.
1662 * Ethernet device flow is being created on.
1664 * Pointer to error structure.
1667 * 0 on success, a negative errno value otherwise and rte_errno is set.
1670 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
1671 uint64_t item_flags,
1672 struct rte_eth_dev *dev,
1673 struct rte_flow_error *error)
1675 const struct rte_flow_item_vlan *mask = item->mask;
1676 const struct rte_flow_item_vlan nic_mask = {
1677 .tci = RTE_BE16(UINT16_MAX),
1678 .inner_type = RTE_BE16(UINT16_MAX),
1680 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1682 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1683 MLX5_FLOW_LAYER_INNER_L4) :
1684 (MLX5_FLOW_LAYER_OUTER_L3 |
1685 MLX5_FLOW_LAYER_OUTER_L4);
1686 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1687 MLX5_FLOW_LAYER_OUTER_VLAN;
1689 if (item_flags & vlanm)
1690 return rte_flow_error_set(error, EINVAL,
1691 RTE_FLOW_ERROR_TYPE_ITEM, item,
1692 "multiple VLAN layers not supported");
1693 else if ((item_flags & l34m) != 0)
1694 return rte_flow_error_set(error, EINVAL,
1695 RTE_FLOW_ERROR_TYPE_ITEM, item,
1696 "VLAN cannot follow L3/L4 layer");
1698 mask = &rte_flow_item_vlan_mask;
1699 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1700 (const uint8_t *)&nic_mask,
1701 sizeof(struct rte_flow_item_vlan),
1705 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1706 struct mlx5_priv *priv = dev->data->dev_private;
1708 if (priv->vmwa_context) {
1710 * Non-NULL context means we have a virtual machine
1711 * and SR-IOV enabled, we have to create VLAN interface
1712 * to make hypervisor to setup E-Switch vport
1713 * context correctly. We avoid creating the multiple
1714 * VLAN interfaces, so we cannot support VLAN tag mask.
1716 return rte_flow_error_set(error, EINVAL,
1717 RTE_FLOW_ERROR_TYPE_ITEM,
1719 "VLAN tag mask is not"
1720 " supported in virtual"
1728 * Validate GTP item.
1731 * Pointer to the rte_eth_dev structure.
1733 * Item specification.
1734 * @param[in] item_flags
1735 * Bit-fields that holds the items detected until now.
1737 * Pointer to error structure.
1740 * 0 on success, a negative errno value otherwise and rte_errno is set.
1743 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1744 const struct rte_flow_item *item,
1745 uint64_t item_flags,
1746 struct rte_flow_error *error)
1748 struct mlx5_priv *priv = dev->data->dev_private;
1749 const struct rte_flow_item_gtp *spec = item->spec;
1750 const struct rte_flow_item_gtp *mask = item->mask;
1751 const struct rte_flow_item_gtp nic_mask = {
1752 .v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
1754 .teid = RTE_BE32(0xffffffff),
1757 if (!priv->config.hca_attr.tunnel_stateless_gtp)
1758 return rte_flow_error_set(error, ENOTSUP,
1759 RTE_FLOW_ERROR_TYPE_ITEM, item,
1760 "GTP support is not enabled");
1761 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1762 return rte_flow_error_set(error, ENOTSUP,
1763 RTE_FLOW_ERROR_TYPE_ITEM, item,
1764 "multiple tunnel layers not"
1766 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1767 return rte_flow_error_set(error, EINVAL,
1768 RTE_FLOW_ERROR_TYPE_ITEM, item,
1769 "no outer UDP layer found");
1771 mask = &rte_flow_item_gtp_mask;
1772 if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
1773 return rte_flow_error_set(error, ENOTSUP,
1774 RTE_FLOW_ERROR_TYPE_ITEM, item,
1775 "Match is supported for GTP"
1777 return mlx5_flow_item_acceptable
1778 (item, (const uint8_t *)mask,
1779 (const uint8_t *)&nic_mask,
1780 sizeof(struct rte_flow_item_gtp),
1785 * Validate the pop VLAN action.
1788 * Pointer to the rte_eth_dev structure.
1789 * @param[in] action_flags
1790 * Holds the actions detected until now.
1792 * Pointer to the pop vlan action.
1793 * @param[in] item_flags
1794 * The items found in this flow rule.
1796 * Pointer to flow attributes.
1798 * Pointer to error structure.
1801 * 0 on success, a negative errno value otherwise and rte_errno is set.
1804 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
1805 uint64_t action_flags,
1806 const struct rte_flow_action *action,
1807 uint64_t item_flags,
1808 const struct rte_flow_attr *attr,
1809 struct rte_flow_error *error)
1811 const struct mlx5_priv *priv = dev->data->dev_private;
1815 if (!priv->sh->pop_vlan_action)
1816 return rte_flow_error_set(error, ENOTSUP,
1817 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1819 "pop vlan action is not supported");
1821 return rte_flow_error_set(error, ENOTSUP,
1822 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
1824 "pop vlan action not supported for "
1826 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1827 return rte_flow_error_set(error, ENOTSUP,
1828 RTE_FLOW_ERROR_TYPE_ACTION, action,
1829 "no support for multiple VLAN "
1831 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
1832 return rte_flow_error_set(error, ENOTSUP,
1833 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1835 "cannot pop vlan without a "
1836 "match on (outer) vlan in the flow");
1837 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1838 return rte_flow_error_set(error, EINVAL,
1839 RTE_FLOW_ERROR_TYPE_ACTION, action,
1840 "wrong action order, port_id should "
1841 "be after pop VLAN action");
1842 if (!attr->transfer && priv->representor)
1843 return rte_flow_error_set(error, ENOTSUP,
1844 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1845 "pop vlan action for VF representor "
1846 "not supported on NIC table");
1851 * Get VLAN default info from vlan match info.
1854 * the list of item specifications.
1856 * pointer VLAN info to fill to.
1859 * 0 on success, a negative errno value otherwise and rte_errno is set.
1862 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
1863 struct rte_vlan_hdr *vlan)
1865 const struct rte_flow_item_vlan nic_mask = {
1866 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
1867 MLX5DV_FLOW_VLAN_VID_MASK),
1868 .inner_type = RTE_BE16(0xffff),
1873 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1874 int type = items->type;
1876 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
1877 type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
1880 if (items->type != RTE_FLOW_ITEM_TYPE_END) {
1881 const struct rte_flow_item_vlan *vlan_m = items->mask;
1882 const struct rte_flow_item_vlan *vlan_v = items->spec;
1886 /* Only full match values are accepted */
1887 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
1888 MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
1889 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
1891 rte_be_to_cpu_16(vlan_v->tci &
1892 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
1894 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
1895 MLX5DV_FLOW_VLAN_VID_MASK_BE) {
1896 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
1898 rte_be_to_cpu_16(vlan_v->tci &
1899 MLX5DV_FLOW_VLAN_VID_MASK_BE);
1901 if (vlan_m->inner_type == nic_mask.inner_type)
1902 vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
1903 vlan_m->inner_type);
1908 * Validate the push VLAN action.
1911 * Pointer to the rte_eth_dev structure.
1912 * @param[in] action_flags
1913 * Holds the actions detected until now.
1914 * @param[in] item_flags
1915 * The items found in this flow rule.
1917 * Pointer to the action structure.
1919 * Pointer to flow attributes
1921 * Pointer to error structure.
1924 * 0 on success, a negative errno value otherwise and rte_errno is set.
1927 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
1928 uint64_t action_flags,
1929 const struct rte_flow_item_vlan *vlan_m,
1930 const struct rte_flow_action *action,
1931 const struct rte_flow_attr *attr,
1932 struct rte_flow_error *error)
1934 const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
1935 const struct mlx5_priv *priv = dev->data->dev_private;
1937 if (!attr->transfer && attr->ingress)
1938 return rte_flow_error_set(error, ENOTSUP,
1939 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1941 "push VLAN action not supported for "
1943 if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
1944 push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
1945 return rte_flow_error_set(error, EINVAL,
1946 RTE_FLOW_ERROR_TYPE_ACTION, action,
1947 "invalid vlan ethertype");
1948 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1949 return rte_flow_error_set(error, ENOTSUP,
1950 RTE_FLOW_ERROR_TYPE_ACTION, action,
1951 "no support for multiple VLAN "
1953 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1954 return rte_flow_error_set(error, EINVAL,
1955 RTE_FLOW_ERROR_TYPE_ACTION, action,
1956 "wrong action order, port_id should "
1957 "be after push VLAN");
1958 if (!attr->transfer && priv->representor)
1959 return rte_flow_error_set(error, ENOTSUP,
1960 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1961 "push vlan action for VF representor "
1962 "not supported on NIC table");
1964 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
1965 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
1966 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
1967 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
1968 !(mlx5_flow_find_action
1969 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
1970 return rte_flow_error_set(error, EINVAL,
1971 RTE_FLOW_ERROR_TYPE_ACTION, action,
1972 "not full match mask on VLAN PCP and "
1973 "there is no of_set_vlan_pcp action, "
1974 "push VLAN action cannot figure out "
1977 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
1978 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
1979 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
1980 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
1981 !(mlx5_flow_find_action
1982 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
1983 return rte_flow_error_set(error, EINVAL,
1984 RTE_FLOW_ERROR_TYPE_ACTION, action,
1985 "not full match mask on VLAN VID and "
1986 "there is no of_set_vlan_vid action, "
1987 "push VLAN action cannot figure out "
1994 * Validate the set VLAN PCP.
1996 * @param[in] action_flags
1997 * Holds the actions detected until now.
1998 * @param[in] actions
1999 * Pointer to the list of actions remaining in the flow rule.
2001 * Pointer to error structure.
2004 * 0 on success, a negative errno value otherwise and rte_errno is set.
2007 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
2008 const struct rte_flow_action actions[],
2009 struct rte_flow_error *error)
2011 const struct rte_flow_action *action = actions;
2012 const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2014 if (conf->vlan_pcp > 7)
2015 return rte_flow_error_set(error, EINVAL,
2016 RTE_FLOW_ERROR_TYPE_ACTION, action,
2017 "VLAN PCP value is too big");
2018 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2019 return rte_flow_error_set(error, ENOTSUP,
2020 RTE_FLOW_ERROR_TYPE_ACTION, action,
2021 "set VLAN PCP action must follow "
2022 "the push VLAN action");
2023 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2024 return rte_flow_error_set(error, ENOTSUP,
2025 RTE_FLOW_ERROR_TYPE_ACTION, action,
2026 "Multiple VLAN PCP modification are "
2028 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2029 return rte_flow_error_set(error, EINVAL,
2030 RTE_FLOW_ERROR_TYPE_ACTION, action,
2031 "wrong action order, port_id should "
2032 "be after set VLAN PCP");
2037 * Validate the set VLAN VID.
2039 * @param[in] item_flags
2040 * Holds the items detected in this rule.
2041 * @param[in] action_flags
2042 * Holds the actions detected until now.
2043 * @param[in] actions
2044 * Pointer to the list of actions remaining in the flow rule.
2046 * Pointer to error structure.
2049 * 0 on success, a negative errno value otherwise and rte_errno is set.
2052 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
2053 uint64_t action_flags,
2054 const struct rte_flow_action actions[],
2055 struct rte_flow_error *error)
2057 const struct rte_flow_action *action = actions;
2058 const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
2060 if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
2061 return rte_flow_error_set(error, EINVAL,
2062 RTE_FLOW_ERROR_TYPE_ACTION, action,
2063 "VLAN VID value is too big");
2064 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2065 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2066 return rte_flow_error_set(error, ENOTSUP,
2067 RTE_FLOW_ERROR_TYPE_ACTION, action,
2068 "set VLAN VID action must follow push"
2069 " VLAN action or match on VLAN item");
2070 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
2071 return rte_flow_error_set(error, ENOTSUP,
2072 RTE_FLOW_ERROR_TYPE_ACTION, action,
2073 "Multiple VLAN VID modifications are "
2075 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2076 return rte_flow_error_set(error, EINVAL,
2077 RTE_FLOW_ERROR_TYPE_ACTION, action,
2078 "wrong action order, port_id should "
2079 "be after set VLAN VID");
2084 * Validate the FLAG action.
2087 * Pointer to the rte_eth_dev structure.
2088 * @param[in] action_flags
2089 * Holds the actions detected until now.
2091 * Pointer to flow attributes
2093 * Pointer to error structure.
2096 * 0 on success, a negative errno value otherwise and rte_errno is set.
2099 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2100 uint64_t action_flags,
2101 const struct rte_flow_attr *attr,
2102 struct rte_flow_error *error)
2104 struct mlx5_priv *priv = dev->data->dev_private;
2105 struct mlx5_dev_config *config = &priv->config;
2108 /* Fall back if no extended metadata register support. */
2109 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2110 return mlx5_flow_validate_action_flag(action_flags, attr,
2112 /* Extensive metadata mode requires registers. */
2113 if (!mlx5_flow_ext_mreg_supported(dev))
2114 return rte_flow_error_set(error, ENOTSUP,
2115 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2116 "no metadata registers "
2117 "to support flag action");
2118 if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2119 return rte_flow_error_set(error, ENOTSUP,
2120 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2121 "extended metadata register"
2122 " isn't available");
2123 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2126 MLX5_ASSERT(ret > 0);
2127 if (action_flags & MLX5_FLOW_ACTION_MARK)
2128 return rte_flow_error_set(error, EINVAL,
2129 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2130 "can't mark and flag in same flow");
2131 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2132 return rte_flow_error_set(error, EINVAL,
2133 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2135 " actions in same flow");
2140 * Validate MARK action.
2143 * Pointer to the rte_eth_dev structure.
2145 * Pointer to action.
2146 * @param[in] action_flags
2147 * Holds the actions detected until now.
2149 * Pointer to flow attributes
2151 * Pointer to error structure.
2154 * 0 on success, a negative errno value otherwise and rte_errno is set.
2157 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2158 const struct rte_flow_action *action,
2159 uint64_t action_flags,
2160 const struct rte_flow_attr *attr,
2161 struct rte_flow_error *error)
2163 struct mlx5_priv *priv = dev->data->dev_private;
2164 struct mlx5_dev_config *config = &priv->config;
2165 const struct rte_flow_action_mark *mark = action->conf;
2168 /* Fall back if no extended metadata register support. */
2169 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2170 return mlx5_flow_validate_action_mark(action, action_flags,
2172 /* Extensive metadata mode requires registers. */
2173 if (!mlx5_flow_ext_mreg_supported(dev))
2174 return rte_flow_error_set(error, ENOTSUP,
2175 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2176 "no metadata registers "
2177 "to support mark action");
2178 if (!priv->sh->dv_mark_mask)
2179 return rte_flow_error_set(error, ENOTSUP,
2180 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2181 "extended metadata register"
2182 " isn't available");
2183 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2186 MLX5_ASSERT(ret > 0);
2188 return rte_flow_error_set(error, EINVAL,
2189 RTE_FLOW_ERROR_TYPE_ACTION, action,
2190 "configuration cannot be null");
2191 if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2192 return rte_flow_error_set(error, EINVAL,
2193 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2195 "mark id exceeds the limit");
2196 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2197 return rte_flow_error_set(error, EINVAL,
2198 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2199 "can't flag and mark in same flow");
2200 if (action_flags & MLX5_FLOW_ACTION_MARK)
2201 return rte_flow_error_set(error, EINVAL,
2202 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2203 "can't have 2 mark actions in same"
2209 * Validate SET_META action.
2212 * Pointer to the rte_eth_dev structure.
2214 * Pointer to the action structure.
2215 * @param[in] action_flags
2216 * Holds the actions detected until now.
2218 * Pointer to flow attributes
2220 * Pointer to error structure.
2223 * 0 on success, a negative errno value otherwise and rte_errno is set.
2226 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2227 const struct rte_flow_action *action,
2228 uint64_t action_flags __rte_unused,
2229 const struct rte_flow_attr *attr,
2230 struct rte_flow_error *error)
2232 const struct rte_flow_action_set_meta *conf;
2233 uint32_t nic_mask = UINT32_MAX;
2236 if (!mlx5_flow_ext_mreg_supported(dev))
2237 return rte_flow_error_set(error, ENOTSUP,
2238 RTE_FLOW_ERROR_TYPE_ACTION, action,
2239 "extended metadata register"
2240 " isn't supported");
2241 reg = flow_dv_get_metadata_reg(dev, attr, error);
2244 if (reg != REG_A && reg != REG_B) {
2245 struct mlx5_priv *priv = dev->data->dev_private;
2247 nic_mask = priv->sh->dv_meta_mask;
2249 if (!(action->conf))
2250 return rte_flow_error_set(error, EINVAL,
2251 RTE_FLOW_ERROR_TYPE_ACTION, action,
2252 "configuration cannot be null");
2253 conf = (const struct rte_flow_action_set_meta *)action->conf;
2255 return rte_flow_error_set(error, EINVAL,
2256 RTE_FLOW_ERROR_TYPE_ACTION, action,
2257 "zero mask doesn't have any effect");
2258 if (conf->mask & ~nic_mask)
2259 return rte_flow_error_set(error, EINVAL,
2260 RTE_FLOW_ERROR_TYPE_ACTION, action,
2261 "meta data must be within reg C0");
2266 * Validate SET_TAG action.
2269 * Pointer to the rte_eth_dev structure.
2271 * Pointer to the action structure.
2272 * @param[in] action_flags
2273 * Holds the actions detected until now.
2275 * Pointer to flow attributes
2277 * Pointer to error structure.
2280 * 0 on success, a negative errno value otherwise and rte_errno is set.
2283 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2284 const struct rte_flow_action *action,
2285 uint64_t action_flags,
2286 const struct rte_flow_attr *attr,
2287 struct rte_flow_error *error)
2289 const struct rte_flow_action_set_tag *conf;
2290 const uint64_t terminal_action_flags =
2291 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2292 MLX5_FLOW_ACTION_RSS;
2295 if (!mlx5_flow_ext_mreg_supported(dev))
2296 return rte_flow_error_set(error, ENOTSUP,
2297 RTE_FLOW_ERROR_TYPE_ACTION, action,
2298 "extensive metadata register"
2299 " isn't supported");
2300 if (!(action->conf))
2301 return rte_flow_error_set(error, EINVAL,
2302 RTE_FLOW_ERROR_TYPE_ACTION, action,
2303 "configuration cannot be null");
2304 conf = (const struct rte_flow_action_set_tag *)action->conf;
2306 return rte_flow_error_set(error, EINVAL,
2307 RTE_FLOW_ERROR_TYPE_ACTION, action,
2308 "zero mask doesn't have any effect");
2309 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2312 if (!attr->transfer && attr->ingress &&
2313 (action_flags & terminal_action_flags))
2314 return rte_flow_error_set(error, EINVAL,
2315 RTE_FLOW_ERROR_TYPE_ACTION, action,
2316 "set_tag has no effect"
2317 " with terminal actions");
2322 * Validate count action.
2325 * Pointer to rte_eth_dev structure.
2327 * Pointer to error structure.
2330 * 0 on success, a negative errno value otherwise and rte_errno is set.
2333 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2334 struct rte_flow_error *error)
2336 struct mlx5_priv *priv = dev->data->dev_private;
2338 if (!priv->config.devx)
2340 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2344 return rte_flow_error_set
2346 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2348 "count action not supported");
2352 * Validate the L2 encap action.
2355 * Pointer to the rte_eth_dev structure.
2356 * @param[in] action_flags
2357 * Holds the actions detected until now.
2359 * Pointer to the action structure.
2361 * Pointer to flow attributes.
2363 * Pointer to error structure.
2366 * 0 on success, a negative errno value otherwise and rte_errno is set.
2369 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2370 uint64_t action_flags,
2371 const struct rte_flow_action *action,
2372 const struct rte_flow_attr *attr,
2373 struct rte_flow_error *error)
2375 const struct mlx5_priv *priv = dev->data->dev_private;
2377 if (!(action->conf))
2378 return rte_flow_error_set(error, EINVAL,
2379 RTE_FLOW_ERROR_TYPE_ACTION, action,
2380 "configuration cannot be null");
2381 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2382 return rte_flow_error_set(error, EINVAL,
2383 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2384 "can only have a single encap action "
2386 if (!attr->transfer && priv->representor)
2387 return rte_flow_error_set(error, ENOTSUP,
2388 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2389 "encap action for VF representor "
2390 "not supported on NIC table");
2395 * Validate a decap action.
2398 * Pointer to the rte_eth_dev structure.
2399 * @param[in] action_flags
2400 * Holds the actions detected until now.
2402 * Pointer to flow attributes
2404 * Pointer to error structure.
2407 * 0 on success, a negative errno value otherwise and rte_errno is set.
2410 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2411 uint64_t action_flags,
2412 const struct rte_flow_attr *attr,
2413 struct rte_flow_error *error)
2415 const struct mlx5_priv *priv = dev->data->dev_private;
2417 if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2418 return rte_flow_error_set(error, ENOTSUP,
2419 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2421 MLX5_FLOW_ACTION_DECAP ? "can only "
2422 "have a single decap action" : "decap "
2423 "after encap is not supported");
2424 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2425 return rte_flow_error_set(error, EINVAL,
2426 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2427 "can't have decap action after"
2430 return rte_flow_error_set(error, ENOTSUP,
2431 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2433 "decap action not supported for "
2435 if (!attr->transfer && priv->representor)
2436 return rte_flow_error_set(error, ENOTSUP,
2437 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2438 "decap action for VF representor "
2439 "not supported on NIC table");
2443 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2446 * Validate the raw encap and decap actions.
2449 * Pointer to the rte_eth_dev structure.
2451 * Pointer to the decap action.
2453 * Pointer to the encap action.
2455 * Pointer to flow attributes
2456 * @param[in/out] action_flags
2457 * Holds the actions detected until now.
2458 * @param[out] actions_n
2459 * pointer to the number of actions counter.
2461 * Pointer to error structure.
2464 * 0 on success, a negative errno value otherwise and rte_errno is set.
2467 flow_dv_validate_action_raw_encap_decap
2468 (struct rte_eth_dev *dev,
2469 const struct rte_flow_action_raw_decap *decap,
2470 const struct rte_flow_action_raw_encap *encap,
2471 const struct rte_flow_attr *attr, uint64_t *action_flags,
2472 int *actions_n, struct rte_flow_error *error)
2474 const struct mlx5_priv *priv = dev->data->dev_private;
2477 if (encap && (!encap->size || !encap->data))
2478 return rte_flow_error_set(error, EINVAL,
2479 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2480 "raw encap data cannot be empty");
2481 if (decap && encap) {
2482 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2483 encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2486 else if (encap->size <=
2487 MLX5_ENCAPSULATION_DECISION_SIZE &&
2489 MLX5_ENCAPSULATION_DECISION_SIZE)
2492 else if (encap->size >
2493 MLX5_ENCAPSULATION_DECISION_SIZE &&
2495 MLX5_ENCAPSULATION_DECISION_SIZE)
2496 /* 2 L2 actions: encap and decap. */
2499 return rte_flow_error_set(error,
2501 RTE_FLOW_ERROR_TYPE_ACTION,
2502 NULL, "unsupported too small "
2503 "raw decap and too small raw "
2504 "encap combination");
2507 ret = flow_dv_validate_action_decap(dev, *action_flags, attr,
2511 *action_flags |= MLX5_FLOW_ACTION_DECAP;
2515 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2516 return rte_flow_error_set(error, ENOTSUP,
2517 RTE_FLOW_ERROR_TYPE_ACTION,
2519 "small raw encap size");
2520 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2521 return rte_flow_error_set(error, EINVAL,
2522 RTE_FLOW_ERROR_TYPE_ACTION,
2524 "more than one encap action");
2525 if (!attr->transfer && priv->representor)
2526 return rte_flow_error_set
2528 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2529 "encap action for VF representor "
2530 "not supported on NIC table");
2531 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
2538 * Find existing encap/decap resource or create and register a new one.
2540 * @param[in, out] dev
2541 * Pointer to rte_eth_dev structure.
2542 * @param[in, out] resource
2543 * Pointer to encap/decap resource.
2544 * @parm[in, out] dev_flow
2545 * Pointer to the dev_flow.
2547 * pointer to error structure.
2550 * 0 on success otherwise -errno and errno is set.
2553 flow_dv_encap_decap_resource_register
2554 (struct rte_eth_dev *dev,
2555 struct mlx5_flow_dv_encap_decap_resource *resource,
2556 struct mlx5_flow *dev_flow,
2557 struct rte_flow_error *error)
2559 struct mlx5_priv *priv = dev->data->dev_private;
2560 struct mlx5_ibv_shared *sh = priv->sh;
2561 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2562 struct mlx5dv_dr_domain *domain;
2565 resource->flags = dev_flow->dv.group ? 0 : 1;
2566 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2567 domain = sh->fdb_domain;
2568 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2569 domain = sh->rx_domain;
2571 domain = sh->tx_domain;
2572 /* Lookup a matching resource from cache. */
2573 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], sh->encaps_decaps, idx,
2574 cache_resource, next) {
2575 if (resource->reformat_type == cache_resource->reformat_type &&
2576 resource->ft_type == cache_resource->ft_type &&
2577 resource->flags == cache_resource->flags &&
2578 resource->size == cache_resource->size &&
2579 !memcmp((const void *)resource->buf,
2580 (const void *)cache_resource->buf,
2582 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
2583 (void *)cache_resource,
2584 rte_atomic32_read(&cache_resource->refcnt));
2585 rte_atomic32_inc(&cache_resource->refcnt);
2586 dev_flow->handle->dvh.rix_encap_decap = idx;
2587 dev_flow->dv.encap_decap = cache_resource;
2591 /* Register new encap/decap resource. */
2592 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2593 &dev_flow->handle->dvh.rix_encap_decap);
2594 if (!cache_resource)
2595 return rte_flow_error_set(error, ENOMEM,
2596 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2597 "cannot allocate resource memory");
2598 *cache_resource = *resource;
2599 cache_resource->verbs_action =
2600 mlx5_glue->dv_create_flow_action_packet_reformat
2601 (sh->ctx, cache_resource->reformat_type,
2602 cache_resource->ft_type, domain, cache_resource->flags,
2603 cache_resource->size,
2604 (cache_resource->size ? cache_resource->buf : NULL));
2605 if (!cache_resource->verbs_action) {
2606 rte_free(cache_resource);
2607 return rte_flow_error_set(error, ENOMEM,
2608 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2609 NULL, "cannot create action");
2611 rte_atomic32_init(&cache_resource->refcnt);
2612 rte_atomic32_inc(&cache_resource->refcnt);
2613 ILIST_INSERT(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], &sh->encaps_decaps,
2614 dev_flow->handle->dvh.rix_encap_decap, cache_resource,
2616 dev_flow->dv.encap_decap = cache_resource;
2617 DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
2618 (void *)cache_resource,
2619 rte_atomic32_read(&cache_resource->refcnt));
2624 * Find existing table jump resource or create and register a new one.
2626 * @param[in, out] dev
2627 * Pointer to rte_eth_dev structure.
2628 * @param[in, out] tbl
2629 * Pointer to flow table resource.
2630 * @parm[in, out] dev_flow
2631 * Pointer to the dev_flow.
2633 * pointer to error structure.
2636 * 0 on success otherwise -errno and errno is set.
2639 flow_dv_jump_tbl_resource_register
2640 (struct rte_eth_dev *dev __rte_unused,
2641 struct mlx5_flow_tbl_resource *tbl,
2642 struct mlx5_flow *dev_flow,
2643 struct rte_flow_error *error)
2645 struct mlx5_flow_tbl_data_entry *tbl_data =
2646 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2650 cnt = rte_atomic32_read(&tbl_data->jump.refcnt);
2652 tbl_data->jump.action =
2653 mlx5_glue->dr_create_flow_action_dest_flow_tbl
2655 if (!tbl_data->jump.action)
2656 return rte_flow_error_set(error, ENOMEM,
2657 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2658 NULL, "cannot create jump action");
2659 DRV_LOG(DEBUG, "new jump table resource %p: refcnt %d++",
2660 (void *)&tbl_data->jump, cnt);
2662 /* old jump should not make the table ref++. */
2663 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
2664 MLX5_ASSERT(tbl_data->jump.action);
2665 DRV_LOG(DEBUG, "existed jump table resource %p: refcnt %d++",
2666 (void *)&tbl_data->jump, cnt);
2668 rte_atomic32_inc(&tbl_data->jump.refcnt);
2669 dev_flow->handle->rix_jump = tbl_data->idx;
2670 dev_flow->dv.jump = &tbl_data->jump;
2675 * Find existing table port ID resource or create and register a new one.
2677 * @param[in, out] dev
2678 * Pointer to rte_eth_dev structure.
2679 * @param[in, out] resource
2680 * Pointer to port ID action resource.
2681 * @parm[in, out] dev_flow
2682 * Pointer to the dev_flow.
2684 * pointer to error structure.
2687 * 0 on success otherwise -errno and errno is set.
2690 flow_dv_port_id_action_resource_register
2691 (struct rte_eth_dev *dev,
2692 struct mlx5_flow_dv_port_id_action_resource *resource,
2693 struct mlx5_flow *dev_flow,
2694 struct rte_flow_error *error)
2696 struct mlx5_priv *priv = dev->data->dev_private;
2697 struct mlx5_ibv_shared *sh = priv->sh;
2698 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
2701 /* Lookup a matching resource from cache. */
2702 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
2703 idx, cache_resource, next) {
2704 if (resource->port_id == cache_resource->port_id) {
2705 DRV_LOG(DEBUG, "port id action resource resource %p: "
2707 (void *)cache_resource,
2708 rte_atomic32_read(&cache_resource->refcnt));
2709 rte_atomic32_inc(&cache_resource->refcnt);
2710 dev_flow->handle->rix_port_id_action = idx;
2711 dev_flow->dv.port_id_action = cache_resource;
2715 /* Register new port id action resource. */
2716 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
2717 &dev_flow->handle->rix_port_id_action);
2718 if (!cache_resource)
2719 return rte_flow_error_set(error, ENOMEM,
2720 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2721 "cannot allocate resource memory");
2722 *cache_resource = *resource;
2724 * Depending on rdma_core version the glue routine calls
2725 * either mlx5dv_dr_action_create_dest_ib_port(domain, ibv_port)
2726 * or mlx5dv_dr_action_create_dest_vport(domain, vport_id).
2728 cache_resource->action =
2729 mlx5_glue->dr_create_flow_action_dest_port
2730 (priv->sh->fdb_domain, resource->port_id);
2731 if (!cache_resource->action) {
2732 rte_free(cache_resource);
2733 return rte_flow_error_set(error, ENOMEM,
2734 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2735 NULL, "cannot create action");
2737 rte_atomic32_init(&cache_resource->refcnt);
2738 rte_atomic32_inc(&cache_resource->refcnt);
2739 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
2740 dev_flow->handle->rix_port_id_action, cache_resource,
2742 dev_flow->dv.port_id_action = cache_resource;
2743 DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
2744 (void *)cache_resource,
2745 rte_atomic32_read(&cache_resource->refcnt));
2750 * Find existing push vlan resource or create and register a new one.
2752 * @param [in, out] dev
2753 * Pointer to rte_eth_dev structure.
2754 * @param[in, out] resource
2755 * Pointer to port ID action resource.
2756 * @parm[in, out] dev_flow
2757 * Pointer to the dev_flow.
2759 * pointer to error structure.
2762 * 0 on success otherwise -errno and errno is set.
2765 flow_dv_push_vlan_action_resource_register
2766 (struct rte_eth_dev *dev,
2767 struct mlx5_flow_dv_push_vlan_action_resource *resource,
2768 struct mlx5_flow *dev_flow,
2769 struct rte_flow_error *error)
2771 struct mlx5_priv *priv = dev->data->dev_private;
2772 struct mlx5_ibv_shared *sh = priv->sh;
2773 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
2774 struct mlx5dv_dr_domain *domain;
2777 /* Lookup a matching resource from cache. */
2778 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2779 sh->push_vlan_action_list, idx, cache_resource, next) {
2780 if (resource->vlan_tag == cache_resource->vlan_tag &&
2781 resource->ft_type == cache_resource->ft_type) {
2782 DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
2784 (void *)cache_resource,
2785 rte_atomic32_read(&cache_resource->refcnt));
2786 rte_atomic32_inc(&cache_resource->refcnt);
2787 dev_flow->handle->dvh.rix_push_vlan = idx;
2788 dev_flow->dv.push_vlan_res = cache_resource;
2792 /* Register new push_vlan action resource. */
2793 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2794 &dev_flow->handle->dvh.rix_push_vlan);
2795 if (!cache_resource)
2796 return rte_flow_error_set(error, ENOMEM,
2797 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2798 "cannot allocate resource memory");
2799 *cache_resource = *resource;
2800 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2801 domain = sh->fdb_domain;
2802 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2803 domain = sh->rx_domain;
2805 domain = sh->tx_domain;
2806 cache_resource->action =
2807 mlx5_glue->dr_create_flow_action_push_vlan(domain,
2808 resource->vlan_tag);
2809 if (!cache_resource->action) {
2810 rte_free(cache_resource);
2811 return rte_flow_error_set(error, ENOMEM,
2812 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2813 NULL, "cannot create action");
2815 rte_atomic32_init(&cache_resource->refcnt);
2816 rte_atomic32_inc(&cache_resource->refcnt);
2817 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2818 &sh->push_vlan_action_list,
2819 dev_flow->handle->dvh.rix_push_vlan,
2820 cache_resource, next);
2821 dev_flow->dv.push_vlan_res = cache_resource;
2822 DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
2823 (void *)cache_resource,
2824 rte_atomic32_read(&cache_resource->refcnt));
2828 * Get the size of specific rte_flow_item_type
2830 * @param[in] item_type
2831 * Tested rte_flow_item_type.
2834 * sizeof struct item_type, 0 if void or irrelevant.
2837 flow_dv_get_item_len(const enum rte_flow_item_type item_type)
2841 switch (item_type) {
2842 case RTE_FLOW_ITEM_TYPE_ETH:
2843 retval = sizeof(struct rte_flow_item_eth);
2845 case RTE_FLOW_ITEM_TYPE_VLAN:
2846 retval = sizeof(struct rte_flow_item_vlan);
2848 case RTE_FLOW_ITEM_TYPE_IPV4:
2849 retval = sizeof(struct rte_flow_item_ipv4);
2851 case RTE_FLOW_ITEM_TYPE_IPV6:
2852 retval = sizeof(struct rte_flow_item_ipv6);
2854 case RTE_FLOW_ITEM_TYPE_UDP:
2855 retval = sizeof(struct rte_flow_item_udp);
2857 case RTE_FLOW_ITEM_TYPE_TCP:
2858 retval = sizeof(struct rte_flow_item_tcp);
2860 case RTE_FLOW_ITEM_TYPE_VXLAN:
2861 retval = sizeof(struct rte_flow_item_vxlan);
2863 case RTE_FLOW_ITEM_TYPE_GRE:
2864 retval = sizeof(struct rte_flow_item_gre);
2866 case RTE_FLOW_ITEM_TYPE_NVGRE:
2867 retval = sizeof(struct rte_flow_item_nvgre);
2869 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2870 retval = sizeof(struct rte_flow_item_vxlan_gpe);
2872 case RTE_FLOW_ITEM_TYPE_MPLS:
2873 retval = sizeof(struct rte_flow_item_mpls);
2875 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
2883 #define MLX5_ENCAP_IPV4_VERSION 0x40
2884 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
2885 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
2886 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
2887 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
2888 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
2889 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
2892 * Convert the encap action data from list of rte_flow_item to raw buffer
2895 * Pointer to rte_flow_item objects list.
2897 * Pointer to the output buffer.
2899 * Pointer to the output buffer size.
2901 * Pointer to the error structure.
2904 * 0 on success, a negative errno value otherwise and rte_errno is set.
2907 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
2908 size_t *size, struct rte_flow_error *error)
2910 struct rte_ether_hdr *eth = NULL;
2911 struct rte_vlan_hdr *vlan = NULL;
2912 struct rte_ipv4_hdr *ipv4 = NULL;
2913 struct rte_ipv6_hdr *ipv6 = NULL;
2914 struct rte_udp_hdr *udp = NULL;
2915 struct rte_vxlan_hdr *vxlan = NULL;
2916 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
2917 struct rte_gre_hdr *gre = NULL;
2919 size_t temp_size = 0;
2922 return rte_flow_error_set(error, EINVAL,
2923 RTE_FLOW_ERROR_TYPE_ACTION,
2924 NULL, "invalid empty data");
2925 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2926 len = flow_dv_get_item_len(items->type);
2927 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
2928 return rte_flow_error_set(error, EINVAL,
2929 RTE_FLOW_ERROR_TYPE_ACTION,
2930 (void *)items->type,
2931 "items total size is too big"
2932 " for encap action");
2933 rte_memcpy((void *)&buf[temp_size], items->spec, len);
2934 switch (items->type) {
2935 case RTE_FLOW_ITEM_TYPE_ETH:
2936 eth = (struct rte_ether_hdr *)&buf[temp_size];
2938 case RTE_FLOW_ITEM_TYPE_VLAN:
2939 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
2941 return rte_flow_error_set(error, EINVAL,
2942 RTE_FLOW_ERROR_TYPE_ACTION,
2943 (void *)items->type,
2944 "eth header not found");
2945 if (!eth->ether_type)
2946 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
2948 case RTE_FLOW_ITEM_TYPE_IPV4:
2949 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
2951 return rte_flow_error_set(error, EINVAL,
2952 RTE_FLOW_ERROR_TYPE_ACTION,
2953 (void *)items->type,
2954 "neither eth nor vlan"
2956 if (vlan && !vlan->eth_proto)
2957 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2958 else if (eth && !eth->ether_type)
2959 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2960 if (!ipv4->version_ihl)
2961 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
2962 MLX5_ENCAP_IPV4_IHL_MIN;
2963 if (!ipv4->time_to_live)
2964 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
2966 case RTE_FLOW_ITEM_TYPE_IPV6:
2967 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
2969 return rte_flow_error_set(error, EINVAL,
2970 RTE_FLOW_ERROR_TYPE_ACTION,
2971 (void *)items->type,
2972 "neither eth nor vlan"
2974 if (vlan && !vlan->eth_proto)
2975 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
2976 else if (eth && !eth->ether_type)
2977 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
2978 if (!ipv6->vtc_flow)
2980 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
2981 if (!ipv6->hop_limits)
2982 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
2984 case RTE_FLOW_ITEM_TYPE_UDP:
2985 udp = (struct rte_udp_hdr *)&buf[temp_size];
2987 return rte_flow_error_set(error, EINVAL,
2988 RTE_FLOW_ERROR_TYPE_ACTION,
2989 (void *)items->type,
2990 "ip header not found");
2991 if (ipv4 && !ipv4->next_proto_id)
2992 ipv4->next_proto_id = IPPROTO_UDP;
2993 else if (ipv6 && !ipv6->proto)
2994 ipv6->proto = IPPROTO_UDP;
2996 case RTE_FLOW_ITEM_TYPE_VXLAN:
2997 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
2999 return rte_flow_error_set(error, EINVAL,
3000 RTE_FLOW_ERROR_TYPE_ACTION,
3001 (void *)items->type,
3002 "udp header not found");
3004 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
3005 if (!vxlan->vx_flags)
3007 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
3009 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3010 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
3012 return rte_flow_error_set(error, EINVAL,
3013 RTE_FLOW_ERROR_TYPE_ACTION,
3014 (void *)items->type,
3015 "udp header not found");
3016 if (!vxlan_gpe->proto)
3017 return rte_flow_error_set(error, EINVAL,
3018 RTE_FLOW_ERROR_TYPE_ACTION,
3019 (void *)items->type,
3020 "next protocol not found");
3023 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
3024 if (!vxlan_gpe->vx_flags)
3025 vxlan_gpe->vx_flags =
3026 MLX5_ENCAP_VXLAN_GPE_FLAGS;
3028 case RTE_FLOW_ITEM_TYPE_GRE:
3029 case RTE_FLOW_ITEM_TYPE_NVGRE:
3030 gre = (struct rte_gre_hdr *)&buf[temp_size];
3032 return rte_flow_error_set(error, EINVAL,
3033 RTE_FLOW_ERROR_TYPE_ACTION,
3034 (void *)items->type,
3035 "next protocol not found");
3037 return rte_flow_error_set(error, EINVAL,
3038 RTE_FLOW_ERROR_TYPE_ACTION,
3039 (void *)items->type,
3040 "ip header not found");
3041 if (ipv4 && !ipv4->next_proto_id)
3042 ipv4->next_proto_id = IPPROTO_GRE;
3043 else if (ipv6 && !ipv6->proto)
3044 ipv6->proto = IPPROTO_GRE;
3046 case RTE_FLOW_ITEM_TYPE_VOID:
3049 return rte_flow_error_set(error, EINVAL,
3050 RTE_FLOW_ERROR_TYPE_ACTION,
3051 (void *)items->type,
3052 "unsupported item type");
3062 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
3064 struct rte_ether_hdr *eth = NULL;
3065 struct rte_vlan_hdr *vlan = NULL;
3066 struct rte_ipv6_hdr *ipv6 = NULL;
3067 struct rte_udp_hdr *udp = NULL;
3071 eth = (struct rte_ether_hdr *)data;
3072 next_hdr = (char *)(eth + 1);
3073 proto = RTE_BE16(eth->ether_type);
3076 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
3077 vlan = (struct rte_vlan_hdr *)next_hdr;
3078 proto = RTE_BE16(vlan->eth_proto);
3079 next_hdr += sizeof(struct rte_vlan_hdr);
3082 /* HW calculates IPv4 csum. no need to proceed */
3083 if (proto == RTE_ETHER_TYPE_IPV4)
3086 /* non IPv4/IPv6 header. not supported */
3087 if (proto != RTE_ETHER_TYPE_IPV6) {
3088 return rte_flow_error_set(error, ENOTSUP,
3089 RTE_FLOW_ERROR_TYPE_ACTION,
3090 NULL, "Cannot offload non IPv4/IPv6");
3093 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3095 /* ignore non UDP */
3096 if (ipv6->proto != IPPROTO_UDP)
3099 udp = (struct rte_udp_hdr *)(ipv6 + 1);
3100 udp->dgram_cksum = 0;
3106 * Convert L2 encap action to DV specification.
3109 * Pointer to rte_eth_dev structure.
3111 * Pointer to action structure.
3112 * @param[in, out] dev_flow
3113 * Pointer to the mlx5_flow.
3114 * @param[in] transfer
3115 * Mark if the flow is E-Switch flow.
3117 * Pointer to the error structure.
3120 * 0 on success, a negative errno value otherwise and rte_errno is set.
3123 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3124 const struct rte_flow_action *action,
3125 struct mlx5_flow *dev_flow,
3127 struct rte_flow_error *error)
3129 const struct rte_flow_item *encap_data;
3130 const struct rte_flow_action_raw_encap *raw_encap_data;
3131 struct mlx5_flow_dv_encap_decap_resource res = {
3133 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3134 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3135 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3138 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3140 (const struct rte_flow_action_raw_encap *)action->conf;
3141 res.size = raw_encap_data->size;
3142 memcpy(res.buf, raw_encap_data->data, res.size);
3144 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3146 ((const struct rte_flow_action_vxlan_encap *)
3147 action->conf)->definition;
3150 ((const struct rte_flow_action_nvgre_encap *)
3151 action->conf)->definition;
3152 if (flow_dv_convert_encap_data(encap_data, res.buf,
3156 if (flow_dv_zero_encap_udp_csum(res.buf, error))
3158 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3159 return rte_flow_error_set(error, EINVAL,
3160 RTE_FLOW_ERROR_TYPE_ACTION,
3161 NULL, "can't create L2 encap action");
3166 * Convert L2 decap action to DV specification.
3169 * Pointer to rte_eth_dev structure.
3170 * @param[in, out] dev_flow
3171 * Pointer to the mlx5_flow.
3172 * @param[in] transfer
3173 * Mark if the flow is E-Switch flow.
3175 * Pointer to the error structure.
3178 * 0 on success, a negative errno value otherwise and rte_errno is set.
3181 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3182 struct mlx5_flow *dev_flow,
3184 struct rte_flow_error *error)
3186 struct mlx5_flow_dv_encap_decap_resource res = {
3189 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3190 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3191 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3194 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3195 return rte_flow_error_set(error, EINVAL,
3196 RTE_FLOW_ERROR_TYPE_ACTION,
3197 NULL, "can't create L2 decap action");
3202 * Convert raw decap/encap (L3 tunnel) action to DV specification.
3205 * Pointer to rte_eth_dev structure.
3207 * Pointer to action structure.
3208 * @param[in, out] dev_flow
3209 * Pointer to the mlx5_flow.
3211 * Pointer to the flow attributes.
3213 * Pointer to the error structure.
3216 * 0 on success, a negative errno value otherwise and rte_errno is set.
3219 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3220 const struct rte_flow_action *action,
3221 struct mlx5_flow *dev_flow,
3222 const struct rte_flow_attr *attr,
3223 struct rte_flow_error *error)
3225 const struct rte_flow_action_raw_encap *encap_data;
3226 struct mlx5_flow_dv_encap_decap_resource res;
3228 memset(&res, 0, sizeof(res));
3229 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3230 res.size = encap_data->size;
3231 memcpy(res.buf, encap_data->data, res.size);
3232 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3233 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3234 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3236 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3238 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3239 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3240 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3241 return rte_flow_error_set(error, EINVAL,
3242 RTE_FLOW_ERROR_TYPE_ACTION,
3243 NULL, "can't create encap action");
3248 * Create action push VLAN.
3251 * Pointer to rte_eth_dev structure.
3253 * Pointer to the flow attributes.
3255 * Pointer to the vlan to push to the Ethernet header.
3256 * @param[in, out] dev_flow
3257 * Pointer to the mlx5_flow.
3259 * Pointer to the error structure.
3262 * 0 on success, a negative errno value otherwise and rte_errno is set.
3265 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3266 const struct rte_flow_attr *attr,
3267 const struct rte_vlan_hdr *vlan,
3268 struct mlx5_flow *dev_flow,
3269 struct rte_flow_error *error)
3271 struct mlx5_flow_dv_push_vlan_action_resource res;
3273 memset(&res, 0, sizeof(res));
3275 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3278 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3280 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3281 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3282 return flow_dv_push_vlan_action_resource_register
3283 (dev, &res, dev_flow, error);
3287 * Validate the modify-header actions.
3289 * @param[in] action_flags
3290 * Holds the actions detected until now.
3292 * Pointer to the modify action.
3294 * Pointer to error structure.
3297 * 0 on success, a negative errno value otherwise and rte_errno is set.
3300 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3301 const struct rte_flow_action *action,
3302 struct rte_flow_error *error)
3304 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3305 return rte_flow_error_set(error, EINVAL,
3306 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3307 NULL, "action configuration not set");
3308 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3309 return rte_flow_error_set(error, EINVAL,
3310 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3311 "can't have encap action before"
3317 * Validate the modify-header MAC address actions.
3319 * @param[in] action_flags
3320 * Holds the actions detected until now.
3322 * Pointer to the modify action.
3323 * @param[in] item_flags
3324 * Holds the items detected.
3326 * Pointer to error structure.
3329 * 0 on success, a negative errno value otherwise and rte_errno is set.
3332 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3333 const struct rte_flow_action *action,
3334 const uint64_t item_flags,
3335 struct rte_flow_error *error)
3339 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3341 if (!(item_flags & MLX5_FLOW_LAYER_L2))
3342 return rte_flow_error_set(error, EINVAL,
3343 RTE_FLOW_ERROR_TYPE_ACTION,
3345 "no L2 item in pattern");
3351 * Validate the modify-header IPv4 address actions.
3353 * @param[in] action_flags
3354 * Holds the actions detected until now.
3356 * Pointer to the modify action.
3357 * @param[in] item_flags
3358 * Holds the items detected.
3360 * Pointer to error structure.
3363 * 0 on success, a negative errno value otherwise and rte_errno is set.
3366 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3367 const struct rte_flow_action *action,
3368 const uint64_t item_flags,
3369 struct rte_flow_error *error)
3374 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3376 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3377 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3378 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3379 if (!(item_flags & layer))
3380 return rte_flow_error_set(error, EINVAL,
3381 RTE_FLOW_ERROR_TYPE_ACTION,
3383 "no ipv4 item in pattern");
3389 * Validate the modify-header IPv6 address actions.
3391 * @param[in] action_flags
3392 * Holds the actions detected until now.
3394 * Pointer to the modify action.
3395 * @param[in] item_flags
3396 * Holds the items detected.
3398 * Pointer to error structure.
3401 * 0 on success, a negative errno value otherwise and rte_errno is set.
3404 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3405 const struct rte_flow_action *action,
3406 const uint64_t item_flags,
3407 struct rte_flow_error *error)
3412 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3414 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3415 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3416 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3417 if (!(item_flags & layer))
3418 return rte_flow_error_set(error, EINVAL,
3419 RTE_FLOW_ERROR_TYPE_ACTION,
3421 "no ipv6 item in pattern");
3427 * Validate the modify-header TP actions.
3429 * @param[in] action_flags
3430 * Holds the actions detected until now.
3432 * Pointer to the modify action.
3433 * @param[in] item_flags
3434 * Holds the items detected.
3436 * Pointer to error structure.
3439 * 0 on success, a negative errno value otherwise and rte_errno is set.
3442 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3443 const struct rte_flow_action *action,
3444 const uint64_t item_flags,
3445 struct rte_flow_error *error)
3450 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3452 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3453 MLX5_FLOW_LAYER_INNER_L4 :
3454 MLX5_FLOW_LAYER_OUTER_L4;
3455 if (!(item_flags & layer))
3456 return rte_flow_error_set(error, EINVAL,
3457 RTE_FLOW_ERROR_TYPE_ACTION,
3458 NULL, "no transport layer "
3465 * Validate the modify-header actions of increment/decrement
3466 * TCP Sequence-number.
3468 * @param[in] action_flags
3469 * Holds the actions detected until now.
3471 * Pointer to the modify action.
3472 * @param[in] item_flags
3473 * Holds the items detected.
3475 * Pointer to error structure.
3478 * 0 on success, a negative errno value otherwise and rte_errno is set.
3481 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3482 const struct rte_flow_action *action,
3483 const uint64_t item_flags,
3484 struct rte_flow_error *error)
3489 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3491 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3492 MLX5_FLOW_LAYER_INNER_L4_TCP :
3493 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3494 if (!(item_flags & layer))
3495 return rte_flow_error_set(error, EINVAL,
3496 RTE_FLOW_ERROR_TYPE_ACTION,
3497 NULL, "no TCP item in"
3499 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3500 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3501 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3502 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3503 return rte_flow_error_set(error, EINVAL,
3504 RTE_FLOW_ERROR_TYPE_ACTION,
3506 "cannot decrease and increase"
3507 " TCP sequence number"
3508 " at the same time");
3514 * Validate the modify-header actions of increment/decrement
3515 * TCP Acknowledgment number.
3517 * @param[in] action_flags
3518 * Holds the actions detected until now.
3520 * Pointer to the modify action.
3521 * @param[in] item_flags
3522 * Holds the items detected.
3524 * Pointer to error structure.
3527 * 0 on success, a negative errno value otherwise and rte_errno is set.
3530 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3531 const struct rte_flow_action *action,
3532 const uint64_t item_flags,
3533 struct rte_flow_error *error)
3538 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3540 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3541 MLX5_FLOW_LAYER_INNER_L4_TCP :
3542 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3543 if (!(item_flags & layer))
3544 return rte_flow_error_set(error, EINVAL,
3545 RTE_FLOW_ERROR_TYPE_ACTION,
3546 NULL, "no TCP item in"
3548 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3549 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3550 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3551 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3552 return rte_flow_error_set(error, EINVAL,
3553 RTE_FLOW_ERROR_TYPE_ACTION,
3555 "cannot decrease and increase"
3556 " TCP acknowledgment number"
3557 " at the same time");
3563 * Validate the modify-header TTL actions.
3565 * @param[in] action_flags
3566 * Holds the actions detected until now.
3568 * Pointer to the modify action.
3569 * @param[in] item_flags
3570 * Holds the items detected.
3572 * Pointer to error structure.
3575 * 0 on success, a negative errno value otherwise and rte_errno is set.
3578 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3579 const struct rte_flow_action *action,
3580 const uint64_t item_flags,
3581 struct rte_flow_error *error)
3586 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3588 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3589 MLX5_FLOW_LAYER_INNER_L3 :
3590 MLX5_FLOW_LAYER_OUTER_L3;
3591 if (!(item_flags & layer))
3592 return rte_flow_error_set(error, EINVAL,
3593 RTE_FLOW_ERROR_TYPE_ACTION,
3595 "no IP protocol in pattern");
3601 * Validate jump action.
3604 * Pointer to the jump action.
3605 * @param[in] action_flags
3606 * Holds the actions detected until now.
3607 * @param[in] attributes
3608 * Pointer to flow attributes
3609 * @param[in] external
3610 * Action belongs to flow rule created by request external to PMD.
3612 * Pointer to error structure.
3615 * 0 on success, a negative errno value otherwise and rte_errno is set.
3618 flow_dv_validate_action_jump(const struct rte_flow_action *action,
3619 uint64_t action_flags,
3620 const struct rte_flow_attr *attributes,
3621 bool external, struct rte_flow_error *error)
3623 uint32_t target_group, table;
3626 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3627 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3628 return rte_flow_error_set(error, EINVAL,
3629 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3630 "can't have 2 fate actions in"
3632 if (action_flags & MLX5_FLOW_ACTION_METER)
3633 return rte_flow_error_set(error, ENOTSUP,
3634 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3635 "jump with meter not support");
3637 return rte_flow_error_set(error, EINVAL,
3638 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3639 NULL, "action configuration not set");
3641 ((const struct rte_flow_action_jump *)action->conf)->group;
3642 ret = mlx5_flow_group_to_table(attributes, external, target_group,
3643 true, &table, error);
3646 if (attributes->group == target_group)
3647 return rte_flow_error_set(error, EINVAL,
3648 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3649 "target group must be other than"
3650 " the current flow group");
3655 * Validate the port_id action.
3658 * Pointer to rte_eth_dev structure.
3659 * @param[in] action_flags
3660 * Bit-fields that holds the actions detected until now.
3662 * Port_id RTE action structure.
3664 * Attributes of flow that includes this action.
3666 * Pointer to error structure.
3669 * 0 on success, a negative errno value otherwise and rte_errno is set.
3672 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
3673 uint64_t action_flags,
3674 const struct rte_flow_action *action,
3675 const struct rte_flow_attr *attr,
3676 struct rte_flow_error *error)
3678 const struct rte_flow_action_port_id *port_id;
3679 struct mlx5_priv *act_priv;
3680 struct mlx5_priv *dev_priv;
3683 if (!attr->transfer)
3684 return rte_flow_error_set(error, ENOTSUP,
3685 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3687 "port id action is valid in transfer"
3689 if (!action || !action->conf)
3690 return rte_flow_error_set(error, ENOTSUP,
3691 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3693 "port id action parameters must be"
3695 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3696 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3697 return rte_flow_error_set(error, EINVAL,
3698 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3699 "can have only one fate actions in"
3701 dev_priv = mlx5_dev_to_eswitch_info(dev);
3703 return rte_flow_error_set(error, rte_errno,
3704 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3706 "failed to obtain E-Switch info");
3707 port_id = action->conf;
3708 port = port_id->original ? dev->data->port_id : port_id->id;
3709 act_priv = mlx5_port_to_eswitch_info(port, false);
3711 return rte_flow_error_set
3713 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
3714 "failed to obtain E-Switch port id for port");
3715 if (act_priv->domain_id != dev_priv->domain_id)
3716 return rte_flow_error_set
3718 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3719 "port does not belong to"
3720 " E-Switch being configured");
3725 * Get the maximum number of modify header actions.
3728 * Pointer to rte_eth_dev structure.
3730 * Flags bits to check if root level.
3733 * Max number of modify header actions device can support.
3735 static inline unsigned int
3736 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
3740 * There's no way to directly query the max capacity from FW.
3741 * The maximal value on root table should be assumed to be supported.
3743 if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
3744 return MLX5_MAX_MODIFY_NUM;
3746 return MLX5_ROOT_TBL_MODIFY_NUM;
3750 * Validate the meter action.
3753 * Pointer to rte_eth_dev structure.
3754 * @param[in] action_flags
3755 * Bit-fields that holds the actions detected until now.
3757 * Pointer to the meter action.
3759 * Attributes of flow that includes this action.
3761 * Pointer to error structure.
3764 * 0 on success, a negative errno value otherwise and rte_ernno is set.
3767 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
3768 uint64_t action_flags,
3769 const struct rte_flow_action *action,
3770 const struct rte_flow_attr *attr,
3771 struct rte_flow_error *error)
3773 struct mlx5_priv *priv = dev->data->dev_private;
3774 const struct rte_flow_action_meter *am = action->conf;
3775 struct mlx5_flow_meter *fm;
3778 return rte_flow_error_set(error, EINVAL,
3779 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3780 "meter action conf is NULL");
3782 if (action_flags & MLX5_FLOW_ACTION_METER)
3783 return rte_flow_error_set(error, ENOTSUP,
3784 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3785 "meter chaining not support");
3786 if (action_flags & MLX5_FLOW_ACTION_JUMP)
3787 return rte_flow_error_set(error, ENOTSUP,
3788 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3789 "meter with jump not support");
3791 return rte_flow_error_set(error, ENOTSUP,
3792 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3794 "meter action not supported");
3795 fm = mlx5_flow_meter_find(priv, am->mtr_id);
3797 return rte_flow_error_set(error, EINVAL,
3798 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3800 if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
3801 (!fm->ingress && !attr->ingress && attr->egress) ||
3802 (!fm->egress && !attr->egress && attr->ingress))))
3803 return rte_flow_error_set(error, EINVAL,
3804 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3805 "Flow attributes are either invalid "
3806 "or have a conflict with current "
3807 "meter attributes");
3812 * Validate the age action.
3814 * @param[in] action_flags
3815 * Holds the actions detected until now.
3817 * Pointer to the age action.
3819 * Pointer to the Ethernet device structure.
3821 * Pointer to error structure.
3824 * 0 on success, a negative errno value otherwise and rte_errno is set.
3827 flow_dv_validate_action_age(uint64_t action_flags,
3828 const struct rte_flow_action *action,
3829 struct rte_eth_dev *dev,
3830 struct rte_flow_error *error)
3832 struct mlx5_priv *priv = dev->data->dev_private;
3833 const struct rte_flow_action_age *age = action->conf;
3835 if (!priv->config.devx || priv->counter_fallback)
3836 return rte_flow_error_set(error, ENOTSUP,
3837 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3839 "age action not supported");
3840 if (!(action->conf))
3841 return rte_flow_error_set(error, EINVAL,
3842 RTE_FLOW_ERROR_TYPE_ACTION, action,
3843 "configuration cannot be null");
3844 if (age->timeout >= UINT16_MAX / 2 / 10)
3845 return rte_flow_error_set(error, ENOTSUP,
3846 RTE_FLOW_ERROR_TYPE_ACTION, action,
3847 "Max age time: 3275 seconds");
3848 if (action_flags & MLX5_FLOW_ACTION_AGE)
3849 return rte_flow_error_set(error, EINVAL,
3850 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3851 "Duplicate age ctions set");
3856 * Validate the modify-header IPv4 DSCP actions.
3858 * @param[in] action_flags
3859 * Holds the actions detected until now.
3861 * Pointer to the modify action.
3862 * @param[in] item_flags
3863 * Holds the items detected.
3865 * Pointer to error structure.
3868 * 0 on success, a negative errno value otherwise and rte_errno is set.
3871 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
3872 const struct rte_flow_action *action,
3873 const uint64_t item_flags,
3874 struct rte_flow_error *error)
3878 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3880 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
3881 return rte_flow_error_set(error, EINVAL,
3882 RTE_FLOW_ERROR_TYPE_ACTION,
3884 "no ipv4 item in pattern");
3890 * Validate the modify-header IPv6 DSCP actions.
3892 * @param[in] action_flags
3893 * Holds the actions detected until now.
3895 * Pointer to the modify action.
3896 * @param[in] item_flags
3897 * Holds the items detected.
3899 * Pointer to error structure.
3902 * 0 on success, a negative errno value otherwise and rte_errno is set.
3905 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
3906 const struct rte_flow_action *action,
3907 const uint64_t item_flags,
3908 struct rte_flow_error *error)
3912 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3914 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
3915 return rte_flow_error_set(error, EINVAL,
3916 RTE_FLOW_ERROR_TYPE_ACTION,
3918 "no ipv6 item in pattern");
3924 * Find existing modify-header resource or create and register a new one.
3926 * @param dev[in, out]
3927 * Pointer to rte_eth_dev structure.
3928 * @param[in, out] resource
3929 * Pointer to modify-header resource.
3930 * @parm[in, out] dev_flow
3931 * Pointer to the dev_flow.
3933 * pointer to error structure.
3936 * 0 on success otherwise -errno and errno is set.
3939 flow_dv_modify_hdr_resource_register
3940 (struct rte_eth_dev *dev,
3941 struct mlx5_flow_dv_modify_hdr_resource *resource,
3942 struct mlx5_flow *dev_flow,
3943 struct rte_flow_error *error)
3945 struct mlx5_priv *priv = dev->data->dev_private;
3946 struct mlx5_ibv_shared *sh = priv->sh;
3947 struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
3948 struct mlx5dv_dr_domain *ns;
3949 uint32_t actions_len;
3951 resource->flags = dev_flow->dv.group ? 0 :
3952 MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
3953 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
3955 return rte_flow_error_set(error, EOVERFLOW,
3956 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3957 "too many modify header items");
3958 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3959 ns = sh->fdb_domain;
3960 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
3964 /* Lookup a matching resource from cache. */
3965 actions_len = resource->actions_num * sizeof(resource->actions[0]);
3966 LIST_FOREACH(cache_resource, &sh->modify_cmds, next) {
3967 if (resource->ft_type == cache_resource->ft_type &&
3968 resource->actions_num == cache_resource->actions_num &&
3969 resource->flags == cache_resource->flags &&
3970 !memcmp((const void *)resource->actions,
3971 (const void *)cache_resource->actions,
3973 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
3974 (void *)cache_resource,
3975 rte_atomic32_read(&cache_resource->refcnt));
3976 rte_atomic32_inc(&cache_resource->refcnt);
3977 dev_flow->handle->dvh.modify_hdr = cache_resource;
3981 /* Register new modify-header resource. */
3982 cache_resource = rte_calloc(__func__, 1,
3983 sizeof(*cache_resource) + actions_len, 0);
3984 if (!cache_resource)
3985 return rte_flow_error_set(error, ENOMEM,
3986 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3987 "cannot allocate resource memory");
3988 *cache_resource = *resource;
3989 rte_memcpy(cache_resource->actions, resource->actions, actions_len);
3990 cache_resource->verbs_action =
3991 mlx5_glue->dv_create_flow_action_modify_header
3992 (sh->ctx, cache_resource->ft_type, ns,
3993 cache_resource->flags, actions_len,
3994 (uint64_t *)cache_resource->actions);
3995 if (!cache_resource->verbs_action) {
3996 rte_free(cache_resource);
3997 return rte_flow_error_set(error, ENOMEM,
3998 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3999 NULL, "cannot create action");
4001 rte_atomic32_init(&cache_resource->refcnt);
4002 rte_atomic32_inc(&cache_resource->refcnt);
4003 LIST_INSERT_HEAD(&sh->modify_cmds, cache_resource, next);
4004 dev_flow->handle->dvh.modify_hdr = cache_resource;
4005 DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
4006 (void *)cache_resource,
4007 rte_atomic32_read(&cache_resource->refcnt));
4012 * Get DV flow counter by index.
4015 * Pointer to the Ethernet device structure.
4017 * mlx5 flow counter index in the container.
4019 * mlx5 flow counter pool in the container,
4022 * Pointer to the counter, NULL otherwise.
4024 static struct mlx5_flow_counter *
4025 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
4027 struct mlx5_flow_counter_pool **ppool)
4029 struct mlx5_priv *priv = dev->data->dev_private;
4030 struct mlx5_pools_container *cont;
4031 struct mlx5_flow_counter_pool *pool;
4032 uint32_t batch = 0, age = 0;
4035 age = MLX_CNT_IS_AGE(idx);
4036 idx = age ? idx - MLX5_CNT_AGE_OFFSET : idx;
4037 if (idx >= MLX5_CNT_BATCH_OFFSET) {
4038 idx -= MLX5_CNT_BATCH_OFFSET;
4041 cont = MLX5_CNT_CONTAINER(priv->sh, batch, age);
4042 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cont->n);
4043 pool = cont->pools[idx / MLX5_COUNTERS_PER_POOL];
4047 return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
4051 * Get a pool by devx counter ID.
4054 * Pointer to the counter container.
4056 * The counter devx ID.
4059 * The counter pool pointer if exists, NULL otherwise,
4061 static struct mlx5_flow_counter_pool *
4062 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
4065 uint32_t n_valid = rte_atomic16_read(&cont->n_valid);
4067 for (i = 0; i < n_valid; i++) {
4068 struct mlx5_flow_counter_pool *pool = cont->pools[i];
4069 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
4070 MLX5_COUNTERS_PER_POOL;
4072 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL) {
4074 * Move the pool to the head, as counter allocate
4075 * always gets the first pool in the container.
4077 if (pool != TAILQ_FIRST(&cont->pool_list)) {
4078 TAILQ_REMOVE(&cont->pool_list, pool, next);
4079 TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
4088 * Allocate a new memory for the counter values wrapped by all the needed
4092 * Pointer to the Ethernet device structure.
4094 * The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
4097 * The new memory management pointer on success, otherwise NULL and rte_errno
4100 static struct mlx5_counter_stats_mem_mng *
4101 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
4103 struct mlx5_priv *priv = dev->data->dev_private;
4104 struct mlx5_ibv_shared *sh = priv->sh;
4105 struct mlx5_devx_mkey_attr mkey_attr;
4106 struct mlx5_counter_stats_mem_mng *mem_mng;
4107 volatile struct flow_counter_stats *raw_data;
4108 int size = (sizeof(struct flow_counter_stats) *
4109 MLX5_COUNTERS_PER_POOL +
4110 sizeof(struct mlx5_counter_stats_raw)) * raws_n +
4111 sizeof(struct mlx5_counter_stats_mem_mng);
4112 uint8_t *mem = rte_calloc(__func__, 1, size, sysconf(_SC_PAGESIZE));
4119 mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
4120 size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
4121 mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
4122 IBV_ACCESS_LOCAL_WRITE);
4123 if (!mem_mng->umem) {
4128 mkey_attr.addr = (uintptr_t)mem;
4129 mkey_attr.size = size;
4130 mkey_attr.umem_id = mem_mng->umem->umem_id;
4131 mkey_attr.pd = sh->pdn;
4132 mkey_attr.log_entity_size = 0;
4133 mkey_attr.pg_access = 0;
4134 mkey_attr.klm_array = NULL;
4135 mkey_attr.klm_num = 0;
4136 if (priv->config.hca_attr.relaxed_ordering_write &&
4137 priv->config.hca_attr.relaxed_ordering_read &&
4138 !haswell_broadwell_cpu)
4139 mkey_attr.relaxed_ordering = 1;
4140 mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
4142 mlx5_glue->devx_umem_dereg(mem_mng->umem);
4147 mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
4148 raw_data = (volatile struct flow_counter_stats *)mem;
4149 for (i = 0; i < raws_n; ++i) {
4150 mem_mng->raws[i].mem_mng = mem_mng;
4151 mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
4153 LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
4158 * Resize a counter container.
4161 * Pointer to the Ethernet device structure.
4163 * Whether the pool is for counter that was allocated by batch command.
4165 * Whether the pool is for Aging counter.
4168 * 0 on success, otherwise negative errno value and rte_errno is set.
4171 flow_dv_container_resize(struct rte_eth_dev *dev,
4172 uint32_t batch, uint32_t age)
4174 struct mlx5_priv *priv = dev->data->dev_private;
4175 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4177 struct mlx5_counter_stats_mem_mng *mem_mng = NULL;
4178 void *old_pools = cont->pools;
4179 uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
4180 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4181 void *pools = rte_calloc(__func__, 1, mem_size, 0);
4188 memcpy(pools, old_pools, cont->n *
4189 sizeof(struct mlx5_flow_counter_pool *));
4191 * Fallback mode query the counter directly, no background query
4192 * resources are needed.
4194 if (!priv->counter_fallback) {
4197 mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
4198 MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES);
4203 for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
4204 LIST_INSERT_HEAD(&priv->sh->cmng.free_stat_raws,
4206 MLX5_CNT_CONTAINER_RESIZE +
4209 rte_spinlock_lock(&cont->resize_sl);
4211 cont->mem_mng = mem_mng;
4212 cont->pools = pools;
4213 rte_spinlock_unlock(&cont->resize_sl);
4215 rte_free(old_pools);
4220 * Query a devx flow counter.
4223 * Pointer to the Ethernet device structure.
4225 * Index to the flow counter.
4227 * The statistics value of packets.
4229 * The statistics value of bytes.
4232 * 0 on success, otherwise a negative errno value and rte_errno is set.
4235 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4238 struct mlx5_priv *priv = dev->data->dev_private;
4239 struct mlx5_flow_counter_pool *pool = NULL;
4240 struct mlx5_flow_counter *cnt;
4241 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4244 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4246 if (counter < MLX5_CNT_BATCH_OFFSET) {
4247 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4248 if (priv->counter_fallback)
4249 return mlx5_devx_cmd_flow_counter_query(cnt_ext->dcs, 0,
4250 0, pkts, bytes, 0, NULL, NULL, 0);
4253 rte_spinlock_lock(&pool->sl);
4255 * The single counters allocation may allocate smaller ID than the
4256 * current allocated in parallel to the host reading.
4257 * In this case the new counter values must be reported as 0.
4259 if (unlikely(cnt_ext && cnt_ext->dcs->id < pool->raw->min_dcs_id)) {
4263 offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
4264 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4265 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4267 rte_spinlock_unlock(&pool->sl);
4272 * Create and initialize a new counter pool.
4275 * Pointer to the Ethernet device structure.
4277 * The devX counter handle.
4279 * Whether the pool is for counter that was allocated by batch command.
4281 * Whether the pool is for counter that was allocated for aging.
4282 * @param[in/out] cont_cur
4283 * Pointer to the container pointer, it will be update in pool resize.
4286 * The pool container pointer on success, NULL otherwise and rte_errno is set.
4288 static struct mlx5_flow_counter_pool *
4289 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4290 uint32_t batch, uint32_t age)
4292 struct mlx5_priv *priv = dev->data->dev_private;
4293 struct mlx5_flow_counter_pool *pool;
4294 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4296 int16_t n_valid = rte_atomic16_read(&cont->n_valid);
4297 uint32_t size = sizeof(*pool);
4299 if (cont->n == n_valid && flow_dv_container_resize(dev, batch, age))
4301 size += MLX5_COUNTERS_PER_POOL * CNT_SIZE;
4302 size += (batch ? 0 : MLX5_COUNTERS_PER_POOL * CNTEXT_SIZE);
4303 size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * AGE_SIZE);
4304 pool = rte_calloc(__func__, 1, size, 0);
4309 pool->min_dcs = dcs;
4310 if (!priv->counter_fallback)
4311 pool->raw = cont->mem_mng->raws + n_valid %
4312 MLX5_CNT_CONTAINER_RESIZE;
4313 pool->raw_hw = NULL;
4315 pool->type |= (batch ? 0 : CNT_POOL_TYPE_EXT);
4316 pool->type |= (!age ? 0 : CNT_POOL_TYPE_AGE);
4317 rte_spinlock_init(&pool->sl);
4319 * The generation of the new allocated counters in this pool is 0, 2 in
4320 * the pool generation makes all the counters valid for allocation.
4321 * The start and end query generation protect the counters be released
4322 * between the query and update gap period will not be reallocated
4323 * without the last query finished and stats updated to the memory.
4325 rte_atomic64_set(&pool->start_query_gen, 0x2);
4327 * There's no background query thread for fallback mode, set the
4328 * end_query_gen to the maximum value since no need to wait for
4329 * statistics update.
4331 rte_atomic64_set(&pool->end_query_gen, priv->counter_fallback ?
4333 TAILQ_INIT(&pool->counters);
4334 TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
4335 pool->index = n_valid;
4336 cont->pools[n_valid] = pool;
4337 /* Pool initialization must be updated before host thread access. */
4339 rte_atomic16_add(&cont->n_valid, 1);
4344 * Update the minimum dcs-id for aged or no-aged counter pool.
4347 * Pointer to the Ethernet device structure.
4349 * Current counter pool.
4351 * Whether the pool is for counter that was allocated by batch command.
4353 * Whether the counter is for aging.
4356 flow_dv_counter_update_min_dcs(struct rte_eth_dev *dev,
4357 struct mlx5_flow_counter_pool *pool,
4358 uint32_t batch, uint32_t age)
4360 struct mlx5_priv *priv = dev->data->dev_private;
4361 struct mlx5_flow_counter_pool *other;
4362 struct mlx5_pools_container *cont;
4364 cont = MLX5_CNT_CONTAINER(priv->sh, batch, (age ^ 0x1));
4365 other = flow_dv_find_pool_by_id(cont, pool->min_dcs->id);
4368 if (pool->min_dcs->id < other->min_dcs->id) {
4369 rte_atomic64_set(&other->a64_dcs,
4370 rte_atomic64_read(&pool->a64_dcs));
4372 rte_atomic64_set(&pool->a64_dcs,
4373 rte_atomic64_read(&other->a64_dcs));
4377 * Prepare a new counter and/or a new counter pool.
4380 * Pointer to the Ethernet device structure.
4381 * @param[out] cnt_free
4382 * Where to put the pointer of a new counter.
4384 * Whether the pool is for counter that was allocated by batch command.
4386 * Whether the pool is for counter that was allocated for aging.
4389 * The counter pool pointer and @p cnt_free is set on success,
4390 * NULL otherwise and rte_errno is set.
4392 static struct mlx5_flow_counter_pool *
4393 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4394 struct mlx5_flow_counter **cnt_free,
4395 uint32_t batch, uint32_t age)
4397 struct mlx5_priv *priv = dev->data->dev_private;
4398 struct mlx5_pools_container *cont;
4399 struct mlx5_flow_counter_pool *pool;
4400 struct mlx5_devx_obj *dcs = NULL;
4401 struct mlx5_flow_counter *cnt;
4404 cont = MLX5_CNT_CONTAINER(priv->sh, batch, age);
4406 /* bulk_bitmap must be 0 for single counter allocation. */
4407 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4410 pool = flow_dv_find_pool_by_id(cont, dcs->id);
4412 pool = flow_dv_pool_create(dev, dcs, batch, age);
4414 mlx5_devx_cmd_destroy(dcs);
4417 } else if (dcs->id < pool->min_dcs->id) {
4418 rte_atomic64_set(&pool->a64_dcs,
4419 (int64_t)(uintptr_t)dcs);
4421 flow_dv_counter_update_min_dcs(dev,
4423 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4424 cnt = MLX5_POOL_GET_CNT(pool, i);
4425 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
4426 MLX5_GET_POOL_CNT_EXT(pool, i)->dcs = dcs;
4430 /* bulk_bitmap is in 128 counters units. */
4431 if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
4432 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4434 rte_errno = ENODATA;
4437 pool = flow_dv_pool_create(dev, dcs, batch, age);
4439 mlx5_devx_cmd_destroy(dcs);
4442 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
4443 cnt = MLX5_POOL_GET_CNT(pool, i);
4444 TAILQ_INSERT_HEAD(&pool->counters, cnt, next);
4446 *cnt_free = MLX5_POOL_GET_CNT(pool, 0);
4451 * Search for existed shared counter.
4454 * Pointer to the relevant counter pool container.
4456 * The shared counter ID to search.
4458 * mlx5 flow counter pool in the container,
4461 * NULL if not existed, otherwise pointer to the shared extend counter.
4463 static struct mlx5_flow_counter_ext *
4464 flow_dv_counter_shared_search(struct mlx5_pools_container *cont, uint32_t id,
4465 struct mlx5_flow_counter_pool **ppool)
4467 static struct mlx5_flow_counter_ext *cnt;
4468 struct mlx5_flow_counter_pool *pool;
4470 uint32_t n_valid = rte_atomic16_read(&cont->n_valid);
4472 for (i = 0; i < n_valid; i++) {
4473 pool = cont->pools[i];
4474 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
4475 cnt = MLX5_GET_POOL_CNT_EXT(pool, i);
4476 if (cnt->ref_cnt && cnt->shared && cnt->id == id) {
4478 *ppool = cont->pools[i];
4487 * Allocate a flow counter.
4490 * Pointer to the Ethernet device structure.
4492 * Indicate if this counter is shared with other flows.
4494 * Counter identifier.
4496 * Counter flow group.
4498 * Whether the counter was allocated for aging.
4501 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4504 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
4505 uint16_t group, uint32_t age)
4507 struct mlx5_priv *priv = dev->data->dev_private;
4508 struct mlx5_flow_counter_pool *pool = NULL;
4509 struct mlx5_flow_counter *cnt_free = NULL;
4510 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4512 * Currently group 0 flow counter cannot be assigned to a flow if it is
4513 * not the first one in the batch counter allocation, so it is better
4514 * to allocate counters one by one for these flows in a separate
4516 * A counter can be shared between different groups so need to take
4517 * shared counters from the single container.
4519 uint32_t batch = (group && !shared && !priv->counter_fallback) ? 1 : 0;
4520 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4524 if (!priv->config.devx) {
4525 rte_errno = ENOTSUP;
4529 cnt_ext = flow_dv_counter_shared_search(cont, id, &pool);
4531 if (cnt_ext->ref_cnt + 1 == 0) {
4536 cnt_idx = pool->index * MLX5_COUNTERS_PER_POOL +
4537 (cnt_ext->dcs->id % MLX5_COUNTERS_PER_POOL)
4542 /* Pools which has a free counters are in the start. */
4543 TAILQ_FOREACH(pool, &cont->pool_list, next) {
4545 * The free counter reset values must be updated between the
4546 * counter release to the counter allocation, so, at least one
4547 * query must be done in this time. ensure it by saving the
4548 * query generation in the release time.
4549 * The free list is sorted according to the generation - so if
4550 * the first one is not updated, all the others are not
4553 cnt_free = TAILQ_FIRST(&pool->counters);
4554 if (cnt_free && cnt_free->query_gen <
4555 rte_atomic64_read(&pool->end_query_gen))
4560 pool = flow_dv_counter_pool_prepare(dev, &cnt_free, batch, age);
4565 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt_free);
4566 /* Create a DV counter action only in the first time usage. */
4567 if (!cnt_free->action) {
4569 struct mlx5_devx_obj *dcs;
4572 offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
4573 dcs = pool->min_dcs;
4578 cnt_free->action = mlx5_glue->dv_create_flow_action_counter
4580 if (!cnt_free->action) {
4585 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4586 MLX5_CNT_ARRAY_IDX(pool, cnt_free));
4587 cnt_idx += batch * MLX5_CNT_BATCH_OFFSET;
4588 cnt_idx += age * MLX5_CNT_AGE_OFFSET;
4589 /* Update the counter reset values. */
4590 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4594 cnt_ext->shared = shared;
4595 cnt_ext->ref_cnt = 1;
4598 if (!priv->counter_fallback && !priv->sh->cmng.query_thread_on)
4599 /* Start the asynchronous batch query by the host thread. */
4600 mlx5_set_query_alarm(priv->sh);
4601 TAILQ_REMOVE(&pool->counters, cnt_free, next);
4602 if (TAILQ_EMPTY(&pool->counters)) {
4603 /* Move the pool to the end of the container pool list. */
4604 TAILQ_REMOVE(&cont->pool_list, pool, next);
4605 TAILQ_INSERT_TAIL(&cont->pool_list, pool, next);
4611 * Get age param from counter index.
4614 * Pointer to the Ethernet device structure.
4615 * @param[in] counter
4616 * Index to the counter handler.
4619 * The aging parameter specified for the counter index.
4621 static struct mlx5_age_param*
4622 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
4625 struct mlx5_flow_counter *cnt;
4626 struct mlx5_flow_counter_pool *pool = NULL;
4628 flow_dv_counter_get_by_idx(dev, counter, &pool);
4629 counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
4630 cnt = MLX5_POOL_GET_CNT(pool, counter);
4631 return MLX5_CNT_TO_AGE(cnt);
4635 * Remove a flow counter from aged counter list.
4638 * Pointer to the Ethernet device structure.
4639 * @param[in] counter
4640 * Index to the counter handler.
4642 * Pointer to the counter handler.
4645 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
4646 uint32_t counter, struct mlx5_flow_counter *cnt)
4648 struct mlx5_age_info *age_info;
4649 struct mlx5_age_param *age_param;
4650 struct mlx5_priv *priv = dev->data->dev_private;
4652 age_info = GET_PORT_AGE_INFO(priv);
4653 age_param = flow_dv_counter_idx_get_age(dev, counter);
4654 if (rte_atomic16_cmpset((volatile uint16_t *)
4656 AGE_CANDIDATE, AGE_FREE)
4659 * We need the lock even it is age timeout,
4660 * since counter may still in process.
4662 rte_spinlock_lock(&age_info->aged_sl);
4663 TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
4664 rte_spinlock_unlock(&age_info->aged_sl);
4666 rte_atomic16_set(&age_param->state, AGE_FREE);
4669 * Release a flow counter.
4672 * Pointer to the Ethernet device structure.
4673 * @param[in] counter
4674 * Index to the counter handler.
4677 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
4679 struct mlx5_flow_counter_pool *pool = NULL;
4680 struct mlx5_flow_counter *cnt;
4681 struct mlx5_flow_counter_ext *cnt_ext = NULL;
4685 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4687 if (counter < MLX5_CNT_BATCH_OFFSET) {
4688 cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4689 if (cnt_ext && --cnt_ext->ref_cnt)
4692 if (IS_AGE_POOL(pool))
4693 flow_dv_counter_remove_from_age(dev, counter, cnt);
4694 /* Put the counter in the end - the last updated one. */
4695 TAILQ_INSERT_TAIL(&pool->counters, cnt, next);
4697 * Counters released between query trigger and handler need
4698 * to wait the next round of query. Since the packets arrive
4699 * in the gap period will not be taken into account to the
4702 cnt->query_gen = rte_atomic64_read(&pool->start_query_gen);
4706 * Verify the @p attributes will be correctly understood by the NIC and store
4707 * them in the @p flow if everything is correct.
4710 * Pointer to dev struct.
4711 * @param[in] attributes
4712 * Pointer to flow attributes
4713 * @param[in] external
4714 * This flow rule is created by request external to PMD.
4716 * Pointer to error structure.
4719 * - 0 on success and non root table.
4720 * - 1 on success and root table.
4721 * - a negative errno value otherwise and rte_errno is set.
4724 flow_dv_validate_attributes(struct rte_eth_dev *dev,
4725 const struct rte_flow_attr *attributes,
4726 bool external __rte_unused,
4727 struct rte_flow_error *error)
4729 struct mlx5_priv *priv = dev->data->dev_private;
4730 uint32_t priority_max = priv->config.flow_prio - 1;
4733 #ifndef HAVE_MLX5DV_DR
4734 if (attributes->group)
4735 return rte_flow_error_set(error, ENOTSUP,
4736 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
4738 "groups are not supported");
4742 ret = mlx5_flow_group_to_table(attributes, external,
4743 attributes->group, !!priv->fdb_def_rule,
4748 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4750 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
4751 attributes->priority >= priority_max)
4752 return rte_flow_error_set(error, ENOTSUP,
4753 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
4755 "priority out of range");
4756 if (attributes->transfer) {
4757 if (!priv->config.dv_esw_en)
4758 return rte_flow_error_set
4760 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4761 "E-Switch dr is not supported");
4762 if (!(priv->representor || priv->master))
4763 return rte_flow_error_set
4764 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4765 NULL, "E-Switch configuration can only be"
4766 " done by a master or a representor device");
4767 if (attributes->egress)
4768 return rte_flow_error_set
4770 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
4771 "egress is not supported");
4773 if (!(attributes->egress ^ attributes->ingress))
4774 return rte_flow_error_set(error, ENOTSUP,
4775 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
4776 "must specify exactly one of "
4777 "ingress or egress");
4782 * Internal validation function. For validating both actions and items.
4785 * Pointer to the rte_eth_dev structure.
4787 * Pointer to the flow attributes.
4789 * Pointer to the list of items.
4790 * @param[in] actions
4791 * Pointer to the list of actions.
4792 * @param[in] external
4793 * This flow rule is created by request external to PMD.
4794 * @param[in] hairpin
4795 * Number of hairpin TX actions, 0 means classic flow.
4797 * Pointer to the error structure.
4800 * 0 on success, a negative errno value otherwise and rte_errno is set.
4803 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
4804 const struct rte_flow_item items[],
4805 const struct rte_flow_action actions[],
4806 bool external, int hairpin, struct rte_flow_error *error)
4809 uint64_t action_flags = 0;
4810 uint64_t item_flags = 0;
4811 uint64_t last_item = 0;
4812 uint8_t next_protocol = 0xff;
4813 uint16_t ether_type = 0;
4815 uint8_t item_ipv6_proto = 0;
4816 const struct rte_flow_item *gre_item = NULL;
4817 const struct rte_flow_action_raw_decap *decap;
4818 const struct rte_flow_action_raw_encap *encap;
4819 const struct rte_flow_action_rss *rss;
4820 const struct rte_flow_item_tcp nic_tcp_mask = {
4823 .src_port = RTE_BE16(UINT16_MAX),
4824 .dst_port = RTE_BE16(UINT16_MAX),
4827 const struct rte_flow_item_ipv4 nic_ipv4_mask = {
4829 .src_addr = RTE_BE32(0xffffffff),
4830 .dst_addr = RTE_BE32(0xffffffff),
4831 .type_of_service = 0xff,
4832 .next_proto_id = 0xff,
4833 .time_to_live = 0xff,
4836 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
4839 "\xff\xff\xff\xff\xff\xff\xff\xff"
4840 "\xff\xff\xff\xff\xff\xff\xff\xff",
4842 "\xff\xff\xff\xff\xff\xff\xff\xff"
4843 "\xff\xff\xff\xff\xff\xff\xff\xff",
4844 .vtc_flow = RTE_BE32(0xffffffff),
4849 struct mlx5_priv *priv = dev->data->dev_private;
4850 struct mlx5_dev_config *dev_conf = &priv->config;
4851 uint16_t queue_index = 0xFFFF;
4852 const struct rte_flow_item_vlan *vlan_m = NULL;
4853 int16_t rw_act_num = 0;
4858 ret = flow_dv_validate_attributes(dev, attr, external, error);
4861 is_root = (uint64_t)ret;
4862 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4863 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
4864 int type = items->type;
4867 case RTE_FLOW_ITEM_TYPE_VOID:
4869 case RTE_FLOW_ITEM_TYPE_PORT_ID:
4870 ret = flow_dv_validate_item_port_id
4871 (dev, items, attr, item_flags, error);
4874 last_item = MLX5_FLOW_ITEM_PORT_ID;
4876 case RTE_FLOW_ITEM_TYPE_ETH:
4877 ret = mlx5_flow_validate_item_eth(items, item_flags,
4881 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
4882 MLX5_FLOW_LAYER_OUTER_L2;
4883 if (items->mask != NULL && items->spec != NULL) {
4885 ((const struct rte_flow_item_eth *)
4888 ((const struct rte_flow_item_eth *)
4890 ether_type = rte_be_to_cpu_16(ether_type);
4895 case RTE_FLOW_ITEM_TYPE_VLAN:
4896 ret = flow_dv_validate_item_vlan(items, item_flags,
4900 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
4901 MLX5_FLOW_LAYER_OUTER_VLAN;
4902 if (items->mask != NULL && items->spec != NULL) {
4904 ((const struct rte_flow_item_vlan *)
4905 items->spec)->inner_type;
4907 ((const struct rte_flow_item_vlan *)
4908 items->mask)->inner_type;
4909 ether_type = rte_be_to_cpu_16(ether_type);
4913 /* Store outer VLAN mask for of_push_vlan action. */
4915 vlan_m = items->mask;
4917 case RTE_FLOW_ITEM_TYPE_IPV4:
4918 mlx5_flow_tunnel_ip_check(items, next_protocol,
4919 &item_flags, &tunnel);
4920 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
4927 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
4928 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4929 if (items->mask != NULL &&
4930 ((const struct rte_flow_item_ipv4 *)
4931 items->mask)->hdr.next_proto_id) {
4933 ((const struct rte_flow_item_ipv4 *)
4934 (items->spec))->hdr.next_proto_id;
4936 ((const struct rte_flow_item_ipv4 *)
4937 (items->mask))->hdr.next_proto_id;
4939 /* Reset for inner layer. */
4940 next_protocol = 0xff;
4943 case RTE_FLOW_ITEM_TYPE_IPV6:
4944 mlx5_flow_tunnel_ip_check(items, next_protocol,
4945 &item_flags, &tunnel);
4946 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
4953 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
4954 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4955 if (items->mask != NULL &&
4956 ((const struct rte_flow_item_ipv6 *)
4957 items->mask)->hdr.proto) {
4959 ((const struct rte_flow_item_ipv6 *)
4960 items->spec)->hdr.proto;
4962 ((const struct rte_flow_item_ipv6 *)
4963 items->spec)->hdr.proto;
4965 ((const struct rte_flow_item_ipv6 *)
4966 items->mask)->hdr.proto;
4968 /* Reset for inner layer. */
4969 next_protocol = 0xff;
4972 case RTE_FLOW_ITEM_TYPE_TCP:
4973 ret = mlx5_flow_validate_item_tcp
4980 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
4981 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4983 case RTE_FLOW_ITEM_TYPE_UDP:
4984 ret = mlx5_flow_validate_item_udp(items, item_flags,
4989 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
4990 MLX5_FLOW_LAYER_OUTER_L4_UDP;
4992 case RTE_FLOW_ITEM_TYPE_GRE:
4993 ret = mlx5_flow_validate_item_gre(items, item_flags,
4994 next_protocol, error);
4998 last_item = MLX5_FLOW_LAYER_GRE;
5000 case RTE_FLOW_ITEM_TYPE_NVGRE:
5001 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
5006 last_item = MLX5_FLOW_LAYER_NVGRE;
5008 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5009 ret = mlx5_flow_validate_item_gre_key
5010 (items, item_flags, gre_item, error);
5013 last_item = MLX5_FLOW_LAYER_GRE_KEY;
5015 case RTE_FLOW_ITEM_TYPE_VXLAN:
5016 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
5020 last_item = MLX5_FLOW_LAYER_VXLAN;
5022 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5023 ret = mlx5_flow_validate_item_vxlan_gpe(items,
5028 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5030 case RTE_FLOW_ITEM_TYPE_GENEVE:
5031 ret = mlx5_flow_validate_item_geneve(items,
5036 last_item = MLX5_FLOW_LAYER_GENEVE;
5038 case RTE_FLOW_ITEM_TYPE_MPLS:
5039 ret = mlx5_flow_validate_item_mpls(dev, items,
5044 last_item = MLX5_FLOW_LAYER_MPLS;
5047 case RTE_FLOW_ITEM_TYPE_MARK:
5048 ret = flow_dv_validate_item_mark(dev, items, attr,
5052 last_item = MLX5_FLOW_ITEM_MARK;
5054 case RTE_FLOW_ITEM_TYPE_META:
5055 ret = flow_dv_validate_item_meta(dev, items, attr,
5059 last_item = MLX5_FLOW_ITEM_METADATA;
5061 case RTE_FLOW_ITEM_TYPE_ICMP:
5062 ret = mlx5_flow_validate_item_icmp(items, item_flags,
5067 last_item = MLX5_FLOW_LAYER_ICMP;
5069 case RTE_FLOW_ITEM_TYPE_ICMP6:
5070 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
5075 item_ipv6_proto = IPPROTO_ICMPV6;
5076 last_item = MLX5_FLOW_LAYER_ICMP6;
5078 case RTE_FLOW_ITEM_TYPE_TAG:
5079 ret = flow_dv_validate_item_tag(dev, items,
5083 last_item = MLX5_FLOW_ITEM_TAG;
5085 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
5086 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
5088 case RTE_FLOW_ITEM_TYPE_GTP:
5089 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
5093 last_item = MLX5_FLOW_LAYER_GTP;
5096 return rte_flow_error_set(error, ENOTSUP,
5097 RTE_FLOW_ERROR_TYPE_ITEM,
5098 NULL, "item not supported");
5100 item_flags |= last_item;
5102 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5103 int type = actions->type;
5104 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5105 return rte_flow_error_set(error, ENOTSUP,
5106 RTE_FLOW_ERROR_TYPE_ACTION,
5107 actions, "too many actions");
5109 case RTE_FLOW_ACTION_TYPE_VOID:
5111 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5112 ret = flow_dv_validate_action_port_id(dev,
5119 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5122 case RTE_FLOW_ACTION_TYPE_FLAG:
5123 ret = flow_dv_validate_action_flag(dev, action_flags,
5127 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5128 /* Count all modify-header actions as one. */
5129 if (!(action_flags &
5130 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5132 action_flags |= MLX5_FLOW_ACTION_FLAG |
5133 MLX5_FLOW_ACTION_MARK_EXT;
5135 action_flags |= MLX5_FLOW_ACTION_FLAG;
5138 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5140 case RTE_FLOW_ACTION_TYPE_MARK:
5141 ret = flow_dv_validate_action_mark(dev, actions,
5146 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5147 /* Count all modify-header actions as one. */
5148 if (!(action_flags &
5149 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5151 action_flags |= MLX5_FLOW_ACTION_MARK |
5152 MLX5_FLOW_ACTION_MARK_EXT;
5154 action_flags |= MLX5_FLOW_ACTION_MARK;
5157 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5159 case RTE_FLOW_ACTION_TYPE_SET_META:
5160 ret = flow_dv_validate_action_set_meta(dev, actions,
5165 /* Count all modify-header actions as one action. */
5166 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5168 action_flags |= MLX5_FLOW_ACTION_SET_META;
5169 rw_act_num += MLX5_ACT_NUM_SET_META;
5171 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5172 ret = flow_dv_validate_action_set_tag(dev, actions,
5177 /* Count all modify-header actions as one action. */
5178 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5180 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
5181 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5183 case RTE_FLOW_ACTION_TYPE_DROP:
5184 ret = mlx5_flow_validate_action_drop(action_flags,
5188 action_flags |= MLX5_FLOW_ACTION_DROP;
5191 case RTE_FLOW_ACTION_TYPE_QUEUE:
5192 ret = mlx5_flow_validate_action_queue(actions,
5197 queue_index = ((const struct rte_flow_action_queue *)
5198 (actions->conf))->index;
5199 action_flags |= MLX5_FLOW_ACTION_QUEUE;
5202 case RTE_FLOW_ACTION_TYPE_RSS:
5203 rss = actions->conf;
5204 ret = mlx5_flow_validate_action_rss(actions,
5210 if (rss != NULL && rss->queue_num)
5211 queue_index = rss->queue[0];
5212 action_flags |= MLX5_FLOW_ACTION_RSS;
5215 case RTE_FLOW_ACTION_TYPE_COUNT:
5216 ret = flow_dv_validate_action_count(dev, error);
5219 action_flags |= MLX5_FLOW_ACTION_COUNT;
5222 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5223 if (flow_dv_validate_action_pop_vlan(dev,
5229 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5232 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5233 ret = flow_dv_validate_action_push_vlan(dev,
5240 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5243 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5244 ret = flow_dv_validate_action_set_vlan_pcp
5245 (action_flags, actions, error);
5248 /* Count PCP with push_vlan command. */
5249 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5251 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5252 ret = flow_dv_validate_action_set_vlan_vid
5253 (item_flags, action_flags,
5257 /* Count VID with push_vlan command. */
5258 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5259 rw_act_num += MLX5_ACT_NUM_MDF_VID;
5261 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5262 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5263 ret = flow_dv_validate_action_l2_encap(dev,
5269 action_flags |= MLX5_FLOW_ACTION_ENCAP;
5272 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5273 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5274 ret = flow_dv_validate_action_decap(dev, action_flags,
5278 action_flags |= MLX5_FLOW_ACTION_DECAP;
5281 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5282 ret = flow_dv_validate_action_raw_encap_decap
5283 (dev, NULL, actions->conf, attr, &action_flags,
5288 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5289 decap = actions->conf;
5290 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5292 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5296 encap = actions->conf;
5298 ret = flow_dv_validate_action_raw_encap_decap
5300 decap ? decap : &empty_decap, encap,
5301 attr, &action_flags, &actions_n,
5306 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5307 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5308 ret = flow_dv_validate_action_modify_mac(action_flags,
5314 /* Count all modify-header actions as one action. */
5315 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5317 action_flags |= actions->type ==
5318 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5319 MLX5_FLOW_ACTION_SET_MAC_SRC :
5320 MLX5_FLOW_ACTION_SET_MAC_DST;
5322 * Even if the source and destination MAC addresses have
5323 * overlap in the header with 4B alignment, the convert
5324 * function will handle them separately and 4 SW actions
5325 * will be created. And 2 actions will be added each
5326 * time no matter how many bytes of address will be set.
5328 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5330 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5331 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5332 ret = flow_dv_validate_action_modify_ipv4(action_flags,
5338 /* Count all modify-header actions as one action. */
5339 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5341 action_flags |= actions->type ==
5342 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5343 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5344 MLX5_FLOW_ACTION_SET_IPV4_DST;
5345 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5347 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5348 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5349 ret = flow_dv_validate_action_modify_ipv6(action_flags,
5355 if (item_ipv6_proto == IPPROTO_ICMPV6)
5356 return rte_flow_error_set(error, ENOTSUP,
5357 RTE_FLOW_ERROR_TYPE_ACTION,
5359 "Can't change header "
5360 "with ICMPv6 proto");
5361 /* Count all modify-header actions as one action. */
5362 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5364 action_flags |= actions->type ==
5365 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5366 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5367 MLX5_FLOW_ACTION_SET_IPV6_DST;
5368 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5370 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5371 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5372 ret = flow_dv_validate_action_modify_tp(action_flags,
5378 /* Count all modify-header actions as one action. */
5379 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5381 action_flags |= actions->type ==
5382 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5383 MLX5_FLOW_ACTION_SET_TP_SRC :
5384 MLX5_FLOW_ACTION_SET_TP_DST;
5385 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5387 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5388 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5389 ret = flow_dv_validate_action_modify_ttl(action_flags,
5395 /* Count all modify-header actions as one action. */
5396 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5398 action_flags |= actions->type ==
5399 RTE_FLOW_ACTION_TYPE_SET_TTL ?
5400 MLX5_FLOW_ACTION_SET_TTL :
5401 MLX5_FLOW_ACTION_DEC_TTL;
5402 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5404 case RTE_FLOW_ACTION_TYPE_JUMP:
5405 ret = flow_dv_validate_action_jump(actions,
5412 action_flags |= MLX5_FLOW_ACTION_JUMP;
5414 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5415 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5416 ret = flow_dv_validate_action_modify_tcp_seq
5423 /* Count all modify-header actions as one action. */
5424 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5426 action_flags |= actions->type ==
5427 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5428 MLX5_FLOW_ACTION_INC_TCP_SEQ :
5429 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5430 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5432 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5433 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5434 ret = flow_dv_validate_action_modify_tcp_ack
5441 /* Count all modify-header actions as one action. */
5442 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5444 action_flags |= actions->type ==
5445 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5446 MLX5_FLOW_ACTION_INC_TCP_ACK :
5447 MLX5_FLOW_ACTION_DEC_TCP_ACK;
5448 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5450 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5452 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5453 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5454 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5456 case RTE_FLOW_ACTION_TYPE_METER:
5457 ret = mlx5_flow_validate_action_meter(dev,
5463 action_flags |= MLX5_FLOW_ACTION_METER;
5465 /* Meter action will add one more TAG action. */
5466 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5468 case RTE_FLOW_ACTION_TYPE_AGE:
5469 ret = flow_dv_validate_action_age(action_flags,
5474 action_flags |= MLX5_FLOW_ACTION_AGE;
5477 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5478 ret = flow_dv_validate_action_modify_ipv4_dscp
5485 /* Count all modify-header actions as one action. */
5486 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5488 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5489 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5491 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5492 ret = flow_dv_validate_action_modify_ipv6_dscp
5499 /* Count all modify-header actions as one action. */
5500 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5502 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5503 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5506 return rte_flow_error_set(error, ENOTSUP,
5507 RTE_FLOW_ERROR_TYPE_ACTION,
5509 "action not supported");
5513 * Validate the drop action mutual exclusion with other actions.
5514 * Drop action is mutually-exclusive with any other action, except for
5517 if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
5518 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
5519 return rte_flow_error_set(error, EINVAL,
5520 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5521 "Drop action is mutually-exclusive "
5522 "with any other action, except for "
5524 /* Eswitch has few restrictions on using items and actions */
5525 if (attr->transfer) {
5526 if (!mlx5_flow_ext_mreg_supported(dev) &&
5527 action_flags & MLX5_FLOW_ACTION_FLAG)
5528 return rte_flow_error_set(error, ENOTSUP,
5529 RTE_FLOW_ERROR_TYPE_ACTION,
5531 "unsupported action FLAG");
5532 if (!mlx5_flow_ext_mreg_supported(dev) &&
5533 action_flags & MLX5_FLOW_ACTION_MARK)
5534 return rte_flow_error_set(error, ENOTSUP,
5535 RTE_FLOW_ERROR_TYPE_ACTION,
5537 "unsupported action MARK");
5538 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
5539 return rte_flow_error_set(error, ENOTSUP,
5540 RTE_FLOW_ERROR_TYPE_ACTION,
5542 "unsupported action QUEUE");
5543 if (action_flags & MLX5_FLOW_ACTION_RSS)
5544 return rte_flow_error_set(error, ENOTSUP,
5545 RTE_FLOW_ERROR_TYPE_ACTION,
5547 "unsupported action RSS");
5548 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
5549 return rte_flow_error_set(error, EINVAL,
5550 RTE_FLOW_ERROR_TYPE_ACTION,
5552 "no fate action is found");
5554 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
5555 return rte_flow_error_set(error, EINVAL,
5556 RTE_FLOW_ERROR_TYPE_ACTION,
5558 "no fate action is found");
5560 /* Continue validation for Xcap actions.*/
5561 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) && (queue_index == 0xFFFF ||
5562 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
5563 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
5564 MLX5_FLOW_XCAP_ACTIONS)
5565 return rte_flow_error_set(error, ENOTSUP,
5566 RTE_FLOW_ERROR_TYPE_ACTION,
5567 NULL, "encap and decap "
5568 "combination aren't supported");
5569 if (!attr->transfer && attr->ingress && (action_flags &
5570 MLX5_FLOW_ACTION_ENCAP))
5571 return rte_flow_error_set(error, ENOTSUP,
5572 RTE_FLOW_ERROR_TYPE_ACTION,
5573 NULL, "encap is not supported"
5574 " for ingress traffic");
5576 /* Hairpin flow will add one more TAG action. */
5578 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5579 /* extra metadata enabled: one more TAG action will be add. */
5580 if (dev_conf->dv_flow_en &&
5581 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
5582 mlx5_flow_ext_mreg_supported(dev))
5583 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5584 if ((uint32_t)rw_act_num >
5585 flow_dv_modify_hdr_action_max(dev, is_root)) {
5586 return rte_flow_error_set(error, ENOTSUP,
5587 RTE_FLOW_ERROR_TYPE_ACTION,
5588 NULL, "too many header modify"
5589 " actions to support");
5595 * Internal preparation function. Allocates the DV flow size,
5596 * this size is constant.
5599 * Pointer to the rte_eth_dev structure.
5601 * Pointer to the flow attributes.
5603 * Pointer to the list of items.
5604 * @param[in] actions
5605 * Pointer to the list of actions.
5607 * Pointer to the error structure.
5610 * Pointer to mlx5_flow object on success,
5611 * otherwise NULL and rte_errno is set.
5613 static struct mlx5_flow *
5614 flow_dv_prepare(struct rte_eth_dev *dev,
5615 const struct rte_flow_attr *attr __rte_unused,
5616 const struct rte_flow_item items[] __rte_unused,
5617 const struct rte_flow_action actions[] __rte_unused,
5618 struct rte_flow_error *error)
5620 uint32_t handle_idx = 0;
5621 struct mlx5_flow *dev_flow;
5622 struct mlx5_flow_handle *dev_handle;
5623 struct mlx5_priv *priv = dev->data->dev_private;
5625 /* In case of corrupting the memory. */
5626 if (priv->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
5627 rte_flow_error_set(error, ENOSPC,
5628 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5629 "not free temporary device flow");
5632 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
5635 rte_flow_error_set(error, ENOMEM,
5636 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5637 "not enough memory to create flow handle");
5640 /* No multi-thread supporting. */
5641 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[priv->flow_idx++];
5642 dev_flow->handle = dev_handle;
5643 dev_flow->handle_idx = handle_idx;
5644 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
5646 * The matching value needs to be cleared to 0 before using. In the
5647 * past, it will be automatically cleared when using rte_*alloc
5648 * API. The time consumption will be almost the same as before.
5650 memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
5651 dev_flow->ingress = attr->ingress;
5652 dev_flow->dv.transfer = attr->transfer;
5656 #ifdef RTE_LIBRTE_MLX5_DEBUG
5658 * Sanity check for match mask and value. Similar to check_valid_spec() in
5659 * kernel driver. If unmasked bit is present in value, it returns failure.
5662 * pointer to match mask buffer.
5663 * @param match_value
5664 * pointer to match value buffer.
5667 * 0 if valid, -EINVAL otherwise.
5670 flow_dv_check_valid_spec(void *match_mask, void *match_value)
5672 uint8_t *m = match_mask;
5673 uint8_t *v = match_value;
5676 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
5679 "match_value differs from match_criteria"
5680 " %p[%u] != %p[%u]",
5681 match_value, i, match_mask, i);
5690 * Add match of ip_version.
5694 * @param[in] headers_v
5695 * Values header pointer.
5696 * @param[in] headers_m
5697 * Masks header pointer.
5698 * @param[in] ip_version
5699 * The IP version to set.
5702 flow_dv_set_match_ip_version(uint32_t group,
5708 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5710 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
5712 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
5713 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
5714 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
5718 * Add Ethernet item to matcher and to the value.
5720 * @param[in, out] matcher
5722 * @param[in, out] key
5723 * Flow matcher value.
5725 * Flow pattern to translate.
5727 * Item is inner pattern.
5730 flow_dv_translate_item_eth(void *matcher, void *key,
5731 const struct rte_flow_item *item, int inner,
5734 const struct rte_flow_item_eth *eth_m = item->mask;
5735 const struct rte_flow_item_eth *eth_v = item->spec;
5736 const struct rte_flow_item_eth nic_mask = {
5737 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5738 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5739 .type = RTE_BE16(0xffff),
5751 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5753 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5755 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5757 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5759 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
5760 ð_m->dst, sizeof(eth_m->dst));
5761 /* The value must be in the range of the mask. */
5762 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
5763 for (i = 0; i < sizeof(eth_m->dst); ++i)
5764 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
5765 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
5766 ð_m->src, sizeof(eth_m->src));
5767 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
5768 /* The value must be in the range of the mask. */
5769 for (i = 0; i < sizeof(eth_m->dst); ++i)
5770 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
5772 /* When ethertype is present set mask for tagged VLAN. */
5773 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5774 /* Set value for tagged VLAN if ethertype is 802.1Q. */
5775 if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
5776 eth_v->type == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
5777 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag,
5779 /* Return here to avoid setting match on ethertype. */
5784 * HW supports match on one Ethertype, the Ethertype following the last
5785 * VLAN tag of the packet (see PRM).
5786 * Set match on ethertype only if ETH header is not followed by VLAN.
5787 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
5788 * ethertype, and use ip_version field instead.
5790 if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
5791 eth_m->type == 0xFFFF) {
5792 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
5793 } else if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
5794 eth_m->type == 0xFFFF) {
5795 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
5797 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
5798 rte_be_to_cpu_16(eth_m->type));
5799 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5801 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
5806 * Add VLAN item to matcher and to the value.
5808 * @param[in, out] dev_flow
5810 * @param[in, out] matcher
5812 * @param[in, out] key
5813 * Flow matcher value.
5815 * Flow pattern to translate.
5817 * Item is inner pattern.
5820 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
5821 void *matcher, void *key,
5822 const struct rte_flow_item *item,
5823 int inner, uint32_t group)
5825 const struct rte_flow_item_vlan *vlan_m = item->mask;
5826 const struct rte_flow_item_vlan *vlan_v = item->spec;
5833 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5835 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5837 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5839 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5841 * This is workaround, masks are not supported,
5842 * and pre-validated.
5845 dev_flow->handle->vf_vlan.tag =
5846 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
5849 * When VLAN item exists in flow, mark packet as tagged,
5850 * even if TCI is not specified.
5852 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5853 MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
5857 vlan_m = &rte_flow_item_vlan_mask;
5858 tci_m = rte_be_to_cpu_16(vlan_m->tci);
5859 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
5860 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
5861 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
5862 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
5863 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
5864 MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
5865 MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
5867 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
5868 * ethertype, and use ip_version field instead.
5870 if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
5871 vlan_m->inner_type == 0xFFFF) {
5872 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
5873 } else if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
5874 vlan_m->inner_type == 0xFFFF) {
5875 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
5877 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
5878 rte_be_to_cpu_16(vlan_m->inner_type));
5879 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
5880 rte_be_to_cpu_16(vlan_m->inner_type &
5881 vlan_v->inner_type));
5886 * Add IPV4 item to matcher and to the value.
5888 * @param[in, out] matcher
5890 * @param[in, out] key
5891 * Flow matcher value.
5893 * Flow pattern to translate.
5894 * @param[in] item_flags
5895 * Bit-fields that holds the items detected until now.
5897 * Item is inner pattern.
5899 * The group to insert the rule.
5902 flow_dv_translate_item_ipv4(void *matcher, void *key,
5903 const struct rte_flow_item *item,
5904 const uint64_t item_flags,
5905 int inner, uint32_t group)
5907 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
5908 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
5909 const struct rte_flow_item_ipv4 nic_mask = {
5911 .src_addr = RTE_BE32(0xffffffff),
5912 .dst_addr = RTE_BE32(0xffffffff),
5913 .type_of_service = 0xff,
5914 .next_proto_id = 0xff,
5915 .time_to_live = 0xff,
5925 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5927 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5929 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5931 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5933 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
5935 * On outer header (which must contains L2), or inner header with L2,
5936 * set cvlan_tag mask bit to mark this packet as untagged.
5937 * This should be done even if item->spec is empty.
5939 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
5940 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5945 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5946 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
5947 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5948 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
5949 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
5950 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
5951 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
5952 src_ipv4_src_ipv6.ipv4_layout.ipv4);
5953 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
5954 src_ipv4_src_ipv6.ipv4_layout.ipv4);
5955 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
5956 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
5957 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
5958 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
5959 ipv4_m->hdr.type_of_service);
5960 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
5961 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
5962 ipv4_m->hdr.type_of_service >> 2);
5963 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
5964 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
5965 ipv4_m->hdr.next_proto_id);
5966 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
5967 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
5968 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
5969 ipv4_m->hdr.time_to_live);
5970 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
5971 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
5975 * Add IPV6 item to matcher and to the value.
5977 * @param[in, out] matcher
5979 * @param[in, out] key
5980 * Flow matcher value.
5982 * Flow pattern to translate.
5983 * @param[in] item_flags
5984 * Bit-fields that holds the items detected until now.
5986 * Item is inner pattern.
5988 * The group to insert the rule.
5991 flow_dv_translate_item_ipv6(void *matcher, void *key,
5992 const struct rte_flow_item *item,
5993 const uint64_t item_flags,
5994 int inner, uint32_t group)
5996 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
5997 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
5998 const struct rte_flow_item_ipv6 nic_mask = {
6001 "\xff\xff\xff\xff\xff\xff\xff\xff"
6002 "\xff\xff\xff\xff\xff\xff\xff\xff",
6004 "\xff\xff\xff\xff\xff\xff\xff\xff"
6005 "\xff\xff\xff\xff\xff\xff\xff\xff",
6006 .vtc_flow = RTE_BE32(0xffffffff),
6013 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6014 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6023 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6025 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6027 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6029 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6031 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6033 * On outer header (which must contains L2), or inner header with L2,
6034 * set cvlan_tag mask bit to mark this packet as untagged.
6035 * This should be done even if item->spec is empty.
6037 if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6038 MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6043 size = sizeof(ipv6_m->hdr.dst_addr);
6044 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6045 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6046 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6047 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6048 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
6049 for (i = 0; i < size; ++i)
6050 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
6051 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6052 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6053 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6054 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6055 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
6056 for (i = 0; i < size; ++i)
6057 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
6059 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
6060 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
6061 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
6062 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
6063 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
6064 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
6067 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
6069 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
6072 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
6074 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
6078 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6080 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6081 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
6083 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6084 ipv6_m->hdr.hop_limits);
6085 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6086 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
6090 * Add TCP item to matcher and to the value.
6092 * @param[in, out] matcher
6094 * @param[in, out] key
6095 * Flow matcher value.
6097 * Flow pattern to translate.
6099 * Item is inner pattern.
6102 flow_dv_translate_item_tcp(void *matcher, void *key,
6103 const struct rte_flow_item *item,
6106 const struct rte_flow_item_tcp *tcp_m = item->mask;
6107 const struct rte_flow_item_tcp *tcp_v = item->spec;
6112 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6114 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6116 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6118 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6120 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6121 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
6125 tcp_m = &rte_flow_item_tcp_mask;
6126 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
6127 rte_be_to_cpu_16(tcp_m->hdr.src_port));
6128 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
6129 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
6130 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
6131 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
6132 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
6133 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
6134 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
6135 tcp_m->hdr.tcp_flags);
6136 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
6137 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
6141 * Add UDP item to matcher and to the value.
6143 * @param[in, out] matcher
6145 * @param[in, out] key
6146 * Flow matcher value.
6148 * Flow pattern to translate.
6150 * Item is inner pattern.
6153 flow_dv_translate_item_udp(void *matcher, void *key,
6154 const struct rte_flow_item *item,
6157 const struct rte_flow_item_udp *udp_m = item->mask;
6158 const struct rte_flow_item_udp *udp_v = item->spec;
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 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6172 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
6176 udp_m = &rte_flow_item_udp_mask;
6177 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
6178 rte_be_to_cpu_16(udp_m->hdr.src_port));
6179 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
6180 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
6181 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
6182 rte_be_to_cpu_16(udp_m->hdr.dst_port));
6183 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6184 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
6188 * Add GRE optional Key item to matcher and to the value.
6190 * @param[in, out] matcher
6192 * @param[in, out] key
6193 * Flow matcher value.
6195 * Flow pattern to translate.
6197 * Item is inner pattern.
6200 flow_dv_translate_item_gre_key(void *matcher, void *key,
6201 const struct rte_flow_item *item)
6203 const rte_be32_t *key_m = item->mask;
6204 const rte_be32_t *key_v = item->spec;
6205 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6206 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6207 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
6209 /* GRE K bit must be on and should already be validated */
6210 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
6211 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
6215 key_m = &gre_key_default_mask;
6216 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
6217 rte_be_to_cpu_32(*key_m) >> 8);
6218 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
6219 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
6220 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
6221 rte_be_to_cpu_32(*key_m) & 0xFF);
6222 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
6223 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
6227 * Add GRE item to matcher and to the value.
6229 * @param[in, out] matcher
6231 * @param[in, out] key
6232 * Flow matcher value.
6234 * Flow pattern to translate.
6236 * Item is inner pattern.
6239 flow_dv_translate_item_gre(void *matcher, void *key,
6240 const struct rte_flow_item *item,
6243 const struct rte_flow_item_gre *gre_m = item->mask;
6244 const struct rte_flow_item_gre *gre_v = item->spec;
6247 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6248 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6255 uint16_t s_present:1;
6256 uint16_t k_present:1;
6257 uint16_t rsvd_bit1:1;
6258 uint16_t c_present:1;
6262 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
6265 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6267 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6269 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6271 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6273 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6274 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
6278 gre_m = &rte_flow_item_gre_mask;
6279 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
6280 rte_be_to_cpu_16(gre_m->protocol));
6281 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6282 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
6283 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
6284 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
6285 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
6286 gre_crks_rsvd0_ver_m.c_present);
6287 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6288 gre_crks_rsvd0_ver_v.c_present &
6289 gre_crks_rsvd0_ver_m.c_present);
6290 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6291 gre_crks_rsvd0_ver_m.k_present);
6292 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6293 gre_crks_rsvd0_ver_v.k_present &
6294 gre_crks_rsvd0_ver_m.k_present);
6295 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6296 gre_crks_rsvd0_ver_m.s_present);
6297 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6298 gre_crks_rsvd0_ver_v.s_present &
6299 gre_crks_rsvd0_ver_m.s_present);
6303 * Add NVGRE item to matcher and to the value.
6305 * @param[in, out] matcher
6307 * @param[in, out] key
6308 * Flow matcher value.
6310 * Flow pattern to translate.
6312 * Item is inner pattern.
6315 flow_dv_translate_item_nvgre(void *matcher, void *key,
6316 const struct rte_flow_item *item,
6319 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6320 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6321 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6322 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6323 const char *tni_flow_id_m = (const char *)nvgre_m->tni;
6324 const char *tni_flow_id_v = (const char *)nvgre_v->tni;
6330 /* For NVGRE, GRE header fields must be set with defined values. */
6331 const struct rte_flow_item_gre gre_spec = {
6332 .c_rsvd0_ver = RTE_BE16(0x2000),
6333 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6335 const struct rte_flow_item_gre gre_mask = {
6336 .c_rsvd0_ver = RTE_BE16(0xB000),
6337 .protocol = RTE_BE16(UINT16_MAX),
6339 const struct rte_flow_item gre_item = {
6344 flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6348 nvgre_m = &rte_flow_item_nvgre_mask;
6349 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6350 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6351 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6352 memcpy(gre_key_m, tni_flow_id_m, size);
6353 for (i = 0; i < size; ++i)
6354 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6358 * Add VXLAN item to matcher and to the value.
6360 * @param[in, out] matcher
6362 * @param[in, out] key
6363 * Flow matcher value.
6365 * Flow pattern to translate.
6367 * Item is inner pattern.
6370 flow_dv_translate_item_vxlan(void *matcher, void *key,
6371 const struct rte_flow_item *item,
6374 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
6375 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
6378 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6379 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6387 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6389 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6391 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6393 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6395 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6396 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6397 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6398 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6399 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6404 vxlan_m = &rte_flow_item_vxlan_mask;
6405 size = sizeof(vxlan_m->vni);
6406 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
6407 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
6408 memcpy(vni_m, vxlan_m->vni, size);
6409 for (i = 0; i < size; ++i)
6410 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6414 * Add VXLAN-GPE item to matcher and to the value.
6416 * @param[in, out] matcher
6418 * @param[in, out] key
6419 * Flow matcher value.
6421 * Flow pattern to translate.
6423 * Item is inner pattern.
6427 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
6428 const struct rte_flow_item *item, int inner)
6430 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
6431 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
6435 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
6437 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6443 uint8_t flags_m = 0xff;
6444 uint8_t flags_v = 0xc;
6447 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6449 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6451 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6453 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6455 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6456 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6457 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6458 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6459 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6464 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
6465 size = sizeof(vxlan_m->vni);
6466 vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
6467 vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
6468 memcpy(vni_m, vxlan_m->vni, size);
6469 for (i = 0; i < size; ++i)
6470 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6471 if (vxlan_m->flags) {
6472 flags_m = vxlan_m->flags;
6473 flags_v = vxlan_v->flags;
6475 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
6476 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
6477 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
6479 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
6484 * Add Geneve item to matcher and to the value.
6486 * @param[in, out] matcher
6488 * @param[in, out] key
6489 * Flow matcher value.
6491 * Flow pattern to translate.
6493 * Item is inner pattern.
6497 flow_dv_translate_item_geneve(void *matcher, void *key,
6498 const struct rte_flow_item *item, int inner)
6500 const struct rte_flow_item_geneve *geneve_m = item->mask;
6501 const struct rte_flow_item_geneve *geneve_v = item->spec;
6504 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6505 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6514 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6516 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6518 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6520 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6522 dport = MLX5_UDP_PORT_GENEVE;
6523 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6524 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6525 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6530 geneve_m = &rte_flow_item_geneve_mask;
6531 size = sizeof(geneve_m->vni);
6532 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
6533 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
6534 memcpy(vni_m, geneve_m->vni, size);
6535 for (i = 0; i < size; ++i)
6536 vni_v[i] = vni_m[i] & geneve_v->vni[i];
6537 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
6538 rte_be_to_cpu_16(geneve_m->protocol));
6539 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
6540 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
6541 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
6542 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
6543 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
6544 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6545 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
6546 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6547 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
6548 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6549 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
6550 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
6551 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6555 * Add MPLS item to matcher and to the value.
6557 * @param[in, out] matcher
6559 * @param[in, out] key
6560 * Flow matcher value.
6562 * Flow pattern to translate.
6563 * @param[in] prev_layer
6564 * The protocol layer indicated in previous item.
6566 * Item is inner pattern.
6569 flow_dv_translate_item_mpls(void *matcher, void *key,
6570 const struct rte_flow_item *item,
6571 uint64_t prev_layer,
6574 const uint32_t *in_mpls_m = item->mask;
6575 const uint32_t *in_mpls_v = item->spec;
6576 uint32_t *out_mpls_m = 0;
6577 uint32_t *out_mpls_v = 0;
6578 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6579 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6580 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
6582 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6583 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
6584 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6586 switch (prev_layer) {
6587 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6588 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
6589 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6590 MLX5_UDP_PORT_MPLS);
6592 case MLX5_FLOW_LAYER_GRE:
6593 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
6594 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6595 RTE_ETHER_TYPE_MPLS);
6598 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6599 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6606 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
6607 switch (prev_layer) {
6608 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6610 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6611 outer_first_mpls_over_udp);
6613 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6614 outer_first_mpls_over_udp);
6616 case MLX5_FLOW_LAYER_GRE:
6618 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6619 outer_first_mpls_over_gre);
6621 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6622 outer_first_mpls_over_gre);
6625 /* Inner MPLS not over GRE is not supported. */
6628 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6632 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6638 if (out_mpls_m && out_mpls_v) {
6639 *out_mpls_m = *in_mpls_m;
6640 *out_mpls_v = *in_mpls_v & *in_mpls_m;
6645 * Add metadata register item to matcher
6647 * @param[in, out] matcher
6649 * @param[in, out] key
6650 * Flow matcher value.
6651 * @param[in] reg_type
6652 * Type of device metadata register
6659 flow_dv_match_meta_reg(void *matcher, void *key,
6660 enum modify_reg reg_type,
6661 uint32_t data, uint32_t mask)
6664 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
6666 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6672 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
6673 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
6676 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
6677 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
6681 * The metadata register C0 field might be divided into
6682 * source vport index and META item value, we should set
6683 * this field according to specified mask, not as whole one.
6685 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
6687 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
6688 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
6691 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
6694 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
6695 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
6698 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
6699 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
6702 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
6703 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
6706 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
6707 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
6710 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
6711 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
6714 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
6715 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
6718 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
6719 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
6728 * Add MARK item to matcher
6731 * The device to configure through.
6732 * @param[in, out] matcher
6734 * @param[in, out] key
6735 * Flow matcher value.
6737 * Flow pattern to translate.
6740 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
6741 void *matcher, void *key,
6742 const struct rte_flow_item *item)
6744 struct mlx5_priv *priv = dev->data->dev_private;
6745 const struct rte_flow_item_mark *mark;
6749 mark = item->mask ? (const void *)item->mask :
6750 &rte_flow_item_mark_mask;
6751 mask = mark->id & priv->sh->dv_mark_mask;
6752 mark = (const void *)item->spec;
6754 value = mark->id & priv->sh->dv_mark_mask & mask;
6756 enum modify_reg reg;
6758 /* Get the metadata register index for the mark. */
6759 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
6760 MLX5_ASSERT(reg > 0);
6761 if (reg == REG_C_0) {
6762 struct mlx5_priv *priv = dev->data->dev_private;
6763 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6764 uint32_t shl_c0 = rte_bsf32(msk_c0);
6770 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6775 * Add META item to matcher
6778 * The devich to configure through.
6779 * @param[in, out] matcher
6781 * @param[in, out] key
6782 * Flow matcher value.
6784 * Attributes of flow that includes this item.
6786 * Flow pattern to translate.
6789 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
6790 void *matcher, void *key,
6791 const struct rte_flow_attr *attr,
6792 const struct rte_flow_item *item)
6794 const struct rte_flow_item_meta *meta_m;
6795 const struct rte_flow_item_meta *meta_v;
6797 meta_m = (const void *)item->mask;
6799 meta_m = &rte_flow_item_meta_mask;
6800 meta_v = (const void *)item->spec;
6803 uint32_t value = meta_v->data;
6804 uint32_t mask = meta_m->data;
6806 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
6810 * In datapath code there is no endianness
6811 * coversions for perfromance reasons, all
6812 * pattern conversions are done in rte_flow.
6814 value = rte_cpu_to_be_32(value);
6815 mask = rte_cpu_to_be_32(mask);
6816 if (reg == REG_C_0) {
6817 struct mlx5_priv *priv = dev->data->dev_private;
6818 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6819 uint32_t shl_c0 = rte_bsf32(msk_c0);
6820 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
6821 uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
6828 MLX5_ASSERT(msk_c0);
6829 MLX5_ASSERT(!(~msk_c0 & mask));
6831 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6836 * Add vport metadata Reg C0 item to matcher
6838 * @param[in, out] matcher
6840 * @param[in, out] key
6841 * Flow matcher value.
6843 * Flow pattern to translate.
6846 flow_dv_translate_item_meta_vport(void *matcher, void *key,
6847 uint32_t value, uint32_t mask)
6849 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
6853 * Add tag item to matcher
6856 * The devich to configure through.
6857 * @param[in, out] matcher
6859 * @param[in, out] key
6860 * Flow matcher value.
6862 * Flow pattern to translate.
6865 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
6866 void *matcher, void *key,
6867 const struct rte_flow_item *item)
6869 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
6870 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
6871 uint32_t mask, value;
6874 value = tag_v->data;
6875 mask = tag_m ? tag_m->data : UINT32_MAX;
6876 if (tag_v->id == REG_C_0) {
6877 struct mlx5_priv *priv = dev->data->dev_private;
6878 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6879 uint32_t shl_c0 = rte_bsf32(msk_c0);
6885 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
6889 * Add TAG item to matcher
6892 * The devich to configure through.
6893 * @param[in, out] matcher
6895 * @param[in, out] key
6896 * Flow matcher value.
6898 * Flow pattern to translate.
6901 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
6902 void *matcher, void *key,
6903 const struct rte_flow_item *item)
6905 const struct rte_flow_item_tag *tag_v = item->spec;
6906 const struct rte_flow_item_tag *tag_m = item->mask;
6907 enum modify_reg reg;
6910 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
6911 /* Get the metadata register index for the tag. */
6912 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
6913 MLX5_ASSERT(reg > 0);
6914 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
6918 * Add source vport match to the specified matcher.
6920 * @param[in, out] matcher
6922 * @param[in, out] key
6923 * Flow matcher value.
6925 * Source vport value to match
6930 flow_dv_translate_item_source_vport(void *matcher, void *key,
6931 int16_t port, uint16_t mask)
6933 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6934 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6936 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
6937 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
6941 * Translate port-id item to eswitch match on port-id.
6944 * The devich to configure through.
6945 * @param[in, out] matcher
6947 * @param[in, out] key
6948 * Flow matcher value.
6950 * Flow pattern to translate.
6953 * 0 on success, a negative errno value otherwise.
6956 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
6957 void *key, const struct rte_flow_item *item)
6959 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
6960 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
6961 struct mlx5_priv *priv;
6964 mask = pid_m ? pid_m->id : 0xffff;
6965 id = pid_v ? pid_v->id : dev->data->port_id;
6966 priv = mlx5_port_to_eswitch_info(id, item == NULL);
6969 /* Translate to vport field or to metadata, depending on mode. */
6970 if (priv->vport_meta_mask)
6971 flow_dv_translate_item_meta_vport(matcher, key,
6972 priv->vport_meta_tag,
6973 priv->vport_meta_mask);
6975 flow_dv_translate_item_source_vport(matcher, key,
6976 priv->vport_id, mask);
6981 * Add ICMP6 item to matcher and to the value.
6983 * @param[in, out] matcher
6985 * @param[in, out] key
6986 * Flow matcher value.
6988 * Flow pattern to translate.
6990 * Item is inner pattern.
6993 flow_dv_translate_item_icmp6(void *matcher, void *key,
6994 const struct rte_flow_item *item,
6997 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
6998 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
7001 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7003 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7005 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7007 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7009 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7011 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7013 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7014 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
7018 icmp6_m = &rte_flow_item_icmp6_mask;
7020 * Force flow only to match the non-fragmented IPv6 ICMPv6 packets.
7021 * If only the protocol is specified, no need to match the frag.
7023 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
7024 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
7025 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
7026 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
7027 icmp6_v->type & icmp6_m->type);
7028 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
7029 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
7030 icmp6_v->code & icmp6_m->code);
7034 * Add ICMP item to matcher and to the value.
7036 * @param[in, out] matcher
7038 * @param[in, out] key
7039 * Flow matcher value.
7041 * Flow pattern to translate.
7043 * Item is inner pattern.
7046 flow_dv_translate_item_icmp(void *matcher, void *key,
7047 const struct rte_flow_item *item,
7050 const struct rte_flow_item_icmp *icmp_m = item->mask;
7051 const struct rte_flow_item_icmp *icmp_v = item->spec;
7054 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7056 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7058 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7060 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7062 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7064 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7066 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7067 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
7071 icmp_m = &rte_flow_item_icmp_mask;
7073 * Force flow only to match the non-fragmented IPv4 ICMP packets.
7074 * If only the protocol is specified, no need to match the frag.
7076 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
7077 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
7078 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
7079 icmp_m->hdr.icmp_type);
7080 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
7081 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
7082 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
7083 icmp_m->hdr.icmp_code);
7084 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
7085 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
7089 * Add GTP item to matcher and to the value.
7091 * @param[in, out] matcher
7093 * @param[in, out] key
7094 * Flow matcher value.
7096 * Flow pattern to translate.
7098 * Item is inner pattern.
7101 flow_dv_translate_item_gtp(void *matcher, void *key,
7102 const struct rte_flow_item *item, int inner)
7104 const struct rte_flow_item_gtp *gtp_m = item->mask;
7105 const struct rte_flow_item_gtp *gtp_v = item->spec;
7108 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7110 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7111 uint16_t dport = RTE_GTPU_UDP_PORT;
7114 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7116 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7118 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7120 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7122 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7123 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7124 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7129 gtp_m = &rte_flow_item_gtp_mask;
7130 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
7131 gtp_m->v_pt_rsv_flags);
7132 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
7133 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
7134 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
7135 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
7136 gtp_v->msg_type & gtp_m->msg_type);
7137 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
7138 rte_be_to_cpu_32(gtp_m->teid));
7139 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
7140 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
7143 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
7145 #define HEADER_IS_ZERO(match_criteria, headers) \
7146 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
7147 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
7150 * Calculate flow matcher enable bitmap.
7152 * @param match_criteria
7153 * Pointer to flow matcher criteria.
7156 * Bitmap of enabled fields.
7159 flow_dv_matcher_enable(uint32_t *match_criteria)
7161 uint8_t match_criteria_enable;
7163 match_criteria_enable =
7164 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
7165 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
7166 match_criteria_enable |=
7167 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
7168 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
7169 match_criteria_enable |=
7170 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
7171 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
7172 match_criteria_enable |=
7173 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
7174 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
7175 match_criteria_enable |=
7176 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
7177 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
7178 return match_criteria_enable;
7185 * @param[in, out] dev
7186 * Pointer to rte_eth_dev structure.
7187 * @param[in] table_id
7190 * Direction of the table.
7191 * @param[in] transfer
7192 * E-Switch or NIC flow.
7194 * pointer to error structure.
7197 * Returns tables resource based on the index, NULL in case of failed.
7199 static struct mlx5_flow_tbl_resource *
7200 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
7201 uint32_t table_id, uint8_t egress,
7203 struct rte_flow_error *error)
7205 struct mlx5_priv *priv = dev->data->dev_private;
7206 struct mlx5_ibv_shared *sh = priv->sh;
7207 struct mlx5_flow_tbl_resource *tbl;
7208 union mlx5_flow_tbl_key table_key = {
7210 .table_id = table_id,
7212 .domain = !!transfer,
7213 .direction = !!egress,
7216 struct mlx5_hlist_entry *pos = mlx5_hlist_lookup(sh->flow_tbls,
7218 struct mlx5_flow_tbl_data_entry *tbl_data;
7224 tbl_data = container_of(pos, struct mlx5_flow_tbl_data_entry,
7226 tbl = &tbl_data->tbl;
7227 rte_atomic32_inc(&tbl->refcnt);
7230 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
7232 rte_flow_error_set(error, ENOMEM,
7233 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7235 "cannot allocate flow table data entry");
7238 tbl_data->idx = idx;
7239 tbl = &tbl_data->tbl;
7240 pos = &tbl_data->entry;
7242 domain = sh->fdb_domain;
7244 domain = sh->tx_domain;
7246 domain = sh->rx_domain;
7247 tbl->obj = mlx5_glue->dr_create_flow_tbl(domain, table_id);
7249 rte_flow_error_set(error, ENOMEM,
7250 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7251 NULL, "cannot create flow table object");
7252 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7256 * No multi-threads now, but still better to initialize the reference
7257 * count before insert it into the hash list.
7259 rte_atomic32_init(&tbl->refcnt);
7260 /* Jump action reference count is initialized here. */
7261 rte_atomic32_init(&tbl_data->jump.refcnt);
7262 pos->key = table_key.v64;
7263 ret = mlx5_hlist_insert(sh->flow_tbls, pos);
7265 rte_flow_error_set(error, -ret,
7266 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7267 "cannot insert flow table data entry");
7268 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
7269 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7271 rte_atomic32_inc(&tbl->refcnt);
7276 * Release a flow table.
7279 * Pointer to rte_eth_dev structure.
7281 * Table resource to be released.
7284 * Returns 0 if table was released, else return 1;
7287 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
7288 struct mlx5_flow_tbl_resource *tbl)
7290 struct mlx5_priv *priv = dev->data->dev_private;
7291 struct mlx5_ibv_shared *sh = priv->sh;
7292 struct mlx5_flow_tbl_data_entry *tbl_data =
7293 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7297 if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
7298 struct mlx5_hlist_entry *pos = &tbl_data->entry;
7300 mlx5_glue->dr_destroy_flow_tbl(tbl->obj);
7302 /* remove the entry from the hash list and free memory. */
7303 mlx5_hlist_remove(sh->flow_tbls, pos);
7304 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_JUMP],
7312 * Register the flow matcher.
7314 * @param[in, out] dev
7315 * Pointer to rte_eth_dev structure.
7316 * @param[in, out] matcher
7317 * Pointer to flow matcher.
7318 * @param[in, out] key
7319 * Pointer to flow table key.
7320 * @parm[in, out] dev_flow
7321 * Pointer to the dev_flow.
7323 * pointer to error structure.
7326 * 0 on success otherwise -errno and errno is set.
7329 flow_dv_matcher_register(struct rte_eth_dev *dev,
7330 struct mlx5_flow_dv_matcher *matcher,
7331 union mlx5_flow_tbl_key *key,
7332 struct mlx5_flow *dev_flow,
7333 struct rte_flow_error *error)
7335 struct mlx5_priv *priv = dev->data->dev_private;
7336 struct mlx5_ibv_shared *sh = priv->sh;
7337 struct mlx5_flow_dv_matcher *cache_matcher;
7338 struct mlx5dv_flow_matcher_attr dv_attr = {
7339 .type = IBV_FLOW_ATTR_NORMAL,
7340 .match_mask = (void *)&matcher->mask,
7342 struct mlx5_flow_tbl_resource *tbl;
7343 struct mlx5_flow_tbl_data_entry *tbl_data;
7345 tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
7346 key->domain, error);
7348 return -rte_errno; /* No need to refill the error info */
7349 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7350 /* Lookup from cache. */
7351 LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
7352 if (matcher->crc == cache_matcher->crc &&
7353 matcher->priority == cache_matcher->priority &&
7354 !memcmp((const void *)matcher->mask.buf,
7355 (const void *)cache_matcher->mask.buf,
7356 cache_matcher->mask.size)) {
7358 "%s group %u priority %hd use %s "
7359 "matcher %p: refcnt %d++",
7360 key->domain ? "FDB" : "NIC", key->table_id,
7361 cache_matcher->priority,
7362 key->direction ? "tx" : "rx",
7363 (void *)cache_matcher,
7364 rte_atomic32_read(&cache_matcher->refcnt));
7365 rte_atomic32_inc(&cache_matcher->refcnt);
7366 dev_flow->handle->dvh.matcher = cache_matcher;
7367 /* old matcher should not make the table ref++. */
7368 flow_dv_tbl_resource_release(dev, tbl);
7372 /* Register new matcher. */
7373 cache_matcher = rte_calloc(__func__, 1, sizeof(*cache_matcher), 0);
7374 if (!cache_matcher) {
7375 flow_dv_tbl_resource_release(dev, tbl);
7376 return rte_flow_error_set(error, ENOMEM,
7377 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7378 "cannot allocate matcher memory");
7380 *cache_matcher = *matcher;
7381 dv_attr.match_criteria_enable =
7382 flow_dv_matcher_enable(cache_matcher->mask.buf);
7383 dv_attr.priority = matcher->priority;
7385 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
7386 cache_matcher->matcher_object =
7387 mlx5_glue->dv_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj);
7388 if (!cache_matcher->matcher_object) {
7389 rte_free(cache_matcher);
7390 #ifdef HAVE_MLX5DV_DR
7391 flow_dv_tbl_resource_release(dev, tbl);
7393 return rte_flow_error_set(error, ENOMEM,
7394 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7395 NULL, "cannot create matcher");
7397 /* Save the table information */
7398 cache_matcher->tbl = tbl;
7399 rte_atomic32_init(&cache_matcher->refcnt);
7400 /* only matcher ref++, table ref++ already done above in get API. */
7401 rte_atomic32_inc(&cache_matcher->refcnt);
7402 LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
7403 dev_flow->handle->dvh.matcher = cache_matcher;
7404 DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
7405 key->domain ? "FDB" : "NIC", key->table_id,
7406 cache_matcher->priority,
7407 key->direction ? "tx" : "rx", (void *)cache_matcher,
7408 rte_atomic32_read(&cache_matcher->refcnt));
7413 * Find existing tag resource or create and register a new one.
7415 * @param dev[in, out]
7416 * Pointer to rte_eth_dev structure.
7417 * @param[in, out] tag_be24
7418 * Tag value in big endian then R-shift 8.
7419 * @parm[in, out] dev_flow
7420 * Pointer to the dev_flow.
7422 * pointer to error structure.
7425 * 0 on success otherwise -errno and errno is set.
7428 flow_dv_tag_resource_register
7429 (struct rte_eth_dev *dev,
7431 struct mlx5_flow *dev_flow,
7432 struct rte_flow_error *error)
7434 struct mlx5_priv *priv = dev->data->dev_private;
7435 struct mlx5_ibv_shared *sh = priv->sh;
7436 struct mlx5_flow_dv_tag_resource *cache_resource;
7437 struct mlx5_hlist_entry *entry;
7439 /* Lookup a matching resource from cache. */
7440 entry = mlx5_hlist_lookup(sh->tag_table, (uint64_t)tag_be24);
7442 cache_resource = container_of
7443 (entry, struct mlx5_flow_dv_tag_resource, entry);
7444 rte_atomic32_inc(&cache_resource->refcnt);
7445 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
7446 dev_flow->dv.tag_resource = cache_resource;
7447 DRV_LOG(DEBUG, "cached tag resource %p: refcnt now %d++",
7448 (void *)cache_resource,
7449 rte_atomic32_read(&cache_resource->refcnt));
7452 /* Register new resource. */
7453 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG],
7454 &dev_flow->handle->dvh.rix_tag);
7455 if (!cache_resource)
7456 return rte_flow_error_set(error, ENOMEM,
7457 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7458 "cannot allocate resource memory");
7459 cache_resource->entry.key = (uint64_t)tag_be24;
7460 cache_resource->action = mlx5_glue->dv_create_flow_action_tag(tag_be24);
7461 if (!cache_resource->action) {
7462 rte_free(cache_resource);
7463 return rte_flow_error_set(error, ENOMEM,
7464 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7465 NULL, "cannot create action");
7467 rte_atomic32_init(&cache_resource->refcnt);
7468 rte_atomic32_inc(&cache_resource->refcnt);
7469 if (mlx5_hlist_insert(sh->tag_table, &cache_resource->entry)) {
7470 mlx5_glue->destroy_flow_action(cache_resource->action);
7471 rte_free(cache_resource);
7472 return rte_flow_error_set(error, EEXIST,
7473 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7474 NULL, "cannot insert tag");
7476 dev_flow->dv.tag_resource = cache_resource;
7477 DRV_LOG(DEBUG, "new tag resource %p: refcnt now %d++",
7478 (void *)cache_resource,
7479 rte_atomic32_read(&cache_resource->refcnt));
7487 * Pointer to Ethernet device.
7492 * 1 while a reference on it exists, 0 when freed.
7495 flow_dv_tag_release(struct rte_eth_dev *dev,
7498 struct mlx5_priv *priv = dev->data->dev_private;
7499 struct mlx5_ibv_shared *sh = priv->sh;
7500 struct mlx5_flow_dv_tag_resource *tag;
7502 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7505 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
7506 dev->data->port_id, (void *)tag,
7507 rte_atomic32_read(&tag->refcnt));
7508 if (rte_atomic32_dec_and_test(&tag->refcnt)) {
7509 claim_zero(mlx5_glue->destroy_flow_action(tag->action));
7510 mlx5_hlist_remove(sh->tag_table, &tag->entry);
7511 DRV_LOG(DEBUG, "port %u tag %p: removed",
7512 dev->data->port_id, (void *)tag);
7513 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7520 * Translate port ID action to vport.
7523 * Pointer to rte_eth_dev structure.
7525 * Pointer to the port ID action.
7526 * @param[out] dst_port_id
7527 * The target port ID.
7529 * Pointer to the error structure.
7532 * 0 on success, a negative errno value otherwise and rte_errno is set.
7535 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
7536 const struct rte_flow_action *action,
7537 uint32_t *dst_port_id,
7538 struct rte_flow_error *error)
7541 struct mlx5_priv *priv;
7542 const struct rte_flow_action_port_id *conf =
7543 (const struct rte_flow_action_port_id *)action->conf;
7545 port = conf->original ? dev->data->port_id : conf->id;
7546 priv = mlx5_port_to_eswitch_info(port, false);
7548 return rte_flow_error_set(error, -rte_errno,
7549 RTE_FLOW_ERROR_TYPE_ACTION,
7551 "No eswitch info was found for port");
7552 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
7554 * This parameter is transferred to
7555 * mlx5dv_dr_action_create_dest_ib_port().
7557 *dst_port_id = priv->ibv_port;
7560 * Legacy mode, no LAG configurations is supported.
7561 * This parameter is transferred to
7562 * mlx5dv_dr_action_create_dest_vport().
7564 *dst_port_id = priv->vport_id;
7570 * Create a counter with aging configuration.
7573 * Pointer to rte_eth_dev structure.
7575 * Pointer to the counter action configuration.
7577 * Pointer to the aging action configuration.
7580 * Index to flow counter on success, 0 otherwise.
7583 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
7584 struct mlx5_flow *dev_flow,
7585 const struct rte_flow_action_count *count,
7586 const struct rte_flow_action_age *age)
7589 struct mlx5_age_param *age_param;
7591 counter = flow_dv_counter_alloc(dev,
7592 count ? count->shared : 0,
7593 count ? count->id : 0,
7594 dev_flow->dv.group, !!age);
7595 if (!counter || age == NULL)
7597 age_param = flow_dv_counter_idx_get_age(dev, counter);
7599 * The counter age accuracy may have a bit delay. Have 3/4
7600 * second bias on the timeount in order to let it age in time.
7602 age_param->context = age->context ? age->context :
7603 (void *)(uintptr_t)(dev_flow->flow_idx);
7605 * The counter age accuracy may have a bit delay. Have 3/4
7606 * second bias on the timeount in order to let it age in time.
7608 age_param->timeout = age->timeout * 10 - MLX5_AGING_TIME_DELAY;
7609 /* Set expire time in unit of 0.1 sec. */
7610 age_param->port_id = dev->data->port_id;
7611 age_param->expire = age_param->timeout +
7612 rte_rdtsc() / (rte_get_tsc_hz() / 10);
7613 rte_atomic16_set(&age_param->state, AGE_CANDIDATE);
7617 * Add Tx queue matcher
7620 * Pointer to the dev struct.
7621 * @param[in, out] matcher
7623 * @param[in, out] key
7624 * Flow matcher value.
7626 * Flow pattern to translate.
7628 * Item is inner pattern.
7631 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
7632 void *matcher, void *key,
7633 const struct rte_flow_item *item)
7635 const struct mlx5_rte_flow_item_tx_queue *queue_m;
7636 const struct mlx5_rte_flow_item_tx_queue *queue_v;
7638 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7640 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7641 struct mlx5_txq_ctrl *txq;
7645 queue_m = (const void *)item->mask;
7648 queue_v = (const void *)item->spec;
7651 txq = mlx5_txq_get(dev, queue_v->queue);
7654 queue = txq->obj->sq->id;
7655 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
7656 MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
7657 queue & queue_m->queue);
7658 mlx5_txq_release(dev, queue_v->queue);
7662 * Set the hash fields according to the @p flow information.
7664 * @param[in] dev_flow
7665 * Pointer to the mlx5_flow.
7666 * @param[in] rss_desc
7667 * Pointer to the mlx5_flow_rss_desc.
7670 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
7671 struct mlx5_flow_rss_desc *rss_desc)
7673 uint64_t items = dev_flow->handle->layers;
7675 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
7677 dev_flow->hash_fields = 0;
7678 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
7679 if (rss_desc->level >= 2) {
7680 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
7684 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
7685 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
7686 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
7687 if (rss_types & ETH_RSS_L3_SRC_ONLY)
7688 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
7689 else if (rss_types & ETH_RSS_L3_DST_ONLY)
7690 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
7692 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
7694 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
7695 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
7696 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
7697 if (rss_types & ETH_RSS_L3_SRC_ONLY)
7698 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
7699 else if (rss_types & ETH_RSS_L3_DST_ONLY)
7700 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
7702 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
7705 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
7706 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
7707 if (rss_types & ETH_RSS_UDP) {
7708 if (rss_types & ETH_RSS_L4_SRC_ONLY)
7709 dev_flow->hash_fields |=
7710 IBV_RX_HASH_SRC_PORT_UDP;
7711 else if (rss_types & ETH_RSS_L4_DST_ONLY)
7712 dev_flow->hash_fields |=
7713 IBV_RX_HASH_DST_PORT_UDP;
7715 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
7717 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
7718 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
7719 if (rss_types & ETH_RSS_TCP) {
7720 if (rss_types & ETH_RSS_L4_SRC_ONLY)
7721 dev_flow->hash_fields |=
7722 IBV_RX_HASH_SRC_PORT_TCP;
7723 else if (rss_types & ETH_RSS_L4_DST_ONLY)
7724 dev_flow->hash_fields |=
7725 IBV_RX_HASH_DST_PORT_TCP;
7727 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
7733 * Fill the flow with DV spec, lock free
7734 * (mutex should be acquired by caller).
7737 * Pointer to rte_eth_dev structure.
7738 * @param[in, out] dev_flow
7739 * Pointer to the sub flow.
7741 * Pointer to the flow attributes.
7743 * Pointer to the list of items.
7744 * @param[in] actions
7745 * Pointer to the list of actions.
7747 * Pointer to the error structure.
7750 * 0 on success, a negative errno value otherwise and rte_errno is set.
7753 __flow_dv_translate(struct rte_eth_dev *dev,
7754 struct mlx5_flow *dev_flow,
7755 const struct rte_flow_attr *attr,
7756 const struct rte_flow_item items[],
7757 const struct rte_flow_action actions[],
7758 struct rte_flow_error *error)
7760 struct mlx5_priv *priv = dev->data->dev_private;
7761 struct mlx5_dev_config *dev_conf = &priv->config;
7762 struct rte_flow *flow = dev_flow->flow;
7763 struct mlx5_flow_handle *handle = dev_flow->handle;
7764 struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
7766 [!!priv->flow_nested_idx];
7767 uint64_t item_flags = 0;
7768 uint64_t last_item = 0;
7769 uint64_t action_flags = 0;
7770 uint64_t priority = attr->priority;
7771 struct mlx5_flow_dv_matcher matcher = {
7773 .size = sizeof(matcher.mask.buf),
7777 bool actions_end = false;
7779 struct mlx5_flow_dv_modify_hdr_resource res;
7780 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
7781 sizeof(struct mlx5_modification_cmd) *
7782 (MLX5_MAX_MODIFY_NUM + 1)];
7784 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
7785 const struct rte_flow_action_count *count = NULL;
7786 const struct rte_flow_action_age *age = NULL;
7787 union flow_dv_attr flow_attr = { .attr = 0 };
7789 union mlx5_flow_tbl_key tbl_key;
7790 uint32_t modify_action_position = UINT32_MAX;
7791 void *match_mask = matcher.mask.buf;
7792 void *match_value = dev_flow->dv.value.buf;
7793 uint8_t next_protocol = 0xff;
7794 struct rte_vlan_hdr vlan = { 0 };
7798 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
7799 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
7800 ret = mlx5_flow_group_to_table(attr, dev_flow->external, attr->group,
7801 !!priv->fdb_def_rule, &table, error);
7804 dev_flow->dv.group = table;
7806 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
7807 if (priority == MLX5_FLOW_PRIO_RSVD)
7808 priority = dev_conf->flow_prio - 1;
7809 /* number of actions must be set to 0 in case of dirty stack. */
7810 mhdr_res->actions_num = 0;
7811 for (; !actions_end ; actions++) {
7812 const struct rte_flow_action_queue *queue;
7813 const struct rte_flow_action_rss *rss;
7814 const struct rte_flow_action *action = actions;
7815 const uint8_t *rss_key;
7816 const struct rte_flow_action_jump *jump_data;
7817 const struct rte_flow_action_meter *mtr;
7818 struct mlx5_flow_tbl_resource *tbl;
7819 uint32_t port_id = 0;
7820 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
7821 int action_type = actions->type;
7822 const struct rte_flow_action *found_action = NULL;
7823 struct mlx5_flow_meter *fm = NULL;
7825 switch (action_type) {
7826 case RTE_FLOW_ACTION_TYPE_VOID:
7828 case RTE_FLOW_ACTION_TYPE_PORT_ID:
7829 if (flow_dv_translate_action_port_id(dev, action,
7832 memset(&port_id_resource, 0, sizeof(port_id_resource));
7833 port_id_resource.port_id = port_id;
7834 if (flow_dv_port_id_action_resource_register
7835 (dev, &port_id_resource, dev_flow, error))
7837 MLX5_ASSERT(!handle->rix_port_id_action);
7838 dev_flow->dv.actions[actions_n++] =
7839 dev_flow->dv.port_id_action->action;
7840 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
7841 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
7843 case RTE_FLOW_ACTION_TYPE_FLAG:
7844 action_flags |= MLX5_FLOW_ACTION_FLAG;
7845 dev_flow->handle->mark = 1;
7846 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7847 struct rte_flow_action_mark mark = {
7848 .id = MLX5_FLOW_MARK_DEFAULT,
7851 if (flow_dv_convert_action_mark(dev, &mark,
7855 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
7858 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
7860 * Only one FLAG or MARK is supported per device flow
7861 * right now. So the pointer to the tag resource must be
7862 * zero before the register process.
7864 MLX5_ASSERT(!handle->dvh.rix_tag);
7865 if (flow_dv_tag_resource_register(dev, tag_be,
7868 MLX5_ASSERT(dev_flow->dv.tag_resource);
7869 dev_flow->dv.actions[actions_n++] =
7870 dev_flow->dv.tag_resource->action;
7872 case RTE_FLOW_ACTION_TYPE_MARK:
7873 action_flags |= MLX5_FLOW_ACTION_MARK;
7874 dev_flow->handle->mark = 1;
7875 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7876 const struct rte_flow_action_mark *mark =
7877 (const struct rte_flow_action_mark *)
7880 if (flow_dv_convert_action_mark(dev, mark,
7884 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
7888 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
7889 /* Legacy (non-extensive) MARK action. */
7890 tag_be = mlx5_flow_mark_set
7891 (((const struct rte_flow_action_mark *)
7892 (actions->conf))->id);
7893 MLX5_ASSERT(!handle->dvh.rix_tag);
7894 if (flow_dv_tag_resource_register(dev, tag_be,
7897 MLX5_ASSERT(dev_flow->dv.tag_resource);
7898 dev_flow->dv.actions[actions_n++] =
7899 dev_flow->dv.tag_resource->action;
7901 case RTE_FLOW_ACTION_TYPE_SET_META:
7902 if (flow_dv_convert_action_set_meta
7903 (dev, mhdr_res, attr,
7904 (const struct rte_flow_action_set_meta *)
7905 actions->conf, error))
7907 action_flags |= MLX5_FLOW_ACTION_SET_META;
7909 case RTE_FLOW_ACTION_TYPE_SET_TAG:
7910 if (flow_dv_convert_action_set_tag
7912 (const struct rte_flow_action_set_tag *)
7913 actions->conf, error))
7915 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7917 case RTE_FLOW_ACTION_TYPE_DROP:
7918 action_flags |= MLX5_FLOW_ACTION_DROP;
7919 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
7921 case RTE_FLOW_ACTION_TYPE_QUEUE:
7922 queue = actions->conf;
7923 rss_desc->queue_num = 1;
7924 rss_desc->queue[0] = queue->index;
7925 action_flags |= MLX5_FLOW_ACTION_QUEUE;
7926 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
7928 case RTE_FLOW_ACTION_TYPE_RSS:
7929 rss = actions->conf;
7930 memcpy(rss_desc->queue, rss->queue,
7931 rss->queue_num * sizeof(uint16_t));
7932 rss_desc->queue_num = rss->queue_num;
7933 /* NULL RSS key indicates default RSS key. */
7934 rss_key = !rss->key ? rss_hash_default_key : rss->key;
7935 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
7937 * rss->level and rss.types should be set in advance
7938 * when expanding items for RSS.
7940 action_flags |= MLX5_FLOW_ACTION_RSS;
7941 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
7943 case RTE_FLOW_ACTION_TYPE_AGE:
7944 case RTE_FLOW_ACTION_TYPE_COUNT:
7945 if (!dev_conf->devx) {
7946 return rte_flow_error_set
7948 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7950 "count action not supported");
7952 /* Save information first, will apply later. */
7953 if (actions->type == RTE_FLOW_ACTION_TYPE_COUNT)
7954 count = action->conf;
7957 action_flags |= MLX5_FLOW_ACTION_COUNT;
7959 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
7960 dev_flow->dv.actions[actions_n++] =
7961 priv->sh->pop_vlan_action;
7962 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
7964 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
7965 if (!(action_flags &
7966 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
7967 flow_dev_get_vlan_info_from_items(items, &vlan);
7968 vlan.eth_proto = rte_be_to_cpu_16
7969 ((((const struct rte_flow_action_of_push_vlan *)
7970 actions->conf)->ethertype));
7971 found_action = mlx5_flow_find_action
7973 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
7975 mlx5_update_vlan_vid_pcp(found_action, &vlan);
7976 found_action = mlx5_flow_find_action
7978 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
7980 mlx5_update_vlan_vid_pcp(found_action, &vlan);
7981 if (flow_dv_create_action_push_vlan
7982 (dev, attr, &vlan, dev_flow, error))
7984 dev_flow->dv.actions[actions_n++] =
7985 dev_flow->dv.push_vlan_res->action;
7986 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
7988 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
7989 /* of_vlan_push action handled this action */
7990 MLX5_ASSERT(action_flags &
7991 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
7993 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
7994 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
7996 flow_dev_get_vlan_info_from_items(items, &vlan);
7997 mlx5_update_vlan_vid_pcp(actions, &vlan);
7998 /* If no VLAN push - this is a modify header action */
7999 if (flow_dv_convert_action_modify_vlan_vid
8000 (mhdr_res, actions, error))
8002 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
8004 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
8005 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
8006 if (flow_dv_create_action_l2_encap(dev, actions,
8011 dev_flow->dv.actions[actions_n++] =
8012 dev_flow->dv.encap_decap->verbs_action;
8013 action_flags |= MLX5_FLOW_ACTION_ENCAP;
8015 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
8016 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
8017 if (flow_dv_create_action_l2_decap(dev, dev_flow,
8021 dev_flow->dv.actions[actions_n++] =
8022 dev_flow->dv.encap_decap->verbs_action;
8023 action_flags |= MLX5_FLOW_ACTION_DECAP;
8025 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
8026 /* Handle encap with preceding decap. */
8027 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
8028 if (flow_dv_create_action_raw_encap
8029 (dev, actions, dev_flow, attr, error))
8031 dev_flow->dv.actions[actions_n++] =
8032 dev_flow->dv.encap_decap->verbs_action;
8034 /* Handle encap without preceding decap. */
8035 if (flow_dv_create_action_l2_encap
8036 (dev, actions, dev_flow, attr->transfer,
8039 dev_flow->dv.actions[actions_n++] =
8040 dev_flow->dv.encap_decap->verbs_action;
8042 action_flags |= MLX5_FLOW_ACTION_ENCAP;
8044 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
8045 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
8047 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
8048 if (flow_dv_create_action_l2_decap
8049 (dev, dev_flow, attr->transfer, error))
8051 dev_flow->dv.actions[actions_n++] =
8052 dev_flow->dv.encap_decap->verbs_action;
8054 /* If decap is followed by encap, handle it at encap. */
8055 action_flags |= MLX5_FLOW_ACTION_DECAP;
8057 case RTE_FLOW_ACTION_TYPE_JUMP:
8058 jump_data = action->conf;
8059 ret = mlx5_flow_group_to_table(attr, dev_flow->external,
8061 !!priv->fdb_def_rule,
8065 tbl = flow_dv_tbl_resource_get(dev, table,
8067 attr->transfer, error);
8069 return rte_flow_error_set
8071 RTE_FLOW_ERROR_TYPE_ACTION,
8073 "cannot create jump action.");
8074 if (flow_dv_jump_tbl_resource_register
8075 (dev, tbl, dev_flow, error)) {
8076 flow_dv_tbl_resource_release(dev, tbl);
8077 return rte_flow_error_set
8079 RTE_FLOW_ERROR_TYPE_ACTION,
8081 "cannot create jump action.");
8083 dev_flow->dv.actions[actions_n++] =
8084 dev_flow->dv.jump->action;
8085 action_flags |= MLX5_FLOW_ACTION_JUMP;
8086 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
8088 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
8089 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
8090 if (flow_dv_convert_action_modify_mac
8091 (mhdr_res, actions, error))
8093 action_flags |= actions->type ==
8094 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
8095 MLX5_FLOW_ACTION_SET_MAC_SRC :
8096 MLX5_FLOW_ACTION_SET_MAC_DST;
8098 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
8099 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
8100 if (flow_dv_convert_action_modify_ipv4
8101 (mhdr_res, actions, error))
8103 action_flags |= actions->type ==
8104 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
8105 MLX5_FLOW_ACTION_SET_IPV4_SRC :
8106 MLX5_FLOW_ACTION_SET_IPV4_DST;
8108 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
8109 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
8110 if (flow_dv_convert_action_modify_ipv6
8111 (mhdr_res, actions, error))
8113 action_flags |= actions->type ==
8114 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
8115 MLX5_FLOW_ACTION_SET_IPV6_SRC :
8116 MLX5_FLOW_ACTION_SET_IPV6_DST;
8118 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
8119 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
8120 if (flow_dv_convert_action_modify_tp
8121 (mhdr_res, actions, items,
8122 &flow_attr, dev_flow, !!(action_flags &
8123 MLX5_FLOW_ACTION_DECAP), error))
8125 action_flags |= actions->type ==
8126 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
8127 MLX5_FLOW_ACTION_SET_TP_SRC :
8128 MLX5_FLOW_ACTION_SET_TP_DST;
8130 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
8131 if (flow_dv_convert_action_modify_dec_ttl
8132 (mhdr_res, items, &flow_attr, dev_flow,
8134 MLX5_FLOW_ACTION_DECAP), error))
8136 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
8138 case RTE_FLOW_ACTION_TYPE_SET_TTL:
8139 if (flow_dv_convert_action_modify_ttl
8140 (mhdr_res, actions, items, &flow_attr,
8141 dev_flow, !!(action_flags &
8142 MLX5_FLOW_ACTION_DECAP), error))
8144 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
8146 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
8147 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
8148 if (flow_dv_convert_action_modify_tcp_seq
8149 (mhdr_res, actions, error))
8151 action_flags |= actions->type ==
8152 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
8153 MLX5_FLOW_ACTION_INC_TCP_SEQ :
8154 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
8157 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
8158 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
8159 if (flow_dv_convert_action_modify_tcp_ack
8160 (mhdr_res, actions, error))
8162 action_flags |= actions->type ==
8163 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
8164 MLX5_FLOW_ACTION_INC_TCP_ACK :
8165 MLX5_FLOW_ACTION_DEC_TCP_ACK;
8167 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
8168 if (flow_dv_convert_action_set_reg
8169 (mhdr_res, actions, error))
8171 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
8173 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
8174 if (flow_dv_convert_action_copy_mreg
8175 (dev, mhdr_res, actions, error))
8177 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
8179 case RTE_FLOW_ACTION_TYPE_METER:
8180 mtr = actions->conf;
8182 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
8185 return rte_flow_error_set(error,
8187 RTE_FLOW_ERROR_TYPE_ACTION,
8190 "or invalid parameters");
8191 flow->meter = fm->idx;
8193 /* Set the meter action. */
8195 fm = mlx5_ipool_get(priv->sh->ipool
8196 [MLX5_IPOOL_MTR], flow->meter);
8198 return rte_flow_error_set(error,
8200 RTE_FLOW_ERROR_TYPE_ACTION,
8203 "or invalid parameters");
8205 dev_flow->dv.actions[actions_n++] =
8206 fm->mfts->meter_action;
8207 action_flags |= MLX5_FLOW_ACTION_METER;
8209 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
8210 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
8213 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
8215 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
8216 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
8219 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
8221 case RTE_FLOW_ACTION_TYPE_END:
8223 if (mhdr_res->actions_num) {
8224 /* create modify action if needed. */
8225 if (flow_dv_modify_hdr_resource_register
8226 (dev, mhdr_res, dev_flow, error))
8228 dev_flow->dv.actions[modify_action_position] =
8229 handle->dvh.modify_hdr->verbs_action;
8231 if (action_flags & MLX5_FLOW_ACTION_COUNT) {
8233 flow_dv_translate_create_counter(dev,
8234 dev_flow, count, age);
8237 return rte_flow_error_set
8239 RTE_FLOW_ERROR_TYPE_ACTION,
8241 "cannot create counter"
8243 dev_flow->dv.actions[actions_n++] =
8244 (flow_dv_counter_get_by_idx(dev,
8245 flow->counter, NULL))->action;
8251 if (mhdr_res->actions_num &&
8252 modify_action_position == UINT32_MAX)
8253 modify_action_position = actions_n++;
8255 dev_flow->dv.actions_n = actions_n;
8256 dev_flow->act_flags = action_flags;
8257 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
8258 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
8259 int item_type = items->type;
8261 switch (item_type) {
8262 case RTE_FLOW_ITEM_TYPE_PORT_ID:
8263 flow_dv_translate_item_port_id(dev, match_mask,
8264 match_value, items);
8265 last_item = MLX5_FLOW_ITEM_PORT_ID;
8267 case RTE_FLOW_ITEM_TYPE_ETH:
8268 flow_dv_translate_item_eth(match_mask, match_value,
8270 dev_flow->dv.group);
8271 matcher.priority = MLX5_PRIORITY_MAP_L2;
8272 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
8273 MLX5_FLOW_LAYER_OUTER_L2;
8275 case RTE_FLOW_ITEM_TYPE_VLAN:
8276 flow_dv_translate_item_vlan(dev_flow,
8277 match_mask, match_value,
8279 dev_flow->dv.group);
8280 matcher.priority = MLX5_PRIORITY_MAP_L2;
8281 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
8282 MLX5_FLOW_LAYER_INNER_VLAN) :
8283 (MLX5_FLOW_LAYER_OUTER_L2 |
8284 MLX5_FLOW_LAYER_OUTER_VLAN);
8286 case RTE_FLOW_ITEM_TYPE_IPV4:
8287 mlx5_flow_tunnel_ip_check(items, next_protocol,
8288 &item_flags, &tunnel);
8289 flow_dv_translate_item_ipv4(match_mask, match_value,
8290 items, item_flags, tunnel,
8291 dev_flow->dv.group);
8292 matcher.priority = MLX5_PRIORITY_MAP_L3;
8293 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
8294 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
8295 if (items->mask != NULL &&
8296 ((const struct rte_flow_item_ipv4 *)
8297 items->mask)->hdr.next_proto_id) {
8299 ((const struct rte_flow_item_ipv4 *)
8300 (items->spec))->hdr.next_proto_id;
8302 ((const struct rte_flow_item_ipv4 *)
8303 (items->mask))->hdr.next_proto_id;
8305 /* Reset for inner layer. */
8306 next_protocol = 0xff;
8309 case RTE_FLOW_ITEM_TYPE_IPV6:
8310 mlx5_flow_tunnel_ip_check(items, next_protocol,
8311 &item_flags, &tunnel);
8312 flow_dv_translate_item_ipv6(match_mask, match_value,
8313 items, item_flags, tunnel,
8314 dev_flow->dv.group);
8315 matcher.priority = MLX5_PRIORITY_MAP_L3;
8316 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
8317 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
8318 if (items->mask != NULL &&
8319 ((const struct rte_flow_item_ipv6 *)
8320 items->mask)->hdr.proto) {
8322 ((const struct rte_flow_item_ipv6 *)
8323 items->spec)->hdr.proto;
8325 ((const struct rte_flow_item_ipv6 *)
8326 items->mask)->hdr.proto;
8328 /* Reset for inner layer. */
8329 next_protocol = 0xff;
8332 case RTE_FLOW_ITEM_TYPE_TCP:
8333 flow_dv_translate_item_tcp(match_mask, match_value,
8335 matcher.priority = MLX5_PRIORITY_MAP_L4;
8336 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
8337 MLX5_FLOW_LAYER_OUTER_L4_TCP;
8339 case RTE_FLOW_ITEM_TYPE_UDP:
8340 flow_dv_translate_item_udp(match_mask, match_value,
8342 matcher.priority = MLX5_PRIORITY_MAP_L4;
8343 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
8344 MLX5_FLOW_LAYER_OUTER_L4_UDP;
8346 case RTE_FLOW_ITEM_TYPE_GRE:
8347 flow_dv_translate_item_gre(match_mask, match_value,
8349 matcher.priority = rss_desc->level >= 2 ?
8350 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8351 last_item = MLX5_FLOW_LAYER_GRE;
8353 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
8354 flow_dv_translate_item_gre_key(match_mask,
8355 match_value, items);
8356 last_item = MLX5_FLOW_LAYER_GRE_KEY;
8358 case RTE_FLOW_ITEM_TYPE_NVGRE:
8359 flow_dv_translate_item_nvgre(match_mask, match_value,
8361 matcher.priority = rss_desc->level >= 2 ?
8362 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8363 last_item = MLX5_FLOW_LAYER_GRE;
8365 case RTE_FLOW_ITEM_TYPE_VXLAN:
8366 flow_dv_translate_item_vxlan(match_mask, match_value,
8368 matcher.priority = rss_desc->level >= 2 ?
8369 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8370 last_item = MLX5_FLOW_LAYER_VXLAN;
8372 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
8373 flow_dv_translate_item_vxlan_gpe(match_mask,
8376 matcher.priority = rss_desc->level >= 2 ?
8377 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8378 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
8380 case RTE_FLOW_ITEM_TYPE_GENEVE:
8381 flow_dv_translate_item_geneve(match_mask, match_value,
8383 matcher.priority = rss_desc->level >= 2 ?
8384 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8385 last_item = MLX5_FLOW_LAYER_GENEVE;
8387 case RTE_FLOW_ITEM_TYPE_MPLS:
8388 flow_dv_translate_item_mpls(match_mask, match_value,
8389 items, last_item, tunnel);
8390 matcher.priority = rss_desc->level >= 2 ?
8391 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8392 last_item = MLX5_FLOW_LAYER_MPLS;
8394 case RTE_FLOW_ITEM_TYPE_MARK:
8395 flow_dv_translate_item_mark(dev, match_mask,
8396 match_value, items);
8397 last_item = MLX5_FLOW_ITEM_MARK;
8399 case RTE_FLOW_ITEM_TYPE_META:
8400 flow_dv_translate_item_meta(dev, match_mask,
8401 match_value, attr, items);
8402 last_item = MLX5_FLOW_ITEM_METADATA;
8404 case RTE_FLOW_ITEM_TYPE_ICMP:
8405 flow_dv_translate_item_icmp(match_mask, match_value,
8407 last_item = MLX5_FLOW_LAYER_ICMP;
8409 case RTE_FLOW_ITEM_TYPE_ICMP6:
8410 flow_dv_translate_item_icmp6(match_mask, match_value,
8412 last_item = MLX5_FLOW_LAYER_ICMP6;
8414 case RTE_FLOW_ITEM_TYPE_TAG:
8415 flow_dv_translate_item_tag(dev, match_mask,
8416 match_value, items);
8417 last_item = MLX5_FLOW_ITEM_TAG;
8419 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
8420 flow_dv_translate_mlx5_item_tag(dev, match_mask,
8421 match_value, items);
8422 last_item = MLX5_FLOW_ITEM_TAG;
8424 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
8425 flow_dv_translate_item_tx_queue(dev, match_mask,
8428 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
8430 case RTE_FLOW_ITEM_TYPE_GTP:
8431 flow_dv_translate_item_gtp(match_mask, match_value,
8433 matcher.priority = rss_desc->level >= 2 ?
8434 MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
8435 last_item = MLX5_FLOW_LAYER_GTP;
8440 item_flags |= last_item;
8443 * When E-Switch mode is enabled, we have two cases where we need to
8444 * set the source port manually.
8445 * The first one, is in case of Nic steering rule, and the second is
8446 * E-Switch rule where no port_id item was found. In both cases
8447 * the source port is set according the current port in use.
8449 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
8450 (priv->representor || priv->master)) {
8451 if (flow_dv_translate_item_port_id(dev, match_mask,
8455 #ifdef RTE_LIBRTE_MLX5_DEBUG
8456 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
8457 dev_flow->dv.value.buf));
8460 * Layers may be already initialized from prefix flow if this dev_flow
8461 * is the suffix flow.
8463 handle->layers |= item_flags;
8464 if (action_flags & MLX5_FLOW_ACTION_RSS)
8465 flow_dv_hashfields_set(dev_flow, rss_desc);
8466 /* Register matcher. */
8467 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
8469 matcher.priority = mlx5_flow_adjust_priority(dev, priority,
8471 /* reserved field no needs to be set to 0 here. */
8472 tbl_key.domain = attr->transfer;
8473 tbl_key.direction = attr->egress;
8474 tbl_key.table_id = dev_flow->dv.group;
8475 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
8481 * Apply the flow to the NIC, lock free,
8482 * (mutex should be acquired by caller).
8485 * Pointer to the Ethernet device structure.
8486 * @param[in, out] flow
8487 * Pointer to flow structure.
8489 * Pointer to error structure.
8492 * 0 on success, a negative errno value otherwise and rte_errno is set.
8495 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
8496 struct rte_flow_error *error)
8498 struct mlx5_flow_dv_workspace *dv;
8499 struct mlx5_flow_handle *dh;
8500 struct mlx5_flow_handle_dv *dv_h;
8501 struct mlx5_flow *dev_flow;
8502 struct mlx5_priv *priv = dev->data->dev_private;
8503 uint32_t handle_idx;
8508 for (idx = priv->flow_idx - 1; idx >= priv->flow_nested_idx; idx--) {
8509 dev_flow = &((struct mlx5_flow *)priv->inter_flows)[idx];
8511 dh = dev_flow->handle;
8514 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8516 dv->actions[n++] = priv->sh->esw_drop_action;
8518 struct mlx5_hrxq *drop_hrxq;
8519 drop_hrxq = mlx5_hrxq_drop_new(dev);
8523 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8525 "cannot get drop hash queue");
8529 * Drop queues will be released by the specify
8530 * mlx5_hrxq_drop_release() function. Assign
8531 * the special index to hrxq to mark the queue
8532 * has been allocated.
8534 dh->rix_hrxq = UINT32_MAX;
8535 dv->actions[n++] = drop_hrxq->action;
8537 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8538 struct mlx5_hrxq *hrxq;
8540 struct mlx5_flow_rss_desc *rss_desc =
8541 &((struct mlx5_flow_rss_desc *)priv->rss_desc)
8542 [!!priv->flow_nested_idx];
8544 MLX5_ASSERT(rss_desc->queue_num);
8545 hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
8546 MLX5_RSS_HASH_KEY_LEN,
8547 dev_flow->hash_fields,
8549 rss_desc->queue_num);
8551 hrxq_idx = mlx5_hrxq_new
8552 (dev, rss_desc->key,
8553 MLX5_RSS_HASH_KEY_LEN,
8554 dev_flow->hash_fields,
8556 rss_desc->queue_num,
8558 MLX5_FLOW_LAYER_TUNNEL));
8560 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8565 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8566 "cannot get hash queue");
8569 dh->rix_hrxq = hrxq_idx;
8570 dv->actions[n++] = hrxq->action;
8573 mlx5_glue->dv_create_flow(dv_h->matcher->matcher_object,
8574 (void *)&dv->value, n,
8577 rte_flow_error_set(error, errno,
8578 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8580 "hardware refuses to create flow");
8583 if (priv->vmwa_context &&
8584 dh->vf_vlan.tag && !dh->vf_vlan.created) {
8586 * The rule contains the VLAN pattern.
8587 * For VF we are going to create VLAN
8588 * interface to make hypervisor set correct
8589 * e-Switch vport context.
8591 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
8596 err = rte_errno; /* Save rte_errno before cleanup. */
8597 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
8598 handle_idx, dh, next) {
8599 /* hrxq is union, don't clear it if the flag is not set. */
8601 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8602 mlx5_hrxq_drop_release(dev);
8604 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8605 mlx5_hrxq_release(dev, dh->rix_hrxq);
8609 if (dh->vf_vlan.tag && dh->vf_vlan.created)
8610 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8612 rte_errno = err; /* Restore rte_errno. */
8617 * Release the flow matcher.
8620 * Pointer to Ethernet device.
8622 * Pointer to mlx5_flow_handle.
8625 * 1 while a reference on it exists, 0 when freed.
8628 flow_dv_matcher_release(struct rte_eth_dev *dev,
8629 struct mlx5_flow_handle *handle)
8631 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
8633 MLX5_ASSERT(matcher->matcher_object);
8634 DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
8635 dev->data->port_id, (void *)matcher,
8636 rte_atomic32_read(&matcher->refcnt));
8637 if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
8638 claim_zero(mlx5_glue->dv_destroy_flow_matcher
8639 (matcher->matcher_object));
8640 LIST_REMOVE(matcher, next);
8641 /* table ref-- in release interface. */
8642 flow_dv_tbl_resource_release(dev, matcher->tbl);
8644 DRV_LOG(DEBUG, "port %u matcher %p: removed",
8645 dev->data->port_id, (void *)matcher);
8652 * Release an encap/decap resource.
8655 * Pointer to Ethernet device.
8657 * Pointer to mlx5_flow_handle.
8660 * 1 while a reference on it exists, 0 when freed.
8663 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
8664 struct mlx5_flow_handle *handle)
8666 struct mlx5_priv *priv = dev->data->dev_private;
8667 uint32_t idx = handle->dvh.rix_encap_decap;
8668 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
8670 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
8672 if (!cache_resource)
8674 MLX5_ASSERT(cache_resource->verbs_action);
8675 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
8676 (void *)cache_resource,
8677 rte_atomic32_read(&cache_resource->refcnt));
8678 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8679 claim_zero(mlx5_glue->destroy_flow_action
8680 (cache_resource->verbs_action));
8681 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
8682 &priv->sh->encaps_decaps, idx,
8683 cache_resource, next);
8684 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
8685 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
8686 (void *)cache_resource);
8693 * Release an jump to table action resource.
8696 * Pointer to Ethernet device.
8698 * Pointer to mlx5_flow_handle.
8701 * 1 while a reference on it exists, 0 when freed.
8704 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
8705 struct mlx5_flow_handle *handle)
8707 struct mlx5_priv *priv = dev->data->dev_private;
8708 struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
8709 struct mlx5_flow_tbl_data_entry *tbl_data;
8711 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
8715 cache_resource = &tbl_data->jump;
8716 MLX5_ASSERT(cache_resource->action);
8717 DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
8718 (void *)cache_resource,
8719 rte_atomic32_read(&cache_resource->refcnt));
8720 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8721 claim_zero(mlx5_glue->destroy_flow_action
8722 (cache_resource->action));
8723 /* jump action memory free is inside the table release. */
8724 flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
8725 DRV_LOG(DEBUG, "jump table resource %p: removed",
8726 (void *)cache_resource);
8733 * Release a modify-header resource.
8736 * Pointer to mlx5_flow_handle.
8739 * 1 while a reference on it exists, 0 when freed.
8742 flow_dv_modify_hdr_resource_release(struct mlx5_flow_handle *handle)
8744 struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
8745 handle->dvh.modify_hdr;
8747 MLX5_ASSERT(cache_resource->verbs_action);
8748 DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
8749 (void *)cache_resource,
8750 rte_atomic32_read(&cache_resource->refcnt));
8751 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8752 claim_zero(mlx5_glue->destroy_flow_action
8753 (cache_resource->verbs_action));
8754 LIST_REMOVE(cache_resource, next);
8755 rte_free(cache_resource);
8756 DRV_LOG(DEBUG, "modify-header resource %p: removed",
8757 (void *)cache_resource);
8764 * Release port ID action resource.
8767 * Pointer to Ethernet device.
8769 * Pointer to mlx5_flow_handle.
8772 * 1 while a reference on it exists, 0 when freed.
8775 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
8776 struct mlx5_flow_handle *handle)
8778 struct mlx5_priv *priv = dev->data->dev_private;
8779 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
8780 uint32_t idx = handle->rix_port_id_action;
8782 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
8784 if (!cache_resource)
8786 MLX5_ASSERT(cache_resource->action);
8787 DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
8788 (void *)cache_resource,
8789 rte_atomic32_read(&cache_resource->refcnt));
8790 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8791 claim_zero(mlx5_glue->destroy_flow_action
8792 (cache_resource->action));
8793 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
8794 &priv->sh->port_id_action_list, idx,
8795 cache_resource, next);
8796 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
8797 DRV_LOG(DEBUG, "port id action resource %p: removed",
8798 (void *)cache_resource);
8805 * Release push vlan action resource.
8808 * Pointer to Ethernet device.
8810 * Pointer to mlx5_flow_handle.
8813 * 1 while a reference on it exists, 0 when freed.
8816 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
8817 struct mlx5_flow_handle *handle)
8819 struct mlx5_priv *priv = dev->data->dev_private;
8820 uint32_t idx = handle->dvh.rix_push_vlan;
8821 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
8823 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
8825 if (!cache_resource)
8827 MLX5_ASSERT(cache_resource->action);
8828 DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
8829 (void *)cache_resource,
8830 rte_atomic32_read(&cache_resource->refcnt));
8831 if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
8832 claim_zero(mlx5_glue->destroy_flow_action
8833 (cache_resource->action));
8834 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
8835 &priv->sh->push_vlan_action_list, idx,
8836 cache_resource, next);
8837 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
8838 DRV_LOG(DEBUG, "push vlan action resource %p: removed",
8839 (void *)cache_resource);
8846 * Release the fate resource.
8849 * Pointer to Ethernet device.
8851 * Pointer to mlx5_flow_handle.
8854 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
8855 struct mlx5_flow_handle *handle)
8857 if (!handle->rix_fate)
8859 if (handle->fate_action == MLX5_FLOW_FATE_DROP)
8860 mlx5_hrxq_drop_release(dev);
8861 else if (handle->fate_action == MLX5_FLOW_FATE_QUEUE)
8862 mlx5_hrxq_release(dev, handle->rix_hrxq);
8863 else if (handle->fate_action == MLX5_FLOW_FATE_JUMP)
8864 flow_dv_jump_tbl_resource_release(dev, handle);
8865 else if (handle->fate_action == MLX5_FLOW_FATE_PORT_ID)
8866 flow_dv_port_id_action_resource_release(dev, handle);
8868 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
8869 handle->rix_fate = 0;
8873 * Remove the flow from the NIC but keeps it in memory.
8874 * Lock free, (mutex should be acquired by caller).
8877 * Pointer to Ethernet device.
8878 * @param[in, out] flow
8879 * Pointer to flow structure.
8882 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
8884 struct mlx5_flow_handle *dh;
8885 uint32_t handle_idx;
8886 struct mlx5_priv *priv = dev->data->dev_private;
8890 handle_idx = flow->dev_handles;
8891 while (handle_idx) {
8892 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8897 claim_zero(mlx5_glue->dv_destroy_flow(dh->ib_flow));
8900 if (dh->fate_action == MLX5_FLOW_FATE_DROP ||
8901 dh->fate_action == MLX5_FLOW_FATE_QUEUE)
8902 flow_dv_fate_resource_release(dev, dh);
8903 if (dh->vf_vlan.tag && dh->vf_vlan.created)
8904 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8905 handle_idx = dh->next.next;
8910 * Remove the flow from the NIC and the memory.
8911 * Lock free, (mutex should be acquired by caller).
8914 * Pointer to the Ethernet device structure.
8915 * @param[in, out] flow
8916 * Pointer to flow structure.
8919 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
8921 struct mlx5_flow_handle *dev_handle;
8922 struct mlx5_priv *priv = dev->data->dev_private;
8926 __flow_dv_remove(dev, flow);
8927 if (flow->counter) {
8928 flow_dv_counter_release(dev, flow->counter);
8932 struct mlx5_flow_meter *fm;
8934 fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
8937 mlx5_flow_meter_detach(fm);
8940 while (flow->dev_handles) {
8941 uint32_t tmp_idx = flow->dev_handles;
8943 dev_handle = mlx5_ipool_get(priv->sh->ipool
8944 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
8947 flow->dev_handles = dev_handle->next.next;
8948 if (dev_handle->dvh.matcher)
8949 flow_dv_matcher_release(dev, dev_handle);
8950 if (dev_handle->dvh.rix_encap_decap)
8951 flow_dv_encap_decap_resource_release(dev, dev_handle);
8952 if (dev_handle->dvh.modify_hdr)
8953 flow_dv_modify_hdr_resource_release(dev_handle);
8954 if (dev_handle->dvh.rix_push_vlan)
8955 flow_dv_push_vlan_action_resource_release(dev,
8957 if (dev_handle->dvh.rix_tag)
8958 flow_dv_tag_release(dev,
8959 dev_handle->dvh.rix_tag);
8960 flow_dv_fate_resource_release(dev, dev_handle);
8961 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8967 * Query a dv flow rule for its statistics via devx.
8970 * Pointer to Ethernet device.
8972 * Pointer to the sub flow.
8974 * data retrieved by the query.
8976 * Perform verbose error reporting if not NULL.
8979 * 0 on success, a negative errno value otherwise and rte_errno is set.
8982 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
8983 void *data, struct rte_flow_error *error)
8985 struct mlx5_priv *priv = dev->data->dev_private;
8986 struct rte_flow_query_count *qc = data;
8988 if (!priv->config.devx)
8989 return rte_flow_error_set(error, ENOTSUP,
8990 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8992 "counters are not supported");
8993 if (flow->counter) {
8994 uint64_t pkts, bytes;
8995 struct mlx5_flow_counter *cnt;
8997 cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
8999 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
9003 return rte_flow_error_set(error, -err,
9004 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9005 NULL, "cannot read counters");
9008 qc->hits = pkts - cnt->hits;
9009 qc->bytes = bytes - cnt->bytes;
9016 return rte_flow_error_set(error, EINVAL,
9017 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9019 "counters are not available");
9025 * @see rte_flow_query()
9029 flow_dv_query(struct rte_eth_dev *dev,
9030 struct rte_flow *flow __rte_unused,
9031 const struct rte_flow_action *actions __rte_unused,
9032 void *data __rte_unused,
9033 struct rte_flow_error *error __rte_unused)
9037 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
9038 switch (actions->type) {
9039 case RTE_FLOW_ACTION_TYPE_VOID:
9041 case RTE_FLOW_ACTION_TYPE_COUNT:
9042 ret = flow_dv_query_count(dev, flow, data, error);
9045 return rte_flow_error_set(error, ENOTSUP,
9046 RTE_FLOW_ERROR_TYPE_ACTION,
9048 "action not supported");
9055 * Destroy the meter table set.
9056 * Lock free, (mutex should be acquired by caller).
9059 * Pointer to Ethernet device.
9061 * Pointer to the meter table set.
9067 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
9068 struct mlx5_meter_domains_infos *tbl)
9070 struct mlx5_priv *priv = dev->data->dev_private;
9071 struct mlx5_meter_domains_infos *mtd =
9072 (struct mlx5_meter_domains_infos *)tbl;
9074 if (!mtd || !priv->config.dv_flow_en)
9076 if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
9077 claim_zero(mlx5_glue->dv_destroy_flow
9078 (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
9079 if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
9080 claim_zero(mlx5_glue->dv_destroy_flow
9081 (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
9082 if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
9083 claim_zero(mlx5_glue->dv_destroy_flow
9084 (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
9085 if (mtd->egress.color_matcher)
9086 claim_zero(mlx5_glue->dv_destroy_flow_matcher
9087 (mtd->egress.color_matcher));
9088 if (mtd->egress.any_matcher)
9089 claim_zero(mlx5_glue->dv_destroy_flow_matcher
9090 (mtd->egress.any_matcher));
9091 if (mtd->egress.tbl)
9092 flow_dv_tbl_resource_release(dev, mtd->egress.tbl);
9093 if (mtd->egress.sfx_tbl)
9094 flow_dv_tbl_resource_release(dev, mtd->egress.sfx_tbl);
9095 if (mtd->ingress.color_matcher)
9096 claim_zero(mlx5_glue->dv_destroy_flow_matcher
9097 (mtd->ingress.color_matcher));
9098 if (mtd->ingress.any_matcher)
9099 claim_zero(mlx5_glue->dv_destroy_flow_matcher
9100 (mtd->ingress.any_matcher));
9101 if (mtd->ingress.tbl)
9102 flow_dv_tbl_resource_release(dev, mtd->ingress.tbl);
9103 if (mtd->ingress.sfx_tbl)
9104 flow_dv_tbl_resource_release(dev, mtd->ingress.sfx_tbl);
9105 if (mtd->transfer.color_matcher)
9106 claim_zero(mlx5_glue->dv_destroy_flow_matcher
9107 (mtd->transfer.color_matcher));
9108 if (mtd->transfer.any_matcher)
9109 claim_zero(mlx5_glue->dv_destroy_flow_matcher
9110 (mtd->transfer.any_matcher));
9111 if (mtd->transfer.tbl)
9112 flow_dv_tbl_resource_release(dev, mtd->transfer.tbl);
9113 if (mtd->transfer.sfx_tbl)
9114 flow_dv_tbl_resource_release(dev, mtd->transfer.sfx_tbl);
9116 claim_zero(mlx5_glue->destroy_flow_action(mtd->drop_actn));
9121 /* Number of meter flow actions, count and jump or count and drop. */
9122 #define METER_ACTIONS 2
9125 * Create specify domain meter table and suffix table.
9128 * Pointer to Ethernet device.
9129 * @param[in,out] mtb
9130 * Pointer to DV meter table set.
9133 * @param[in] transfer
9135 * @param[in] color_reg_c_idx
9136 * Reg C index for color match.
9139 * 0 on success, -1 otherwise and rte_errno is set.
9142 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
9143 struct mlx5_meter_domains_infos *mtb,
9144 uint8_t egress, uint8_t transfer,
9145 uint32_t color_reg_c_idx)
9147 struct mlx5_priv *priv = dev->data->dev_private;
9148 struct mlx5_ibv_shared *sh = priv->sh;
9149 struct mlx5_flow_dv_match_params mask = {
9150 .size = sizeof(mask.buf),
9152 struct mlx5_flow_dv_match_params value = {
9153 .size = sizeof(value.buf),
9155 struct mlx5dv_flow_matcher_attr dv_attr = {
9156 .type = IBV_FLOW_ATTR_NORMAL,
9158 .match_criteria_enable = 0,
9159 .match_mask = (void *)&mask,
9161 void *actions[METER_ACTIONS];
9162 struct mlx5_meter_domain_info *dtb;
9163 struct rte_flow_error error;
9167 dtb = &mtb->transfer;
9171 dtb = &mtb->ingress;
9172 /* Create the meter table with METER level. */
9173 dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
9174 egress, transfer, &error);
9176 DRV_LOG(ERR, "Failed to create meter policer table.");
9179 /* Create the meter suffix table with SUFFIX level. */
9180 dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
9181 MLX5_FLOW_TABLE_LEVEL_SUFFIX,
9182 egress, transfer, &error);
9183 if (!dtb->sfx_tbl) {
9184 DRV_LOG(ERR, "Failed to create meter suffix table.");
9187 /* Create matchers, Any and Color. */
9188 dv_attr.priority = 3;
9189 dv_attr.match_criteria_enable = 0;
9190 dtb->any_matcher = mlx5_glue->dv_create_flow_matcher(sh->ctx,
9193 if (!dtb->any_matcher) {
9194 DRV_LOG(ERR, "Failed to create meter"
9195 " policer default matcher.");
9198 dv_attr.priority = 0;
9199 dv_attr.match_criteria_enable =
9200 1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
9201 flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
9202 rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
9203 dtb->color_matcher = mlx5_glue->dv_create_flow_matcher(sh->ctx,
9206 if (!dtb->color_matcher) {
9207 DRV_LOG(ERR, "Failed to create meter policer color matcher.");
9210 if (mtb->count_actns[RTE_MTR_DROPPED])
9211 actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
9212 actions[i++] = mtb->drop_actn;
9213 /* Default rule: lowest priority, match any, actions: drop. */
9214 dtb->policer_rules[RTE_MTR_DROPPED] =
9215 mlx5_glue->dv_create_flow(dtb->any_matcher,
9216 (void *)&value, i, actions);
9217 if (!dtb->policer_rules[RTE_MTR_DROPPED]) {
9218 DRV_LOG(ERR, "Failed to create meter policer drop rule.");
9227 * Create the needed meter and suffix tables.
9228 * Lock free, (mutex should be acquired by caller).
9231 * Pointer to Ethernet device.
9233 * Pointer to the flow meter.
9236 * Pointer to table set on success, NULL otherwise and rte_errno is set.
9238 static struct mlx5_meter_domains_infos *
9239 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
9240 const struct mlx5_flow_meter *fm)
9242 struct mlx5_priv *priv = dev->data->dev_private;
9243 struct mlx5_meter_domains_infos *mtb;
9247 if (!priv->mtr_en) {
9248 rte_errno = ENOTSUP;
9251 mtb = rte_calloc(__func__, 1, sizeof(*mtb), 0);
9253 DRV_LOG(ERR, "Failed to allocate memory for meter.");
9256 /* Create meter count actions */
9257 for (i = 0; i <= RTE_MTR_DROPPED; i++) {
9258 struct mlx5_flow_counter *cnt;
9259 if (!fm->policer_stats.cnt[i])
9261 cnt = flow_dv_counter_get_by_idx(dev,
9262 fm->policer_stats.cnt[i], NULL);
9263 mtb->count_actns[i] = cnt->action;
9265 /* Create drop action. */
9266 mtb->drop_actn = mlx5_glue->dr_create_flow_action_drop();
9267 if (!mtb->drop_actn) {
9268 DRV_LOG(ERR, "Failed to create drop action.");
9271 /* Egress meter table. */
9272 ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
9274 DRV_LOG(ERR, "Failed to prepare egress meter table.");
9277 /* Ingress meter table. */
9278 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
9280 DRV_LOG(ERR, "Failed to prepare ingress meter table.");
9283 /* FDB meter table. */
9284 if (priv->config.dv_esw_en) {
9285 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
9286 priv->mtr_color_reg);
9288 DRV_LOG(ERR, "Failed to prepare fdb meter table.");
9294 flow_dv_destroy_mtr_tbl(dev, mtb);
9299 * Destroy domain policer rule.
9302 * Pointer to domain table.
9305 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
9309 for (i = 0; i < RTE_MTR_DROPPED; i++) {
9310 if (dt->policer_rules[i]) {
9311 claim_zero(mlx5_glue->dv_destroy_flow
9312 (dt->policer_rules[i]));
9313 dt->policer_rules[i] = NULL;
9316 if (dt->jump_actn) {
9317 claim_zero(mlx5_glue->destroy_flow_action(dt->jump_actn));
9318 dt->jump_actn = NULL;
9323 * Destroy policer rules.
9326 * Pointer to Ethernet device.
9328 * Pointer to flow meter structure.
9330 * Pointer to flow attributes.
9336 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
9337 const struct mlx5_flow_meter *fm,
9338 const struct rte_flow_attr *attr)
9340 struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
9345 flow_dv_destroy_domain_policer_rule(&mtb->egress);
9347 flow_dv_destroy_domain_policer_rule(&mtb->ingress);
9349 flow_dv_destroy_domain_policer_rule(&mtb->transfer);
9354 * Create specify domain meter policer rule.
9357 * Pointer to flow meter structure.
9359 * Pointer to DV meter table set.
9360 * @param[in] mtr_reg_c
9361 * Color match REG_C.
9364 * 0 on success, -1 otherwise.
9367 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
9368 struct mlx5_meter_domain_info *dtb,
9371 struct mlx5_flow_dv_match_params matcher = {
9372 .size = sizeof(matcher.buf),
9374 struct mlx5_flow_dv_match_params value = {
9375 .size = sizeof(value.buf),
9377 struct mlx5_meter_domains_infos *mtb = fm->mfts;
9378 void *actions[METER_ACTIONS];
9381 /* Create jump action. */
9382 if (!dtb->jump_actn)
9384 mlx5_glue->dr_create_flow_action_dest_flow_tbl
9385 (dtb->sfx_tbl->obj);
9386 if (!dtb->jump_actn) {
9387 DRV_LOG(ERR, "Failed to create policer jump action.");
9390 for (i = 0; i < RTE_MTR_DROPPED; i++) {
9393 flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
9394 rte_col_2_mlx5_col(i), UINT8_MAX);
9395 if (mtb->count_actns[i])
9396 actions[j++] = mtb->count_actns[i];
9397 if (fm->action[i] == MTR_POLICER_ACTION_DROP)
9398 actions[j++] = mtb->drop_actn;
9400 actions[j++] = dtb->jump_actn;
9401 dtb->policer_rules[i] =
9402 mlx5_glue->dv_create_flow(dtb->color_matcher,
9405 if (!dtb->policer_rules[i]) {
9406 DRV_LOG(ERR, "Failed to create policer rule.");
9417 * Create policer rules.
9420 * Pointer to Ethernet device.
9422 * Pointer to flow meter structure.
9424 * Pointer to flow attributes.
9427 * 0 on success, -1 otherwise.
9430 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
9431 struct mlx5_flow_meter *fm,
9432 const struct rte_flow_attr *attr)
9434 struct mlx5_priv *priv = dev->data->dev_private;
9435 struct mlx5_meter_domains_infos *mtb = fm->mfts;
9439 ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
9440 priv->mtr_color_reg);
9442 DRV_LOG(ERR, "Failed to create egress policer.");
9446 if (attr->ingress) {
9447 ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
9448 priv->mtr_color_reg);
9450 DRV_LOG(ERR, "Failed to create ingress policer.");
9454 if (attr->transfer) {
9455 ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
9456 priv->mtr_color_reg);
9458 DRV_LOG(ERR, "Failed to create transfer policer.");
9464 flow_dv_destroy_policer_rules(dev, fm, attr);
9469 * Query a devx counter.
9472 * Pointer to the Ethernet device structure.
9474 * Index to the flow counter.
9476 * Set to clear the counter statistics.
9478 * The statistics value of packets.
9480 * The statistics value of bytes.
9483 * 0 on success, otherwise return -1.
9486 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
9487 uint64_t *pkts, uint64_t *bytes)
9489 struct mlx5_priv *priv = dev->data->dev_private;
9490 struct mlx5_flow_counter *cnt;
9491 uint64_t inn_pkts, inn_bytes;
9494 if (!priv->config.devx)
9497 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
9500 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
9501 *pkts = inn_pkts - cnt->hits;
9502 *bytes = inn_bytes - cnt->bytes;
9504 cnt->hits = inn_pkts;
9505 cnt->bytes = inn_bytes;
9511 * Get aged-out flows.
9514 * Pointer to the Ethernet device structure.
9515 * @param[in] context
9516 * The address of an array of pointers to the aged-out flows contexts.
9517 * @param[in] nb_contexts
9518 * The length of context array pointers.
9520 * Perform verbose error reporting if not NULL. Initialized in case of
9524 * how many contexts get in success, otherwise negative errno value.
9525 * if nb_contexts is 0, return the amount of all aged contexts.
9526 * if nb_contexts is not 0 , return the amount of aged flows reported
9527 * in the context array.
9528 * @note: only stub for now
9531 flow_get_aged_flows(struct rte_eth_dev *dev,
9533 uint32_t nb_contexts,
9534 struct rte_flow_error *error)
9536 struct mlx5_priv *priv = dev->data->dev_private;
9537 struct mlx5_age_info *age_info;
9538 struct mlx5_age_param *age_param;
9539 struct mlx5_flow_counter *counter;
9542 if (nb_contexts && !context)
9543 return rte_flow_error_set(error, EINVAL,
9544 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9546 "Should assign at least one flow or"
9547 " context to get if nb_contexts != 0");
9548 age_info = GET_PORT_AGE_INFO(priv);
9549 rte_spinlock_lock(&age_info->aged_sl);
9550 TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
9553 age_param = MLX5_CNT_TO_AGE(counter);
9554 context[nb_flows - 1] = age_param->context;
9555 if (!(--nb_contexts))
9559 rte_spinlock_unlock(&age_info->aged_sl);
9560 MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
9565 * Mutex-protected thunk to lock-free __flow_dv_translate().
9568 flow_dv_translate(struct rte_eth_dev *dev,
9569 struct mlx5_flow *dev_flow,
9570 const struct rte_flow_attr *attr,
9571 const struct rte_flow_item items[],
9572 const struct rte_flow_action actions[],
9573 struct rte_flow_error *error)
9577 flow_dv_shared_lock(dev);
9578 ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
9579 flow_dv_shared_unlock(dev);
9584 * Mutex-protected thunk to lock-free __flow_dv_apply().
9587 flow_dv_apply(struct rte_eth_dev *dev,
9588 struct rte_flow *flow,
9589 struct rte_flow_error *error)
9593 flow_dv_shared_lock(dev);
9594 ret = __flow_dv_apply(dev, flow, error);
9595 flow_dv_shared_unlock(dev);
9600 * Mutex-protected thunk to lock-free __flow_dv_remove().
9603 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
9605 flow_dv_shared_lock(dev);
9606 __flow_dv_remove(dev, flow);
9607 flow_dv_shared_unlock(dev);
9611 * Mutex-protected thunk to lock-free __flow_dv_destroy().
9614 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
9616 flow_dv_shared_lock(dev);
9617 __flow_dv_destroy(dev, flow);
9618 flow_dv_shared_unlock(dev);
9622 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
9625 flow_dv_counter_allocate(struct rte_eth_dev *dev)
9629 flow_dv_shared_lock(dev);
9630 cnt = flow_dv_counter_alloc(dev, 0, 0, 1, 0);
9631 flow_dv_shared_unlock(dev);
9636 * Mutex-protected thunk to lock-free flow_dv_counter_release().
9639 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
9641 flow_dv_shared_lock(dev);
9642 flow_dv_counter_release(dev, cnt);
9643 flow_dv_shared_unlock(dev);
9646 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
9647 .validate = flow_dv_validate,
9648 .prepare = flow_dv_prepare,
9649 .translate = flow_dv_translate,
9650 .apply = flow_dv_apply,
9651 .remove = flow_dv_remove,
9652 .destroy = flow_dv_destroy,
9653 .query = flow_dv_query,
9654 .create_mtr_tbls = flow_dv_create_mtr_tbl,
9655 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
9656 .create_policer_rules = flow_dv_create_policer_rules,
9657 .destroy_policer_rules = flow_dv_destroy_policer_rules,
9658 .counter_alloc = flow_dv_counter_allocate,
9659 .counter_free = flow_dv_counter_free,
9660 .counter_query = flow_dv_counter_query,
9661 .get_aged_flows = flow_get_aged_flows,
9664 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */