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 the hash list.
2790 * Pointer to exist resource entry object.
2792 * Key of the new entry.
2794 * Pointer to new encap_decap resource.
2797 * 0 on matching, none-zero otherwise.
2800 flow_dv_encap_decap_match_cb(struct mlx5_hlist *list __rte_unused,
2801 struct mlx5_hlist_entry *entry,
2802 uint64_t key __rte_unused, void *cb_ctx)
2804 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
2805 struct mlx5_flow_dv_encap_decap_resource *resource = ctx->data;
2806 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2808 cache_resource = container_of(entry,
2809 struct mlx5_flow_dv_encap_decap_resource,
2811 if (resource->entry.key == cache_resource->entry.key &&
2812 resource->reformat_type == cache_resource->reformat_type &&
2813 resource->ft_type == cache_resource->ft_type &&
2814 resource->flags == cache_resource->flags &&
2815 resource->size == cache_resource->size &&
2816 !memcmp((const void *)resource->buf,
2817 (const void *)cache_resource->buf,
2824 * Allocate encap_decap resource.
2827 * Pointer to the hash list.
2829 * Pointer to exist resource entry object.
2831 * Pointer to new encap_decap resource.
2834 * 0 on matching, none-zero otherwise.
2836 struct mlx5_hlist_entry *
2837 flow_dv_encap_decap_create_cb(struct mlx5_hlist *list,
2838 uint64_t key __rte_unused,
2841 struct mlx5_dev_ctx_shared *sh = list->ctx;
2842 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
2843 struct mlx5dv_dr_domain *domain;
2844 struct mlx5_flow_dv_encap_decap_resource *resource = ctx->data;
2845 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2849 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2850 domain = sh->fdb_domain;
2851 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2852 domain = sh->rx_domain;
2854 domain = sh->tx_domain;
2855 /* Register new encap/decap resource. */
2856 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2858 if (!cache_resource) {
2859 rte_flow_error_set(ctx->error, ENOMEM,
2860 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2861 "cannot allocate resource memory");
2864 *cache_resource = *resource;
2865 cache_resource->idx = idx;
2866 ret = mlx5_flow_os_create_flow_action_packet_reformat
2867 (sh->ctx, domain, cache_resource,
2868 &cache_resource->action);
2870 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
2871 rte_flow_error_set(ctx->error, ENOMEM,
2872 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2873 NULL, "cannot create action");
2877 return &cache_resource->entry;
2881 * Find existing encap/decap resource or create and register a new one.
2883 * @param[in, out] dev
2884 * Pointer to rte_eth_dev structure.
2885 * @param[in, out] resource
2886 * Pointer to encap/decap resource.
2887 * @parm[in, out] dev_flow
2888 * Pointer to the dev_flow.
2890 * pointer to error structure.
2893 * 0 on success otherwise -errno and errno is set.
2896 flow_dv_encap_decap_resource_register
2897 (struct rte_eth_dev *dev,
2898 struct mlx5_flow_dv_encap_decap_resource *resource,
2899 struct mlx5_flow *dev_flow,
2900 struct rte_flow_error *error)
2902 struct mlx5_priv *priv = dev->data->dev_private;
2903 struct mlx5_dev_ctx_shared *sh = priv->sh;
2904 struct mlx5_hlist_entry *entry;
2905 union mlx5_flow_encap_decap_key encap_decap_key = {
2907 .ft_type = resource->ft_type,
2908 .refmt_type = resource->reformat_type,
2909 .buf_size = resource->size,
2910 .table_level = !!dev_flow->dv.group,
2914 struct mlx5_flow_cb_ctx ctx = {
2919 resource->flags = dev_flow->dv.group ? 0 : 1;
2920 encap_decap_key.cksum = __rte_raw_cksum(resource->buf,
2922 resource->entry.key = encap_decap_key.v64;
2923 entry = mlx5_hlist_register(sh->encaps_decaps, resource->entry.key,
2927 resource = container_of(entry, typeof(*resource), entry);
2928 dev_flow->dv.encap_decap = resource;
2929 dev_flow->handle->dvh.rix_encap_decap = resource->idx;
2934 * Find existing table jump resource or create and register a new one.
2936 * @param[in, out] dev
2937 * Pointer to rte_eth_dev structure.
2938 * @param[in, out] tbl
2939 * Pointer to flow table resource.
2940 * @parm[in, out] dev_flow
2941 * Pointer to the dev_flow.
2943 * pointer to error structure.
2946 * 0 on success otherwise -errno and errno is set.
2949 flow_dv_jump_tbl_resource_register
2950 (struct rte_eth_dev *dev __rte_unused,
2951 struct mlx5_flow_tbl_resource *tbl,
2952 struct mlx5_flow *dev_flow,
2953 struct rte_flow_error *error __rte_unused)
2955 struct mlx5_flow_tbl_data_entry *tbl_data =
2956 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2959 MLX5_ASSERT(tbl_data->jump.action);
2960 dev_flow->handle->rix_jump = tbl_data->idx;
2961 dev_flow->dv.jump = &tbl_data->jump;
2966 * Find existing table port ID resource or create and register a new one.
2968 * @param[in, out] dev
2969 * Pointer to rte_eth_dev structure.
2970 * @param[in, out] resource
2971 * Pointer to port ID action resource.
2972 * @parm[in, out] dev_flow
2973 * Pointer to the dev_flow.
2975 * pointer to error structure.
2978 * 0 on success otherwise -errno and errno is set.
2981 flow_dv_port_id_action_resource_register
2982 (struct rte_eth_dev *dev,
2983 struct mlx5_flow_dv_port_id_action_resource *resource,
2984 struct mlx5_flow *dev_flow,
2985 struct rte_flow_error *error)
2987 struct mlx5_priv *priv = dev->data->dev_private;
2988 struct mlx5_dev_ctx_shared *sh = priv->sh;
2989 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
2993 /* Lookup a matching resource from cache. */
2994 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
2995 idx, cache_resource, next) {
2996 if (resource->port_id == cache_resource->port_id) {
2997 DRV_LOG(DEBUG, "port id action resource resource %p: "
2999 (void *)cache_resource,
3000 __atomic_load_n(&cache_resource->refcnt,
3002 __atomic_fetch_add(&cache_resource->refcnt, 1,
3004 dev_flow->handle->rix_port_id_action = idx;
3005 dev_flow->dv.port_id_action = cache_resource;
3009 /* Register new port id action resource. */
3010 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
3011 &dev_flow->handle->rix_port_id_action);
3012 if (!cache_resource)
3013 return rte_flow_error_set(error, ENOMEM,
3014 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3015 "cannot allocate resource memory");
3016 *cache_resource = *resource;
3017 ret = mlx5_flow_os_create_flow_action_dest_port
3018 (priv->sh->fdb_domain, resource->port_id,
3019 &cache_resource->action);
3021 mlx5_free(cache_resource);
3022 return rte_flow_error_set(error, ENOMEM,
3023 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3024 NULL, "cannot create action");
3026 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
3027 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
3028 dev_flow->handle->rix_port_id_action, cache_resource,
3030 dev_flow->dv.port_id_action = cache_resource;
3031 DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
3032 (void *)cache_resource,
3033 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
3038 * Find existing push vlan resource or create and register a new one.
3040 * @param [in, out] dev
3041 * Pointer to rte_eth_dev structure.
3042 * @param[in, out] resource
3043 * Pointer to port ID action resource.
3044 * @parm[in, out] dev_flow
3045 * Pointer to the dev_flow.
3047 * pointer to error structure.
3050 * 0 on success otherwise -errno and errno is set.
3053 flow_dv_push_vlan_action_resource_register
3054 (struct rte_eth_dev *dev,
3055 struct mlx5_flow_dv_push_vlan_action_resource *resource,
3056 struct mlx5_flow *dev_flow,
3057 struct rte_flow_error *error)
3059 struct mlx5_priv *priv = dev->data->dev_private;
3060 struct mlx5_dev_ctx_shared *sh = priv->sh;
3061 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
3062 struct mlx5dv_dr_domain *domain;
3066 /* Lookup a matching resource from cache. */
3067 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3068 sh->push_vlan_action_list, idx, cache_resource, next) {
3069 if (resource->vlan_tag == cache_resource->vlan_tag &&
3070 resource->ft_type == cache_resource->ft_type) {
3071 DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
3073 (void *)cache_resource,
3074 __atomic_load_n(&cache_resource->refcnt,
3076 __atomic_fetch_add(&cache_resource->refcnt, 1,
3078 dev_flow->handle->dvh.rix_push_vlan = idx;
3079 dev_flow->dv.push_vlan_res = cache_resource;
3083 /* Register new push_vlan action resource. */
3084 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3085 &dev_flow->handle->dvh.rix_push_vlan);
3086 if (!cache_resource)
3087 return rte_flow_error_set(error, ENOMEM,
3088 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3089 "cannot allocate resource memory");
3090 *cache_resource = *resource;
3091 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3092 domain = sh->fdb_domain;
3093 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
3094 domain = sh->rx_domain;
3096 domain = sh->tx_domain;
3097 ret = mlx5_flow_os_create_flow_action_push_vlan
3098 (domain, resource->vlan_tag,
3099 &cache_resource->action);
3101 mlx5_free(cache_resource);
3102 return rte_flow_error_set(error, ENOMEM,
3103 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3104 NULL, "cannot create action");
3106 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
3107 ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
3108 &sh->push_vlan_action_list,
3109 dev_flow->handle->dvh.rix_push_vlan,
3110 cache_resource, next);
3111 dev_flow->dv.push_vlan_res = cache_resource;
3112 DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
3113 (void *)cache_resource,
3114 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
3118 * Get the size of specific rte_flow_item_type hdr size
3120 * @param[in] item_type
3121 * Tested rte_flow_item_type.
3124 * sizeof struct item_type, 0 if void or irrelevant.
3127 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
3131 switch (item_type) {
3132 case RTE_FLOW_ITEM_TYPE_ETH:
3133 retval = sizeof(struct rte_ether_hdr);
3135 case RTE_FLOW_ITEM_TYPE_VLAN:
3136 retval = sizeof(struct rte_vlan_hdr);
3138 case RTE_FLOW_ITEM_TYPE_IPV4:
3139 retval = sizeof(struct rte_ipv4_hdr);
3141 case RTE_FLOW_ITEM_TYPE_IPV6:
3142 retval = sizeof(struct rte_ipv6_hdr);
3144 case RTE_FLOW_ITEM_TYPE_UDP:
3145 retval = sizeof(struct rte_udp_hdr);
3147 case RTE_FLOW_ITEM_TYPE_TCP:
3148 retval = sizeof(struct rte_tcp_hdr);
3150 case RTE_FLOW_ITEM_TYPE_VXLAN:
3151 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3152 retval = sizeof(struct rte_vxlan_hdr);
3154 case RTE_FLOW_ITEM_TYPE_GRE:
3155 case RTE_FLOW_ITEM_TYPE_NVGRE:
3156 retval = sizeof(struct rte_gre_hdr);
3158 case RTE_FLOW_ITEM_TYPE_MPLS:
3159 retval = sizeof(struct rte_mpls_hdr);
3161 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
3169 #define MLX5_ENCAP_IPV4_VERSION 0x40
3170 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
3171 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
3172 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
3173 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
3174 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
3175 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
3178 * Convert the encap action data from list of rte_flow_item to raw buffer
3181 * Pointer to rte_flow_item objects list.
3183 * Pointer to the output buffer.
3185 * Pointer to the output buffer size.
3187 * Pointer to the error structure.
3190 * 0 on success, a negative errno value otherwise and rte_errno is set.
3193 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
3194 size_t *size, struct rte_flow_error *error)
3196 struct rte_ether_hdr *eth = NULL;
3197 struct rte_vlan_hdr *vlan = NULL;
3198 struct rte_ipv4_hdr *ipv4 = NULL;
3199 struct rte_ipv6_hdr *ipv6 = NULL;
3200 struct rte_udp_hdr *udp = NULL;
3201 struct rte_vxlan_hdr *vxlan = NULL;
3202 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
3203 struct rte_gre_hdr *gre = NULL;
3205 size_t temp_size = 0;
3208 return rte_flow_error_set(error, EINVAL,
3209 RTE_FLOW_ERROR_TYPE_ACTION,
3210 NULL, "invalid empty data");
3211 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3212 len = flow_dv_get_item_hdr_len(items->type);
3213 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
3214 return rte_flow_error_set(error, EINVAL,
3215 RTE_FLOW_ERROR_TYPE_ACTION,
3216 (void *)items->type,
3217 "items total size is too big"
3218 " for encap action");
3219 rte_memcpy((void *)&buf[temp_size], items->spec, len);
3220 switch (items->type) {
3221 case RTE_FLOW_ITEM_TYPE_ETH:
3222 eth = (struct rte_ether_hdr *)&buf[temp_size];
3224 case RTE_FLOW_ITEM_TYPE_VLAN:
3225 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
3227 return rte_flow_error_set(error, EINVAL,
3228 RTE_FLOW_ERROR_TYPE_ACTION,
3229 (void *)items->type,
3230 "eth header not found");
3231 if (!eth->ether_type)
3232 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
3234 case RTE_FLOW_ITEM_TYPE_IPV4:
3235 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
3237 return rte_flow_error_set(error, EINVAL,
3238 RTE_FLOW_ERROR_TYPE_ACTION,
3239 (void *)items->type,
3240 "neither eth nor vlan"
3242 if (vlan && !vlan->eth_proto)
3243 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3244 else if (eth && !eth->ether_type)
3245 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3246 if (!ipv4->version_ihl)
3247 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
3248 MLX5_ENCAP_IPV4_IHL_MIN;
3249 if (!ipv4->time_to_live)
3250 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
3252 case RTE_FLOW_ITEM_TYPE_IPV6:
3253 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
3255 return rte_flow_error_set(error, EINVAL,
3256 RTE_FLOW_ERROR_TYPE_ACTION,
3257 (void *)items->type,
3258 "neither eth nor vlan"
3260 if (vlan && !vlan->eth_proto)
3261 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3262 else if (eth && !eth->ether_type)
3263 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3264 if (!ipv6->vtc_flow)
3266 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
3267 if (!ipv6->hop_limits)
3268 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
3270 case RTE_FLOW_ITEM_TYPE_UDP:
3271 udp = (struct rte_udp_hdr *)&buf[temp_size];
3273 return rte_flow_error_set(error, EINVAL,
3274 RTE_FLOW_ERROR_TYPE_ACTION,
3275 (void *)items->type,
3276 "ip header not found");
3277 if (ipv4 && !ipv4->next_proto_id)
3278 ipv4->next_proto_id = IPPROTO_UDP;
3279 else if (ipv6 && !ipv6->proto)
3280 ipv6->proto = IPPROTO_UDP;
3282 case RTE_FLOW_ITEM_TYPE_VXLAN:
3283 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
3285 return rte_flow_error_set(error, EINVAL,
3286 RTE_FLOW_ERROR_TYPE_ACTION,
3287 (void *)items->type,
3288 "udp header not found");
3290 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
3291 if (!vxlan->vx_flags)
3293 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
3295 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3296 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
3298 return rte_flow_error_set(error, EINVAL,
3299 RTE_FLOW_ERROR_TYPE_ACTION,
3300 (void *)items->type,
3301 "udp header not found");
3302 if (!vxlan_gpe->proto)
3303 return rte_flow_error_set(error, EINVAL,
3304 RTE_FLOW_ERROR_TYPE_ACTION,
3305 (void *)items->type,
3306 "next protocol not found");
3309 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
3310 if (!vxlan_gpe->vx_flags)
3311 vxlan_gpe->vx_flags =
3312 MLX5_ENCAP_VXLAN_GPE_FLAGS;
3314 case RTE_FLOW_ITEM_TYPE_GRE:
3315 case RTE_FLOW_ITEM_TYPE_NVGRE:
3316 gre = (struct rte_gre_hdr *)&buf[temp_size];
3318 return rte_flow_error_set(error, EINVAL,
3319 RTE_FLOW_ERROR_TYPE_ACTION,
3320 (void *)items->type,
3321 "next protocol not found");
3323 return rte_flow_error_set(error, EINVAL,
3324 RTE_FLOW_ERROR_TYPE_ACTION,
3325 (void *)items->type,
3326 "ip header not found");
3327 if (ipv4 && !ipv4->next_proto_id)
3328 ipv4->next_proto_id = IPPROTO_GRE;
3329 else if (ipv6 && !ipv6->proto)
3330 ipv6->proto = IPPROTO_GRE;
3332 case RTE_FLOW_ITEM_TYPE_VOID:
3335 return rte_flow_error_set(error, EINVAL,
3336 RTE_FLOW_ERROR_TYPE_ACTION,
3337 (void *)items->type,
3338 "unsupported item type");
3348 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
3350 struct rte_ether_hdr *eth = NULL;
3351 struct rte_vlan_hdr *vlan = NULL;
3352 struct rte_ipv6_hdr *ipv6 = NULL;
3353 struct rte_udp_hdr *udp = NULL;
3357 eth = (struct rte_ether_hdr *)data;
3358 next_hdr = (char *)(eth + 1);
3359 proto = RTE_BE16(eth->ether_type);
3362 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
3363 vlan = (struct rte_vlan_hdr *)next_hdr;
3364 proto = RTE_BE16(vlan->eth_proto);
3365 next_hdr += sizeof(struct rte_vlan_hdr);
3368 /* HW calculates IPv4 csum. no need to proceed */
3369 if (proto == RTE_ETHER_TYPE_IPV4)
3372 /* non IPv4/IPv6 header. not supported */
3373 if (proto != RTE_ETHER_TYPE_IPV6) {
3374 return rte_flow_error_set(error, ENOTSUP,
3375 RTE_FLOW_ERROR_TYPE_ACTION,
3376 NULL, "Cannot offload non IPv4/IPv6");
3379 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3381 /* ignore non UDP */
3382 if (ipv6->proto != IPPROTO_UDP)
3385 udp = (struct rte_udp_hdr *)(ipv6 + 1);
3386 udp->dgram_cksum = 0;
3392 * Convert L2 encap action to DV specification.
3395 * Pointer to rte_eth_dev structure.
3397 * Pointer to action structure.
3398 * @param[in, out] dev_flow
3399 * Pointer to the mlx5_flow.
3400 * @param[in] transfer
3401 * Mark if the flow is E-Switch flow.
3403 * Pointer to the error structure.
3406 * 0 on success, a negative errno value otherwise and rte_errno is set.
3409 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3410 const struct rte_flow_action *action,
3411 struct mlx5_flow *dev_flow,
3413 struct rte_flow_error *error)
3415 const struct rte_flow_item *encap_data;
3416 const struct rte_flow_action_raw_encap *raw_encap_data;
3417 struct mlx5_flow_dv_encap_decap_resource res = {
3419 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3420 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3421 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3424 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3426 (const struct rte_flow_action_raw_encap *)action->conf;
3427 res.size = raw_encap_data->size;
3428 memcpy(res.buf, raw_encap_data->data, res.size);
3430 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3432 ((const struct rte_flow_action_vxlan_encap *)
3433 action->conf)->definition;
3436 ((const struct rte_flow_action_nvgre_encap *)
3437 action->conf)->definition;
3438 if (flow_dv_convert_encap_data(encap_data, res.buf,
3442 if (flow_dv_zero_encap_udp_csum(res.buf, error))
3444 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3445 return rte_flow_error_set(error, EINVAL,
3446 RTE_FLOW_ERROR_TYPE_ACTION,
3447 NULL, "can't create L2 encap action");
3452 * Convert L2 decap action to DV specification.
3455 * Pointer to rte_eth_dev structure.
3456 * @param[in, out] dev_flow
3457 * Pointer to the mlx5_flow.
3458 * @param[in] transfer
3459 * Mark if the flow is E-Switch flow.
3461 * Pointer to the error structure.
3464 * 0 on success, a negative errno value otherwise and rte_errno is set.
3467 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3468 struct mlx5_flow *dev_flow,
3470 struct rte_flow_error *error)
3472 struct mlx5_flow_dv_encap_decap_resource res = {
3475 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3476 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3477 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3480 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3481 return rte_flow_error_set(error, EINVAL,
3482 RTE_FLOW_ERROR_TYPE_ACTION,
3483 NULL, "can't create L2 decap action");
3488 * Convert raw decap/encap (L3 tunnel) action to DV specification.
3491 * Pointer to rte_eth_dev structure.
3493 * Pointer to action structure.
3494 * @param[in, out] dev_flow
3495 * Pointer to the mlx5_flow.
3497 * Pointer to the flow attributes.
3499 * Pointer to the error structure.
3502 * 0 on success, a negative errno value otherwise and rte_errno is set.
3505 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3506 const struct rte_flow_action *action,
3507 struct mlx5_flow *dev_flow,
3508 const struct rte_flow_attr *attr,
3509 struct rte_flow_error *error)
3511 const struct rte_flow_action_raw_encap *encap_data;
3512 struct mlx5_flow_dv_encap_decap_resource res;
3514 memset(&res, 0, sizeof(res));
3515 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3516 res.size = encap_data->size;
3517 memcpy(res.buf, encap_data->data, res.size);
3518 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3519 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3520 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3522 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3524 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3525 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3526 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3527 return rte_flow_error_set(error, EINVAL,
3528 RTE_FLOW_ERROR_TYPE_ACTION,
3529 NULL, "can't create encap action");
3534 * Create action push VLAN.
3537 * Pointer to rte_eth_dev structure.
3539 * Pointer to the flow attributes.
3541 * Pointer to the vlan to push to the Ethernet header.
3542 * @param[in, out] dev_flow
3543 * Pointer to the mlx5_flow.
3545 * Pointer to the error structure.
3548 * 0 on success, a negative errno value otherwise and rte_errno is set.
3551 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3552 const struct rte_flow_attr *attr,
3553 const struct rte_vlan_hdr *vlan,
3554 struct mlx5_flow *dev_flow,
3555 struct rte_flow_error *error)
3557 struct mlx5_flow_dv_push_vlan_action_resource res;
3559 memset(&res, 0, sizeof(res));
3561 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3564 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3566 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3567 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3568 return flow_dv_push_vlan_action_resource_register
3569 (dev, &res, dev_flow, error);
3572 static int fdb_mirror;
3575 * Validate the modify-header actions.
3577 * @param[in] action_flags
3578 * Holds the actions detected until now.
3580 * Pointer to the modify action.
3582 * Pointer to error structure.
3585 * 0 on success, a negative errno value otherwise and rte_errno is set.
3588 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3589 const struct rte_flow_action *action,
3590 struct rte_flow_error *error)
3592 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3593 return rte_flow_error_set(error, EINVAL,
3594 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3595 NULL, "action configuration not set");
3596 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3597 return rte_flow_error_set(error, EINVAL,
3598 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3599 "can't have encap action before"
3601 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3602 return rte_flow_error_set(error, EINVAL,
3603 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3604 "can't support sample action before"
3605 " modify action for E-Switch"
3611 * Validate the modify-header MAC address actions.
3613 * @param[in] action_flags
3614 * Holds the actions detected until now.
3616 * Pointer to the modify action.
3617 * @param[in] item_flags
3618 * Holds the items detected.
3620 * Pointer to error structure.
3623 * 0 on success, a negative errno value otherwise and rte_errno is set.
3626 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3627 const struct rte_flow_action *action,
3628 const uint64_t item_flags,
3629 struct rte_flow_error *error)
3633 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3635 if (!(item_flags & MLX5_FLOW_LAYER_L2))
3636 return rte_flow_error_set(error, EINVAL,
3637 RTE_FLOW_ERROR_TYPE_ACTION,
3639 "no L2 item in pattern");
3645 * Validate the modify-header IPv4 address actions.
3647 * @param[in] action_flags
3648 * Holds the actions detected until now.
3650 * Pointer to the modify action.
3651 * @param[in] item_flags
3652 * Holds the items detected.
3654 * Pointer to error structure.
3657 * 0 on success, a negative errno value otherwise and rte_errno is set.
3660 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3661 const struct rte_flow_action *action,
3662 const uint64_t item_flags,
3663 struct rte_flow_error *error)
3668 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3670 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3671 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3672 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3673 if (!(item_flags & layer))
3674 return rte_flow_error_set(error, EINVAL,
3675 RTE_FLOW_ERROR_TYPE_ACTION,
3677 "no ipv4 item in pattern");
3683 * Validate the modify-header IPv6 address actions.
3685 * @param[in] action_flags
3686 * Holds the actions detected until now.
3688 * Pointer to the modify action.
3689 * @param[in] item_flags
3690 * Holds the items detected.
3692 * Pointer to error structure.
3695 * 0 on success, a negative errno value otherwise and rte_errno is set.
3698 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3699 const struct rte_flow_action *action,
3700 const uint64_t item_flags,
3701 struct rte_flow_error *error)
3706 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3708 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3709 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3710 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3711 if (!(item_flags & layer))
3712 return rte_flow_error_set(error, EINVAL,
3713 RTE_FLOW_ERROR_TYPE_ACTION,
3715 "no ipv6 item in pattern");
3721 * Validate the modify-header TP actions.
3723 * @param[in] action_flags
3724 * Holds the actions detected until now.
3726 * Pointer to the modify action.
3727 * @param[in] item_flags
3728 * Holds the items detected.
3730 * Pointer to error structure.
3733 * 0 on success, a negative errno value otherwise and rte_errno is set.
3736 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3737 const struct rte_flow_action *action,
3738 const uint64_t item_flags,
3739 struct rte_flow_error *error)
3744 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3746 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3747 MLX5_FLOW_LAYER_INNER_L4 :
3748 MLX5_FLOW_LAYER_OUTER_L4;
3749 if (!(item_flags & layer))
3750 return rte_flow_error_set(error, EINVAL,
3751 RTE_FLOW_ERROR_TYPE_ACTION,
3752 NULL, "no transport layer "
3759 * Validate the modify-header actions of increment/decrement
3760 * TCP Sequence-number.
3762 * @param[in] action_flags
3763 * Holds the actions detected until now.
3765 * Pointer to the modify action.
3766 * @param[in] item_flags
3767 * Holds the items detected.
3769 * Pointer to error structure.
3772 * 0 on success, a negative errno value otherwise and rte_errno is set.
3775 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3776 const struct rte_flow_action *action,
3777 const uint64_t item_flags,
3778 struct rte_flow_error *error)
3783 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3785 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3786 MLX5_FLOW_LAYER_INNER_L4_TCP :
3787 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3788 if (!(item_flags & layer))
3789 return rte_flow_error_set(error, EINVAL,
3790 RTE_FLOW_ERROR_TYPE_ACTION,
3791 NULL, "no TCP item in"
3793 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3794 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3795 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3796 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3797 return rte_flow_error_set(error, EINVAL,
3798 RTE_FLOW_ERROR_TYPE_ACTION,
3800 "cannot decrease and increase"
3801 " TCP sequence number"
3802 " at the same time");
3808 * Validate the modify-header actions of increment/decrement
3809 * TCP Acknowledgment number.
3811 * @param[in] action_flags
3812 * Holds the actions detected until now.
3814 * Pointer to the modify action.
3815 * @param[in] item_flags
3816 * Holds the items detected.
3818 * Pointer to error structure.
3821 * 0 on success, a negative errno value otherwise and rte_errno is set.
3824 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3825 const struct rte_flow_action *action,
3826 const uint64_t item_flags,
3827 struct rte_flow_error *error)
3832 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3834 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3835 MLX5_FLOW_LAYER_INNER_L4_TCP :
3836 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3837 if (!(item_flags & layer))
3838 return rte_flow_error_set(error, EINVAL,
3839 RTE_FLOW_ERROR_TYPE_ACTION,
3840 NULL, "no TCP item in"
3842 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3843 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3844 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3845 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3846 return rte_flow_error_set(error, EINVAL,
3847 RTE_FLOW_ERROR_TYPE_ACTION,
3849 "cannot decrease and increase"
3850 " TCP acknowledgment number"
3851 " at the same time");
3857 * Validate the modify-header TTL actions.
3859 * @param[in] action_flags
3860 * Holds the actions detected until now.
3862 * Pointer to the modify action.
3863 * @param[in] item_flags
3864 * Holds the items detected.
3866 * Pointer to error structure.
3869 * 0 on success, a negative errno value otherwise and rte_errno is set.
3872 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3873 const struct rte_flow_action *action,
3874 const uint64_t item_flags,
3875 struct rte_flow_error *error)
3880 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3882 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3883 MLX5_FLOW_LAYER_INNER_L3 :
3884 MLX5_FLOW_LAYER_OUTER_L3;
3885 if (!(item_flags & layer))
3886 return rte_flow_error_set(error, EINVAL,
3887 RTE_FLOW_ERROR_TYPE_ACTION,
3889 "no IP protocol in pattern");
3895 * Validate jump action.
3898 * Pointer to the jump action.
3899 * @param[in] action_flags
3900 * Holds the actions detected until now.
3901 * @param[in] attributes
3902 * Pointer to flow attributes
3903 * @param[in] external
3904 * Action belongs to flow rule created by request external to PMD.
3906 * Pointer to error structure.
3909 * 0 on success, a negative errno value otherwise and rte_errno is set.
3912 flow_dv_validate_action_jump(struct rte_eth_dev *dev,
3913 const struct mlx5_flow_tunnel *tunnel,
3914 const struct rte_flow_action *action,
3915 uint64_t action_flags,
3916 const struct rte_flow_attr *attributes,
3917 bool external, struct rte_flow_error *error)
3919 uint32_t target_group, table;
3921 struct flow_grp_info grp_info = {
3922 .external = !!external,
3923 .transfer = !!attributes->transfer,
3927 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3928 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3929 return rte_flow_error_set(error, EINVAL,
3930 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3931 "can't have 2 fate actions in"
3933 if (action_flags & MLX5_FLOW_ACTION_METER)
3934 return rte_flow_error_set(error, ENOTSUP,
3935 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3936 "jump with meter not support");
3937 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3938 return rte_flow_error_set(error, EINVAL,
3939 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3940 "E-Switch mirroring can't support"
3941 " Sample action and jump action in"
3944 return rte_flow_error_set(error, EINVAL,
3945 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3946 NULL, "action configuration not set");
3948 ((const struct rte_flow_action_jump *)action->conf)->group;
3949 ret = mlx5_flow_group_to_table(dev, tunnel, target_group, &table,
3953 if (attributes->group == target_group &&
3954 !(action_flags & (MLX5_FLOW_ACTION_TUNNEL_SET |
3955 MLX5_FLOW_ACTION_TUNNEL_MATCH)))
3956 return rte_flow_error_set(error, EINVAL,
3957 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3958 "target group must be other than"
3959 " the current flow group");
3964 * Validate the port_id action.
3967 * Pointer to rte_eth_dev structure.
3968 * @param[in] action_flags
3969 * Bit-fields that holds the actions detected until now.
3971 * Port_id RTE action structure.
3973 * Attributes of flow that includes this action.
3975 * Pointer to error structure.
3978 * 0 on success, a negative errno value otherwise and rte_errno is set.
3981 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
3982 uint64_t action_flags,
3983 const struct rte_flow_action *action,
3984 const struct rte_flow_attr *attr,
3985 struct rte_flow_error *error)
3987 const struct rte_flow_action_port_id *port_id;
3988 struct mlx5_priv *act_priv;
3989 struct mlx5_priv *dev_priv;
3992 if (!attr->transfer)
3993 return rte_flow_error_set(error, ENOTSUP,
3994 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3996 "port id action is valid in transfer"
3998 if (!action || !action->conf)
3999 return rte_flow_error_set(error, ENOTSUP,
4000 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4002 "port id action parameters must be"
4004 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
4005 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
4006 return rte_flow_error_set(error, EINVAL,
4007 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4008 "can have only one fate actions in"
4010 dev_priv = mlx5_dev_to_eswitch_info(dev);
4012 return rte_flow_error_set(error, rte_errno,
4013 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4015 "failed to obtain E-Switch info");
4016 port_id = action->conf;
4017 port = port_id->original ? dev->data->port_id : port_id->id;
4018 act_priv = mlx5_port_to_eswitch_info(port, false);
4020 return rte_flow_error_set
4022 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
4023 "failed to obtain E-Switch port id for port");
4024 if (act_priv->domain_id != dev_priv->domain_id)
4025 return rte_flow_error_set
4027 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4028 "port does not belong to"
4029 " E-Switch being configured");
4034 * Get the maximum number of modify header actions.
4037 * Pointer to rte_eth_dev structure.
4039 * Flags bits to check if root level.
4042 * Max number of modify header actions device can support.
4044 static inline unsigned int
4045 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
4049 * There's no way to directly query the max capacity from FW.
4050 * The maximal value on root table should be assumed to be supported.
4052 if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
4053 return MLX5_MAX_MODIFY_NUM;
4055 return MLX5_ROOT_TBL_MODIFY_NUM;
4059 * Validate the meter action.
4062 * Pointer to rte_eth_dev structure.
4063 * @param[in] action_flags
4064 * Bit-fields that holds the actions detected until now.
4066 * Pointer to the meter action.
4068 * Attributes of flow that includes this action.
4070 * Pointer to error structure.
4073 * 0 on success, a negative errno value otherwise and rte_ernno is set.
4076 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
4077 uint64_t action_flags,
4078 const struct rte_flow_action *action,
4079 const struct rte_flow_attr *attr,
4080 struct rte_flow_error *error)
4082 struct mlx5_priv *priv = dev->data->dev_private;
4083 const struct rte_flow_action_meter *am = action->conf;
4084 struct mlx5_flow_meter *fm;
4087 return rte_flow_error_set(error, EINVAL,
4088 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4089 "meter action conf is NULL");
4091 if (action_flags & MLX5_FLOW_ACTION_METER)
4092 return rte_flow_error_set(error, ENOTSUP,
4093 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4094 "meter chaining not support");
4095 if (action_flags & MLX5_FLOW_ACTION_JUMP)
4096 return rte_flow_error_set(error, ENOTSUP,
4097 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4098 "meter with jump not support");
4100 return rte_flow_error_set(error, ENOTSUP,
4101 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4103 "meter action not supported");
4104 fm = mlx5_flow_meter_find(priv, am->mtr_id);
4106 return rte_flow_error_set(error, EINVAL,
4107 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4109 if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
4110 (!fm->ingress && !attr->ingress && attr->egress) ||
4111 (!fm->egress && !attr->egress && attr->ingress))))
4112 return rte_flow_error_set(error, EINVAL,
4113 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4114 "Flow attributes are either invalid "
4115 "or have a conflict with current "
4116 "meter attributes");
4121 * Validate the age action.
4123 * @param[in] action_flags
4124 * Holds the actions detected until now.
4126 * Pointer to the age action.
4128 * Pointer to the Ethernet device structure.
4130 * Pointer to error structure.
4133 * 0 on success, a negative errno value otherwise and rte_errno is set.
4136 flow_dv_validate_action_age(uint64_t action_flags,
4137 const struct rte_flow_action *action,
4138 struct rte_eth_dev *dev,
4139 struct rte_flow_error *error)
4141 struct mlx5_priv *priv = dev->data->dev_private;
4142 const struct rte_flow_action_age *age = action->conf;
4144 if (!priv->config.devx || priv->sh->cmng.counter_fallback)
4145 return rte_flow_error_set(error, ENOTSUP,
4146 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4148 "age action not supported");
4149 if (!(action->conf))
4150 return rte_flow_error_set(error, EINVAL,
4151 RTE_FLOW_ERROR_TYPE_ACTION, action,
4152 "configuration cannot be null");
4153 if (!(age->timeout))
4154 return rte_flow_error_set(error, EINVAL,
4155 RTE_FLOW_ERROR_TYPE_ACTION, action,
4156 "invalid timeout value 0");
4157 if (action_flags & MLX5_FLOW_ACTION_AGE)
4158 return rte_flow_error_set(error, EINVAL,
4159 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4160 "duplicate age actions set");
4165 * Validate the modify-header IPv4 DSCP actions.
4167 * @param[in] action_flags
4168 * Holds the actions detected until now.
4170 * Pointer to the modify action.
4171 * @param[in] item_flags
4172 * Holds the items detected.
4174 * Pointer to error structure.
4177 * 0 on success, a negative errno value otherwise and rte_errno is set.
4180 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
4181 const struct rte_flow_action *action,
4182 const uint64_t item_flags,
4183 struct rte_flow_error *error)
4187 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4189 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
4190 return rte_flow_error_set(error, EINVAL,
4191 RTE_FLOW_ERROR_TYPE_ACTION,
4193 "no ipv4 item in pattern");
4199 * Validate the modify-header IPv6 DSCP actions.
4201 * @param[in] action_flags
4202 * Holds the actions detected until now.
4204 * Pointer to the modify action.
4205 * @param[in] item_flags
4206 * Holds the items detected.
4208 * Pointer to error structure.
4211 * 0 on success, a negative errno value otherwise and rte_errno is set.
4214 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
4215 const struct rte_flow_action *action,
4216 const uint64_t item_flags,
4217 struct rte_flow_error *error)
4221 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4223 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
4224 return rte_flow_error_set(error, EINVAL,
4225 RTE_FLOW_ERROR_TYPE_ACTION,
4227 "no ipv6 item in pattern");
4233 * Match modify-header resource.
4236 * Pointer to the hash list.
4238 * Pointer to exist resource entry object.
4240 * Key of the new entry.
4242 * Pointer to new modify-header resource.
4245 * 0 on matching, non-zero otherwise.
4248 flow_dv_modify_match_cb(struct mlx5_hlist *list __rte_unused,
4249 struct mlx5_hlist_entry *entry,
4250 uint64_t key __rte_unused, void *cb_ctx)
4252 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4253 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
4254 struct mlx5_flow_dv_modify_hdr_resource *resource =
4255 container_of(entry, typeof(*resource), entry);
4256 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
4258 key_len += ref->actions_num * sizeof(ref->actions[0]);
4259 return ref->actions_num != resource->actions_num ||
4260 memcmp(&ref->ft_type, &resource->ft_type, key_len);
4263 struct mlx5_hlist_entry *
4264 flow_dv_modify_create_cb(struct mlx5_hlist *list, uint64_t key __rte_unused,
4267 struct mlx5_dev_ctx_shared *sh = list->ctx;
4268 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4269 struct mlx5dv_dr_domain *ns;
4270 struct mlx5_flow_dv_modify_hdr_resource *entry;
4271 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
4273 uint32_t data_len = ref->actions_num * sizeof(ref->actions[0]);
4274 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
4276 entry = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*entry) + data_len, 0,
4279 rte_flow_error_set(ctx->error, ENOMEM,
4280 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4281 "cannot allocate resource memory");
4284 rte_memcpy(&entry->ft_type,
4285 RTE_PTR_ADD(ref, offsetof(typeof(*ref), ft_type)),
4286 key_len + data_len);
4287 if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
4288 ns = sh->fdb_domain;
4289 else if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
4293 ret = mlx5_flow_os_create_flow_action_modify_header
4294 (sh->ctx, ns, entry,
4295 data_len, &entry->action);
4298 rte_flow_error_set(ctx->error, ENOMEM,
4299 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4300 NULL, "cannot create modification action");
4303 return &entry->entry;
4307 * Validate the sample action.
4309 * @param[in] action_flags
4310 * Holds the actions detected until now.
4312 * Pointer to the sample action.
4314 * Pointer to the Ethernet device structure.
4316 * Attributes of flow that includes this action.
4318 * Pointer to error structure.
4321 * 0 on success, a negative errno value otherwise and rte_errno is set.
4324 flow_dv_validate_action_sample(uint64_t action_flags,
4325 const struct rte_flow_action *action,
4326 struct rte_eth_dev *dev,
4327 const struct rte_flow_attr *attr,
4328 struct rte_flow_error *error)
4330 struct mlx5_priv *priv = dev->data->dev_private;
4331 struct mlx5_dev_config *dev_conf = &priv->config;
4332 const struct rte_flow_action_sample *sample = action->conf;
4333 const struct rte_flow_action *act;
4334 uint64_t sub_action_flags = 0;
4335 uint16_t queue_index = 0xFFFF;
4341 return rte_flow_error_set(error, EINVAL,
4342 RTE_FLOW_ERROR_TYPE_ACTION, action,
4343 "configuration cannot be NULL");
4344 if (sample->ratio == 0)
4345 return rte_flow_error_set(error, EINVAL,
4346 RTE_FLOW_ERROR_TYPE_ACTION, action,
4347 "ratio value starts from 1");
4348 if (!priv->config.devx || (sample->ratio > 0 && !priv->sampler_en))
4349 return rte_flow_error_set(error, ENOTSUP,
4350 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4352 "sample action not supported");
4353 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
4354 return rte_flow_error_set(error, EINVAL,
4355 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4356 "Multiple sample actions not "
4358 if (action_flags & MLX5_FLOW_ACTION_METER)
4359 return rte_flow_error_set(error, EINVAL,
4360 RTE_FLOW_ERROR_TYPE_ACTION, action,
4361 "wrong action order, meter should "
4362 "be after sample action");
4363 if (action_flags & MLX5_FLOW_ACTION_JUMP)
4364 return rte_flow_error_set(error, EINVAL,
4365 RTE_FLOW_ERROR_TYPE_ACTION, action,
4366 "wrong action order, jump should "
4367 "be after sample action");
4368 act = sample->actions;
4369 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
4370 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
4371 return rte_flow_error_set(error, ENOTSUP,
4372 RTE_FLOW_ERROR_TYPE_ACTION,
4373 act, "too many actions");
4374 switch (act->type) {
4375 case RTE_FLOW_ACTION_TYPE_QUEUE:
4376 ret = mlx5_flow_validate_action_queue(act,
4382 queue_index = ((const struct rte_flow_action_queue *)
4383 (act->conf))->index;
4384 sub_action_flags |= MLX5_FLOW_ACTION_QUEUE;
4387 case RTE_FLOW_ACTION_TYPE_MARK:
4388 ret = flow_dv_validate_action_mark(dev, act,
4393 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4394 sub_action_flags |= MLX5_FLOW_ACTION_MARK |
4395 MLX5_FLOW_ACTION_MARK_EXT;
4397 sub_action_flags |= MLX5_FLOW_ACTION_MARK;
4400 case RTE_FLOW_ACTION_TYPE_COUNT:
4401 ret = flow_dv_validate_action_count(dev, error);
4404 sub_action_flags |= MLX5_FLOW_ACTION_COUNT;
4407 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4408 ret = flow_dv_validate_action_port_id(dev,
4415 sub_action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4418 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4419 ret = flow_dv_validate_action_raw_encap_decap
4420 (dev, NULL, act->conf, attr, &sub_action_flags,
4427 return rte_flow_error_set(error, ENOTSUP,
4428 RTE_FLOW_ERROR_TYPE_ACTION,
4430 "Doesn't support optional "
4434 if (attr->ingress && !attr->transfer) {
4435 if (!(sub_action_flags & MLX5_FLOW_ACTION_QUEUE))
4436 return rte_flow_error_set(error, EINVAL,
4437 RTE_FLOW_ERROR_TYPE_ACTION,
4439 "Ingress must has a dest "
4440 "QUEUE for Sample");
4441 } else if (attr->egress && !attr->transfer) {
4442 return rte_flow_error_set(error, ENOTSUP,
4443 RTE_FLOW_ERROR_TYPE_ACTION,
4445 "Sample Only support Ingress "
4447 } else if (sample->actions->type != RTE_FLOW_ACTION_TYPE_END) {
4448 MLX5_ASSERT(attr->transfer);
4449 if (sample->ratio > 1)
4450 return rte_flow_error_set(error, ENOTSUP,
4451 RTE_FLOW_ERROR_TYPE_ACTION,
4453 "E-Switch doesn't support "
4454 "any optional action "
4457 if (sub_action_flags & MLX5_FLOW_ACTION_QUEUE)
4458 return rte_flow_error_set(error, ENOTSUP,
4459 RTE_FLOW_ERROR_TYPE_ACTION,
4461 "unsupported action QUEUE");
4462 if (!(sub_action_flags & MLX5_FLOW_ACTION_PORT_ID))
4463 return rte_flow_error_set(error, EINVAL,
4464 RTE_FLOW_ERROR_TYPE_ACTION,
4466 "E-Switch must has a dest "
4467 "port for mirroring");
4469 /* Continue validation for Xcap actions.*/
4470 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) &&
4471 (queue_index == 0xFFFF ||
4472 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
4473 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
4474 MLX5_FLOW_XCAP_ACTIONS)
4475 return rte_flow_error_set(error, ENOTSUP,
4476 RTE_FLOW_ERROR_TYPE_ACTION,
4477 NULL, "encap and decap "
4478 "combination aren't "
4480 if (!attr->transfer && attr->ingress && (sub_action_flags &
4481 MLX5_FLOW_ACTION_ENCAP))
4482 return rte_flow_error_set(error, ENOTSUP,
4483 RTE_FLOW_ERROR_TYPE_ACTION,
4484 NULL, "encap is not supported"
4485 " for ingress traffic");
4491 * Find existing modify-header resource or create and register a new one.
4493 * @param dev[in, out]
4494 * Pointer to rte_eth_dev structure.
4495 * @param[in, out] resource
4496 * Pointer to modify-header resource.
4497 * @parm[in, out] dev_flow
4498 * Pointer to the dev_flow.
4500 * pointer to error structure.
4503 * 0 on success otherwise -errno and errno is set.
4506 flow_dv_modify_hdr_resource_register
4507 (struct rte_eth_dev *dev,
4508 struct mlx5_flow_dv_modify_hdr_resource *resource,
4509 struct mlx5_flow *dev_flow,
4510 struct rte_flow_error *error)
4512 struct mlx5_priv *priv = dev->data->dev_private;
4513 struct mlx5_dev_ctx_shared *sh = priv->sh;
4514 uint32_t key_len = sizeof(*resource) -
4515 offsetof(typeof(*resource), ft_type) +
4516 resource->actions_num * sizeof(resource->actions[0]);
4517 struct mlx5_hlist_entry *entry;
4518 struct mlx5_flow_cb_ctx ctx = {
4523 resource->flags = dev_flow->dv.group ? 0 :
4524 MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4525 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
4527 return rte_flow_error_set(error, EOVERFLOW,
4528 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4529 "too many modify header items");
4530 resource->entry.key = __rte_raw_cksum(&resource->ft_type, key_len, 0);
4531 entry = mlx5_hlist_register(sh->modify_cmds, resource->entry.key, &ctx);
4534 resource = container_of(entry, typeof(*resource), entry);
4535 dev_flow->handle->dvh.modify_hdr = resource;
4540 * Get DV flow counter by index.
4543 * Pointer to the Ethernet device structure.
4545 * mlx5 flow counter index in the container.
4547 * mlx5 flow counter pool in the container,
4550 * Pointer to the counter, NULL otherwise.
4552 static struct mlx5_flow_counter *
4553 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
4555 struct mlx5_flow_counter_pool **ppool)
4557 struct mlx5_priv *priv = dev->data->dev_private;
4558 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4559 struct mlx5_flow_counter_pool *pool;
4561 /* Decrease to original index and clear shared bit. */
4562 idx = (idx - 1) & (MLX5_CNT_SHARED_OFFSET - 1);
4563 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cmng->n);
4564 pool = cmng->pools[idx / MLX5_COUNTERS_PER_POOL];
4568 return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
4572 * Check the devx counter belongs to the pool.
4575 * Pointer to the counter pool.
4577 * The counter devx ID.
4580 * True if counter belongs to the pool, false otherwise.
4583 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
4585 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
4586 MLX5_COUNTERS_PER_POOL;
4588 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
4594 * Get a pool by devx counter ID.
4597 * Pointer to the counter management.
4599 * The counter devx ID.
4602 * The counter pool pointer if exists, NULL otherwise,
4604 static struct mlx5_flow_counter_pool *
4605 flow_dv_find_pool_by_id(struct mlx5_flow_counter_mng *cmng, int id)
4608 struct mlx5_flow_counter_pool *pool = NULL;
4610 rte_spinlock_lock(&cmng->pool_update_sl);
4611 /* Check last used pool. */
4612 if (cmng->last_pool_idx != POOL_IDX_INVALID &&
4613 flow_dv_is_counter_in_pool(cmng->pools[cmng->last_pool_idx], id)) {
4614 pool = cmng->pools[cmng->last_pool_idx];
4617 /* ID out of range means no suitable pool in the container. */
4618 if (id > cmng->max_id || id < cmng->min_id)
4621 * Find the pool from the end of the container, since mostly counter
4622 * ID is sequence increasing, and the last pool should be the needed
4627 struct mlx5_flow_counter_pool *pool_tmp = cmng->pools[i];
4629 if (flow_dv_is_counter_in_pool(pool_tmp, id)) {
4635 rte_spinlock_unlock(&cmng->pool_update_sl);
4640 * Resize a counter container.
4643 * Pointer to the Ethernet device structure.
4646 * 0 on success, otherwise negative errno value and rte_errno is set.
4649 flow_dv_container_resize(struct rte_eth_dev *dev)
4651 struct mlx5_priv *priv = dev->data->dev_private;
4652 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4653 void *old_pools = cmng->pools;
4654 uint32_t resize = cmng->n + MLX5_CNT_CONTAINER_RESIZE;
4655 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4656 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
4663 memcpy(pools, old_pools, cmng->n *
4664 sizeof(struct mlx5_flow_counter_pool *));
4666 cmng->pools = pools;
4668 mlx5_free(old_pools);
4673 * Query a devx flow counter.
4676 * Pointer to the Ethernet device structure.
4678 * Index to the flow counter.
4680 * The statistics value of packets.
4682 * The statistics value of bytes.
4685 * 0 on success, otherwise a negative errno value and rte_errno is set.
4688 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4691 struct mlx5_priv *priv = dev->data->dev_private;
4692 struct mlx5_flow_counter_pool *pool = NULL;
4693 struct mlx5_flow_counter *cnt;
4696 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4698 if (priv->sh->cmng.counter_fallback)
4699 return mlx5_devx_cmd_flow_counter_query(cnt->dcs_when_active, 0,
4700 0, pkts, bytes, 0, NULL, NULL, 0);
4701 rte_spinlock_lock(&pool->sl);
4706 offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
4707 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4708 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4710 rte_spinlock_unlock(&pool->sl);
4715 * Create and initialize a new counter pool.
4718 * Pointer to the Ethernet device structure.
4720 * The devX counter handle.
4722 * Whether the pool is for counter that was allocated for aging.
4723 * @param[in/out] cont_cur
4724 * Pointer to the container pointer, it will be update in pool resize.
4727 * The pool container pointer on success, NULL otherwise and rte_errno is set.
4729 static struct mlx5_flow_counter_pool *
4730 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4733 struct mlx5_priv *priv = dev->data->dev_private;
4734 struct mlx5_flow_counter_pool *pool;
4735 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4736 bool fallback = priv->sh->cmng.counter_fallback;
4737 uint32_t size = sizeof(*pool);
4739 size += MLX5_COUNTERS_PER_POOL * MLX5_CNT_SIZE;
4740 size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * MLX5_AGE_SIZE);
4741 pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
4747 pool->is_aged = !!age;
4748 pool->query_gen = 0;
4749 pool->min_dcs = dcs;
4750 rte_spinlock_init(&pool->sl);
4751 rte_spinlock_init(&pool->csl);
4752 TAILQ_INIT(&pool->counters[0]);
4753 TAILQ_INIT(&pool->counters[1]);
4754 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
4755 rte_spinlock_lock(&cmng->pool_update_sl);
4756 pool->index = cmng->n_valid;
4757 if (pool->index == cmng->n && flow_dv_container_resize(dev)) {
4759 rte_spinlock_unlock(&cmng->pool_update_sl);
4762 cmng->pools[pool->index] = pool;
4764 if (unlikely(fallback)) {
4765 int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
4767 if (base < cmng->min_id)
4768 cmng->min_id = base;
4769 if (base > cmng->max_id)
4770 cmng->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
4771 cmng->last_pool_idx = pool->index;
4773 rte_spinlock_unlock(&cmng->pool_update_sl);
4778 * Prepare a new counter and/or a new counter pool.
4781 * Pointer to the Ethernet device structure.
4782 * @param[out] cnt_free
4783 * Where to put the pointer of a new counter.
4785 * Whether the pool is for counter that was allocated for aging.
4788 * The counter pool pointer and @p cnt_free is set on success,
4789 * NULL otherwise and rte_errno is set.
4791 static struct mlx5_flow_counter_pool *
4792 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4793 struct mlx5_flow_counter **cnt_free,
4796 struct mlx5_priv *priv = dev->data->dev_private;
4797 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4798 struct mlx5_flow_counter_pool *pool;
4799 struct mlx5_counters tmp_tq;
4800 struct mlx5_devx_obj *dcs = NULL;
4801 struct mlx5_flow_counter *cnt;
4802 enum mlx5_counter_type cnt_type =
4803 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
4804 bool fallback = priv->sh->cmng.counter_fallback;
4808 /* bulk_bitmap must be 0 for single counter allocation. */
4809 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4812 pool = flow_dv_find_pool_by_id(cmng, dcs->id);
4814 pool = flow_dv_pool_create(dev, dcs, age);
4816 mlx5_devx_cmd_destroy(dcs);
4820 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4821 cnt = MLX5_POOL_GET_CNT(pool, i);
4823 cnt->dcs_when_free = dcs;
4827 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4829 rte_errno = ENODATA;
4832 pool = flow_dv_pool_create(dev, dcs, age);
4834 mlx5_devx_cmd_destroy(dcs);
4837 TAILQ_INIT(&tmp_tq);
4838 for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
4839 cnt = MLX5_POOL_GET_CNT(pool, i);
4841 TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
4843 rte_spinlock_lock(&cmng->csl[cnt_type]);
4844 TAILQ_CONCAT(&cmng->counters[cnt_type], &tmp_tq, next);
4845 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4846 *cnt_free = MLX5_POOL_GET_CNT(pool, 0);
4847 (*cnt_free)->pool = pool;
4852 * Allocate a flow counter.
4855 * Pointer to the Ethernet device structure.
4857 * Whether the counter was allocated for aging.
4860 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4863 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t age)
4865 struct mlx5_priv *priv = dev->data->dev_private;
4866 struct mlx5_flow_counter_pool *pool = NULL;
4867 struct mlx5_flow_counter *cnt_free = NULL;
4868 bool fallback = priv->sh->cmng.counter_fallback;
4869 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4870 enum mlx5_counter_type cnt_type =
4871 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
4874 if (!priv->config.devx) {
4875 rte_errno = ENOTSUP;
4878 /* Get free counters from container. */
4879 rte_spinlock_lock(&cmng->csl[cnt_type]);
4880 cnt_free = TAILQ_FIRST(&cmng->counters[cnt_type]);
4882 TAILQ_REMOVE(&cmng->counters[cnt_type], cnt_free, next);
4883 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4884 if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free, age))
4886 pool = cnt_free->pool;
4888 cnt_free->dcs_when_active = cnt_free->dcs_when_free;
4889 /* Create a DV counter action only in the first time usage. */
4890 if (!cnt_free->action) {
4892 struct mlx5_devx_obj *dcs;
4896 offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
4897 dcs = pool->min_dcs;
4900 dcs = cnt_free->dcs_when_free;
4902 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
4909 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4910 MLX5_CNT_ARRAY_IDX(pool, cnt_free));
4911 /* Update the counter reset values. */
4912 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4915 if (!fallback && !priv->sh->cmng.query_thread_on)
4916 /* Start the asynchronous batch query by the host thread. */
4917 mlx5_set_query_alarm(priv->sh);
4921 cnt_free->pool = pool;
4923 cnt_free->dcs_when_free = cnt_free->dcs_when_active;
4924 rte_spinlock_lock(&cmng->csl[cnt_type]);
4925 TAILQ_INSERT_TAIL(&cmng->counters[cnt_type], cnt_free, next);
4926 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4932 * Allocate a shared flow counter.
4935 * Pointer to the shared counter configuration.
4937 * Pointer to save the allocated counter index.
4940 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4944 flow_dv_counter_alloc_shared_cb(void *ctx, union mlx5_l3t_data *data)
4946 struct mlx5_shared_counter_conf *conf = ctx;
4947 struct rte_eth_dev *dev = conf->dev;
4948 struct mlx5_flow_counter *cnt;
4950 data->dword = flow_dv_counter_alloc(dev, 0);
4951 data->dword |= MLX5_CNT_SHARED_OFFSET;
4952 cnt = flow_dv_counter_get_by_idx(dev, data->dword, NULL);
4953 cnt->shared_info.id = conf->id;
4958 * Get a shared flow counter.
4961 * Pointer to the Ethernet device structure.
4963 * Counter identifier.
4966 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4969 flow_dv_counter_get_shared(struct rte_eth_dev *dev, uint32_t id)
4971 struct mlx5_priv *priv = dev->data->dev_private;
4972 struct mlx5_shared_counter_conf conf = {
4976 union mlx5_l3t_data data = {
4980 mlx5_l3t_prepare_entry(priv->sh->cnt_id_tbl, id, &data,
4981 flow_dv_counter_alloc_shared_cb, &conf);
4986 * Get age param from counter index.
4989 * Pointer to the Ethernet device structure.
4990 * @param[in] counter
4991 * Index to the counter handler.
4994 * The aging parameter specified for the counter index.
4996 static struct mlx5_age_param*
4997 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
5000 struct mlx5_flow_counter *cnt;
5001 struct mlx5_flow_counter_pool *pool = NULL;
5003 flow_dv_counter_get_by_idx(dev, counter, &pool);
5004 counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
5005 cnt = MLX5_POOL_GET_CNT(pool, counter);
5006 return MLX5_CNT_TO_AGE(cnt);
5010 * Remove a flow counter from aged counter list.
5013 * Pointer to the Ethernet device structure.
5014 * @param[in] counter
5015 * Index to the counter handler.
5017 * Pointer to the counter handler.
5020 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
5021 uint32_t counter, struct mlx5_flow_counter *cnt)
5023 struct mlx5_age_info *age_info;
5024 struct mlx5_age_param *age_param;
5025 struct mlx5_priv *priv = dev->data->dev_private;
5026 uint16_t expected = AGE_CANDIDATE;
5028 age_info = GET_PORT_AGE_INFO(priv);
5029 age_param = flow_dv_counter_idx_get_age(dev, counter);
5030 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
5031 AGE_FREE, false, __ATOMIC_RELAXED,
5032 __ATOMIC_RELAXED)) {
5034 * We need the lock even it is age timeout,
5035 * since counter may still in process.
5037 rte_spinlock_lock(&age_info->aged_sl);
5038 TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
5039 rte_spinlock_unlock(&age_info->aged_sl);
5040 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
5045 * Release a flow counter.
5048 * Pointer to the Ethernet device structure.
5049 * @param[in] counter
5050 * Index to the counter handler.
5053 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
5055 struct mlx5_priv *priv = dev->data->dev_private;
5056 struct mlx5_flow_counter_pool *pool = NULL;
5057 struct mlx5_flow_counter *cnt;
5058 enum mlx5_counter_type cnt_type;
5062 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
5064 if (IS_SHARED_CNT(counter) &&
5065 mlx5_l3t_clear_entry(priv->sh->cnt_id_tbl, cnt->shared_info.id))
5068 flow_dv_counter_remove_from_age(dev, counter, cnt);
5071 * Put the counter back to list to be updated in none fallback mode.
5072 * Currently, we are using two list alternately, while one is in query,
5073 * add the freed counter to the other list based on the pool query_gen
5074 * value. After query finishes, add counter the list to the global
5075 * container counter list. The list changes while query starts. In
5076 * this case, lock will not be needed as query callback and release
5077 * function both operate with the different list.
5080 if (!priv->sh->cmng.counter_fallback) {
5081 rte_spinlock_lock(&pool->csl);
5082 TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
5083 rte_spinlock_unlock(&pool->csl);
5085 cnt->dcs_when_free = cnt->dcs_when_active;
5086 cnt_type = pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
5087 MLX5_COUNTER_TYPE_ORIGIN;
5088 rte_spinlock_lock(&priv->sh->cmng.csl[cnt_type]);
5089 TAILQ_INSERT_TAIL(&priv->sh->cmng.counters[cnt_type],
5091 rte_spinlock_unlock(&priv->sh->cmng.csl[cnt_type]);
5096 * Verify the @p attributes will be correctly understood by the NIC and store
5097 * them in the @p flow if everything is correct.
5100 * Pointer to dev struct.
5101 * @param[in] attributes
5102 * Pointer to flow attributes
5103 * @param[in] external
5104 * This flow rule is created by request external to PMD.
5106 * Pointer to error structure.
5109 * - 0 on success and non root table.
5110 * - 1 on success and root table.
5111 * - a negative errno value otherwise and rte_errno is set.
5114 flow_dv_validate_attributes(struct rte_eth_dev *dev,
5115 const struct mlx5_flow_tunnel *tunnel,
5116 const struct rte_flow_attr *attributes,
5117 struct flow_grp_info grp_info,
5118 struct rte_flow_error *error)
5120 struct mlx5_priv *priv = dev->data->dev_private;
5121 uint32_t priority_max = priv->config.flow_prio - 1;
5124 #ifndef HAVE_MLX5DV_DR
5125 RTE_SET_USED(tunnel);
5126 RTE_SET_USED(grp_info);
5127 if (attributes->group)
5128 return rte_flow_error_set(error, ENOTSUP,
5129 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5131 "groups are not supported");
5135 ret = mlx5_flow_group_to_table(dev, tunnel, attributes->group, &table,
5140 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
5142 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
5143 attributes->priority >= priority_max)
5144 return rte_flow_error_set(error, ENOTSUP,
5145 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
5147 "priority out of range");
5148 if (attributes->transfer) {
5149 if (!priv->config.dv_esw_en)
5150 return rte_flow_error_set
5152 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5153 "E-Switch dr is not supported");
5154 if (!(priv->representor || priv->master))
5155 return rte_flow_error_set
5156 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5157 NULL, "E-Switch configuration can only be"
5158 " done by a master or a representor device");
5159 if (attributes->egress)
5160 return rte_flow_error_set
5162 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
5163 "egress is not supported");
5165 if (!(attributes->egress ^ attributes->ingress))
5166 return rte_flow_error_set(error, ENOTSUP,
5167 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
5168 "must specify exactly one of "
5169 "ingress or egress");
5174 * Internal validation function. For validating both actions and items.
5177 * Pointer to the rte_eth_dev structure.
5179 * Pointer to the flow attributes.
5181 * Pointer to the list of items.
5182 * @param[in] actions
5183 * Pointer to the list of actions.
5184 * @param[in] external
5185 * This flow rule is created by request external to PMD.
5186 * @param[in] hairpin
5187 * Number of hairpin TX actions, 0 means classic flow.
5189 * Pointer to the error structure.
5192 * 0 on success, a negative errno value otherwise and rte_errno is set.
5195 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
5196 const struct rte_flow_item items[],
5197 const struct rte_flow_action actions[],
5198 bool external, int hairpin, struct rte_flow_error *error)
5201 uint64_t action_flags = 0;
5202 uint64_t item_flags = 0;
5203 uint64_t last_item = 0;
5204 uint8_t next_protocol = 0xff;
5205 uint16_t ether_type = 0;
5207 uint8_t item_ipv6_proto = 0;
5208 const struct rte_flow_item *gre_item = NULL;
5209 const struct rte_flow_action_raw_decap *decap;
5210 const struct rte_flow_action_raw_encap *encap;
5211 const struct rte_flow_action_rss *rss;
5212 const struct rte_flow_item_tcp nic_tcp_mask = {
5215 .src_port = RTE_BE16(UINT16_MAX),
5216 .dst_port = RTE_BE16(UINT16_MAX),
5219 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
5222 "\xff\xff\xff\xff\xff\xff\xff\xff"
5223 "\xff\xff\xff\xff\xff\xff\xff\xff",
5225 "\xff\xff\xff\xff\xff\xff\xff\xff"
5226 "\xff\xff\xff\xff\xff\xff\xff\xff",
5227 .vtc_flow = RTE_BE32(0xffffffff),
5233 const struct rte_flow_item_ecpri nic_ecpri_mask = {
5237 RTE_BE32(((const struct rte_ecpri_common_hdr) {
5241 .dummy[0] = 0xffffffff,
5244 struct mlx5_priv *priv = dev->data->dev_private;
5245 struct mlx5_dev_config *dev_conf = &priv->config;
5246 uint16_t queue_index = 0xFFFF;
5247 const struct rte_flow_item_vlan *vlan_m = NULL;
5248 int16_t rw_act_num = 0;
5250 const struct mlx5_flow_tunnel *tunnel;
5251 struct flow_grp_info grp_info = {
5252 .external = !!external,
5253 .transfer = !!attr->transfer,
5254 .fdb_def_rule = !!priv->fdb_def_rule,
5256 const struct rte_eth_hairpin_conf *conf;
5260 if (is_flow_tunnel_match_rule(dev, attr, items, actions)) {
5261 tunnel = flow_items_to_tunnel(items);
5262 action_flags |= MLX5_FLOW_ACTION_TUNNEL_MATCH |
5263 MLX5_FLOW_ACTION_DECAP;
5264 } else if (is_flow_tunnel_steer_rule(dev, attr, items, actions)) {
5265 tunnel = flow_actions_to_tunnel(actions);
5266 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
5270 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
5271 (dev, tunnel, attr, items, actions);
5272 ret = flow_dv_validate_attributes(dev, tunnel, attr, grp_info, error);
5275 is_root = (uint64_t)ret;
5276 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5277 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
5278 int type = items->type;
5280 if (!mlx5_flow_os_item_supported(type))
5281 return rte_flow_error_set(error, ENOTSUP,
5282 RTE_FLOW_ERROR_TYPE_ITEM,
5283 NULL, "item not supported");
5285 case MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL:
5286 if (items[0].type != (typeof(items[0].type))
5287 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL)
5288 return rte_flow_error_set
5290 RTE_FLOW_ERROR_TYPE_ITEM,
5291 NULL, "MLX5 private items "
5292 "must be the first");
5294 case RTE_FLOW_ITEM_TYPE_VOID:
5296 case RTE_FLOW_ITEM_TYPE_PORT_ID:
5297 ret = flow_dv_validate_item_port_id
5298 (dev, items, attr, item_flags, error);
5301 last_item = MLX5_FLOW_ITEM_PORT_ID;
5303 case RTE_FLOW_ITEM_TYPE_ETH:
5304 ret = mlx5_flow_validate_item_eth(items, item_flags,
5308 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
5309 MLX5_FLOW_LAYER_OUTER_L2;
5310 if (items->mask != NULL && items->spec != NULL) {
5312 ((const struct rte_flow_item_eth *)
5315 ((const struct rte_flow_item_eth *)
5317 ether_type = rte_be_to_cpu_16(ether_type);
5322 case RTE_FLOW_ITEM_TYPE_VLAN:
5323 ret = flow_dv_validate_item_vlan(items, item_flags,
5327 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
5328 MLX5_FLOW_LAYER_OUTER_VLAN;
5329 if (items->mask != NULL && items->spec != NULL) {
5331 ((const struct rte_flow_item_vlan *)
5332 items->spec)->inner_type;
5334 ((const struct rte_flow_item_vlan *)
5335 items->mask)->inner_type;
5336 ether_type = rte_be_to_cpu_16(ether_type);
5340 /* Store outer VLAN mask for of_push_vlan action. */
5342 vlan_m = items->mask;
5344 case RTE_FLOW_ITEM_TYPE_IPV4:
5345 mlx5_flow_tunnel_ip_check(items, next_protocol,
5346 &item_flags, &tunnel);
5347 ret = flow_dv_validate_item_ipv4(items, item_flags,
5348 last_item, ether_type,
5352 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
5353 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
5354 if (items->mask != NULL &&
5355 ((const struct rte_flow_item_ipv4 *)
5356 items->mask)->hdr.next_proto_id) {
5358 ((const struct rte_flow_item_ipv4 *)
5359 (items->spec))->hdr.next_proto_id;
5361 ((const struct rte_flow_item_ipv4 *)
5362 (items->mask))->hdr.next_proto_id;
5364 /* Reset for inner layer. */
5365 next_protocol = 0xff;
5368 case RTE_FLOW_ITEM_TYPE_IPV6:
5369 mlx5_flow_tunnel_ip_check(items, next_protocol,
5370 &item_flags, &tunnel);
5371 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
5378 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
5379 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
5380 if (items->mask != NULL &&
5381 ((const struct rte_flow_item_ipv6 *)
5382 items->mask)->hdr.proto) {
5384 ((const struct rte_flow_item_ipv6 *)
5385 items->spec)->hdr.proto;
5387 ((const struct rte_flow_item_ipv6 *)
5388 items->spec)->hdr.proto;
5390 ((const struct rte_flow_item_ipv6 *)
5391 items->mask)->hdr.proto;
5393 /* Reset for inner layer. */
5394 next_protocol = 0xff;
5397 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
5398 ret = flow_dv_validate_item_ipv6_frag_ext(items,
5403 last_item = tunnel ?
5404 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
5405 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
5406 if (items->mask != NULL &&
5407 ((const struct rte_flow_item_ipv6_frag_ext *)
5408 items->mask)->hdr.next_header) {
5410 ((const struct rte_flow_item_ipv6_frag_ext *)
5411 items->spec)->hdr.next_header;
5413 ((const struct rte_flow_item_ipv6_frag_ext *)
5414 items->mask)->hdr.next_header;
5416 /* Reset for inner layer. */
5417 next_protocol = 0xff;
5420 case RTE_FLOW_ITEM_TYPE_TCP:
5421 ret = mlx5_flow_validate_item_tcp
5428 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
5429 MLX5_FLOW_LAYER_OUTER_L4_TCP;
5431 case RTE_FLOW_ITEM_TYPE_UDP:
5432 ret = mlx5_flow_validate_item_udp(items, item_flags,
5437 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
5438 MLX5_FLOW_LAYER_OUTER_L4_UDP;
5440 case RTE_FLOW_ITEM_TYPE_GRE:
5441 ret = mlx5_flow_validate_item_gre(items, item_flags,
5442 next_protocol, error);
5446 last_item = MLX5_FLOW_LAYER_GRE;
5448 case RTE_FLOW_ITEM_TYPE_NVGRE:
5449 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
5454 last_item = MLX5_FLOW_LAYER_NVGRE;
5456 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5457 ret = mlx5_flow_validate_item_gre_key
5458 (items, item_flags, gre_item, error);
5461 last_item = MLX5_FLOW_LAYER_GRE_KEY;
5463 case RTE_FLOW_ITEM_TYPE_VXLAN:
5464 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
5468 last_item = MLX5_FLOW_LAYER_VXLAN;
5470 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5471 ret = mlx5_flow_validate_item_vxlan_gpe(items,
5476 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5478 case RTE_FLOW_ITEM_TYPE_GENEVE:
5479 ret = mlx5_flow_validate_item_geneve(items,
5484 last_item = MLX5_FLOW_LAYER_GENEVE;
5486 case RTE_FLOW_ITEM_TYPE_MPLS:
5487 ret = mlx5_flow_validate_item_mpls(dev, items,
5492 last_item = MLX5_FLOW_LAYER_MPLS;
5495 case RTE_FLOW_ITEM_TYPE_MARK:
5496 ret = flow_dv_validate_item_mark(dev, items, attr,
5500 last_item = MLX5_FLOW_ITEM_MARK;
5502 case RTE_FLOW_ITEM_TYPE_META:
5503 ret = flow_dv_validate_item_meta(dev, items, attr,
5507 last_item = MLX5_FLOW_ITEM_METADATA;
5509 case RTE_FLOW_ITEM_TYPE_ICMP:
5510 ret = mlx5_flow_validate_item_icmp(items, item_flags,
5515 last_item = MLX5_FLOW_LAYER_ICMP;
5517 case RTE_FLOW_ITEM_TYPE_ICMP6:
5518 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
5523 item_ipv6_proto = IPPROTO_ICMPV6;
5524 last_item = MLX5_FLOW_LAYER_ICMP6;
5526 case RTE_FLOW_ITEM_TYPE_TAG:
5527 ret = flow_dv_validate_item_tag(dev, items,
5531 last_item = MLX5_FLOW_ITEM_TAG;
5533 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
5534 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
5536 case RTE_FLOW_ITEM_TYPE_GTP:
5537 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
5541 last_item = MLX5_FLOW_LAYER_GTP;
5543 case RTE_FLOW_ITEM_TYPE_ECPRI:
5544 /* Capacity will be checked in the translate stage. */
5545 ret = mlx5_flow_validate_item_ecpri(items, item_flags,
5552 last_item = MLX5_FLOW_LAYER_ECPRI;
5555 return rte_flow_error_set(error, ENOTSUP,
5556 RTE_FLOW_ERROR_TYPE_ITEM,
5557 NULL, "item not supported");
5559 item_flags |= last_item;
5561 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5562 int type = actions->type;
5564 if (!mlx5_flow_os_action_supported(type))
5565 return rte_flow_error_set(error, ENOTSUP,
5566 RTE_FLOW_ERROR_TYPE_ACTION,
5568 "action not supported");
5569 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5570 return rte_flow_error_set(error, ENOTSUP,
5571 RTE_FLOW_ERROR_TYPE_ACTION,
5572 actions, "too many actions");
5574 case RTE_FLOW_ACTION_TYPE_VOID:
5576 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5577 ret = flow_dv_validate_action_port_id(dev,
5584 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5587 case RTE_FLOW_ACTION_TYPE_FLAG:
5588 ret = flow_dv_validate_action_flag(dev, action_flags,
5592 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5593 /* Count all modify-header actions as one. */
5594 if (!(action_flags &
5595 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5597 action_flags |= MLX5_FLOW_ACTION_FLAG |
5598 MLX5_FLOW_ACTION_MARK_EXT;
5600 action_flags |= MLX5_FLOW_ACTION_FLAG;
5603 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5605 case RTE_FLOW_ACTION_TYPE_MARK:
5606 ret = flow_dv_validate_action_mark(dev, actions,
5611 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5612 /* Count all modify-header actions as one. */
5613 if (!(action_flags &
5614 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5616 action_flags |= MLX5_FLOW_ACTION_MARK |
5617 MLX5_FLOW_ACTION_MARK_EXT;
5619 action_flags |= MLX5_FLOW_ACTION_MARK;
5622 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5624 case RTE_FLOW_ACTION_TYPE_SET_META:
5625 ret = flow_dv_validate_action_set_meta(dev, actions,
5630 /* Count all modify-header actions as one action. */
5631 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5633 action_flags |= MLX5_FLOW_ACTION_SET_META;
5634 rw_act_num += MLX5_ACT_NUM_SET_META;
5636 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5637 ret = flow_dv_validate_action_set_tag(dev, actions,
5642 /* Count all modify-header actions as one action. */
5643 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5645 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
5646 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5648 case RTE_FLOW_ACTION_TYPE_DROP:
5649 ret = mlx5_flow_validate_action_drop(action_flags,
5653 action_flags |= MLX5_FLOW_ACTION_DROP;
5656 case RTE_FLOW_ACTION_TYPE_QUEUE:
5657 ret = mlx5_flow_validate_action_queue(actions,
5662 queue_index = ((const struct rte_flow_action_queue *)
5663 (actions->conf))->index;
5664 action_flags |= MLX5_FLOW_ACTION_QUEUE;
5667 case RTE_FLOW_ACTION_TYPE_RSS:
5668 rss = actions->conf;
5669 ret = mlx5_flow_validate_action_rss(actions,
5675 if (rss != NULL && rss->queue_num)
5676 queue_index = rss->queue[0];
5677 action_flags |= MLX5_FLOW_ACTION_RSS;
5680 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
5682 mlx5_flow_validate_action_default_miss(action_flags,
5686 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
5689 case RTE_FLOW_ACTION_TYPE_COUNT:
5690 ret = flow_dv_validate_action_count(dev, error);
5693 action_flags |= MLX5_FLOW_ACTION_COUNT;
5696 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5697 if (flow_dv_validate_action_pop_vlan(dev,
5703 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5706 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5707 ret = flow_dv_validate_action_push_vlan(dev,
5714 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5717 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5718 ret = flow_dv_validate_action_set_vlan_pcp
5719 (action_flags, actions, error);
5722 /* Count PCP with push_vlan command. */
5723 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5725 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5726 ret = flow_dv_validate_action_set_vlan_vid
5727 (item_flags, action_flags,
5731 /* Count VID with push_vlan command. */
5732 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5733 rw_act_num += MLX5_ACT_NUM_MDF_VID;
5735 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5736 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5737 ret = flow_dv_validate_action_l2_encap(dev,
5743 action_flags |= MLX5_FLOW_ACTION_ENCAP;
5746 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5747 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5748 ret = flow_dv_validate_action_decap(dev, action_flags,
5752 action_flags |= MLX5_FLOW_ACTION_DECAP;
5755 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5756 ret = flow_dv_validate_action_raw_encap_decap
5757 (dev, NULL, actions->conf, attr, &action_flags,
5762 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5763 decap = actions->conf;
5764 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5766 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5770 encap = actions->conf;
5772 ret = flow_dv_validate_action_raw_encap_decap
5774 decap ? decap : &empty_decap, encap,
5775 attr, &action_flags, &actions_n,
5780 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5781 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5782 ret = flow_dv_validate_action_modify_mac(action_flags,
5788 /* Count all modify-header actions as one action. */
5789 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5791 action_flags |= actions->type ==
5792 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5793 MLX5_FLOW_ACTION_SET_MAC_SRC :
5794 MLX5_FLOW_ACTION_SET_MAC_DST;
5796 * Even if the source and destination MAC addresses have
5797 * overlap in the header with 4B alignment, the convert
5798 * function will handle them separately and 4 SW actions
5799 * will be created. And 2 actions will be added each
5800 * time no matter how many bytes of address will be set.
5802 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5804 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5805 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5806 ret = flow_dv_validate_action_modify_ipv4(action_flags,
5812 /* Count all modify-header actions as one action. */
5813 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5815 action_flags |= actions->type ==
5816 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5817 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5818 MLX5_FLOW_ACTION_SET_IPV4_DST;
5819 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5821 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5822 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5823 ret = flow_dv_validate_action_modify_ipv6(action_flags,
5829 if (item_ipv6_proto == IPPROTO_ICMPV6)
5830 return rte_flow_error_set(error, ENOTSUP,
5831 RTE_FLOW_ERROR_TYPE_ACTION,
5833 "Can't change header "
5834 "with ICMPv6 proto");
5835 /* Count all modify-header actions as one action. */
5836 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5838 action_flags |= actions->type ==
5839 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5840 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5841 MLX5_FLOW_ACTION_SET_IPV6_DST;
5842 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5844 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5845 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5846 ret = flow_dv_validate_action_modify_tp(action_flags,
5852 /* Count all modify-header actions as one action. */
5853 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5855 action_flags |= actions->type ==
5856 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5857 MLX5_FLOW_ACTION_SET_TP_SRC :
5858 MLX5_FLOW_ACTION_SET_TP_DST;
5859 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5861 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5862 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5863 ret = flow_dv_validate_action_modify_ttl(action_flags,
5869 /* Count all modify-header actions as one action. */
5870 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5872 action_flags |= actions->type ==
5873 RTE_FLOW_ACTION_TYPE_SET_TTL ?
5874 MLX5_FLOW_ACTION_SET_TTL :
5875 MLX5_FLOW_ACTION_DEC_TTL;
5876 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5878 case RTE_FLOW_ACTION_TYPE_JUMP:
5879 ret = flow_dv_validate_action_jump(dev, tunnel, actions,
5886 action_flags |= MLX5_FLOW_ACTION_JUMP;
5888 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5889 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5890 ret = flow_dv_validate_action_modify_tcp_seq
5897 /* Count all modify-header actions as one action. */
5898 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5900 action_flags |= actions->type ==
5901 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5902 MLX5_FLOW_ACTION_INC_TCP_SEQ :
5903 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5904 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5906 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5907 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5908 ret = flow_dv_validate_action_modify_tcp_ack
5915 /* Count all modify-header actions as one action. */
5916 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5918 action_flags |= actions->type ==
5919 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5920 MLX5_FLOW_ACTION_INC_TCP_ACK :
5921 MLX5_FLOW_ACTION_DEC_TCP_ACK;
5922 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5924 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5926 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5927 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5928 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5930 case RTE_FLOW_ACTION_TYPE_METER:
5931 ret = mlx5_flow_validate_action_meter(dev,
5937 action_flags |= MLX5_FLOW_ACTION_METER;
5939 /* Meter action will add one more TAG action. */
5940 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5942 case RTE_FLOW_ACTION_TYPE_AGE:
5943 ret = flow_dv_validate_action_age(action_flags,
5948 action_flags |= MLX5_FLOW_ACTION_AGE;
5951 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5952 ret = flow_dv_validate_action_modify_ipv4_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_IPV4_DSCP;
5963 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5965 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5966 ret = flow_dv_validate_action_modify_ipv6_dscp
5973 /* Count all modify-header actions as one action. */
5974 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5976 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5977 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5979 case RTE_FLOW_ACTION_TYPE_SAMPLE:
5980 ret = flow_dv_validate_action_sample(action_flags,
5985 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
5988 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
5989 if (actions[0].type != (typeof(actions[0].type))
5990 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET)
5991 return rte_flow_error_set
5993 RTE_FLOW_ERROR_TYPE_ACTION,
5994 NULL, "MLX5 private action "
5995 "must be the first");
5997 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
6000 return rte_flow_error_set(error, ENOTSUP,
6001 RTE_FLOW_ERROR_TYPE_ACTION,
6003 "action not supported");
6007 * Validate actions in flow rules
6008 * - Explicit decap action is prohibited by the tunnel offload API.
6009 * - Drop action in tunnel steer rule is prohibited by the API.
6010 * - Application cannot use MARK action because it's value can mask
6011 * tunnel default miss nitification.
6012 * - JUMP in tunnel match rule has no support in current PMD
6014 * - TAG & META are reserved for future uses.
6016 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_SET) {
6017 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_DECAP |
6018 MLX5_FLOW_ACTION_MARK |
6019 MLX5_FLOW_ACTION_SET_TAG |
6020 MLX5_FLOW_ACTION_SET_META |
6021 MLX5_FLOW_ACTION_DROP;
6023 if (action_flags & bad_actions_mask)
6024 return rte_flow_error_set
6026 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6027 "Invalid RTE action in tunnel "
6029 if (!(action_flags & MLX5_FLOW_ACTION_JUMP))
6030 return rte_flow_error_set
6032 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6033 "tunnel set decap rule must terminate "
6036 return rte_flow_error_set
6038 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6039 "tunnel flows for ingress traffic only");
6041 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_MATCH) {
6042 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_JUMP |
6043 MLX5_FLOW_ACTION_MARK |
6044 MLX5_FLOW_ACTION_SET_TAG |
6045 MLX5_FLOW_ACTION_SET_META;
6047 if (action_flags & bad_actions_mask)
6048 return rte_flow_error_set
6050 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6051 "Invalid RTE action in tunnel "
6055 * Validate the drop action mutual exclusion with other actions.
6056 * Drop action is mutually-exclusive with any other action, except for
6059 if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
6060 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
6061 return rte_flow_error_set(error, EINVAL,
6062 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6063 "Drop action is mutually-exclusive "
6064 "with any other action, except for "
6066 /* Eswitch has few restrictions on using items and actions */
6067 if (attr->transfer) {
6068 if (!mlx5_flow_ext_mreg_supported(dev) &&
6069 action_flags & MLX5_FLOW_ACTION_FLAG)
6070 return rte_flow_error_set(error, ENOTSUP,
6071 RTE_FLOW_ERROR_TYPE_ACTION,
6073 "unsupported action FLAG");
6074 if (!mlx5_flow_ext_mreg_supported(dev) &&
6075 action_flags & MLX5_FLOW_ACTION_MARK)
6076 return rte_flow_error_set(error, ENOTSUP,
6077 RTE_FLOW_ERROR_TYPE_ACTION,
6079 "unsupported action MARK");
6080 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
6081 return rte_flow_error_set(error, ENOTSUP,
6082 RTE_FLOW_ERROR_TYPE_ACTION,
6084 "unsupported action QUEUE");
6085 if (action_flags & MLX5_FLOW_ACTION_RSS)
6086 return rte_flow_error_set(error, ENOTSUP,
6087 RTE_FLOW_ERROR_TYPE_ACTION,
6089 "unsupported action RSS");
6090 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
6091 return rte_flow_error_set(error, EINVAL,
6092 RTE_FLOW_ERROR_TYPE_ACTION,
6094 "no fate action is found");
6096 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
6097 return rte_flow_error_set(error, EINVAL,
6098 RTE_FLOW_ERROR_TYPE_ACTION,
6100 "no fate action is found");
6103 * Continue validation for Xcap and VLAN actions.
6104 * If hairpin is working in explicit TX rule mode, there is no actions
6105 * splitting and the validation of hairpin ingress flow should be the
6106 * same as other standard flows.
6108 if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
6109 MLX5_FLOW_VLAN_ACTIONS)) &&
6110 (queue_index == 0xFFFF ||
6111 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN ||
6112 ((conf = mlx5_rxq_get_hairpin_conf(dev, queue_index)) != NULL &&
6113 conf->tx_explicit != 0))) {
6114 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
6115 MLX5_FLOW_XCAP_ACTIONS)
6116 return rte_flow_error_set(error, ENOTSUP,
6117 RTE_FLOW_ERROR_TYPE_ACTION,
6118 NULL, "encap and decap "
6119 "combination aren't supported");
6120 if (!attr->transfer && attr->ingress) {
6121 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
6122 return rte_flow_error_set
6124 RTE_FLOW_ERROR_TYPE_ACTION,
6125 NULL, "encap is not supported"
6126 " for ingress traffic");
6127 else if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
6128 return rte_flow_error_set
6130 RTE_FLOW_ERROR_TYPE_ACTION,
6131 NULL, "push VLAN action not "
6132 "supported for ingress");
6133 else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
6134 MLX5_FLOW_VLAN_ACTIONS)
6135 return rte_flow_error_set
6137 RTE_FLOW_ERROR_TYPE_ACTION,
6138 NULL, "no support for "
6139 "multiple VLAN actions");
6143 * Hairpin flow will add one more TAG action in TX implicit mode.
6144 * In TX explicit mode, there will be no hairpin flow ID.
6147 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6148 /* extra metadata enabled: one more TAG action will be add. */
6149 if (dev_conf->dv_flow_en &&
6150 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
6151 mlx5_flow_ext_mreg_supported(dev))
6152 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6153 if ((uint32_t)rw_act_num >
6154 flow_dv_modify_hdr_action_max(dev, is_root)) {
6155 return rte_flow_error_set(error, ENOTSUP,
6156 RTE_FLOW_ERROR_TYPE_ACTION,
6157 NULL, "too many header modify"
6158 " actions to support");
6164 * Internal preparation function. Allocates the DV flow size,
6165 * this size is constant.
6168 * Pointer to the rte_eth_dev structure.
6170 * Pointer to the flow attributes.
6172 * Pointer to the list of items.
6173 * @param[in] actions
6174 * Pointer to the list of actions.
6176 * Pointer to the error structure.
6179 * Pointer to mlx5_flow object on success,
6180 * otherwise NULL and rte_errno is set.
6182 static struct mlx5_flow *
6183 flow_dv_prepare(struct rte_eth_dev *dev,
6184 const struct rte_flow_attr *attr __rte_unused,
6185 const struct rte_flow_item items[] __rte_unused,
6186 const struct rte_flow_action actions[] __rte_unused,
6187 struct rte_flow_error *error)
6189 uint32_t handle_idx = 0;
6190 struct mlx5_flow *dev_flow;
6191 struct mlx5_flow_handle *dev_handle;
6192 struct mlx5_priv *priv = dev->data->dev_private;
6193 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6196 /* In case of corrupting the memory. */
6197 if (wks->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
6198 rte_flow_error_set(error, ENOSPC,
6199 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6200 "not free temporary device flow");
6203 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
6206 rte_flow_error_set(error, ENOMEM,
6207 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6208 "not enough memory to create flow handle");
6211 MLX5_ASSERT(wks->flow_idx + 1 < RTE_DIM(wks->flows));
6212 dev_flow = &wks->flows[wks->flow_idx++];
6213 dev_flow->handle = dev_handle;
6214 dev_flow->handle_idx = handle_idx;
6216 * In some old rdma-core releases, before continuing, a check of the
6217 * length of matching parameter will be done at first. It needs to use
6218 * the length without misc4 param. If the flow has misc4 support, then
6219 * the length needs to be adjusted accordingly. Each param member is
6220 * aligned with a 64B boundary naturally.
6222 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param) -
6223 MLX5_ST_SZ_BYTES(fte_match_set_misc4);
6225 * The matching value needs to be cleared to 0 before using. In the
6226 * past, it will be automatically cleared when using rte_*alloc
6227 * API. The time consumption will be almost the same as before.
6229 memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
6230 dev_flow->ingress = attr->ingress;
6231 dev_flow->dv.transfer = attr->transfer;
6235 #ifdef RTE_LIBRTE_MLX5_DEBUG
6237 * Sanity check for match mask and value. Similar to check_valid_spec() in
6238 * kernel driver. If unmasked bit is present in value, it returns failure.
6241 * pointer to match mask buffer.
6242 * @param match_value
6243 * pointer to match value buffer.
6246 * 0 if valid, -EINVAL otherwise.
6249 flow_dv_check_valid_spec(void *match_mask, void *match_value)
6251 uint8_t *m = match_mask;
6252 uint8_t *v = match_value;
6255 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
6258 "match_value differs from match_criteria"
6259 " %p[%u] != %p[%u]",
6260 match_value, i, match_mask, i);
6269 * Add match of ip_version.
6273 * @param[in] headers_v
6274 * Values header pointer.
6275 * @param[in] headers_m
6276 * Masks header pointer.
6277 * @param[in] ip_version
6278 * The IP version to set.
6281 flow_dv_set_match_ip_version(uint32_t group,
6287 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
6289 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
6291 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
6292 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
6293 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
6297 * Add Ethernet item to matcher and to the value.
6299 * @param[in, out] matcher
6301 * @param[in, out] key
6302 * Flow matcher value.
6304 * Flow pattern to translate.
6306 * Item is inner pattern.
6309 flow_dv_translate_item_eth(void *matcher, void *key,
6310 const struct rte_flow_item *item, int inner,
6313 const struct rte_flow_item_eth *eth_m = item->mask;
6314 const struct rte_flow_item_eth *eth_v = item->spec;
6315 const struct rte_flow_item_eth nic_mask = {
6316 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6317 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6318 .type = RTE_BE16(0xffff),
6331 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6333 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6335 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6337 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6339 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, dmac_47_16),
6340 ð_m->dst, sizeof(eth_m->dst));
6341 /* The value must be in the range of the mask. */
6342 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, dmac_47_16);
6343 for (i = 0; i < sizeof(eth_m->dst); ++i)
6344 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
6345 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, smac_47_16),
6346 ð_m->src, sizeof(eth_m->src));
6347 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, smac_47_16);
6348 /* The value must be in the range of the mask. */
6349 for (i = 0; i < sizeof(eth_m->dst); ++i)
6350 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
6352 * HW supports match on one Ethertype, the Ethertype following the last
6353 * VLAN tag of the packet (see PRM).
6354 * Set match on ethertype only if ETH header is not followed by VLAN.
6355 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6356 * ethertype, and use ip_version field instead.
6357 * eCPRI over Ether layer will use type value 0xAEFE.
6359 if (eth_m->type == 0xFFFF) {
6360 /* Set cvlan_tag mask for any single\multi\un-tagged case. */
6361 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6362 switch (eth_v->type) {
6363 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
6364 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6366 case RTE_BE16(RTE_ETHER_TYPE_QINQ):
6367 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6368 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6370 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
6371 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
6373 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
6374 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
6380 if (eth_m->has_vlan) {
6381 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6382 if (eth_v->has_vlan) {
6384 * Here, when also has_more_vlan field in VLAN item is
6385 * not set, only single-tagged packets will be matched.
6387 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6391 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
6392 rte_be_to_cpu_16(eth_m->type));
6393 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, ethertype);
6394 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
6398 * Add VLAN item to matcher and to the value.
6400 * @param[in, out] dev_flow
6402 * @param[in, out] matcher
6404 * @param[in, out] key
6405 * Flow matcher value.
6407 * Flow pattern to translate.
6409 * Item is inner pattern.
6412 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
6413 void *matcher, void *key,
6414 const struct rte_flow_item *item,
6415 int inner, uint32_t group)
6417 const struct rte_flow_item_vlan *vlan_m = item->mask;
6418 const struct rte_flow_item_vlan *vlan_v = item->spec;
6425 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6427 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6429 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6431 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6433 * This is workaround, masks are not supported,
6434 * and pre-validated.
6437 dev_flow->handle->vf_vlan.tag =
6438 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
6441 * When VLAN item exists in flow, mark packet as tagged,
6442 * even if TCI is not specified.
6444 if (!MLX5_GET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag)) {
6445 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6446 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6451 vlan_m = &rte_flow_item_vlan_mask;
6452 tci_m = rte_be_to_cpu_16(vlan_m->tci);
6453 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
6454 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_vid, tci_m);
6455 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_vid, tci_v);
6456 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_cfi, tci_m >> 12);
6457 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_cfi, tci_v >> 12);
6458 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_prio, tci_m >> 13);
6459 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_prio, tci_v >> 13);
6461 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6462 * ethertype, and use ip_version field instead.
6464 if (vlan_m->inner_type == 0xFFFF) {
6465 switch (vlan_v->inner_type) {
6466 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
6467 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6468 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6469 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
6471 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
6472 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
6474 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
6475 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
6481 if (vlan_m->has_more_vlan && vlan_v->has_more_vlan) {
6482 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6483 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6484 /* Only one vlan_tag bit can be set. */
6485 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
6488 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
6489 rte_be_to_cpu_16(vlan_m->inner_type));
6490 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, ethertype,
6491 rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
6495 * Add IPV4 item to matcher and to the value.
6497 * @param[in, out] matcher
6499 * @param[in, out] key
6500 * Flow matcher value.
6502 * Flow pattern to translate.
6504 * Item is inner pattern.
6506 * The group to insert the rule.
6509 flow_dv_translate_item_ipv4(void *matcher, void *key,
6510 const struct rte_flow_item *item,
6511 int inner, uint32_t group)
6513 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
6514 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
6515 const struct rte_flow_item_ipv4 nic_mask = {
6517 .src_addr = RTE_BE32(0xffffffff),
6518 .dst_addr = RTE_BE32(0xffffffff),
6519 .type_of_service = 0xff,
6520 .next_proto_id = 0xff,
6521 .time_to_live = 0xff,
6531 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6533 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6535 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6537 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6539 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6544 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6545 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6546 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6547 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6548 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
6549 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
6550 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6551 src_ipv4_src_ipv6.ipv4_layout.ipv4);
6552 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6553 src_ipv4_src_ipv6.ipv4_layout.ipv4);
6554 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
6555 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
6556 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
6557 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
6558 ipv4_m->hdr.type_of_service);
6559 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
6560 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
6561 ipv4_m->hdr.type_of_service >> 2);
6562 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
6563 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6564 ipv4_m->hdr.next_proto_id);
6565 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6566 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
6567 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6568 ipv4_m->hdr.time_to_live);
6569 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6570 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
6571 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
6572 !!(ipv4_m->hdr.fragment_offset));
6573 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
6574 !!(ipv4_v->hdr.fragment_offset & ipv4_m->hdr.fragment_offset));
6578 * Add IPV6 item to matcher and to the value.
6580 * @param[in, out] matcher
6582 * @param[in, out] key
6583 * Flow matcher value.
6585 * Flow pattern to translate.
6587 * Item is inner pattern.
6589 * The group to insert the rule.
6592 flow_dv_translate_item_ipv6(void *matcher, void *key,
6593 const struct rte_flow_item *item,
6594 int inner, uint32_t group)
6596 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
6597 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
6598 const struct rte_flow_item_ipv6 nic_mask = {
6601 "\xff\xff\xff\xff\xff\xff\xff\xff"
6602 "\xff\xff\xff\xff\xff\xff\xff\xff",
6604 "\xff\xff\xff\xff\xff\xff\xff\xff"
6605 "\xff\xff\xff\xff\xff\xff\xff\xff",
6606 .vtc_flow = RTE_BE32(0xffffffff),
6613 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6614 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6623 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6625 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6627 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6629 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6631 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6636 size = sizeof(ipv6_m->hdr.dst_addr);
6637 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6638 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6639 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6640 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6641 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
6642 for (i = 0; i < size; ++i)
6643 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
6644 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6645 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6646 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6647 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6648 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
6649 for (i = 0; i < size; ++i)
6650 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
6652 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
6653 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
6654 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
6655 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
6656 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
6657 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
6660 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
6662 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
6665 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
6667 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
6671 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6673 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6674 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
6676 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6677 ipv6_m->hdr.hop_limits);
6678 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6679 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
6680 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
6681 !!(ipv6_m->has_frag_ext));
6682 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
6683 !!(ipv6_v->has_frag_ext & ipv6_m->has_frag_ext));
6687 * Add IPV6 fragment extension item to matcher and to the value.
6689 * @param[in, out] matcher
6691 * @param[in, out] key
6692 * Flow matcher value.
6694 * Flow pattern to translate.
6696 * Item is inner pattern.
6699 flow_dv_translate_item_ipv6_frag_ext(void *matcher, void *key,
6700 const struct rte_flow_item *item,
6703 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_m = item->mask;
6704 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_v = item->spec;
6705 const struct rte_flow_item_ipv6_frag_ext nic_mask = {
6707 .next_header = 0xff,
6708 .frag_data = RTE_BE16(0xffff),
6715 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6717 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6719 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6721 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6723 /* IPv6 fragment extension item exists, so packet is IP fragment. */
6724 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6725 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 1);
6726 if (!ipv6_frag_ext_v)
6728 if (!ipv6_frag_ext_m)
6729 ipv6_frag_ext_m = &nic_mask;
6730 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6731 ipv6_frag_ext_m->hdr.next_header);
6732 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6733 ipv6_frag_ext_v->hdr.next_header &
6734 ipv6_frag_ext_m->hdr.next_header);
6738 * Add TCP item to matcher and to the value.
6740 * @param[in, out] matcher
6742 * @param[in, out] key
6743 * Flow matcher value.
6745 * Flow pattern to translate.
6747 * Item is inner pattern.
6750 flow_dv_translate_item_tcp(void *matcher, void *key,
6751 const struct rte_flow_item *item,
6754 const struct rte_flow_item_tcp *tcp_m = item->mask;
6755 const struct rte_flow_item_tcp *tcp_v = item->spec;
6760 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6762 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6764 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6766 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6768 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6769 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
6773 tcp_m = &rte_flow_item_tcp_mask;
6774 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
6775 rte_be_to_cpu_16(tcp_m->hdr.src_port));
6776 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
6777 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
6778 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
6779 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
6780 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
6781 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
6782 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
6783 tcp_m->hdr.tcp_flags);
6784 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
6785 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
6789 * Add UDP item to matcher and to the value.
6791 * @param[in, out] matcher
6793 * @param[in, out] key
6794 * Flow matcher value.
6796 * Flow pattern to translate.
6798 * Item is inner pattern.
6801 flow_dv_translate_item_udp(void *matcher, void *key,
6802 const struct rte_flow_item *item,
6805 const struct rte_flow_item_udp *udp_m = item->mask;
6806 const struct rte_flow_item_udp *udp_v = item->spec;
6811 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6813 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6815 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6817 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6819 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6820 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
6824 udp_m = &rte_flow_item_udp_mask;
6825 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
6826 rte_be_to_cpu_16(udp_m->hdr.src_port));
6827 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
6828 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
6829 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
6830 rte_be_to_cpu_16(udp_m->hdr.dst_port));
6831 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6832 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
6836 * Add GRE optional Key item to matcher and to the value.
6838 * @param[in, out] matcher
6840 * @param[in, out] key
6841 * Flow matcher value.
6843 * Flow pattern to translate.
6845 * Item is inner pattern.
6848 flow_dv_translate_item_gre_key(void *matcher, void *key,
6849 const struct rte_flow_item *item)
6851 const rte_be32_t *key_m = item->mask;
6852 const rte_be32_t *key_v = item->spec;
6853 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6854 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6855 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
6857 /* GRE K bit must be on and should already be validated */
6858 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
6859 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
6863 key_m = &gre_key_default_mask;
6864 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
6865 rte_be_to_cpu_32(*key_m) >> 8);
6866 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
6867 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
6868 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
6869 rte_be_to_cpu_32(*key_m) & 0xFF);
6870 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
6871 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
6875 * Add GRE item to matcher and to the value.
6877 * @param[in, out] matcher
6879 * @param[in, out] key
6880 * Flow matcher value.
6882 * Flow pattern to translate.
6884 * Item is inner pattern.
6887 flow_dv_translate_item_gre(void *matcher, void *key,
6888 const struct rte_flow_item *item,
6891 const struct rte_flow_item_gre *gre_m = item->mask;
6892 const struct rte_flow_item_gre *gre_v = item->spec;
6895 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6896 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6903 uint16_t s_present:1;
6904 uint16_t k_present:1;
6905 uint16_t rsvd_bit1:1;
6906 uint16_t c_present:1;
6910 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
6913 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6915 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6917 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6919 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6921 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6922 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
6926 gre_m = &rte_flow_item_gre_mask;
6927 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
6928 rte_be_to_cpu_16(gre_m->protocol));
6929 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6930 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
6931 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
6932 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
6933 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
6934 gre_crks_rsvd0_ver_m.c_present);
6935 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6936 gre_crks_rsvd0_ver_v.c_present &
6937 gre_crks_rsvd0_ver_m.c_present);
6938 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6939 gre_crks_rsvd0_ver_m.k_present);
6940 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6941 gre_crks_rsvd0_ver_v.k_present &
6942 gre_crks_rsvd0_ver_m.k_present);
6943 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6944 gre_crks_rsvd0_ver_m.s_present);
6945 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6946 gre_crks_rsvd0_ver_v.s_present &
6947 gre_crks_rsvd0_ver_m.s_present);
6951 * Add NVGRE item to matcher and to the value.
6953 * @param[in, out] matcher
6955 * @param[in, out] key
6956 * Flow matcher value.
6958 * Flow pattern to translate.
6960 * Item is inner pattern.
6963 flow_dv_translate_item_nvgre(void *matcher, void *key,
6964 const struct rte_flow_item *item,
6967 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6968 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6969 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6970 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6971 const char *tni_flow_id_m;
6972 const char *tni_flow_id_v;
6978 /* For NVGRE, GRE header fields must be set with defined values. */
6979 const struct rte_flow_item_gre gre_spec = {
6980 .c_rsvd0_ver = RTE_BE16(0x2000),
6981 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6983 const struct rte_flow_item_gre gre_mask = {
6984 .c_rsvd0_ver = RTE_BE16(0xB000),
6985 .protocol = RTE_BE16(UINT16_MAX),
6987 const struct rte_flow_item gre_item = {
6992 flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6996 nvgre_m = &rte_flow_item_nvgre_mask;
6997 tni_flow_id_m = (const char *)nvgre_m->tni;
6998 tni_flow_id_v = (const char *)nvgre_v->tni;
6999 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
7000 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
7001 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
7002 memcpy(gre_key_m, tni_flow_id_m, size);
7003 for (i = 0; i < size; ++i)
7004 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
7008 * Add VXLAN item to matcher and to the value.
7010 * @param[in, out] matcher
7012 * @param[in, out] key
7013 * Flow matcher value.
7015 * Flow pattern to translate.
7017 * Item is inner pattern.
7020 flow_dv_translate_item_vxlan(void *matcher, void *key,
7021 const struct rte_flow_item *item,
7024 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
7025 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
7028 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7029 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7037 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7039 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7041 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7043 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7045 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
7046 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
7047 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7048 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7049 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7054 vxlan_m = &rte_flow_item_vxlan_mask;
7055 size = sizeof(vxlan_m->vni);
7056 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
7057 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
7058 memcpy(vni_m, vxlan_m->vni, size);
7059 for (i = 0; i < size; ++i)
7060 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7064 * Add VXLAN-GPE item to matcher and to the value.
7066 * @param[in, out] matcher
7068 * @param[in, out] key
7069 * Flow matcher value.
7071 * Flow pattern to translate.
7073 * Item is inner pattern.
7077 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
7078 const struct rte_flow_item *item, int inner)
7080 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
7081 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
7085 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
7087 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7093 uint8_t flags_m = 0xff;
7094 uint8_t flags_v = 0xc;
7097 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7099 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7101 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7103 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7105 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
7106 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
7107 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7108 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7109 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7114 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
7115 size = sizeof(vxlan_m->vni);
7116 vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
7117 vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
7118 memcpy(vni_m, vxlan_m->vni, size);
7119 for (i = 0; i < size; ++i)
7120 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7121 if (vxlan_m->flags) {
7122 flags_m = vxlan_m->flags;
7123 flags_v = vxlan_v->flags;
7125 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
7126 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
7127 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
7129 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
7134 * Add Geneve item to matcher and to the value.
7136 * @param[in, out] matcher
7138 * @param[in, out] key
7139 * Flow matcher value.
7141 * Flow pattern to translate.
7143 * Item is inner pattern.
7147 flow_dv_translate_item_geneve(void *matcher, void *key,
7148 const struct rte_flow_item *item, int inner)
7150 const struct rte_flow_item_geneve *geneve_m = item->mask;
7151 const struct rte_flow_item_geneve *geneve_v = item->spec;
7154 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7155 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7164 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7166 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7168 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7170 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7172 dport = MLX5_UDP_PORT_GENEVE;
7173 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7174 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7175 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7180 geneve_m = &rte_flow_item_geneve_mask;
7181 size = sizeof(geneve_m->vni);
7182 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
7183 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
7184 memcpy(vni_m, geneve_m->vni, size);
7185 for (i = 0; i < size; ++i)
7186 vni_v[i] = vni_m[i] & geneve_v->vni[i];
7187 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
7188 rte_be_to_cpu_16(geneve_m->protocol));
7189 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
7190 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
7191 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
7192 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
7193 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
7194 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7195 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
7196 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7197 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
7198 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7199 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
7200 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
7201 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7205 * Add MPLS item to matcher and to the value.
7207 * @param[in, out] matcher
7209 * @param[in, out] key
7210 * Flow matcher value.
7212 * Flow pattern to translate.
7213 * @param[in] prev_layer
7214 * The protocol layer indicated in previous item.
7216 * Item is inner pattern.
7219 flow_dv_translate_item_mpls(void *matcher, void *key,
7220 const struct rte_flow_item *item,
7221 uint64_t prev_layer,
7224 const uint32_t *in_mpls_m = item->mask;
7225 const uint32_t *in_mpls_v = item->spec;
7226 uint32_t *out_mpls_m = 0;
7227 uint32_t *out_mpls_v = 0;
7228 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7229 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7230 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
7232 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7233 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
7234 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7236 switch (prev_layer) {
7237 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7238 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
7239 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
7240 MLX5_UDP_PORT_MPLS);
7242 case MLX5_FLOW_LAYER_GRE:
7243 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
7244 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
7245 RTE_ETHER_TYPE_MPLS);
7248 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
7249 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
7256 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
7257 switch (prev_layer) {
7258 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7260 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7261 outer_first_mpls_over_udp);
7263 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7264 outer_first_mpls_over_udp);
7266 case MLX5_FLOW_LAYER_GRE:
7268 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7269 outer_first_mpls_over_gre);
7271 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7272 outer_first_mpls_over_gre);
7275 /* Inner MPLS not over GRE is not supported. */
7278 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7282 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7288 if (out_mpls_m && out_mpls_v) {
7289 *out_mpls_m = *in_mpls_m;
7290 *out_mpls_v = *in_mpls_v & *in_mpls_m;
7295 * Add metadata register item to matcher
7297 * @param[in, out] matcher
7299 * @param[in, out] key
7300 * Flow matcher value.
7301 * @param[in] reg_type
7302 * Type of device metadata register
7309 flow_dv_match_meta_reg(void *matcher, void *key,
7310 enum modify_reg reg_type,
7311 uint32_t data, uint32_t mask)
7314 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
7316 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7322 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
7323 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
7326 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
7327 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
7331 * The metadata register C0 field might be divided into
7332 * source vport index and META item value, we should set
7333 * this field according to specified mask, not as whole one.
7335 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
7337 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
7338 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
7341 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
7344 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
7345 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
7348 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
7349 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
7352 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
7353 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
7356 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
7357 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
7360 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
7361 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
7364 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
7365 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
7368 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
7369 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
7378 * Add MARK item to matcher
7381 * The device to configure through.
7382 * @param[in, out] matcher
7384 * @param[in, out] key
7385 * Flow matcher value.
7387 * Flow pattern to translate.
7390 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
7391 void *matcher, void *key,
7392 const struct rte_flow_item *item)
7394 struct mlx5_priv *priv = dev->data->dev_private;
7395 const struct rte_flow_item_mark *mark;
7399 mark = item->mask ? (const void *)item->mask :
7400 &rte_flow_item_mark_mask;
7401 mask = mark->id & priv->sh->dv_mark_mask;
7402 mark = (const void *)item->spec;
7404 value = mark->id & priv->sh->dv_mark_mask & mask;
7406 enum modify_reg reg;
7408 /* Get the metadata register index for the mark. */
7409 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
7410 MLX5_ASSERT(reg > 0);
7411 if (reg == REG_C_0) {
7412 struct mlx5_priv *priv = dev->data->dev_private;
7413 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7414 uint32_t shl_c0 = rte_bsf32(msk_c0);
7420 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7425 * Add META item to matcher
7428 * The devich to configure through.
7429 * @param[in, out] matcher
7431 * @param[in, out] key
7432 * Flow matcher value.
7434 * Attributes of flow that includes this item.
7436 * Flow pattern to translate.
7439 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
7440 void *matcher, void *key,
7441 const struct rte_flow_attr *attr,
7442 const struct rte_flow_item *item)
7444 const struct rte_flow_item_meta *meta_m;
7445 const struct rte_flow_item_meta *meta_v;
7447 meta_m = (const void *)item->mask;
7449 meta_m = &rte_flow_item_meta_mask;
7450 meta_v = (const void *)item->spec;
7453 uint32_t value = meta_v->data;
7454 uint32_t mask = meta_m->data;
7456 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
7460 * In datapath code there is no endianness
7461 * coversions for perfromance reasons, all
7462 * pattern conversions are done in rte_flow.
7464 value = rte_cpu_to_be_32(value);
7465 mask = rte_cpu_to_be_32(mask);
7466 if (reg == REG_C_0) {
7467 struct mlx5_priv *priv = dev->data->dev_private;
7468 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7469 uint32_t shl_c0 = rte_bsf32(msk_c0);
7470 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
7471 uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
7478 MLX5_ASSERT(msk_c0);
7479 MLX5_ASSERT(!(~msk_c0 & mask));
7481 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7486 * Add vport metadata Reg C0 item to matcher
7488 * @param[in, out] matcher
7490 * @param[in, out] key
7491 * Flow matcher value.
7493 * Flow pattern to translate.
7496 flow_dv_translate_item_meta_vport(void *matcher, void *key,
7497 uint32_t value, uint32_t mask)
7499 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
7503 * Add tag item to matcher
7506 * The devich to configure through.
7507 * @param[in, out] matcher
7509 * @param[in, out] key
7510 * Flow matcher value.
7512 * Flow pattern to translate.
7515 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
7516 void *matcher, void *key,
7517 const struct rte_flow_item *item)
7519 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
7520 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
7521 uint32_t mask, value;
7524 value = tag_v->data;
7525 mask = tag_m ? tag_m->data : UINT32_MAX;
7526 if (tag_v->id == REG_C_0) {
7527 struct mlx5_priv *priv = dev->data->dev_private;
7528 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7529 uint32_t shl_c0 = rte_bsf32(msk_c0);
7535 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
7539 * Add TAG item to matcher
7542 * The devich to configure through.
7543 * @param[in, out] matcher
7545 * @param[in, out] key
7546 * Flow matcher value.
7548 * Flow pattern to translate.
7551 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
7552 void *matcher, void *key,
7553 const struct rte_flow_item *item)
7555 const struct rte_flow_item_tag *tag_v = item->spec;
7556 const struct rte_flow_item_tag *tag_m = item->mask;
7557 enum modify_reg reg;
7560 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
7561 /* Get the metadata register index for the tag. */
7562 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
7563 MLX5_ASSERT(reg > 0);
7564 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
7568 * Add source vport match to the specified matcher.
7570 * @param[in, out] matcher
7572 * @param[in, out] key
7573 * Flow matcher value.
7575 * Source vport value to match
7580 flow_dv_translate_item_source_vport(void *matcher, void *key,
7581 int16_t port, uint16_t mask)
7583 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7584 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7586 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
7587 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
7591 * Translate port-id item to eswitch match on port-id.
7594 * The devich to configure through.
7595 * @param[in, out] matcher
7597 * @param[in, out] key
7598 * Flow matcher value.
7600 * Flow pattern to translate.
7603 * 0 on success, a negative errno value otherwise.
7606 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
7607 void *key, const struct rte_flow_item *item)
7609 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
7610 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
7611 struct mlx5_priv *priv;
7614 mask = pid_m ? pid_m->id : 0xffff;
7615 id = pid_v ? pid_v->id : dev->data->port_id;
7616 priv = mlx5_port_to_eswitch_info(id, item == NULL);
7619 /* Translate to vport field or to metadata, depending on mode. */
7620 if (priv->vport_meta_mask)
7621 flow_dv_translate_item_meta_vport(matcher, key,
7622 priv->vport_meta_tag,
7623 priv->vport_meta_mask);
7625 flow_dv_translate_item_source_vport(matcher, key,
7626 priv->vport_id, mask);
7631 * Add ICMP6 item to matcher and to the value.
7633 * @param[in, out] matcher
7635 * @param[in, out] key
7636 * Flow matcher value.
7638 * Flow pattern to translate.
7640 * Item is inner pattern.
7643 flow_dv_translate_item_icmp6(void *matcher, void *key,
7644 const struct rte_flow_item *item,
7647 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
7648 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
7651 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7653 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7655 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7657 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7659 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7661 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7663 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7664 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
7668 icmp6_m = &rte_flow_item_icmp6_mask;
7669 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
7670 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
7671 icmp6_v->type & icmp6_m->type);
7672 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
7673 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
7674 icmp6_v->code & icmp6_m->code);
7678 * Add ICMP item to matcher and to the value.
7680 * @param[in, out] matcher
7682 * @param[in, out] key
7683 * Flow matcher value.
7685 * Flow pattern to translate.
7687 * Item is inner pattern.
7690 flow_dv_translate_item_icmp(void *matcher, void *key,
7691 const struct rte_flow_item *item,
7694 const struct rte_flow_item_icmp *icmp_m = item->mask;
7695 const struct rte_flow_item_icmp *icmp_v = item->spec;
7696 uint32_t icmp_header_data_m = 0;
7697 uint32_t icmp_header_data_v = 0;
7700 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7702 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7704 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7706 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7708 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7710 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7712 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7713 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
7717 icmp_m = &rte_flow_item_icmp_mask;
7718 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
7719 icmp_m->hdr.icmp_type);
7720 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
7721 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
7722 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
7723 icmp_m->hdr.icmp_code);
7724 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
7725 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
7726 icmp_header_data_m = rte_be_to_cpu_16(icmp_m->hdr.icmp_seq_nb);
7727 icmp_header_data_m |= rte_be_to_cpu_16(icmp_m->hdr.icmp_ident) << 16;
7728 if (icmp_header_data_m) {
7729 icmp_header_data_v = rte_be_to_cpu_16(icmp_v->hdr.icmp_seq_nb);
7730 icmp_header_data_v |=
7731 rte_be_to_cpu_16(icmp_v->hdr.icmp_ident) << 16;
7732 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_header_data,
7733 icmp_header_data_m);
7734 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_header_data,
7735 icmp_header_data_v & icmp_header_data_m);
7740 * Add GTP item to matcher and to the value.
7742 * @param[in, out] matcher
7744 * @param[in, out] key
7745 * Flow matcher value.
7747 * Flow pattern to translate.
7749 * Item is inner pattern.
7752 flow_dv_translate_item_gtp(void *matcher, void *key,
7753 const struct rte_flow_item *item, int inner)
7755 const struct rte_flow_item_gtp *gtp_m = item->mask;
7756 const struct rte_flow_item_gtp *gtp_v = item->spec;
7759 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7761 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7762 uint16_t dport = RTE_GTPU_UDP_PORT;
7765 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7767 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7769 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7771 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7773 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7774 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7775 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7780 gtp_m = &rte_flow_item_gtp_mask;
7781 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
7782 gtp_m->v_pt_rsv_flags);
7783 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
7784 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
7785 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
7786 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
7787 gtp_v->msg_type & gtp_m->msg_type);
7788 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
7789 rte_be_to_cpu_32(gtp_m->teid));
7790 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
7791 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
7795 * Add eCPRI item to matcher and to the value.
7798 * The devich to configure through.
7799 * @param[in, out] matcher
7801 * @param[in, out] key
7802 * Flow matcher value.
7804 * Flow pattern to translate.
7805 * @param[in] samples
7806 * Sample IDs to be used in the matching.
7809 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
7810 void *key, const struct rte_flow_item *item)
7812 struct mlx5_priv *priv = dev->data->dev_private;
7813 const struct rte_flow_item_ecpri *ecpri_m = item->mask;
7814 const struct rte_flow_item_ecpri *ecpri_v = item->spec;
7815 void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
7817 void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
7825 ecpri_m = &rte_flow_item_ecpri_mask;
7827 * Maximal four DW samples are supported in a single matching now.
7828 * Two are used now for a eCPRI matching:
7829 * 1. Type: one byte, mask should be 0x00ff0000 in network order
7830 * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
7833 if (!ecpri_m->hdr.common.u32)
7835 samples = priv->sh->fp[MLX5_FLEX_PARSER_ECPRI_0].ids;
7836 /* Need to take the whole DW as the mask to fill the entry. */
7837 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7838 prog_sample_field_value_0);
7839 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7840 prog_sample_field_value_0);
7841 /* Already big endian (network order) in the header. */
7842 *(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
7843 *(uint32_t *)dw_v = ecpri_v->hdr.common.u32;
7844 /* Sample#0, used for matching type, offset 0. */
7845 MLX5_SET(fte_match_set_misc4, misc4_m,
7846 prog_sample_field_id_0, samples[0]);
7847 /* It makes no sense to set the sample ID in the mask field. */
7848 MLX5_SET(fte_match_set_misc4, misc4_v,
7849 prog_sample_field_id_0, samples[0]);
7851 * Checking if message body part needs to be matched.
7852 * Some wildcard rules only matching type field should be supported.
7854 if (ecpri_m->hdr.dummy[0]) {
7855 switch (ecpri_v->hdr.common.type) {
7856 case RTE_ECPRI_MSG_TYPE_IQ_DATA:
7857 case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
7858 case RTE_ECPRI_MSG_TYPE_DLY_MSR:
7859 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7860 prog_sample_field_value_1);
7861 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7862 prog_sample_field_value_1);
7863 *(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
7864 *(uint32_t *)dw_v = ecpri_v->hdr.dummy[0];
7865 /* Sample#1, to match message body, offset 4. */
7866 MLX5_SET(fte_match_set_misc4, misc4_m,
7867 prog_sample_field_id_1, samples[1]);
7868 MLX5_SET(fte_match_set_misc4, misc4_v,
7869 prog_sample_field_id_1, samples[1]);
7872 /* Others, do not match any sample ID. */
7878 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
7880 #define HEADER_IS_ZERO(match_criteria, headers) \
7881 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
7882 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
7885 * Calculate flow matcher enable bitmap.
7887 * @param match_criteria
7888 * Pointer to flow matcher criteria.
7891 * Bitmap of enabled fields.
7894 flow_dv_matcher_enable(uint32_t *match_criteria)
7896 uint8_t match_criteria_enable;
7898 match_criteria_enable =
7899 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
7900 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
7901 match_criteria_enable |=
7902 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
7903 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
7904 match_criteria_enable |=
7905 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
7906 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
7907 match_criteria_enable |=
7908 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
7909 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
7910 match_criteria_enable |=
7911 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
7912 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
7913 match_criteria_enable |=
7914 (!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
7915 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
7916 return match_criteria_enable;
7919 struct mlx5_hlist_entry *
7920 flow_dv_tbl_create_cb(struct mlx5_hlist *list, uint64_t key64, void *cb_ctx)
7922 struct mlx5_dev_ctx_shared *sh = list->ctx;
7923 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
7924 struct rte_eth_dev *dev = ctx->dev;
7925 struct mlx5_flow_tbl_data_entry *tbl_data;
7926 struct mlx5_flow_tbl_tunnel_prm *tt_prm = ctx->data;
7927 struct rte_flow_error *error = ctx->error;
7928 union mlx5_flow_tbl_key key = { .v64 = key64 };
7929 struct mlx5_flow_tbl_resource *tbl;
7934 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
7936 rte_flow_error_set(error, ENOMEM,
7937 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7939 "cannot allocate flow table data entry");
7942 tbl_data->idx = idx;
7943 tbl_data->tunnel = tt_prm->tunnel;
7944 tbl_data->group_id = tt_prm->group_id;
7945 tbl_data->external = tt_prm->external;
7946 tbl_data->tunnel_offload = is_tunnel_offload_active(dev);
7947 tbl = &tbl_data->tbl;
7949 return &tbl_data->entry;
7951 domain = sh->fdb_domain;
7952 else if (key.direction)
7953 domain = sh->tx_domain;
7955 domain = sh->rx_domain;
7956 ret = mlx5_flow_os_create_flow_tbl(domain, key.table_id, &tbl->obj);
7958 rte_flow_error_set(error, ENOMEM,
7959 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7960 NULL, "cannot create flow table object");
7961 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7965 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
7966 (tbl->obj, &tbl_data->jump.action);
7968 rte_flow_error_set(error, ENOMEM,
7969 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7971 "cannot create flow jump action");
7972 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7973 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7977 return &tbl_data->entry;
7983 * @param[in, out] dev
7984 * Pointer to rte_eth_dev structure.
7985 * @param[in] table_id
7988 * Direction of the table.
7989 * @param[in] transfer
7990 * E-Switch or NIC flow.
7992 * Dummy entry for dv API.
7994 * pointer to error structure.
7997 * Returns tables resource based on the index, NULL in case of failed.
7999 struct mlx5_flow_tbl_resource *
8000 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
8001 uint32_t table_id, uint8_t egress,
8004 const struct mlx5_flow_tunnel *tunnel,
8005 uint32_t group_id, uint8_t dummy,
8006 struct rte_flow_error *error)
8008 struct mlx5_priv *priv = dev->data->dev_private;
8009 union mlx5_flow_tbl_key table_key = {
8011 .table_id = table_id,
8013 .domain = !!transfer,
8014 .direction = !!egress,
8017 struct mlx5_flow_tbl_tunnel_prm tt_prm = {
8019 .group_id = group_id,
8020 .external = external,
8022 struct mlx5_flow_cb_ctx ctx = {
8027 struct mlx5_hlist_entry *entry;
8028 struct mlx5_flow_tbl_data_entry *tbl_data;
8030 entry = mlx5_hlist_register(priv->sh->flow_tbls, table_key.v64, &ctx);
8032 rte_flow_error_set(error, ENOMEM,
8033 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8034 "cannot get table");
8037 tbl_data = container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
8038 return &tbl_data->tbl;
8042 flow_dv_tbl_remove_cb(struct mlx5_hlist *list,
8043 struct mlx5_hlist_entry *entry)
8045 struct mlx5_dev_ctx_shared *sh = list->ctx;
8046 struct mlx5_flow_tbl_data_entry *tbl_data =
8047 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
8049 MLX5_ASSERT(entry && sh);
8050 if (tbl_data->jump.action)
8051 mlx5_flow_os_destroy_flow_action(tbl_data->jump.action);
8052 if (tbl_data->tbl.obj)
8053 mlx5_flow_os_destroy_flow_tbl(tbl_data->tbl.obj);
8054 if (tbl_data->tunnel_offload && tbl_data->external) {
8055 struct mlx5_hlist_entry *he;
8056 struct mlx5_hlist *tunnel_grp_hash;
8057 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8058 union tunnel_tbl_key tunnel_key = {
8059 .tunnel_id = tbl_data->tunnel ?
8060 tbl_data->tunnel->tunnel_id : 0,
8061 .group = tbl_data->group_id
8063 union mlx5_flow_tbl_key table_key = {
8066 uint32_t table_id = table_key.table_id;
8068 tunnel_grp_hash = tbl_data->tunnel ?
8069 tbl_data->tunnel->groups :
8071 he = mlx5_hlist_lookup(tunnel_grp_hash, tunnel_key.val, NULL);
8073 struct tunnel_tbl_entry *tte;
8074 tte = container_of(he, typeof(*tte), hash);
8075 MLX5_ASSERT(tte->flow_table == table_id);
8076 mlx5_hlist_remove(tunnel_grp_hash, he);
8079 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8080 tunnel_flow_tbl_to_id(table_id));
8082 "Table_id %#x tunnel %u group %u released.",
8085 tbl_data->tunnel->tunnel_id : 0,
8086 tbl_data->group_id);
8088 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], tbl_data->idx);
8092 * Release a flow table.
8095 * Pointer to rte_eth_dev structure.
8097 * Table resource to be released.
8100 * Returns 0 if table was released, else return 1;
8103 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
8104 struct mlx5_flow_tbl_resource *tbl)
8106 struct mlx5_priv *priv = dev->data->dev_private;
8107 struct mlx5_dev_ctx_shared *sh = priv->sh;
8108 struct mlx5_flow_tbl_data_entry *tbl_data =
8109 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
8113 return mlx5_hlist_unregister(sh->flow_tbls, &tbl_data->entry);
8117 * Register the flow matcher.
8119 * @param[in, out] dev
8120 * Pointer to rte_eth_dev structure.
8121 * @param[in, out] matcher
8122 * Pointer to flow matcher.
8123 * @param[in, out] key
8124 * Pointer to flow table key.
8125 * @parm[in, out] dev_flow
8126 * Pointer to the dev_flow.
8128 * pointer to error structure.
8131 * 0 on success otherwise -errno and errno is set.
8134 flow_dv_matcher_register(struct rte_eth_dev *dev,
8135 struct mlx5_flow_dv_matcher *matcher,
8136 union mlx5_flow_tbl_key *key,
8137 struct mlx5_flow *dev_flow,
8138 struct rte_flow_error *error)
8140 struct mlx5_priv *priv = dev->data->dev_private;
8141 struct mlx5_dev_ctx_shared *sh = priv->sh;
8142 struct mlx5_flow_dv_matcher *cache_matcher;
8143 struct mlx5dv_flow_matcher_attr dv_attr = {
8144 .type = IBV_FLOW_ATTR_NORMAL,
8145 .match_mask = (void *)&matcher->mask,
8147 struct mlx5_flow_tbl_resource *tbl;
8148 struct mlx5_flow_tbl_data_entry *tbl_data;
8151 tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
8152 key->domain, false, NULL, 0, 0, error);
8154 return -rte_errno; /* No need to refill the error info */
8155 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
8156 /* Lookup from cache. */
8157 LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
8158 if (matcher->crc == cache_matcher->crc &&
8159 matcher->priority == cache_matcher->priority &&
8160 !memcmp((const void *)matcher->mask.buf,
8161 (const void *)cache_matcher->mask.buf,
8162 cache_matcher->mask.size)) {
8164 "%s group %u priority %hd use %s "
8165 "matcher %p: refcnt %d++",
8166 key->domain ? "FDB" : "NIC", key->table_id,
8167 cache_matcher->priority,
8168 key->direction ? "tx" : "rx",
8169 (void *)cache_matcher,
8170 __atomic_load_n(&cache_matcher->refcnt,
8172 __atomic_fetch_add(&cache_matcher->refcnt, 1,
8174 dev_flow->handle->dvh.matcher = cache_matcher;
8175 /* old matcher should not make the table ref++. */
8176 flow_dv_tbl_resource_release(dev, tbl);
8180 /* Register new matcher. */
8181 cache_matcher = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*cache_matcher), 0,
8183 if (!cache_matcher) {
8184 flow_dv_tbl_resource_release(dev, tbl);
8185 return rte_flow_error_set(error, ENOMEM,
8186 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8187 "cannot allocate matcher memory");
8189 *cache_matcher = *matcher;
8190 dv_attr.match_criteria_enable =
8191 flow_dv_matcher_enable(cache_matcher->mask.buf);
8192 dv_attr.priority = matcher->priority;
8194 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
8195 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
8196 &cache_matcher->matcher_object);
8198 mlx5_free(cache_matcher);
8199 #ifdef HAVE_MLX5DV_DR
8200 flow_dv_tbl_resource_release(dev, tbl);
8202 return rte_flow_error_set(error, ENOMEM,
8203 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8204 NULL, "cannot create matcher");
8206 /* Save the table information */
8207 cache_matcher->tbl = tbl;
8208 /* only matcher ref++, table ref++ already done above in get API. */
8209 __atomic_store_n(&cache_matcher->refcnt, 1, __ATOMIC_RELAXED);
8210 LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
8211 dev_flow->handle->dvh.matcher = cache_matcher;
8212 DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
8213 key->domain ? "FDB" : "NIC", key->table_id,
8214 cache_matcher->priority,
8215 key->direction ? "tx" : "rx", (void *)cache_matcher,
8216 __atomic_load_n(&cache_matcher->refcnt, __ATOMIC_RELAXED));
8220 struct mlx5_hlist_entry *
8221 flow_dv_tag_create_cb(struct mlx5_hlist *list, uint64_t key, void *ctx)
8223 struct mlx5_dev_ctx_shared *sh = list->ctx;
8224 struct rte_flow_error *error = ctx;
8225 struct mlx5_flow_dv_tag_resource *entry;
8229 entry = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG], &idx);
8231 rte_flow_error_set(error, ENOMEM,
8232 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8233 "cannot allocate resource memory");
8237 ret = mlx5_flow_os_create_flow_action_tag(key,
8240 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], idx);
8241 rte_flow_error_set(error, ENOMEM,
8242 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8243 NULL, "cannot create action");
8246 return &entry->entry;
8250 * Find existing tag resource or create and register a new one.
8252 * @param dev[in, out]
8253 * Pointer to rte_eth_dev structure.
8254 * @param[in, out] tag_be24
8255 * Tag value in big endian then R-shift 8.
8256 * @parm[in, out] dev_flow
8257 * Pointer to the dev_flow.
8259 * pointer to error structure.
8262 * 0 on success otherwise -errno and errno is set.
8265 flow_dv_tag_resource_register
8266 (struct rte_eth_dev *dev,
8268 struct mlx5_flow *dev_flow,
8269 struct rte_flow_error *error)
8271 struct mlx5_priv *priv = dev->data->dev_private;
8272 struct mlx5_flow_dv_tag_resource *cache_resource;
8273 struct mlx5_hlist_entry *entry;
8275 entry = mlx5_hlist_register(priv->sh->tag_table, tag_be24, error);
8277 cache_resource = container_of
8278 (entry, struct mlx5_flow_dv_tag_resource, entry);
8279 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
8280 dev_flow->dv.tag_resource = cache_resource;
8287 flow_dv_tag_remove_cb(struct mlx5_hlist *list,
8288 struct mlx5_hlist_entry *entry)
8290 struct mlx5_dev_ctx_shared *sh = list->ctx;
8291 struct mlx5_flow_dv_tag_resource *tag =
8292 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
8294 MLX5_ASSERT(tag && sh && tag->action);
8295 claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
8296 DRV_LOG(DEBUG, "Tag %p: removed.", (void *)tag);
8297 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], tag->idx);
8304 * Pointer to Ethernet device.
8309 * 1 while a reference on it exists, 0 when freed.
8312 flow_dv_tag_release(struct rte_eth_dev *dev,
8315 struct mlx5_priv *priv = dev->data->dev_private;
8316 struct mlx5_flow_dv_tag_resource *tag;
8318 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
8321 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
8322 dev->data->port_id, (void *)tag, tag->entry.ref_cnt);
8323 return mlx5_hlist_unregister(priv->sh->tag_table, &tag->entry);
8327 * Translate port ID action to vport.
8330 * Pointer to rte_eth_dev structure.
8332 * Pointer to the port ID action.
8333 * @param[out] dst_port_id
8334 * The target port ID.
8336 * Pointer to the error structure.
8339 * 0 on success, a negative errno value otherwise and rte_errno is set.
8342 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
8343 const struct rte_flow_action *action,
8344 uint32_t *dst_port_id,
8345 struct rte_flow_error *error)
8348 struct mlx5_priv *priv;
8349 const struct rte_flow_action_port_id *conf =
8350 (const struct rte_flow_action_port_id *)action->conf;
8352 port = conf->original ? dev->data->port_id : conf->id;
8353 priv = mlx5_port_to_eswitch_info(port, false);
8355 return rte_flow_error_set(error, -rte_errno,
8356 RTE_FLOW_ERROR_TYPE_ACTION,
8358 "No eswitch info was found for port");
8359 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
8361 * This parameter is transferred to
8362 * mlx5dv_dr_action_create_dest_ib_port().
8364 *dst_port_id = priv->dev_port;
8367 * Legacy mode, no LAG configurations is supported.
8368 * This parameter is transferred to
8369 * mlx5dv_dr_action_create_dest_vport().
8371 *dst_port_id = priv->vport_id;
8377 * Create a counter with aging configuration.
8380 * Pointer to rte_eth_dev structure.
8382 * Pointer to the counter action configuration.
8384 * Pointer to the aging action configuration.
8387 * Index to flow counter on success, 0 otherwise.
8390 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
8391 struct mlx5_flow *dev_flow,
8392 const struct rte_flow_action_count *count,
8393 const struct rte_flow_action_age *age)
8396 struct mlx5_age_param *age_param;
8398 if (count && count->shared)
8399 counter = flow_dv_counter_get_shared(dev, count->id);
8401 counter = flow_dv_counter_alloc(dev, !!age);
8402 if (!counter || age == NULL)
8404 age_param = flow_dv_counter_idx_get_age(dev, counter);
8405 age_param->context = age->context ? age->context :
8406 (void *)(uintptr_t)(dev_flow->flow_idx);
8407 age_param->timeout = age->timeout;
8408 age_param->port_id = dev->data->port_id;
8409 __atomic_store_n(&age_param->sec_since_last_hit, 0, __ATOMIC_RELAXED);
8410 __atomic_store_n(&age_param->state, AGE_CANDIDATE, __ATOMIC_RELAXED);
8414 * Add Tx queue matcher
8417 * Pointer to the dev struct.
8418 * @param[in, out] matcher
8420 * @param[in, out] key
8421 * Flow matcher value.
8423 * Flow pattern to translate.
8425 * Item is inner pattern.
8428 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
8429 void *matcher, void *key,
8430 const struct rte_flow_item *item)
8432 const struct mlx5_rte_flow_item_tx_queue *queue_m;
8433 const struct mlx5_rte_flow_item_tx_queue *queue_v;
8435 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8437 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8438 struct mlx5_txq_ctrl *txq;
8442 queue_m = (const void *)item->mask;
8445 queue_v = (const void *)item->spec;
8448 txq = mlx5_txq_get(dev, queue_v->queue);
8451 queue = txq->obj->sq->id;
8452 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
8453 MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
8454 queue & queue_m->queue);
8455 mlx5_txq_release(dev, queue_v->queue);
8459 * Set the hash fields according to the @p flow information.
8461 * @param[in] dev_flow
8462 * Pointer to the mlx5_flow.
8463 * @param[in] rss_desc
8464 * Pointer to the mlx5_flow_rss_desc.
8467 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
8468 struct mlx5_flow_rss_desc *rss_desc)
8470 uint64_t items = dev_flow->handle->layers;
8472 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
8474 dev_flow->hash_fields = 0;
8475 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
8476 if (rss_desc->level >= 2) {
8477 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
8481 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
8482 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
8483 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
8484 if (rss_types & ETH_RSS_L3_SRC_ONLY)
8485 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
8486 else if (rss_types & ETH_RSS_L3_DST_ONLY)
8487 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
8489 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
8491 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
8492 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
8493 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
8494 if (rss_types & ETH_RSS_L3_SRC_ONLY)
8495 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
8496 else if (rss_types & ETH_RSS_L3_DST_ONLY)
8497 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
8499 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
8502 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
8503 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
8504 if (rss_types & ETH_RSS_UDP) {
8505 if (rss_types & ETH_RSS_L4_SRC_ONLY)
8506 dev_flow->hash_fields |=
8507 IBV_RX_HASH_SRC_PORT_UDP;
8508 else if (rss_types & ETH_RSS_L4_DST_ONLY)
8509 dev_flow->hash_fields |=
8510 IBV_RX_HASH_DST_PORT_UDP;
8512 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
8514 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
8515 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
8516 if (rss_types & ETH_RSS_TCP) {
8517 if (rss_types & ETH_RSS_L4_SRC_ONLY)
8518 dev_flow->hash_fields |=
8519 IBV_RX_HASH_SRC_PORT_TCP;
8520 else if (rss_types & ETH_RSS_L4_DST_ONLY)
8521 dev_flow->hash_fields |=
8522 IBV_RX_HASH_DST_PORT_TCP;
8524 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
8530 * Prepare an Rx Hash queue.
8533 * Pointer to Ethernet device.
8534 * @param[in] dev_flow
8535 * Pointer to the mlx5_flow.
8536 * @param[in] rss_desc
8537 * Pointer to the mlx5_flow_rss_desc.
8538 * @param[out] hrxq_idx
8539 * Hash Rx queue index.
8542 * The Verbs/DevX object initialised, NULL otherwise and rte_errno is set.
8544 static struct mlx5_hrxq *
8545 flow_dv_hrxq_prepare(struct rte_eth_dev *dev,
8546 struct mlx5_flow *dev_flow,
8547 struct mlx5_flow_rss_desc *rss_desc,
8550 struct mlx5_priv *priv = dev->data->dev_private;
8551 struct mlx5_flow_handle *dh = dev_flow->handle;
8552 struct mlx5_hrxq *hrxq;
8554 MLX5_ASSERT(rss_desc->queue_num);
8555 rss_desc->key_len = MLX5_RSS_HASH_KEY_LEN;
8556 rss_desc->hash_fields = dev_flow->hash_fields;
8557 rss_desc->tunnel = !!(dh->layers & MLX5_FLOW_LAYER_TUNNEL);
8558 rss_desc->standalone = false;
8559 *hrxq_idx = mlx5_hrxq_get(dev, rss_desc);
8562 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8568 * Find existing sample resource or create and register a new one.
8570 * @param[in, out] dev
8571 * Pointer to rte_eth_dev structure.
8573 * Attributes of flow that includes this item.
8574 * @param[in] resource
8575 * Pointer to sample resource.
8576 * @parm[in, out] dev_flow
8577 * Pointer to the dev_flow.
8578 * @param[in, out] sample_dv_actions
8579 * Pointer to sample actions list.
8581 * pointer to error structure.
8584 * 0 on success otherwise -errno and errno is set.
8587 flow_dv_sample_resource_register(struct rte_eth_dev *dev,
8588 const struct rte_flow_attr *attr,
8589 struct mlx5_flow_dv_sample_resource *resource,
8590 struct mlx5_flow *dev_flow,
8591 void **sample_dv_actions,
8592 struct rte_flow_error *error)
8594 struct mlx5_flow_dv_sample_resource *cache_resource;
8595 struct mlx5dv_dr_flow_sampler_attr sampler_attr;
8596 struct mlx5_priv *priv = dev->data->dev_private;
8597 struct mlx5_dev_ctx_shared *sh = priv->sh;
8598 struct mlx5_flow_tbl_resource *tbl;
8600 const uint32_t next_ft_step = 1;
8601 uint32_t next_ft_id = resource->ft_id + next_ft_step;
8603 /* Lookup a matching resource from cache. */
8604 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_SAMPLE], sh->sample_action_list,
8605 idx, cache_resource, next) {
8606 if (resource->ratio == cache_resource->ratio &&
8607 resource->ft_type == cache_resource->ft_type &&
8608 resource->ft_id == cache_resource->ft_id &&
8609 resource->set_action == cache_resource->set_action &&
8610 !memcmp((void *)&resource->sample_act,
8611 (void *)&cache_resource->sample_act,
8612 sizeof(struct mlx5_flow_sub_actions_list))) {
8613 DRV_LOG(DEBUG, "sample resource %p: refcnt %d++",
8614 (void *)cache_resource,
8615 __atomic_load_n(&cache_resource->refcnt,
8617 __atomic_fetch_add(&cache_resource->refcnt, 1,
8619 dev_flow->handle->dvh.rix_sample = idx;
8620 dev_flow->dv.sample_res = cache_resource;
8624 /* Register new sample resource. */
8625 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE],
8626 &dev_flow->handle->dvh.rix_sample);
8627 if (!cache_resource)
8628 return rte_flow_error_set(error, ENOMEM,
8629 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8631 "cannot allocate resource memory");
8632 *cache_resource = *resource;
8633 /* Create normal path table level */
8634 tbl = flow_dv_tbl_resource_get(dev, next_ft_id,
8635 attr->egress, attr->transfer,
8636 dev_flow->external, NULL, 0, 0, error);
8638 rte_flow_error_set(error, ENOMEM,
8639 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8641 "fail to create normal path table "
8645 cache_resource->normal_path_tbl = tbl;
8646 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
8647 cache_resource->default_miss =
8648 mlx5_glue->dr_create_flow_action_default_miss();
8649 if (!cache_resource->default_miss) {
8650 rte_flow_error_set(error, ENOMEM,
8651 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8653 "cannot create default miss "
8657 sample_dv_actions[resource->sample_act.actions_num++] =
8658 cache_resource->default_miss;
8660 /* Create a DR sample action */
8661 sampler_attr.sample_ratio = cache_resource->ratio;
8662 sampler_attr.default_next_table = tbl->obj;
8663 sampler_attr.num_sample_actions = resource->sample_act.actions_num;
8664 sampler_attr.sample_actions = (struct mlx5dv_dr_action **)
8665 &sample_dv_actions[0];
8666 sampler_attr.action = cache_resource->set_action;
8667 cache_resource->verbs_action =
8668 mlx5_glue->dr_create_flow_action_sampler(&sampler_attr);
8669 if (!cache_resource->verbs_action) {
8670 rte_flow_error_set(error, ENOMEM,
8671 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8672 NULL, "cannot create sample action");
8675 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
8676 ILIST_INSERT(sh->ipool[MLX5_IPOOL_SAMPLE], &sh->sample_action_list,
8677 dev_flow->handle->dvh.rix_sample, cache_resource,
8679 dev_flow->dv.sample_res = cache_resource;
8680 DRV_LOG(DEBUG, "new sample resource %p: refcnt %d++",
8681 (void *)cache_resource,
8682 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
8685 if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
8686 if (cache_resource->default_miss)
8687 claim_zero(mlx5_glue->destroy_flow_action
8688 (cache_resource->default_miss));
8690 if (cache_resource->sample_idx.rix_hrxq &&
8691 !mlx5_hrxq_release(dev,
8692 cache_resource->sample_idx.rix_hrxq))
8693 cache_resource->sample_idx.rix_hrxq = 0;
8694 if (cache_resource->sample_idx.rix_tag &&
8695 !flow_dv_tag_release(dev,
8696 cache_resource->sample_idx.rix_tag))
8697 cache_resource->sample_idx.rix_tag = 0;
8698 if (cache_resource->sample_idx.cnt) {
8699 flow_dv_counter_release(dev,
8700 cache_resource->sample_idx.cnt);
8701 cache_resource->sample_idx.cnt = 0;
8704 if (cache_resource->normal_path_tbl)
8705 flow_dv_tbl_resource_release(dev,
8706 cache_resource->normal_path_tbl);
8707 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_SAMPLE],
8708 dev_flow->handle->dvh.rix_sample);
8709 dev_flow->handle->dvh.rix_sample = 0;
8714 * Find existing destination array resource or create and register a new one.
8716 * @param[in, out] dev
8717 * Pointer to rte_eth_dev structure.
8719 * Attributes of flow that includes this item.
8720 * @param[in] resource
8721 * Pointer to destination array resource.
8722 * @parm[in, out] dev_flow
8723 * Pointer to the dev_flow.
8725 * pointer to error structure.
8728 * 0 on success otherwise -errno and errno is set.
8731 flow_dv_dest_array_resource_register(struct rte_eth_dev *dev,
8732 const struct rte_flow_attr *attr,
8733 struct mlx5_flow_dv_dest_array_resource *resource,
8734 struct mlx5_flow *dev_flow,
8735 struct rte_flow_error *error)
8737 struct mlx5_flow_dv_dest_array_resource *cache_resource;
8738 struct mlx5dv_dr_action_dest_attr *dest_attr[MLX5_MAX_DEST_NUM] = { 0 };
8739 struct mlx5dv_dr_action_dest_reformat dest_reformat[MLX5_MAX_DEST_NUM];
8740 struct mlx5_priv *priv = dev->data->dev_private;
8741 struct mlx5_dev_ctx_shared *sh = priv->sh;
8742 struct mlx5_flow_sub_actions_list *sample_act;
8743 struct mlx5dv_dr_domain *domain;
8746 /* Lookup a matching resource from cache. */
8747 ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8748 sh->dest_array_list,
8749 idx, cache_resource, next) {
8750 if (resource->num_of_dest == cache_resource->num_of_dest &&
8751 resource->ft_type == cache_resource->ft_type &&
8752 !memcmp((void *)cache_resource->sample_act,
8753 (void *)resource->sample_act,
8754 (resource->num_of_dest *
8755 sizeof(struct mlx5_flow_sub_actions_list)))) {
8756 DRV_LOG(DEBUG, "dest array resource %p: refcnt %d++",
8757 (void *)cache_resource,
8758 __atomic_load_n(&cache_resource->refcnt,
8760 __atomic_fetch_add(&cache_resource->refcnt, 1,
8762 dev_flow->handle->dvh.rix_dest_array = idx;
8763 dev_flow->dv.dest_array_res = cache_resource;
8767 /* Register new destination array resource. */
8768 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8769 &dev_flow->handle->dvh.rix_dest_array);
8770 if (!cache_resource)
8771 return rte_flow_error_set(error, ENOMEM,
8772 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8774 "cannot allocate resource memory");
8775 *cache_resource = *resource;
8777 domain = sh->fdb_domain;
8778 else if (attr->ingress)
8779 domain = sh->rx_domain;
8781 domain = sh->tx_domain;
8782 for (idx = 0; idx < resource->num_of_dest; idx++) {
8783 dest_attr[idx] = (struct mlx5dv_dr_action_dest_attr *)
8784 mlx5_malloc(MLX5_MEM_ZERO,
8785 sizeof(struct mlx5dv_dr_action_dest_attr),
8787 if (!dest_attr[idx]) {
8788 rte_flow_error_set(error, ENOMEM,
8789 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8791 "cannot allocate resource memory");
8794 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST;
8795 sample_act = &resource->sample_act[idx];
8796 if (sample_act->action_flags == MLX5_FLOW_ACTION_QUEUE) {
8797 dest_attr[idx]->dest = sample_act->dr_queue_action;
8798 } else if (sample_act->action_flags ==
8799 (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_ENCAP)) {
8800 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST_REFORMAT;
8801 dest_attr[idx]->dest_reformat = &dest_reformat[idx];
8802 dest_attr[idx]->dest_reformat->reformat =
8803 sample_act->dr_encap_action;
8804 dest_attr[idx]->dest_reformat->dest =
8805 sample_act->dr_port_id_action;
8806 } else if (sample_act->action_flags ==
8807 MLX5_FLOW_ACTION_PORT_ID) {
8808 dest_attr[idx]->dest = sample_act->dr_port_id_action;
8811 /* create a dest array actioin */
8812 cache_resource->action = mlx5_glue->dr_create_flow_action_dest_array
8814 cache_resource->num_of_dest,
8816 if (!cache_resource->action) {
8817 rte_flow_error_set(error, ENOMEM,
8818 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8820 "cannot create destination array action");
8823 __atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
8824 ILIST_INSERT(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8825 &sh->dest_array_list,
8826 dev_flow->handle->dvh.rix_dest_array, cache_resource,
8828 dev_flow->dv.dest_array_res = cache_resource;
8829 DRV_LOG(DEBUG, "new destination array resource %p: refcnt %d++",
8830 (void *)cache_resource,
8831 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
8832 for (idx = 0; idx < resource->num_of_dest; idx++)
8833 mlx5_free(dest_attr[idx]);
8836 for (idx = 0; idx < resource->num_of_dest; idx++) {
8837 struct mlx5_flow_sub_actions_idx *act_res =
8838 &cache_resource->sample_idx[idx];
8839 if (act_res->rix_hrxq &&
8840 !mlx5_hrxq_release(dev,
8842 act_res->rix_hrxq = 0;
8843 if (act_res->rix_encap_decap &&
8844 !flow_dv_encap_decap_resource_release(dev,
8845 act_res->rix_encap_decap))
8846 act_res->rix_encap_decap = 0;
8847 if (act_res->rix_port_id_action &&
8848 !flow_dv_port_id_action_resource_release(dev,
8849 act_res->rix_port_id_action))
8850 act_res->rix_port_id_action = 0;
8852 mlx5_free(dest_attr[idx]);
8855 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8856 dev_flow->handle->dvh.rix_dest_array);
8857 dev_flow->handle->dvh.rix_dest_array = 0;
8862 * Convert Sample action to DV specification.
8865 * Pointer to rte_eth_dev structure.
8867 * Pointer to action structure.
8868 * @param[in, out] dev_flow
8869 * Pointer to the mlx5_flow.
8871 * Pointer to the flow attributes.
8872 * @param[in, out] num_of_dest
8873 * Pointer to the num of destination.
8874 * @param[in, out] sample_actions
8875 * Pointer to sample actions list.
8876 * @param[in, out] res
8877 * Pointer to sample resource.
8879 * Pointer to the error structure.
8882 * 0 on success, a negative errno value otherwise and rte_errno is set.
8885 flow_dv_translate_action_sample(struct rte_eth_dev *dev,
8886 const struct rte_flow_action *action,
8887 struct mlx5_flow *dev_flow,
8888 const struct rte_flow_attr *attr,
8889 uint32_t *num_of_dest,
8890 void **sample_actions,
8891 struct mlx5_flow_dv_sample_resource *res,
8892 struct rte_flow_error *error)
8894 struct mlx5_priv *priv = dev->data->dev_private;
8895 const struct rte_flow_action_sample *sample_action;
8896 const struct rte_flow_action *sub_actions;
8897 const struct rte_flow_action_queue *queue;
8898 struct mlx5_flow_sub_actions_list *sample_act;
8899 struct mlx5_flow_sub_actions_idx *sample_idx;
8900 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
8901 struct mlx5_flow_rss_desc *rss_desc;
8902 uint64_t action_flags = 0;
8905 rss_desc = &wks->rss_desc[!!wks->flow_nested_idx];
8906 sample_act = &res->sample_act;
8907 sample_idx = &res->sample_idx;
8908 sample_action = (const struct rte_flow_action_sample *)action->conf;
8909 res->ratio = sample_action->ratio;
8910 sub_actions = sample_action->actions;
8911 for (; sub_actions->type != RTE_FLOW_ACTION_TYPE_END; sub_actions++) {
8912 int type = sub_actions->type;
8913 uint32_t pre_rix = 0;
8916 case RTE_FLOW_ACTION_TYPE_QUEUE:
8918 struct mlx5_hrxq *hrxq;
8921 queue = sub_actions->conf;
8922 rss_desc->queue_num = 1;
8923 rss_desc->queue[0] = queue->index;
8924 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
8925 rss_desc, &hrxq_idx);
8927 return rte_flow_error_set
8929 RTE_FLOW_ERROR_TYPE_ACTION,
8931 "cannot create fate queue");
8932 sample_act->dr_queue_action = hrxq->action;
8933 sample_idx->rix_hrxq = hrxq_idx;
8934 sample_actions[sample_act->actions_num++] =
8937 action_flags |= MLX5_FLOW_ACTION_QUEUE;
8938 if (action_flags & MLX5_FLOW_ACTION_MARK)
8939 dev_flow->handle->rix_hrxq = hrxq_idx;
8940 dev_flow->handle->fate_action =
8941 MLX5_FLOW_FATE_QUEUE;
8944 case RTE_FLOW_ACTION_TYPE_MARK:
8946 uint32_t tag_be = mlx5_flow_mark_set
8947 (((const struct rte_flow_action_mark *)
8948 (sub_actions->conf))->id);
8950 dev_flow->handle->mark = 1;
8951 pre_rix = dev_flow->handle->dvh.rix_tag;
8952 /* Save the mark resource before sample */
8953 pre_r = dev_flow->dv.tag_resource;
8954 if (flow_dv_tag_resource_register(dev, tag_be,
8957 MLX5_ASSERT(dev_flow->dv.tag_resource);
8958 sample_act->dr_tag_action =
8959 dev_flow->dv.tag_resource->action;
8960 sample_idx->rix_tag =
8961 dev_flow->handle->dvh.rix_tag;
8962 sample_actions[sample_act->actions_num++] =
8963 sample_act->dr_tag_action;
8964 /* Recover the mark resource after sample */
8965 dev_flow->dv.tag_resource = pre_r;
8966 dev_flow->handle->dvh.rix_tag = pre_rix;
8967 action_flags |= MLX5_FLOW_ACTION_MARK;
8970 case RTE_FLOW_ACTION_TYPE_COUNT:
8974 counter = flow_dv_translate_create_counter(dev,
8975 dev_flow, sub_actions->conf, 0);
8977 return rte_flow_error_set
8979 RTE_FLOW_ERROR_TYPE_ACTION,
8981 "cannot create counter"
8983 sample_idx->cnt = counter;
8984 sample_act->dr_cnt_action =
8985 (flow_dv_counter_get_by_idx(dev,
8986 counter, NULL))->action;
8987 sample_actions[sample_act->actions_num++] =
8988 sample_act->dr_cnt_action;
8989 action_flags |= MLX5_FLOW_ACTION_COUNT;
8992 case RTE_FLOW_ACTION_TYPE_PORT_ID:
8994 struct mlx5_flow_dv_port_id_action_resource
8996 uint32_t port_id = 0;
8998 memset(&port_id_resource, 0, sizeof(port_id_resource));
8999 /* Save the port id resource before sample */
9000 pre_rix = dev_flow->handle->rix_port_id_action;
9001 pre_r = dev_flow->dv.port_id_action;
9002 if (flow_dv_translate_action_port_id(dev, sub_actions,
9005 port_id_resource.port_id = port_id;
9006 if (flow_dv_port_id_action_resource_register
9007 (dev, &port_id_resource, dev_flow, error))
9009 sample_act->dr_port_id_action =
9010 dev_flow->dv.port_id_action->action;
9011 sample_idx->rix_port_id_action =
9012 dev_flow->handle->rix_port_id_action;
9013 sample_actions[sample_act->actions_num++] =
9014 sample_act->dr_port_id_action;
9015 /* Recover the port id resource after sample */
9016 dev_flow->dv.port_id_action = pre_r;
9017 dev_flow->handle->rix_port_id_action = pre_rix;
9019 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9022 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
9023 /* Save the encap resource before sample */
9024 pre_rix = dev_flow->handle->dvh.rix_encap_decap;
9025 pre_r = dev_flow->dv.encap_decap;
9026 if (flow_dv_create_action_l2_encap(dev, sub_actions,
9031 sample_act->dr_encap_action =
9032 dev_flow->dv.encap_decap->action;
9033 sample_idx->rix_encap_decap =
9034 dev_flow->handle->dvh.rix_encap_decap;
9035 sample_actions[sample_act->actions_num++] =
9036 sample_act->dr_encap_action;
9037 /* Recover the encap resource after sample */
9038 dev_flow->dv.encap_decap = pre_r;
9039 dev_flow->handle->dvh.rix_encap_decap = pre_rix;
9040 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9043 return rte_flow_error_set(error, EINVAL,
9044 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9046 "Not support for sampler action");
9049 sample_act->action_flags = action_flags;
9050 res->ft_id = dev_flow->dv.group;
9051 if (attr->transfer) {
9053 uint32_t action_in[MLX5_ST_SZ_DW(set_action_in)];
9054 uint64_t set_action;
9055 } action_ctx = { .set_action = 0 };
9057 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
9058 MLX5_SET(set_action_in, action_ctx.action_in, action_type,
9059 MLX5_MODIFICATION_TYPE_SET);
9060 MLX5_SET(set_action_in, action_ctx.action_in, field,
9061 MLX5_MODI_META_REG_C_0);
9062 MLX5_SET(set_action_in, action_ctx.action_in, data,
9063 priv->vport_meta_tag);
9064 res->set_action = action_ctx.set_action;
9065 } else if (attr->ingress) {
9066 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9072 * Convert Sample action to DV specification.
9075 * Pointer to rte_eth_dev structure.
9076 * @param[in, out] dev_flow
9077 * Pointer to the mlx5_flow.
9079 * Pointer to the flow attributes.
9080 * @param[in] num_of_dest
9081 * The num of destination.
9082 * @param[in, out] res
9083 * Pointer to sample resource.
9084 * @param[in, out] mdest_res
9085 * Pointer to destination array resource.
9086 * @param[in] sample_actions
9087 * Pointer to sample path actions list.
9088 * @param[in] action_flags
9089 * Holds the actions detected until now.
9091 * Pointer to the error structure.
9094 * 0 on success, a negative errno value otherwise and rte_errno is set.
9097 flow_dv_create_action_sample(struct rte_eth_dev *dev,
9098 struct mlx5_flow *dev_flow,
9099 const struct rte_flow_attr *attr,
9100 uint32_t num_of_dest,
9101 struct mlx5_flow_dv_sample_resource *res,
9102 struct mlx5_flow_dv_dest_array_resource *mdest_res,
9103 void **sample_actions,
9104 uint64_t action_flags,
9105 struct rte_flow_error *error)
9107 /* update normal path action resource into last index of array */
9108 uint32_t dest_index = MLX5_MAX_DEST_NUM - 1;
9109 struct mlx5_flow_sub_actions_list *sample_act =
9110 &mdest_res->sample_act[dest_index];
9111 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
9112 struct mlx5_flow_rss_desc *rss_desc;
9113 uint32_t normal_idx = 0;
9114 struct mlx5_hrxq *hrxq;
9118 rss_desc = &wks->rss_desc[!!wks->flow_nested_idx];
9119 if (num_of_dest > 1) {
9120 if (sample_act->action_flags & MLX5_FLOW_ACTION_QUEUE) {
9121 /* Handle QP action for mirroring */
9122 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
9123 rss_desc, &hrxq_idx);
9125 return rte_flow_error_set
9127 RTE_FLOW_ERROR_TYPE_ACTION,
9129 "cannot create rx queue");
9131 mdest_res->sample_idx[dest_index].rix_hrxq = hrxq_idx;
9132 sample_act->dr_queue_action = hrxq->action;
9133 if (action_flags & MLX5_FLOW_ACTION_MARK)
9134 dev_flow->handle->rix_hrxq = hrxq_idx;
9135 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9137 if (sample_act->action_flags & MLX5_FLOW_ACTION_ENCAP) {
9139 mdest_res->sample_idx[dest_index].rix_encap_decap =
9140 dev_flow->handle->dvh.rix_encap_decap;
9141 sample_act->dr_encap_action =
9142 dev_flow->dv.encap_decap->action;
9144 if (sample_act->action_flags & MLX5_FLOW_ACTION_PORT_ID) {
9146 mdest_res->sample_idx[dest_index].rix_port_id_action =
9147 dev_flow->handle->rix_port_id_action;
9148 sample_act->dr_port_id_action =
9149 dev_flow->dv.port_id_action->action;
9151 sample_act->actions_num = normal_idx;
9152 /* update sample action resource into first index of array */
9153 mdest_res->ft_type = res->ft_type;
9154 memcpy(&mdest_res->sample_idx[0], &res->sample_idx,
9155 sizeof(struct mlx5_flow_sub_actions_idx));
9156 memcpy(&mdest_res->sample_act[0], &res->sample_act,
9157 sizeof(struct mlx5_flow_sub_actions_list));
9158 mdest_res->num_of_dest = num_of_dest;
9159 if (flow_dv_dest_array_resource_register(dev, attr, mdest_res,
9161 return rte_flow_error_set(error, EINVAL,
9162 RTE_FLOW_ERROR_TYPE_ACTION,
9163 NULL, "can't create sample "
9166 if (flow_dv_sample_resource_register(dev, attr, res, dev_flow,
9167 sample_actions, error))
9168 return rte_flow_error_set(error, EINVAL,
9169 RTE_FLOW_ERROR_TYPE_ACTION,
9171 "can't create sample action");
9177 * Fill the flow with DV spec, lock free
9178 * (mutex should be acquired by caller).
9181 * Pointer to rte_eth_dev structure.
9182 * @param[in, out] dev_flow
9183 * Pointer to the sub flow.
9185 * Pointer to the flow attributes.
9187 * Pointer to the list of items.
9188 * @param[in] actions
9189 * Pointer to the list of actions.
9191 * Pointer to the error structure.
9194 * 0 on success, a negative errno value otherwise and rte_errno is set.
9197 __flow_dv_translate(struct rte_eth_dev *dev,
9198 struct mlx5_flow *dev_flow,
9199 const struct rte_flow_attr *attr,
9200 const struct rte_flow_item items[],
9201 const struct rte_flow_action actions[],
9202 struct rte_flow_error *error)
9204 struct mlx5_priv *priv = dev->data->dev_private;
9205 struct mlx5_dev_config *dev_conf = &priv->config;
9206 struct rte_flow *flow = dev_flow->flow;
9207 struct mlx5_flow_handle *handle = dev_flow->handle;
9208 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
9209 struct mlx5_flow_rss_desc *rss_desc;
9210 uint64_t item_flags = 0;
9211 uint64_t last_item = 0;
9212 uint64_t action_flags = 0;
9213 uint64_t priority = attr->priority;
9214 struct mlx5_flow_dv_matcher matcher = {
9216 .size = sizeof(matcher.mask.buf) -
9217 MLX5_ST_SZ_BYTES(fte_match_set_misc4),
9221 bool actions_end = false;
9223 struct mlx5_flow_dv_modify_hdr_resource res;
9224 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
9225 sizeof(struct mlx5_modification_cmd) *
9226 (MLX5_MAX_MODIFY_NUM + 1)];
9228 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
9229 const struct rte_flow_action_count *count = NULL;
9230 const struct rte_flow_action_age *age = NULL;
9231 union flow_dv_attr flow_attr = { .attr = 0 };
9233 union mlx5_flow_tbl_key tbl_key;
9234 uint32_t modify_action_position = UINT32_MAX;
9235 void *match_mask = matcher.mask.buf;
9236 void *match_value = dev_flow->dv.value.buf;
9237 uint8_t next_protocol = 0xff;
9238 struct rte_vlan_hdr vlan = { 0 };
9239 struct mlx5_flow_dv_dest_array_resource mdest_res;
9240 struct mlx5_flow_dv_sample_resource sample_res;
9241 void *sample_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9242 struct mlx5_flow_sub_actions_list *sample_act;
9243 uint32_t sample_act_pos = UINT32_MAX;
9244 uint32_t num_of_dest = 0;
9245 int tmp_actions_n = 0;
9248 const struct mlx5_flow_tunnel *tunnel;
9249 struct flow_grp_info grp_info = {
9250 .external = !!dev_flow->external,
9251 .transfer = !!attr->transfer,
9252 .fdb_def_rule = !!priv->fdb_def_rule,
9256 rss_desc = &wks->rss_desc[!!wks->flow_nested_idx];
9257 memset(&mdest_res, 0, sizeof(struct mlx5_flow_dv_dest_array_resource));
9258 memset(&sample_res, 0, sizeof(struct mlx5_flow_dv_sample_resource));
9259 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
9260 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9261 /* update normal path action resource into last index of array */
9262 sample_act = &mdest_res.sample_act[MLX5_MAX_DEST_NUM - 1];
9263 tunnel = is_flow_tunnel_match_rule(dev, attr, items, actions) ?
9264 flow_items_to_tunnel(items) :
9265 is_flow_tunnel_steer_rule(dev, attr, items, actions) ?
9266 flow_actions_to_tunnel(actions) :
9267 dev_flow->tunnel ? dev_flow->tunnel : NULL;
9268 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
9269 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9270 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
9271 (dev, tunnel, attr, items, actions);
9272 ret = mlx5_flow_group_to_table(dev, tunnel, attr->group, &table,
9276 dev_flow->dv.group = table;
9278 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
9279 if (priority == MLX5_FLOW_PRIO_RSVD)
9280 priority = dev_conf->flow_prio - 1;
9281 /* number of actions must be set to 0 in case of dirty stack. */
9282 mhdr_res->actions_num = 0;
9283 if (is_flow_tunnel_match_rule(dev, attr, items, actions)) {
9285 * do not add decap action if match rule drops packet
9286 * HW rejects rules with decap & drop
9288 bool add_decap = true;
9289 const struct rte_flow_action *ptr = actions;
9290 struct mlx5_flow_tbl_resource *tbl;
9292 for (; ptr->type != RTE_FLOW_ACTION_TYPE_END; ptr++) {
9293 if (ptr->type == RTE_FLOW_ACTION_TYPE_DROP) {
9299 if (flow_dv_create_action_l2_decap(dev, dev_flow,
9303 dev_flow->dv.actions[actions_n++] =
9304 dev_flow->dv.encap_decap->action;
9305 action_flags |= MLX5_FLOW_ACTION_DECAP;
9308 * bind table_id with <group, table> for tunnel match rule.
9309 * Tunnel set rule establishes that bind in JUMP action handler.
9310 * Required for scenario when application creates tunnel match
9311 * rule before tunnel set rule.
9313 tbl = flow_dv_tbl_resource_get(dev, table, attr->egress,
9315 !!dev_flow->external, tunnel,
9316 attr->group, 0, error);
9318 return rte_flow_error_set
9319 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
9320 actions, "cannot register tunnel group");
9322 for (; !actions_end ; actions++) {
9323 const struct rte_flow_action_queue *queue;
9324 const struct rte_flow_action_rss *rss;
9325 const struct rte_flow_action *action = actions;
9326 const uint8_t *rss_key;
9327 const struct rte_flow_action_meter *mtr;
9328 struct mlx5_flow_tbl_resource *tbl;
9329 uint32_t port_id = 0;
9330 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
9331 int action_type = actions->type;
9332 const struct rte_flow_action *found_action = NULL;
9333 struct mlx5_flow_meter *fm = NULL;
9334 uint32_t jump_group = 0;
9336 if (!mlx5_flow_os_action_supported(action_type))
9337 return rte_flow_error_set(error, ENOTSUP,
9338 RTE_FLOW_ERROR_TYPE_ACTION,
9340 "action not supported");
9341 switch (action_type) {
9342 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
9343 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
9345 case RTE_FLOW_ACTION_TYPE_VOID:
9347 case RTE_FLOW_ACTION_TYPE_PORT_ID:
9348 if (flow_dv_translate_action_port_id(dev, action,
9351 port_id_resource.port_id = port_id;
9352 MLX5_ASSERT(!handle->rix_port_id_action);
9353 if (flow_dv_port_id_action_resource_register
9354 (dev, &port_id_resource, dev_flow, error))
9356 dev_flow->dv.actions[actions_n++] =
9357 dev_flow->dv.port_id_action->action;
9358 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9359 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
9360 sample_act->action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9363 case RTE_FLOW_ACTION_TYPE_FLAG:
9364 action_flags |= MLX5_FLOW_ACTION_FLAG;
9365 dev_flow->handle->mark = 1;
9366 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
9367 struct rte_flow_action_mark mark = {
9368 .id = MLX5_FLOW_MARK_DEFAULT,
9371 if (flow_dv_convert_action_mark(dev, &mark,
9375 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
9378 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
9380 * Only one FLAG or MARK is supported per device flow
9381 * right now. So the pointer to the tag resource must be
9382 * zero before the register process.
9384 MLX5_ASSERT(!handle->dvh.rix_tag);
9385 if (flow_dv_tag_resource_register(dev, tag_be,
9388 MLX5_ASSERT(dev_flow->dv.tag_resource);
9389 dev_flow->dv.actions[actions_n++] =
9390 dev_flow->dv.tag_resource->action;
9392 case RTE_FLOW_ACTION_TYPE_MARK:
9393 action_flags |= MLX5_FLOW_ACTION_MARK;
9394 dev_flow->handle->mark = 1;
9395 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
9396 const struct rte_flow_action_mark *mark =
9397 (const struct rte_flow_action_mark *)
9400 if (flow_dv_convert_action_mark(dev, mark,
9404 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
9408 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
9409 /* Legacy (non-extensive) MARK action. */
9410 tag_be = mlx5_flow_mark_set
9411 (((const struct rte_flow_action_mark *)
9412 (actions->conf))->id);
9413 MLX5_ASSERT(!handle->dvh.rix_tag);
9414 if (flow_dv_tag_resource_register(dev, tag_be,
9417 MLX5_ASSERT(dev_flow->dv.tag_resource);
9418 dev_flow->dv.actions[actions_n++] =
9419 dev_flow->dv.tag_resource->action;
9421 case RTE_FLOW_ACTION_TYPE_SET_META:
9422 if (flow_dv_convert_action_set_meta
9423 (dev, mhdr_res, attr,
9424 (const struct rte_flow_action_set_meta *)
9425 actions->conf, error))
9427 action_flags |= MLX5_FLOW_ACTION_SET_META;
9429 case RTE_FLOW_ACTION_TYPE_SET_TAG:
9430 if (flow_dv_convert_action_set_tag
9432 (const struct rte_flow_action_set_tag *)
9433 actions->conf, error))
9435 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9437 case RTE_FLOW_ACTION_TYPE_DROP:
9438 action_flags |= MLX5_FLOW_ACTION_DROP;
9439 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
9441 case RTE_FLOW_ACTION_TYPE_QUEUE:
9442 queue = actions->conf;
9443 rss_desc->queue_num = 1;
9444 rss_desc->queue[0] = queue->index;
9445 action_flags |= MLX5_FLOW_ACTION_QUEUE;
9446 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9447 sample_act->action_flags |= MLX5_FLOW_ACTION_QUEUE;
9450 case RTE_FLOW_ACTION_TYPE_RSS:
9451 rss = actions->conf;
9452 memcpy(rss_desc->queue, rss->queue,
9453 rss->queue_num * sizeof(uint16_t));
9454 rss_desc->queue_num = rss->queue_num;
9455 /* NULL RSS key indicates default RSS key. */
9456 rss_key = !rss->key ? rss_hash_default_key : rss->key;
9457 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
9459 * rss->level and rss.types should be set in advance
9460 * when expanding items for RSS.
9462 action_flags |= MLX5_FLOW_ACTION_RSS;
9463 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9465 case RTE_FLOW_ACTION_TYPE_AGE:
9466 case RTE_FLOW_ACTION_TYPE_COUNT:
9467 if (!dev_conf->devx) {
9468 return rte_flow_error_set
9470 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9472 "count action not supported");
9474 /* Save information first, will apply later. */
9475 if (actions->type == RTE_FLOW_ACTION_TYPE_COUNT)
9476 count = action->conf;
9479 action_flags |= MLX5_FLOW_ACTION_COUNT;
9481 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
9482 dev_flow->dv.actions[actions_n++] =
9483 priv->sh->pop_vlan_action;
9484 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
9486 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
9487 if (!(action_flags &
9488 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
9489 flow_dev_get_vlan_info_from_items(items, &vlan);
9490 vlan.eth_proto = rte_be_to_cpu_16
9491 ((((const struct rte_flow_action_of_push_vlan *)
9492 actions->conf)->ethertype));
9493 found_action = mlx5_flow_find_action
9495 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
9497 mlx5_update_vlan_vid_pcp(found_action, &vlan);
9498 found_action = mlx5_flow_find_action
9500 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
9502 mlx5_update_vlan_vid_pcp(found_action, &vlan);
9503 if (flow_dv_create_action_push_vlan
9504 (dev, attr, &vlan, dev_flow, error))
9506 dev_flow->dv.actions[actions_n++] =
9507 dev_flow->dv.push_vlan_res->action;
9508 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
9510 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
9511 /* of_vlan_push action handled this action */
9512 MLX5_ASSERT(action_flags &
9513 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
9515 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
9516 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
9518 flow_dev_get_vlan_info_from_items(items, &vlan);
9519 mlx5_update_vlan_vid_pcp(actions, &vlan);
9520 /* If no VLAN push - this is a modify header action */
9521 if (flow_dv_convert_action_modify_vlan_vid
9522 (mhdr_res, actions, error))
9524 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
9526 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
9527 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
9528 if (flow_dv_create_action_l2_encap(dev, actions,
9533 dev_flow->dv.actions[actions_n++] =
9534 dev_flow->dv.encap_decap->action;
9535 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9536 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
9537 sample_act->action_flags |=
9538 MLX5_FLOW_ACTION_ENCAP;
9540 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
9541 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
9542 if (flow_dv_create_action_l2_decap(dev, dev_flow,
9546 dev_flow->dv.actions[actions_n++] =
9547 dev_flow->dv.encap_decap->action;
9548 action_flags |= MLX5_FLOW_ACTION_DECAP;
9550 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
9551 /* Handle encap with preceding decap. */
9552 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
9553 if (flow_dv_create_action_raw_encap
9554 (dev, actions, dev_flow, attr, error))
9556 dev_flow->dv.actions[actions_n++] =
9557 dev_flow->dv.encap_decap->action;
9559 /* Handle encap without preceding decap. */
9560 if (flow_dv_create_action_l2_encap
9561 (dev, actions, dev_flow, attr->transfer,
9564 dev_flow->dv.actions[actions_n++] =
9565 dev_flow->dv.encap_decap->action;
9567 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9568 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
9569 sample_act->action_flags |=
9570 MLX5_FLOW_ACTION_ENCAP;
9572 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
9573 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
9575 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
9576 if (flow_dv_create_action_l2_decap
9577 (dev, dev_flow, attr->transfer, error))
9579 dev_flow->dv.actions[actions_n++] =
9580 dev_flow->dv.encap_decap->action;
9582 /* If decap is followed by encap, handle it at encap. */
9583 action_flags |= MLX5_FLOW_ACTION_DECAP;
9585 case RTE_FLOW_ACTION_TYPE_JUMP:
9586 jump_group = ((const struct rte_flow_action_jump *)
9587 action->conf)->group;
9588 grp_info.std_tbl_fix = 0;
9589 ret = mlx5_flow_group_to_table(dev, tunnel,
9595 tbl = flow_dv_tbl_resource_get(dev, table, attr->egress,
9597 !!dev_flow->external,
9598 tunnel, jump_group, 0,
9601 return rte_flow_error_set
9603 RTE_FLOW_ERROR_TYPE_ACTION,
9605 "cannot create jump action.");
9606 if (flow_dv_jump_tbl_resource_register
9607 (dev, tbl, dev_flow, error)) {
9608 flow_dv_tbl_resource_release(dev, tbl);
9609 return rte_flow_error_set
9611 RTE_FLOW_ERROR_TYPE_ACTION,
9613 "cannot create jump action.");
9615 dev_flow->dv.actions[actions_n++] =
9616 dev_flow->dv.jump->action;
9617 action_flags |= MLX5_FLOW_ACTION_JUMP;
9618 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
9620 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
9621 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
9622 if (flow_dv_convert_action_modify_mac
9623 (mhdr_res, actions, error))
9625 action_flags |= actions->type ==
9626 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
9627 MLX5_FLOW_ACTION_SET_MAC_SRC :
9628 MLX5_FLOW_ACTION_SET_MAC_DST;
9630 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
9631 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
9632 if (flow_dv_convert_action_modify_ipv4
9633 (mhdr_res, actions, error))
9635 action_flags |= actions->type ==
9636 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
9637 MLX5_FLOW_ACTION_SET_IPV4_SRC :
9638 MLX5_FLOW_ACTION_SET_IPV4_DST;
9640 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
9641 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
9642 if (flow_dv_convert_action_modify_ipv6
9643 (mhdr_res, actions, error))
9645 action_flags |= actions->type ==
9646 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
9647 MLX5_FLOW_ACTION_SET_IPV6_SRC :
9648 MLX5_FLOW_ACTION_SET_IPV6_DST;
9650 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
9651 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
9652 if (flow_dv_convert_action_modify_tp
9653 (mhdr_res, actions, items,
9654 &flow_attr, dev_flow, !!(action_flags &
9655 MLX5_FLOW_ACTION_DECAP), error))
9657 action_flags |= actions->type ==
9658 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
9659 MLX5_FLOW_ACTION_SET_TP_SRC :
9660 MLX5_FLOW_ACTION_SET_TP_DST;
9662 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
9663 if (flow_dv_convert_action_modify_dec_ttl
9664 (mhdr_res, items, &flow_attr, dev_flow,
9666 MLX5_FLOW_ACTION_DECAP), error))
9668 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
9670 case RTE_FLOW_ACTION_TYPE_SET_TTL:
9671 if (flow_dv_convert_action_modify_ttl
9672 (mhdr_res, actions, items, &flow_attr,
9673 dev_flow, !!(action_flags &
9674 MLX5_FLOW_ACTION_DECAP), error))
9676 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
9678 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
9679 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
9680 if (flow_dv_convert_action_modify_tcp_seq
9681 (mhdr_res, actions, error))
9683 action_flags |= actions->type ==
9684 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
9685 MLX5_FLOW_ACTION_INC_TCP_SEQ :
9686 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
9689 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
9690 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
9691 if (flow_dv_convert_action_modify_tcp_ack
9692 (mhdr_res, actions, error))
9694 action_flags |= actions->type ==
9695 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
9696 MLX5_FLOW_ACTION_INC_TCP_ACK :
9697 MLX5_FLOW_ACTION_DEC_TCP_ACK;
9699 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
9700 if (flow_dv_convert_action_set_reg
9701 (mhdr_res, actions, error))
9703 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9705 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
9706 if (flow_dv_convert_action_copy_mreg
9707 (dev, mhdr_res, actions, error))
9709 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9711 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
9712 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
9713 dev_flow->handle->fate_action =
9714 MLX5_FLOW_FATE_DEFAULT_MISS;
9716 case RTE_FLOW_ACTION_TYPE_METER:
9717 mtr = actions->conf;
9719 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
9722 return rte_flow_error_set(error,
9724 RTE_FLOW_ERROR_TYPE_ACTION,
9727 "or invalid parameters");
9728 flow->meter = fm->idx;
9730 /* Set the meter action. */
9732 fm = mlx5_ipool_get(priv->sh->ipool
9733 [MLX5_IPOOL_MTR], flow->meter);
9735 return rte_flow_error_set(error,
9737 RTE_FLOW_ERROR_TYPE_ACTION,
9740 "or invalid parameters");
9742 dev_flow->dv.actions[actions_n++] =
9743 fm->mfts->meter_action;
9744 action_flags |= MLX5_FLOW_ACTION_METER;
9746 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
9747 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
9750 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
9752 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
9753 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
9756 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
9758 case RTE_FLOW_ACTION_TYPE_SAMPLE:
9759 sample_act_pos = actions_n;
9760 ret = flow_dv_translate_action_sample(dev,
9770 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
9771 /* put encap action into group if work with port id */
9772 if ((action_flags & MLX5_FLOW_ACTION_ENCAP) &&
9773 (action_flags & MLX5_FLOW_ACTION_PORT_ID))
9774 sample_act->action_flags |=
9775 MLX5_FLOW_ACTION_ENCAP;
9777 case RTE_FLOW_ACTION_TYPE_END:
9779 if (mhdr_res->actions_num) {
9780 /* create modify action if needed. */
9781 if (flow_dv_modify_hdr_resource_register
9782 (dev, mhdr_res, dev_flow, error))
9784 dev_flow->dv.actions[modify_action_position] =
9785 handle->dvh.modify_hdr->action;
9787 if (action_flags & MLX5_FLOW_ACTION_COUNT) {
9789 flow_dv_translate_create_counter(dev,
9790 dev_flow, count, age);
9793 return rte_flow_error_set
9795 RTE_FLOW_ERROR_TYPE_ACTION,
9797 "cannot create counter"
9799 dev_flow->dv.actions[actions_n] =
9800 (flow_dv_counter_get_by_idx(dev,
9801 flow->counter, NULL))->action;
9804 if (action_flags & MLX5_FLOW_ACTION_SAMPLE) {
9805 ret = flow_dv_create_action_sample(dev,
9814 return rte_flow_error_set
9816 RTE_FLOW_ERROR_TYPE_ACTION,
9818 "cannot create sample action");
9819 if (num_of_dest > 1) {
9820 dev_flow->dv.actions[sample_act_pos] =
9821 dev_flow->dv.dest_array_res->action;
9823 dev_flow->dv.actions[sample_act_pos] =
9824 dev_flow->dv.sample_res->verbs_action;
9831 if (mhdr_res->actions_num &&
9832 modify_action_position == UINT32_MAX)
9833 modify_action_position = actions_n++;
9836 * For multiple destination (sample action with ratio=1), the encap
9837 * action and port id action will be combined into group action.
9838 * So need remove the original these actions in the flow and only
9839 * use the sample action instead of.
9841 if (num_of_dest > 1 && sample_act->dr_port_id_action) {
9843 void *temp_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9845 for (i = 0; i < actions_n; i++) {
9846 if ((sample_act->dr_encap_action &&
9847 sample_act->dr_encap_action ==
9848 dev_flow->dv.actions[i]) ||
9849 (sample_act->dr_port_id_action &&
9850 sample_act->dr_port_id_action ==
9851 dev_flow->dv.actions[i]))
9853 temp_actions[tmp_actions_n++] = dev_flow->dv.actions[i];
9855 memcpy((void *)dev_flow->dv.actions,
9856 (void *)temp_actions,
9857 tmp_actions_n * sizeof(void *));
9858 actions_n = tmp_actions_n;
9860 dev_flow->dv.actions_n = actions_n;
9861 dev_flow->act_flags = action_flags;
9862 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
9863 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
9864 int item_type = items->type;
9866 if (!mlx5_flow_os_item_supported(item_type))
9867 return rte_flow_error_set(error, ENOTSUP,
9868 RTE_FLOW_ERROR_TYPE_ITEM,
9869 NULL, "item not supported");
9870 switch (item_type) {
9871 case RTE_FLOW_ITEM_TYPE_PORT_ID:
9872 flow_dv_translate_item_port_id(dev, match_mask,
9873 match_value, items);
9874 last_item = MLX5_FLOW_ITEM_PORT_ID;
9876 case RTE_FLOW_ITEM_TYPE_ETH:
9877 flow_dv_translate_item_eth(match_mask, match_value,
9879 dev_flow->dv.group);
9880 matcher.priority = action_flags &
9881 MLX5_FLOW_ACTION_DEFAULT_MISS &&
9882 !dev_flow->external ?
9883 MLX5_PRIORITY_MAP_L3 :
9884 MLX5_PRIORITY_MAP_L2;
9885 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
9886 MLX5_FLOW_LAYER_OUTER_L2;
9888 case RTE_FLOW_ITEM_TYPE_VLAN:
9889 flow_dv_translate_item_vlan(dev_flow,
9890 match_mask, match_value,
9892 dev_flow->dv.group);
9893 matcher.priority = MLX5_PRIORITY_MAP_L2;
9894 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
9895 MLX5_FLOW_LAYER_INNER_VLAN) :
9896 (MLX5_FLOW_LAYER_OUTER_L2 |
9897 MLX5_FLOW_LAYER_OUTER_VLAN);
9899 case RTE_FLOW_ITEM_TYPE_IPV4:
9900 mlx5_flow_tunnel_ip_check(items, next_protocol,
9901 &item_flags, &tunnel);
9902 flow_dv_translate_item_ipv4(match_mask, match_value,
9904 dev_flow->dv.group);
9905 matcher.priority = MLX5_PRIORITY_MAP_L3;
9906 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
9907 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
9908 if (items->mask != NULL &&
9909 ((const struct rte_flow_item_ipv4 *)
9910 items->mask)->hdr.next_proto_id) {
9912 ((const struct rte_flow_item_ipv4 *)
9913 (items->spec))->hdr.next_proto_id;
9915 ((const struct rte_flow_item_ipv4 *)
9916 (items->mask))->hdr.next_proto_id;
9918 /* Reset for inner layer. */
9919 next_protocol = 0xff;
9922 case RTE_FLOW_ITEM_TYPE_IPV6:
9923 mlx5_flow_tunnel_ip_check(items, next_protocol,
9924 &item_flags, &tunnel);
9925 flow_dv_translate_item_ipv6(match_mask, match_value,
9927 dev_flow->dv.group);
9928 matcher.priority = MLX5_PRIORITY_MAP_L3;
9929 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
9930 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
9931 if (items->mask != NULL &&
9932 ((const struct rte_flow_item_ipv6 *)
9933 items->mask)->hdr.proto) {
9935 ((const struct rte_flow_item_ipv6 *)
9936 items->spec)->hdr.proto;
9938 ((const struct rte_flow_item_ipv6 *)
9939 items->mask)->hdr.proto;
9941 /* Reset for inner layer. */
9942 next_protocol = 0xff;
9945 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
9946 flow_dv_translate_item_ipv6_frag_ext(match_mask,
9949 last_item = tunnel ?
9950 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
9951 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
9952 if (items->mask != NULL &&
9953 ((const struct rte_flow_item_ipv6_frag_ext *)
9954 items->mask)->hdr.next_header) {
9956 ((const struct rte_flow_item_ipv6_frag_ext *)
9957 items->spec)->hdr.next_header;
9959 ((const struct rte_flow_item_ipv6_frag_ext *)
9960 items->mask)->hdr.next_header;
9962 /* Reset for inner layer. */
9963 next_protocol = 0xff;
9966 case RTE_FLOW_ITEM_TYPE_TCP:
9967 flow_dv_translate_item_tcp(match_mask, match_value,
9969 matcher.priority = MLX5_PRIORITY_MAP_L4;
9970 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
9971 MLX5_FLOW_LAYER_OUTER_L4_TCP;
9973 case RTE_FLOW_ITEM_TYPE_UDP:
9974 flow_dv_translate_item_udp(match_mask, match_value,
9976 matcher.priority = MLX5_PRIORITY_MAP_L4;
9977 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
9978 MLX5_FLOW_LAYER_OUTER_L4_UDP;
9980 case RTE_FLOW_ITEM_TYPE_GRE:
9981 flow_dv_translate_item_gre(match_mask, match_value,
9983 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9984 last_item = MLX5_FLOW_LAYER_GRE;
9986 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
9987 flow_dv_translate_item_gre_key(match_mask,
9988 match_value, items);
9989 last_item = MLX5_FLOW_LAYER_GRE_KEY;
9991 case RTE_FLOW_ITEM_TYPE_NVGRE:
9992 flow_dv_translate_item_nvgre(match_mask, match_value,
9994 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9995 last_item = MLX5_FLOW_LAYER_GRE;
9997 case RTE_FLOW_ITEM_TYPE_VXLAN:
9998 flow_dv_translate_item_vxlan(match_mask, match_value,
10000 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10001 last_item = MLX5_FLOW_LAYER_VXLAN;
10003 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
10004 flow_dv_translate_item_vxlan_gpe(match_mask,
10005 match_value, items,
10007 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10008 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
10010 case RTE_FLOW_ITEM_TYPE_GENEVE:
10011 flow_dv_translate_item_geneve(match_mask, match_value,
10013 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10014 last_item = MLX5_FLOW_LAYER_GENEVE;
10016 case RTE_FLOW_ITEM_TYPE_MPLS:
10017 flow_dv_translate_item_mpls(match_mask, match_value,
10018 items, last_item, tunnel);
10019 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10020 last_item = MLX5_FLOW_LAYER_MPLS;
10022 case RTE_FLOW_ITEM_TYPE_MARK:
10023 flow_dv_translate_item_mark(dev, match_mask,
10024 match_value, items);
10025 last_item = MLX5_FLOW_ITEM_MARK;
10027 case RTE_FLOW_ITEM_TYPE_META:
10028 flow_dv_translate_item_meta(dev, match_mask,
10029 match_value, attr, items);
10030 last_item = MLX5_FLOW_ITEM_METADATA;
10032 case RTE_FLOW_ITEM_TYPE_ICMP:
10033 flow_dv_translate_item_icmp(match_mask, match_value,
10035 last_item = MLX5_FLOW_LAYER_ICMP;
10037 case RTE_FLOW_ITEM_TYPE_ICMP6:
10038 flow_dv_translate_item_icmp6(match_mask, match_value,
10040 last_item = MLX5_FLOW_LAYER_ICMP6;
10042 case RTE_FLOW_ITEM_TYPE_TAG:
10043 flow_dv_translate_item_tag(dev, match_mask,
10044 match_value, items);
10045 last_item = MLX5_FLOW_ITEM_TAG;
10047 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
10048 flow_dv_translate_mlx5_item_tag(dev, match_mask,
10049 match_value, items);
10050 last_item = MLX5_FLOW_ITEM_TAG;
10052 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
10053 flow_dv_translate_item_tx_queue(dev, match_mask,
10056 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
10058 case RTE_FLOW_ITEM_TYPE_GTP:
10059 flow_dv_translate_item_gtp(match_mask, match_value,
10061 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10062 last_item = MLX5_FLOW_LAYER_GTP;
10064 case RTE_FLOW_ITEM_TYPE_ECPRI:
10065 if (!mlx5_flex_parser_ecpri_exist(dev)) {
10066 /* Create it only the first time to be used. */
10067 ret = mlx5_flex_parser_ecpri_alloc(dev);
10069 return rte_flow_error_set
10071 RTE_FLOW_ERROR_TYPE_ITEM,
10073 "cannot create eCPRI parser");
10075 /* Adjust the length matcher and device flow value. */
10076 matcher.mask.size = MLX5_ST_SZ_BYTES(fte_match_param);
10077 dev_flow->dv.value.size =
10078 MLX5_ST_SZ_BYTES(fte_match_param);
10079 flow_dv_translate_item_ecpri(dev, match_mask,
10080 match_value, items);
10081 /* No other protocol should follow eCPRI layer. */
10082 last_item = MLX5_FLOW_LAYER_ECPRI;
10087 item_flags |= last_item;
10090 * When E-Switch mode is enabled, we have two cases where we need to
10091 * set the source port manually.
10092 * The first one, is in case of Nic steering rule, and the second is
10093 * E-Switch rule where no port_id item was found. In both cases
10094 * the source port is set according the current port in use.
10096 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
10097 (priv->representor || priv->master)) {
10098 if (flow_dv_translate_item_port_id(dev, match_mask,
10099 match_value, NULL))
10102 #ifdef RTE_LIBRTE_MLX5_DEBUG
10103 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
10104 dev_flow->dv.value.buf));
10107 * Layers may be already initialized from prefix flow if this dev_flow
10108 * is the suffix flow.
10110 handle->layers |= item_flags;
10111 if (action_flags & MLX5_FLOW_ACTION_RSS)
10112 flow_dv_hashfields_set(dev_flow, rss_desc);
10113 /* Register matcher. */
10114 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
10115 matcher.mask.size);
10116 matcher.priority = mlx5_flow_adjust_priority(dev, priority,
10118 /* reserved field no needs to be set to 0 here. */
10119 tbl_key.domain = attr->transfer;
10120 tbl_key.direction = attr->egress;
10121 tbl_key.table_id = dev_flow->dv.group;
10122 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
10128 * Set hash RX queue by hash fields (see enum ibv_rx_hash_fields)
10131 * @param[in, out] action
10132 * Shred RSS action holding hash RX queue objects.
10133 * @param[in] hash_fields
10134 * Defines combination of packet fields to participate in RX hash.
10135 * @param[in] tunnel
10137 * @param[in] hrxq_idx
10138 * Hash RX queue index to set.
10141 * 0 on success, otherwise negative errno value.
10144 __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
10145 const uint64_t hash_fields,
10149 uint32_t *hrxqs = tunnel ? action->hrxq : action->hrxq_tunnel;
10151 switch (hash_fields & ~IBV_RX_HASH_INNER) {
10152 case MLX5_RSS_HASH_IPV4:
10153 hrxqs[0] = hrxq_idx;
10155 case MLX5_RSS_HASH_IPV4_TCP:
10156 hrxqs[1] = hrxq_idx;
10158 case MLX5_RSS_HASH_IPV4_UDP:
10159 hrxqs[2] = hrxq_idx;
10161 case MLX5_RSS_HASH_IPV6:
10162 hrxqs[3] = hrxq_idx;
10164 case MLX5_RSS_HASH_IPV6_TCP:
10165 hrxqs[4] = hrxq_idx;
10167 case MLX5_RSS_HASH_IPV6_UDP:
10168 hrxqs[5] = hrxq_idx;
10170 case MLX5_RSS_HASH_NONE:
10171 hrxqs[6] = hrxq_idx;
10179 * Look up for hash RX queue by hash fields (see enum ibv_rx_hash_fields)
10182 * @param[in] action
10183 * Shred RSS action holding hash RX queue objects.
10184 * @param[in] hash_fields
10185 * Defines combination of packet fields to participate in RX hash.
10186 * @param[in] tunnel
10190 * Valid hash RX queue index, otherwise 0.
10193 __flow_dv_action_rss_hrxq_lookup(const struct mlx5_shared_action_rss *action,
10194 const uint64_t hash_fields,
10197 const uint32_t *hrxqs = tunnel ? action->hrxq : action->hrxq_tunnel;
10199 switch (hash_fields & ~IBV_RX_HASH_INNER) {
10200 case MLX5_RSS_HASH_IPV4:
10202 case MLX5_RSS_HASH_IPV4_TCP:
10204 case MLX5_RSS_HASH_IPV4_UDP:
10206 case MLX5_RSS_HASH_IPV6:
10208 case MLX5_RSS_HASH_IPV6_TCP:
10210 case MLX5_RSS_HASH_IPV6_UDP:
10212 case MLX5_RSS_HASH_NONE:
10220 * Retrieves hash RX queue suitable for the *flow*.
10221 * If shared action configured for *flow* suitable hash RX queue will be
10222 * retrieved from attached shared action.
10225 * Shred RSS action holding hash RX queue objects.
10226 * @param[in] dev_flow
10227 * Pointer to the sub flow.
10229 * Pointer to retrieved hash RX queue object.
10232 * Valid hash RX queue index, otherwise 0 and rte_errno is set.
10235 __flow_dv_rss_get_hrxq(struct rte_eth_dev *dev, struct rte_flow *flow,
10236 struct mlx5_flow *dev_flow,
10237 struct mlx5_hrxq **hrxq)
10239 struct mlx5_priv *priv = dev->data->dev_private;
10240 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
10243 if (flow->shared_rss) {
10244 hrxq_idx = __flow_dv_action_rss_hrxq_lookup
10245 (flow->shared_rss, dev_flow->hash_fields,
10246 !!(dev_flow->handle->layers &
10247 MLX5_FLOW_LAYER_TUNNEL));
10249 *hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
10251 __atomic_fetch_add(&(*hrxq)->refcnt, 1,
10255 struct mlx5_flow_rss_desc *rss_desc =
10256 &wks->rss_desc[!!wks->flow_nested_idx];
10258 *hrxq = flow_dv_hrxq_prepare(dev, dev_flow, rss_desc,
10265 * Apply the flow to the NIC, lock free,
10266 * (mutex should be acquired by caller).
10269 * Pointer to the Ethernet device structure.
10270 * @param[in, out] flow
10271 * Pointer to flow structure.
10272 * @param[out] error
10273 * Pointer to error structure.
10276 * 0 on success, a negative errno value otherwise and rte_errno is set.
10279 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
10280 struct rte_flow_error *error)
10282 struct mlx5_flow_dv_workspace *dv;
10283 struct mlx5_flow_handle *dh;
10284 struct mlx5_flow_handle_dv *dv_h;
10285 struct mlx5_flow *dev_flow;
10286 struct mlx5_priv *priv = dev->data->dev_private;
10287 uint32_t handle_idx;
10291 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
10294 for (idx = wks->flow_idx - 1; idx >= wks->flow_nested_idx; idx--) {
10295 dev_flow = &wks->flows[idx];
10296 dv = &dev_flow->dv;
10297 dh = dev_flow->handle;
10300 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
10301 if (dv->transfer) {
10302 dv->actions[n++] = priv->sh->esw_drop_action;
10304 MLX5_ASSERT(priv->drop_queue.hrxq);
10306 priv->drop_queue.hrxq->action;
10308 } else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
10309 !dv_h->rix_sample && !dv_h->rix_dest_array) {
10310 struct mlx5_hrxq *hrxq = NULL;
10311 uint32_t hrxq_idx = __flow_dv_rss_get_hrxq
10312 (dev, flow, dev_flow, &hrxq);
10316 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10317 "cannot get hash queue");
10320 dh->rix_hrxq = hrxq_idx;
10321 dv->actions[n++] = hrxq->action;
10322 } else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
10323 if (!priv->sh->default_miss_action) {
10326 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10327 "default miss action not be created.");
10330 dv->actions[n++] = priv->sh->default_miss_action;
10332 err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
10333 (void *)&dv->value, n,
10334 dv->actions, &dh->drv_flow);
10336 rte_flow_error_set(error, errno,
10337 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10339 "hardware refuses to create flow");
10342 if (priv->vmwa_context &&
10343 dh->vf_vlan.tag && !dh->vf_vlan.created) {
10345 * The rule contains the VLAN pattern.
10346 * For VF we are going to create VLAN
10347 * interface to make hypervisor set correct
10348 * e-Switch vport context.
10350 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
10355 err = rte_errno; /* Save rte_errno before cleanup. */
10356 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
10357 handle_idx, dh, next) {
10358 /* hrxq is union, don't clear it if the flag is not set. */
10359 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE && dh->rix_hrxq) {
10360 mlx5_hrxq_release(dev, dh->rix_hrxq);
10363 if (dh->vf_vlan.tag && dh->vf_vlan.created)
10364 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
10366 rte_errno = err; /* Restore rte_errno. */
10371 * Release the flow matcher.
10374 * Pointer to Ethernet device.
10376 * Pointer to mlx5_flow_handle.
10379 * 1 while a reference on it exists, 0 when freed.
10382 flow_dv_matcher_release(struct rte_eth_dev *dev,
10383 struct mlx5_flow_handle *handle)
10385 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
10387 MLX5_ASSERT(matcher->matcher_object);
10388 DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
10389 dev->data->port_id, (void *)matcher,
10390 __atomic_load_n(&matcher->refcnt, __ATOMIC_RELAXED));
10391 if (__atomic_sub_fetch(&matcher->refcnt, 1, __ATOMIC_RELAXED) == 0) {
10392 claim_zero(mlx5_flow_os_destroy_flow_matcher
10393 (matcher->matcher_object));
10394 LIST_REMOVE(matcher, next);
10395 /* table ref-- in release interface. */
10396 flow_dv_tbl_resource_release(dev, matcher->tbl);
10397 mlx5_free(matcher);
10398 DRV_LOG(DEBUG, "port %u matcher %p: removed",
10399 dev->data->port_id, (void *)matcher);
10406 * Release encap_decap resource.
10409 * Pointer to the hash list.
10411 * Pointer to exist resource entry object.
10414 flow_dv_encap_decap_remove_cb(struct mlx5_hlist *list,
10415 struct mlx5_hlist_entry *entry)
10417 struct mlx5_dev_ctx_shared *sh = list->ctx;
10418 struct mlx5_flow_dv_encap_decap_resource *res =
10419 container_of(entry, typeof(*res), entry);
10421 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
10422 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], res->idx);
10426 * Release an encap/decap resource.
10429 * Pointer to Ethernet device.
10430 * @param encap_decap_idx
10431 * Index of encap decap resource.
10434 * 1 while a reference on it exists, 0 when freed.
10437 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
10438 uint32_t encap_decap_idx)
10440 struct mlx5_priv *priv = dev->data->dev_private;
10441 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
10443 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
10445 if (!cache_resource)
10447 MLX5_ASSERT(cache_resource->action);
10448 return mlx5_hlist_unregister(priv->sh->encaps_decaps,
10449 &cache_resource->entry);
10453 * Release an jump to table action resource.
10456 * Pointer to Ethernet device.
10458 * Pointer to mlx5_flow_handle.
10461 * 1 while a reference on it exists, 0 when freed.
10464 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
10465 struct mlx5_flow_handle *handle)
10467 struct mlx5_priv *priv = dev->data->dev_private;
10468 struct mlx5_flow_tbl_data_entry *tbl_data;
10470 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
10474 return flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
10478 flow_dv_modify_remove_cb(struct mlx5_hlist *list __rte_unused,
10479 struct mlx5_hlist_entry *entry)
10481 struct mlx5_flow_dv_modify_hdr_resource *res =
10482 container_of(entry, typeof(*res), entry);
10484 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
10489 * Release a modify-header resource.
10492 * Pointer to Ethernet device.
10494 * Pointer to mlx5_flow_handle.
10497 * 1 while a reference on it exists, 0 when freed.
10500 flow_dv_modify_hdr_resource_release(struct rte_eth_dev *dev,
10501 struct mlx5_flow_handle *handle)
10503 struct mlx5_priv *priv = dev->data->dev_private;
10504 struct mlx5_flow_dv_modify_hdr_resource *entry = handle->dvh.modify_hdr;
10506 MLX5_ASSERT(entry->action);
10507 return mlx5_hlist_unregister(priv->sh->modify_cmds, &entry->entry);
10511 * Release port ID action resource.
10514 * Pointer to Ethernet device.
10516 * Pointer to mlx5_flow_handle.
10519 * 1 while a reference on it exists, 0 when freed.
10522 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
10525 struct mlx5_priv *priv = dev->data->dev_private;
10526 struct mlx5_flow_dv_port_id_action_resource *cache_resource;
10527 uint32_t idx = port_id;
10529 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
10531 if (!cache_resource)
10533 MLX5_ASSERT(cache_resource->action);
10534 DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
10535 (void *)cache_resource,
10536 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10537 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10538 __ATOMIC_RELAXED) == 0) {
10539 claim_zero(mlx5_flow_os_destroy_flow_action
10540 (cache_resource->action));
10541 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
10542 &priv->sh->port_id_action_list, idx,
10543 cache_resource, next);
10544 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
10545 DRV_LOG(DEBUG, "port id action resource %p: removed",
10546 (void *)cache_resource);
10553 * Release push vlan action resource.
10556 * Pointer to Ethernet device.
10558 * Pointer to mlx5_flow_handle.
10561 * 1 while a reference on it exists, 0 when freed.
10564 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
10565 struct mlx5_flow_handle *handle)
10567 struct mlx5_priv *priv = dev->data->dev_private;
10568 uint32_t idx = handle->dvh.rix_push_vlan;
10569 struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
10571 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
10573 if (!cache_resource)
10575 MLX5_ASSERT(cache_resource->action);
10576 DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
10577 (void *)cache_resource,
10578 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10579 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10580 __ATOMIC_RELAXED) == 0) {
10581 claim_zero(mlx5_flow_os_destroy_flow_action
10582 (cache_resource->action));
10583 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
10584 &priv->sh->push_vlan_action_list, idx,
10585 cache_resource, next);
10586 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
10587 DRV_LOG(DEBUG, "push vlan action resource %p: removed",
10588 (void *)cache_resource);
10595 * Release the fate resource.
10598 * Pointer to Ethernet device.
10600 * Pointer to mlx5_flow_handle.
10603 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
10604 struct mlx5_flow_handle *handle)
10606 if (!handle->rix_fate)
10608 switch (handle->fate_action) {
10609 case MLX5_FLOW_FATE_QUEUE:
10610 mlx5_hrxq_release(dev, handle->rix_hrxq);
10612 case MLX5_FLOW_FATE_JUMP:
10613 flow_dv_jump_tbl_resource_release(dev, handle);
10615 case MLX5_FLOW_FATE_PORT_ID:
10616 flow_dv_port_id_action_resource_release(dev,
10617 handle->rix_port_id_action);
10620 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
10623 handle->rix_fate = 0;
10627 * Release an sample resource.
10630 * Pointer to Ethernet device.
10632 * Pointer to mlx5_flow_handle.
10635 * 1 while a reference on it exists, 0 when freed.
10638 flow_dv_sample_resource_release(struct rte_eth_dev *dev,
10639 struct mlx5_flow_handle *handle)
10641 struct mlx5_priv *priv = dev->data->dev_private;
10642 uint32_t idx = handle->dvh.rix_sample;
10643 struct mlx5_flow_dv_sample_resource *cache_resource;
10645 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
10647 if (!cache_resource)
10649 MLX5_ASSERT(cache_resource->verbs_action);
10650 DRV_LOG(DEBUG, "sample resource %p: refcnt %d--",
10651 (void *)cache_resource,
10652 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10653 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10654 __ATOMIC_RELAXED) == 0) {
10655 if (cache_resource->verbs_action)
10656 claim_zero(mlx5_glue->destroy_flow_action
10657 (cache_resource->verbs_action));
10658 if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
10659 if (cache_resource->default_miss)
10660 claim_zero(mlx5_glue->destroy_flow_action
10661 (cache_resource->default_miss));
10663 if (cache_resource->normal_path_tbl)
10664 flow_dv_tbl_resource_release(dev,
10665 cache_resource->normal_path_tbl);
10667 if (cache_resource->sample_idx.rix_hrxq &&
10668 !mlx5_hrxq_release(dev,
10669 cache_resource->sample_idx.rix_hrxq))
10670 cache_resource->sample_idx.rix_hrxq = 0;
10671 if (cache_resource->sample_idx.rix_tag &&
10672 !flow_dv_tag_release(dev,
10673 cache_resource->sample_idx.rix_tag))
10674 cache_resource->sample_idx.rix_tag = 0;
10675 if (cache_resource->sample_idx.cnt) {
10676 flow_dv_counter_release(dev,
10677 cache_resource->sample_idx.cnt);
10678 cache_resource->sample_idx.cnt = 0;
10680 if (!__atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED)) {
10681 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
10682 &priv->sh->sample_action_list, idx,
10683 cache_resource, next);
10684 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE], idx);
10685 DRV_LOG(DEBUG, "sample resource %p: removed",
10686 (void *)cache_resource);
10693 * Release an destination array resource.
10696 * Pointer to Ethernet device.
10698 * Pointer to mlx5_flow_handle.
10701 * 1 while a reference on it exists, 0 when freed.
10704 flow_dv_dest_array_resource_release(struct rte_eth_dev *dev,
10705 struct mlx5_flow_handle *handle)
10707 struct mlx5_priv *priv = dev->data->dev_private;
10708 struct mlx5_flow_dv_dest_array_resource *cache_resource;
10709 struct mlx5_flow_sub_actions_idx *mdest_act_res;
10710 uint32_t idx = handle->dvh.rix_dest_array;
10713 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
10715 if (!cache_resource)
10717 MLX5_ASSERT(cache_resource->action);
10718 DRV_LOG(DEBUG, "destination array resource %p: refcnt %d--",
10719 (void *)cache_resource,
10720 __atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10721 if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10722 __ATOMIC_RELAXED) == 0) {
10723 if (cache_resource->action)
10724 claim_zero(mlx5_glue->destroy_flow_action
10725 (cache_resource->action));
10726 for (; i < cache_resource->num_of_dest; i++) {
10727 mdest_act_res = &cache_resource->sample_idx[i];
10728 if (mdest_act_res->rix_hrxq) {
10729 mlx5_hrxq_release(dev,
10730 mdest_act_res->rix_hrxq);
10731 mdest_act_res->rix_hrxq = 0;
10733 if (mdest_act_res->rix_encap_decap) {
10734 flow_dv_encap_decap_resource_release(dev,
10735 mdest_act_res->rix_encap_decap);
10736 mdest_act_res->rix_encap_decap = 0;
10738 if (mdest_act_res->rix_port_id_action) {
10739 flow_dv_port_id_action_resource_release(dev,
10740 mdest_act_res->rix_port_id_action);
10741 mdest_act_res->rix_port_id_action = 0;
10743 if (mdest_act_res->rix_tag) {
10744 flow_dv_tag_release(dev,
10745 mdest_act_res->rix_tag);
10746 mdest_act_res->rix_tag = 0;
10749 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
10750 &priv->sh->dest_array_list, idx,
10751 cache_resource, next);
10752 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY], idx);
10753 DRV_LOG(DEBUG, "destination array resource %p: removed",
10754 (void *)cache_resource);
10761 * Remove the flow from the NIC but keeps it in memory.
10762 * Lock free, (mutex should be acquired by caller).
10765 * Pointer to Ethernet device.
10766 * @param[in, out] flow
10767 * Pointer to flow structure.
10770 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
10772 struct mlx5_flow_handle *dh;
10773 uint32_t handle_idx;
10774 struct mlx5_priv *priv = dev->data->dev_private;
10778 handle_idx = flow->dev_handles;
10779 while (handle_idx) {
10780 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
10784 if (dh->drv_flow) {
10785 claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
10786 dh->drv_flow = NULL;
10788 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE)
10789 flow_dv_fate_resource_release(dev, dh);
10790 if (dh->vf_vlan.tag && dh->vf_vlan.created)
10791 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
10792 handle_idx = dh->next.next;
10797 * Remove the flow from the NIC and the memory.
10798 * Lock free, (mutex should be acquired by caller).
10801 * Pointer to the Ethernet device structure.
10802 * @param[in, out] flow
10803 * Pointer to flow structure.
10806 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
10808 struct rte_flow_shared_action *shared;
10809 struct mlx5_flow_handle *dev_handle;
10810 struct mlx5_priv *priv = dev->data->dev_private;
10814 __flow_dv_remove(dev, flow);
10815 shared = mlx5_flow_get_shared_rss(flow);
10817 __atomic_sub_fetch(&shared->refcnt, 1, __ATOMIC_RELAXED);
10818 if (flow->counter) {
10819 flow_dv_counter_release(dev, flow->counter);
10823 struct mlx5_flow_meter *fm;
10825 fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
10828 mlx5_flow_meter_detach(fm);
10831 while (flow->dev_handles) {
10832 uint32_t tmp_idx = flow->dev_handles;
10834 dev_handle = mlx5_ipool_get(priv->sh->ipool
10835 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
10838 flow->dev_handles = dev_handle->next.next;
10839 if (dev_handle->dvh.matcher)
10840 flow_dv_matcher_release(dev, dev_handle);
10841 if (dev_handle->dvh.rix_sample)
10842 flow_dv_sample_resource_release(dev, dev_handle);
10843 if (dev_handle->dvh.rix_dest_array)
10844 flow_dv_dest_array_resource_release(dev, dev_handle);
10845 if (dev_handle->dvh.rix_encap_decap)
10846 flow_dv_encap_decap_resource_release(dev,
10847 dev_handle->dvh.rix_encap_decap);
10848 if (dev_handle->dvh.modify_hdr)
10849 flow_dv_modify_hdr_resource_release(dev, dev_handle);
10850 if (dev_handle->dvh.rix_push_vlan)
10851 flow_dv_push_vlan_action_resource_release(dev,
10853 if (dev_handle->dvh.rix_tag)
10854 flow_dv_tag_release(dev,
10855 dev_handle->dvh.rix_tag);
10856 flow_dv_fate_resource_release(dev, dev_handle);
10857 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
10863 * Release array of hash RX queue objects.
10867 * Pointer to the Ethernet device structure.
10868 * @param[in, out] hrxqs
10869 * Array of hash RX queue objects.
10872 * Total number of references to hash RX queue objects in *hrxqs* array
10873 * after this operation.
10876 __flow_dv_hrxqs_release(struct rte_eth_dev *dev,
10877 uint32_t (*hrxqs)[MLX5_RSS_HASH_FIELDS_LEN])
10882 for (i = 0; i < RTE_DIM(*hrxqs); i++) {
10883 int ret = mlx5_hrxq_release(dev, (*hrxqs)[i]);
10893 * Release all hash RX queue objects representing shared RSS action.
10896 * Pointer to the Ethernet device structure.
10897 * @param[in, out] action
10898 * Shared RSS action to remove hash RX queue objects from.
10901 * Total number of references to hash RX queue objects stored in *action*
10902 * after this operation.
10903 * Expected to be 0 if no external references held.
10906 __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
10907 struct mlx5_shared_action_rss *action)
10909 return __flow_dv_hrxqs_release(dev, &action->hrxq) +
10910 __flow_dv_hrxqs_release(dev, &action->hrxq_tunnel);
10914 * Setup shared RSS action.
10915 * Prepare set of hash RX queue objects sufficient to handle all valid
10916 * hash_fields combinations (see enum ibv_rx_hash_fields).
10919 * Pointer to the Ethernet device structure.
10920 * @param[in, out] action
10921 * Partially initialized shared RSS action.
10922 * @param[out] error
10923 * Perform verbose error reporting if not NULL. Initialized in case of
10927 * 0 on success, otherwise negative errno value.
10930 __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
10931 struct mlx5_shared_action_rss *action,
10932 struct rte_flow_error *error)
10934 struct mlx5_flow_rss_desc rss_desc = { 0 };
10938 memcpy(rss_desc.key, action->origin.key, MLX5_RSS_HASH_KEY_LEN);
10939 rss_desc.key_len = MLX5_RSS_HASH_KEY_LEN;
10940 rss_desc.const_q = action->origin.queue;
10941 rss_desc.queue_num = action->origin.queue_num;
10942 rss_desc.standalone = true;
10943 for (i = 0; i < MLX5_RSS_HASH_FIELDS_LEN; i++) {
10945 uint64_t hash_fields = mlx5_rss_hash_fields[i];
10948 for (tunnel = 0; tunnel < 2; tunnel++) {
10949 rss_desc.tunnel = tunnel;
10950 rss_desc.hash_fields = hash_fields;
10951 hrxq_idx = mlx5_hrxq_get(dev, &rss_desc);
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 */