1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018 Mellanox Technologies, Ltd
11 #include <rte_common.h>
12 #include <rte_ether.h>
13 #include <rte_ethdev_driver.h>
15 #include <rte_flow_driver.h>
16 #include <rte_malloc.h>
17 #include <rte_cycles.h>
20 #include <rte_vxlan.h>
22 #include <rte_eal_paging.h>
25 #include <mlx5_glue.h>
26 #include <mlx5_devx_cmds.h>
28 #include <mlx5_malloc.h>
30 #include "mlx5_defs.h"
32 #include "mlx5_common_os.h"
33 #include "mlx5_flow.h"
34 #include "mlx5_flow_os.h"
35 #include "mlx5_rxtx.h"
36 #include "rte_pmd_mlx5.h"
38 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
40 #ifndef HAVE_IBV_FLOW_DEVX_COUNTERS
41 #define MLX5DV_FLOW_ACTION_COUNTERS_DEVX 0
44 #ifndef HAVE_MLX5DV_DR_ESWITCH
45 #ifndef MLX5DV_FLOW_TABLE_TYPE_FDB
46 #define MLX5DV_FLOW_TABLE_TYPE_FDB 0
50 #ifndef HAVE_MLX5DV_DR
51 #define MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL 1
54 /* VLAN header definitions */
55 #define MLX5DV_FLOW_VLAN_PCP_SHIFT 13
56 #define MLX5DV_FLOW_VLAN_PCP_MASK (0x7 << MLX5DV_FLOW_VLAN_PCP_SHIFT)
57 #define MLX5DV_FLOW_VLAN_VID_MASK 0x0fff
58 #define MLX5DV_FLOW_VLAN_PCP_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK)
59 #define MLX5DV_FLOW_VLAN_VID_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_VID_MASK)
74 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
75 struct mlx5_flow_tbl_resource *tbl);
78 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
79 uint32_t encap_decap_idx);
82 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
86 * Initialize flow attributes structure according to flow items' types.
88 * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
89 * mode. For tunnel mode, the items to be modified are the outermost ones.
92 * Pointer to item specification.
94 * Pointer to flow attributes structure.
96 * Pointer to the sub flow.
97 * @param[in] tunnel_decap
98 * Whether action is after tunnel decapsulation.
101 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
102 struct mlx5_flow *dev_flow, bool tunnel_decap)
104 uint64_t layers = dev_flow->handle->layers;
107 * If layers is already initialized, it means this dev_flow is the
108 * suffix flow, the layers flags is set by the prefix flow. Need to
109 * use the layer flags from prefix flow as the suffix flow may not
110 * have the user defined items as the flow is split.
113 if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
115 else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
117 if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
119 else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
124 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
125 uint8_t next_protocol = 0xff;
126 switch (item->type) {
127 case RTE_FLOW_ITEM_TYPE_GRE:
128 case RTE_FLOW_ITEM_TYPE_NVGRE:
129 case RTE_FLOW_ITEM_TYPE_VXLAN:
130 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
131 case RTE_FLOW_ITEM_TYPE_GENEVE:
132 case RTE_FLOW_ITEM_TYPE_MPLS:
136 case RTE_FLOW_ITEM_TYPE_IPV4:
139 if (item->mask != NULL &&
140 ((const struct rte_flow_item_ipv4 *)
141 item->mask)->hdr.next_proto_id)
143 ((const struct rte_flow_item_ipv4 *)
144 (item->spec))->hdr.next_proto_id &
145 ((const struct rte_flow_item_ipv4 *)
146 (item->mask))->hdr.next_proto_id;
147 if ((next_protocol == IPPROTO_IPIP ||
148 next_protocol == IPPROTO_IPV6) && tunnel_decap)
151 case RTE_FLOW_ITEM_TYPE_IPV6:
154 if (item->mask != NULL &&
155 ((const struct rte_flow_item_ipv6 *)
156 item->mask)->hdr.proto)
158 ((const struct rte_flow_item_ipv6 *)
159 (item->spec))->hdr.proto &
160 ((const struct rte_flow_item_ipv6 *)
161 (item->mask))->hdr.proto;
162 if ((next_protocol == IPPROTO_IPIP ||
163 next_protocol == IPPROTO_IPV6) && tunnel_decap)
166 case RTE_FLOW_ITEM_TYPE_UDP:
170 case RTE_FLOW_ITEM_TYPE_TCP:
182 * Convert rte_mtr_color to mlx5 color.
191 rte_col_2_mlx5_col(enum rte_color rcol)
194 case RTE_COLOR_GREEN:
195 return MLX5_FLOW_COLOR_GREEN;
196 case RTE_COLOR_YELLOW:
197 return MLX5_FLOW_COLOR_YELLOW;
199 return MLX5_FLOW_COLOR_RED;
203 return MLX5_FLOW_COLOR_UNDEFINED;
206 struct field_modify_info {
207 uint32_t size; /* Size of field in protocol header, in bytes. */
208 uint32_t offset; /* Offset of field in protocol header, in bytes. */
209 enum mlx5_modification_field id;
212 struct field_modify_info modify_eth[] = {
213 {4, 0, MLX5_MODI_OUT_DMAC_47_16},
214 {2, 4, MLX5_MODI_OUT_DMAC_15_0},
215 {4, 6, MLX5_MODI_OUT_SMAC_47_16},
216 {2, 10, MLX5_MODI_OUT_SMAC_15_0},
220 struct field_modify_info modify_vlan_out_first_vid[] = {
221 /* Size in bits !!! */
222 {12, 0, MLX5_MODI_OUT_FIRST_VID},
226 struct field_modify_info modify_ipv4[] = {
227 {1, 1, MLX5_MODI_OUT_IP_DSCP},
228 {1, 8, MLX5_MODI_OUT_IPV4_TTL},
229 {4, 12, MLX5_MODI_OUT_SIPV4},
230 {4, 16, MLX5_MODI_OUT_DIPV4},
234 struct field_modify_info modify_ipv6[] = {
235 {1, 0, MLX5_MODI_OUT_IP_DSCP},
236 {1, 7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
237 {4, 8, MLX5_MODI_OUT_SIPV6_127_96},
238 {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
239 {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
240 {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
241 {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
242 {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
243 {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
244 {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
248 struct field_modify_info modify_udp[] = {
249 {2, 0, MLX5_MODI_OUT_UDP_SPORT},
250 {2, 2, MLX5_MODI_OUT_UDP_DPORT},
254 struct field_modify_info modify_tcp[] = {
255 {2, 0, MLX5_MODI_OUT_TCP_SPORT},
256 {2, 2, MLX5_MODI_OUT_TCP_DPORT},
257 {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
258 {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
263 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
264 uint8_t next_protocol, uint64_t *item_flags,
267 MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
268 item->type == RTE_FLOW_ITEM_TYPE_IPV6);
269 if (next_protocol == IPPROTO_IPIP) {
270 *item_flags |= MLX5_FLOW_LAYER_IPIP;
273 if (next_protocol == IPPROTO_IPV6) {
274 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
280 * Acquire the synchronizing object to protect multithreaded access
281 * to shared dv context. Lock occurs only if context is actually
282 * shared, i.e. we have multiport IB device and representors are
286 * Pointer to the rte_eth_dev structure.
289 flow_dv_shared_lock(struct rte_eth_dev *dev)
291 struct mlx5_priv *priv = dev->data->dev_private;
292 struct mlx5_dev_ctx_shared *sh = priv->sh;
294 if (sh->refcnt > 1) {
297 ret = pthread_mutex_lock(&sh->dv_mutex);
304 flow_dv_shared_unlock(struct rte_eth_dev *dev)
306 struct mlx5_priv *priv = dev->data->dev_private;
307 struct mlx5_dev_ctx_shared *sh = priv->sh;
309 if (sh->refcnt > 1) {
312 ret = pthread_mutex_unlock(&sh->dv_mutex);
318 /* Update VLAN's VID/PCP based on input rte_flow_action.
321 * Pointer to struct rte_flow_action.
323 * Pointer to struct rte_vlan_hdr.
326 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
327 struct rte_vlan_hdr *vlan)
330 if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
332 ((const struct rte_flow_action_of_set_vlan_pcp *)
333 action->conf)->vlan_pcp;
334 vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
335 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
336 vlan->vlan_tci |= vlan_tci;
337 } else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
338 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
339 vlan->vlan_tci |= rte_be_to_cpu_16
340 (((const struct rte_flow_action_of_set_vlan_vid *)
341 action->conf)->vlan_vid);
346 * Fetch 1, 2, 3 or 4 byte field from the byte array
347 * and return as unsigned integer in host-endian format.
350 * Pointer to data array.
352 * Size of field to extract.
355 * converted field in host endian format.
357 static inline uint32_t
358 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
367 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
370 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
371 ret = (ret << 8) | *(data + sizeof(uint16_t));
374 ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
385 * Convert modify-header action to DV specification.
387 * Data length of each action is determined by provided field description
388 * and the item mask. Data bit offset and width of each action is determined
389 * by provided item mask.
392 * Pointer to item specification.
394 * Pointer to field modification information.
395 * For MLX5_MODIFICATION_TYPE_SET specifies destination field.
396 * For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
397 * For MLX5_MODIFICATION_TYPE_COPY specifies source field.
399 * Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
400 * Negative offset value sets the same offset as source offset.
401 * size field is ignored, value is taken from source field.
402 * @param[in,out] resource
403 * Pointer to the modify-header resource.
405 * Type of modification.
407 * Pointer to the error structure.
410 * 0 on success, a negative errno value otherwise and rte_errno is set.
413 flow_dv_convert_modify_action(struct rte_flow_item *item,
414 struct field_modify_info *field,
415 struct field_modify_info *dcopy,
416 struct mlx5_flow_dv_modify_hdr_resource *resource,
417 uint32_t type, struct rte_flow_error *error)
419 uint32_t i = resource->actions_num;
420 struct mlx5_modification_cmd *actions = resource->actions;
423 * The item and mask are provided in big-endian format.
424 * The fields should be presented as in big-endian format either.
425 * Mask must be always present, it defines the actual field width.
427 MLX5_ASSERT(item->mask);
428 MLX5_ASSERT(field->size);
435 if (i >= MLX5_MAX_MODIFY_NUM)
436 return rte_flow_error_set(error, EINVAL,
437 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
438 "too many items to modify");
439 /* Fetch variable byte size mask from the array. */
440 mask = flow_dv_fetch_field((const uint8_t *)item->mask +
441 field->offset, field->size);
446 /* Deduce actual data width in bits from mask value. */
447 off_b = rte_bsf32(mask);
448 size_b = sizeof(uint32_t) * CHAR_BIT -
449 off_b - __builtin_clz(mask);
451 size_b = size_b == sizeof(uint32_t) * CHAR_BIT ? 0 : size_b;
452 actions[i] = (struct mlx5_modification_cmd) {
458 /* Convert entire record to expected big-endian format. */
459 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
460 if (type == MLX5_MODIFICATION_TYPE_COPY) {
462 actions[i].dst_field = dcopy->id;
463 actions[i].dst_offset =
464 (int)dcopy->offset < 0 ? off_b : dcopy->offset;
465 /* Convert entire record to big-endian format. */
466 actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
468 MLX5_ASSERT(item->spec);
469 data = flow_dv_fetch_field((const uint8_t *)item->spec +
470 field->offset, field->size);
471 /* Shift out the trailing masked bits from data. */
472 data = (data & mask) >> off_b;
473 actions[i].data1 = rte_cpu_to_be_32(data);
477 } while (field->size);
478 if (resource->actions_num == i)
479 return rte_flow_error_set(error, EINVAL,
480 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
481 "invalid modification flow item");
482 resource->actions_num = i;
487 * Convert modify-header set IPv4 address action to DV specification.
489 * @param[in,out] resource
490 * Pointer to the modify-header resource.
492 * Pointer to action specification.
494 * Pointer to the error structure.
497 * 0 on success, a negative errno value otherwise and rte_errno is set.
500 flow_dv_convert_action_modify_ipv4
501 (struct mlx5_flow_dv_modify_hdr_resource *resource,
502 const struct rte_flow_action *action,
503 struct rte_flow_error *error)
505 const struct rte_flow_action_set_ipv4 *conf =
506 (const struct rte_flow_action_set_ipv4 *)(action->conf);
507 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
508 struct rte_flow_item_ipv4 ipv4;
509 struct rte_flow_item_ipv4 ipv4_mask;
511 memset(&ipv4, 0, sizeof(ipv4));
512 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
513 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
514 ipv4.hdr.src_addr = conf->ipv4_addr;
515 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
517 ipv4.hdr.dst_addr = conf->ipv4_addr;
518 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
521 item.mask = &ipv4_mask;
522 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
523 MLX5_MODIFICATION_TYPE_SET, error);
527 * Convert modify-header set IPv6 address action to DV specification.
529 * @param[in,out] resource
530 * Pointer to the modify-header resource.
532 * Pointer to action specification.
534 * Pointer to the error structure.
537 * 0 on success, a negative errno value otherwise and rte_errno is set.
540 flow_dv_convert_action_modify_ipv6
541 (struct mlx5_flow_dv_modify_hdr_resource *resource,
542 const struct rte_flow_action *action,
543 struct rte_flow_error *error)
545 const struct rte_flow_action_set_ipv6 *conf =
546 (const struct rte_flow_action_set_ipv6 *)(action->conf);
547 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
548 struct rte_flow_item_ipv6 ipv6;
549 struct rte_flow_item_ipv6 ipv6_mask;
551 memset(&ipv6, 0, sizeof(ipv6));
552 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
553 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
554 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
555 sizeof(ipv6.hdr.src_addr));
556 memcpy(&ipv6_mask.hdr.src_addr,
557 &rte_flow_item_ipv6_mask.hdr.src_addr,
558 sizeof(ipv6.hdr.src_addr));
560 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
561 sizeof(ipv6.hdr.dst_addr));
562 memcpy(&ipv6_mask.hdr.dst_addr,
563 &rte_flow_item_ipv6_mask.hdr.dst_addr,
564 sizeof(ipv6.hdr.dst_addr));
567 item.mask = &ipv6_mask;
568 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
569 MLX5_MODIFICATION_TYPE_SET, error);
573 * Convert modify-header set MAC address action to DV specification.
575 * @param[in,out] resource
576 * Pointer to the modify-header resource.
578 * Pointer to action specification.
580 * Pointer to the error structure.
583 * 0 on success, a negative errno value otherwise and rte_errno is set.
586 flow_dv_convert_action_modify_mac
587 (struct mlx5_flow_dv_modify_hdr_resource *resource,
588 const struct rte_flow_action *action,
589 struct rte_flow_error *error)
591 const struct rte_flow_action_set_mac *conf =
592 (const struct rte_flow_action_set_mac *)(action->conf);
593 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
594 struct rte_flow_item_eth eth;
595 struct rte_flow_item_eth eth_mask;
597 memset(ð, 0, sizeof(eth));
598 memset(ð_mask, 0, sizeof(eth_mask));
599 if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
600 memcpy(ð.src.addr_bytes, &conf->mac_addr,
601 sizeof(eth.src.addr_bytes));
602 memcpy(ð_mask.src.addr_bytes,
603 &rte_flow_item_eth_mask.src.addr_bytes,
604 sizeof(eth_mask.src.addr_bytes));
606 memcpy(ð.dst.addr_bytes, &conf->mac_addr,
607 sizeof(eth.dst.addr_bytes));
608 memcpy(ð_mask.dst.addr_bytes,
609 &rte_flow_item_eth_mask.dst.addr_bytes,
610 sizeof(eth_mask.dst.addr_bytes));
613 item.mask = ð_mask;
614 return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
615 MLX5_MODIFICATION_TYPE_SET, error);
619 * Convert modify-header set VLAN VID action to DV specification.
621 * @param[in,out] resource
622 * Pointer to the modify-header resource.
624 * Pointer to action specification.
626 * Pointer to the error structure.
629 * 0 on success, a negative errno value otherwise and rte_errno is set.
632 flow_dv_convert_action_modify_vlan_vid
633 (struct mlx5_flow_dv_modify_hdr_resource *resource,
634 const struct rte_flow_action *action,
635 struct rte_flow_error *error)
637 const struct rte_flow_action_of_set_vlan_vid *conf =
638 (const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
639 int i = resource->actions_num;
640 struct mlx5_modification_cmd *actions = resource->actions;
641 struct field_modify_info *field = modify_vlan_out_first_vid;
643 if (i >= MLX5_MAX_MODIFY_NUM)
644 return rte_flow_error_set(error, EINVAL,
645 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
646 "too many items to modify");
647 actions[i] = (struct mlx5_modification_cmd) {
648 .action_type = MLX5_MODIFICATION_TYPE_SET,
650 .length = field->size,
651 .offset = field->offset,
653 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
654 actions[i].data1 = conf->vlan_vid;
655 actions[i].data1 = actions[i].data1 << 16;
656 resource->actions_num = ++i;
661 * Convert modify-header set TP action to DV specification.
663 * @param[in,out] resource
664 * Pointer to the modify-header resource.
666 * Pointer to action specification.
668 * Pointer to rte_flow_item objects list.
670 * Pointer to flow attributes structure.
671 * @param[in] dev_flow
672 * Pointer to the sub flow.
673 * @param[in] tunnel_decap
674 * Whether action is after tunnel decapsulation.
676 * Pointer to the error structure.
679 * 0 on success, a negative errno value otherwise and rte_errno is set.
682 flow_dv_convert_action_modify_tp
683 (struct mlx5_flow_dv_modify_hdr_resource *resource,
684 const struct rte_flow_action *action,
685 const struct rte_flow_item *items,
686 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
687 bool tunnel_decap, struct rte_flow_error *error)
689 const struct rte_flow_action_set_tp *conf =
690 (const struct rte_flow_action_set_tp *)(action->conf);
691 struct rte_flow_item item;
692 struct rte_flow_item_udp udp;
693 struct rte_flow_item_udp udp_mask;
694 struct rte_flow_item_tcp tcp;
695 struct rte_flow_item_tcp tcp_mask;
696 struct field_modify_info *field;
699 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
701 memset(&udp, 0, sizeof(udp));
702 memset(&udp_mask, 0, sizeof(udp_mask));
703 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
704 udp.hdr.src_port = conf->port;
705 udp_mask.hdr.src_port =
706 rte_flow_item_udp_mask.hdr.src_port;
708 udp.hdr.dst_port = conf->port;
709 udp_mask.hdr.dst_port =
710 rte_flow_item_udp_mask.hdr.dst_port;
712 item.type = RTE_FLOW_ITEM_TYPE_UDP;
714 item.mask = &udp_mask;
717 MLX5_ASSERT(attr->tcp);
718 memset(&tcp, 0, sizeof(tcp));
719 memset(&tcp_mask, 0, sizeof(tcp_mask));
720 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
721 tcp.hdr.src_port = conf->port;
722 tcp_mask.hdr.src_port =
723 rte_flow_item_tcp_mask.hdr.src_port;
725 tcp.hdr.dst_port = conf->port;
726 tcp_mask.hdr.dst_port =
727 rte_flow_item_tcp_mask.hdr.dst_port;
729 item.type = RTE_FLOW_ITEM_TYPE_TCP;
731 item.mask = &tcp_mask;
734 return flow_dv_convert_modify_action(&item, field, NULL, resource,
735 MLX5_MODIFICATION_TYPE_SET, error);
739 * Convert modify-header set TTL action to DV specification.
741 * @param[in,out] resource
742 * Pointer to the modify-header resource.
744 * Pointer to action specification.
746 * Pointer to rte_flow_item objects list.
748 * Pointer to flow attributes structure.
749 * @param[in] dev_flow
750 * Pointer to the sub flow.
751 * @param[in] tunnel_decap
752 * Whether action is after tunnel decapsulation.
754 * Pointer to the error structure.
757 * 0 on success, a negative errno value otherwise and rte_errno is set.
760 flow_dv_convert_action_modify_ttl
761 (struct mlx5_flow_dv_modify_hdr_resource *resource,
762 const struct rte_flow_action *action,
763 const struct rte_flow_item *items,
764 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
765 bool tunnel_decap, struct rte_flow_error *error)
767 const struct rte_flow_action_set_ttl *conf =
768 (const struct rte_flow_action_set_ttl *)(action->conf);
769 struct rte_flow_item item;
770 struct rte_flow_item_ipv4 ipv4;
771 struct rte_flow_item_ipv4 ipv4_mask;
772 struct rte_flow_item_ipv6 ipv6;
773 struct rte_flow_item_ipv6 ipv6_mask;
774 struct field_modify_info *field;
777 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
779 memset(&ipv4, 0, sizeof(ipv4));
780 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
781 ipv4.hdr.time_to_live = conf->ttl_value;
782 ipv4_mask.hdr.time_to_live = 0xFF;
783 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
785 item.mask = &ipv4_mask;
788 MLX5_ASSERT(attr->ipv6);
789 memset(&ipv6, 0, sizeof(ipv6));
790 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
791 ipv6.hdr.hop_limits = conf->ttl_value;
792 ipv6_mask.hdr.hop_limits = 0xFF;
793 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
795 item.mask = &ipv6_mask;
798 return flow_dv_convert_modify_action(&item, field, NULL, resource,
799 MLX5_MODIFICATION_TYPE_SET, error);
803 * Convert modify-header decrement TTL action to DV specification.
805 * @param[in,out] resource
806 * Pointer to the modify-header resource.
808 * Pointer to action specification.
810 * Pointer to rte_flow_item objects list.
812 * Pointer to flow attributes structure.
813 * @param[in] dev_flow
814 * Pointer to the sub flow.
815 * @param[in] tunnel_decap
816 * Whether action is after tunnel decapsulation.
818 * Pointer to the error structure.
821 * 0 on success, a negative errno value otherwise and rte_errno is set.
824 flow_dv_convert_action_modify_dec_ttl
825 (struct mlx5_flow_dv_modify_hdr_resource *resource,
826 const struct rte_flow_item *items,
827 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
828 bool tunnel_decap, struct rte_flow_error *error)
830 struct rte_flow_item item;
831 struct rte_flow_item_ipv4 ipv4;
832 struct rte_flow_item_ipv4 ipv4_mask;
833 struct rte_flow_item_ipv6 ipv6;
834 struct rte_flow_item_ipv6 ipv6_mask;
835 struct field_modify_info *field;
838 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
840 memset(&ipv4, 0, sizeof(ipv4));
841 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
842 ipv4.hdr.time_to_live = 0xFF;
843 ipv4_mask.hdr.time_to_live = 0xFF;
844 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
846 item.mask = &ipv4_mask;
849 MLX5_ASSERT(attr->ipv6);
850 memset(&ipv6, 0, sizeof(ipv6));
851 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
852 ipv6.hdr.hop_limits = 0xFF;
853 ipv6_mask.hdr.hop_limits = 0xFF;
854 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
856 item.mask = &ipv6_mask;
859 return flow_dv_convert_modify_action(&item, field, NULL, resource,
860 MLX5_MODIFICATION_TYPE_ADD, error);
864 * Convert modify-header increment/decrement TCP Sequence number
865 * to DV specification.
867 * @param[in,out] resource
868 * Pointer to the modify-header resource.
870 * Pointer to action specification.
872 * Pointer to the error structure.
875 * 0 on success, a negative errno value otherwise and rte_errno is set.
878 flow_dv_convert_action_modify_tcp_seq
879 (struct mlx5_flow_dv_modify_hdr_resource *resource,
880 const struct rte_flow_action *action,
881 struct rte_flow_error *error)
883 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
884 uint64_t value = rte_be_to_cpu_32(*conf);
885 struct rte_flow_item item;
886 struct rte_flow_item_tcp tcp;
887 struct rte_flow_item_tcp tcp_mask;
889 memset(&tcp, 0, sizeof(tcp));
890 memset(&tcp_mask, 0, sizeof(tcp_mask));
891 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
893 * The HW has no decrement operation, only increment operation.
894 * To simulate decrement X from Y using increment operation
895 * we need to add UINT32_MAX X times to Y.
896 * Each adding of UINT32_MAX decrements Y by 1.
899 tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
900 tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
901 item.type = RTE_FLOW_ITEM_TYPE_TCP;
903 item.mask = &tcp_mask;
904 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
905 MLX5_MODIFICATION_TYPE_ADD, error);
909 * Convert modify-header increment/decrement TCP Acknowledgment number
910 * to DV specification.
912 * @param[in,out] resource
913 * Pointer to the modify-header resource.
915 * Pointer to action specification.
917 * Pointer to the error structure.
920 * 0 on success, a negative errno value otherwise and rte_errno is set.
923 flow_dv_convert_action_modify_tcp_ack
924 (struct mlx5_flow_dv_modify_hdr_resource *resource,
925 const struct rte_flow_action *action,
926 struct rte_flow_error *error)
928 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
929 uint64_t value = rte_be_to_cpu_32(*conf);
930 struct rte_flow_item item;
931 struct rte_flow_item_tcp tcp;
932 struct rte_flow_item_tcp tcp_mask;
934 memset(&tcp, 0, sizeof(tcp));
935 memset(&tcp_mask, 0, sizeof(tcp_mask));
936 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
938 * The HW has no decrement operation, only increment operation.
939 * To simulate decrement X from Y using increment operation
940 * we need to add UINT32_MAX X times to Y.
941 * Each adding of UINT32_MAX decrements Y by 1.
944 tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
945 tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
946 item.type = RTE_FLOW_ITEM_TYPE_TCP;
948 item.mask = &tcp_mask;
949 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
950 MLX5_MODIFICATION_TYPE_ADD, error);
953 static enum mlx5_modification_field reg_to_field[] = {
954 [REG_NON] = MLX5_MODI_OUT_NONE,
955 [REG_A] = MLX5_MODI_META_DATA_REG_A,
956 [REG_B] = MLX5_MODI_META_DATA_REG_B,
957 [REG_C_0] = MLX5_MODI_META_REG_C_0,
958 [REG_C_1] = MLX5_MODI_META_REG_C_1,
959 [REG_C_2] = MLX5_MODI_META_REG_C_2,
960 [REG_C_3] = MLX5_MODI_META_REG_C_3,
961 [REG_C_4] = MLX5_MODI_META_REG_C_4,
962 [REG_C_5] = MLX5_MODI_META_REG_C_5,
963 [REG_C_6] = MLX5_MODI_META_REG_C_6,
964 [REG_C_7] = MLX5_MODI_META_REG_C_7,
968 * Convert register set to DV specification.
970 * @param[in,out] resource
971 * Pointer to the modify-header resource.
973 * Pointer to action specification.
975 * Pointer to the error structure.
978 * 0 on success, a negative errno value otherwise and rte_errno is set.
981 flow_dv_convert_action_set_reg
982 (struct mlx5_flow_dv_modify_hdr_resource *resource,
983 const struct rte_flow_action *action,
984 struct rte_flow_error *error)
986 const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
987 struct mlx5_modification_cmd *actions = resource->actions;
988 uint32_t i = resource->actions_num;
990 if (i >= MLX5_MAX_MODIFY_NUM)
991 return rte_flow_error_set(error, EINVAL,
992 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
993 "too many items to modify");
994 MLX5_ASSERT(conf->id != REG_NON);
995 MLX5_ASSERT(conf->id < RTE_DIM(reg_to_field));
996 actions[i] = (struct mlx5_modification_cmd) {
997 .action_type = MLX5_MODIFICATION_TYPE_SET,
998 .field = reg_to_field[conf->id],
1000 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
1001 actions[i].data1 = rte_cpu_to_be_32(conf->data);
1003 resource->actions_num = i;
1008 * Convert SET_TAG action to DV specification.
1011 * Pointer to the rte_eth_dev structure.
1012 * @param[in,out] resource
1013 * Pointer to the modify-header resource.
1015 * Pointer to action specification.
1017 * Pointer to the error structure.
1020 * 0 on success, a negative errno value otherwise and rte_errno is set.
1023 flow_dv_convert_action_set_tag
1024 (struct rte_eth_dev *dev,
1025 struct mlx5_flow_dv_modify_hdr_resource *resource,
1026 const struct rte_flow_action_set_tag *conf,
1027 struct rte_flow_error *error)
1029 rte_be32_t data = rte_cpu_to_be_32(conf->data);
1030 rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1031 struct rte_flow_item item = {
1035 struct field_modify_info reg_c_x[] = {
1038 enum mlx5_modification_field reg_type;
1041 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1044 MLX5_ASSERT(ret != REG_NON);
1045 MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1046 reg_type = reg_to_field[ret];
1047 MLX5_ASSERT(reg_type > 0);
1048 reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1049 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1050 MLX5_MODIFICATION_TYPE_SET, error);
1054 * Convert internal COPY_REG action to DV specification.
1057 * Pointer to the rte_eth_dev structure.
1058 * @param[in,out] res
1059 * Pointer to the modify-header resource.
1061 * Pointer to action specification.
1063 * Pointer to the error structure.
1066 * 0 on success, a negative errno value otherwise and rte_errno is set.
1069 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1070 struct mlx5_flow_dv_modify_hdr_resource *res,
1071 const struct rte_flow_action *action,
1072 struct rte_flow_error *error)
1074 const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1075 rte_be32_t mask = RTE_BE32(UINT32_MAX);
1076 struct rte_flow_item item = {
1080 struct field_modify_info reg_src[] = {
1081 {4, 0, reg_to_field[conf->src]},
1084 struct field_modify_info reg_dst = {
1086 .id = reg_to_field[conf->dst],
1088 /* Adjust reg_c[0] usage according to reported mask. */
1089 if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1090 struct mlx5_priv *priv = dev->data->dev_private;
1091 uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1093 MLX5_ASSERT(reg_c0);
1094 MLX5_ASSERT(priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY);
1095 if (conf->dst == REG_C_0) {
1096 /* Copy to reg_c[0], within mask only. */
1097 reg_dst.offset = rte_bsf32(reg_c0);
1099 * Mask is ignoring the enianness, because
1100 * there is no conversion in datapath.
1102 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1103 /* Copy from destination lower bits to reg_c[0]. */
1104 mask = reg_c0 >> reg_dst.offset;
1106 /* Copy from destination upper bits to reg_c[0]. */
1107 mask = reg_c0 << (sizeof(reg_c0) * CHAR_BIT -
1108 rte_fls_u32(reg_c0));
1111 mask = rte_cpu_to_be_32(reg_c0);
1112 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1113 /* Copy from reg_c[0] to destination lower bits. */
1116 /* Copy from reg_c[0] to destination upper bits. */
1117 reg_dst.offset = sizeof(reg_c0) * CHAR_BIT -
1118 (rte_fls_u32(reg_c0) -
1123 return flow_dv_convert_modify_action(&item,
1124 reg_src, ®_dst, res,
1125 MLX5_MODIFICATION_TYPE_COPY,
1130 * Convert MARK action to DV specification. This routine is used
1131 * in extensive metadata only and requires metadata register to be
1132 * handled. In legacy mode hardware tag resource is engaged.
1135 * Pointer to the rte_eth_dev structure.
1137 * Pointer to MARK action specification.
1138 * @param[in,out] resource
1139 * Pointer to the modify-header resource.
1141 * Pointer to the error structure.
1144 * 0 on success, a negative errno value otherwise and rte_errno is set.
1147 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1148 const struct rte_flow_action_mark *conf,
1149 struct mlx5_flow_dv_modify_hdr_resource *resource,
1150 struct rte_flow_error *error)
1152 struct mlx5_priv *priv = dev->data->dev_private;
1153 rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1154 priv->sh->dv_mark_mask);
1155 rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1156 struct rte_flow_item item = {
1160 struct field_modify_info reg_c_x[] = {
1166 return rte_flow_error_set(error, EINVAL,
1167 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1168 NULL, "zero mark action mask");
1169 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1172 MLX5_ASSERT(reg > 0);
1173 if (reg == REG_C_0) {
1174 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1175 uint32_t shl_c0 = rte_bsf32(msk_c0);
1177 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1178 mask = rte_cpu_to_be_32(mask) & msk_c0;
1179 mask = rte_cpu_to_be_32(mask << shl_c0);
1181 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1182 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1183 MLX5_MODIFICATION_TYPE_SET, error);
1187 * Get metadata register index for specified steering domain.
1190 * Pointer to the rte_eth_dev structure.
1192 * Attributes of flow to determine steering domain.
1194 * Pointer to the error structure.
1197 * positive index on success, a negative errno value otherwise
1198 * and rte_errno is set.
1200 static enum modify_reg
1201 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1202 const struct rte_flow_attr *attr,
1203 struct rte_flow_error *error)
1206 mlx5_flow_get_reg_id(dev, attr->transfer ?
1210 MLX5_METADATA_RX, 0, error);
1212 return rte_flow_error_set(error,
1213 ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1214 NULL, "unavailable "
1215 "metadata register");
1220 * Convert SET_META action to DV specification.
1223 * Pointer to the rte_eth_dev structure.
1224 * @param[in,out] resource
1225 * Pointer to the modify-header resource.
1227 * Attributes of flow that includes this item.
1229 * Pointer to action specification.
1231 * Pointer to the error structure.
1234 * 0 on success, a negative errno value otherwise and rte_errno is set.
1237 flow_dv_convert_action_set_meta
1238 (struct rte_eth_dev *dev,
1239 struct mlx5_flow_dv_modify_hdr_resource *resource,
1240 const struct rte_flow_attr *attr,
1241 const struct rte_flow_action_set_meta *conf,
1242 struct rte_flow_error *error)
1244 uint32_t data = conf->data;
1245 uint32_t mask = conf->mask;
1246 struct rte_flow_item item = {
1250 struct field_modify_info reg_c_x[] = {
1253 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1258 * In datapath code there is no endianness
1259 * coversions for perfromance reasons, all
1260 * pattern conversions are done in rte_flow.
1262 if (reg == REG_C_0) {
1263 struct mlx5_priv *priv = dev->data->dev_private;
1264 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1267 MLX5_ASSERT(msk_c0);
1268 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1269 shl_c0 = rte_bsf32(msk_c0);
1271 shl_c0 = sizeof(msk_c0) * CHAR_BIT - rte_fls_u32(msk_c0);
1275 MLX5_ASSERT(!(~msk_c0 & rte_cpu_to_be_32(mask)));
1277 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1278 /* The routine expects parameters in memory as big-endian ones. */
1279 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1280 MLX5_MODIFICATION_TYPE_SET, error);
1284 * Convert modify-header set IPv4 DSCP action to DV specification.
1286 * @param[in,out] resource
1287 * Pointer to the modify-header resource.
1289 * Pointer to action specification.
1291 * Pointer to the error structure.
1294 * 0 on success, a negative errno value otherwise and rte_errno is set.
1297 flow_dv_convert_action_modify_ipv4_dscp
1298 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1299 const struct rte_flow_action *action,
1300 struct rte_flow_error *error)
1302 const struct rte_flow_action_set_dscp *conf =
1303 (const struct rte_flow_action_set_dscp *)(action->conf);
1304 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1305 struct rte_flow_item_ipv4 ipv4;
1306 struct rte_flow_item_ipv4 ipv4_mask;
1308 memset(&ipv4, 0, sizeof(ipv4));
1309 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1310 ipv4.hdr.type_of_service = conf->dscp;
1311 ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1313 item.mask = &ipv4_mask;
1314 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1315 MLX5_MODIFICATION_TYPE_SET, error);
1319 * Convert modify-header set IPv6 DSCP action to DV specification.
1321 * @param[in,out] resource
1322 * Pointer to the modify-header resource.
1324 * Pointer to action specification.
1326 * Pointer to the error structure.
1329 * 0 on success, a negative errno value otherwise and rte_errno is set.
1332 flow_dv_convert_action_modify_ipv6_dscp
1333 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1334 const struct rte_flow_action *action,
1335 struct rte_flow_error *error)
1337 const struct rte_flow_action_set_dscp *conf =
1338 (const struct rte_flow_action_set_dscp *)(action->conf);
1339 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1340 struct rte_flow_item_ipv6 ipv6;
1341 struct rte_flow_item_ipv6 ipv6_mask;
1343 memset(&ipv6, 0, sizeof(ipv6));
1344 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1346 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1347 * rdma-core only accept the DSCP bits byte aligned start from
1348 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1349 * bits in IPv6 case as rdma-core requires byte aligned value.
1351 ipv6.hdr.vtc_flow = conf->dscp;
1352 ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1354 item.mask = &ipv6_mask;
1355 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1356 MLX5_MODIFICATION_TYPE_SET, error);
1360 * Validate MARK item.
1363 * Pointer to the rte_eth_dev structure.
1365 * Item specification.
1367 * Attributes of flow that includes this item.
1369 * Pointer to error structure.
1372 * 0 on success, a negative errno value otherwise and rte_errno is set.
1375 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1376 const struct rte_flow_item *item,
1377 const struct rte_flow_attr *attr __rte_unused,
1378 struct rte_flow_error *error)
1380 struct mlx5_priv *priv = dev->data->dev_private;
1381 struct mlx5_dev_config *config = &priv->config;
1382 const struct rte_flow_item_mark *spec = item->spec;
1383 const struct rte_flow_item_mark *mask = item->mask;
1384 const struct rte_flow_item_mark nic_mask = {
1385 .id = priv->sh->dv_mark_mask,
1389 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1390 return rte_flow_error_set(error, ENOTSUP,
1391 RTE_FLOW_ERROR_TYPE_ITEM, item,
1392 "extended metadata feature"
1394 if (!mlx5_flow_ext_mreg_supported(dev))
1395 return rte_flow_error_set(error, ENOTSUP,
1396 RTE_FLOW_ERROR_TYPE_ITEM, item,
1397 "extended metadata register"
1398 " isn't supported");
1400 return rte_flow_error_set(error, ENOTSUP,
1401 RTE_FLOW_ERROR_TYPE_ITEM, item,
1402 "extended metadata register"
1403 " isn't available");
1404 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1408 return rte_flow_error_set(error, EINVAL,
1409 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1411 "data cannot be empty");
1412 if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1413 return rte_flow_error_set(error, EINVAL,
1414 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1416 "mark id exceeds the limit");
1420 return rte_flow_error_set(error, EINVAL,
1421 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1422 "mask cannot be zero");
1424 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1425 (const uint8_t *)&nic_mask,
1426 sizeof(struct rte_flow_item_mark),
1427 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1434 * Validate META item.
1437 * Pointer to the rte_eth_dev structure.
1439 * Item specification.
1441 * Attributes of flow that includes this item.
1443 * Pointer to error structure.
1446 * 0 on success, a negative errno value otherwise and rte_errno is set.
1449 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1450 const struct rte_flow_item *item,
1451 const struct rte_flow_attr *attr,
1452 struct rte_flow_error *error)
1454 struct mlx5_priv *priv = dev->data->dev_private;
1455 struct mlx5_dev_config *config = &priv->config;
1456 const struct rte_flow_item_meta *spec = item->spec;
1457 const struct rte_flow_item_meta *mask = item->mask;
1458 struct rte_flow_item_meta nic_mask = {
1465 return rte_flow_error_set(error, EINVAL,
1466 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1468 "data cannot be empty");
1469 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1470 if (!mlx5_flow_ext_mreg_supported(dev))
1471 return rte_flow_error_set(error, ENOTSUP,
1472 RTE_FLOW_ERROR_TYPE_ITEM, item,
1473 "extended metadata register"
1474 " isn't supported");
1475 reg = flow_dv_get_metadata_reg(dev, attr, error);
1479 return rte_flow_error_set(error, ENOTSUP,
1480 RTE_FLOW_ERROR_TYPE_ITEM, item,
1484 nic_mask.data = priv->sh->dv_meta_mask;
1485 } else if (attr->transfer) {
1486 return rte_flow_error_set(error, ENOTSUP,
1487 RTE_FLOW_ERROR_TYPE_ITEM, item,
1488 "extended metadata feature "
1489 "should be enabled when "
1490 "meta item is requested "
1491 "with e-switch mode ");
1494 mask = &rte_flow_item_meta_mask;
1496 return rte_flow_error_set(error, EINVAL,
1497 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1498 "mask cannot be zero");
1500 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1501 (const uint8_t *)&nic_mask,
1502 sizeof(struct rte_flow_item_meta),
1503 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1508 * Validate TAG item.
1511 * Pointer to the rte_eth_dev structure.
1513 * Item specification.
1515 * Attributes of flow that includes this item.
1517 * Pointer to error structure.
1520 * 0 on success, a negative errno value otherwise and rte_errno is set.
1523 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
1524 const struct rte_flow_item *item,
1525 const struct rte_flow_attr *attr __rte_unused,
1526 struct rte_flow_error *error)
1528 const struct rte_flow_item_tag *spec = item->spec;
1529 const struct rte_flow_item_tag *mask = item->mask;
1530 const struct rte_flow_item_tag nic_mask = {
1531 .data = RTE_BE32(UINT32_MAX),
1536 if (!mlx5_flow_ext_mreg_supported(dev))
1537 return rte_flow_error_set(error, ENOTSUP,
1538 RTE_FLOW_ERROR_TYPE_ITEM, item,
1539 "extensive metadata register"
1540 " isn't supported");
1542 return rte_flow_error_set(error, EINVAL,
1543 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1545 "data cannot be empty");
1547 mask = &rte_flow_item_tag_mask;
1549 return rte_flow_error_set(error, EINVAL,
1550 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1551 "mask cannot be zero");
1553 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1554 (const uint8_t *)&nic_mask,
1555 sizeof(struct rte_flow_item_tag),
1556 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1559 if (mask->index != 0xff)
1560 return rte_flow_error_set(error, EINVAL,
1561 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1562 "partial mask for tag index"
1563 " is not supported");
1564 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
1567 MLX5_ASSERT(ret != REG_NON);
1572 * Validate vport item.
1575 * Pointer to the rte_eth_dev structure.
1577 * Item specification.
1579 * Attributes of flow that includes this item.
1580 * @param[in] item_flags
1581 * Bit-fields that holds the items detected until now.
1583 * Pointer to error structure.
1586 * 0 on success, a negative errno value otherwise and rte_errno is set.
1589 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
1590 const struct rte_flow_item *item,
1591 const struct rte_flow_attr *attr,
1592 uint64_t item_flags,
1593 struct rte_flow_error *error)
1595 const struct rte_flow_item_port_id *spec = item->spec;
1596 const struct rte_flow_item_port_id *mask = item->mask;
1597 const struct rte_flow_item_port_id switch_mask = {
1600 struct mlx5_priv *esw_priv;
1601 struct mlx5_priv *dev_priv;
1604 if (!attr->transfer)
1605 return rte_flow_error_set(error, EINVAL,
1606 RTE_FLOW_ERROR_TYPE_ITEM,
1608 "match on port id is valid only"
1609 " when transfer flag is enabled");
1610 if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
1611 return rte_flow_error_set(error, ENOTSUP,
1612 RTE_FLOW_ERROR_TYPE_ITEM, item,
1613 "multiple source ports are not"
1616 mask = &switch_mask;
1617 if (mask->id != 0xffffffff)
1618 return rte_flow_error_set(error, ENOTSUP,
1619 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1621 "no support for partial mask on"
1623 ret = mlx5_flow_item_acceptable
1624 (item, (const uint8_t *)mask,
1625 (const uint8_t *)&rte_flow_item_port_id_mask,
1626 sizeof(struct rte_flow_item_port_id),
1627 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1632 esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
1634 return rte_flow_error_set(error, rte_errno,
1635 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1636 "failed to obtain E-Switch info for"
1638 dev_priv = mlx5_dev_to_eswitch_info(dev);
1640 return rte_flow_error_set(error, rte_errno,
1641 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1643 "failed to obtain E-Switch info");
1644 if (esw_priv->domain_id != dev_priv->domain_id)
1645 return rte_flow_error_set(error, EINVAL,
1646 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1647 "cannot match on a port from a"
1648 " different E-Switch");
1653 * Validate VLAN item.
1656 * Item specification.
1657 * @param[in] item_flags
1658 * Bit-fields that holds the items detected until now.
1660 * Ethernet device flow is being created on.
1662 * Pointer to error structure.
1665 * 0 on success, a negative errno value otherwise and rte_errno is set.
1668 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
1669 uint64_t item_flags,
1670 struct rte_eth_dev *dev,
1671 struct rte_flow_error *error)
1673 const struct rte_flow_item_vlan *mask = item->mask;
1674 const struct rte_flow_item_vlan nic_mask = {
1675 .tci = RTE_BE16(UINT16_MAX),
1676 .inner_type = RTE_BE16(UINT16_MAX),
1679 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1681 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1682 MLX5_FLOW_LAYER_INNER_L4) :
1683 (MLX5_FLOW_LAYER_OUTER_L3 |
1684 MLX5_FLOW_LAYER_OUTER_L4);
1685 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1686 MLX5_FLOW_LAYER_OUTER_VLAN;
1688 if (item_flags & vlanm)
1689 return rte_flow_error_set(error, EINVAL,
1690 RTE_FLOW_ERROR_TYPE_ITEM, item,
1691 "multiple VLAN layers not supported");
1692 else if ((item_flags & l34m) != 0)
1693 return rte_flow_error_set(error, EINVAL,
1694 RTE_FLOW_ERROR_TYPE_ITEM, item,
1695 "VLAN cannot follow L3/L4 layer");
1697 mask = &rte_flow_item_vlan_mask;
1698 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1699 (const uint8_t *)&nic_mask,
1700 sizeof(struct rte_flow_item_vlan),
1701 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1704 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1705 struct mlx5_priv *priv = dev->data->dev_private;
1707 if (priv->vmwa_context) {
1709 * Non-NULL context means we have a virtual machine
1710 * and SR-IOV enabled, we have to create VLAN interface
1711 * to make hypervisor to setup E-Switch vport
1712 * context correctly. We avoid creating the multiple
1713 * VLAN interfaces, so we cannot support VLAN tag mask.
1715 return rte_flow_error_set(error, EINVAL,
1716 RTE_FLOW_ERROR_TYPE_ITEM,
1718 "VLAN tag mask is not"
1719 " supported in virtual"
1727 * GTP flags are contained in 1 byte of the format:
1728 * -------------------------------------------
1729 * | bit | 0 - 2 | 3 | 4 | 5 | 6 | 7 |
1730 * |-----------------------------------------|
1731 * | value | Version | PT | Res | E | S | PN |
1732 * -------------------------------------------
1734 * Matching is supported only for GTP flags E, S, PN.
1736 #define MLX5_GTP_FLAGS_MASK 0x07
1739 * Validate GTP item.
1742 * Pointer to the rte_eth_dev structure.
1744 * Item specification.
1745 * @param[in] item_flags
1746 * Bit-fields that holds the items detected until now.
1748 * Pointer to error structure.
1751 * 0 on success, a negative errno value otherwise and rte_errno is set.
1754 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1755 const struct rte_flow_item *item,
1756 uint64_t item_flags,
1757 struct rte_flow_error *error)
1759 struct mlx5_priv *priv = dev->data->dev_private;
1760 const struct rte_flow_item_gtp *spec = item->spec;
1761 const struct rte_flow_item_gtp *mask = item->mask;
1762 const struct rte_flow_item_gtp nic_mask = {
1763 .v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
1765 .teid = RTE_BE32(0xffffffff),
1768 if (!priv->config.hca_attr.tunnel_stateless_gtp)
1769 return rte_flow_error_set(error, ENOTSUP,
1770 RTE_FLOW_ERROR_TYPE_ITEM, item,
1771 "GTP support is not enabled");
1772 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1773 return rte_flow_error_set(error, ENOTSUP,
1774 RTE_FLOW_ERROR_TYPE_ITEM, item,
1775 "multiple tunnel layers not"
1777 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1778 return rte_flow_error_set(error, EINVAL,
1779 RTE_FLOW_ERROR_TYPE_ITEM, item,
1780 "no outer UDP layer found");
1782 mask = &rte_flow_item_gtp_mask;
1783 if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
1784 return rte_flow_error_set(error, ENOTSUP,
1785 RTE_FLOW_ERROR_TYPE_ITEM, item,
1786 "Match is supported for GTP"
1788 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1789 (const uint8_t *)&nic_mask,
1790 sizeof(struct rte_flow_item_gtp),
1791 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1795 * Validate IPV4 item.
1796 * Use existing validation function mlx5_flow_validate_item_ipv4(), and
1797 * add specific validation of fragment_offset field,
1800 * Item specification.
1801 * @param[in] item_flags
1802 * Bit-fields that holds the items detected until now.
1804 * Pointer to error structure.
1807 * 0 on success, a negative errno value otherwise and rte_errno is set.
1810 flow_dv_validate_item_ipv4(const struct rte_flow_item *item,
1811 uint64_t item_flags,
1813 uint16_t ether_type,
1814 struct rte_flow_error *error)
1817 const struct rte_flow_item_ipv4 *spec = item->spec;
1818 const struct rte_flow_item_ipv4 *last = item->last;
1819 const struct rte_flow_item_ipv4 *mask = item->mask;
1820 rte_be16_t fragment_offset_spec = 0;
1821 rte_be16_t fragment_offset_last = 0;
1822 const struct rte_flow_item_ipv4 nic_ipv4_mask = {
1824 .src_addr = RTE_BE32(0xffffffff),
1825 .dst_addr = RTE_BE32(0xffffffff),
1826 .type_of_service = 0xff,
1827 .fragment_offset = RTE_BE16(0xffff),
1828 .next_proto_id = 0xff,
1829 .time_to_live = 0xff,
1833 ret = mlx5_flow_validate_item_ipv4(item, item_flags, last_item,
1834 ether_type, &nic_ipv4_mask,
1835 MLX5_ITEM_RANGE_ACCEPTED, error);
1839 fragment_offset_spec = spec->hdr.fragment_offset &
1840 mask->hdr.fragment_offset;
1841 if (!fragment_offset_spec)
1844 * spec and mask are valid, enforce using full mask to make sure the
1845 * complete value is used correctly.
1847 if ((mask->hdr.fragment_offset & RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1848 != RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1849 return rte_flow_error_set(error, EINVAL,
1850 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1851 item, "must use full mask for"
1852 " fragment_offset");
1854 * Match on fragment_offset 0x2000 means MF is 1 and frag-offset is 0,
1855 * indicating this is 1st fragment of fragmented packet.
1856 * This is not yet supported in MLX5, return appropriate error message.
1858 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG))
1859 return rte_flow_error_set(error, ENOTSUP,
1860 RTE_FLOW_ERROR_TYPE_ITEM, item,
1861 "match on first fragment not "
1863 if (fragment_offset_spec && !last)
1864 return rte_flow_error_set(error, ENOTSUP,
1865 RTE_FLOW_ERROR_TYPE_ITEM, item,
1866 "specified value not supported");
1867 /* spec and last are valid, validate the specified range. */
1868 fragment_offset_last = last->hdr.fragment_offset &
1869 mask->hdr.fragment_offset;
1871 * Match on fragment_offset spec 0x2001 and last 0x3fff
1872 * means MF is 1 and frag-offset is > 0.
1873 * This packet is fragment 2nd and onward, excluding last.
1874 * This is not yet supported in MLX5, return appropriate
1877 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG + 1) &&
1878 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1879 return rte_flow_error_set(error, ENOTSUP,
1880 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1881 last, "match on following "
1882 "fragments not supported");
1884 * Match on fragment_offset spec 0x0001 and last 0x1fff
1885 * means MF is 0 and frag-offset is > 0.
1886 * This packet is last fragment of fragmented packet.
1887 * This is not yet supported in MLX5, return appropriate
1890 if (fragment_offset_spec == RTE_BE16(1) &&
1891 fragment_offset_last == RTE_BE16(RTE_IPV4_HDR_OFFSET_MASK))
1892 return rte_flow_error_set(error, ENOTSUP,
1893 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1894 last, "match on last "
1895 "fragment not supported");
1897 * Match on fragment_offset spec 0x0001 and last 0x3fff
1898 * means MF and/or frag-offset is not 0.
1899 * This is a fragmented packet.
1900 * Other range values are invalid and rejected.
1902 if (!(fragment_offset_spec == RTE_BE16(1) &&
1903 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK)))
1904 return rte_flow_error_set(error, ENOTSUP,
1905 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
1906 "specified range not supported");
1911 * Validate IPV6 fragment extension item.
1914 * Item specification.
1915 * @param[in] item_flags
1916 * Bit-fields that holds the items detected until now.
1918 * Pointer to error structure.
1921 * 0 on success, a negative errno value otherwise and rte_errno is set.
1924 flow_dv_validate_item_ipv6_frag_ext(const struct rte_flow_item *item,
1925 uint64_t item_flags,
1926 struct rte_flow_error *error)
1928 const struct rte_flow_item_ipv6_frag_ext *spec = item->spec;
1929 const struct rte_flow_item_ipv6_frag_ext *last = item->last;
1930 const struct rte_flow_item_ipv6_frag_ext *mask = item->mask;
1931 rte_be16_t frag_data_spec = 0;
1932 rte_be16_t frag_data_last = 0;
1933 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1934 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1935 MLX5_FLOW_LAYER_OUTER_L4;
1937 struct rte_flow_item_ipv6_frag_ext nic_mask = {
1939 .next_header = 0xff,
1940 .frag_data = RTE_BE16(0xffff),
1944 if (item_flags & l4m)
1945 return rte_flow_error_set(error, EINVAL,
1946 RTE_FLOW_ERROR_TYPE_ITEM, item,
1947 "ipv6 fragment extension item cannot "
1949 if ((tunnel && !(item_flags & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
1950 (!tunnel && !(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
1951 return rte_flow_error_set(error, EINVAL,
1952 RTE_FLOW_ERROR_TYPE_ITEM, item,
1953 "ipv6 fragment extension item must "
1954 "follow ipv6 item");
1956 frag_data_spec = spec->hdr.frag_data & mask->hdr.frag_data;
1957 if (!frag_data_spec)
1960 * spec and mask are valid, enforce using full mask to make sure the
1961 * complete value is used correctly.
1963 if ((mask->hdr.frag_data & RTE_BE16(RTE_IPV6_FRAG_USED_MASK)) !=
1964 RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
1965 return rte_flow_error_set(error, EINVAL,
1966 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1967 item, "must use full mask for"
1970 * Match on frag_data 0x00001 means M is 1 and frag-offset is 0.
1971 * This is 1st fragment of fragmented packet.
1973 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_MF_MASK))
1974 return rte_flow_error_set(error, ENOTSUP,
1975 RTE_FLOW_ERROR_TYPE_ITEM, item,
1976 "match on first fragment not "
1978 if (frag_data_spec && !last)
1979 return rte_flow_error_set(error, EINVAL,
1980 RTE_FLOW_ERROR_TYPE_ITEM, item,
1981 "specified value not supported");
1982 ret = mlx5_flow_item_acceptable
1983 (item, (const uint8_t *)mask,
1984 (const uint8_t *)&nic_mask,
1985 sizeof(struct rte_flow_item_ipv6_frag_ext),
1986 MLX5_ITEM_RANGE_ACCEPTED, error);
1989 /* spec and last are valid, validate the specified range. */
1990 frag_data_last = last->hdr.frag_data & mask->hdr.frag_data;
1992 * Match on frag_data spec 0x0009 and last 0xfff9
1993 * means M is 1 and frag-offset is > 0.
1994 * This packet is fragment 2nd and onward, excluding last.
1995 * This is not yet supported in MLX5, return appropriate
1998 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN |
1999 RTE_IPV6_EHDR_MF_MASK) &&
2000 frag_data_last == RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
2001 return rte_flow_error_set(error, ENOTSUP,
2002 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2003 last, "match on following "
2004 "fragments not supported");
2006 * Match on frag_data spec 0x0008 and last 0xfff8
2007 * means M is 0 and frag-offset is > 0.
2008 * This packet is last fragment of fragmented packet.
2009 * This is not yet supported in MLX5, return appropriate
2012 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN) &&
2013 frag_data_last == RTE_BE16(RTE_IPV6_EHDR_FO_MASK))
2014 return rte_flow_error_set(error, ENOTSUP,
2015 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2016 last, "match on last "
2017 "fragment not supported");
2018 /* Other range values are invalid and rejected. */
2019 return rte_flow_error_set(error, EINVAL,
2020 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
2021 "specified range not supported");
2025 * Validate the pop VLAN action.
2028 * Pointer to the rte_eth_dev structure.
2029 * @param[in] action_flags
2030 * Holds the actions detected until now.
2032 * Pointer to the pop vlan action.
2033 * @param[in] item_flags
2034 * The items found in this flow rule.
2036 * Pointer to flow attributes.
2038 * Pointer to error structure.
2041 * 0 on success, a negative errno value otherwise and rte_errno is set.
2044 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
2045 uint64_t action_flags,
2046 const struct rte_flow_action *action,
2047 uint64_t item_flags,
2048 const struct rte_flow_attr *attr,
2049 struct rte_flow_error *error)
2051 const struct mlx5_priv *priv = dev->data->dev_private;
2055 if (!priv->sh->pop_vlan_action)
2056 return rte_flow_error_set(error, ENOTSUP,
2057 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2059 "pop vlan action is not supported");
2061 return rte_flow_error_set(error, ENOTSUP,
2062 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2064 "pop vlan action not supported for "
2066 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
2067 return rte_flow_error_set(error, ENOTSUP,
2068 RTE_FLOW_ERROR_TYPE_ACTION, action,
2069 "no support for multiple VLAN "
2071 /* Pop VLAN with preceding Decap requires inner header with VLAN. */
2072 if ((action_flags & MLX5_FLOW_ACTION_DECAP) &&
2073 !(item_flags & MLX5_FLOW_LAYER_INNER_VLAN))
2074 return rte_flow_error_set(error, ENOTSUP,
2075 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2077 "cannot pop vlan after decap without "
2078 "match on inner vlan in the flow");
2079 /* Pop VLAN without preceding Decap requires outer header with VLAN. */
2080 if (!(action_flags & MLX5_FLOW_ACTION_DECAP) &&
2081 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2082 return rte_flow_error_set(error, ENOTSUP,
2083 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2085 "cannot pop vlan without a "
2086 "match on (outer) vlan in the flow");
2087 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2088 return rte_flow_error_set(error, EINVAL,
2089 RTE_FLOW_ERROR_TYPE_ACTION, action,
2090 "wrong action order, port_id should "
2091 "be after pop VLAN action");
2092 if (!attr->transfer && priv->representor)
2093 return rte_flow_error_set(error, ENOTSUP,
2094 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2095 "pop vlan action for VF representor "
2096 "not supported on NIC table");
2101 * Get VLAN default info from vlan match info.
2104 * the list of item specifications.
2106 * pointer VLAN info to fill to.
2109 * 0 on success, a negative errno value otherwise and rte_errno is set.
2112 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
2113 struct rte_vlan_hdr *vlan)
2115 const struct rte_flow_item_vlan nic_mask = {
2116 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
2117 MLX5DV_FLOW_VLAN_VID_MASK),
2118 .inner_type = RTE_BE16(0xffff),
2123 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2124 int type = items->type;
2126 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
2127 type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
2130 if (items->type != RTE_FLOW_ITEM_TYPE_END) {
2131 const struct rte_flow_item_vlan *vlan_m = items->mask;
2132 const struct rte_flow_item_vlan *vlan_v = items->spec;
2134 /* If VLAN item in pattern doesn't contain data, return here. */
2139 /* Only full match values are accepted */
2140 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
2141 MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
2142 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
2144 rte_be_to_cpu_16(vlan_v->tci &
2145 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
2147 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
2148 MLX5DV_FLOW_VLAN_VID_MASK_BE) {
2149 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
2151 rte_be_to_cpu_16(vlan_v->tci &
2152 MLX5DV_FLOW_VLAN_VID_MASK_BE);
2154 if (vlan_m->inner_type == nic_mask.inner_type)
2155 vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
2156 vlan_m->inner_type);
2161 * Validate the push VLAN action.
2164 * Pointer to the rte_eth_dev structure.
2165 * @param[in] action_flags
2166 * Holds the actions detected until now.
2167 * @param[in] item_flags
2168 * The items found in this flow rule.
2170 * Pointer to the action structure.
2172 * Pointer to flow attributes
2174 * Pointer to error structure.
2177 * 0 on success, a negative errno value otherwise and rte_errno is set.
2180 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
2181 uint64_t action_flags,
2182 const struct rte_flow_item_vlan *vlan_m,
2183 const struct rte_flow_action *action,
2184 const struct rte_flow_attr *attr,
2185 struct rte_flow_error *error)
2187 const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
2188 const struct mlx5_priv *priv = dev->data->dev_private;
2190 if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
2191 push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
2192 return rte_flow_error_set(error, EINVAL,
2193 RTE_FLOW_ERROR_TYPE_ACTION, action,
2194 "invalid vlan ethertype");
2195 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2196 return rte_flow_error_set(error, EINVAL,
2197 RTE_FLOW_ERROR_TYPE_ACTION, action,
2198 "wrong action order, port_id should "
2199 "be after push VLAN");
2200 if (!attr->transfer && priv->representor)
2201 return rte_flow_error_set(error, ENOTSUP,
2202 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2203 "push vlan action for VF representor "
2204 "not supported on NIC table");
2206 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
2207 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
2208 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
2209 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
2210 !(mlx5_flow_find_action
2211 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
2212 return rte_flow_error_set(error, EINVAL,
2213 RTE_FLOW_ERROR_TYPE_ACTION, action,
2214 "not full match mask on VLAN PCP and "
2215 "there is no of_set_vlan_pcp action, "
2216 "push VLAN action cannot figure out "
2219 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
2220 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
2221 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
2222 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
2223 !(mlx5_flow_find_action
2224 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
2225 return rte_flow_error_set(error, EINVAL,
2226 RTE_FLOW_ERROR_TYPE_ACTION, action,
2227 "not full match mask on VLAN VID and "
2228 "there is no of_set_vlan_vid action, "
2229 "push VLAN action cannot figure out "
2236 * Validate the set VLAN PCP.
2238 * @param[in] action_flags
2239 * Holds the actions detected until now.
2240 * @param[in] actions
2241 * Pointer to the list of actions remaining in the flow rule.
2243 * Pointer to error structure.
2246 * 0 on success, a negative errno value otherwise and rte_errno is set.
2249 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
2250 const struct rte_flow_action actions[],
2251 struct rte_flow_error *error)
2253 const struct rte_flow_action *action = actions;
2254 const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2256 if (conf->vlan_pcp > 7)
2257 return rte_flow_error_set(error, EINVAL,
2258 RTE_FLOW_ERROR_TYPE_ACTION, action,
2259 "VLAN PCP value is too big");
2260 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2261 return rte_flow_error_set(error, ENOTSUP,
2262 RTE_FLOW_ERROR_TYPE_ACTION, action,
2263 "set VLAN PCP action must follow "
2264 "the push VLAN action");
2265 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2266 return rte_flow_error_set(error, ENOTSUP,
2267 RTE_FLOW_ERROR_TYPE_ACTION, action,
2268 "Multiple VLAN PCP modification are "
2270 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2271 return rte_flow_error_set(error, EINVAL,
2272 RTE_FLOW_ERROR_TYPE_ACTION, action,
2273 "wrong action order, port_id should "
2274 "be after set VLAN PCP");
2279 * Validate the set VLAN VID.
2281 * @param[in] item_flags
2282 * Holds the items detected in this rule.
2283 * @param[in] action_flags
2284 * Holds the actions detected until now.
2285 * @param[in] actions
2286 * Pointer to the list of actions remaining in the flow rule.
2288 * Pointer to error structure.
2291 * 0 on success, a negative errno value otherwise and rte_errno is set.
2294 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
2295 uint64_t action_flags,
2296 const struct rte_flow_action actions[],
2297 struct rte_flow_error *error)
2299 const struct rte_flow_action *action = actions;
2300 const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
2302 if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
2303 return rte_flow_error_set(error, EINVAL,
2304 RTE_FLOW_ERROR_TYPE_ACTION, action,
2305 "VLAN VID value is too big");
2306 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2307 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2308 return rte_flow_error_set(error, ENOTSUP,
2309 RTE_FLOW_ERROR_TYPE_ACTION, action,
2310 "set VLAN VID action must follow push"
2311 " VLAN action or match on VLAN item");
2312 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
2313 return rte_flow_error_set(error, ENOTSUP,
2314 RTE_FLOW_ERROR_TYPE_ACTION, action,
2315 "Multiple VLAN VID modifications are "
2317 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2318 return rte_flow_error_set(error, EINVAL,
2319 RTE_FLOW_ERROR_TYPE_ACTION, action,
2320 "wrong action order, port_id should "
2321 "be after set VLAN VID");
2326 * Validate the FLAG action.
2329 * Pointer to the rte_eth_dev structure.
2330 * @param[in] action_flags
2331 * Holds the actions detected until now.
2333 * Pointer to flow attributes
2335 * Pointer to error structure.
2338 * 0 on success, a negative errno value otherwise and rte_errno is set.
2341 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2342 uint64_t action_flags,
2343 const struct rte_flow_attr *attr,
2344 struct rte_flow_error *error)
2346 struct mlx5_priv *priv = dev->data->dev_private;
2347 struct mlx5_dev_config *config = &priv->config;
2350 /* Fall back if no extended metadata register support. */
2351 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2352 return mlx5_flow_validate_action_flag(action_flags, attr,
2354 /* Extensive metadata mode requires registers. */
2355 if (!mlx5_flow_ext_mreg_supported(dev))
2356 return rte_flow_error_set(error, ENOTSUP,
2357 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2358 "no metadata registers "
2359 "to support flag action");
2360 if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2361 return rte_flow_error_set(error, ENOTSUP,
2362 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2363 "extended metadata register"
2364 " isn't available");
2365 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2368 MLX5_ASSERT(ret > 0);
2369 if (action_flags & MLX5_FLOW_ACTION_MARK)
2370 return rte_flow_error_set(error, EINVAL,
2371 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2372 "can't mark and flag in same flow");
2373 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2374 return rte_flow_error_set(error, EINVAL,
2375 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2377 " actions in same flow");
2382 * Validate MARK action.
2385 * Pointer to the rte_eth_dev structure.
2387 * Pointer to action.
2388 * @param[in] action_flags
2389 * Holds the actions detected until now.
2391 * Pointer to flow attributes
2393 * Pointer to error structure.
2396 * 0 on success, a negative errno value otherwise and rte_errno is set.
2399 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2400 const struct rte_flow_action *action,
2401 uint64_t action_flags,
2402 const struct rte_flow_attr *attr,
2403 struct rte_flow_error *error)
2405 struct mlx5_priv *priv = dev->data->dev_private;
2406 struct mlx5_dev_config *config = &priv->config;
2407 const struct rte_flow_action_mark *mark = action->conf;
2410 /* Fall back if no extended metadata register support. */
2411 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2412 return mlx5_flow_validate_action_mark(action, action_flags,
2414 /* Extensive metadata mode requires registers. */
2415 if (!mlx5_flow_ext_mreg_supported(dev))
2416 return rte_flow_error_set(error, ENOTSUP,
2417 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2418 "no metadata registers "
2419 "to support mark action");
2420 if (!priv->sh->dv_mark_mask)
2421 return rte_flow_error_set(error, ENOTSUP,
2422 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2423 "extended metadata register"
2424 " isn't available");
2425 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2428 MLX5_ASSERT(ret > 0);
2430 return rte_flow_error_set(error, EINVAL,
2431 RTE_FLOW_ERROR_TYPE_ACTION, action,
2432 "configuration cannot be null");
2433 if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2434 return rte_flow_error_set(error, EINVAL,
2435 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2437 "mark id exceeds the limit");
2438 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2439 return rte_flow_error_set(error, EINVAL,
2440 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2441 "can't flag and mark in same flow");
2442 if (action_flags & MLX5_FLOW_ACTION_MARK)
2443 return rte_flow_error_set(error, EINVAL,
2444 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2445 "can't have 2 mark actions in same"
2451 * Validate SET_META action.
2454 * Pointer to the rte_eth_dev structure.
2456 * Pointer to the action structure.
2457 * @param[in] action_flags
2458 * Holds the actions detected until now.
2460 * Pointer to flow attributes
2462 * Pointer to error structure.
2465 * 0 on success, a negative errno value otherwise and rte_errno is set.
2468 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2469 const struct rte_flow_action *action,
2470 uint64_t action_flags __rte_unused,
2471 const struct rte_flow_attr *attr,
2472 struct rte_flow_error *error)
2474 const struct rte_flow_action_set_meta *conf;
2475 uint32_t nic_mask = UINT32_MAX;
2478 if (!mlx5_flow_ext_mreg_supported(dev))
2479 return rte_flow_error_set(error, ENOTSUP,
2480 RTE_FLOW_ERROR_TYPE_ACTION, action,
2481 "extended metadata register"
2482 " isn't supported");
2483 reg = flow_dv_get_metadata_reg(dev, attr, error);
2486 if (reg != REG_A && reg != REG_B) {
2487 struct mlx5_priv *priv = dev->data->dev_private;
2489 nic_mask = priv->sh->dv_meta_mask;
2491 if (!(action->conf))
2492 return rte_flow_error_set(error, EINVAL,
2493 RTE_FLOW_ERROR_TYPE_ACTION, action,
2494 "configuration cannot be null");
2495 conf = (const struct rte_flow_action_set_meta *)action->conf;
2497 return rte_flow_error_set(error, EINVAL,
2498 RTE_FLOW_ERROR_TYPE_ACTION, action,
2499 "zero mask doesn't have any effect");
2500 if (conf->mask & ~nic_mask)
2501 return rte_flow_error_set(error, EINVAL,
2502 RTE_FLOW_ERROR_TYPE_ACTION, action,
2503 "meta data must be within reg C0");
2508 * Validate SET_TAG action.
2511 * Pointer to the rte_eth_dev structure.
2513 * Pointer to the action structure.
2514 * @param[in] action_flags
2515 * Holds the actions detected until now.
2517 * Pointer to flow attributes
2519 * Pointer to error structure.
2522 * 0 on success, a negative errno value otherwise and rte_errno is set.
2525 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2526 const struct rte_flow_action *action,
2527 uint64_t action_flags,
2528 const struct rte_flow_attr *attr,
2529 struct rte_flow_error *error)
2531 const struct rte_flow_action_set_tag *conf;
2532 const uint64_t terminal_action_flags =
2533 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2534 MLX5_FLOW_ACTION_RSS;
2537 if (!mlx5_flow_ext_mreg_supported(dev))
2538 return rte_flow_error_set(error, ENOTSUP,
2539 RTE_FLOW_ERROR_TYPE_ACTION, action,
2540 "extensive metadata register"
2541 " isn't supported");
2542 if (!(action->conf))
2543 return rte_flow_error_set(error, EINVAL,
2544 RTE_FLOW_ERROR_TYPE_ACTION, action,
2545 "configuration cannot be null");
2546 conf = (const struct rte_flow_action_set_tag *)action->conf;
2548 return rte_flow_error_set(error, EINVAL,
2549 RTE_FLOW_ERROR_TYPE_ACTION, action,
2550 "zero mask doesn't have any effect");
2551 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2554 if (!attr->transfer && attr->ingress &&
2555 (action_flags & terminal_action_flags))
2556 return rte_flow_error_set(error, EINVAL,
2557 RTE_FLOW_ERROR_TYPE_ACTION, action,
2558 "set_tag has no effect"
2559 " with terminal actions");
2564 * Validate count action.
2567 * Pointer to rte_eth_dev structure.
2569 * Pointer to error structure.
2572 * 0 on success, a negative errno value otherwise and rte_errno is set.
2575 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2576 struct rte_flow_error *error)
2578 struct mlx5_priv *priv = dev->data->dev_private;
2580 if (!priv->config.devx)
2582 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2586 return rte_flow_error_set
2588 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2590 "count action not supported");
2594 * Validate the L2 encap action.
2597 * Pointer to the rte_eth_dev structure.
2598 * @param[in] action_flags
2599 * Holds the actions detected until now.
2601 * Pointer to the action structure.
2603 * Pointer to flow attributes.
2605 * Pointer to error structure.
2608 * 0 on success, a negative errno value otherwise and rte_errno is set.
2611 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2612 uint64_t action_flags,
2613 const struct rte_flow_action *action,
2614 const struct rte_flow_attr *attr,
2615 struct rte_flow_error *error)
2617 const struct mlx5_priv *priv = dev->data->dev_private;
2619 if (!(action->conf))
2620 return rte_flow_error_set(error, EINVAL,
2621 RTE_FLOW_ERROR_TYPE_ACTION, action,
2622 "configuration cannot be null");
2623 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2624 return rte_flow_error_set(error, EINVAL,
2625 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2626 "can only have a single encap action "
2628 if (!attr->transfer && priv->representor)
2629 return rte_flow_error_set(error, ENOTSUP,
2630 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2631 "encap action for VF representor "
2632 "not supported on NIC table");
2637 * Validate a decap action.
2640 * Pointer to the rte_eth_dev structure.
2641 * @param[in] action_flags
2642 * Holds the actions detected until now.
2644 * Pointer to flow attributes
2646 * Pointer to error structure.
2649 * 0 on success, a negative errno value otherwise and rte_errno is set.
2652 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2653 uint64_t action_flags,
2654 const struct rte_flow_attr *attr,
2655 struct rte_flow_error *error)
2657 const struct mlx5_priv *priv = dev->data->dev_private;
2659 if (priv->config.hca_attr.scatter_fcs_w_decap_disable &&
2660 !priv->config.decap_en)
2661 return rte_flow_error_set(error, ENOTSUP,
2662 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2663 "decap is not enabled");
2664 if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2665 return rte_flow_error_set(error, ENOTSUP,
2666 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2668 MLX5_FLOW_ACTION_DECAP ? "can only "
2669 "have a single decap action" : "decap "
2670 "after encap is not supported");
2671 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2672 return rte_flow_error_set(error, EINVAL,
2673 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2674 "can't have decap action after"
2677 return rte_flow_error_set(error, ENOTSUP,
2678 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2680 "decap action not supported for "
2682 if (!attr->transfer && priv->representor)
2683 return rte_flow_error_set(error, ENOTSUP,
2684 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2685 "decap action for VF representor "
2686 "not supported on NIC table");
2690 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2693 * Validate the raw encap and decap actions.
2696 * Pointer to the rte_eth_dev structure.
2698 * Pointer to the decap action.
2700 * Pointer to the encap action.
2702 * Pointer to flow attributes
2703 * @param[in/out] action_flags
2704 * Holds the actions detected until now.
2705 * @param[out] actions_n
2706 * pointer to the number of actions counter.
2708 * Pointer to error structure.
2711 * 0 on success, a negative errno value otherwise and rte_errno is set.
2714 flow_dv_validate_action_raw_encap_decap
2715 (struct rte_eth_dev *dev,
2716 const struct rte_flow_action_raw_decap *decap,
2717 const struct rte_flow_action_raw_encap *encap,
2718 const struct rte_flow_attr *attr, uint64_t *action_flags,
2719 int *actions_n, struct rte_flow_error *error)
2721 const struct mlx5_priv *priv = dev->data->dev_private;
2724 if (encap && (!encap->size || !encap->data))
2725 return rte_flow_error_set(error, EINVAL,
2726 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2727 "raw encap data cannot be empty");
2728 if (decap && encap) {
2729 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2730 encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2733 else if (encap->size <=
2734 MLX5_ENCAPSULATION_DECISION_SIZE &&
2736 MLX5_ENCAPSULATION_DECISION_SIZE)
2739 else if (encap->size >
2740 MLX5_ENCAPSULATION_DECISION_SIZE &&
2742 MLX5_ENCAPSULATION_DECISION_SIZE)
2743 /* 2 L2 actions: encap and decap. */
2746 return rte_flow_error_set(error,
2748 RTE_FLOW_ERROR_TYPE_ACTION,
2749 NULL, "unsupported too small "
2750 "raw decap and too small raw "
2751 "encap combination");
2754 ret = flow_dv_validate_action_decap(dev, *action_flags, attr,
2758 *action_flags |= MLX5_FLOW_ACTION_DECAP;
2762 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2763 return rte_flow_error_set(error, ENOTSUP,
2764 RTE_FLOW_ERROR_TYPE_ACTION,
2766 "small raw encap size");
2767 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2768 return rte_flow_error_set(error, EINVAL,
2769 RTE_FLOW_ERROR_TYPE_ACTION,
2771 "more than one encap action");
2772 if (!attr->transfer && priv->representor)
2773 return rte_flow_error_set
2775 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2776 "encap action for VF representor "
2777 "not supported on NIC table");
2778 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
2785 * Match encap_decap resource.
2788 * Pointer to exist resource entry object.
2790 * Pointer to new encap_decap resource.
2793 * 0 on matching, -1 otherwise.
2796 flow_dv_encap_decap_resource_match(struct mlx5_hlist_entry *entry, void *ctx)
2798 struct mlx5_flow_dv_encap_decap_resource *resource;
2799 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2801 resource = (struct mlx5_flow_dv_encap_decap_resource *)ctx;
2802 cache_resource = container_of(entry,
2803 struct mlx5_flow_dv_encap_decap_resource,
2805 if (resource->entry.key == cache_resource->entry.key &&
2806 resource->reformat_type == cache_resource->reformat_type &&
2807 resource->ft_type == cache_resource->ft_type &&
2808 resource->flags == cache_resource->flags &&
2809 resource->size == cache_resource->size &&
2810 !memcmp((const void *)resource->buf,
2811 (const void *)cache_resource->buf,
2818 * Find existing encap/decap resource or create and register a new one.
2820 * @param[in, out] dev
2821 * Pointer to rte_eth_dev structure.
2822 * @param[in, out] resource
2823 * Pointer to encap/decap resource.
2824 * @parm[in, out] dev_flow
2825 * Pointer to the dev_flow.
2827 * pointer to error structure.
2830 * 0 on success otherwise -errno and errno is set.
2833 flow_dv_encap_decap_resource_register
2834 (struct rte_eth_dev *dev,
2835 struct mlx5_flow_dv_encap_decap_resource *resource,
2836 struct mlx5_flow *dev_flow,
2837 struct rte_flow_error *error)
2839 struct mlx5_priv *priv = dev->data->dev_private;
2840 struct mlx5_dev_ctx_shared *sh = priv->sh;
2841 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2842 struct mlx5dv_dr_domain *domain;
2843 struct mlx5_hlist_entry *entry;
2844 union mlx5_flow_encap_decap_key encap_decap_key = {
2846 .ft_type = resource->ft_type,
2847 .refmt_type = resource->reformat_type,
2848 .buf_size = resource->size,
2849 .table_level = !!dev_flow->dv.group,
2855 resource->flags = dev_flow->dv.group ? 0 : 1;
2856 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2857 domain = sh->fdb_domain;
2858 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2859 domain = sh->rx_domain;
2861 domain = sh->tx_domain;
2862 encap_decap_key.cksum = __rte_raw_cksum(resource->buf,
2864 resource->entry.key = encap_decap_key.v64;
2865 /* Lookup a matching resource from cache. */
2866 entry = mlx5_hlist_lookup_ex(sh->encaps_decaps, resource->entry.key,
2867 flow_dv_encap_decap_resource_match,
2870 cache_resource = container_of(entry,
2871 struct mlx5_flow_dv_encap_decap_resource, entry);
2872 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
2873 (void *)cache_resource,
2874 __atomic_load_n(&cache_resource->refcnt,
2876 __atomic_fetch_add(&cache_resource->refcnt, 1,
2878 dev_flow->handle->dvh.rix_encap_decap = cache_resource->idx;
2879 dev_flow->dv.encap_decap = cache_resource;
2882 /* Register new encap/decap resource. */
2883 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2884 &dev_flow->handle->dvh.rix_encap_decap);
2885 if (!cache_resource)
2886 return rte_flow_error_set(error, ENOMEM,
2887 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2888 "cannot allocate resource memory");
2889 *cache_resource = *resource;
2890 cache_resource->idx = dev_flow->handle->dvh.rix_encap_decap;
2891 ret = mlx5_flow_os_create_flow_action_packet_reformat
2892 (sh->ctx, domain, cache_resource,
2893 &cache_resource->action);
2895 mlx5_free(cache_resource);
2896 return rte_flow_error_set(error, ENOMEM,
2897 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2898 NULL, "cannot create action");
2900 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
2901 if (mlx5_hlist_insert_ex(sh->encaps_decaps, &cache_resource->entry,
2902 flow_dv_encap_decap_resource_match,
2903 (void *)cache_resource)) {
2904 claim_zero(mlx5_flow_os_destroy_flow_action
2905 (cache_resource->action));
2906 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2907 cache_resource->idx);
2908 return rte_flow_error_set(error, EEXIST,
2909 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2910 NULL, "action exist");
2912 dev_flow->dv.encap_decap = cache_resource;
2913 DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
2914 (void *)cache_resource,
2915 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
2920 * Find existing table jump resource or create and register a new one.
2922 * @param[in, out] dev
2923 * Pointer to rte_eth_dev structure.
2924 * @param[in, out] tbl
2925 * Pointer to flow table resource.
2926 * @parm[in, out] dev_flow
2927 * Pointer to the dev_flow.
2929 * pointer to error structure.
2932 * 0 on success otherwise -errno and errno is set.
2935 flow_dv_jump_tbl_resource_register
2936 (struct rte_eth_dev *dev __rte_unused,
2937 struct mlx5_flow_tbl_resource *tbl,
2938 struct mlx5_flow *dev_flow,
2939 struct rte_flow_error *error __rte_unused)
2941 struct mlx5_flow_tbl_data_entry *tbl_data =
2942 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2945 MLX5_ASSERT(tbl_data->jump.action);
2946 dev_flow->handle->rix_jump = tbl_data->idx;
2947 dev_flow->dv.jump = &tbl_data->jump;
2952 * Find existing table port ID resource or create and register a new one.
2954 * @param[in, out] dev
2955 * Pointer to rte_eth_dev structure.
2956 * @param[in, out] resource
2957 * Pointer to port ID action resource.
2958 * @parm[in, out] dev_flow
2959 * Pointer to the dev_flow.
2961 * pointer to error structure.
2964 * 0 on success otherwise -errno and errno is set.
2967 flow_dv_port_id_action_resource_register
2968 (struct rte_eth_dev *dev,
2969 struct mlx5_flow_dv_port_id_action_resource *resource,
2970 struct mlx5_flow *dev_flow,
2971 struct rte_flow_error *error)
2973 struct mlx5_priv *priv = dev->data->dev_private;
2974 struct mlx5_dev_ctx_shared *sh = priv->sh;
2975 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
2979 /* Lookup a matching resource from cache. */
2980 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
2981 idx, cache_resource, next) {
2982 if (resource->port_id == cache_resource->port_id) {
2983 DRV_LOG(DEBUG, "port id action resource resource %p: "
2985 (void *)cache_resource,
2986 __atomic_load_n(&cache_resource->refcnt,
2988 __atomic_fetch_add(&cache_resource->refcnt, 1,
2990 dev_flow->handle->rix_port_id_action = idx;
2991 dev_flow->dv.port_id_action = cache_resource;
2995 /* Register new port id action resource. */
2996 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
2997 &dev_flow->handle->rix_port_id_action);
2998 if (!cache_resource)
2999 return rte_flow_error_set(error, ENOMEM,
3000 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3001 "cannot allocate resource memory");
3002 *cache_resource = *resource;
3003 ret = mlx5_flow_os_create_flow_action_dest_port
3004 (priv->sh->fdb_domain, resource->port_id,
3005 &cache_resource->action);
3007 mlx5_free(cache_resource);
3008 return rte_flow_error_set(error, ENOMEM,
3009 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3010 NULL, "cannot create action");
3012 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
3013 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
3014 dev_flow->handle->rix_port_id_action, cache_resource,
3016 dev_flow->dv.port_id_action = cache_resource;
3017 DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
3018 (void *)cache_resource,
3019 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
3024 * Find existing push vlan resource or create and register a new one.
3026 * @param [in, out] dev
3027 * Pointer to rte_eth_dev structure.
3028 * @param[in, out] resource
3029 * Pointer to port ID action resource.
3030 * @parm[in, out] dev_flow
3031 * Pointer to the dev_flow.
3033 * pointer to error structure.
3036 * 0 on success otherwise -errno and errno is set.
3039 flow_dv_push_vlan_action_resource_register
3040 (struct rte_eth_dev *dev,
3041 struct mlx5_flow_dv_push_vlan_action_resource *resource,
3042 struct mlx5_flow *dev_flow,
3043 struct rte_flow_error *error)
3045 struct mlx5_priv *priv = dev->data->dev_private;
3046 struct mlx5_dev_ctx_shared *sh = priv->sh;
3047 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
3048 struct mlx5dv_dr_domain *domain;
3052 /* Lookup a matching resource from cache. */
3053 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3054 sh->push_vlan_action_list, idx, cache_resource, next) {
3055 if (resource->vlan_tag == cache_resource->vlan_tag &&
3056 resource->ft_type == cache_resource->ft_type) {
3057 DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
3059 (void *)cache_resource,
3060 __atomic_load_n(&cache_resource->refcnt,
3062 __atomic_fetch_add(&cache_resource->refcnt, 1,
3064 dev_flow->handle->dvh.rix_push_vlan = idx;
3065 dev_flow->dv.push_vlan_res = cache_resource;
3069 /* Register new push_vlan action resource. */
3070 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3071 &dev_flow->handle->dvh.rix_push_vlan);
3072 if (!cache_resource)
3073 return rte_flow_error_set(error, ENOMEM,
3074 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3075 "cannot allocate resource memory");
3076 *cache_resource = *resource;
3077 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3078 domain = sh->fdb_domain;
3079 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
3080 domain = sh->rx_domain;
3082 domain = sh->tx_domain;
3083 ret = mlx5_flow_os_create_flow_action_push_vlan
3084 (domain, resource->vlan_tag,
3085 &cache_resource->action);
3087 mlx5_free(cache_resource);
3088 return rte_flow_error_set(error, ENOMEM,
3089 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3090 NULL, "cannot create action");
3092 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
3093 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3094 &sh->push_vlan_action_list,
3095 dev_flow->handle->dvh.rix_push_vlan,
3096 cache_resource, next);
3097 dev_flow->dv.push_vlan_res = cache_resource;
3098 DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
3099 (void *)cache_resource,
3100 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
3104 * Get the size of specific rte_flow_item_type hdr size
3106 * @param[in] item_type
3107 * Tested rte_flow_item_type.
3110 * sizeof struct item_type, 0 if void or irrelevant.
3113 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
3117 switch (item_type) {
3118 case RTE_FLOW_ITEM_TYPE_ETH:
3119 retval = sizeof(struct rte_ether_hdr);
3121 case RTE_FLOW_ITEM_TYPE_VLAN:
3122 retval = sizeof(struct rte_vlan_hdr);
3124 case RTE_FLOW_ITEM_TYPE_IPV4:
3125 retval = sizeof(struct rte_ipv4_hdr);
3127 case RTE_FLOW_ITEM_TYPE_IPV6:
3128 retval = sizeof(struct rte_ipv6_hdr);
3130 case RTE_FLOW_ITEM_TYPE_UDP:
3131 retval = sizeof(struct rte_udp_hdr);
3133 case RTE_FLOW_ITEM_TYPE_TCP:
3134 retval = sizeof(struct rte_tcp_hdr);
3136 case RTE_FLOW_ITEM_TYPE_VXLAN:
3137 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3138 retval = sizeof(struct rte_vxlan_hdr);
3140 case RTE_FLOW_ITEM_TYPE_GRE:
3141 case RTE_FLOW_ITEM_TYPE_NVGRE:
3142 retval = sizeof(struct rte_gre_hdr);
3144 case RTE_FLOW_ITEM_TYPE_MPLS:
3145 retval = sizeof(struct rte_mpls_hdr);
3147 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
3155 #define MLX5_ENCAP_IPV4_VERSION 0x40
3156 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
3157 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
3158 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
3159 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
3160 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
3161 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
3164 * Convert the encap action data from list of rte_flow_item to raw buffer
3167 * Pointer to rte_flow_item objects list.
3169 * Pointer to the output buffer.
3171 * Pointer to the output buffer size.
3173 * Pointer to the error structure.
3176 * 0 on success, a negative errno value otherwise and rte_errno is set.
3179 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
3180 size_t *size, struct rte_flow_error *error)
3182 struct rte_ether_hdr *eth = NULL;
3183 struct rte_vlan_hdr *vlan = NULL;
3184 struct rte_ipv4_hdr *ipv4 = NULL;
3185 struct rte_ipv6_hdr *ipv6 = NULL;
3186 struct rte_udp_hdr *udp = NULL;
3187 struct rte_vxlan_hdr *vxlan = NULL;
3188 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
3189 struct rte_gre_hdr *gre = NULL;
3191 size_t temp_size = 0;
3194 return rte_flow_error_set(error, EINVAL,
3195 RTE_FLOW_ERROR_TYPE_ACTION,
3196 NULL, "invalid empty data");
3197 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3198 len = flow_dv_get_item_hdr_len(items->type);
3199 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
3200 return rte_flow_error_set(error, EINVAL,
3201 RTE_FLOW_ERROR_TYPE_ACTION,
3202 (void *)items->type,
3203 "items total size is too big"
3204 " for encap action");
3205 rte_memcpy((void *)&buf[temp_size], items->spec, len);
3206 switch (items->type) {
3207 case RTE_FLOW_ITEM_TYPE_ETH:
3208 eth = (struct rte_ether_hdr *)&buf[temp_size];
3210 case RTE_FLOW_ITEM_TYPE_VLAN:
3211 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
3213 return rte_flow_error_set(error, EINVAL,
3214 RTE_FLOW_ERROR_TYPE_ACTION,
3215 (void *)items->type,
3216 "eth header not found");
3217 if (!eth->ether_type)
3218 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
3220 case RTE_FLOW_ITEM_TYPE_IPV4:
3221 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
3223 return rte_flow_error_set(error, EINVAL,
3224 RTE_FLOW_ERROR_TYPE_ACTION,
3225 (void *)items->type,
3226 "neither eth nor vlan"
3228 if (vlan && !vlan->eth_proto)
3229 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3230 else if (eth && !eth->ether_type)
3231 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3232 if (!ipv4->version_ihl)
3233 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
3234 MLX5_ENCAP_IPV4_IHL_MIN;
3235 if (!ipv4->time_to_live)
3236 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
3238 case RTE_FLOW_ITEM_TYPE_IPV6:
3239 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
3241 return rte_flow_error_set(error, EINVAL,
3242 RTE_FLOW_ERROR_TYPE_ACTION,
3243 (void *)items->type,
3244 "neither eth nor vlan"
3246 if (vlan && !vlan->eth_proto)
3247 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3248 else if (eth && !eth->ether_type)
3249 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3250 if (!ipv6->vtc_flow)
3252 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
3253 if (!ipv6->hop_limits)
3254 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
3256 case RTE_FLOW_ITEM_TYPE_UDP:
3257 udp = (struct rte_udp_hdr *)&buf[temp_size];
3259 return rte_flow_error_set(error, EINVAL,
3260 RTE_FLOW_ERROR_TYPE_ACTION,
3261 (void *)items->type,
3262 "ip header not found");
3263 if (ipv4 && !ipv4->next_proto_id)
3264 ipv4->next_proto_id = IPPROTO_UDP;
3265 else if (ipv6 && !ipv6->proto)
3266 ipv6->proto = IPPROTO_UDP;
3268 case RTE_FLOW_ITEM_TYPE_VXLAN:
3269 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
3271 return rte_flow_error_set(error, EINVAL,
3272 RTE_FLOW_ERROR_TYPE_ACTION,
3273 (void *)items->type,
3274 "udp header not found");
3276 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
3277 if (!vxlan->vx_flags)
3279 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
3281 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3282 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
3284 return rte_flow_error_set(error, EINVAL,
3285 RTE_FLOW_ERROR_TYPE_ACTION,
3286 (void *)items->type,
3287 "udp header not found");
3288 if (!vxlan_gpe->proto)
3289 return rte_flow_error_set(error, EINVAL,
3290 RTE_FLOW_ERROR_TYPE_ACTION,
3291 (void *)items->type,
3292 "next protocol not found");
3295 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
3296 if (!vxlan_gpe->vx_flags)
3297 vxlan_gpe->vx_flags =
3298 MLX5_ENCAP_VXLAN_GPE_FLAGS;
3300 case RTE_FLOW_ITEM_TYPE_GRE:
3301 case RTE_FLOW_ITEM_TYPE_NVGRE:
3302 gre = (struct rte_gre_hdr *)&buf[temp_size];
3304 return rte_flow_error_set(error, EINVAL,
3305 RTE_FLOW_ERROR_TYPE_ACTION,
3306 (void *)items->type,
3307 "next protocol not found");
3309 return rte_flow_error_set(error, EINVAL,
3310 RTE_FLOW_ERROR_TYPE_ACTION,
3311 (void *)items->type,
3312 "ip header not found");
3313 if (ipv4 && !ipv4->next_proto_id)
3314 ipv4->next_proto_id = IPPROTO_GRE;
3315 else if (ipv6 && !ipv6->proto)
3316 ipv6->proto = IPPROTO_GRE;
3318 case RTE_FLOW_ITEM_TYPE_VOID:
3321 return rte_flow_error_set(error, EINVAL,
3322 RTE_FLOW_ERROR_TYPE_ACTION,
3323 (void *)items->type,
3324 "unsupported item type");
3334 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
3336 struct rte_ether_hdr *eth = NULL;
3337 struct rte_vlan_hdr *vlan = NULL;
3338 struct rte_ipv6_hdr *ipv6 = NULL;
3339 struct rte_udp_hdr *udp = NULL;
3343 eth = (struct rte_ether_hdr *)data;
3344 next_hdr = (char *)(eth + 1);
3345 proto = RTE_BE16(eth->ether_type);
3348 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
3349 vlan = (struct rte_vlan_hdr *)next_hdr;
3350 proto = RTE_BE16(vlan->eth_proto);
3351 next_hdr += sizeof(struct rte_vlan_hdr);
3354 /* HW calculates IPv4 csum. no need to proceed */
3355 if (proto == RTE_ETHER_TYPE_IPV4)
3358 /* non IPv4/IPv6 header. not supported */
3359 if (proto != RTE_ETHER_TYPE_IPV6) {
3360 return rte_flow_error_set(error, ENOTSUP,
3361 RTE_FLOW_ERROR_TYPE_ACTION,
3362 NULL, "Cannot offload non IPv4/IPv6");
3365 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3367 /* ignore non UDP */
3368 if (ipv6->proto != IPPROTO_UDP)
3371 udp = (struct rte_udp_hdr *)(ipv6 + 1);
3372 udp->dgram_cksum = 0;
3378 * Convert L2 encap action to DV specification.
3381 * Pointer to rte_eth_dev structure.
3383 * Pointer to action structure.
3384 * @param[in, out] dev_flow
3385 * Pointer to the mlx5_flow.
3386 * @param[in] transfer
3387 * Mark if the flow is E-Switch flow.
3389 * Pointer to the error structure.
3392 * 0 on success, a negative errno value otherwise and rte_errno is set.
3395 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3396 const struct rte_flow_action *action,
3397 struct mlx5_flow *dev_flow,
3399 struct rte_flow_error *error)
3401 const struct rte_flow_item *encap_data;
3402 const struct rte_flow_action_raw_encap *raw_encap_data;
3403 struct mlx5_flow_dv_encap_decap_resource res = {
3405 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3406 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3407 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3410 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3412 (const struct rte_flow_action_raw_encap *)action->conf;
3413 res.size = raw_encap_data->size;
3414 memcpy(res.buf, raw_encap_data->data, res.size);
3416 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3418 ((const struct rte_flow_action_vxlan_encap *)
3419 action->conf)->definition;
3422 ((const struct rte_flow_action_nvgre_encap *)
3423 action->conf)->definition;
3424 if (flow_dv_convert_encap_data(encap_data, res.buf,
3428 if (flow_dv_zero_encap_udp_csum(res.buf, error))
3430 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3431 return rte_flow_error_set(error, EINVAL,
3432 RTE_FLOW_ERROR_TYPE_ACTION,
3433 NULL, "can't create L2 encap action");
3438 * Convert L2 decap action to DV specification.
3441 * Pointer to rte_eth_dev structure.
3442 * @param[in, out] dev_flow
3443 * Pointer to the mlx5_flow.
3444 * @param[in] transfer
3445 * Mark if the flow is E-Switch flow.
3447 * Pointer to the error structure.
3450 * 0 on success, a negative errno value otherwise and rte_errno is set.
3453 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3454 struct mlx5_flow *dev_flow,
3456 struct rte_flow_error *error)
3458 struct mlx5_flow_dv_encap_decap_resource res = {
3461 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3462 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3463 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3466 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3467 return rte_flow_error_set(error, EINVAL,
3468 RTE_FLOW_ERROR_TYPE_ACTION,
3469 NULL, "can't create L2 decap action");
3474 * Convert raw decap/encap (L3 tunnel) action to DV specification.
3477 * Pointer to rte_eth_dev structure.
3479 * Pointer to action structure.
3480 * @param[in, out] dev_flow
3481 * Pointer to the mlx5_flow.
3483 * Pointer to the flow attributes.
3485 * Pointer to the error structure.
3488 * 0 on success, a negative errno value otherwise and rte_errno is set.
3491 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3492 const struct rte_flow_action *action,
3493 struct mlx5_flow *dev_flow,
3494 const struct rte_flow_attr *attr,
3495 struct rte_flow_error *error)
3497 const struct rte_flow_action_raw_encap *encap_data;
3498 struct mlx5_flow_dv_encap_decap_resource res;
3500 memset(&res, 0, sizeof(res));
3501 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3502 res.size = encap_data->size;
3503 memcpy(res.buf, encap_data->data, res.size);
3504 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3505 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3506 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3508 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3510 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3511 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3512 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3513 return rte_flow_error_set(error, EINVAL,
3514 RTE_FLOW_ERROR_TYPE_ACTION,
3515 NULL, "can't create encap action");
3520 * Create action push VLAN.
3523 * Pointer to rte_eth_dev structure.
3525 * Pointer to the flow attributes.
3527 * Pointer to the vlan to push to the Ethernet header.
3528 * @param[in, out] dev_flow
3529 * Pointer to the mlx5_flow.
3531 * Pointer to the error structure.
3534 * 0 on success, a negative errno value otherwise and rte_errno is set.
3537 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3538 const struct rte_flow_attr *attr,
3539 const struct rte_vlan_hdr *vlan,
3540 struct mlx5_flow *dev_flow,
3541 struct rte_flow_error *error)
3543 struct mlx5_flow_dv_push_vlan_action_resource res;
3545 memset(&res, 0, sizeof(res));
3547 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3550 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3552 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3553 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3554 return flow_dv_push_vlan_action_resource_register
3555 (dev, &res, dev_flow, error);
3558 static int fdb_mirror;
3561 * Validate the modify-header actions.
3563 * @param[in] action_flags
3564 * Holds the actions detected until now.
3566 * Pointer to the modify action.
3568 * Pointer to error structure.
3571 * 0 on success, a negative errno value otherwise and rte_errno is set.
3574 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3575 const struct rte_flow_action *action,
3576 struct rte_flow_error *error)
3578 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3579 return rte_flow_error_set(error, EINVAL,
3580 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3581 NULL, "action configuration not set");
3582 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3583 return rte_flow_error_set(error, EINVAL,
3584 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3585 "can't have encap action before"
3587 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3588 return rte_flow_error_set(error, EINVAL,
3589 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3590 "can't support sample action before"
3591 " modify action for E-Switch"
3597 * Validate the modify-header MAC address actions.
3599 * @param[in] action_flags
3600 * Holds the actions detected until now.
3602 * Pointer to the modify action.
3603 * @param[in] item_flags
3604 * Holds the items detected.
3606 * Pointer to error structure.
3609 * 0 on success, a negative errno value otherwise and rte_errno is set.
3612 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3613 const struct rte_flow_action *action,
3614 const uint64_t item_flags,
3615 struct rte_flow_error *error)
3619 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3621 if (!(item_flags & MLX5_FLOW_LAYER_L2))
3622 return rte_flow_error_set(error, EINVAL,
3623 RTE_FLOW_ERROR_TYPE_ACTION,
3625 "no L2 item in pattern");
3631 * Validate the modify-header IPv4 address actions.
3633 * @param[in] action_flags
3634 * Holds the actions detected until now.
3636 * Pointer to the modify action.
3637 * @param[in] item_flags
3638 * Holds the items detected.
3640 * Pointer to error structure.
3643 * 0 on success, a negative errno value otherwise and rte_errno is set.
3646 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3647 const struct rte_flow_action *action,
3648 const uint64_t item_flags,
3649 struct rte_flow_error *error)
3654 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3656 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3657 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3658 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3659 if (!(item_flags & layer))
3660 return rte_flow_error_set(error, EINVAL,
3661 RTE_FLOW_ERROR_TYPE_ACTION,
3663 "no ipv4 item in pattern");
3669 * Validate the modify-header IPv6 address actions.
3671 * @param[in] action_flags
3672 * Holds the actions detected until now.
3674 * Pointer to the modify action.
3675 * @param[in] item_flags
3676 * Holds the items detected.
3678 * Pointer to error structure.
3681 * 0 on success, a negative errno value otherwise and rte_errno is set.
3684 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3685 const struct rte_flow_action *action,
3686 const uint64_t item_flags,
3687 struct rte_flow_error *error)
3692 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3694 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3695 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3696 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3697 if (!(item_flags & layer))
3698 return rte_flow_error_set(error, EINVAL,
3699 RTE_FLOW_ERROR_TYPE_ACTION,
3701 "no ipv6 item in pattern");
3707 * Validate the modify-header TP actions.
3709 * @param[in] action_flags
3710 * Holds the actions detected until now.
3712 * Pointer to the modify action.
3713 * @param[in] item_flags
3714 * Holds the items detected.
3716 * Pointer to error structure.
3719 * 0 on success, a negative errno value otherwise and rte_errno is set.
3722 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3723 const struct rte_flow_action *action,
3724 const uint64_t item_flags,
3725 struct rte_flow_error *error)
3730 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3732 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3733 MLX5_FLOW_LAYER_INNER_L4 :
3734 MLX5_FLOW_LAYER_OUTER_L4;
3735 if (!(item_flags & layer))
3736 return rte_flow_error_set(error, EINVAL,
3737 RTE_FLOW_ERROR_TYPE_ACTION,
3738 NULL, "no transport layer "
3745 * Validate the modify-header actions of increment/decrement
3746 * TCP Sequence-number.
3748 * @param[in] action_flags
3749 * Holds the actions detected until now.
3751 * Pointer to the modify action.
3752 * @param[in] item_flags
3753 * Holds the items detected.
3755 * Pointer to error structure.
3758 * 0 on success, a negative errno value otherwise and rte_errno is set.
3761 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3762 const struct rte_flow_action *action,
3763 const uint64_t item_flags,
3764 struct rte_flow_error *error)
3769 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3771 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3772 MLX5_FLOW_LAYER_INNER_L4_TCP :
3773 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3774 if (!(item_flags & layer))
3775 return rte_flow_error_set(error, EINVAL,
3776 RTE_FLOW_ERROR_TYPE_ACTION,
3777 NULL, "no TCP item in"
3779 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3780 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3781 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3782 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3783 return rte_flow_error_set(error, EINVAL,
3784 RTE_FLOW_ERROR_TYPE_ACTION,
3786 "cannot decrease and increase"
3787 " TCP sequence number"
3788 " at the same time");
3794 * Validate the modify-header actions of increment/decrement
3795 * TCP Acknowledgment number.
3797 * @param[in] action_flags
3798 * Holds the actions detected until now.
3800 * Pointer to the modify action.
3801 * @param[in] item_flags
3802 * Holds the items detected.
3804 * Pointer to error structure.
3807 * 0 on success, a negative errno value otherwise and rte_errno is set.
3810 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3811 const struct rte_flow_action *action,
3812 const uint64_t item_flags,
3813 struct rte_flow_error *error)
3818 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3820 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3821 MLX5_FLOW_LAYER_INNER_L4_TCP :
3822 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3823 if (!(item_flags & layer))
3824 return rte_flow_error_set(error, EINVAL,
3825 RTE_FLOW_ERROR_TYPE_ACTION,
3826 NULL, "no TCP item in"
3828 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3829 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3830 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3831 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3832 return rte_flow_error_set(error, EINVAL,
3833 RTE_FLOW_ERROR_TYPE_ACTION,
3835 "cannot decrease and increase"
3836 " TCP acknowledgment number"
3837 " at the same time");
3843 * Validate the modify-header TTL actions.
3845 * @param[in] action_flags
3846 * Holds the actions detected until now.
3848 * Pointer to the modify action.
3849 * @param[in] item_flags
3850 * Holds the items detected.
3852 * Pointer to error structure.
3855 * 0 on success, a negative errno value otherwise and rte_errno is set.
3858 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3859 const struct rte_flow_action *action,
3860 const uint64_t item_flags,
3861 struct rte_flow_error *error)
3866 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3868 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3869 MLX5_FLOW_LAYER_INNER_L3 :
3870 MLX5_FLOW_LAYER_OUTER_L3;
3871 if (!(item_flags & layer))
3872 return rte_flow_error_set(error, EINVAL,
3873 RTE_FLOW_ERROR_TYPE_ACTION,
3875 "no IP protocol in pattern");
3881 * Validate jump action.
3884 * Pointer to the jump action.
3885 * @param[in] action_flags
3886 * Holds the actions detected until now.
3887 * @param[in] attributes
3888 * Pointer to flow attributes
3889 * @param[in] external
3890 * Action belongs to flow rule created by request external to PMD.
3892 * Pointer to error structure.
3895 * 0 on success, a negative errno value otherwise and rte_errno is set.
3898 flow_dv_validate_action_jump(struct rte_eth_dev *dev,
3899 const struct mlx5_flow_tunnel *tunnel,
3900 const struct rte_flow_action *action,
3901 uint64_t action_flags,
3902 const struct rte_flow_attr *attributes,
3903 bool external, struct rte_flow_error *error)
3905 uint32_t target_group, table;
3907 struct flow_grp_info grp_info = {
3908 .external = !!external,
3909 .transfer = !!attributes->transfer,
3913 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3914 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3915 return rte_flow_error_set(error, EINVAL,
3916 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3917 "can't have 2 fate actions in"
3919 if (action_flags & MLX5_FLOW_ACTION_METER)
3920 return rte_flow_error_set(error, ENOTSUP,
3921 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3922 "jump with meter not support");
3923 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3924 return rte_flow_error_set(error, EINVAL,
3925 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3926 "E-Switch mirroring can't support"
3927 " Sample action and jump action in"
3930 return rte_flow_error_set(error, EINVAL,
3931 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3932 NULL, "action configuration not set");
3934 ((const struct rte_flow_action_jump *)action->conf)->group;
3935 ret = mlx5_flow_group_to_table(dev, tunnel, target_group, &table,
3939 if (attributes->group == target_group &&
3940 !(action_flags & (MLX5_FLOW_ACTION_TUNNEL_SET |
3941 MLX5_FLOW_ACTION_TUNNEL_MATCH)))
3942 return rte_flow_error_set(error, EINVAL,
3943 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3944 "target group must be other than"
3945 " the current flow group");
3950 * Validate the port_id action.
3953 * Pointer to rte_eth_dev structure.
3954 * @param[in] action_flags
3955 * Bit-fields that holds the actions detected until now.
3957 * Port_id RTE action structure.
3959 * Attributes of flow that includes this action.
3961 * Pointer to error structure.
3964 * 0 on success, a negative errno value otherwise and rte_errno is set.
3967 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
3968 uint64_t action_flags,
3969 const struct rte_flow_action *action,
3970 const struct rte_flow_attr *attr,
3971 struct rte_flow_error *error)
3973 const struct rte_flow_action_port_id *port_id;
3974 struct mlx5_priv *act_priv;
3975 struct mlx5_priv *dev_priv;
3978 if (!attr->transfer)
3979 return rte_flow_error_set(error, ENOTSUP,
3980 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3982 "port id action is valid in transfer"
3984 if (!action || !action->conf)
3985 return rte_flow_error_set(error, ENOTSUP,
3986 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3988 "port id action parameters must be"
3990 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3991 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3992 return rte_flow_error_set(error, EINVAL,
3993 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3994 "can have only one fate actions in"
3996 dev_priv = mlx5_dev_to_eswitch_info(dev);
3998 return rte_flow_error_set(error, rte_errno,
3999 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4001 "failed to obtain E-Switch info");
4002 port_id = action->conf;
4003 port = port_id->original ? dev->data->port_id : port_id->id;
4004 act_priv = mlx5_port_to_eswitch_info(port, false);
4006 return rte_flow_error_set
4008 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
4009 "failed to obtain E-Switch port id for port");
4010 if (act_priv->domain_id != dev_priv->domain_id)
4011 return rte_flow_error_set
4013 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4014 "port does not belong to"
4015 " E-Switch being configured");
4020 * Get the maximum number of modify header actions.
4023 * Pointer to rte_eth_dev structure.
4025 * Flags bits to check if root level.
4028 * Max number of modify header actions device can support.
4030 static inline unsigned int
4031 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
4035 * There's no way to directly query the max capacity from FW.
4036 * The maximal value on root table should be assumed to be supported.
4038 if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
4039 return MLX5_MAX_MODIFY_NUM;
4041 return MLX5_ROOT_TBL_MODIFY_NUM;
4045 * Validate the meter action.
4048 * Pointer to rte_eth_dev structure.
4049 * @param[in] action_flags
4050 * Bit-fields that holds the actions detected until now.
4052 * Pointer to the meter action.
4054 * Attributes of flow that includes this action.
4056 * Pointer to error structure.
4059 * 0 on success, a negative errno value otherwise and rte_ernno is set.
4062 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
4063 uint64_t action_flags,
4064 const struct rte_flow_action *action,
4065 const struct rte_flow_attr *attr,
4066 struct rte_flow_error *error)
4068 struct mlx5_priv *priv = dev->data->dev_private;
4069 const struct rte_flow_action_meter *am = action->conf;
4070 struct mlx5_flow_meter *fm;
4073 return rte_flow_error_set(error, EINVAL,
4074 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4075 "meter action conf is NULL");
4077 if (action_flags & MLX5_FLOW_ACTION_METER)
4078 return rte_flow_error_set(error, ENOTSUP,
4079 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4080 "meter chaining not support");
4081 if (action_flags & MLX5_FLOW_ACTION_JUMP)
4082 return rte_flow_error_set(error, ENOTSUP,
4083 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4084 "meter with jump not support");
4086 return rte_flow_error_set(error, ENOTSUP,
4087 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4089 "meter action not supported");
4090 fm = mlx5_flow_meter_find(priv, am->mtr_id);
4092 return rte_flow_error_set(error, EINVAL,
4093 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4095 if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
4096 (!fm->ingress && !attr->ingress && attr->egress) ||
4097 (!fm->egress && !attr->egress && attr->ingress))))
4098 return rte_flow_error_set(error, EINVAL,
4099 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4100 "Flow attributes are either invalid "
4101 "or have a conflict with current "
4102 "meter attributes");
4107 * Validate the age action.
4109 * @param[in] action_flags
4110 * Holds the actions detected until now.
4112 * Pointer to the age action.
4114 * Pointer to the Ethernet device structure.
4116 * Pointer to error structure.
4119 * 0 on success, a negative errno value otherwise and rte_errno is set.
4122 flow_dv_validate_action_age(uint64_t action_flags,
4123 const struct rte_flow_action *action,
4124 struct rte_eth_dev *dev,
4125 struct rte_flow_error *error)
4127 struct mlx5_priv *priv = dev->data->dev_private;
4128 const struct rte_flow_action_age *age = action->conf;
4130 if (!priv->config.devx || priv->sh->cmng.counter_fallback)
4131 return rte_flow_error_set(error, ENOTSUP,
4132 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4134 "age action not supported");
4135 if (!(action->conf))
4136 return rte_flow_error_set(error, EINVAL,
4137 RTE_FLOW_ERROR_TYPE_ACTION, action,
4138 "configuration cannot be null");
4139 if (!(age->timeout))
4140 return rte_flow_error_set(error, EINVAL,
4141 RTE_FLOW_ERROR_TYPE_ACTION, action,
4142 "invalid timeout value 0");
4143 if (action_flags & MLX5_FLOW_ACTION_AGE)
4144 return rte_flow_error_set(error, EINVAL,
4145 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4146 "duplicate age actions set");
4151 * Validate the modify-header IPv4 DSCP actions.
4153 * @param[in] action_flags
4154 * Holds the actions detected until now.
4156 * Pointer to the modify action.
4157 * @param[in] item_flags
4158 * Holds the items detected.
4160 * Pointer to error structure.
4163 * 0 on success, a negative errno value otherwise and rte_errno is set.
4166 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
4167 const struct rte_flow_action *action,
4168 const uint64_t item_flags,
4169 struct rte_flow_error *error)
4173 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4175 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
4176 return rte_flow_error_set(error, EINVAL,
4177 RTE_FLOW_ERROR_TYPE_ACTION,
4179 "no ipv4 item in pattern");
4185 * Validate the modify-header IPv6 DSCP actions.
4187 * @param[in] action_flags
4188 * Holds the actions detected until now.
4190 * Pointer to the modify action.
4191 * @param[in] item_flags
4192 * Holds the items detected.
4194 * Pointer to error structure.
4197 * 0 on success, a negative errno value otherwise and rte_errno is set.
4200 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
4201 const struct rte_flow_action *action,
4202 const uint64_t item_flags,
4203 struct rte_flow_error *error)
4207 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4209 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
4210 return rte_flow_error_set(error, EINVAL,
4211 RTE_FLOW_ERROR_TYPE_ACTION,
4213 "no ipv6 item in pattern");
4219 * Match modify-header resource.
4222 * Pointer to the hash list.
4224 * Pointer to exist resource entry object.
4226 * Key of the new entry.
4228 * Pointer to new modify-header resource.
4231 * 0 on matching, non-zero otherwise.
4234 flow_dv_modify_match_cb(struct mlx5_hlist *list __rte_unused,
4235 struct mlx5_hlist_entry *entry,
4236 uint64_t key __rte_unused, void *cb_ctx)
4238 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4239 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
4240 struct mlx5_flow_dv_modify_hdr_resource *resource =
4241 container_of(entry, typeof(*resource), entry);
4242 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
4244 key_len += ref->actions_num * sizeof(ref->actions[0]);
4245 return ref->actions_num != resource->actions_num ||
4246 memcmp(&ref->ft_type, &resource->ft_type, key_len);
4249 struct mlx5_hlist_entry *
4250 flow_dv_modify_create_cb(struct mlx5_hlist *list, uint64_t key __rte_unused,
4253 struct mlx5_dev_ctx_shared *sh = list->ctx;
4254 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4255 struct mlx5dv_dr_domain *ns;
4256 struct mlx5_flow_dv_modify_hdr_resource *entry;
4257 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
4259 uint32_t data_len = ref->actions_num * sizeof(ref->actions[0]);
4260 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
4262 entry = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*entry) + data_len, 0,
4265 rte_flow_error_set(ctx->error, ENOMEM,
4266 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4267 "cannot allocate resource memory");
4270 rte_memcpy(&entry->ft_type,
4271 RTE_PTR_ADD(ref, offsetof(typeof(*ref), ft_type)),
4272 key_len + data_len);
4273 if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
4274 ns = sh->fdb_domain;
4275 else if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
4279 ret = mlx5_flow_os_create_flow_action_modify_header
4280 (sh->ctx, ns, entry,
4281 data_len, &entry->action);
4284 rte_flow_error_set(ctx->error, ENOMEM,
4285 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4286 NULL, "cannot create modification action");
4289 return &entry->entry;
4293 * Validate the sample action.
4295 * @param[in] action_flags
4296 * Holds the actions detected until now.
4298 * Pointer to the sample action.
4300 * Pointer to the Ethernet device structure.
4302 * Attributes of flow that includes this action.
4304 * Pointer to error structure.
4307 * 0 on success, a negative errno value otherwise and rte_errno is set.
4310 flow_dv_validate_action_sample(uint64_t action_flags,
4311 const struct rte_flow_action *action,
4312 struct rte_eth_dev *dev,
4313 const struct rte_flow_attr *attr,
4314 struct rte_flow_error *error)
4316 struct mlx5_priv *priv = dev->data->dev_private;
4317 struct mlx5_dev_config *dev_conf = &priv->config;
4318 const struct rte_flow_action_sample *sample = action->conf;
4319 const struct rte_flow_action *act;
4320 uint64_t sub_action_flags = 0;
4321 uint16_t queue_index = 0xFFFF;
4327 return rte_flow_error_set(error, EINVAL,
4328 RTE_FLOW_ERROR_TYPE_ACTION, action,
4329 "configuration cannot be NULL");
4330 if (sample->ratio == 0)
4331 return rte_flow_error_set(error, EINVAL,
4332 RTE_FLOW_ERROR_TYPE_ACTION, action,
4333 "ratio value starts from 1");
4334 if (!priv->config.devx || (sample->ratio > 0 && !priv->sampler_en))
4335 return rte_flow_error_set(error, ENOTSUP,
4336 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4338 "sample action not supported");
4339 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
4340 return rte_flow_error_set(error, EINVAL,
4341 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4342 "Multiple sample actions not "
4344 if (action_flags & MLX5_FLOW_ACTION_METER)
4345 return rte_flow_error_set(error, EINVAL,
4346 RTE_FLOW_ERROR_TYPE_ACTION, action,
4347 "wrong action order, meter should "
4348 "be after sample action");
4349 if (action_flags & MLX5_FLOW_ACTION_JUMP)
4350 return rte_flow_error_set(error, EINVAL,
4351 RTE_FLOW_ERROR_TYPE_ACTION, action,
4352 "wrong action order, jump should "
4353 "be after sample action");
4354 act = sample->actions;
4355 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
4356 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
4357 return rte_flow_error_set(error, ENOTSUP,
4358 RTE_FLOW_ERROR_TYPE_ACTION,
4359 act, "too many actions");
4360 switch (act->type) {
4361 case RTE_FLOW_ACTION_TYPE_QUEUE:
4362 ret = mlx5_flow_validate_action_queue(act,
4368 queue_index = ((const struct rte_flow_action_queue *)
4369 (act->conf))->index;
4370 sub_action_flags |= MLX5_FLOW_ACTION_QUEUE;
4373 case RTE_FLOW_ACTION_TYPE_MARK:
4374 ret = flow_dv_validate_action_mark(dev, act,
4379 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4380 sub_action_flags |= MLX5_FLOW_ACTION_MARK |
4381 MLX5_FLOW_ACTION_MARK_EXT;
4383 sub_action_flags |= MLX5_FLOW_ACTION_MARK;
4386 case RTE_FLOW_ACTION_TYPE_COUNT:
4387 ret = flow_dv_validate_action_count(dev, error);
4390 sub_action_flags |= MLX5_FLOW_ACTION_COUNT;
4393 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4394 ret = flow_dv_validate_action_port_id(dev,
4401 sub_action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4404 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4405 ret = flow_dv_validate_action_raw_encap_decap
4406 (dev, NULL, act->conf, attr, &sub_action_flags,
4413 return rte_flow_error_set(error, ENOTSUP,
4414 RTE_FLOW_ERROR_TYPE_ACTION,
4416 "Doesn't support optional "
4420 if (attr->ingress && !attr->transfer) {
4421 if (!(sub_action_flags & MLX5_FLOW_ACTION_QUEUE))
4422 return rte_flow_error_set(error, EINVAL,
4423 RTE_FLOW_ERROR_TYPE_ACTION,
4425 "Ingress must has a dest "
4426 "QUEUE for Sample");
4427 } else if (attr->egress && !attr->transfer) {
4428 return rte_flow_error_set(error, ENOTSUP,
4429 RTE_FLOW_ERROR_TYPE_ACTION,
4431 "Sample Only support Ingress "
4433 } else if (sample->actions->type != RTE_FLOW_ACTION_TYPE_END) {
4434 MLX5_ASSERT(attr->transfer);
4435 if (sample->ratio > 1)
4436 return rte_flow_error_set(error, ENOTSUP,
4437 RTE_FLOW_ERROR_TYPE_ACTION,
4439 "E-Switch doesn't support "
4440 "any optional action "
4443 if (sub_action_flags & MLX5_FLOW_ACTION_QUEUE)
4444 return rte_flow_error_set(error, ENOTSUP,
4445 RTE_FLOW_ERROR_TYPE_ACTION,
4447 "unsupported action QUEUE");
4448 if (!(sub_action_flags & MLX5_FLOW_ACTION_PORT_ID))
4449 return rte_flow_error_set(error, EINVAL,
4450 RTE_FLOW_ERROR_TYPE_ACTION,
4452 "E-Switch must has a dest "
4453 "port for mirroring");
4455 /* Continue validation for Xcap actions.*/
4456 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) &&
4457 (queue_index == 0xFFFF ||
4458 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
4459 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
4460 MLX5_FLOW_XCAP_ACTIONS)
4461 return rte_flow_error_set(error, ENOTSUP,
4462 RTE_FLOW_ERROR_TYPE_ACTION,
4463 NULL, "encap and decap "
4464 "combination aren't "
4466 if (!attr->transfer && attr->ingress && (sub_action_flags &
4467 MLX5_FLOW_ACTION_ENCAP))
4468 return rte_flow_error_set(error, ENOTSUP,
4469 RTE_FLOW_ERROR_TYPE_ACTION,
4470 NULL, "encap is not supported"
4471 " for ingress traffic");
4477 * Find existing modify-header resource or create and register a new one.
4479 * @param dev[in, out]
4480 * Pointer to rte_eth_dev structure.
4481 * @param[in, out] resource
4482 * Pointer to modify-header resource.
4483 * @parm[in, out] dev_flow
4484 * Pointer to the dev_flow.
4486 * pointer to error structure.
4489 * 0 on success otherwise -errno and errno is set.
4492 flow_dv_modify_hdr_resource_register
4493 (struct rte_eth_dev *dev,
4494 struct mlx5_flow_dv_modify_hdr_resource *resource,
4495 struct mlx5_flow *dev_flow,
4496 struct rte_flow_error *error)
4498 struct mlx5_priv *priv = dev->data->dev_private;
4499 struct mlx5_dev_ctx_shared *sh = priv->sh;
4500 uint32_t key_len = sizeof(*resource) -
4501 offsetof(typeof(*resource), ft_type) +
4502 resource->actions_num * sizeof(resource->actions[0]);
4503 struct mlx5_hlist_entry *entry;
4504 struct mlx5_flow_cb_ctx ctx = {
4509 resource->flags = dev_flow->dv.group ? 0 :
4510 MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4511 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
4513 return rte_flow_error_set(error, EOVERFLOW,
4514 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4515 "too many modify header items");
4516 resource->entry.key = __rte_raw_cksum(&resource->ft_type, key_len, 0);
4517 entry = mlx5_hlist_register(sh->modify_cmds, resource->entry.key, &ctx);
4520 resource = container_of(entry, typeof(*resource), entry);
4521 dev_flow->handle->dvh.modify_hdr = resource;
4526 * Get DV flow counter by index.
4529 * Pointer to the Ethernet device structure.
4531 * mlx5 flow counter index in the container.
4533 * mlx5 flow counter pool in the container,
4536 * Pointer to the counter, NULL otherwise.
4538 static struct mlx5_flow_counter *
4539 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
4541 struct mlx5_flow_counter_pool **ppool)
4543 struct mlx5_priv *priv = dev->data->dev_private;
4544 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4545 struct mlx5_flow_counter_pool *pool;
4547 /* Decrease to original index and clear shared bit. */
4548 idx = (idx - 1) & (MLX5_CNT_SHARED_OFFSET - 1);
4549 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cmng->n);
4550 pool = cmng->pools[idx / MLX5_COUNTERS_PER_POOL];
4554 return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
4558 * Check the devx counter belongs to the pool.
4561 * Pointer to the counter pool.
4563 * The counter devx ID.
4566 * True if counter belongs to the pool, false otherwise.
4569 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
4571 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
4572 MLX5_COUNTERS_PER_POOL;
4574 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
4580 * Get a pool by devx counter ID.
4583 * Pointer to the counter management.
4585 * The counter devx ID.
4588 * The counter pool pointer if exists, NULL otherwise,
4590 static struct mlx5_flow_counter_pool *
4591 flow_dv_find_pool_by_id(struct mlx5_flow_counter_mng *cmng, int id)
4594 struct mlx5_flow_counter_pool *pool = NULL;
4596 rte_spinlock_lock(&cmng->pool_update_sl);
4597 /* Check last used pool. */
4598 if (cmng->last_pool_idx != POOL_IDX_INVALID &&
4599 flow_dv_is_counter_in_pool(cmng->pools[cmng->last_pool_idx], id)) {
4600 pool = cmng->pools[cmng->last_pool_idx];
4603 /* ID out of range means no suitable pool in the container. */
4604 if (id > cmng->max_id || id < cmng->min_id)
4607 * Find the pool from the end of the container, since mostly counter
4608 * ID is sequence increasing, and the last pool should be the needed
4613 struct mlx5_flow_counter_pool *pool_tmp = cmng->pools[i];
4615 if (flow_dv_is_counter_in_pool(pool_tmp, id)) {
4621 rte_spinlock_unlock(&cmng->pool_update_sl);
4626 * Resize a counter container.
4629 * Pointer to the Ethernet device structure.
4632 * 0 on success, otherwise negative errno value and rte_errno is set.
4635 flow_dv_container_resize(struct rte_eth_dev *dev)
4637 struct mlx5_priv *priv = dev->data->dev_private;
4638 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4639 void *old_pools = cmng->pools;
4640 uint32_t resize = cmng->n + MLX5_CNT_CONTAINER_RESIZE;
4641 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4642 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
4649 memcpy(pools, old_pools, cmng->n *
4650 sizeof(struct mlx5_flow_counter_pool *));
4652 cmng->pools = pools;
4654 mlx5_free(old_pools);
4659 * Query a devx flow counter.
4662 * Pointer to the Ethernet device structure.
4664 * Index to the flow counter.
4666 * The statistics value of packets.
4668 * The statistics value of bytes.
4671 * 0 on success, otherwise a negative errno value and rte_errno is set.
4674 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4677 struct mlx5_priv *priv = dev->data->dev_private;
4678 struct mlx5_flow_counter_pool *pool = NULL;
4679 struct mlx5_flow_counter *cnt;
4682 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4684 if (priv->sh->cmng.counter_fallback)
4685 return mlx5_devx_cmd_flow_counter_query(cnt->dcs_when_active, 0,
4686 0, pkts, bytes, 0, NULL, NULL, 0);
4687 rte_spinlock_lock(&pool->sl);
4692 offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
4693 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4694 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4696 rte_spinlock_unlock(&pool->sl);
4701 * Create and initialize a new counter pool.
4704 * Pointer to the Ethernet device structure.
4706 * The devX counter handle.
4708 * Whether the pool is for counter that was allocated for aging.
4709 * @param[in/out] cont_cur
4710 * Pointer to the container pointer, it will be update in pool resize.
4713 * The pool container pointer on success, NULL otherwise and rte_errno is set.
4715 static struct mlx5_flow_counter_pool *
4716 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4719 struct mlx5_priv *priv = dev->data->dev_private;
4720 struct mlx5_flow_counter_pool *pool;
4721 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4722 bool fallback = priv->sh->cmng.counter_fallback;
4723 uint32_t size = sizeof(*pool);
4725 size += MLX5_COUNTERS_PER_POOL * MLX5_CNT_SIZE;
4726 size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * MLX5_AGE_SIZE);
4727 pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
4733 pool->is_aged = !!age;
4734 pool->query_gen = 0;
4735 pool->min_dcs = dcs;
4736 rte_spinlock_init(&pool->sl);
4737 rte_spinlock_init(&pool->csl);
4738 TAILQ_INIT(&pool->counters[0]);
4739 TAILQ_INIT(&pool->counters[1]);
4740 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
4741 rte_spinlock_lock(&cmng->pool_update_sl);
4742 pool->index = cmng->n_valid;
4743 if (pool->index == cmng->n && flow_dv_container_resize(dev)) {
4745 rte_spinlock_unlock(&cmng->pool_update_sl);
4748 cmng->pools[pool->index] = pool;
4750 if (unlikely(fallback)) {
4751 int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
4753 if (base < cmng->min_id)
4754 cmng->min_id = base;
4755 if (base > cmng->max_id)
4756 cmng->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
4757 cmng->last_pool_idx = pool->index;
4759 rte_spinlock_unlock(&cmng->pool_update_sl);
4764 * Prepare a new counter and/or a new counter pool.
4767 * Pointer to the Ethernet device structure.
4768 * @param[out] cnt_free
4769 * Where to put the pointer of a new counter.
4771 * Whether the pool is for counter that was allocated for aging.
4774 * The counter pool pointer and @p cnt_free is set on success,
4775 * NULL otherwise and rte_errno is set.
4777 static struct mlx5_flow_counter_pool *
4778 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4779 struct mlx5_flow_counter **cnt_free,
4782 struct mlx5_priv *priv = dev->data->dev_private;
4783 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4784 struct mlx5_flow_counter_pool *pool;
4785 struct mlx5_counters tmp_tq;
4786 struct mlx5_devx_obj *dcs = NULL;
4787 struct mlx5_flow_counter *cnt;
4788 enum mlx5_counter_type cnt_type =
4789 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
4790 bool fallback = priv->sh->cmng.counter_fallback;
4794 /* bulk_bitmap must be 0 for single counter allocation. */
4795 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4798 pool = flow_dv_find_pool_by_id(cmng, dcs->id);
4800 pool = flow_dv_pool_create(dev, dcs, age);
4802 mlx5_devx_cmd_destroy(dcs);
4806 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4807 cnt = MLX5_POOL_GET_CNT(pool, i);
4809 cnt->dcs_when_free = dcs;
4813 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4815 rte_errno = ENODATA;
4818 pool = flow_dv_pool_create(dev, dcs, age);
4820 mlx5_devx_cmd_destroy(dcs);
4823 TAILQ_INIT(&tmp_tq);
4824 for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
4825 cnt = MLX5_POOL_GET_CNT(pool, i);
4827 TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
4829 rte_spinlock_lock(&cmng->csl[cnt_type]);
4830 TAILQ_CONCAT(&cmng->counters[cnt_type], &tmp_tq, next);
4831 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4832 *cnt_free = MLX5_POOL_GET_CNT(pool, 0);
4833 (*cnt_free)->pool = pool;
4838 * Allocate a flow counter.
4841 * Pointer to the Ethernet device structure.
4843 * Whether the counter was allocated for aging.
4846 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4849 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t age)
4851 struct mlx5_priv *priv = dev->data->dev_private;
4852 struct mlx5_flow_counter_pool *pool = NULL;
4853 struct mlx5_flow_counter *cnt_free = NULL;
4854 bool fallback = priv->sh->cmng.counter_fallback;
4855 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4856 enum mlx5_counter_type cnt_type =
4857 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
4860 if (!priv->config.devx) {
4861 rte_errno = ENOTSUP;
4864 /* Get free counters from container. */
4865 rte_spinlock_lock(&cmng->csl[cnt_type]);
4866 cnt_free = TAILQ_FIRST(&cmng->counters[cnt_type]);
4868 TAILQ_REMOVE(&cmng->counters[cnt_type], cnt_free, next);
4869 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4870 if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free, age))
4872 pool = cnt_free->pool;
4874 cnt_free->dcs_when_active = cnt_free->dcs_when_free;
4875 /* Create a DV counter action only in the first time usage. */
4876 if (!cnt_free->action) {
4878 struct mlx5_devx_obj *dcs;
4882 offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
4883 dcs = pool->min_dcs;
4886 dcs = cnt_free->dcs_when_free;
4888 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
4895 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4896 MLX5_CNT_ARRAY_IDX(pool, cnt_free));
4897 /* Update the counter reset values. */
4898 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4901 if (!fallback && !priv->sh->cmng.query_thread_on)
4902 /* Start the asynchronous batch query by the host thread. */
4903 mlx5_set_query_alarm(priv->sh);
4907 cnt_free->pool = pool;
4909 cnt_free->dcs_when_free = cnt_free->dcs_when_active;
4910 rte_spinlock_lock(&cmng->csl[cnt_type]);
4911 TAILQ_INSERT_TAIL(&cmng->counters[cnt_type], cnt_free, next);
4912 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4918 * Allocate a shared flow counter.
4921 * Pointer to the shared counter configuration.
4923 * Pointer to save the allocated counter index.
4926 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4930 flow_dv_counter_alloc_shared_cb(void *ctx, union mlx5_l3t_data *data)
4932 struct mlx5_shared_counter_conf *conf = ctx;
4933 struct rte_eth_dev *dev = conf->dev;
4934 struct mlx5_flow_counter *cnt;
4936 data->dword = flow_dv_counter_alloc(dev, 0);
4937 data->dword |= MLX5_CNT_SHARED_OFFSET;
4938 cnt = flow_dv_counter_get_by_idx(dev, data->dword, NULL);
4939 cnt->shared_info.id = conf->id;
4944 * Get a shared flow counter.
4947 * Pointer to the Ethernet device structure.
4949 * Counter identifier.
4952 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4955 flow_dv_counter_get_shared(struct rte_eth_dev *dev, uint32_t id)
4957 struct mlx5_priv *priv = dev->data->dev_private;
4958 struct mlx5_shared_counter_conf conf = {
4962 union mlx5_l3t_data data = {
4966 mlx5_l3t_prepare_entry(priv->sh->cnt_id_tbl, id, &data,
4967 flow_dv_counter_alloc_shared_cb, &conf);
4972 * Get age param from counter index.
4975 * Pointer to the Ethernet device structure.
4976 * @param[in] counter
4977 * Index to the counter handler.
4980 * The aging parameter specified for the counter index.
4982 static struct mlx5_age_param*
4983 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
4986 struct mlx5_flow_counter *cnt;
4987 struct mlx5_flow_counter_pool *pool = NULL;
4989 flow_dv_counter_get_by_idx(dev, counter, &pool);
4990 counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
4991 cnt = MLX5_POOL_GET_CNT(pool, counter);
4992 return MLX5_CNT_TO_AGE(cnt);
4996 * Remove a flow counter from aged counter list.
4999 * Pointer to the Ethernet device structure.
5000 * @param[in] counter
5001 * Index to the counter handler.
5003 * Pointer to the counter handler.
5006 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
5007 uint32_t counter, struct mlx5_flow_counter *cnt)
5009 struct mlx5_age_info *age_info;
5010 struct mlx5_age_param *age_param;
5011 struct mlx5_priv *priv = dev->data->dev_private;
5012 uint16_t expected = AGE_CANDIDATE;
5014 age_info = GET_PORT_AGE_INFO(priv);
5015 age_param = flow_dv_counter_idx_get_age(dev, counter);
5016 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
5017 AGE_FREE, false, __ATOMIC_RELAXED,
5018 __ATOMIC_RELAXED)) {
5020 * We need the lock even it is age timeout,
5021 * since counter may still in process.
5023 rte_spinlock_lock(&age_info->aged_sl);
5024 TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
5025 rte_spinlock_unlock(&age_info->aged_sl);
5026 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
5031 * Release a flow counter.
5034 * Pointer to the Ethernet device structure.
5035 * @param[in] counter
5036 * Index to the counter handler.
5039 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
5041 struct mlx5_priv *priv = dev->data->dev_private;
5042 struct mlx5_flow_counter_pool *pool = NULL;
5043 struct mlx5_flow_counter *cnt;
5044 enum mlx5_counter_type cnt_type;
5048 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
5050 if (IS_SHARED_CNT(counter) &&
5051 mlx5_l3t_clear_entry(priv->sh->cnt_id_tbl, cnt->shared_info.id))
5054 flow_dv_counter_remove_from_age(dev, counter, cnt);
5057 * Put the counter back to list to be updated in none fallback mode.
5058 * Currently, we are using two list alternately, while one is in query,
5059 * add the freed counter to the other list based on the pool query_gen
5060 * value. After query finishes, add counter the list to the global
5061 * container counter list. The list changes while query starts. In
5062 * this case, lock will not be needed as query callback and release
5063 * function both operate with the different list.
5066 if (!priv->sh->cmng.counter_fallback) {
5067 rte_spinlock_lock(&pool->csl);
5068 TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
5069 rte_spinlock_unlock(&pool->csl);
5071 cnt->dcs_when_free = cnt->dcs_when_active;
5072 cnt_type = pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
5073 MLX5_COUNTER_TYPE_ORIGIN;
5074 rte_spinlock_lock(&priv->sh->cmng.csl[cnt_type]);
5075 TAILQ_INSERT_TAIL(&priv->sh->cmng.counters[cnt_type],
5077 rte_spinlock_unlock(&priv->sh->cmng.csl[cnt_type]);
5082 * Verify the @p attributes will be correctly understood by the NIC and store
5083 * them in the @p flow if everything is correct.
5086 * Pointer to dev struct.
5087 * @param[in] attributes
5088 * Pointer to flow attributes
5089 * @param[in] external
5090 * This flow rule is created by request external to PMD.
5092 * Pointer to error structure.
5095 * - 0 on success and non root table.
5096 * - 1 on success and root table.
5097 * - a negative errno value otherwise and rte_errno is set.
5100 flow_dv_validate_attributes(struct rte_eth_dev *dev,
5101 const struct mlx5_flow_tunnel *tunnel,
5102 const struct rte_flow_attr *attributes,
5103 struct flow_grp_info grp_info,
5104 struct rte_flow_error *error)
5106 struct mlx5_priv *priv = dev->data->dev_private;
5107 uint32_t priority_max = priv->config.flow_prio - 1;
5110 #ifndef HAVE_MLX5DV_DR
5111 RTE_SET_USED(tunnel);
5112 RTE_SET_USED(grp_info);
5113 if (attributes->group)
5114 return rte_flow_error_set(error, ENOTSUP,
5115 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5117 "groups are not supported");
5121 ret = mlx5_flow_group_to_table(dev, tunnel, attributes->group, &table,
5126 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
5128 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
5129 attributes->priority >= priority_max)
5130 return rte_flow_error_set(error, ENOTSUP,
5131 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
5133 "priority out of range");
5134 if (attributes->transfer) {
5135 if (!priv->config.dv_esw_en)
5136 return rte_flow_error_set
5138 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5139 "E-Switch dr is not supported");
5140 if (!(priv->representor || priv->master))
5141 return rte_flow_error_set
5142 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5143 NULL, "E-Switch configuration can only be"
5144 " done by a master or a representor device");
5145 if (attributes->egress)
5146 return rte_flow_error_set
5148 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
5149 "egress is not supported");
5151 if (!(attributes->egress ^ attributes->ingress))
5152 return rte_flow_error_set(error, ENOTSUP,
5153 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
5154 "must specify exactly one of "
5155 "ingress or egress");
5160 * Internal validation function. For validating both actions and items.
5163 * Pointer to the rte_eth_dev structure.
5165 * Pointer to the flow attributes.
5167 * Pointer to the list of items.
5168 * @param[in] actions
5169 * Pointer to the list of actions.
5170 * @param[in] external
5171 * This flow rule is created by request external to PMD.
5172 * @param[in] hairpin
5173 * Number of hairpin TX actions, 0 means classic flow.
5175 * Pointer to the error structure.
5178 * 0 on success, a negative errno value otherwise and rte_errno is set.
5181 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
5182 const struct rte_flow_item items[],
5183 const struct rte_flow_action actions[],
5184 bool external, int hairpin, struct rte_flow_error *error)
5187 uint64_t action_flags = 0;
5188 uint64_t item_flags = 0;
5189 uint64_t last_item = 0;
5190 uint8_t next_protocol = 0xff;
5191 uint16_t ether_type = 0;
5193 uint8_t item_ipv6_proto = 0;
5194 const struct rte_flow_item *gre_item = NULL;
5195 const struct rte_flow_action_raw_decap *decap;
5196 const struct rte_flow_action_raw_encap *encap;
5197 const struct rte_flow_action_rss *rss;
5198 const struct rte_flow_item_tcp nic_tcp_mask = {
5201 .src_port = RTE_BE16(UINT16_MAX),
5202 .dst_port = RTE_BE16(UINT16_MAX),
5205 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
5208 "\xff\xff\xff\xff\xff\xff\xff\xff"
5209 "\xff\xff\xff\xff\xff\xff\xff\xff",
5211 "\xff\xff\xff\xff\xff\xff\xff\xff"
5212 "\xff\xff\xff\xff\xff\xff\xff\xff",
5213 .vtc_flow = RTE_BE32(0xffffffff),
5219 const struct rte_flow_item_ecpri nic_ecpri_mask = {
5223 RTE_BE32(((const struct rte_ecpri_common_hdr) {
5227 .dummy[0] = 0xffffffff,
5230 struct mlx5_priv *priv = dev->data->dev_private;
5231 struct mlx5_dev_config *dev_conf = &priv->config;
5232 uint16_t queue_index = 0xFFFF;
5233 const struct rte_flow_item_vlan *vlan_m = NULL;
5234 int16_t rw_act_num = 0;
5236 const struct mlx5_flow_tunnel *tunnel;
5237 struct flow_grp_info grp_info = {
5238 .external = !!external,
5239 .transfer = !!attr->transfer,
5240 .fdb_def_rule = !!priv->fdb_def_rule,
5242 const struct rte_eth_hairpin_conf *conf;
5246 if (is_flow_tunnel_match_rule(dev, attr, items, actions)) {
5247 tunnel = flow_items_to_tunnel(items);
5248 action_flags |= MLX5_FLOW_ACTION_TUNNEL_MATCH |
5249 MLX5_FLOW_ACTION_DECAP;
5250 } else if (is_flow_tunnel_steer_rule(dev, attr, items, actions)) {
5251 tunnel = flow_actions_to_tunnel(actions);
5252 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
5256 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
5257 (dev, tunnel, attr, items, actions);
5258 ret = flow_dv_validate_attributes(dev, tunnel, attr, grp_info, error);
5261 is_root = (uint64_t)ret;
5262 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5263 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
5264 int type = items->type;
5266 if (!mlx5_flow_os_item_supported(type))
5267 return rte_flow_error_set(error, ENOTSUP,
5268 RTE_FLOW_ERROR_TYPE_ITEM,
5269 NULL, "item not supported");
5271 case MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL:
5272 if (items[0].type != (typeof(items[0].type))
5273 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL)
5274 return rte_flow_error_set
5276 RTE_FLOW_ERROR_TYPE_ITEM,
5277 NULL, "MLX5 private items "
5278 "must be the first");
5280 case RTE_FLOW_ITEM_TYPE_VOID:
5282 case RTE_FLOW_ITEM_TYPE_PORT_ID:
5283 ret = flow_dv_validate_item_port_id
5284 (dev, items, attr, item_flags, error);
5287 last_item = MLX5_FLOW_ITEM_PORT_ID;
5289 case RTE_FLOW_ITEM_TYPE_ETH:
5290 ret = mlx5_flow_validate_item_eth(items, item_flags,
5294 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
5295 MLX5_FLOW_LAYER_OUTER_L2;
5296 if (items->mask != NULL && items->spec != NULL) {
5298 ((const struct rte_flow_item_eth *)
5301 ((const struct rte_flow_item_eth *)
5303 ether_type = rte_be_to_cpu_16(ether_type);
5308 case RTE_FLOW_ITEM_TYPE_VLAN:
5309 ret = flow_dv_validate_item_vlan(items, item_flags,
5313 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
5314 MLX5_FLOW_LAYER_OUTER_VLAN;
5315 if (items->mask != NULL && items->spec != NULL) {
5317 ((const struct rte_flow_item_vlan *)
5318 items->spec)->inner_type;
5320 ((const struct rte_flow_item_vlan *)
5321 items->mask)->inner_type;
5322 ether_type = rte_be_to_cpu_16(ether_type);
5326 /* Store outer VLAN mask for of_push_vlan action. */
5328 vlan_m = items->mask;
5330 case RTE_FLOW_ITEM_TYPE_IPV4:
5331 mlx5_flow_tunnel_ip_check(items, next_protocol,
5332 &item_flags, &tunnel);
5333 ret = flow_dv_validate_item_ipv4(items, item_flags,
5334 last_item, ether_type,
5338 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
5339 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
5340 if (items->mask != NULL &&
5341 ((const struct rte_flow_item_ipv4 *)
5342 items->mask)->hdr.next_proto_id) {
5344 ((const struct rte_flow_item_ipv4 *)
5345 (items->spec))->hdr.next_proto_id;
5347 ((const struct rte_flow_item_ipv4 *)
5348 (items->mask))->hdr.next_proto_id;
5350 /* Reset for inner layer. */
5351 next_protocol = 0xff;
5354 case RTE_FLOW_ITEM_TYPE_IPV6:
5355 mlx5_flow_tunnel_ip_check(items, next_protocol,
5356 &item_flags, &tunnel);
5357 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
5364 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
5365 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
5366 if (items->mask != NULL &&
5367 ((const struct rte_flow_item_ipv6 *)
5368 items->mask)->hdr.proto) {
5370 ((const struct rte_flow_item_ipv6 *)
5371 items->spec)->hdr.proto;
5373 ((const struct rte_flow_item_ipv6 *)
5374 items->spec)->hdr.proto;
5376 ((const struct rte_flow_item_ipv6 *)
5377 items->mask)->hdr.proto;
5379 /* Reset for inner layer. */
5380 next_protocol = 0xff;
5383 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
5384 ret = flow_dv_validate_item_ipv6_frag_ext(items,
5389 last_item = tunnel ?
5390 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
5391 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
5392 if (items->mask != NULL &&
5393 ((const struct rte_flow_item_ipv6_frag_ext *)
5394 items->mask)->hdr.next_header) {
5396 ((const struct rte_flow_item_ipv6_frag_ext *)
5397 items->spec)->hdr.next_header;
5399 ((const struct rte_flow_item_ipv6_frag_ext *)
5400 items->mask)->hdr.next_header;
5402 /* Reset for inner layer. */
5403 next_protocol = 0xff;
5406 case RTE_FLOW_ITEM_TYPE_TCP:
5407 ret = mlx5_flow_validate_item_tcp
5414 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
5415 MLX5_FLOW_LAYER_OUTER_L4_TCP;
5417 case RTE_FLOW_ITEM_TYPE_UDP:
5418 ret = mlx5_flow_validate_item_udp(items, item_flags,
5423 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
5424 MLX5_FLOW_LAYER_OUTER_L4_UDP;
5426 case RTE_FLOW_ITEM_TYPE_GRE:
5427 ret = mlx5_flow_validate_item_gre(items, item_flags,
5428 next_protocol, error);
5432 last_item = MLX5_FLOW_LAYER_GRE;
5434 case RTE_FLOW_ITEM_TYPE_NVGRE:
5435 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
5440 last_item = MLX5_FLOW_LAYER_NVGRE;
5442 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5443 ret = mlx5_flow_validate_item_gre_key
5444 (items, item_flags, gre_item, error);
5447 last_item = MLX5_FLOW_LAYER_GRE_KEY;
5449 case RTE_FLOW_ITEM_TYPE_VXLAN:
5450 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
5454 last_item = MLX5_FLOW_LAYER_VXLAN;
5456 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5457 ret = mlx5_flow_validate_item_vxlan_gpe(items,
5462 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5464 case RTE_FLOW_ITEM_TYPE_GENEVE:
5465 ret = mlx5_flow_validate_item_geneve(items,
5470 last_item = MLX5_FLOW_LAYER_GENEVE;
5472 case RTE_FLOW_ITEM_TYPE_MPLS:
5473 ret = mlx5_flow_validate_item_mpls(dev, items,
5478 last_item = MLX5_FLOW_LAYER_MPLS;
5481 case RTE_FLOW_ITEM_TYPE_MARK:
5482 ret = flow_dv_validate_item_mark(dev, items, attr,
5486 last_item = MLX5_FLOW_ITEM_MARK;
5488 case RTE_FLOW_ITEM_TYPE_META:
5489 ret = flow_dv_validate_item_meta(dev, items, attr,
5493 last_item = MLX5_FLOW_ITEM_METADATA;
5495 case RTE_FLOW_ITEM_TYPE_ICMP:
5496 ret = mlx5_flow_validate_item_icmp(items, item_flags,
5501 last_item = MLX5_FLOW_LAYER_ICMP;
5503 case RTE_FLOW_ITEM_TYPE_ICMP6:
5504 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
5509 item_ipv6_proto = IPPROTO_ICMPV6;
5510 last_item = MLX5_FLOW_LAYER_ICMP6;
5512 case RTE_FLOW_ITEM_TYPE_TAG:
5513 ret = flow_dv_validate_item_tag(dev, items,
5517 last_item = MLX5_FLOW_ITEM_TAG;
5519 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
5520 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
5522 case RTE_FLOW_ITEM_TYPE_GTP:
5523 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
5527 last_item = MLX5_FLOW_LAYER_GTP;
5529 case RTE_FLOW_ITEM_TYPE_ECPRI:
5530 /* Capacity will be checked in the translate stage. */
5531 ret = mlx5_flow_validate_item_ecpri(items, item_flags,
5538 last_item = MLX5_FLOW_LAYER_ECPRI;
5541 return rte_flow_error_set(error, ENOTSUP,
5542 RTE_FLOW_ERROR_TYPE_ITEM,
5543 NULL, "item not supported");
5545 item_flags |= last_item;
5547 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5548 int type = actions->type;
5550 if (!mlx5_flow_os_action_supported(type))
5551 return rte_flow_error_set(error, ENOTSUP,
5552 RTE_FLOW_ERROR_TYPE_ACTION,
5554 "action not supported");
5555 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5556 return rte_flow_error_set(error, ENOTSUP,
5557 RTE_FLOW_ERROR_TYPE_ACTION,
5558 actions, "too many actions");
5560 case RTE_FLOW_ACTION_TYPE_VOID:
5562 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5563 ret = flow_dv_validate_action_port_id(dev,
5570 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5573 case RTE_FLOW_ACTION_TYPE_FLAG:
5574 ret = flow_dv_validate_action_flag(dev, action_flags,
5578 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5579 /* Count all modify-header actions as one. */
5580 if (!(action_flags &
5581 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5583 action_flags |= MLX5_FLOW_ACTION_FLAG |
5584 MLX5_FLOW_ACTION_MARK_EXT;
5586 action_flags |= MLX5_FLOW_ACTION_FLAG;
5589 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5591 case RTE_FLOW_ACTION_TYPE_MARK:
5592 ret = flow_dv_validate_action_mark(dev, actions,
5597 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5598 /* Count all modify-header actions as one. */
5599 if (!(action_flags &
5600 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5602 action_flags |= MLX5_FLOW_ACTION_MARK |
5603 MLX5_FLOW_ACTION_MARK_EXT;
5605 action_flags |= MLX5_FLOW_ACTION_MARK;
5608 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5610 case RTE_FLOW_ACTION_TYPE_SET_META:
5611 ret = flow_dv_validate_action_set_meta(dev, actions,
5616 /* Count all modify-header actions as one action. */
5617 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5619 action_flags |= MLX5_FLOW_ACTION_SET_META;
5620 rw_act_num += MLX5_ACT_NUM_SET_META;
5622 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5623 ret = flow_dv_validate_action_set_tag(dev, actions,
5628 /* Count all modify-header actions as one action. */
5629 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5631 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
5632 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5634 case RTE_FLOW_ACTION_TYPE_DROP:
5635 ret = mlx5_flow_validate_action_drop(action_flags,
5639 action_flags |= MLX5_FLOW_ACTION_DROP;
5642 case RTE_FLOW_ACTION_TYPE_QUEUE:
5643 ret = mlx5_flow_validate_action_queue(actions,
5648 queue_index = ((const struct rte_flow_action_queue *)
5649 (actions->conf))->index;
5650 action_flags |= MLX5_FLOW_ACTION_QUEUE;
5653 case RTE_FLOW_ACTION_TYPE_RSS:
5654 rss = actions->conf;
5655 ret = mlx5_flow_validate_action_rss(actions,
5661 if (rss != NULL && rss->queue_num)
5662 queue_index = rss->queue[0];
5663 action_flags |= MLX5_FLOW_ACTION_RSS;
5666 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
5668 mlx5_flow_validate_action_default_miss(action_flags,
5672 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
5675 case RTE_FLOW_ACTION_TYPE_COUNT:
5676 ret = flow_dv_validate_action_count(dev, error);
5679 action_flags |= MLX5_FLOW_ACTION_COUNT;
5682 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5683 if (flow_dv_validate_action_pop_vlan(dev,
5689 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5692 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5693 ret = flow_dv_validate_action_push_vlan(dev,
5700 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5703 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5704 ret = flow_dv_validate_action_set_vlan_pcp
5705 (action_flags, actions, error);
5708 /* Count PCP with push_vlan command. */
5709 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5711 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5712 ret = flow_dv_validate_action_set_vlan_vid
5713 (item_flags, action_flags,
5717 /* Count VID with push_vlan command. */
5718 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5719 rw_act_num += MLX5_ACT_NUM_MDF_VID;
5721 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5722 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5723 ret = flow_dv_validate_action_l2_encap(dev,
5729 action_flags |= MLX5_FLOW_ACTION_ENCAP;
5732 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5733 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5734 ret = flow_dv_validate_action_decap(dev, action_flags,
5738 action_flags |= MLX5_FLOW_ACTION_DECAP;
5741 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5742 ret = flow_dv_validate_action_raw_encap_decap
5743 (dev, NULL, actions->conf, attr, &action_flags,
5748 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5749 decap = actions->conf;
5750 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5752 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5756 encap = actions->conf;
5758 ret = flow_dv_validate_action_raw_encap_decap
5760 decap ? decap : &empty_decap, encap,
5761 attr, &action_flags, &actions_n,
5766 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5767 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5768 ret = flow_dv_validate_action_modify_mac(action_flags,
5774 /* Count all modify-header actions as one action. */
5775 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5777 action_flags |= actions->type ==
5778 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5779 MLX5_FLOW_ACTION_SET_MAC_SRC :
5780 MLX5_FLOW_ACTION_SET_MAC_DST;
5782 * Even if the source and destination MAC addresses have
5783 * overlap in the header with 4B alignment, the convert
5784 * function will handle them separately and 4 SW actions
5785 * will be created. And 2 actions will be added each
5786 * time no matter how many bytes of address will be set.
5788 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5790 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5791 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5792 ret = flow_dv_validate_action_modify_ipv4(action_flags,
5798 /* Count all modify-header actions as one action. */
5799 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5801 action_flags |= actions->type ==
5802 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5803 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5804 MLX5_FLOW_ACTION_SET_IPV4_DST;
5805 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5807 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5808 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5809 ret = flow_dv_validate_action_modify_ipv6(action_flags,
5815 if (item_ipv6_proto == IPPROTO_ICMPV6)
5816 return rte_flow_error_set(error, ENOTSUP,
5817 RTE_FLOW_ERROR_TYPE_ACTION,
5819 "Can't change header "
5820 "with ICMPv6 proto");
5821 /* Count all modify-header actions as one action. */
5822 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5824 action_flags |= actions->type ==
5825 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5826 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5827 MLX5_FLOW_ACTION_SET_IPV6_DST;
5828 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5830 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5831 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5832 ret = flow_dv_validate_action_modify_tp(action_flags,
5838 /* Count all modify-header actions as one action. */
5839 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5841 action_flags |= actions->type ==
5842 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5843 MLX5_FLOW_ACTION_SET_TP_SRC :
5844 MLX5_FLOW_ACTION_SET_TP_DST;
5845 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5847 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5848 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5849 ret = flow_dv_validate_action_modify_ttl(action_flags,
5855 /* Count all modify-header actions as one action. */
5856 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5858 action_flags |= actions->type ==
5859 RTE_FLOW_ACTION_TYPE_SET_TTL ?
5860 MLX5_FLOW_ACTION_SET_TTL :
5861 MLX5_FLOW_ACTION_DEC_TTL;
5862 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5864 case RTE_FLOW_ACTION_TYPE_JUMP:
5865 ret = flow_dv_validate_action_jump(dev, tunnel, actions,
5872 action_flags |= MLX5_FLOW_ACTION_JUMP;
5874 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5875 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5876 ret = flow_dv_validate_action_modify_tcp_seq
5883 /* Count all modify-header actions as one action. */
5884 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5886 action_flags |= actions->type ==
5887 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5888 MLX5_FLOW_ACTION_INC_TCP_SEQ :
5889 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5890 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5892 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5893 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5894 ret = flow_dv_validate_action_modify_tcp_ack
5901 /* Count all modify-header actions as one action. */
5902 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5904 action_flags |= actions->type ==
5905 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5906 MLX5_FLOW_ACTION_INC_TCP_ACK :
5907 MLX5_FLOW_ACTION_DEC_TCP_ACK;
5908 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5910 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5912 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5913 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5914 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5916 case RTE_FLOW_ACTION_TYPE_METER:
5917 ret = mlx5_flow_validate_action_meter(dev,
5923 action_flags |= MLX5_FLOW_ACTION_METER;
5925 /* Meter action will add one more TAG action. */
5926 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5928 case RTE_FLOW_ACTION_TYPE_AGE:
5929 ret = flow_dv_validate_action_age(action_flags,
5934 action_flags |= MLX5_FLOW_ACTION_AGE;
5937 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5938 ret = flow_dv_validate_action_modify_ipv4_dscp
5945 /* Count all modify-header actions as one action. */
5946 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5948 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5949 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5951 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5952 ret = flow_dv_validate_action_modify_ipv6_dscp
5959 /* Count all modify-header actions as one action. */
5960 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5962 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5963 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5965 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5966 ret = flow_dv_validate_action_sample(action_flags,
5971 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
5974 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
5975 if (actions[0].type != (typeof(actions[0].type))
5976 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET)
5977 return rte_flow_error_set
5979 RTE_FLOW_ERROR_TYPE_ACTION,
5980 NULL, "MLX5 private action "
5981 "must be the first");
5983 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
5986 return rte_flow_error_set(error, ENOTSUP,
5987 RTE_FLOW_ERROR_TYPE_ACTION,
5989 "action not supported");
5993 * Validate actions in flow rules
5994 * - Explicit decap action is prohibited by the tunnel offload API.
5995 * - Drop action in tunnel steer rule is prohibited by the API.
5996 * - Application cannot use MARK action because it's value can mask
5997 * tunnel default miss nitification.
5998 * - JUMP in tunnel match rule has no support in current PMD
6000 * - TAG & META are reserved for future uses.
6002 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_SET) {
6003 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_DECAP |
6004 MLX5_FLOW_ACTION_MARK |
6005 MLX5_FLOW_ACTION_SET_TAG |
6006 MLX5_FLOW_ACTION_SET_META |
6007 MLX5_FLOW_ACTION_DROP;
6009 if (action_flags & bad_actions_mask)
6010 return rte_flow_error_set
6012 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6013 "Invalid RTE action in tunnel "
6015 if (!(action_flags & MLX5_FLOW_ACTION_JUMP))
6016 return rte_flow_error_set
6018 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6019 "tunnel set decap rule must terminate "
6022 return rte_flow_error_set
6024 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6025 "tunnel flows for ingress traffic only");
6027 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_MATCH) {
6028 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_JUMP |
6029 MLX5_FLOW_ACTION_MARK |
6030 MLX5_FLOW_ACTION_SET_TAG |
6031 MLX5_FLOW_ACTION_SET_META;
6033 if (action_flags & bad_actions_mask)
6034 return rte_flow_error_set
6036 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6037 "Invalid RTE action in tunnel "
6041 * Validate the drop action mutual exclusion with other actions.
6042 * Drop action is mutually-exclusive with any other action, except for
6045 if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
6046 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
6047 return rte_flow_error_set(error, EINVAL,
6048 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6049 "Drop action is mutually-exclusive "
6050 "with any other action, except for "
6052 /* Eswitch has few restrictions on using items and actions */
6053 if (attr->transfer) {
6054 if (!mlx5_flow_ext_mreg_supported(dev) &&
6055 action_flags & MLX5_FLOW_ACTION_FLAG)
6056 return rte_flow_error_set(error, ENOTSUP,
6057 RTE_FLOW_ERROR_TYPE_ACTION,
6059 "unsupported action FLAG");
6060 if (!mlx5_flow_ext_mreg_supported(dev) &&
6061 action_flags & MLX5_FLOW_ACTION_MARK)
6062 return rte_flow_error_set(error, ENOTSUP,
6063 RTE_FLOW_ERROR_TYPE_ACTION,
6065 "unsupported action MARK");
6066 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
6067 return rte_flow_error_set(error, ENOTSUP,
6068 RTE_FLOW_ERROR_TYPE_ACTION,
6070 "unsupported action QUEUE");
6071 if (action_flags & MLX5_FLOW_ACTION_RSS)
6072 return rte_flow_error_set(error, ENOTSUP,
6073 RTE_FLOW_ERROR_TYPE_ACTION,
6075 "unsupported action RSS");
6076 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
6077 return rte_flow_error_set(error, EINVAL,
6078 RTE_FLOW_ERROR_TYPE_ACTION,
6080 "no fate action is found");
6082 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
6083 return rte_flow_error_set(error, EINVAL,
6084 RTE_FLOW_ERROR_TYPE_ACTION,
6086 "no fate action is found");
6089 * Continue validation for Xcap and VLAN actions.
6090 * If hairpin is working in explicit TX rule mode, there is no actions
6091 * splitting and the validation of hairpin ingress flow should be the
6092 * same as other standard flows.
6094 if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
6095 MLX5_FLOW_VLAN_ACTIONS)) &&
6096 (queue_index == 0xFFFF ||
6097 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN ||
6098 ((conf = mlx5_rxq_get_hairpin_conf(dev, queue_index)) != NULL &&
6099 conf->tx_explicit != 0))) {
6100 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
6101 MLX5_FLOW_XCAP_ACTIONS)
6102 return rte_flow_error_set(error, ENOTSUP,
6103 RTE_FLOW_ERROR_TYPE_ACTION,
6104 NULL, "encap and decap "
6105 "combination aren't supported");
6106 if (!attr->transfer && attr->ingress) {
6107 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
6108 return rte_flow_error_set
6110 RTE_FLOW_ERROR_TYPE_ACTION,
6111 NULL, "encap is not supported"
6112 " for ingress traffic");
6113 else if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
6114 return rte_flow_error_set
6116 RTE_FLOW_ERROR_TYPE_ACTION,
6117 NULL, "push VLAN action not "
6118 "supported for ingress");
6119 else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
6120 MLX5_FLOW_VLAN_ACTIONS)
6121 return rte_flow_error_set
6123 RTE_FLOW_ERROR_TYPE_ACTION,
6124 NULL, "no support for "
6125 "multiple VLAN actions");
6129 * Hairpin flow will add one more TAG action in TX implicit mode.
6130 * In TX explicit mode, there will be no hairpin flow ID.
6133 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6134 /* extra metadata enabled: one more TAG action will be add. */
6135 if (dev_conf->dv_flow_en &&
6136 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
6137 mlx5_flow_ext_mreg_supported(dev))
6138 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6139 if ((uint32_t)rw_act_num >
6140 flow_dv_modify_hdr_action_max(dev, is_root)) {
6141 return rte_flow_error_set(error, ENOTSUP,
6142 RTE_FLOW_ERROR_TYPE_ACTION,
6143 NULL, "too many header modify"
6144 " actions to support");
6150 * Internal preparation function. Allocates the DV flow size,
6151 * this size is constant.
6154 * Pointer to the rte_eth_dev structure.
6156 * Pointer to the flow attributes.
6158 * Pointer to the list of items.
6159 * @param[in] actions
6160 * Pointer to the list of actions.
6162 * Pointer to the error structure.
6165 * Pointer to mlx5_flow object on success,
6166 * otherwise NULL and rte_errno is set.
6168 static struct mlx5_flow *
6169 flow_dv_prepare(struct rte_eth_dev *dev,
6170 const struct rte_flow_attr *attr __rte_unused,
6171 const struct rte_flow_item items[] __rte_unused,
6172 const struct rte_flow_action actions[] __rte_unused,
6173 struct rte_flow_error *error)
6175 uint32_t handle_idx = 0;
6176 struct mlx5_flow *dev_flow;
6177 struct mlx5_flow_handle *dev_handle;
6178 struct mlx5_priv *priv = dev->data->dev_private;
6179 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6182 /* In case of corrupting the memory. */
6183 if (wks->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
6184 rte_flow_error_set(error, ENOSPC,
6185 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6186 "not free temporary device flow");
6189 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
6192 rte_flow_error_set(error, ENOMEM,
6193 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6194 "not enough memory to create flow handle");
6197 MLX5_ASSERT(wks->flow_idx + 1 < RTE_DIM(wks->flows));
6198 dev_flow = &wks->flows[wks->flow_idx++];
6199 dev_flow->handle = dev_handle;
6200 dev_flow->handle_idx = handle_idx;
6202 * In some old rdma-core releases, before continuing, a check of the
6203 * length of matching parameter will be done at first. It needs to use
6204 * the length without misc4 param. If the flow has misc4 support, then
6205 * the length needs to be adjusted accordingly. Each param member is
6206 * aligned with a 64B boundary naturally.
6208 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param) -
6209 MLX5_ST_SZ_BYTES(fte_match_set_misc4);
6211 * The matching value needs to be cleared to 0 before using. In the
6212 * past, it will be automatically cleared when using rte_*alloc
6213 * API. The time consumption will be almost the same as before.
6215 memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
6216 dev_flow->ingress = attr->ingress;
6217 dev_flow->dv.transfer = attr->transfer;
6221 #ifdef RTE_LIBRTE_MLX5_DEBUG
6223 * Sanity check for match mask and value. Similar to check_valid_spec() in
6224 * kernel driver. If unmasked bit is present in value, it returns failure.
6227 * pointer to match mask buffer.
6228 * @param match_value
6229 * pointer to match value buffer.
6232 * 0 if valid, -EINVAL otherwise.
6235 flow_dv_check_valid_spec(void *match_mask, void *match_value)
6237 uint8_t *m = match_mask;
6238 uint8_t *v = match_value;
6241 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
6244 "match_value differs from match_criteria"
6245 " %p[%u] != %p[%u]",
6246 match_value, i, match_mask, i);
6255 * Add match of ip_version.
6259 * @param[in] headers_v
6260 * Values header pointer.
6261 * @param[in] headers_m
6262 * Masks header pointer.
6263 * @param[in] ip_version
6264 * The IP version to set.
6267 flow_dv_set_match_ip_version(uint32_t group,
6273 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
6275 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
6277 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
6278 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
6279 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
6283 * Add Ethernet item to matcher and to the value.
6285 * @param[in, out] matcher
6287 * @param[in, out] key
6288 * Flow matcher value.
6290 * Flow pattern to translate.
6292 * Item is inner pattern.
6295 flow_dv_translate_item_eth(void *matcher, void *key,
6296 const struct rte_flow_item *item, int inner,
6299 const struct rte_flow_item_eth *eth_m = item->mask;
6300 const struct rte_flow_item_eth *eth_v = item->spec;
6301 const struct rte_flow_item_eth nic_mask = {
6302 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6303 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6304 .type = RTE_BE16(0xffff),
6317 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6319 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6321 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6323 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6325 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, dmac_47_16),
6326 ð_m->dst, sizeof(eth_m->dst));
6327 /* The value must be in the range of the mask. */
6328 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, dmac_47_16);
6329 for (i = 0; i < sizeof(eth_m->dst); ++i)
6330 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
6331 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, smac_47_16),
6332 ð_m->src, sizeof(eth_m->src));
6333 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, smac_47_16);
6334 /* The value must be in the range of the mask. */
6335 for (i = 0; i < sizeof(eth_m->dst); ++i)
6336 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
6338 * HW supports match on one Ethertype, the Ethertype following the last
6339 * VLAN tag of the packet (see PRM).
6340 * Set match on ethertype only if ETH header is not followed by VLAN.
6341 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6342 * ethertype, and use ip_version field instead.
6343 * eCPRI over Ether layer will use type value 0xAEFE.
6345 if (eth_m->type == 0xFFFF) {
6346 /* Set cvlan_tag mask for any single\multi\un-tagged case. */
6347 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6348 switch (eth_v->type) {
6349 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
6350 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6352 case RTE_BE16(RTE_ETHER_TYPE_QINQ):
6353 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6354 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6356 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
6357 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
6359 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
6360 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
6366 if (eth_m->has_vlan) {
6367 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6368 if (eth_v->has_vlan) {
6370 * Here, when also has_more_vlan field in VLAN item is
6371 * not set, only single-tagged packets will be matched.
6373 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6377 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
6378 rte_be_to_cpu_16(eth_m->type));
6379 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, ethertype);
6380 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
6384 * Add VLAN item to matcher and to the value.
6386 * @param[in, out] dev_flow
6388 * @param[in, out] matcher
6390 * @param[in, out] key
6391 * Flow matcher value.
6393 * Flow pattern to translate.
6395 * Item is inner pattern.
6398 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
6399 void *matcher, void *key,
6400 const struct rte_flow_item *item,
6401 int inner, uint32_t group)
6403 const struct rte_flow_item_vlan *vlan_m = item->mask;
6404 const struct rte_flow_item_vlan *vlan_v = item->spec;
6411 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6413 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6415 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6417 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6419 * This is workaround, masks are not supported,
6420 * and pre-validated.
6423 dev_flow->handle->vf_vlan.tag =
6424 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
6427 * When VLAN item exists in flow, mark packet as tagged,
6428 * even if TCI is not specified.
6430 if (!MLX5_GET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag)) {
6431 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6432 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6437 vlan_m = &rte_flow_item_vlan_mask;
6438 tci_m = rte_be_to_cpu_16(vlan_m->tci);
6439 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
6440 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_vid, tci_m);
6441 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_vid, tci_v);
6442 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_cfi, tci_m >> 12);
6443 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_cfi, tci_v >> 12);
6444 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_prio, tci_m >> 13);
6445 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_prio, tci_v >> 13);
6447 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6448 * ethertype, and use ip_version field instead.
6450 if (vlan_m->inner_type == 0xFFFF) {
6451 switch (vlan_v->inner_type) {
6452 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
6453 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6454 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6455 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
6457 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
6458 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
6460 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
6461 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
6467 if (vlan_m->has_more_vlan && vlan_v->has_more_vlan) {
6468 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6469 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6470 /* Only one vlan_tag bit can be set. */
6471 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
6474 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
6475 rte_be_to_cpu_16(vlan_m->inner_type));
6476 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, ethertype,
6477 rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
6481 * Add IPV4 item to matcher and to the value.
6483 * @param[in, out] matcher
6485 * @param[in, out] key
6486 * Flow matcher value.
6488 * Flow pattern to translate.
6490 * Item is inner pattern.
6492 * The group to insert the rule.
6495 flow_dv_translate_item_ipv4(void *matcher, void *key,
6496 const struct rte_flow_item *item,
6497 int inner, uint32_t group)
6499 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
6500 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
6501 const struct rte_flow_item_ipv4 nic_mask = {
6503 .src_addr = RTE_BE32(0xffffffff),
6504 .dst_addr = RTE_BE32(0xffffffff),
6505 .type_of_service = 0xff,
6506 .next_proto_id = 0xff,
6507 .time_to_live = 0xff,
6517 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6519 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6521 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6523 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6525 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6530 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6531 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6532 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6533 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6534 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
6535 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
6536 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6537 src_ipv4_src_ipv6.ipv4_layout.ipv4);
6538 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6539 src_ipv4_src_ipv6.ipv4_layout.ipv4);
6540 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
6541 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
6542 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
6543 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
6544 ipv4_m->hdr.type_of_service);
6545 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
6546 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
6547 ipv4_m->hdr.type_of_service >> 2);
6548 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
6549 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6550 ipv4_m->hdr.next_proto_id);
6551 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6552 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
6553 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6554 ipv4_m->hdr.time_to_live);
6555 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6556 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
6557 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
6558 !!(ipv4_m->hdr.fragment_offset));
6559 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
6560 !!(ipv4_v->hdr.fragment_offset & ipv4_m->hdr.fragment_offset));
6564 * Add IPV6 item to matcher and to the value.
6566 * @param[in, out] matcher
6568 * @param[in, out] key
6569 * Flow matcher value.
6571 * Flow pattern to translate.
6573 * Item is inner pattern.
6575 * The group to insert the rule.
6578 flow_dv_translate_item_ipv6(void *matcher, void *key,
6579 const struct rte_flow_item *item,
6580 int inner, uint32_t group)
6582 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
6583 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
6584 const struct rte_flow_item_ipv6 nic_mask = {
6587 "\xff\xff\xff\xff\xff\xff\xff\xff"
6588 "\xff\xff\xff\xff\xff\xff\xff\xff",
6590 "\xff\xff\xff\xff\xff\xff\xff\xff"
6591 "\xff\xff\xff\xff\xff\xff\xff\xff",
6592 .vtc_flow = RTE_BE32(0xffffffff),
6599 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6600 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6609 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6611 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6613 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6615 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6617 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6622 size = sizeof(ipv6_m->hdr.dst_addr);
6623 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6624 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6625 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6626 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6627 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
6628 for (i = 0; i < size; ++i)
6629 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
6630 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6631 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6632 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6633 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6634 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
6635 for (i = 0; i < size; ++i)
6636 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
6638 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
6639 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
6640 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
6641 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
6642 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
6643 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
6646 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
6648 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
6651 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
6653 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
6657 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6659 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6660 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
6662 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6663 ipv6_m->hdr.hop_limits);
6664 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6665 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
6666 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
6667 !!(ipv6_m->has_frag_ext));
6668 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
6669 !!(ipv6_v->has_frag_ext & ipv6_m->has_frag_ext));
6673 * Add IPV6 fragment extension item to matcher and to the value.
6675 * @param[in, out] matcher
6677 * @param[in, out] key
6678 * Flow matcher value.
6680 * Flow pattern to translate.
6682 * Item is inner pattern.
6685 flow_dv_translate_item_ipv6_frag_ext(void *matcher, void *key,
6686 const struct rte_flow_item *item,
6689 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_m = item->mask;
6690 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_v = item->spec;
6691 const struct rte_flow_item_ipv6_frag_ext nic_mask = {
6693 .next_header = 0xff,
6694 .frag_data = RTE_BE16(0xffff),
6701 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6703 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6705 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6707 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6709 /* IPv6 fragment extension item exists, so packet is IP fragment. */
6710 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6711 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 1);
6712 if (!ipv6_frag_ext_v)
6714 if (!ipv6_frag_ext_m)
6715 ipv6_frag_ext_m = &nic_mask;
6716 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6717 ipv6_frag_ext_m->hdr.next_header);
6718 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6719 ipv6_frag_ext_v->hdr.next_header &
6720 ipv6_frag_ext_m->hdr.next_header);
6724 * Add TCP item to matcher and to the value.
6726 * @param[in, out] matcher
6728 * @param[in, out] key
6729 * Flow matcher value.
6731 * Flow pattern to translate.
6733 * Item is inner pattern.
6736 flow_dv_translate_item_tcp(void *matcher, void *key,
6737 const struct rte_flow_item *item,
6740 const struct rte_flow_item_tcp *tcp_m = item->mask;
6741 const struct rte_flow_item_tcp *tcp_v = item->spec;
6746 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6748 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6750 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6752 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6754 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6755 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
6759 tcp_m = &rte_flow_item_tcp_mask;
6760 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
6761 rte_be_to_cpu_16(tcp_m->hdr.src_port));
6762 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
6763 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
6764 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
6765 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
6766 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
6767 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
6768 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
6769 tcp_m->hdr.tcp_flags);
6770 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
6771 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
6775 * Add UDP item to matcher and to the value.
6777 * @param[in, out] matcher
6779 * @param[in, out] key
6780 * Flow matcher value.
6782 * Flow pattern to translate.
6784 * Item is inner pattern.
6787 flow_dv_translate_item_udp(void *matcher, void *key,
6788 const struct rte_flow_item *item,
6791 const struct rte_flow_item_udp *udp_m = item->mask;
6792 const struct rte_flow_item_udp *udp_v = item->spec;
6797 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6799 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6801 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6803 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6805 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6806 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
6810 udp_m = &rte_flow_item_udp_mask;
6811 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
6812 rte_be_to_cpu_16(udp_m->hdr.src_port));
6813 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
6814 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
6815 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
6816 rte_be_to_cpu_16(udp_m->hdr.dst_port));
6817 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6818 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
6822 * Add GRE optional Key item to matcher and to the value.
6824 * @param[in, out] matcher
6826 * @param[in, out] key
6827 * Flow matcher value.
6829 * Flow pattern to translate.
6831 * Item is inner pattern.
6834 flow_dv_translate_item_gre_key(void *matcher, void *key,
6835 const struct rte_flow_item *item)
6837 const rte_be32_t *key_m = item->mask;
6838 const rte_be32_t *key_v = item->spec;
6839 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6840 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6841 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
6843 /* GRE K bit must be on and should already be validated */
6844 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
6845 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
6849 key_m = &gre_key_default_mask;
6850 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
6851 rte_be_to_cpu_32(*key_m) >> 8);
6852 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
6853 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
6854 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
6855 rte_be_to_cpu_32(*key_m) & 0xFF);
6856 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
6857 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
6861 * Add GRE item to matcher and to the value.
6863 * @param[in, out] matcher
6865 * @param[in, out] key
6866 * Flow matcher value.
6868 * Flow pattern to translate.
6870 * Item is inner pattern.
6873 flow_dv_translate_item_gre(void *matcher, void *key,
6874 const struct rte_flow_item *item,
6877 const struct rte_flow_item_gre *gre_m = item->mask;
6878 const struct rte_flow_item_gre *gre_v = item->spec;
6881 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6882 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6889 uint16_t s_present:1;
6890 uint16_t k_present:1;
6891 uint16_t rsvd_bit1:1;
6892 uint16_t c_present:1;
6896 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
6899 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6901 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6903 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6905 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6907 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6908 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
6912 gre_m = &rte_flow_item_gre_mask;
6913 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
6914 rte_be_to_cpu_16(gre_m->protocol));
6915 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6916 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
6917 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
6918 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
6919 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
6920 gre_crks_rsvd0_ver_m.c_present);
6921 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6922 gre_crks_rsvd0_ver_v.c_present &
6923 gre_crks_rsvd0_ver_m.c_present);
6924 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6925 gre_crks_rsvd0_ver_m.k_present);
6926 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6927 gre_crks_rsvd0_ver_v.k_present &
6928 gre_crks_rsvd0_ver_m.k_present);
6929 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6930 gre_crks_rsvd0_ver_m.s_present);
6931 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6932 gre_crks_rsvd0_ver_v.s_present &
6933 gre_crks_rsvd0_ver_m.s_present);
6937 * Add NVGRE item to matcher and to the value.
6939 * @param[in, out] matcher
6941 * @param[in, out] key
6942 * Flow matcher value.
6944 * Flow pattern to translate.
6946 * Item is inner pattern.
6949 flow_dv_translate_item_nvgre(void *matcher, void *key,
6950 const struct rte_flow_item *item,
6953 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6954 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6955 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6956 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6957 const char *tni_flow_id_m;
6958 const char *tni_flow_id_v;
6964 /* For NVGRE, GRE header fields must be set with defined values. */
6965 const struct rte_flow_item_gre gre_spec = {
6966 .c_rsvd0_ver = RTE_BE16(0x2000),
6967 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6969 const struct rte_flow_item_gre gre_mask = {
6970 .c_rsvd0_ver = RTE_BE16(0xB000),
6971 .protocol = RTE_BE16(UINT16_MAX),
6973 const struct rte_flow_item gre_item = {
6978 flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6982 nvgre_m = &rte_flow_item_nvgre_mask;
6983 tni_flow_id_m = (const char *)nvgre_m->tni;
6984 tni_flow_id_v = (const char *)nvgre_v->tni;
6985 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6986 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6987 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6988 memcpy(gre_key_m, tni_flow_id_m, size);
6989 for (i = 0; i < size; ++i)
6990 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6994 * Add VXLAN item to matcher and to the value.
6996 * @param[in, out] matcher
6998 * @param[in, out] key
6999 * Flow matcher value.
7001 * Flow pattern to translate.
7003 * Item is inner pattern.
7006 flow_dv_translate_item_vxlan(void *matcher, void *key,
7007 const struct rte_flow_item *item,
7010 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
7011 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
7014 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7015 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7023 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7025 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7027 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7029 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7031 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
7032 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
7033 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7034 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7035 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7040 vxlan_m = &rte_flow_item_vxlan_mask;
7041 size = sizeof(vxlan_m->vni);
7042 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
7043 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
7044 memcpy(vni_m, vxlan_m->vni, size);
7045 for (i = 0; i < size; ++i)
7046 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7050 * Add VXLAN-GPE item to matcher and to the value.
7052 * @param[in, out] matcher
7054 * @param[in, out] key
7055 * Flow matcher value.
7057 * Flow pattern to translate.
7059 * Item is inner pattern.
7063 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
7064 const struct rte_flow_item *item, int inner)
7066 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
7067 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
7071 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
7073 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7079 uint8_t flags_m = 0xff;
7080 uint8_t flags_v = 0xc;
7083 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7085 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7087 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7089 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7091 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
7092 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
7093 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7094 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7095 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7100 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
7101 size = sizeof(vxlan_m->vni);
7102 vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
7103 vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
7104 memcpy(vni_m, vxlan_m->vni, size);
7105 for (i = 0; i < size; ++i)
7106 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7107 if (vxlan_m->flags) {
7108 flags_m = vxlan_m->flags;
7109 flags_v = vxlan_v->flags;
7111 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
7112 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
7113 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
7115 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
7120 * Add Geneve item to matcher and to the value.
7122 * @param[in, out] matcher
7124 * @param[in, out] key
7125 * Flow matcher value.
7127 * Flow pattern to translate.
7129 * Item is inner pattern.
7133 flow_dv_translate_item_geneve(void *matcher, void *key,
7134 const struct rte_flow_item *item, int inner)
7136 const struct rte_flow_item_geneve *geneve_m = item->mask;
7137 const struct rte_flow_item_geneve *geneve_v = item->spec;
7140 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7141 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7150 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7152 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7154 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7156 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7158 dport = MLX5_UDP_PORT_GENEVE;
7159 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7160 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7161 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7166 geneve_m = &rte_flow_item_geneve_mask;
7167 size = sizeof(geneve_m->vni);
7168 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
7169 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
7170 memcpy(vni_m, geneve_m->vni, size);
7171 for (i = 0; i < size; ++i)
7172 vni_v[i] = vni_m[i] & geneve_v->vni[i];
7173 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
7174 rte_be_to_cpu_16(geneve_m->protocol));
7175 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
7176 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
7177 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
7178 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
7179 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
7180 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7181 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
7182 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7183 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
7184 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7185 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
7186 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
7187 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7191 * Add MPLS item to matcher and to the value.
7193 * @param[in, out] matcher
7195 * @param[in, out] key
7196 * Flow matcher value.
7198 * Flow pattern to translate.
7199 * @param[in] prev_layer
7200 * The protocol layer indicated in previous item.
7202 * Item is inner pattern.
7205 flow_dv_translate_item_mpls(void *matcher, void *key,
7206 const struct rte_flow_item *item,
7207 uint64_t prev_layer,
7210 const uint32_t *in_mpls_m = item->mask;
7211 const uint32_t *in_mpls_v = item->spec;
7212 uint32_t *out_mpls_m = 0;
7213 uint32_t *out_mpls_v = 0;
7214 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7215 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7216 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
7218 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7219 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
7220 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7222 switch (prev_layer) {
7223 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7224 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
7225 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
7226 MLX5_UDP_PORT_MPLS);
7228 case MLX5_FLOW_LAYER_GRE:
7229 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
7230 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
7231 RTE_ETHER_TYPE_MPLS);
7234 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
7235 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
7242 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
7243 switch (prev_layer) {
7244 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7246 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7247 outer_first_mpls_over_udp);
7249 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7250 outer_first_mpls_over_udp);
7252 case MLX5_FLOW_LAYER_GRE:
7254 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7255 outer_first_mpls_over_gre);
7257 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7258 outer_first_mpls_over_gre);
7261 /* Inner MPLS not over GRE is not supported. */
7264 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7268 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7274 if (out_mpls_m && out_mpls_v) {
7275 *out_mpls_m = *in_mpls_m;
7276 *out_mpls_v = *in_mpls_v & *in_mpls_m;
7281 * Add metadata register item to matcher
7283 * @param[in, out] matcher
7285 * @param[in, out] key
7286 * Flow matcher value.
7287 * @param[in] reg_type
7288 * Type of device metadata register
7295 flow_dv_match_meta_reg(void *matcher, void *key,
7296 enum modify_reg reg_type,
7297 uint32_t data, uint32_t mask)
7300 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
7302 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7308 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
7309 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
7312 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
7313 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
7317 * The metadata register C0 field might be divided into
7318 * source vport index and META item value, we should set
7319 * this field according to specified mask, not as whole one.
7321 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
7323 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
7324 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
7327 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
7330 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
7331 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
7334 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
7335 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
7338 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
7339 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
7342 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
7343 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
7346 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
7347 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
7350 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
7351 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
7354 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
7355 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
7364 * Add MARK item to matcher
7367 * The device to configure through.
7368 * @param[in, out] matcher
7370 * @param[in, out] key
7371 * Flow matcher value.
7373 * Flow pattern to translate.
7376 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
7377 void *matcher, void *key,
7378 const struct rte_flow_item *item)
7380 struct mlx5_priv *priv = dev->data->dev_private;
7381 const struct rte_flow_item_mark *mark;
7385 mark = item->mask ? (const void *)item->mask :
7386 &rte_flow_item_mark_mask;
7387 mask = mark->id & priv->sh->dv_mark_mask;
7388 mark = (const void *)item->spec;
7390 value = mark->id & priv->sh->dv_mark_mask & mask;
7392 enum modify_reg reg;
7394 /* Get the metadata register index for the mark. */
7395 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
7396 MLX5_ASSERT(reg > 0);
7397 if (reg == REG_C_0) {
7398 struct mlx5_priv *priv = dev->data->dev_private;
7399 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7400 uint32_t shl_c0 = rte_bsf32(msk_c0);
7406 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7411 * Add META item to matcher
7414 * The devich to configure through.
7415 * @param[in, out] matcher
7417 * @param[in, out] key
7418 * Flow matcher value.
7420 * Attributes of flow that includes this item.
7422 * Flow pattern to translate.
7425 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
7426 void *matcher, void *key,
7427 const struct rte_flow_attr *attr,
7428 const struct rte_flow_item *item)
7430 const struct rte_flow_item_meta *meta_m;
7431 const struct rte_flow_item_meta *meta_v;
7433 meta_m = (const void *)item->mask;
7435 meta_m = &rte_flow_item_meta_mask;
7436 meta_v = (const void *)item->spec;
7439 uint32_t value = meta_v->data;
7440 uint32_t mask = meta_m->data;
7442 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
7446 * In datapath code there is no endianness
7447 * coversions for perfromance reasons, all
7448 * pattern conversions are done in rte_flow.
7450 value = rte_cpu_to_be_32(value);
7451 mask = rte_cpu_to_be_32(mask);
7452 if (reg == REG_C_0) {
7453 struct mlx5_priv *priv = dev->data->dev_private;
7454 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7455 uint32_t shl_c0 = rte_bsf32(msk_c0);
7456 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
7457 uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
7464 MLX5_ASSERT(msk_c0);
7465 MLX5_ASSERT(!(~msk_c0 & mask));
7467 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7472 * Add vport metadata Reg C0 item to matcher
7474 * @param[in, out] matcher
7476 * @param[in, out] key
7477 * Flow matcher value.
7479 * Flow pattern to translate.
7482 flow_dv_translate_item_meta_vport(void *matcher, void *key,
7483 uint32_t value, uint32_t mask)
7485 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
7489 * Add tag item to matcher
7492 * The devich to configure through.
7493 * @param[in, out] matcher
7495 * @param[in, out] key
7496 * Flow matcher value.
7498 * Flow pattern to translate.
7501 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
7502 void *matcher, void *key,
7503 const struct rte_flow_item *item)
7505 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
7506 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
7507 uint32_t mask, value;
7510 value = tag_v->data;
7511 mask = tag_m ? tag_m->data : UINT32_MAX;
7512 if (tag_v->id == REG_C_0) {
7513 struct mlx5_priv *priv = dev->data->dev_private;
7514 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7515 uint32_t shl_c0 = rte_bsf32(msk_c0);
7521 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
7525 * Add TAG item to matcher
7528 * The devich to configure through.
7529 * @param[in, out] matcher
7531 * @param[in, out] key
7532 * Flow matcher value.
7534 * Flow pattern to translate.
7537 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
7538 void *matcher, void *key,
7539 const struct rte_flow_item *item)
7541 const struct rte_flow_item_tag *tag_v = item->spec;
7542 const struct rte_flow_item_tag *tag_m = item->mask;
7543 enum modify_reg reg;
7546 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
7547 /* Get the metadata register index for the tag. */
7548 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
7549 MLX5_ASSERT(reg > 0);
7550 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
7554 * Add source vport match to the specified matcher.
7556 * @param[in, out] matcher
7558 * @param[in, out] key
7559 * Flow matcher value.
7561 * Source vport value to match
7566 flow_dv_translate_item_source_vport(void *matcher, void *key,
7567 int16_t port, uint16_t mask)
7569 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7570 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7572 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
7573 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
7577 * Translate port-id item to eswitch match on port-id.
7580 * The devich to configure through.
7581 * @param[in, out] matcher
7583 * @param[in, out] key
7584 * Flow matcher value.
7586 * Flow pattern to translate.
7589 * 0 on success, a negative errno value otherwise.
7592 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
7593 void *key, const struct rte_flow_item *item)
7595 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
7596 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
7597 struct mlx5_priv *priv;
7600 mask = pid_m ? pid_m->id : 0xffff;
7601 id = pid_v ? pid_v->id : dev->data->port_id;
7602 priv = mlx5_port_to_eswitch_info(id, item == NULL);
7605 /* Translate to vport field or to metadata, depending on mode. */
7606 if (priv->vport_meta_mask)
7607 flow_dv_translate_item_meta_vport(matcher, key,
7608 priv->vport_meta_tag,
7609 priv->vport_meta_mask);
7611 flow_dv_translate_item_source_vport(matcher, key,
7612 priv->vport_id, mask);
7617 * Add ICMP6 item to matcher and to the value.
7619 * @param[in, out] matcher
7621 * @param[in, out] key
7622 * Flow matcher value.
7624 * Flow pattern to translate.
7626 * Item is inner pattern.
7629 flow_dv_translate_item_icmp6(void *matcher, void *key,
7630 const struct rte_flow_item *item,
7633 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
7634 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
7637 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7639 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7641 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7643 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7645 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7647 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7649 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7650 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
7654 icmp6_m = &rte_flow_item_icmp6_mask;
7655 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
7656 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
7657 icmp6_v->type & icmp6_m->type);
7658 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
7659 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
7660 icmp6_v->code & icmp6_m->code);
7664 * Add ICMP item to matcher and to the value.
7666 * @param[in, out] matcher
7668 * @param[in, out] key
7669 * Flow matcher value.
7671 * Flow pattern to translate.
7673 * Item is inner pattern.
7676 flow_dv_translate_item_icmp(void *matcher, void *key,
7677 const struct rte_flow_item *item,
7680 const struct rte_flow_item_icmp *icmp_m = item->mask;
7681 const struct rte_flow_item_icmp *icmp_v = item->spec;
7682 uint32_t icmp_header_data_m = 0;
7683 uint32_t icmp_header_data_v = 0;
7686 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7688 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7690 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7692 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7694 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7696 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7698 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7699 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
7703 icmp_m = &rte_flow_item_icmp_mask;
7704 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
7705 icmp_m->hdr.icmp_type);
7706 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
7707 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
7708 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
7709 icmp_m->hdr.icmp_code);
7710 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
7711 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
7712 icmp_header_data_m = rte_be_to_cpu_16(icmp_m->hdr.icmp_seq_nb);
7713 icmp_header_data_m |= rte_be_to_cpu_16(icmp_m->hdr.icmp_ident) << 16;
7714 if (icmp_header_data_m) {
7715 icmp_header_data_v = rte_be_to_cpu_16(icmp_v->hdr.icmp_seq_nb);
7716 icmp_header_data_v |=
7717 rte_be_to_cpu_16(icmp_v->hdr.icmp_ident) << 16;
7718 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_header_data,
7719 icmp_header_data_m);
7720 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_header_data,
7721 icmp_header_data_v & icmp_header_data_m);
7726 * Add GTP item to matcher and to the value.
7728 * @param[in, out] matcher
7730 * @param[in, out] key
7731 * Flow matcher value.
7733 * Flow pattern to translate.
7735 * Item is inner pattern.
7738 flow_dv_translate_item_gtp(void *matcher, void *key,
7739 const struct rte_flow_item *item, int inner)
7741 const struct rte_flow_item_gtp *gtp_m = item->mask;
7742 const struct rte_flow_item_gtp *gtp_v = item->spec;
7745 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7747 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7748 uint16_t dport = RTE_GTPU_UDP_PORT;
7751 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7753 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7755 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7757 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7759 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7760 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7761 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7766 gtp_m = &rte_flow_item_gtp_mask;
7767 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
7768 gtp_m->v_pt_rsv_flags);
7769 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
7770 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
7771 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
7772 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
7773 gtp_v->msg_type & gtp_m->msg_type);
7774 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
7775 rte_be_to_cpu_32(gtp_m->teid));
7776 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
7777 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
7781 * Add eCPRI item to matcher and to the value.
7784 * The devich to configure through.
7785 * @param[in, out] matcher
7787 * @param[in, out] key
7788 * Flow matcher value.
7790 * Flow pattern to translate.
7791 * @param[in] samples
7792 * Sample IDs to be used in the matching.
7795 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
7796 void *key, const struct rte_flow_item *item)
7798 struct mlx5_priv *priv = dev->data->dev_private;
7799 const struct rte_flow_item_ecpri *ecpri_m = item->mask;
7800 const struct rte_flow_item_ecpri *ecpri_v = item->spec;
7801 void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
7803 void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
7811 ecpri_m = &rte_flow_item_ecpri_mask;
7813 * Maximal four DW samples are supported in a single matching now.
7814 * Two are used now for a eCPRI matching:
7815 * 1. Type: one byte, mask should be 0x00ff0000 in network order
7816 * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
7819 if (!ecpri_m->hdr.common.u32)
7821 samples = priv->sh->fp[MLX5_FLEX_PARSER_ECPRI_0].ids;
7822 /* Need to take the whole DW as the mask to fill the entry. */
7823 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7824 prog_sample_field_value_0);
7825 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7826 prog_sample_field_value_0);
7827 /* Already big endian (network order) in the header. */
7828 *(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
7829 *(uint32_t *)dw_v = ecpri_v->hdr.common.u32;
7830 /* Sample#0, used for matching type, offset 0. */
7831 MLX5_SET(fte_match_set_misc4, misc4_m,
7832 prog_sample_field_id_0, samples[0]);
7833 /* It makes no sense to set the sample ID in the mask field. */
7834 MLX5_SET(fte_match_set_misc4, misc4_v,
7835 prog_sample_field_id_0, samples[0]);
7837 * Checking if message body part needs to be matched.
7838 * Some wildcard rules only matching type field should be supported.
7840 if (ecpri_m->hdr.dummy[0]) {
7841 switch (ecpri_v->hdr.common.type) {
7842 case RTE_ECPRI_MSG_TYPE_IQ_DATA:
7843 case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
7844 case RTE_ECPRI_MSG_TYPE_DLY_MSR:
7845 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7846 prog_sample_field_value_1);
7847 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7848 prog_sample_field_value_1);
7849 *(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
7850 *(uint32_t *)dw_v = ecpri_v->hdr.dummy[0];
7851 /* Sample#1, to match message body, offset 4. */
7852 MLX5_SET(fte_match_set_misc4, misc4_m,
7853 prog_sample_field_id_1, samples[1]);
7854 MLX5_SET(fte_match_set_misc4, misc4_v,
7855 prog_sample_field_id_1, samples[1]);
7858 /* Others, do not match any sample ID. */
7864 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
7866 #define HEADER_IS_ZERO(match_criteria, headers) \
7867 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
7868 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
7871 * Calculate flow matcher enable bitmap.
7873 * @param match_criteria
7874 * Pointer to flow matcher criteria.
7877 * Bitmap of enabled fields.
7880 flow_dv_matcher_enable(uint32_t *match_criteria)
7882 uint8_t match_criteria_enable;
7884 match_criteria_enable =
7885 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
7886 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
7887 match_criteria_enable |=
7888 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
7889 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
7890 match_criteria_enable |=
7891 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
7892 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
7893 match_criteria_enable |=
7894 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
7895 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
7896 match_criteria_enable |=
7897 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
7898 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
7899 match_criteria_enable |=
7900 (!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
7901 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
7902 return match_criteria_enable;
7905 struct mlx5_hlist_entry *
7906 flow_dv_tbl_create_cb(struct mlx5_hlist *list, uint64_t key64, void *cb_ctx)
7908 struct mlx5_dev_ctx_shared *sh = list->ctx;
7909 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
7910 struct rte_eth_dev *dev = ctx->dev;
7911 struct mlx5_flow_tbl_data_entry *tbl_data;
7912 struct mlx5_flow_tbl_tunnel_prm *tt_prm = ctx->data;
7913 struct rte_flow_error *error = ctx->error;
7914 union mlx5_flow_tbl_key key = { .v64 = key64 };
7915 struct mlx5_flow_tbl_resource *tbl;
7920 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
7922 rte_flow_error_set(error, ENOMEM,
7923 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7925 "cannot allocate flow table data entry");
7928 tbl_data->idx = idx;
7929 tbl_data->tunnel = tt_prm->tunnel;
7930 tbl_data->group_id = tt_prm->group_id;
7931 tbl_data->external = tt_prm->external;
7932 tbl_data->tunnel_offload = is_tunnel_offload_active(dev);
7933 tbl = &tbl_data->tbl;
7935 return &tbl_data->entry;
7937 domain = sh->fdb_domain;
7938 else if (key.direction)
7939 domain = sh->tx_domain;
7941 domain = sh->rx_domain;
7942 ret = mlx5_flow_os_create_flow_tbl(domain, key.table_id, &tbl->obj);
7944 rte_flow_error_set(error, ENOMEM,
7945 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7946 NULL, "cannot create flow table object");
7947 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7951 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
7952 (tbl->obj, &tbl_data->jump.action);
7954 rte_flow_error_set(error, ENOMEM,
7955 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7957 "cannot create flow jump action");
7958 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7959 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7963 return &tbl_data->entry;
7969 * @param[in, out] dev
7970 * Pointer to rte_eth_dev structure.
7971 * @param[in] table_id
7974 * Direction of the table.
7975 * @param[in] transfer
7976 * E-Switch or NIC flow.
7978 * Dummy entry for dv API.
7980 * pointer to error structure.
7983 * Returns tables resource based on the index, NULL in case of failed.
7985 struct mlx5_flow_tbl_resource *
7986 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
7987 uint32_t table_id, uint8_t egress,
7990 const struct mlx5_flow_tunnel *tunnel,
7991 uint32_t group_id, uint8_t dummy,
7992 struct rte_flow_error *error)
7994 struct mlx5_priv *priv = dev->data->dev_private;
7995 union mlx5_flow_tbl_key table_key = {
7997 .table_id = table_id,
7999 .domain = !!transfer,
8000 .direction = !!egress,
8003 struct mlx5_flow_tbl_tunnel_prm tt_prm = {
8005 .group_id = group_id,
8006 .external = external,
8008 struct mlx5_flow_cb_ctx ctx = {
8013 struct mlx5_hlist_entry *entry;
8014 struct mlx5_flow_tbl_data_entry *tbl_data;
8016 entry = mlx5_hlist_register(priv->sh->flow_tbls, table_key.v64, &ctx);
8018 rte_flow_error_set(error, ENOMEM,
8019 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8020 "cannot get table");
8023 tbl_data = container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
8024 return &tbl_data->tbl;
8028 flow_dv_tbl_remove_cb(struct mlx5_hlist *list,
8029 struct mlx5_hlist_entry *entry)
8031 struct mlx5_dev_ctx_shared *sh = list->ctx;
8032 struct mlx5_flow_tbl_data_entry *tbl_data =
8033 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
8035 MLX5_ASSERT(entry && sh);
8036 if (tbl_data->jump.action)
8037 mlx5_flow_os_destroy_flow_action(tbl_data->jump.action);
8038 if (tbl_data->tbl.obj)
8039 mlx5_flow_os_destroy_flow_tbl(tbl_data->tbl.obj);
8040 if (tbl_data->tunnel_offload && tbl_data->external) {
8041 struct mlx5_hlist_entry *he;
8042 struct mlx5_hlist *tunnel_grp_hash;
8043 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8044 union tunnel_tbl_key tunnel_key = {
8045 .tunnel_id = tbl_data->tunnel ?
8046 tbl_data->tunnel->tunnel_id : 0,
8047 .group = tbl_data->group_id
8049 union mlx5_flow_tbl_key table_key = {
8052 uint32_t table_id = table_key.table_id;
8054 tunnel_grp_hash = tbl_data->tunnel ?
8055 tbl_data->tunnel->groups :
8057 he = mlx5_hlist_lookup(tunnel_grp_hash, tunnel_key.val, NULL);
8059 struct tunnel_tbl_entry *tte;
8060 tte = container_of(he, typeof(*tte), hash);
8061 MLX5_ASSERT(tte->flow_table == table_id);
8062 mlx5_hlist_remove(tunnel_grp_hash, he);
8065 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8066 tunnel_flow_tbl_to_id(table_id));
8068 "Table_id %#x tunnel %u group %u released.",
8071 tbl_data->tunnel->tunnel_id : 0,
8072 tbl_data->group_id);
8074 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], tbl_data->idx);
8078 * Release a flow table.
8081 * Pointer to rte_eth_dev structure.
8083 * Table resource to be released.
8086 * Returns 0 if table was released, else return 1;
8089 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
8090 struct mlx5_flow_tbl_resource *tbl)
8092 struct mlx5_priv *priv = dev->data->dev_private;
8093 struct mlx5_dev_ctx_shared *sh = priv->sh;
8094 struct mlx5_flow_tbl_data_entry *tbl_data =
8095 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
8099 return mlx5_hlist_unregister(sh->flow_tbls, &tbl_data->entry);
8103 * Register the flow matcher.
8105 * @param[in, out] dev
8106 * Pointer to rte_eth_dev structure.
8107 * @param[in, out] matcher
8108 * Pointer to flow matcher.
8109 * @param[in, out] key
8110 * Pointer to flow table key.
8111 * @parm[in, out] dev_flow
8112 * Pointer to the dev_flow.
8114 * pointer to error structure.
8117 * 0 on success otherwise -errno and errno is set.
8120 flow_dv_matcher_register(struct rte_eth_dev *dev,
8121 struct mlx5_flow_dv_matcher *matcher,
8122 union mlx5_flow_tbl_key *key,
8123 struct mlx5_flow *dev_flow,
8124 struct rte_flow_error *error)
8126 struct mlx5_priv *priv = dev->data->dev_private;
8127 struct mlx5_dev_ctx_shared *sh = priv->sh;
8128 struct mlx5_flow_dv_matcher *cache_matcher;
8129 struct mlx5dv_flow_matcher_attr dv_attr = {
8130 .type = IBV_FLOW_ATTR_NORMAL,
8131 .match_mask = (void *)&matcher->mask,
8133 struct mlx5_flow_tbl_resource *tbl;
8134 struct mlx5_flow_tbl_data_entry *tbl_data;
8137 tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
8138 key->domain, false, NULL, 0, 0, error);
8140 return -rte_errno; /* No need to refill the error info */
8141 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
8142 /* Lookup from cache. */
8143 LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
8144 if (matcher->crc == cache_matcher->crc &&
8145 matcher->priority == cache_matcher->priority &&
8146 !memcmp((const void *)matcher->mask.buf,
8147 (const void *)cache_matcher->mask.buf,
8148 cache_matcher->mask.size)) {
8150 "%s group %u priority %hd use %s "
8151 "matcher %p: refcnt %d++",
8152 key->domain ? "FDB" : "NIC", key->table_id,
8153 cache_matcher->priority,
8154 key->direction ? "tx" : "rx",
8155 (void *)cache_matcher,
8156 __atomic_load_n(&cache_matcher->refcnt,
8158 __atomic_fetch_add(&cache_matcher->refcnt, 1,
8160 dev_flow->handle->dvh.matcher = cache_matcher;
8161 /* old matcher should not make the table ref++. */
8162 flow_dv_tbl_resource_release(dev, tbl);
8166 /* Register new matcher. */
8167 cache_matcher = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*cache_matcher), 0,
8169 if (!cache_matcher) {
8170 flow_dv_tbl_resource_release(dev, tbl);
8171 return rte_flow_error_set(error, ENOMEM,
8172 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8173 "cannot allocate matcher memory");
8175 *cache_matcher = *matcher;
8176 dv_attr.match_criteria_enable =
8177 flow_dv_matcher_enable(cache_matcher->mask.buf);
8178 dv_attr.priority = matcher->priority;
8180 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
8181 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
8182 &cache_matcher->matcher_object);
8184 mlx5_free(cache_matcher);
8185 #ifdef HAVE_MLX5DV_DR
8186 flow_dv_tbl_resource_release(dev, tbl);
8188 return rte_flow_error_set(error, ENOMEM,
8189 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8190 NULL, "cannot create matcher");
8192 /* Save the table information */
8193 cache_matcher->tbl = tbl;
8194 /* only matcher ref++, table ref++ already done above in get API. */
8195 __atomic_store_n(&cache_matcher->refcnt, 1, __ATOMIC_RELAXED);
8196 LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
8197 dev_flow->handle->dvh.matcher = cache_matcher;
8198 DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
8199 key->domain ? "FDB" : "NIC", key->table_id,
8200 cache_matcher->priority,
8201 key->direction ? "tx" : "rx", (void *)cache_matcher,
8202 __atomic_load_n(&cache_matcher->refcnt, __ATOMIC_RELAXED));
8206 struct mlx5_hlist_entry *
8207 flow_dv_tag_create_cb(struct mlx5_hlist *list, uint64_t key, void *ctx)
8209 struct mlx5_dev_ctx_shared *sh = list->ctx;
8210 struct rte_flow_error *error = ctx;
8211 struct mlx5_flow_dv_tag_resource *entry;
8215 entry = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG], &idx);
8217 rte_flow_error_set(error, ENOMEM,
8218 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8219 "cannot allocate resource memory");
8223 ret = mlx5_flow_os_create_flow_action_tag(key,
8226 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], idx);
8227 rte_flow_error_set(error, ENOMEM,
8228 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8229 NULL, "cannot create action");
8232 return &entry->entry;
8236 * Find existing tag resource or create and register a new one.
8238 * @param dev[in, out]
8239 * Pointer to rte_eth_dev structure.
8240 * @param[in, out] tag_be24
8241 * Tag value in big endian then R-shift 8.
8242 * @parm[in, out] dev_flow
8243 * Pointer to the dev_flow.
8245 * pointer to error structure.
8248 * 0 on success otherwise -errno and errno is set.
8251 flow_dv_tag_resource_register
8252 (struct rte_eth_dev *dev,
8254 struct mlx5_flow *dev_flow,
8255 struct rte_flow_error *error)
8257 struct mlx5_priv *priv = dev->data->dev_private;
8258 struct mlx5_flow_dv_tag_resource *cache_resource;
8259 struct mlx5_hlist_entry *entry;
8261 entry = mlx5_hlist_register(priv->sh->tag_table, tag_be24, error);
8263 cache_resource = container_of
8264 (entry, struct mlx5_flow_dv_tag_resource, entry);
8265 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
8266 dev_flow->dv.tag_resource = cache_resource;
8273 flow_dv_tag_remove_cb(struct mlx5_hlist *list,
8274 struct mlx5_hlist_entry *entry)
8276 struct mlx5_dev_ctx_shared *sh = list->ctx;
8277 struct mlx5_flow_dv_tag_resource *tag =
8278 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
8280 MLX5_ASSERT(tag && sh && tag->action);
8281 claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
8282 DRV_LOG(DEBUG, "Tag %p: removed.", (void *)tag);
8283 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], tag->idx);
8290 * Pointer to Ethernet device.
8295 * 1 while a reference on it exists, 0 when freed.
8298 flow_dv_tag_release(struct rte_eth_dev *dev,
8301 struct mlx5_priv *priv = dev->data->dev_private;
8302 struct mlx5_flow_dv_tag_resource *tag;
8304 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
8307 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
8308 dev->data->port_id, (void *)tag, tag->entry.ref_cnt);
8309 return mlx5_hlist_unregister(priv->sh->tag_table, &tag->entry);
8313 * Translate port ID action to vport.
8316 * Pointer to rte_eth_dev structure.
8318 * Pointer to the port ID action.
8319 * @param[out] dst_port_id
8320 * The target port ID.
8322 * Pointer to the error structure.
8325 * 0 on success, a negative errno value otherwise and rte_errno is set.
8328 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
8329 const struct rte_flow_action *action,
8330 uint32_t *dst_port_id,
8331 struct rte_flow_error *error)
8334 struct mlx5_priv *priv;
8335 const struct rte_flow_action_port_id *conf =
8336 (const struct rte_flow_action_port_id *)action->conf;
8338 port = conf->original ? dev->data->port_id : conf->id;
8339 priv = mlx5_port_to_eswitch_info(port, false);
8341 return rte_flow_error_set(error, -rte_errno,
8342 RTE_FLOW_ERROR_TYPE_ACTION,
8344 "No eswitch info was found for port");
8345 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
8347 * This parameter is transferred to
8348 * mlx5dv_dr_action_create_dest_ib_port().
8350 *dst_port_id = priv->dev_port;
8353 * Legacy mode, no LAG configurations is supported.
8354 * This parameter is transferred to
8355 * mlx5dv_dr_action_create_dest_vport().
8357 *dst_port_id = priv->vport_id;
8363 * Create a counter with aging configuration.
8366 * Pointer to rte_eth_dev structure.
8368 * Pointer to the counter action configuration.
8370 * Pointer to the aging action configuration.
8373 * Index to flow counter on success, 0 otherwise.
8376 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
8377 struct mlx5_flow *dev_flow,
8378 const struct rte_flow_action_count *count,
8379 const struct rte_flow_action_age *age)
8382 struct mlx5_age_param *age_param;
8384 if (count && count->shared)
8385 counter = flow_dv_counter_get_shared(dev, count->id);
8387 counter = flow_dv_counter_alloc(dev, !!age);
8388 if (!counter || age == NULL)
8390 age_param = flow_dv_counter_idx_get_age(dev, counter);
8391 age_param->context = age->context ? age->context :
8392 (void *)(uintptr_t)(dev_flow->flow_idx);
8393 age_param->timeout = age->timeout;
8394 age_param->port_id = dev->data->port_id;
8395 __atomic_store_n(&age_param->sec_since_last_hit, 0, __ATOMIC_RELAXED);
8396 __atomic_store_n(&age_param->state, AGE_CANDIDATE, __ATOMIC_RELAXED);
8400 * Add Tx queue matcher
8403 * Pointer to the dev struct.
8404 * @param[in, out] matcher
8406 * @param[in, out] key
8407 * Flow matcher value.
8409 * Flow pattern to translate.
8411 * Item is inner pattern.
8414 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
8415 void *matcher, void *key,
8416 const struct rte_flow_item *item)
8418 const struct mlx5_rte_flow_item_tx_queue *queue_m;
8419 const struct mlx5_rte_flow_item_tx_queue *queue_v;
8421 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8423 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8424 struct mlx5_txq_ctrl *txq;
8428 queue_m = (const void *)item->mask;
8431 queue_v = (const void *)item->spec;
8434 txq = mlx5_txq_get(dev, queue_v->queue);
8437 queue = txq->obj->sq->id;
8438 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
8439 MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
8440 queue & queue_m->queue);
8441 mlx5_txq_release(dev, queue_v->queue);
8445 * Set the hash fields according to the @p flow information.
8447 * @param[in] dev_flow
8448 * Pointer to the mlx5_flow.
8449 * @param[in] rss_desc
8450 * Pointer to the mlx5_flow_rss_desc.
8453 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
8454 struct mlx5_flow_rss_desc *rss_desc)
8456 uint64_t items = dev_flow->handle->layers;
8458 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
8460 dev_flow->hash_fields = 0;
8461 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
8462 if (rss_desc->level >= 2) {
8463 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
8467 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
8468 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
8469 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
8470 if (rss_types & ETH_RSS_L3_SRC_ONLY)
8471 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
8472 else if (rss_types & ETH_RSS_L3_DST_ONLY)
8473 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
8475 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
8477 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
8478 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
8479 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
8480 if (rss_types & ETH_RSS_L3_SRC_ONLY)
8481 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
8482 else if (rss_types & ETH_RSS_L3_DST_ONLY)
8483 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
8485 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
8488 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
8489 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
8490 if (rss_types & ETH_RSS_UDP) {
8491 if (rss_types & ETH_RSS_L4_SRC_ONLY)
8492 dev_flow->hash_fields |=
8493 IBV_RX_HASH_SRC_PORT_UDP;
8494 else if (rss_types & ETH_RSS_L4_DST_ONLY)
8495 dev_flow->hash_fields |=
8496 IBV_RX_HASH_DST_PORT_UDP;
8498 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
8500 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
8501 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
8502 if (rss_types & ETH_RSS_TCP) {
8503 if (rss_types & ETH_RSS_L4_SRC_ONLY)
8504 dev_flow->hash_fields |=
8505 IBV_RX_HASH_SRC_PORT_TCP;
8506 else if (rss_types & ETH_RSS_L4_DST_ONLY)
8507 dev_flow->hash_fields |=
8508 IBV_RX_HASH_DST_PORT_TCP;
8510 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
8516 * Create an Rx Hash queue.
8519 * Pointer to Ethernet device.
8520 * @param[in] dev_flow
8521 * Pointer to the mlx5_flow.
8522 * @param[in] rss_desc
8523 * Pointer to the mlx5_flow_rss_desc.
8524 * @param[out] hrxq_idx
8525 * Hash Rx queue index.
8528 * The Verbs/DevX object initialised, NULL otherwise and rte_errno is set.
8530 static struct mlx5_hrxq *
8531 flow_dv_handle_rx_queue(struct rte_eth_dev *dev,
8532 struct mlx5_flow *dev_flow,
8533 struct mlx5_flow_rss_desc *rss_desc,
8536 struct mlx5_priv *priv = dev->data->dev_private;
8537 struct mlx5_flow_handle *dh = dev_flow->handle;
8538 struct mlx5_hrxq *hrxq;
8540 MLX5_ASSERT(rss_desc->queue_num);
8541 *hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key, MLX5_RSS_HASH_KEY_LEN,
8542 dev_flow->hash_fields,
8543 rss_desc->queue, rss_desc->queue_num);
8545 *hrxq_idx = mlx5_hrxq_new
8546 (dev, rss_desc->key, MLX5_RSS_HASH_KEY_LEN,
8547 dev_flow->hash_fields,
8548 rss_desc->queue, rss_desc->queue_num,
8549 !!(dh->layers & MLX5_FLOW_LAYER_TUNNEL),
8554 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8560 * Find existing sample resource or create and register a new one.
8562 * @param[in, out] dev
8563 * Pointer to rte_eth_dev structure.
8565 * Attributes of flow that includes this item.
8566 * @param[in] resource
8567 * Pointer to sample resource.
8568 * @parm[in, out] dev_flow
8569 * Pointer to the dev_flow.
8570 * @param[in, out] sample_dv_actions
8571 * Pointer to sample actions list.
8573 * pointer to error structure.
8576 * 0 on success otherwise -errno and errno is set.
8579 flow_dv_sample_resource_register(struct rte_eth_dev *dev,
8580 const struct rte_flow_attr *attr,
8581 struct mlx5_flow_dv_sample_resource *resource,
8582 struct mlx5_flow *dev_flow,
8583 void **sample_dv_actions,
8584 struct rte_flow_error *error)
8586 struct mlx5_flow_dv_sample_resource *cache_resource;
8587 struct mlx5dv_dr_flow_sampler_attr sampler_attr;
8588 struct mlx5_priv *priv = dev->data->dev_private;
8589 struct mlx5_dev_ctx_shared *sh = priv->sh;
8590 struct mlx5_flow_tbl_resource *tbl;
8592 const uint32_t next_ft_step = 1;
8593 uint32_t next_ft_id = resource->ft_id + next_ft_step;
8595 /* Lookup a matching resource from cache. */
8596 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_SAMPLE], sh->sample_action_list,
8597 idx, cache_resource, next) {
8598 if (resource->ratio == cache_resource->ratio &&
8599 resource->ft_type == cache_resource->ft_type &&
8600 resource->ft_id == cache_resource->ft_id &&
8601 resource->set_action == cache_resource->set_action &&
8602 !memcmp((void *)&resource->sample_act,
8603 (void *)&cache_resource->sample_act,
8604 sizeof(struct mlx5_flow_sub_actions_list))) {
8605 DRV_LOG(DEBUG, "sample resource %p: refcnt %d++",
8606 (void *)cache_resource,
8607 __atomic_load_n(&cache_resource->refcnt,
8609 __atomic_fetch_add(&cache_resource->refcnt, 1,
8611 dev_flow->handle->dvh.rix_sample = idx;
8612 dev_flow->dv.sample_res = cache_resource;
8616 /* Register new sample resource. */
8617 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE],
8618 &dev_flow->handle->dvh.rix_sample);
8619 if (!cache_resource)
8620 return rte_flow_error_set(error, ENOMEM,
8621 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8623 "cannot allocate resource memory");
8624 *cache_resource = *resource;
8625 /* Create normal path table level */
8626 tbl = flow_dv_tbl_resource_get(dev, next_ft_id,
8627 attr->egress, attr->transfer,
8628 dev_flow->external, NULL, 0, 0, error);
8630 rte_flow_error_set(error, ENOMEM,
8631 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8633 "fail to create normal path table "
8637 cache_resource->normal_path_tbl = tbl;
8638 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
8639 cache_resource->default_miss =
8640 mlx5_glue->dr_create_flow_action_default_miss();
8641 if (!cache_resource->default_miss) {
8642 rte_flow_error_set(error, ENOMEM,
8643 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8645 "cannot create default miss "
8649 sample_dv_actions[resource->sample_act.actions_num++] =
8650 cache_resource->default_miss;
8652 /* Create a DR sample action */
8653 sampler_attr.sample_ratio = cache_resource->ratio;
8654 sampler_attr.default_next_table = tbl->obj;
8655 sampler_attr.num_sample_actions = resource->sample_act.actions_num;
8656 sampler_attr.sample_actions = (struct mlx5dv_dr_action **)
8657 &sample_dv_actions[0];
8658 sampler_attr.action = cache_resource->set_action;
8659 cache_resource->verbs_action =
8660 mlx5_glue->dr_create_flow_action_sampler(&sampler_attr);
8661 if (!cache_resource->verbs_action) {
8662 rte_flow_error_set(error, ENOMEM,
8663 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8664 NULL, "cannot create sample action");
8667 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
8668 ILIST_INSERT(sh->ipool[MLX5_IPOOL_SAMPLE], &sh->sample_action_list,
8669 dev_flow->handle->dvh.rix_sample, cache_resource,
8671 dev_flow->dv.sample_res = cache_resource;
8672 DRV_LOG(DEBUG, "new sample resource %p: refcnt %d++",
8673 (void *)cache_resource,
8674 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
8677 if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
8678 if (cache_resource->default_miss)
8679 claim_zero(mlx5_glue->destroy_flow_action
8680 (cache_resource->default_miss));
8682 if (cache_resource->sample_idx.rix_hrxq &&
8683 !mlx5_hrxq_release(dev,
8684 cache_resource->sample_idx.rix_hrxq))
8685 cache_resource->sample_idx.rix_hrxq = 0;
8686 if (cache_resource->sample_idx.rix_tag &&
8687 !flow_dv_tag_release(dev,
8688 cache_resource->sample_idx.rix_tag))
8689 cache_resource->sample_idx.rix_tag = 0;
8690 if (cache_resource->sample_idx.cnt) {
8691 flow_dv_counter_release(dev,
8692 cache_resource->sample_idx.cnt);
8693 cache_resource->sample_idx.cnt = 0;
8696 if (cache_resource->normal_path_tbl)
8697 flow_dv_tbl_resource_release(dev,
8698 cache_resource->normal_path_tbl);
8699 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_SAMPLE],
8700 dev_flow->handle->dvh.rix_sample);
8701 dev_flow->handle->dvh.rix_sample = 0;
8706 * Find existing destination array resource or create and register a new one.
8708 * @param[in, out] dev
8709 * Pointer to rte_eth_dev structure.
8711 * Attributes of flow that includes this item.
8712 * @param[in] resource
8713 * Pointer to destination array resource.
8714 * @parm[in, out] dev_flow
8715 * Pointer to the dev_flow.
8717 * pointer to error structure.
8720 * 0 on success otherwise -errno and errno is set.
8723 flow_dv_dest_array_resource_register(struct rte_eth_dev *dev,
8724 const struct rte_flow_attr *attr,
8725 struct mlx5_flow_dv_dest_array_resource *resource,
8726 struct mlx5_flow *dev_flow,
8727 struct rte_flow_error *error)
8729 struct mlx5_flow_dv_dest_array_resource *cache_resource;
8730 struct mlx5dv_dr_action_dest_attr *dest_attr[MLX5_MAX_DEST_NUM] = { 0 };
8731 struct mlx5dv_dr_action_dest_reformat dest_reformat[MLX5_MAX_DEST_NUM];
8732 struct mlx5_priv *priv = dev->data->dev_private;
8733 struct mlx5_dev_ctx_shared *sh = priv->sh;
8734 struct mlx5_flow_sub_actions_list *sample_act;
8735 struct mlx5dv_dr_domain *domain;
8738 /* Lookup a matching resource from cache. */
8739 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8740 sh->dest_array_list,
8741 idx, cache_resource, next) {
8742 if (resource->num_of_dest == cache_resource->num_of_dest &&
8743 resource->ft_type == cache_resource->ft_type &&
8744 !memcmp((void *)cache_resource->sample_act,
8745 (void *)resource->sample_act,
8746 (resource->num_of_dest *
8747 sizeof(struct mlx5_flow_sub_actions_list)))) {
8748 DRV_LOG(DEBUG, "dest array resource %p: refcnt %d++",
8749 (void *)cache_resource,
8750 __atomic_load_n(&cache_resource->refcnt,
8752 __atomic_fetch_add(&cache_resource->refcnt, 1,
8754 dev_flow->handle->dvh.rix_dest_array = idx;
8755 dev_flow->dv.dest_array_res = cache_resource;
8759 /* Register new destination array resource. */
8760 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8761 &dev_flow->handle->dvh.rix_dest_array);
8762 if (!cache_resource)
8763 return rte_flow_error_set(error, ENOMEM,
8764 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8766 "cannot allocate resource memory");
8767 *cache_resource = *resource;
8769 domain = sh->fdb_domain;
8770 else if (attr->ingress)
8771 domain = sh->rx_domain;
8773 domain = sh->tx_domain;
8774 for (idx = 0; idx < resource->num_of_dest; idx++) {
8775 dest_attr[idx] = (struct mlx5dv_dr_action_dest_attr *)
8776 mlx5_malloc(MLX5_MEM_ZERO,
8777 sizeof(struct mlx5dv_dr_action_dest_attr),
8779 if (!dest_attr[idx]) {
8780 rte_flow_error_set(error, ENOMEM,
8781 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8783 "cannot allocate resource memory");
8786 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST;
8787 sample_act = &resource->sample_act[idx];
8788 if (sample_act->action_flags == MLX5_FLOW_ACTION_QUEUE) {
8789 dest_attr[idx]->dest = sample_act->dr_queue_action;
8790 } else if (sample_act->action_flags ==
8791 (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_ENCAP)) {
8792 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST_REFORMAT;
8793 dest_attr[idx]->dest_reformat = &dest_reformat[idx];
8794 dest_attr[idx]->dest_reformat->reformat =
8795 sample_act->dr_encap_action;
8796 dest_attr[idx]->dest_reformat->dest =
8797 sample_act->dr_port_id_action;
8798 } else if (sample_act->action_flags ==
8799 MLX5_FLOW_ACTION_PORT_ID) {
8800 dest_attr[idx]->dest = sample_act->dr_port_id_action;
8803 /* create a dest array actioin */
8804 cache_resource->action = mlx5_glue->dr_create_flow_action_dest_array
8806 cache_resource->num_of_dest,
8808 if (!cache_resource->action) {
8809 rte_flow_error_set(error, ENOMEM,
8810 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8812 "cannot create destination array action");
8815 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
8816 ILIST_INSERT(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8817 &sh->dest_array_list,
8818 dev_flow->handle->dvh.rix_dest_array, cache_resource,
8820 dev_flow->dv.dest_array_res = cache_resource;
8821 DRV_LOG(DEBUG, "new destination array resource %p: refcnt %d++",
8822 (void *)cache_resource,
8823 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
8824 for (idx = 0; idx < resource->num_of_dest; idx++)
8825 mlx5_free(dest_attr[idx]);
8828 for (idx = 0; idx < resource->num_of_dest; idx++) {
8829 struct mlx5_flow_sub_actions_idx *act_res =
8830 &cache_resource->sample_idx[idx];
8831 if (act_res->rix_hrxq &&
8832 !mlx5_hrxq_release(dev,
8834 act_res->rix_hrxq = 0;
8835 if (act_res->rix_encap_decap &&
8836 !flow_dv_encap_decap_resource_release(dev,
8837 act_res->rix_encap_decap))
8838 act_res->rix_encap_decap = 0;
8839 if (act_res->rix_port_id_action &&
8840 !flow_dv_port_id_action_resource_release(dev,
8841 act_res->rix_port_id_action))
8842 act_res->rix_port_id_action = 0;
8844 mlx5_free(dest_attr[idx]);
8847 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8848 dev_flow->handle->dvh.rix_dest_array);
8849 dev_flow->handle->dvh.rix_dest_array = 0;
8854 * Convert Sample action to DV specification.
8857 * Pointer to rte_eth_dev structure.
8859 * Pointer to action structure.
8860 * @param[in, out] dev_flow
8861 * Pointer to the mlx5_flow.
8863 * Pointer to the flow attributes.
8864 * @param[in, out] num_of_dest
8865 * Pointer to the num of destination.
8866 * @param[in, out] sample_actions
8867 * Pointer to sample actions list.
8868 * @param[in, out] res
8869 * Pointer to sample resource.
8871 * Pointer to the error structure.
8874 * 0 on success, a negative errno value otherwise and rte_errno is set.
8877 flow_dv_translate_action_sample(struct rte_eth_dev *dev,
8878 const struct rte_flow_action *action,
8879 struct mlx5_flow *dev_flow,
8880 const struct rte_flow_attr *attr,
8881 uint32_t *num_of_dest,
8882 void **sample_actions,
8883 struct mlx5_flow_dv_sample_resource *res,
8884 struct rte_flow_error *error)
8886 struct mlx5_priv *priv = dev->data->dev_private;
8887 const struct rte_flow_action_sample *sample_action;
8888 const struct rte_flow_action *sub_actions;
8889 const struct rte_flow_action_queue *queue;
8890 struct mlx5_flow_sub_actions_list *sample_act;
8891 struct mlx5_flow_sub_actions_idx *sample_idx;
8892 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
8893 struct mlx5_flow_rss_desc *rss_desc;
8894 uint64_t action_flags = 0;
8897 rss_desc = &wks->rss_desc[!!wks->flow_nested_idx];
8898 sample_act = &res->sample_act;
8899 sample_idx = &res->sample_idx;
8900 sample_action = (const struct rte_flow_action_sample *)action->conf;
8901 res->ratio = sample_action->ratio;
8902 sub_actions = sample_action->actions;
8903 for (; sub_actions->type != RTE_FLOW_ACTION_TYPE_END; sub_actions++) {
8904 int type = sub_actions->type;
8905 uint32_t pre_rix = 0;
8908 case RTE_FLOW_ACTION_TYPE_QUEUE:
8910 struct mlx5_hrxq *hrxq;
8913 queue = sub_actions->conf;
8914 rss_desc->queue_num = 1;
8915 rss_desc->queue[0] = queue->index;
8916 hrxq = flow_dv_handle_rx_queue(dev, dev_flow,
8917 rss_desc, &hrxq_idx);
8919 return rte_flow_error_set
8921 RTE_FLOW_ERROR_TYPE_ACTION,
8923 "cannot create fate queue");
8924 sample_act->dr_queue_action = hrxq->action;
8925 sample_idx->rix_hrxq = hrxq_idx;
8926 sample_actions[sample_act->actions_num++] =
8929 action_flags |= MLX5_FLOW_ACTION_QUEUE;
8930 if (action_flags & MLX5_FLOW_ACTION_MARK)
8931 dev_flow->handle->rix_hrxq = hrxq_idx;
8932 dev_flow->handle->fate_action =
8933 MLX5_FLOW_FATE_QUEUE;
8936 case RTE_FLOW_ACTION_TYPE_MARK:
8938 uint32_t tag_be = mlx5_flow_mark_set
8939 (((const struct rte_flow_action_mark *)
8940 (sub_actions->conf))->id);
8942 dev_flow->handle->mark = 1;
8943 pre_rix = dev_flow->handle->dvh.rix_tag;
8944 /* Save the mark resource before sample */
8945 pre_r = dev_flow->dv.tag_resource;
8946 if (flow_dv_tag_resource_register(dev, tag_be,
8949 MLX5_ASSERT(dev_flow->dv.tag_resource);
8950 sample_act->dr_tag_action =
8951 dev_flow->dv.tag_resource->action;
8952 sample_idx->rix_tag =
8953 dev_flow->handle->dvh.rix_tag;
8954 sample_actions[sample_act->actions_num++] =
8955 sample_act->dr_tag_action;
8956 /* Recover the mark resource after sample */
8957 dev_flow->dv.tag_resource = pre_r;
8958 dev_flow->handle->dvh.rix_tag = pre_rix;
8959 action_flags |= MLX5_FLOW_ACTION_MARK;
8962 case RTE_FLOW_ACTION_TYPE_COUNT:
8966 counter = flow_dv_translate_create_counter(dev,
8967 dev_flow, sub_actions->conf, 0);
8969 return rte_flow_error_set
8971 RTE_FLOW_ERROR_TYPE_ACTION,
8973 "cannot create counter"
8975 sample_idx->cnt = counter;
8976 sample_act->dr_cnt_action =
8977 (flow_dv_counter_get_by_idx(dev,
8978 counter, NULL))->action;
8979 sample_actions[sample_act->actions_num++] =
8980 sample_act->dr_cnt_action;
8981 action_flags |= MLX5_FLOW_ACTION_COUNT;
8984 case RTE_FLOW_ACTION_TYPE_PORT_ID:
8986 struct mlx5_flow_dv_port_id_action_resource
8988 uint32_t port_id = 0;
8990 memset(&port_id_resource, 0, sizeof(port_id_resource));
8991 /* Save the port id resource before sample */
8992 pre_rix = dev_flow->handle->rix_port_id_action;
8993 pre_r = dev_flow->dv.port_id_action;
8994 if (flow_dv_translate_action_port_id(dev, sub_actions,
8997 port_id_resource.port_id = port_id;
8998 if (flow_dv_port_id_action_resource_register
8999 (dev, &port_id_resource, dev_flow, error))
9001 sample_act->dr_port_id_action =
9002 dev_flow->dv.port_id_action->action;
9003 sample_idx->rix_port_id_action =
9004 dev_flow->handle->rix_port_id_action;
9005 sample_actions[sample_act->actions_num++] =
9006 sample_act->dr_port_id_action;
9007 /* Recover the port id resource after sample */
9008 dev_flow->dv.port_id_action = pre_r;
9009 dev_flow->handle->rix_port_id_action = pre_rix;
9011 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9014 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
9015 /* Save the encap resource before sample */
9016 pre_rix = dev_flow->handle->dvh.rix_encap_decap;
9017 pre_r = dev_flow->dv.encap_decap;
9018 if (flow_dv_create_action_l2_encap(dev, sub_actions,
9023 sample_act->dr_encap_action =
9024 dev_flow->dv.encap_decap->action;
9025 sample_idx->rix_encap_decap =
9026 dev_flow->handle->dvh.rix_encap_decap;
9027 sample_actions[sample_act->actions_num++] =
9028 sample_act->dr_encap_action;
9029 /* Recover the encap resource after sample */
9030 dev_flow->dv.encap_decap = pre_r;
9031 dev_flow->handle->dvh.rix_encap_decap = pre_rix;
9032 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9035 return rte_flow_error_set(error, EINVAL,
9036 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9038 "Not support for sampler action");
9041 sample_act->action_flags = action_flags;
9042 res->ft_id = dev_flow->dv.group;
9043 if (attr->transfer) {
9045 uint32_t action_in[MLX5_ST_SZ_DW(set_action_in)];
9046 uint64_t set_action;
9047 } action_ctx = { .set_action = 0 };
9049 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
9050 MLX5_SET(set_action_in, action_ctx.action_in, action_type,
9051 MLX5_MODIFICATION_TYPE_SET);
9052 MLX5_SET(set_action_in, action_ctx.action_in, field,
9053 MLX5_MODI_META_REG_C_0);
9054 MLX5_SET(set_action_in, action_ctx.action_in, data,
9055 priv->vport_meta_tag);
9056 res->set_action = action_ctx.set_action;
9057 } else if (attr->ingress) {
9058 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9064 * Convert Sample action to DV specification.
9067 * Pointer to rte_eth_dev structure.
9068 * @param[in, out] dev_flow
9069 * Pointer to the mlx5_flow.
9071 * Pointer to the flow attributes.
9072 * @param[in] num_of_dest
9073 * The num of destination.
9074 * @param[in, out] res
9075 * Pointer to sample resource.
9076 * @param[in, out] mdest_res
9077 * Pointer to destination array resource.
9078 * @param[in] sample_actions
9079 * Pointer to sample path actions list.
9080 * @param[in] action_flags
9081 * Holds the actions detected until now.
9083 * Pointer to the error structure.
9086 * 0 on success, a negative errno value otherwise and rte_errno is set.
9089 flow_dv_create_action_sample(struct rte_eth_dev *dev,
9090 struct mlx5_flow *dev_flow,
9091 const struct rte_flow_attr *attr,
9092 uint32_t num_of_dest,
9093 struct mlx5_flow_dv_sample_resource *res,
9094 struct mlx5_flow_dv_dest_array_resource *mdest_res,
9095 void **sample_actions,
9096 uint64_t action_flags,
9097 struct rte_flow_error *error)
9099 /* update normal path action resource into last index of array */
9100 uint32_t dest_index = MLX5_MAX_DEST_NUM - 1;
9101 struct mlx5_flow_sub_actions_list *sample_act =
9102 &mdest_res->sample_act[dest_index];
9103 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
9104 struct mlx5_flow_rss_desc *rss_desc;
9105 uint32_t normal_idx = 0;
9106 struct mlx5_hrxq *hrxq;
9110 rss_desc = &wks->rss_desc[!!wks->flow_nested_idx];
9111 if (num_of_dest > 1) {
9112 if (sample_act->action_flags & MLX5_FLOW_ACTION_QUEUE) {
9113 /* Handle QP action for mirroring */
9114 hrxq = flow_dv_handle_rx_queue(dev, dev_flow,
9115 rss_desc, &hrxq_idx);
9117 return rte_flow_error_set
9119 RTE_FLOW_ERROR_TYPE_ACTION,
9121 "cannot create rx queue");
9123 mdest_res->sample_idx[dest_index].rix_hrxq = hrxq_idx;
9124 sample_act->dr_queue_action = hrxq->action;
9125 if (action_flags & MLX5_FLOW_ACTION_MARK)
9126 dev_flow->handle->rix_hrxq = hrxq_idx;
9127 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9129 if (sample_act->action_flags & MLX5_FLOW_ACTION_ENCAP) {
9131 mdest_res->sample_idx[dest_index].rix_encap_decap =
9132 dev_flow->handle->dvh.rix_encap_decap;
9133 sample_act->dr_encap_action =
9134 dev_flow->dv.encap_decap->action;
9136 if (sample_act->action_flags & MLX5_FLOW_ACTION_PORT_ID) {
9138 mdest_res->sample_idx[dest_index].rix_port_id_action =
9139 dev_flow->handle->rix_port_id_action;
9140 sample_act->dr_port_id_action =
9141 dev_flow->dv.port_id_action->action;
9143 sample_act->actions_num = normal_idx;
9144 /* update sample action resource into first index of array */
9145 mdest_res->ft_type = res->ft_type;
9146 memcpy(&mdest_res->sample_idx[0], &res->sample_idx,
9147 sizeof(struct mlx5_flow_sub_actions_idx));
9148 memcpy(&mdest_res->sample_act[0], &res->sample_act,
9149 sizeof(struct mlx5_flow_sub_actions_list));
9150 mdest_res->num_of_dest = num_of_dest;
9151 if (flow_dv_dest_array_resource_register(dev, attr, mdest_res,
9153 return rte_flow_error_set(error, EINVAL,
9154 RTE_FLOW_ERROR_TYPE_ACTION,
9155 NULL, "can't create sample "
9158 if (flow_dv_sample_resource_register(dev, attr, res, dev_flow,
9159 sample_actions, error))
9160 return rte_flow_error_set(error, EINVAL,
9161 RTE_FLOW_ERROR_TYPE_ACTION,
9163 "can't create sample action");
9169 * Fill the flow with DV spec, lock free
9170 * (mutex should be acquired by caller).
9173 * Pointer to rte_eth_dev structure.
9174 * @param[in, out] dev_flow
9175 * Pointer to the sub flow.
9177 * Pointer to the flow attributes.
9179 * Pointer to the list of items.
9180 * @param[in] actions
9181 * Pointer to the list of actions.
9183 * Pointer to the error structure.
9186 * 0 on success, a negative errno value otherwise and rte_errno is set.
9189 __flow_dv_translate(struct rte_eth_dev *dev,
9190 struct mlx5_flow *dev_flow,
9191 const struct rte_flow_attr *attr,
9192 const struct rte_flow_item items[],
9193 const struct rte_flow_action actions[],
9194 struct rte_flow_error *error)
9196 struct mlx5_priv *priv = dev->data->dev_private;
9197 struct mlx5_dev_config *dev_conf = &priv->config;
9198 struct rte_flow *flow = dev_flow->flow;
9199 struct mlx5_flow_handle *handle = dev_flow->handle;
9200 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
9201 struct mlx5_flow_rss_desc *rss_desc;
9202 uint64_t item_flags = 0;
9203 uint64_t last_item = 0;
9204 uint64_t action_flags = 0;
9205 uint64_t priority = attr->priority;
9206 struct mlx5_flow_dv_matcher matcher = {
9208 .size = sizeof(matcher.mask.buf) -
9209 MLX5_ST_SZ_BYTES(fte_match_set_misc4),
9213 bool actions_end = false;
9215 struct mlx5_flow_dv_modify_hdr_resource res;
9216 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
9217 sizeof(struct mlx5_modification_cmd) *
9218 (MLX5_MAX_MODIFY_NUM + 1)];
9220 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
9221 const struct rte_flow_action_count *count = NULL;
9222 const struct rte_flow_action_age *age = NULL;
9223 union flow_dv_attr flow_attr = { .attr = 0 };
9225 union mlx5_flow_tbl_key tbl_key;
9226 uint32_t modify_action_position = UINT32_MAX;
9227 void *match_mask = matcher.mask.buf;
9228 void *match_value = dev_flow->dv.value.buf;
9229 uint8_t next_protocol = 0xff;
9230 struct rte_vlan_hdr vlan = { 0 };
9231 struct mlx5_flow_dv_dest_array_resource mdest_res;
9232 struct mlx5_flow_dv_sample_resource sample_res;
9233 void *sample_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9234 struct mlx5_flow_sub_actions_list *sample_act;
9235 uint32_t sample_act_pos = UINT32_MAX;
9236 uint32_t num_of_dest = 0;
9237 int tmp_actions_n = 0;
9240 const struct mlx5_flow_tunnel *tunnel;
9241 struct flow_grp_info grp_info = {
9242 .external = !!dev_flow->external,
9243 .transfer = !!attr->transfer,
9244 .fdb_def_rule = !!priv->fdb_def_rule,
9248 rss_desc = &wks->rss_desc[!!wks->flow_nested_idx];
9249 memset(&mdest_res, 0, sizeof(struct mlx5_flow_dv_dest_array_resource));
9250 memset(&sample_res, 0, sizeof(struct mlx5_flow_dv_sample_resource));
9251 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
9252 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9253 /* update normal path action resource into last index of array */
9254 sample_act = &mdest_res.sample_act[MLX5_MAX_DEST_NUM - 1];
9255 tunnel = is_flow_tunnel_match_rule(dev, attr, items, actions) ?
9256 flow_items_to_tunnel(items) :
9257 is_flow_tunnel_steer_rule(dev, attr, items, actions) ?
9258 flow_actions_to_tunnel(actions) :
9259 dev_flow->tunnel ? dev_flow->tunnel : NULL;
9260 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
9261 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9262 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
9263 (dev, tunnel, attr, items, actions);
9264 ret = mlx5_flow_group_to_table(dev, tunnel, attr->group, &table,
9268 dev_flow->dv.group = table;
9270 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
9271 if (priority == MLX5_FLOW_PRIO_RSVD)
9272 priority = dev_conf->flow_prio - 1;
9273 /* number of actions must be set to 0 in case of dirty stack. */
9274 mhdr_res->actions_num = 0;
9275 if (is_flow_tunnel_match_rule(dev, attr, items, actions)) {
9277 * do not add decap action if match rule drops packet
9278 * HW rejects rules with decap & drop
9280 bool add_decap = true;
9281 const struct rte_flow_action *ptr = actions;
9282 struct mlx5_flow_tbl_resource *tbl;
9284 for (; ptr->type != RTE_FLOW_ACTION_TYPE_END; ptr++) {
9285 if (ptr->type == RTE_FLOW_ACTION_TYPE_DROP) {
9291 if (flow_dv_create_action_l2_decap(dev, dev_flow,
9295 dev_flow->dv.actions[actions_n++] =
9296 dev_flow->dv.encap_decap->action;
9297 action_flags |= MLX5_FLOW_ACTION_DECAP;
9300 * bind table_id with <group, table> for tunnel match rule.
9301 * Tunnel set rule establishes that bind in JUMP action handler.
9302 * Required for scenario when application creates tunnel match
9303 * rule before tunnel set rule.
9305 tbl = flow_dv_tbl_resource_get(dev, table, attr->egress,
9307 !!dev_flow->external, tunnel,
9308 attr->group, 0, error);
9310 return rte_flow_error_set
9311 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
9312 actions, "cannot register tunnel group");
9314 for (; !actions_end ; actions++) {
9315 const struct rte_flow_action_queue *queue;
9316 const struct rte_flow_action_rss *rss;
9317 const struct rte_flow_action *action = actions;
9318 const uint8_t *rss_key;
9319 const struct rte_flow_action_meter *mtr;
9320 struct mlx5_flow_tbl_resource *tbl;
9321 uint32_t port_id = 0;
9322 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
9323 int action_type = actions->type;
9324 const struct rte_flow_action *found_action = NULL;
9325 struct mlx5_flow_meter *fm = NULL;
9326 uint32_t jump_group = 0;
9328 if (!mlx5_flow_os_action_supported(action_type))
9329 return rte_flow_error_set(error, ENOTSUP,
9330 RTE_FLOW_ERROR_TYPE_ACTION,
9332 "action not supported");
9333 switch (action_type) {
9334 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
9335 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
9337 case RTE_FLOW_ACTION_TYPE_VOID:
9339 case RTE_FLOW_ACTION_TYPE_PORT_ID:
9340 if (flow_dv_translate_action_port_id(dev, action,
9343 port_id_resource.port_id = port_id;
9344 MLX5_ASSERT(!handle->rix_port_id_action);
9345 if (flow_dv_port_id_action_resource_register
9346 (dev, &port_id_resource, dev_flow, error))
9348 dev_flow->dv.actions[actions_n++] =
9349 dev_flow->dv.port_id_action->action;
9350 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9351 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
9352 sample_act->action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9355 case RTE_FLOW_ACTION_TYPE_FLAG:
9356 action_flags |= MLX5_FLOW_ACTION_FLAG;
9357 dev_flow->handle->mark = 1;
9358 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
9359 struct rte_flow_action_mark mark = {
9360 .id = MLX5_FLOW_MARK_DEFAULT,
9363 if (flow_dv_convert_action_mark(dev, &mark,
9367 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
9370 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
9372 * Only one FLAG or MARK is supported per device flow
9373 * right now. So the pointer to the tag resource must be
9374 * zero before the register process.
9376 MLX5_ASSERT(!handle->dvh.rix_tag);
9377 if (flow_dv_tag_resource_register(dev, tag_be,
9380 MLX5_ASSERT(dev_flow->dv.tag_resource);
9381 dev_flow->dv.actions[actions_n++] =
9382 dev_flow->dv.tag_resource->action;
9384 case RTE_FLOW_ACTION_TYPE_MARK:
9385 action_flags |= MLX5_FLOW_ACTION_MARK;
9386 dev_flow->handle->mark = 1;
9387 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
9388 const struct rte_flow_action_mark *mark =
9389 (const struct rte_flow_action_mark *)
9392 if (flow_dv_convert_action_mark(dev, mark,
9396 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
9400 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
9401 /* Legacy (non-extensive) MARK action. */
9402 tag_be = mlx5_flow_mark_set
9403 (((const struct rte_flow_action_mark *)
9404 (actions->conf))->id);
9405 MLX5_ASSERT(!handle->dvh.rix_tag);
9406 if (flow_dv_tag_resource_register(dev, tag_be,
9409 MLX5_ASSERT(dev_flow->dv.tag_resource);
9410 dev_flow->dv.actions[actions_n++] =
9411 dev_flow->dv.tag_resource->action;
9413 case RTE_FLOW_ACTION_TYPE_SET_META:
9414 if (flow_dv_convert_action_set_meta
9415 (dev, mhdr_res, attr,
9416 (const struct rte_flow_action_set_meta *)
9417 actions->conf, error))
9419 action_flags |= MLX5_FLOW_ACTION_SET_META;
9421 case RTE_FLOW_ACTION_TYPE_SET_TAG:
9422 if (flow_dv_convert_action_set_tag
9424 (const struct rte_flow_action_set_tag *)
9425 actions->conf, error))
9427 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9429 case RTE_FLOW_ACTION_TYPE_DROP:
9430 action_flags |= MLX5_FLOW_ACTION_DROP;
9431 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
9433 case RTE_FLOW_ACTION_TYPE_QUEUE:
9434 queue = actions->conf;
9435 rss_desc->queue_num = 1;
9436 rss_desc->queue[0] = queue->index;
9437 action_flags |= MLX5_FLOW_ACTION_QUEUE;
9438 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9439 sample_act->action_flags |= MLX5_FLOW_ACTION_QUEUE;
9442 case RTE_FLOW_ACTION_TYPE_RSS:
9443 rss = actions->conf;
9444 memcpy(rss_desc->queue, rss->queue,
9445 rss->queue_num * sizeof(uint16_t));
9446 rss_desc->queue_num = rss->queue_num;
9447 /* NULL RSS key indicates default RSS key. */
9448 rss_key = !rss->key ? rss_hash_default_key : rss->key;
9449 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
9451 * rss->level and rss.types should be set in advance
9452 * when expanding items for RSS.
9454 action_flags |= MLX5_FLOW_ACTION_RSS;
9455 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9457 case RTE_FLOW_ACTION_TYPE_AGE:
9458 case RTE_FLOW_ACTION_TYPE_COUNT:
9459 if (!dev_conf->devx) {
9460 return rte_flow_error_set
9462 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9464 "count action not supported");
9466 /* Save information first, will apply later. */
9467 if (actions->type == RTE_FLOW_ACTION_TYPE_COUNT)
9468 count = action->conf;
9471 action_flags |= MLX5_FLOW_ACTION_COUNT;
9473 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
9474 dev_flow->dv.actions[actions_n++] =
9475 priv->sh->pop_vlan_action;
9476 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
9478 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
9479 if (!(action_flags &
9480 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
9481 flow_dev_get_vlan_info_from_items(items, &vlan);
9482 vlan.eth_proto = rte_be_to_cpu_16
9483 ((((const struct rte_flow_action_of_push_vlan *)
9484 actions->conf)->ethertype));
9485 found_action = mlx5_flow_find_action
9487 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
9489 mlx5_update_vlan_vid_pcp(found_action, &vlan);
9490 found_action = mlx5_flow_find_action
9492 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
9494 mlx5_update_vlan_vid_pcp(found_action, &vlan);
9495 if (flow_dv_create_action_push_vlan
9496 (dev, attr, &vlan, dev_flow, error))
9498 dev_flow->dv.actions[actions_n++] =
9499 dev_flow->dv.push_vlan_res->action;
9500 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
9502 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
9503 /* of_vlan_push action handled this action */
9504 MLX5_ASSERT(action_flags &
9505 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
9507 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
9508 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
9510 flow_dev_get_vlan_info_from_items(items, &vlan);
9511 mlx5_update_vlan_vid_pcp(actions, &vlan);
9512 /* If no VLAN push - this is a modify header action */
9513 if (flow_dv_convert_action_modify_vlan_vid
9514 (mhdr_res, actions, error))
9516 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
9518 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
9519 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
9520 if (flow_dv_create_action_l2_encap(dev, actions,
9525 dev_flow->dv.actions[actions_n++] =
9526 dev_flow->dv.encap_decap->action;
9527 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9528 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
9529 sample_act->action_flags |=
9530 MLX5_FLOW_ACTION_ENCAP;
9532 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
9533 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
9534 if (flow_dv_create_action_l2_decap(dev, dev_flow,
9538 dev_flow->dv.actions[actions_n++] =
9539 dev_flow->dv.encap_decap->action;
9540 action_flags |= MLX5_FLOW_ACTION_DECAP;
9542 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
9543 /* Handle encap with preceding decap. */
9544 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
9545 if (flow_dv_create_action_raw_encap
9546 (dev, actions, dev_flow, attr, error))
9548 dev_flow->dv.actions[actions_n++] =
9549 dev_flow->dv.encap_decap->action;
9551 /* Handle encap without preceding decap. */
9552 if (flow_dv_create_action_l2_encap
9553 (dev, actions, dev_flow, attr->transfer,
9556 dev_flow->dv.actions[actions_n++] =
9557 dev_flow->dv.encap_decap->action;
9559 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9560 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
9561 sample_act->action_flags |=
9562 MLX5_FLOW_ACTION_ENCAP;
9564 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
9565 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
9567 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
9568 if (flow_dv_create_action_l2_decap
9569 (dev, dev_flow, attr->transfer, error))
9571 dev_flow->dv.actions[actions_n++] =
9572 dev_flow->dv.encap_decap->action;
9574 /* If decap is followed by encap, handle it at encap. */
9575 action_flags |= MLX5_FLOW_ACTION_DECAP;
9577 case RTE_FLOW_ACTION_TYPE_JUMP:
9578 jump_group = ((const struct rte_flow_action_jump *)
9579 action->conf)->group;
9580 grp_info.std_tbl_fix = 0;
9581 ret = mlx5_flow_group_to_table(dev, tunnel,
9587 tbl = flow_dv_tbl_resource_get(dev, table, attr->egress,
9589 !!dev_flow->external,
9590 tunnel, jump_group, 0,
9593 return rte_flow_error_set
9595 RTE_FLOW_ERROR_TYPE_ACTION,
9597 "cannot create jump action.");
9598 if (flow_dv_jump_tbl_resource_register
9599 (dev, tbl, dev_flow, error)) {
9600 flow_dv_tbl_resource_release(dev, tbl);
9601 return rte_flow_error_set
9603 RTE_FLOW_ERROR_TYPE_ACTION,
9605 "cannot create jump action.");
9607 dev_flow->dv.actions[actions_n++] =
9608 dev_flow->dv.jump->action;
9609 action_flags |= MLX5_FLOW_ACTION_JUMP;
9610 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
9612 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
9613 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
9614 if (flow_dv_convert_action_modify_mac
9615 (mhdr_res, actions, error))
9617 action_flags |= actions->type ==
9618 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
9619 MLX5_FLOW_ACTION_SET_MAC_SRC :
9620 MLX5_FLOW_ACTION_SET_MAC_DST;
9622 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
9623 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
9624 if (flow_dv_convert_action_modify_ipv4
9625 (mhdr_res, actions, error))
9627 action_flags |= actions->type ==
9628 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
9629 MLX5_FLOW_ACTION_SET_IPV4_SRC :
9630 MLX5_FLOW_ACTION_SET_IPV4_DST;
9632 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
9633 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
9634 if (flow_dv_convert_action_modify_ipv6
9635 (mhdr_res, actions, error))
9637 action_flags |= actions->type ==
9638 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
9639 MLX5_FLOW_ACTION_SET_IPV6_SRC :
9640 MLX5_FLOW_ACTION_SET_IPV6_DST;
9642 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
9643 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
9644 if (flow_dv_convert_action_modify_tp
9645 (mhdr_res, actions, items,
9646 &flow_attr, dev_flow, !!(action_flags &
9647 MLX5_FLOW_ACTION_DECAP), error))
9649 action_flags |= actions->type ==
9650 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
9651 MLX5_FLOW_ACTION_SET_TP_SRC :
9652 MLX5_FLOW_ACTION_SET_TP_DST;
9654 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
9655 if (flow_dv_convert_action_modify_dec_ttl
9656 (mhdr_res, items, &flow_attr, dev_flow,
9658 MLX5_FLOW_ACTION_DECAP), error))
9660 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
9662 case RTE_FLOW_ACTION_TYPE_SET_TTL:
9663 if (flow_dv_convert_action_modify_ttl
9664 (mhdr_res, actions, items, &flow_attr,
9665 dev_flow, !!(action_flags &
9666 MLX5_FLOW_ACTION_DECAP), error))
9668 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
9670 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
9671 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
9672 if (flow_dv_convert_action_modify_tcp_seq
9673 (mhdr_res, actions, error))
9675 action_flags |= actions->type ==
9676 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
9677 MLX5_FLOW_ACTION_INC_TCP_SEQ :
9678 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
9681 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
9682 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
9683 if (flow_dv_convert_action_modify_tcp_ack
9684 (mhdr_res, actions, error))
9686 action_flags |= actions->type ==
9687 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
9688 MLX5_FLOW_ACTION_INC_TCP_ACK :
9689 MLX5_FLOW_ACTION_DEC_TCP_ACK;
9691 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
9692 if (flow_dv_convert_action_set_reg
9693 (mhdr_res, actions, error))
9695 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9697 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
9698 if (flow_dv_convert_action_copy_mreg
9699 (dev, mhdr_res, actions, error))
9701 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9703 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
9704 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
9705 dev_flow->handle->fate_action =
9706 MLX5_FLOW_FATE_DEFAULT_MISS;
9708 case RTE_FLOW_ACTION_TYPE_METER:
9709 mtr = actions->conf;
9711 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
9714 return rte_flow_error_set(error,
9716 RTE_FLOW_ERROR_TYPE_ACTION,
9719 "or invalid parameters");
9720 flow->meter = fm->idx;
9722 /* Set the meter action. */
9724 fm = mlx5_ipool_get(priv->sh->ipool
9725 [MLX5_IPOOL_MTR], flow->meter);
9727 return rte_flow_error_set(error,
9729 RTE_FLOW_ERROR_TYPE_ACTION,
9732 "or invalid parameters");
9734 dev_flow->dv.actions[actions_n++] =
9735 fm->mfts->meter_action;
9736 action_flags |= MLX5_FLOW_ACTION_METER;
9738 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
9739 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
9742 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
9744 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
9745 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
9748 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
9750 case RTE_FLOW_ACTION_TYPE_SAMPLE:
9751 sample_act_pos = actions_n;
9752 ret = flow_dv_translate_action_sample(dev,
9762 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
9763 /* put encap action into group if work with port id */
9764 if ((action_flags & MLX5_FLOW_ACTION_ENCAP) &&
9765 (action_flags & MLX5_FLOW_ACTION_PORT_ID))
9766 sample_act->action_flags |=
9767 MLX5_FLOW_ACTION_ENCAP;
9769 case RTE_FLOW_ACTION_TYPE_END:
9771 if (mhdr_res->actions_num) {
9772 /* create modify action if needed. */
9773 if (flow_dv_modify_hdr_resource_register
9774 (dev, mhdr_res, dev_flow, error))
9776 dev_flow->dv.actions[modify_action_position] =
9777 handle->dvh.modify_hdr->action;
9779 if (action_flags & MLX5_FLOW_ACTION_COUNT) {
9781 flow_dv_translate_create_counter(dev,
9782 dev_flow, count, age);
9785 return rte_flow_error_set
9787 RTE_FLOW_ERROR_TYPE_ACTION,
9789 "cannot create counter"
9791 dev_flow->dv.actions[actions_n] =
9792 (flow_dv_counter_get_by_idx(dev,
9793 flow->counter, NULL))->action;
9796 if (action_flags & MLX5_FLOW_ACTION_SAMPLE) {
9797 ret = flow_dv_create_action_sample(dev,
9806 return rte_flow_error_set
9808 RTE_FLOW_ERROR_TYPE_ACTION,
9810 "cannot create sample action");
9811 if (num_of_dest > 1) {
9812 dev_flow->dv.actions[sample_act_pos] =
9813 dev_flow->dv.dest_array_res->action;
9815 dev_flow->dv.actions[sample_act_pos] =
9816 dev_flow->dv.sample_res->verbs_action;
9823 if (mhdr_res->actions_num &&
9824 modify_action_position == UINT32_MAX)
9825 modify_action_position = actions_n++;
9828 * For multiple destination (sample action with ratio=1), the encap
9829 * action and port id action will be combined into group action.
9830 * So need remove the original these actions in the flow and only
9831 * use the sample action instead of.
9833 if (num_of_dest > 1 && sample_act->dr_port_id_action) {
9835 void *temp_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9837 for (i = 0; i < actions_n; i++) {
9838 if ((sample_act->dr_encap_action &&
9839 sample_act->dr_encap_action ==
9840 dev_flow->dv.actions[i]) ||
9841 (sample_act->dr_port_id_action &&
9842 sample_act->dr_port_id_action ==
9843 dev_flow->dv.actions[i]))
9845 temp_actions[tmp_actions_n++] = dev_flow->dv.actions[i];
9847 memcpy((void *)dev_flow->dv.actions,
9848 (void *)temp_actions,
9849 tmp_actions_n * sizeof(void *));
9850 actions_n = tmp_actions_n;
9852 dev_flow->dv.actions_n = actions_n;
9853 dev_flow->act_flags = action_flags;
9854 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
9855 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
9856 int item_type = items->type;
9858 if (!mlx5_flow_os_item_supported(item_type))
9859 return rte_flow_error_set(error, ENOTSUP,
9860 RTE_FLOW_ERROR_TYPE_ITEM,
9861 NULL, "item not supported");
9862 switch (item_type) {
9863 case RTE_FLOW_ITEM_TYPE_PORT_ID:
9864 flow_dv_translate_item_port_id(dev, match_mask,
9865 match_value, items);
9866 last_item = MLX5_FLOW_ITEM_PORT_ID;
9868 case RTE_FLOW_ITEM_TYPE_ETH:
9869 flow_dv_translate_item_eth(match_mask, match_value,
9871 dev_flow->dv.group);
9872 matcher.priority = action_flags &
9873 MLX5_FLOW_ACTION_DEFAULT_MISS &&
9874 !dev_flow->external ?
9875 MLX5_PRIORITY_MAP_L3 :
9876 MLX5_PRIORITY_MAP_L2;
9877 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
9878 MLX5_FLOW_LAYER_OUTER_L2;
9880 case RTE_FLOW_ITEM_TYPE_VLAN:
9881 flow_dv_translate_item_vlan(dev_flow,
9882 match_mask, match_value,
9884 dev_flow->dv.group);
9885 matcher.priority = MLX5_PRIORITY_MAP_L2;
9886 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
9887 MLX5_FLOW_LAYER_INNER_VLAN) :
9888 (MLX5_FLOW_LAYER_OUTER_L2 |
9889 MLX5_FLOW_LAYER_OUTER_VLAN);
9891 case RTE_FLOW_ITEM_TYPE_IPV4:
9892 mlx5_flow_tunnel_ip_check(items, next_protocol,
9893 &item_flags, &tunnel);
9894 flow_dv_translate_item_ipv4(match_mask, match_value,
9896 dev_flow->dv.group);
9897 matcher.priority = MLX5_PRIORITY_MAP_L3;
9898 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
9899 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
9900 if (items->mask != NULL &&
9901 ((const struct rte_flow_item_ipv4 *)
9902 items->mask)->hdr.next_proto_id) {
9904 ((const struct rte_flow_item_ipv4 *)
9905 (items->spec))->hdr.next_proto_id;
9907 ((const struct rte_flow_item_ipv4 *)
9908 (items->mask))->hdr.next_proto_id;
9910 /* Reset for inner layer. */
9911 next_protocol = 0xff;
9914 case RTE_FLOW_ITEM_TYPE_IPV6:
9915 mlx5_flow_tunnel_ip_check(items, next_protocol,
9916 &item_flags, &tunnel);
9917 flow_dv_translate_item_ipv6(match_mask, match_value,
9919 dev_flow->dv.group);
9920 matcher.priority = MLX5_PRIORITY_MAP_L3;
9921 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
9922 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
9923 if (items->mask != NULL &&
9924 ((const struct rte_flow_item_ipv6 *)
9925 items->mask)->hdr.proto) {
9927 ((const struct rte_flow_item_ipv6 *)
9928 items->spec)->hdr.proto;
9930 ((const struct rte_flow_item_ipv6 *)
9931 items->mask)->hdr.proto;
9933 /* Reset for inner layer. */
9934 next_protocol = 0xff;
9937 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
9938 flow_dv_translate_item_ipv6_frag_ext(match_mask,
9941 last_item = tunnel ?
9942 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
9943 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
9944 if (items->mask != NULL &&
9945 ((const struct rte_flow_item_ipv6_frag_ext *)
9946 items->mask)->hdr.next_header) {
9948 ((const struct rte_flow_item_ipv6_frag_ext *)
9949 items->spec)->hdr.next_header;
9951 ((const struct rte_flow_item_ipv6_frag_ext *)
9952 items->mask)->hdr.next_header;
9954 /* Reset for inner layer. */
9955 next_protocol = 0xff;
9958 case RTE_FLOW_ITEM_TYPE_TCP:
9959 flow_dv_translate_item_tcp(match_mask, match_value,
9961 matcher.priority = MLX5_PRIORITY_MAP_L4;
9962 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
9963 MLX5_FLOW_LAYER_OUTER_L4_TCP;
9965 case RTE_FLOW_ITEM_TYPE_UDP:
9966 flow_dv_translate_item_udp(match_mask, match_value,
9968 matcher.priority = MLX5_PRIORITY_MAP_L4;
9969 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
9970 MLX5_FLOW_LAYER_OUTER_L4_UDP;
9972 case RTE_FLOW_ITEM_TYPE_GRE:
9973 flow_dv_translate_item_gre(match_mask, match_value,
9975 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9976 last_item = MLX5_FLOW_LAYER_GRE;
9978 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
9979 flow_dv_translate_item_gre_key(match_mask,
9980 match_value, items);
9981 last_item = MLX5_FLOW_LAYER_GRE_KEY;
9983 case RTE_FLOW_ITEM_TYPE_NVGRE:
9984 flow_dv_translate_item_nvgre(match_mask, match_value,
9986 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9987 last_item = MLX5_FLOW_LAYER_GRE;
9989 case RTE_FLOW_ITEM_TYPE_VXLAN:
9990 flow_dv_translate_item_vxlan(match_mask, match_value,
9992 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9993 last_item = MLX5_FLOW_LAYER_VXLAN;
9995 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
9996 flow_dv_translate_item_vxlan_gpe(match_mask,
9999 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10000 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
10002 case RTE_FLOW_ITEM_TYPE_GENEVE:
10003 flow_dv_translate_item_geneve(match_mask, match_value,
10005 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10006 last_item = MLX5_FLOW_LAYER_GENEVE;
10008 case RTE_FLOW_ITEM_TYPE_MPLS:
10009 flow_dv_translate_item_mpls(match_mask, match_value,
10010 items, last_item, tunnel);
10011 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10012 last_item = MLX5_FLOW_LAYER_MPLS;
10014 case RTE_FLOW_ITEM_TYPE_MARK:
10015 flow_dv_translate_item_mark(dev, match_mask,
10016 match_value, items);
10017 last_item = MLX5_FLOW_ITEM_MARK;
10019 case RTE_FLOW_ITEM_TYPE_META:
10020 flow_dv_translate_item_meta(dev, match_mask,
10021 match_value, attr, items);
10022 last_item = MLX5_FLOW_ITEM_METADATA;
10024 case RTE_FLOW_ITEM_TYPE_ICMP:
10025 flow_dv_translate_item_icmp(match_mask, match_value,
10027 last_item = MLX5_FLOW_LAYER_ICMP;
10029 case RTE_FLOW_ITEM_TYPE_ICMP6:
10030 flow_dv_translate_item_icmp6(match_mask, match_value,
10032 last_item = MLX5_FLOW_LAYER_ICMP6;
10034 case RTE_FLOW_ITEM_TYPE_TAG:
10035 flow_dv_translate_item_tag(dev, match_mask,
10036 match_value, items);
10037 last_item = MLX5_FLOW_ITEM_TAG;
10039 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
10040 flow_dv_translate_mlx5_item_tag(dev, match_mask,
10041 match_value, items);
10042 last_item = MLX5_FLOW_ITEM_TAG;
10044 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
10045 flow_dv_translate_item_tx_queue(dev, match_mask,
10048 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
10050 case RTE_FLOW_ITEM_TYPE_GTP:
10051 flow_dv_translate_item_gtp(match_mask, match_value,
10053 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10054 last_item = MLX5_FLOW_LAYER_GTP;
10056 case RTE_FLOW_ITEM_TYPE_ECPRI:
10057 if (!mlx5_flex_parser_ecpri_exist(dev)) {
10058 /* Create it only the first time to be used. */
10059 ret = mlx5_flex_parser_ecpri_alloc(dev);
10061 return rte_flow_error_set
10063 RTE_FLOW_ERROR_TYPE_ITEM,
10065 "cannot create eCPRI parser");
10067 /* Adjust the length matcher and device flow value. */
10068 matcher.mask.size = MLX5_ST_SZ_BYTES(fte_match_param);
10069 dev_flow->dv.value.size =
10070 MLX5_ST_SZ_BYTES(fte_match_param);
10071 flow_dv_translate_item_ecpri(dev, match_mask,
10072 match_value, items);
10073 /* No other protocol should follow eCPRI layer. */
10074 last_item = MLX5_FLOW_LAYER_ECPRI;
10079 item_flags |= last_item;
10082 * When E-Switch mode is enabled, we have two cases where we need to
10083 * set the source port manually.
10084 * The first one, is in case of Nic steering rule, and the second is
10085 * E-Switch rule where no port_id item was found. In both cases
10086 * the source port is set according the current port in use.
10088 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
10089 (priv->representor || priv->master)) {
10090 if (flow_dv_translate_item_port_id(dev, match_mask,
10091 match_value, NULL))
10094 #ifdef RTE_LIBRTE_MLX5_DEBUG
10095 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
10096 dev_flow->dv.value.buf));
10099 * Layers may be already initialized from prefix flow if this dev_flow
10100 * is the suffix flow.
10102 handle->layers |= item_flags;
10103 if (action_flags & MLX5_FLOW_ACTION_RSS)
10104 flow_dv_hashfields_set(dev_flow, rss_desc);
10105 /* Register matcher. */
10106 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
10107 matcher.mask.size);
10108 matcher.priority = mlx5_flow_adjust_priority(dev, priority,
10110 /* reserved field no needs to be set to 0 here. */
10111 tbl_key.domain = attr->transfer;
10112 tbl_key.direction = attr->egress;
10113 tbl_key.table_id = dev_flow->dv.group;
10114 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
10120 * Set hash RX queue by hash fields (see enum ibv_rx_hash_fields)
10123 * @param[in, out] action
10124 * Shred RSS action holding hash RX queue objects.
10125 * @param[in] hash_fields
10126 * Defines combination of packet fields to participate in RX hash.
10127 * @param[in] tunnel
10129 * @param[in] hrxq_idx
10130 * Hash RX queue index to set.
10133 * 0 on success, otherwise negative errno value.
10136 __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
10137 const uint64_t hash_fields,
10141 uint32_t *hrxqs = tunnel ? action->hrxq : action->hrxq_tunnel;
10143 switch (hash_fields & ~IBV_RX_HASH_INNER) {
10144 case MLX5_RSS_HASH_IPV4:
10145 hrxqs[0] = hrxq_idx;
10147 case MLX5_RSS_HASH_IPV4_TCP:
10148 hrxqs[1] = hrxq_idx;
10150 case MLX5_RSS_HASH_IPV4_UDP:
10151 hrxqs[2] = hrxq_idx;
10153 case MLX5_RSS_HASH_IPV6:
10154 hrxqs[3] = hrxq_idx;
10156 case MLX5_RSS_HASH_IPV6_TCP:
10157 hrxqs[4] = hrxq_idx;
10159 case MLX5_RSS_HASH_IPV6_UDP:
10160 hrxqs[5] = hrxq_idx;
10162 case MLX5_RSS_HASH_NONE:
10163 hrxqs[6] = hrxq_idx;
10171 * Look up for hash RX queue by hash fields (see enum ibv_rx_hash_fields)
10174 * @param[in] action
10175 * Shred RSS action holding hash RX queue objects.
10176 * @param[in] hash_fields
10177 * Defines combination of packet fields to participate in RX hash.
10178 * @param[in] tunnel
10182 * Valid hash RX queue index, otherwise 0.
10185 __flow_dv_action_rss_hrxq_lookup(const struct mlx5_shared_action_rss *action,
10186 const uint64_t hash_fields,
10189 const uint32_t *hrxqs = tunnel ? action->hrxq : action->hrxq_tunnel;
10191 switch (hash_fields & ~IBV_RX_HASH_INNER) {
10192 case MLX5_RSS_HASH_IPV4:
10194 case MLX5_RSS_HASH_IPV4_TCP:
10196 case MLX5_RSS_HASH_IPV4_UDP:
10198 case MLX5_RSS_HASH_IPV6:
10200 case MLX5_RSS_HASH_IPV6_TCP:
10202 case MLX5_RSS_HASH_IPV6_UDP:
10204 case MLX5_RSS_HASH_NONE:
10212 * Retrieves hash RX queue suitable for the *flow*.
10213 * If shared action configured for *flow* suitable hash RX queue will be
10214 * retrieved from attached shared action.
10217 * Shred RSS action holding hash RX queue objects.
10218 * @param[in] dev_flow
10219 * Pointer to the sub flow.
10221 * Pointer to retrieved hash RX queue object.
10224 * Valid hash RX queue index, otherwise 0 and rte_errno is set.
10227 __flow_dv_rss_get_hrxq(struct rte_eth_dev *dev, struct rte_flow *flow,
10228 struct mlx5_flow *dev_flow,
10229 struct mlx5_hrxq **hrxq)
10231 struct mlx5_priv *priv = dev->data->dev_private;
10232 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
10235 if (flow->shared_rss) {
10236 hrxq_idx = __flow_dv_action_rss_hrxq_lookup
10237 (flow->shared_rss, dev_flow->hash_fields,
10238 !!(dev_flow->handle->layers &
10239 MLX5_FLOW_LAYER_TUNNEL));
10241 *hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
10243 __atomic_fetch_add(&(*hrxq)->refcnt, 1,
10247 struct mlx5_flow_rss_desc *rss_desc =
10248 &wks->rss_desc[!!wks->flow_nested_idx];
10250 MLX5_ASSERT(rss_desc->queue_num);
10251 hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
10252 MLX5_RSS_HASH_KEY_LEN,
10253 dev_flow->hash_fields,
10254 rss_desc->queue, rss_desc->queue_num);
10256 hrxq_idx = mlx5_hrxq_new(dev,
10258 MLX5_RSS_HASH_KEY_LEN,
10259 dev_flow->hash_fields,
10261 rss_desc->queue_num,
10262 !!(dev_flow->handle->layers &
10263 MLX5_FLOW_LAYER_TUNNEL),
10266 *hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
10273 * Apply the flow to the NIC, lock free,
10274 * (mutex should be acquired by caller).
10277 * Pointer to the Ethernet device structure.
10278 * @param[in, out] flow
10279 * Pointer to flow structure.
10280 * @param[out] error
10281 * Pointer to error structure.
10284 * 0 on success, a negative errno value otherwise and rte_errno is set.
10287 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
10288 struct rte_flow_error *error)
10290 struct mlx5_flow_dv_workspace *dv;
10291 struct mlx5_flow_handle *dh;
10292 struct mlx5_flow_handle_dv *dv_h;
10293 struct mlx5_flow *dev_flow;
10294 struct mlx5_priv *priv = dev->data->dev_private;
10295 uint32_t handle_idx;
10299 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
10302 for (idx = wks->flow_idx - 1; idx >= wks->flow_nested_idx; idx--) {
10303 dev_flow = &wks->flows[idx];
10304 dv = &dev_flow->dv;
10305 dh = dev_flow->handle;
10308 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
10309 if (dv->transfer) {
10310 dv->actions[n++] = priv->sh->esw_drop_action;
10312 MLX5_ASSERT(priv->drop_queue.hrxq);
10314 priv->drop_queue.hrxq->action;
10316 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
10317 !dv_h->rix_sample && !dv_h->rix_dest_array) {
10318 struct mlx5_hrxq *hrxq = NULL;
10319 uint32_t hrxq_idx = __flow_dv_rss_get_hrxq
10320 (dev, flow, dev_flow, &hrxq);
10325 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10326 "cannot get hash queue");
10329 dh->rix_hrxq = hrxq_idx;
10330 dv->actions[n++] = hrxq->action;
10331 } else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
10332 if (!priv->sh->default_miss_action) {
10335 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10336 "default miss action not be created.");
10339 dv->actions[n++] = priv->sh->default_miss_action;
10341 err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
10342 (void *)&dv->value, n,
10343 dv->actions, &dh->drv_flow);
10345 rte_flow_error_set(error, errno,
10346 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10348 "hardware refuses to create flow");
10351 if (priv->vmwa_context &&
10352 dh->vf_vlan.tag && !dh->vf_vlan.created) {
10354 * The rule contains the VLAN pattern.
10355 * For VF we are going to create VLAN
10356 * interface to make hypervisor set correct
10357 * e-Switch vport context.
10359 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
10364 err = rte_errno; /* Save rte_errno before cleanup. */
10365 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
10366 handle_idx, dh, next) {
10367 /* hrxq is union, don't clear it if the flag is not set. */
10368 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE && dh->rix_hrxq) {
10369 mlx5_hrxq_release(dev, dh->rix_hrxq);
10372 if (dh->vf_vlan.tag && dh->vf_vlan.created)
10373 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
10375 rte_errno = err; /* Restore rte_errno. */
10380 * Release the flow matcher.
10383 * Pointer to Ethernet device.
10385 * Pointer to mlx5_flow_handle.
10388 * 1 while a reference on it exists, 0 when freed.
10391 flow_dv_matcher_release(struct rte_eth_dev *dev,
10392 struct mlx5_flow_handle *handle)
10394 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
10396 MLX5_ASSERT(matcher->matcher_object);
10397 DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
10398 dev->data->port_id, (void *)matcher,
10399 __atomic_load_n(&matcher->refcnt, __ATOMIC_RELAXED));
10400 if (__atomic_sub_fetch(&matcher->refcnt, 1, __ATOMIC_RELAXED) == 0) {
10401 claim_zero(mlx5_flow_os_destroy_flow_matcher
10402 (matcher->matcher_object));
10403 LIST_REMOVE(matcher, next);
10404 /* table ref-- in release interface. */
10405 flow_dv_tbl_resource_release(dev, matcher->tbl);
10406 mlx5_free(matcher);
10407 DRV_LOG(DEBUG, "port %u matcher %p: removed",
10408 dev->data->port_id, (void *)matcher);
10415 * Release an encap/decap resource.
10418 * Pointer to Ethernet device.
10419 * @param encap_decap_idx
10420 * Index of encap decap resource.
10423 * 1 while a reference on it exists, 0 when freed.
10426 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
10427 uint32_t encap_decap_idx)
10429 struct mlx5_priv *priv = dev->data->dev_private;
10430 uint32_t idx = encap_decap_idx;
10431 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
10433 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
10435 if (!cache_resource)
10437 MLX5_ASSERT(cache_resource->action);
10438 DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
10439 (void *)cache_resource,
10440 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10441 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10442 __ATOMIC_RELAXED) == 0) {
10443 claim_zero(mlx5_flow_os_destroy_flow_action
10444 (cache_resource->action));
10445 mlx5_hlist_remove(priv->sh->encaps_decaps,
10446 &cache_resource->entry);
10447 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
10448 DRV_LOG(DEBUG, "encap/decap resource %p: removed",
10449 (void *)cache_resource);
10456 * Release an jump to table action resource.
10459 * Pointer to Ethernet device.
10461 * Pointer to mlx5_flow_handle.
10464 * 1 while a reference on it exists, 0 when freed.
10467 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
10468 struct mlx5_flow_handle *handle)
10470 struct mlx5_priv *priv = dev->data->dev_private;
10471 struct mlx5_flow_tbl_data_entry *tbl_data;
10473 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
10477 return flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
10481 flow_dv_modify_remove_cb(struct mlx5_hlist *list __rte_unused,
10482 struct mlx5_hlist_entry *entry)
10484 struct mlx5_flow_dv_modify_hdr_resource *res =
10485 container_of(entry, typeof(*res), entry);
10487 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
10492 * Release a modify-header resource.
10495 * Pointer to Ethernet device.
10497 * Pointer to mlx5_flow_handle.
10500 * 1 while a reference on it exists, 0 when freed.
10503 flow_dv_modify_hdr_resource_release(struct rte_eth_dev *dev,
10504 struct mlx5_flow_handle *handle)
10506 struct mlx5_priv *priv = dev->data->dev_private;
10507 struct mlx5_flow_dv_modify_hdr_resource *entry = handle->dvh.modify_hdr;
10509 MLX5_ASSERT(entry->action);
10510 return mlx5_hlist_unregister(priv->sh->modify_cmds, &entry->entry);
10514 * Release port ID action resource.
10517 * Pointer to Ethernet device.
10519 * Pointer to mlx5_flow_handle.
10522 * 1 while a reference on it exists, 0 when freed.
10525 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
10528 struct mlx5_priv *priv = dev->data->dev_private;
10529 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
10530 uint32_t idx = port_id;
10532 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
10534 if (!cache_resource)
10536 MLX5_ASSERT(cache_resource->action);
10537 DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
10538 (void *)cache_resource,
10539 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10540 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10541 __ATOMIC_RELAXED) == 0) {
10542 claim_zero(mlx5_flow_os_destroy_flow_action
10543 (cache_resource->action));
10544 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
10545 &priv->sh->port_id_action_list, idx,
10546 cache_resource, next);
10547 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
10548 DRV_LOG(DEBUG, "port id action resource %p: removed",
10549 (void *)cache_resource);
10556 * Release push vlan action resource.
10559 * Pointer to Ethernet device.
10561 * Pointer to mlx5_flow_handle.
10564 * 1 while a reference on it exists, 0 when freed.
10567 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
10568 struct mlx5_flow_handle *handle)
10570 struct mlx5_priv *priv = dev->data->dev_private;
10571 uint32_t idx = handle->dvh.rix_push_vlan;
10572 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
10574 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
10576 if (!cache_resource)
10578 MLX5_ASSERT(cache_resource->action);
10579 DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
10580 (void *)cache_resource,
10581 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10582 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10583 __ATOMIC_RELAXED) == 0) {
10584 claim_zero(mlx5_flow_os_destroy_flow_action
10585 (cache_resource->action));
10586 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
10587 &priv->sh->push_vlan_action_list, idx,
10588 cache_resource, next);
10589 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
10590 DRV_LOG(DEBUG, "push vlan action resource %p: removed",
10591 (void *)cache_resource);
10598 * Release the fate resource.
10601 * Pointer to Ethernet device.
10603 * Pointer to mlx5_flow_handle.
10606 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
10607 struct mlx5_flow_handle *handle)
10609 if (!handle->rix_fate)
10611 switch (handle->fate_action) {
10612 case MLX5_FLOW_FATE_QUEUE:
10613 mlx5_hrxq_release(dev, handle->rix_hrxq);
10615 case MLX5_FLOW_FATE_JUMP:
10616 flow_dv_jump_tbl_resource_release(dev, handle);
10618 case MLX5_FLOW_FATE_PORT_ID:
10619 flow_dv_port_id_action_resource_release(dev,
10620 handle->rix_port_id_action);
10623 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
10626 handle->rix_fate = 0;
10630 * Release an sample resource.
10633 * Pointer to Ethernet device.
10635 * Pointer to mlx5_flow_handle.
10638 * 1 while a reference on it exists, 0 when freed.
10641 flow_dv_sample_resource_release(struct rte_eth_dev *dev,
10642 struct mlx5_flow_handle *handle)
10644 struct mlx5_priv *priv = dev->data->dev_private;
10645 uint32_t idx = handle->dvh.rix_sample;
10646 struct mlx5_flow_dv_sample_resource *cache_resource;
10648 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
10650 if (!cache_resource)
10652 MLX5_ASSERT(cache_resource->verbs_action);
10653 DRV_LOG(DEBUG, "sample resource %p: refcnt %d--",
10654 (void *)cache_resource,
10655 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10656 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10657 __ATOMIC_RELAXED) == 0) {
10658 if (cache_resource->verbs_action)
10659 claim_zero(mlx5_glue->destroy_flow_action
10660 (cache_resource->verbs_action));
10661 if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
10662 if (cache_resource->default_miss)
10663 claim_zero(mlx5_glue->destroy_flow_action
10664 (cache_resource->default_miss));
10666 if (cache_resource->normal_path_tbl)
10667 flow_dv_tbl_resource_release(dev,
10668 cache_resource->normal_path_tbl);
10670 if (cache_resource->sample_idx.rix_hrxq &&
10671 !mlx5_hrxq_release(dev,
10672 cache_resource->sample_idx.rix_hrxq))
10673 cache_resource->sample_idx.rix_hrxq = 0;
10674 if (cache_resource->sample_idx.rix_tag &&
10675 !flow_dv_tag_release(dev,
10676 cache_resource->sample_idx.rix_tag))
10677 cache_resource->sample_idx.rix_tag = 0;
10678 if (cache_resource->sample_idx.cnt) {
10679 flow_dv_counter_release(dev,
10680 cache_resource->sample_idx.cnt);
10681 cache_resource->sample_idx.cnt = 0;
10683 if (!__atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED)) {
10684 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
10685 &priv->sh->sample_action_list, idx,
10686 cache_resource, next);
10687 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE], idx);
10688 DRV_LOG(DEBUG, "sample resource %p: removed",
10689 (void *)cache_resource);
10696 * Release an destination array resource.
10699 * Pointer to Ethernet device.
10701 * Pointer to mlx5_flow_handle.
10704 * 1 while a reference on it exists, 0 when freed.
10707 flow_dv_dest_array_resource_release(struct rte_eth_dev *dev,
10708 struct mlx5_flow_handle *handle)
10710 struct mlx5_priv *priv = dev->data->dev_private;
10711 struct mlx5_flow_dv_dest_array_resource *cache_resource;
10712 struct mlx5_flow_sub_actions_idx *mdest_act_res;
10713 uint32_t idx = handle->dvh.rix_dest_array;
10716 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
10718 if (!cache_resource)
10720 MLX5_ASSERT(cache_resource->action);
10721 DRV_LOG(DEBUG, "destination array resource %p: refcnt %d--",
10722 (void *)cache_resource,
10723 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10724 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10725 __ATOMIC_RELAXED) == 0) {
10726 if (cache_resource->action)
10727 claim_zero(mlx5_glue->destroy_flow_action
10728 (cache_resource->action));
10729 for (; i < cache_resource->num_of_dest; i++) {
10730 mdest_act_res = &cache_resource->sample_idx[i];
10731 if (mdest_act_res->rix_hrxq) {
10732 mlx5_hrxq_release(dev,
10733 mdest_act_res->rix_hrxq);
10734 mdest_act_res->rix_hrxq = 0;
10736 if (mdest_act_res->rix_encap_decap) {
10737 flow_dv_encap_decap_resource_release(dev,
10738 mdest_act_res->rix_encap_decap);
10739 mdest_act_res->rix_encap_decap = 0;
10741 if (mdest_act_res->rix_port_id_action) {
10742 flow_dv_port_id_action_resource_release(dev,
10743 mdest_act_res->rix_port_id_action);
10744 mdest_act_res->rix_port_id_action = 0;
10746 if (mdest_act_res->rix_tag) {
10747 flow_dv_tag_release(dev,
10748 mdest_act_res->rix_tag);
10749 mdest_act_res->rix_tag = 0;
10752 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
10753 &priv->sh->dest_array_list, idx,
10754 cache_resource, next);
10755 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY], idx);
10756 DRV_LOG(DEBUG, "destination array resource %p: removed",
10757 (void *)cache_resource);
10764 * Remove the flow from the NIC but keeps it in memory.
10765 * Lock free, (mutex should be acquired by caller).
10768 * Pointer to Ethernet device.
10769 * @param[in, out] flow
10770 * Pointer to flow structure.
10773 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
10775 struct mlx5_flow_handle *dh;
10776 uint32_t handle_idx;
10777 struct mlx5_priv *priv = dev->data->dev_private;
10781 handle_idx = flow->dev_handles;
10782 while (handle_idx) {
10783 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
10787 if (dh->drv_flow) {
10788 claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
10789 dh->drv_flow = NULL;
10791 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE)
10792 flow_dv_fate_resource_release(dev, dh);
10793 if (dh->vf_vlan.tag && dh->vf_vlan.created)
10794 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
10795 handle_idx = dh->next.next;
10800 * Remove the flow from the NIC and the memory.
10801 * Lock free, (mutex should be acquired by caller).
10804 * Pointer to the Ethernet device structure.
10805 * @param[in, out] flow
10806 * Pointer to flow structure.
10809 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
10811 struct rte_flow_shared_action *shared;
10812 struct mlx5_flow_handle *dev_handle;
10813 struct mlx5_priv *priv = dev->data->dev_private;
10817 __flow_dv_remove(dev, flow);
10818 shared = mlx5_flow_get_shared_rss(flow);
10820 __atomic_sub_fetch(&shared->refcnt, 1, __ATOMIC_RELAXED);
10821 if (flow->counter) {
10822 flow_dv_counter_release(dev, flow->counter);
10826 struct mlx5_flow_meter *fm;
10828 fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
10831 mlx5_flow_meter_detach(fm);
10834 while (flow->dev_handles) {
10835 uint32_t tmp_idx = flow->dev_handles;
10837 dev_handle = mlx5_ipool_get(priv->sh->ipool
10838 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
10841 flow->dev_handles = dev_handle->next.next;
10842 if (dev_handle->dvh.matcher)
10843 flow_dv_matcher_release(dev, dev_handle);
10844 if (dev_handle->dvh.rix_sample)
10845 flow_dv_sample_resource_release(dev, dev_handle);
10846 if (dev_handle->dvh.rix_dest_array)
10847 flow_dv_dest_array_resource_release(dev, dev_handle);
10848 if (dev_handle->dvh.rix_encap_decap)
10849 flow_dv_encap_decap_resource_release(dev,
10850 dev_handle->dvh.rix_encap_decap);
10851 if (dev_handle->dvh.modify_hdr)
10852 flow_dv_modify_hdr_resource_release(dev, dev_handle);
10853 if (dev_handle->dvh.rix_push_vlan)
10854 flow_dv_push_vlan_action_resource_release(dev,
10856 if (dev_handle->dvh.rix_tag)
10857 flow_dv_tag_release(dev,
10858 dev_handle->dvh.rix_tag);
10859 flow_dv_fate_resource_release(dev, dev_handle);
10860 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
10866 * Release array of hash RX queue objects.
10870 * Pointer to the Ethernet device structure.
10871 * @param[in, out] hrxqs
10872 * Array of hash RX queue objects.
10875 * Total number of references to hash RX queue objects in *hrxqs* array
10876 * after this operation.
10879 __flow_dv_hrxqs_release(struct rte_eth_dev *dev,
10880 uint32_t (*hrxqs)[MLX5_RSS_HASH_FIELDS_LEN])
10885 for (i = 0; i < RTE_DIM(*hrxqs); i++) {
10886 int ret = mlx5_hrxq_release(dev, (*hrxqs)[i]);
10896 * Release all hash RX queue objects representing shared RSS action.
10899 * Pointer to the Ethernet device structure.
10900 * @param[in, out] action
10901 * Shared RSS action to remove hash RX queue objects from.
10904 * Total number of references to hash RX queue objects stored in *action*
10905 * after this operation.
10906 * Expected to be 0 if no external references held.
10909 __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
10910 struct mlx5_shared_action_rss *action)
10912 return __flow_dv_hrxqs_release(dev, &action->hrxq) +
10913 __flow_dv_hrxqs_release(dev, &action->hrxq_tunnel);
10917 * Setup shared RSS action.
10918 * Prepare set of hash RX queue objects sufficient to handle all valid
10919 * hash_fields combinations (see enum ibv_rx_hash_fields).
10922 * Pointer to the Ethernet device structure.
10923 * @param[in, out] action
10924 * Partially initialized shared RSS action.
10925 * @param[out] error
10926 * Perform verbose error reporting if not NULL. Initialized in case of
10930 * 0 on success, otherwise negative errno value.
10933 __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
10934 struct mlx5_shared_action_rss *action,
10935 struct rte_flow_error *error)
10940 for (i = 0; i < MLX5_RSS_HASH_FIELDS_LEN; i++) {
10942 uint64_t hash_fields = mlx5_rss_hash_fields[i];
10945 for (tunnel = 0; tunnel < 2; tunnel++) {
10946 hrxq_idx = mlx5_hrxq_new(dev, action->origin.key,
10947 MLX5_RSS_HASH_KEY_LEN,
10949 action->origin.queue,
10950 action->origin.queue_num,
10955 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10956 "cannot get hash queue");
10957 goto error_hrxq_new;
10959 err = __flow_dv_action_rss_hrxq_set
10960 (action, hash_fields, tunnel, hrxq_idx);
10967 __flow_dv_action_rss_hrxqs_release(dev, action);
10973 * Create shared RSS action.
10976 * Pointer to the Ethernet device structure.
10978 * Shared action configuration.
10980 * RSS action specification used to create shared action.
10981 * @param[out] error
10982 * Perform verbose error reporting if not NULL. Initialized in case of
10986 * A valid shared action handle in case of success, NULL otherwise and
10987 * rte_errno is set.
10989 static struct rte_flow_shared_action *
10990 __flow_dv_action_rss_create(struct rte_eth_dev *dev,
10991 const struct rte_flow_shared_action_conf *conf,
10992 const struct rte_flow_action_rss *rss,
10993 struct rte_flow_error *error)
10995 struct rte_flow_shared_action *shared_action = NULL;
10996 void *queue = NULL;
10997 struct mlx5_shared_action_rss *shared_rss;
10998 struct rte_flow_action_rss *origin;
10999 const uint8_t *rss_key;
11000 uint32_t queue_size = rss->queue_num * sizeof(uint16_t);
11002 RTE_SET_USED(conf);
11003 queue = mlx5_malloc(0, RTE_ALIGN_CEIL(queue_size, sizeof(void *)),
11005 shared_action = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*shared_action), 0,
11007 if (!shared_action || !queue) {
11008 rte_flow_error_set(error, ENOMEM,
11009 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11010 "cannot allocate resource memory");
11011 goto error_rss_init;
11013 shared_rss = &shared_action->rss;
11014 shared_rss->queue = queue;
11015 origin = &shared_rss->origin;
11016 origin->func = rss->func;
11017 origin->level = rss->level;
11018 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
11019 origin->types = !rss->types ? ETH_RSS_IP : rss->types;
11020 /* NULL RSS key indicates default RSS key. */
11021 rss_key = !rss->key ? rss_hash_default_key : rss->key;
11022 memcpy(shared_rss->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
11023 origin->key = &shared_rss->key[0];
11024 origin->key_len = MLX5_RSS_HASH_KEY_LEN;
11025 memcpy(shared_rss->queue, rss->queue, queue_size);
11026 origin->queue = shared_rss->queue;
11027 origin->queue_num = rss->queue_num;
11028 if (__flow_dv_action_rss_setup(dev, shared_rss, error))
11029 goto error_rss_init;
11030 shared_action->type = MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS;
11031 return shared_action;
11033 mlx5_free(shared_action);
11039 * Destroy the shared RSS action.
11040 * Release related hash RX queue objects.
11043 * Pointer to the Ethernet device structure.
11044 * @param[in] shared_rss
11045 * The shared RSS action object to be removed.
11046 * @param[out] error
11047 * Perform verbose error reporting if not NULL. Initialized in case of
11051 * 0 on success, otherwise negative errno value.
11054 __flow_dv_action_rss_release(struct rte_eth_dev *dev,
11055 struct mlx5_shared_action_rss *shared_rss,
11056 struct rte_flow_error *error)
11058 struct rte_flow_shared_action *shared_action = NULL;
11059 uint32_t old_refcnt = 1;
11060 int remaining = __flow_dv_action_rss_hrxqs_release(dev, shared_rss);
11063 return rte_flow_error_set(error, ETOOMANYREFS,
11064 RTE_FLOW_ERROR_TYPE_ACTION,
11066 "shared rss hrxq has references");
11068 shared_action = container_of(shared_rss,
11069 struct rte_flow_shared_action, rss);
11070 if (!__atomic_compare_exchange_n(&shared_action->refcnt, &old_refcnt,
11072 __ATOMIC_ACQUIRE, __ATOMIC_RELAXED)) {
11073 return rte_flow_error_set(error, ETOOMANYREFS,
11074 RTE_FLOW_ERROR_TYPE_ACTION,
11076 "shared rss has references");
11078 rte_free(shared_rss->queue);
11083 * Create shared action, lock free,
11084 * (mutex should be acquired by caller).
11085 * Dispatcher for action type specific call.
11088 * Pointer to the Ethernet device structure.
11090 * Shared action configuration.
11091 * @param[in] action
11092 * Action specification used to create shared action.
11093 * @param[out] error
11094 * Perform verbose error reporting if not NULL. Initialized in case of
11098 * A valid shared action handle in case of success, NULL otherwise and
11099 * rte_errno is set.
11101 static struct rte_flow_shared_action *
11102 __flow_dv_action_create(struct rte_eth_dev *dev,
11103 const struct rte_flow_shared_action_conf *conf,
11104 const struct rte_flow_action *action,
11105 struct rte_flow_error *error)
11107 struct rte_flow_shared_action *shared_action = NULL;
11108 struct mlx5_priv *priv = dev->data->dev_private;
11110 switch (action->type) {
11111 case RTE_FLOW_ACTION_TYPE_RSS:
11112 shared_action = __flow_dv_action_rss_create(dev, conf,
11117 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
11118 NULL, "action type not supported");
11121 if (shared_action) {
11122 __atomic_add_fetch(&shared_action->refcnt, 1,
11124 LIST_INSERT_HEAD(&priv->shared_actions, shared_action, next);
11126 return shared_action;
11130 * Destroy the shared action.
11131 * Release action related resources on the NIC and the memory.
11132 * Lock free, (mutex should be acquired by caller).
11133 * Dispatcher for action type specific call.
11136 * Pointer to the Ethernet device structure.
11137 * @param[in] action
11138 * The shared action object to be removed.
11139 * @param[out] error
11140 * Perform verbose error reporting if not NULL. Initialized in case of
11144 * 0 on success, otherwise negative errno value.
11147 __flow_dv_action_destroy(struct rte_eth_dev *dev,
11148 struct rte_flow_shared_action *action,
11149 struct rte_flow_error *error)
11153 switch (action->type) {
11154 case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
11155 ret = __flow_dv_action_rss_release(dev, &action->rss, error);
11158 return rte_flow_error_set(error, ENOTSUP,
11159 RTE_FLOW_ERROR_TYPE_ACTION,
11161 "action type not supported");
11165 LIST_REMOVE(action, next);
11171 * Updates in place shared RSS action configuration.
11174 * Pointer to the Ethernet device structure.
11175 * @param[in] shared_rss
11176 * The shared RSS action object to be updated.
11177 * @param[in] action_conf
11178 * RSS action specification used to modify *shared_rss*.
11179 * @param[out] error
11180 * Perform verbose error reporting if not NULL. Initialized in case of
11184 * 0 on success, otherwise negative errno value.
11185 * @note: currently only support update of RSS queues.
11188 __flow_dv_action_rss_update(struct rte_eth_dev *dev,
11189 struct mlx5_shared_action_rss *shared_rss,
11190 const struct rte_flow_action_rss *action_conf,
11191 struct rte_flow_error *error)
11195 void *queue = NULL;
11196 const uint8_t *rss_key;
11197 uint32_t rss_key_len;
11198 uint32_t queue_size = action_conf->queue_num * sizeof(uint16_t);
11200 queue = mlx5_malloc(MLX5_MEM_ZERO,
11201 RTE_ALIGN_CEIL(queue_size, sizeof(void *)),
11204 return rte_flow_error_set(error, ENOMEM,
11205 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11207 "cannot allocate resource memory");
11208 if (action_conf->key) {
11209 rss_key = action_conf->key;
11210 rss_key_len = action_conf->key_len;
11212 rss_key = rss_hash_default_key;
11213 rss_key_len = MLX5_RSS_HASH_KEY_LEN;
11215 for (i = 0; i < MLX5_RSS_HASH_FIELDS_LEN; i++) {
11217 uint64_t hash_fields = mlx5_rss_hash_fields[i];
11220 for (tunnel = 0; tunnel < 2; tunnel++) {
11221 hrxq_idx = __flow_dv_action_rss_hrxq_lookup
11222 (shared_rss, hash_fields, tunnel);
11223 MLX5_ASSERT(hrxq_idx);
11224 ret = mlx5_hrxq_modify
11226 rss_key, rss_key_len,
11228 action_conf->queue, action_conf->queue_num);
11231 return rte_flow_error_set
11233 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
11234 "cannot update hash queue");
11238 mlx5_free(shared_rss->queue);
11239 shared_rss->queue = queue;
11240 memcpy(shared_rss->queue, action_conf->queue, queue_size);
11241 shared_rss->origin.queue = shared_rss->queue;
11242 shared_rss->origin.queue_num = action_conf->queue_num;
11247 * Updates in place shared action configuration, lock free,
11248 * (mutex should be acquired by caller).
11251 * Pointer to the Ethernet device structure.
11252 * @param[in] action
11253 * The shared action object to be updated.
11254 * @param[in] action_conf
11255 * Action specification used to modify *action*.
11256 * *action_conf* should be of type correlating with type of the *action*,
11257 * otherwise considered as invalid.
11258 * @param[out] error
11259 * Perform verbose error reporting if not NULL. Initialized in case of
11263 * 0 on success, otherwise negative errno value.
11266 __flow_dv_action_update(struct rte_eth_dev *dev,
11267 struct rte_flow_shared_action *action,
11268 const void *action_conf,
11269 struct rte_flow_error *error)
11271 switch (action->type) {
11272 case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
11273 return __flow_dv_action_rss_update(dev, &action->rss,
11274 action_conf, error);
11276 return rte_flow_error_set(error, ENOTSUP,
11277 RTE_FLOW_ERROR_TYPE_ACTION,
11279 "action type not supported");
11283 * Query a dv flow rule for its statistics via devx.
11286 * Pointer to Ethernet device.
11288 * Pointer to the sub flow.
11290 * data retrieved by the query.
11291 * @param[out] error
11292 * Perform verbose error reporting if not NULL.
11295 * 0 on success, a negative errno value otherwise and rte_errno is set.
11298 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
11299 void *data, struct rte_flow_error *error)
11301 struct mlx5_priv *priv = dev->data->dev_private;
11302 struct rte_flow_query_count *qc = data;
11304 if (!priv->config.devx)
11305 return rte_flow_error_set(error, ENOTSUP,
11306 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11308 "counters are not supported");
11309 if (flow->counter) {
11310 uint64_t pkts, bytes;
11311 struct mlx5_flow_counter *cnt;
11313 cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
11315 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
11319 return rte_flow_error_set(error, -err,
11320 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11321 NULL, "cannot read counters");
11324 qc->hits = pkts - cnt->hits;
11325 qc->bytes = bytes - cnt->bytes;
11328 cnt->bytes = bytes;
11332 return rte_flow_error_set(error, EINVAL,
11333 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11335 "counters are not available");
11339 * Query a flow rule AGE action for aging information.
11342 * Pointer to Ethernet device.
11344 * Pointer to the sub flow.
11346 * data retrieved by the query.
11347 * @param[out] error
11348 * Perform verbose error reporting if not NULL.
11351 * 0 on success, a negative errno value otherwise and rte_errno is set.
11354 flow_dv_query_age(struct rte_eth_dev *dev, struct rte_flow *flow,
11355 void *data, struct rte_flow_error *error)
11357 struct rte_flow_query_age *resp = data;
11359 if (flow->counter) {
11360 struct mlx5_age_param *age_param =
11361 flow_dv_counter_idx_get_age(dev, flow->counter);
11363 if (!age_param || !age_param->timeout)
11364 return rte_flow_error_set
11366 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11367 NULL, "cannot read age data");
11368 resp->aged = __atomic_load_n(&age_param->state,
11369 __ATOMIC_RELAXED) ==
11371 resp->sec_since_last_hit_valid = !resp->aged;
11372 if (resp->sec_since_last_hit_valid)
11373 resp->sec_since_last_hit =
11374 __atomic_load_n(&age_param->sec_since_last_hit,
11378 return rte_flow_error_set(error, EINVAL,
11379 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11381 "age data not available");
11387 * @see rte_flow_query()
11388 * @see rte_flow_ops
11391 flow_dv_query(struct rte_eth_dev *dev,
11392 struct rte_flow *flow __rte_unused,
11393 const struct rte_flow_action *actions __rte_unused,
11394 void *data __rte_unused,
11395 struct rte_flow_error *error __rte_unused)
11399 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
11400 switch (actions->type) {
11401 case RTE_FLOW_ACTION_TYPE_VOID:
11403 case RTE_FLOW_ACTION_TYPE_COUNT:
11404 ret = flow_dv_query_count(dev, flow, data, error);
11406 case RTE_FLOW_ACTION_TYPE_AGE:
11407 ret = flow_dv_query_age(dev, flow, data, error);
11410 return rte_flow_error_set(error, ENOTSUP,
11411 RTE_FLOW_ERROR_TYPE_ACTION,
11413 "action not supported");
11420 * Destroy the meter table set.
11421 * Lock free, (mutex should be acquired by caller).
11424 * Pointer to Ethernet device.
11426 * Pointer to the meter table set.
11432 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
11433 struct mlx5_meter_domains_infos *tbl)
11435 struct mlx5_priv *priv = dev->data->dev_private;
11436 struct mlx5_meter_domains_infos *mtd =
11437 (struct mlx5_meter_domains_infos *)tbl;
11439 if (!mtd || !priv->config.dv_flow_en)
11441 if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
11442 claim_zero(mlx5_flow_os_destroy_flow
11443 (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
11444 if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
11445 claim_zero(mlx5_flow_os_destroy_flow
11446 (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
11447 if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
11448 claim_zero(mlx5_flow_os_destroy_flow
11449 (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
11450 if (mtd->egress.color_matcher)
11451 claim_zero(mlx5_flow_os_destroy_flow_matcher
11452 (mtd->egress.color_matcher));
11453 if (mtd->egress.any_matcher)
11454 claim_zero(mlx5_flow_os_destroy_flow_matcher
11455 (mtd->egress.any_matcher));
11456 if (mtd->egress.tbl)
11457 flow_dv_tbl_resource_release(dev, mtd->egress.tbl);
11458 if (mtd->egress.sfx_tbl)
11459 flow_dv_tbl_resource_release(dev, mtd->egress.sfx_tbl);
11460 if (mtd->ingress.color_matcher)
11461 claim_zero(mlx5_flow_os_destroy_flow_matcher
11462 (mtd->ingress.color_matcher));
11463 if (mtd->ingress.any_matcher)
11464 claim_zero(mlx5_flow_os_destroy_flow_matcher
11465 (mtd->ingress.any_matcher));
11466 if (mtd->ingress.tbl)
11467 flow_dv_tbl_resource_release(dev, mtd->ingress.tbl);
11468 if (mtd->ingress.sfx_tbl)
11469 flow_dv_tbl_resource_release(dev, mtd->ingress.sfx_tbl);
11470 if (mtd->transfer.color_matcher)
11471 claim_zero(mlx5_flow_os_destroy_flow_matcher
11472 (mtd->transfer.color_matcher));
11473 if (mtd->transfer.any_matcher)
11474 claim_zero(mlx5_flow_os_destroy_flow_matcher
11475 (mtd->transfer.any_matcher));
11476 if (mtd->transfer.tbl)
11477 flow_dv_tbl_resource_release(dev, mtd->transfer.tbl);
11478 if (mtd->transfer.sfx_tbl)
11479 flow_dv_tbl_resource_release(dev, mtd->transfer.sfx_tbl);
11480 if (mtd->drop_actn)
11481 claim_zero(mlx5_flow_os_destroy_flow_action(mtd->drop_actn));
11486 /* Number of meter flow actions, count and jump or count and drop. */
11487 #define METER_ACTIONS 2
11490 * Create specify domain meter table and suffix table.
11493 * Pointer to Ethernet device.
11494 * @param[in,out] mtb
11495 * Pointer to DV meter table set.
11496 * @param[in] egress
11498 * @param[in] transfer
11500 * @param[in] color_reg_c_idx
11501 * Reg C index for color match.
11504 * 0 on success, -1 otherwise and rte_errno is set.
11507 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
11508 struct mlx5_meter_domains_infos *mtb,
11509 uint8_t egress, uint8_t transfer,
11510 uint32_t color_reg_c_idx)
11512 struct mlx5_priv *priv = dev->data->dev_private;
11513 struct mlx5_dev_ctx_shared *sh = priv->sh;
11514 struct mlx5_flow_dv_match_params mask = {
11515 .size = sizeof(mask.buf),
11517 struct mlx5_flow_dv_match_params value = {
11518 .size = sizeof(value.buf),
11520 struct mlx5dv_flow_matcher_attr dv_attr = {
11521 .type = IBV_FLOW_ATTR_NORMAL,
11523 .match_criteria_enable = 0,
11524 .match_mask = (void *)&mask,
11526 void *actions[METER_ACTIONS];
11527 struct mlx5_meter_domain_info *dtb;
11528 struct rte_flow_error error;
11533 dtb = &mtb->transfer;
11535 dtb = &mtb->egress;
11537 dtb = &mtb->ingress;
11538 /* Create the meter table with METER level. */
11539 dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
11540 egress, transfer, false, NULL, 0,
11543 DRV_LOG(ERR, "Failed to create meter policer table.");
11546 /* Create the meter suffix table with SUFFIX level. */
11547 dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
11548 MLX5_FLOW_TABLE_LEVEL_SUFFIX,
11549 egress, transfer, false, NULL, 0,
11551 if (!dtb->sfx_tbl) {
11552 DRV_LOG(ERR, "Failed to create meter suffix table.");
11555 /* Create matchers, Any and Color. */
11556 dv_attr.priority = 3;
11557 dv_attr.match_criteria_enable = 0;
11558 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
11559 &dtb->any_matcher);
11561 DRV_LOG(ERR, "Failed to create meter"
11562 " policer default matcher.");
11565 dv_attr.priority = 0;
11566 dv_attr.match_criteria_enable =
11567 1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
11568 flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
11569 rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
11570 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
11571 &dtb->color_matcher);
11573 DRV_LOG(ERR, "Failed to create meter policer color matcher.");
11576 if (mtb->count_actns[RTE_MTR_DROPPED])
11577 actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
11578 actions[i++] = mtb->drop_actn;
11579 /* Default rule: lowest priority, match any, actions: drop. */
11580 ret = mlx5_flow_os_create_flow(dtb->any_matcher, (void *)&value, i,
11582 &dtb->policer_rules[RTE_MTR_DROPPED]);
11584 DRV_LOG(ERR, "Failed to create meter policer drop rule.");
11593 * Create the needed meter and suffix tables.
11594 * Lock free, (mutex should be acquired by caller).
11597 * Pointer to Ethernet device.
11599 * Pointer to the flow meter.
11602 * Pointer to table set on success, NULL otherwise and rte_errno is set.
11604 static struct mlx5_meter_domains_infos *
11605 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
11606 const struct mlx5_flow_meter *fm)
11608 struct mlx5_priv *priv = dev->data->dev_private;
11609 struct mlx5_meter_domains_infos *mtb;
11613 if (!priv->mtr_en) {
11614 rte_errno = ENOTSUP;
11617 mtb = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*mtb), 0, SOCKET_ID_ANY);
11619 DRV_LOG(ERR, "Failed to allocate memory for meter.");
11622 /* Create meter count actions */
11623 for (i = 0; i <= RTE_MTR_DROPPED; i++) {
11624 struct mlx5_flow_counter *cnt;
11625 if (!fm->policer_stats.cnt[i])
11627 cnt = flow_dv_counter_get_by_idx(dev,
11628 fm->policer_stats.cnt[i], NULL);
11629 mtb->count_actns[i] = cnt->action;
11631 /* Create drop action. */
11632 ret = mlx5_flow_os_create_flow_action_drop(&mtb->drop_actn);
11634 DRV_LOG(ERR, "Failed to create drop action.");
11637 /* Egress meter table. */
11638 ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
11640 DRV_LOG(ERR, "Failed to prepare egress meter table.");
11643 /* Ingress meter table. */
11644 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
11646 DRV_LOG(ERR, "Failed to prepare ingress meter table.");
11649 /* FDB meter table. */
11650 if (priv->config.dv_esw_en) {
11651 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
11652 priv->mtr_color_reg);
11654 DRV_LOG(ERR, "Failed to prepare fdb meter table.");
11660 flow_dv_destroy_mtr_tbl(dev, mtb);
11665 * Destroy domain policer rule.
11668 * Pointer to domain table.
11671 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
11675 for (i = 0; i < RTE_MTR_DROPPED; i++) {
11676 if (dt->policer_rules[i]) {
11677 claim_zero(mlx5_flow_os_destroy_flow
11678 (dt->policer_rules[i]));
11679 dt->policer_rules[i] = NULL;
11682 if (dt->jump_actn) {
11683 claim_zero(mlx5_flow_os_destroy_flow_action(dt->jump_actn));
11684 dt->jump_actn = NULL;
11689 * Destroy policer rules.
11692 * Pointer to Ethernet device.
11694 * Pointer to flow meter structure.
11696 * Pointer to flow attributes.
11702 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
11703 const struct mlx5_flow_meter *fm,
11704 const struct rte_flow_attr *attr)
11706 struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
11711 flow_dv_destroy_domain_policer_rule(&mtb->egress);
11713 flow_dv_destroy_domain_policer_rule(&mtb->ingress);
11714 if (attr->transfer)
11715 flow_dv_destroy_domain_policer_rule(&mtb->transfer);
11720 * Create specify domain meter policer rule.
11723 * Pointer to flow meter structure.
11725 * Pointer to DV meter table set.
11726 * @param[in] mtr_reg_c
11727 * Color match REG_C.
11730 * 0 on success, -1 otherwise.
11733 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
11734 struct mlx5_meter_domain_info *dtb,
11737 struct mlx5_flow_dv_match_params matcher = {
11738 .size = sizeof(matcher.buf),
11740 struct mlx5_flow_dv_match_params value = {
11741 .size = sizeof(value.buf),
11743 struct mlx5_meter_domains_infos *mtb = fm->mfts;
11744 void *actions[METER_ACTIONS];
11748 /* Create jump action. */
11749 if (!dtb->jump_actn)
11750 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
11751 (dtb->sfx_tbl->obj, &dtb->jump_actn);
11753 DRV_LOG(ERR, "Failed to create policer jump action.");
11756 for (i = 0; i < RTE_MTR_DROPPED; i++) {
11759 flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
11760 rte_col_2_mlx5_col(i), UINT8_MAX);
11761 if (mtb->count_actns[i])
11762 actions[j++] = mtb->count_actns[i];
11763 if (fm->action[i] == MTR_POLICER_ACTION_DROP)
11764 actions[j++] = mtb->drop_actn;
11766 actions[j++] = dtb->jump_actn;
11767 ret = mlx5_flow_os_create_flow(dtb->color_matcher,
11768 (void *)&value, j, actions,
11769 &dtb->policer_rules[i]);
11771 DRV_LOG(ERR, "Failed to create policer rule.");
11782 * Create policer rules.
11785 * Pointer to Ethernet device.
11787 * Pointer to flow meter structure.
11789 * Pointer to flow attributes.
11792 * 0 on success, -1 otherwise.
11795 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
11796 struct mlx5_flow_meter *fm,
11797 const struct rte_flow_attr *attr)
11799 struct mlx5_priv *priv = dev->data->dev_private;
11800 struct mlx5_meter_domains_infos *mtb = fm->mfts;
11803 if (attr->egress) {
11804 ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
11805 priv->mtr_color_reg);
11807 DRV_LOG(ERR, "Failed to create egress policer.");
11811 if (attr->ingress) {
11812 ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
11813 priv->mtr_color_reg);
11815 DRV_LOG(ERR, "Failed to create ingress policer.");
11819 if (attr->transfer) {
11820 ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
11821 priv->mtr_color_reg);
11823 DRV_LOG(ERR, "Failed to create transfer policer.");
11829 flow_dv_destroy_policer_rules(dev, fm, attr);
11834 * Validate the batch counter support in root table.
11836 * Create a simple flow with invalid counter and drop action on root table to
11837 * validate if batch counter with offset on root table is supported or not.
11840 * Pointer to rte_eth_dev structure.
11843 * 0 on success, a negative errno value otherwise and rte_errno is set.
11846 mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev)
11848 struct mlx5_priv *priv = dev->data->dev_private;
11849 struct mlx5_dev_ctx_shared *sh = priv->sh;
11850 struct mlx5_flow_dv_match_params mask = {
11851 .size = sizeof(mask.buf),
11853 struct mlx5_flow_dv_match_params value = {
11854 .size = sizeof(value.buf),
11856 struct mlx5dv_flow_matcher_attr dv_attr = {
11857 .type = IBV_FLOW_ATTR_NORMAL,
11859 .match_criteria_enable = 0,
11860 .match_mask = (void *)&mask,
11862 void *actions[2] = { 0 };
11863 struct mlx5_flow_tbl_resource *tbl = NULL, *dest_tbl = NULL;
11864 struct mlx5_devx_obj *dcs = NULL;
11865 void *matcher = NULL;
11869 tbl = flow_dv_tbl_resource_get(dev, 0, 0, 0, false, NULL, 0, 0, NULL);
11872 dest_tbl = flow_dv_tbl_resource_get(dev, 1, 0, 0, false,
11876 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
11879 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, UINT16_MAX,
11883 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
11884 (dest_tbl->obj, &actions[1]);
11887 dv_attr.match_criteria_enable = flow_dv_matcher_enable(mask.buf);
11888 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
11892 ret = mlx5_flow_os_create_flow(matcher, (void *)&value, 2,
11896 * If batch counter with offset is not supported, the driver will not
11897 * validate the invalid offset value, flow create should success.
11898 * In this case, it means batch counter is not supported in root table.
11900 * Otherwise, if flow create is failed, counter offset is supported.
11903 DRV_LOG(INFO, "Batch counter is not supported in root "
11904 "table. Switch to fallback mode.");
11905 rte_errno = ENOTSUP;
11907 claim_zero(mlx5_flow_os_destroy_flow(flow));
11909 /* Check matcher to make sure validate fail at flow create. */
11910 if (!matcher || (matcher && errno != EINVAL))
11911 DRV_LOG(ERR, "Unexpected error in counter offset "
11912 "support detection");
11915 for (i = 0; i < 2; i++) {
11917 claim_zero(mlx5_flow_os_destroy_flow_action
11921 claim_zero(mlx5_flow_os_destroy_flow_matcher(matcher));
11923 flow_dv_tbl_resource_release(dev, tbl);
11925 flow_dv_tbl_resource_release(dev, dest_tbl);
11927 claim_zero(mlx5_devx_cmd_destroy(dcs));
11932 * Query a devx counter.
11935 * Pointer to the Ethernet device structure.
11937 * Index to the flow counter.
11939 * Set to clear the counter statistics.
11941 * The statistics value of packets.
11942 * @param[out] bytes
11943 * The statistics value of bytes.
11946 * 0 on success, otherwise return -1.
11949 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
11950 uint64_t *pkts, uint64_t *bytes)
11952 struct mlx5_priv *priv = dev->data->dev_private;
11953 struct mlx5_flow_counter *cnt;
11954 uint64_t inn_pkts, inn_bytes;
11957 if (!priv->config.devx)
11960 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
11963 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
11964 *pkts = inn_pkts - cnt->hits;
11965 *bytes = inn_bytes - cnt->bytes;
11967 cnt->hits = inn_pkts;
11968 cnt->bytes = inn_bytes;
11974 * Get aged-out flows.
11977 * Pointer to the Ethernet device structure.
11978 * @param[in] context
11979 * The address of an array of pointers to the aged-out flows contexts.
11980 * @param[in] nb_contexts
11981 * The length of context array pointers.
11982 * @param[out] error
11983 * Perform verbose error reporting if not NULL. Initialized in case of
11987 * how many contexts get in success, otherwise negative errno value.
11988 * if nb_contexts is 0, return the amount of all aged contexts.
11989 * if nb_contexts is not 0 , return the amount of aged flows reported
11990 * in the context array.
11991 * @note: only stub for now
11994 flow_get_aged_flows(struct rte_eth_dev *dev,
11996 uint32_t nb_contexts,
11997 struct rte_flow_error *error)
11999 struct mlx5_priv *priv = dev->data->dev_private;
12000 struct mlx5_age_info *age_info;
12001 struct mlx5_age_param *age_param;
12002 struct mlx5_flow_counter *counter;
12005 if (nb_contexts && !context)
12006 return rte_flow_error_set(error, EINVAL,
12007 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12009 "Should assign at least one flow or"
12010 " context to get if nb_contexts != 0");
12011 age_info = GET_PORT_AGE_INFO(priv);
12012 rte_spinlock_lock(&age_info->aged_sl);
12013 TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
12016 age_param = MLX5_CNT_TO_AGE(counter);
12017 context[nb_flows - 1] = age_param->context;
12018 if (!(--nb_contexts))
12022 rte_spinlock_unlock(&age_info->aged_sl);
12023 MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
12028 * Mutex-protected thunk to lock-free __flow_dv_translate().
12031 flow_dv_translate(struct rte_eth_dev *dev,
12032 struct mlx5_flow *dev_flow,
12033 const struct rte_flow_attr *attr,
12034 const struct rte_flow_item items[],
12035 const struct rte_flow_action actions[],
12036 struct rte_flow_error *error)
12040 flow_dv_shared_lock(dev);
12041 ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
12042 flow_dv_shared_unlock(dev);
12047 * Mutex-protected thunk to lock-free __flow_dv_apply().
12050 flow_dv_apply(struct rte_eth_dev *dev,
12051 struct rte_flow *flow,
12052 struct rte_flow_error *error)
12056 flow_dv_shared_lock(dev);
12057 ret = __flow_dv_apply(dev, flow, error);
12058 flow_dv_shared_unlock(dev);
12063 * Mutex-protected thunk to lock-free __flow_dv_remove().
12066 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
12068 flow_dv_shared_lock(dev);
12069 __flow_dv_remove(dev, flow);
12070 flow_dv_shared_unlock(dev);
12074 * Mutex-protected thunk to lock-free __flow_dv_destroy().
12077 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
12079 flow_dv_shared_lock(dev);
12080 __flow_dv_destroy(dev, flow);
12081 flow_dv_shared_unlock(dev);
12085 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
12088 flow_dv_counter_allocate(struct rte_eth_dev *dev)
12092 flow_dv_shared_lock(dev);
12093 cnt = flow_dv_counter_alloc(dev, 0);
12094 flow_dv_shared_unlock(dev);
12099 * Mutex-protected thunk to lock-free flow_dv_counter_release().
12102 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
12104 flow_dv_shared_lock(dev);
12105 flow_dv_counter_release(dev, cnt);
12106 flow_dv_shared_unlock(dev);
12110 * Validate shared action.
12111 * Dispatcher for action type specific validation.
12114 * Pointer to the Ethernet device structure.
12116 * Shared action configuration.
12117 * @param[in] action
12118 * The shared action object to validate.
12119 * @param[out] error
12120 * Perform verbose error reporting if not NULL. Initialized in case of
12124 * 0 on success, otherwise negative errno value.
12127 flow_dv_action_validate(struct rte_eth_dev *dev,
12128 const struct rte_flow_shared_action_conf *conf,
12129 const struct rte_flow_action *action,
12130 struct rte_flow_error *error)
12132 RTE_SET_USED(conf);
12133 switch (action->type) {
12134 case RTE_FLOW_ACTION_TYPE_RSS:
12135 return mlx5_validate_action_rss(dev, action, error);
12137 return rte_flow_error_set(error, ENOTSUP,
12138 RTE_FLOW_ERROR_TYPE_ACTION,
12140 "action type not supported");
12145 * Mutex-protected thunk to lock-free __flow_dv_action_create().
12147 static struct rte_flow_shared_action *
12148 flow_dv_action_create(struct rte_eth_dev *dev,
12149 const struct rte_flow_shared_action_conf *conf,
12150 const struct rte_flow_action *action,
12151 struct rte_flow_error *error)
12153 struct rte_flow_shared_action *shared_action = NULL;
12155 flow_dv_shared_lock(dev);
12156 shared_action = __flow_dv_action_create(dev, conf, action, error);
12157 flow_dv_shared_unlock(dev);
12158 return shared_action;
12162 * Mutex-protected thunk to lock-free __flow_dv_action_destroy().
12165 flow_dv_action_destroy(struct rte_eth_dev *dev,
12166 struct rte_flow_shared_action *action,
12167 struct rte_flow_error *error)
12171 flow_dv_shared_lock(dev);
12172 ret = __flow_dv_action_destroy(dev, action, error);
12173 flow_dv_shared_unlock(dev);
12178 * Mutex-protected thunk to lock-free __flow_dv_action_update().
12181 flow_dv_action_update(struct rte_eth_dev *dev,
12182 struct rte_flow_shared_action *action,
12183 const void *action_conf,
12184 struct rte_flow_error *error)
12188 flow_dv_shared_lock(dev);
12189 ret = __flow_dv_action_update(dev, action, action_conf,
12191 flow_dv_shared_unlock(dev);
12196 flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
12198 struct mlx5_priv *priv = dev->data->dev_private;
12201 if ((domains & MLX5_DOMAIN_BIT_NIC_RX) && priv->sh->rx_domain != NULL) {
12202 ret = mlx5_glue->dr_sync_domain(priv->sh->rx_domain,
12207 if ((domains & MLX5_DOMAIN_BIT_NIC_TX) && priv->sh->tx_domain != NULL) {
12208 ret = mlx5_glue->dr_sync_domain(priv->sh->tx_domain, flags);
12212 if ((domains & MLX5_DOMAIN_BIT_FDB) && priv->sh->fdb_domain != NULL) {
12213 ret = mlx5_glue->dr_sync_domain(priv->sh->fdb_domain, flags);
12220 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
12221 .validate = flow_dv_validate,
12222 .prepare = flow_dv_prepare,
12223 .translate = flow_dv_translate,
12224 .apply = flow_dv_apply,
12225 .remove = flow_dv_remove,
12226 .destroy = flow_dv_destroy,
12227 .query = flow_dv_query,
12228 .create_mtr_tbls = flow_dv_create_mtr_tbl,
12229 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
12230 .create_policer_rules = flow_dv_create_policer_rules,
12231 .destroy_policer_rules = flow_dv_destroy_policer_rules,
12232 .counter_alloc = flow_dv_counter_allocate,
12233 .counter_free = flow_dv_counter_free,
12234 .counter_query = flow_dv_counter_query,
12235 .get_aged_flows = flow_get_aged_flows,
12236 .action_validate = flow_dv_action_validate,
12237 .action_create = flow_dv_action_create,
12238 .action_destroy = flow_dv_action_destroy,
12239 .action_update = flow_dv_action_update,
12240 .sync_domain = flow_dv_sync_domain,
12243 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */