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 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
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 mlx5_dev_ctx_shared *sh,
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,
85 flow_dv_shared_rss_action_release(struct rte_eth_dev *dev, uint32_t srss);
88 * Initialize flow attributes structure according to flow items' types.
90 * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
91 * mode. For tunnel mode, the items to be modified are the outermost ones.
94 * Pointer to item specification.
96 * Pointer to flow attributes structure.
98 * Pointer to the sub flow.
99 * @param[in] tunnel_decap
100 * Whether action is after tunnel decapsulation.
103 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
104 struct mlx5_flow *dev_flow, bool tunnel_decap)
106 uint64_t layers = dev_flow->handle->layers;
109 * If layers is already initialized, it means this dev_flow is the
110 * suffix flow, the layers flags is set by the prefix flow. Need to
111 * use the layer flags from prefix flow as the suffix flow may not
112 * have the user defined items as the flow is split.
115 if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
117 else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
119 if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
121 else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
126 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
127 uint8_t next_protocol = 0xff;
128 switch (item->type) {
129 case RTE_FLOW_ITEM_TYPE_GRE:
130 case RTE_FLOW_ITEM_TYPE_NVGRE:
131 case RTE_FLOW_ITEM_TYPE_VXLAN:
132 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
133 case RTE_FLOW_ITEM_TYPE_GENEVE:
134 case RTE_FLOW_ITEM_TYPE_MPLS:
138 case RTE_FLOW_ITEM_TYPE_IPV4:
141 if (item->mask != NULL &&
142 ((const struct rte_flow_item_ipv4 *)
143 item->mask)->hdr.next_proto_id)
145 ((const struct rte_flow_item_ipv4 *)
146 (item->spec))->hdr.next_proto_id &
147 ((const struct rte_flow_item_ipv4 *)
148 (item->mask))->hdr.next_proto_id;
149 if ((next_protocol == IPPROTO_IPIP ||
150 next_protocol == IPPROTO_IPV6) && tunnel_decap)
153 case RTE_FLOW_ITEM_TYPE_IPV6:
156 if (item->mask != NULL &&
157 ((const struct rte_flow_item_ipv6 *)
158 item->mask)->hdr.proto)
160 ((const struct rte_flow_item_ipv6 *)
161 (item->spec))->hdr.proto &
162 ((const struct rte_flow_item_ipv6 *)
163 (item->mask))->hdr.proto;
164 if ((next_protocol == IPPROTO_IPIP ||
165 next_protocol == IPPROTO_IPV6) && tunnel_decap)
168 case RTE_FLOW_ITEM_TYPE_UDP:
172 case RTE_FLOW_ITEM_TYPE_TCP:
184 * Convert rte_mtr_color to mlx5 color.
193 rte_col_2_mlx5_col(enum rte_color rcol)
196 case RTE_COLOR_GREEN:
197 return MLX5_FLOW_COLOR_GREEN;
198 case RTE_COLOR_YELLOW:
199 return MLX5_FLOW_COLOR_YELLOW;
201 return MLX5_FLOW_COLOR_RED;
205 return MLX5_FLOW_COLOR_UNDEFINED;
208 struct field_modify_info {
209 uint32_t size; /* Size of field in protocol header, in bytes. */
210 uint32_t offset; /* Offset of field in protocol header, in bytes. */
211 enum mlx5_modification_field id;
214 struct field_modify_info modify_eth[] = {
215 {4, 0, MLX5_MODI_OUT_DMAC_47_16},
216 {2, 4, MLX5_MODI_OUT_DMAC_15_0},
217 {4, 6, MLX5_MODI_OUT_SMAC_47_16},
218 {2, 10, MLX5_MODI_OUT_SMAC_15_0},
222 struct field_modify_info modify_vlan_out_first_vid[] = {
223 /* Size in bits !!! */
224 {12, 0, MLX5_MODI_OUT_FIRST_VID},
228 struct field_modify_info modify_ipv4[] = {
229 {1, 1, MLX5_MODI_OUT_IP_DSCP},
230 {1, 8, MLX5_MODI_OUT_IPV4_TTL},
231 {4, 12, MLX5_MODI_OUT_SIPV4},
232 {4, 16, MLX5_MODI_OUT_DIPV4},
236 struct field_modify_info modify_ipv6[] = {
237 {1, 0, MLX5_MODI_OUT_IP_DSCP},
238 {1, 7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
239 {4, 8, MLX5_MODI_OUT_SIPV6_127_96},
240 {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
241 {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
242 {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
243 {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
244 {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
245 {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
246 {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
250 struct field_modify_info modify_udp[] = {
251 {2, 0, MLX5_MODI_OUT_UDP_SPORT},
252 {2, 2, MLX5_MODI_OUT_UDP_DPORT},
256 struct field_modify_info modify_tcp[] = {
257 {2, 0, MLX5_MODI_OUT_TCP_SPORT},
258 {2, 2, MLX5_MODI_OUT_TCP_DPORT},
259 {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
260 {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
265 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
266 uint8_t next_protocol, uint64_t *item_flags,
269 MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
270 item->type == RTE_FLOW_ITEM_TYPE_IPV6);
271 if (next_protocol == IPPROTO_IPIP) {
272 *item_flags |= MLX5_FLOW_LAYER_IPIP;
275 if (next_protocol == IPPROTO_IPV6) {
276 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
281 /* Update VLAN's VID/PCP based on input rte_flow_action.
284 * Pointer to struct rte_flow_action.
286 * Pointer to struct rte_vlan_hdr.
289 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
290 struct rte_vlan_hdr *vlan)
293 if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
295 ((const struct rte_flow_action_of_set_vlan_pcp *)
296 action->conf)->vlan_pcp;
297 vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
298 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
299 vlan->vlan_tci |= vlan_tci;
300 } else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
301 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
302 vlan->vlan_tci |= rte_be_to_cpu_16
303 (((const struct rte_flow_action_of_set_vlan_vid *)
304 action->conf)->vlan_vid);
309 * Fetch 1, 2, 3 or 4 byte field from the byte array
310 * and return as unsigned integer in host-endian format.
313 * Pointer to data array.
315 * Size of field to extract.
318 * converted field in host endian format.
320 static inline uint32_t
321 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
330 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
333 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
334 ret = (ret << 8) | *(data + sizeof(uint16_t));
337 ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
348 * Convert modify-header action to DV specification.
350 * Data length of each action is determined by provided field description
351 * and the item mask. Data bit offset and width of each action is determined
352 * by provided item mask.
355 * Pointer to item specification.
357 * Pointer to field modification information.
358 * For MLX5_MODIFICATION_TYPE_SET specifies destination field.
359 * For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
360 * For MLX5_MODIFICATION_TYPE_COPY specifies source field.
362 * Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
363 * Negative offset value sets the same offset as source offset.
364 * size field is ignored, value is taken from source field.
365 * @param[in,out] resource
366 * Pointer to the modify-header resource.
368 * Type of modification.
370 * Pointer to the error structure.
373 * 0 on success, a negative errno value otherwise and rte_errno is set.
376 flow_dv_convert_modify_action(struct rte_flow_item *item,
377 struct field_modify_info *field,
378 struct field_modify_info *dcopy,
379 struct mlx5_flow_dv_modify_hdr_resource *resource,
380 uint32_t type, struct rte_flow_error *error)
382 uint32_t i = resource->actions_num;
383 struct mlx5_modification_cmd *actions = resource->actions;
386 * The item and mask are provided in big-endian format.
387 * The fields should be presented as in big-endian format either.
388 * Mask must be always present, it defines the actual field width.
390 MLX5_ASSERT(item->mask);
391 MLX5_ASSERT(field->size);
398 if (i >= MLX5_MAX_MODIFY_NUM)
399 return rte_flow_error_set(error, EINVAL,
400 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
401 "too many items to modify");
402 /* Fetch variable byte size mask from the array. */
403 mask = flow_dv_fetch_field((const uint8_t *)item->mask +
404 field->offset, field->size);
409 /* Deduce actual data width in bits from mask value. */
410 off_b = rte_bsf32(mask);
411 size_b = sizeof(uint32_t) * CHAR_BIT -
412 off_b - __builtin_clz(mask);
414 size_b = size_b == sizeof(uint32_t) * CHAR_BIT ? 0 : size_b;
415 actions[i] = (struct mlx5_modification_cmd) {
421 /* Convert entire record to expected big-endian format. */
422 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
423 if (type == MLX5_MODIFICATION_TYPE_COPY) {
425 actions[i].dst_field = dcopy->id;
426 actions[i].dst_offset =
427 (int)dcopy->offset < 0 ? off_b : dcopy->offset;
428 /* Convert entire record to big-endian format. */
429 actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
431 MLX5_ASSERT(item->spec);
432 data = flow_dv_fetch_field((const uint8_t *)item->spec +
433 field->offset, field->size);
434 /* Shift out the trailing masked bits from data. */
435 data = (data & mask) >> off_b;
436 actions[i].data1 = rte_cpu_to_be_32(data);
440 } while (field->size);
441 if (resource->actions_num == i)
442 return rte_flow_error_set(error, EINVAL,
443 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
444 "invalid modification flow item");
445 resource->actions_num = i;
450 * Convert modify-header set IPv4 address action to DV specification.
452 * @param[in,out] resource
453 * Pointer to the modify-header resource.
455 * Pointer to action specification.
457 * Pointer to the error structure.
460 * 0 on success, a negative errno value otherwise and rte_errno is set.
463 flow_dv_convert_action_modify_ipv4
464 (struct mlx5_flow_dv_modify_hdr_resource *resource,
465 const struct rte_flow_action *action,
466 struct rte_flow_error *error)
468 const struct rte_flow_action_set_ipv4 *conf =
469 (const struct rte_flow_action_set_ipv4 *)(action->conf);
470 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
471 struct rte_flow_item_ipv4 ipv4;
472 struct rte_flow_item_ipv4 ipv4_mask;
474 memset(&ipv4, 0, sizeof(ipv4));
475 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
476 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
477 ipv4.hdr.src_addr = conf->ipv4_addr;
478 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
480 ipv4.hdr.dst_addr = conf->ipv4_addr;
481 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
484 item.mask = &ipv4_mask;
485 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
486 MLX5_MODIFICATION_TYPE_SET, error);
490 * Convert modify-header set IPv6 address action to DV specification.
492 * @param[in,out] resource
493 * Pointer to the modify-header resource.
495 * Pointer to action specification.
497 * Pointer to the error structure.
500 * 0 on success, a negative errno value otherwise and rte_errno is set.
503 flow_dv_convert_action_modify_ipv6
504 (struct mlx5_flow_dv_modify_hdr_resource *resource,
505 const struct rte_flow_action *action,
506 struct rte_flow_error *error)
508 const struct rte_flow_action_set_ipv6 *conf =
509 (const struct rte_flow_action_set_ipv6 *)(action->conf);
510 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
511 struct rte_flow_item_ipv6 ipv6;
512 struct rte_flow_item_ipv6 ipv6_mask;
514 memset(&ipv6, 0, sizeof(ipv6));
515 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
516 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
517 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
518 sizeof(ipv6.hdr.src_addr));
519 memcpy(&ipv6_mask.hdr.src_addr,
520 &rte_flow_item_ipv6_mask.hdr.src_addr,
521 sizeof(ipv6.hdr.src_addr));
523 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
524 sizeof(ipv6.hdr.dst_addr));
525 memcpy(&ipv6_mask.hdr.dst_addr,
526 &rte_flow_item_ipv6_mask.hdr.dst_addr,
527 sizeof(ipv6.hdr.dst_addr));
530 item.mask = &ipv6_mask;
531 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
532 MLX5_MODIFICATION_TYPE_SET, error);
536 * Convert modify-header set MAC address action to DV specification.
538 * @param[in,out] resource
539 * Pointer to the modify-header resource.
541 * Pointer to action specification.
543 * Pointer to the error structure.
546 * 0 on success, a negative errno value otherwise and rte_errno is set.
549 flow_dv_convert_action_modify_mac
550 (struct mlx5_flow_dv_modify_hdr_resource *resource,
551 const struct rte_flow_action *action,
552 struct rte_flow_error *error)
554 const struct rte_flow_action_set_mac *conf =
555 (const struct rte_flow_action_set_mac *)(action->conf);
556 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
557 struct rte_flow_item_eth eth;
558 struct rte_flow_item_eth eth_mask;
560 memset(ð, 0, sizeof(eth));
561 memset(ð_mask, 0, sizeof(eth_mask));
562 if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
563 memcpy(ð.src.addr_bytes, &conf->mac_addr,
564 sizeof(eth.src.addr_bytes));
565 memcpy(ð_mask.src.addr_bytes,
566 &rte_flow_item_eth_mask.src.addr_bytes,
567 sizeof(eth_mask.src.addr_bytes));
569 memcpy(ð.dst.addr_bytes, &conf->mac_addr,
570 sizeof(eth.dst.addr_bytes));
571 memcpy(ð_mask.dst.addr_bytes,
572 &rte_flow_item_eth_mask.dst.addr_bytes,
573 sizeof(eth_mask.dst.addr_bytes));
576 item.mask = ð_mask;
577 return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
578 MLX5_MODIFICATION_TYPE_SET, error);
582 * Convert modify-header set VLAN VID action to DV specification.
584 * @param[in,out] resource
585 * Pointer to the modify-header resource.
587 * Pointer to action specification.
589 * Pointer to the error structure.
592 * 0 on success, a negative errno value otherwise and rte_errno is set.
595 flow_dv_convert_action_modify_vlan_vid
596 (struct mlx5_flow_dv_modify_hdr_resource *resource,
597 const struct rte_flow_action *action,
598 struct rte_flow_error *error)
600 const struct rte_flow_action_of_set_vlan_vid *conf =
601 (const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
602 int i = resource->actions_num;
603 struct mlx5_modification_cmd *actions = resource->actions;
604 struct field_modify_info *field = modify_vlan_out_first_vid;
606 if (i >= MLX5_MAX_MODIFY_NUM)
607 return rte_flow_error_set(error, EINVAL,
608 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
609 "too many items to modify");
610 actions[i] = (struct mlx5_modification_cmd) {
611 .action_type = MLX5_MODIFICATION_TYPE_SET,
613 .length = field->size,
614 .offset = field->offset,
616 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
617 actions[i].data1 = conf->vlan_vid;
618 actions[i].data1 = actions[i].data1 << 16;
619 resource->actions_num = ++i;
624 * Convert modify-header set TP action to DV specification.
626 * @param[in,out] resource
627 * Pointer to the modify-header resource.
629 * Pointer to action specification.
631 * Pointer to rte_flow_item objects list.
633 * Pointer to flow attributes structure.
634 * @param[in] dev_flow
635 * Pointer to the sub flow.
636 * @param[in] tunnel_decap
637 * Whether action is after tunnel decapsulation.
639 * Pointer to the error structure.
642 * 0 on success, a negative errno value otherwise and rte_errno is set.
645 flow_dv_convert_action_modify_tp
646 (struct mlx5_flow_dv_modify_hdr_resource *resource,
647 const struct rte_flow_action *action,
648 const struct rte_flow_item *items,
649 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
650 bool tunnel_decap, struct rte_flow_error *error)
652 const struct rte_flow_action_set_tp *conf =
653 (const struct rte_flow_action_set_tp *)(action->conf);
654 struct rte_flow_item item;
655 struct rte_flow_item_udp udp;
656 struct rte_flow_item_udp udp_mask;
657 struct rte_flow_item_tcp tcp;
658 struct rte_flow_item_tcp tcp_mask;
659 struct field_modify_info *field;
662 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
664 memset(&udp, 0, sizeof(udp));
665 memset(&udp_mask, 0, sizeof(udp_mask));
666 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
667 udp.hdr.src_port = conf->port;
668 udp_mask.hdr.src_port =
669 rte_flow_item_udp_mask.hdr.src_port;
671 udp.hdr.dst_port = conf->port;
672 udp_mask.hdr.dst_port =
673 rte_flow_item_udp_mask.hdr.dst_port;
675 item.type = RTE_FLOW_ITEM_TYPE_UDP;
677 item.mask = &udp_mask;
680 MLX5_ASSERT(attr->tcp);
681 memset(&tcp, 0, sizeof(tcp));
682 memset(&tcp_mask, 0, sizeof(tcp_mask));
683 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
684 tcp.hdr.src_port = conf->port;
685 tcp_mask.hdr.src_port =
686 rte_flow_item_tcp_mask.hdr.src_port;
688 tcp.hdr.dst_port = conf->port;
689 tcp_mask.hdr.dst_port =
690 rte_flow_item_tcp_mask.hdr.dst_port;
692 item.type = RTE_FLOW_ITEM_TYPE_TCP;
694 item.mask = &tcp_mask;
697 return flow_dv_convert_modify_action(&item, field, NULL, resource,
698 MLX5_MODIFICATION_TYPE_SET, error);
702 * Convert modify-header set TTL action to DV specification.
704 * @param[in,out] resource
705 * Pointer to the modify-header resource.
707 * Pointer to action specification.
709 * Pointer to rte_flow_item objects list.
711 * Pointer to flow attributes structure.
712 * @param[in] dev_flow
713 * Pointer to the sub flow.
714 * @param[in] tunnel_decap
715 * Whether action is after tunnel decapsulation.
717 * Pointer to the error structure.
720 * 0 on success, a negative errno value otherwise and rte_errno is set.
723 flow_dv_convert_action_modify_ttl
724 (struct mlx5_flow_dv_modify_hdr_resource *resource,
725 const struct rte_flow_action *action,
726 const struct rte_flow_item *items,
727 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
728 bool tunnel_decap, struct rte_flow_error *error)
730 const struct rte_flow_action_set_ttl *conf =
731 (const struct rte_flow_action_set_ttl *)(action->conf);
732 struct rte_flow_item item;
733 struct rte_flow_item_ipv4 ipv4;
734 struct rte_flow_item_ipv4 ipv4_mask;
735 struct rte_flow_item_ipv6 ipv6;
736 struct rte_flow_item_ipv6 ipv6_mask;
737 struct field_modify_info *field;
740 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
742 memset(&ipv4, 0, sizeof(ipv4));
743 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
744 ipv4.hdr.time_to_live = conf->ttl_value;
745 ipv4_mask.hdr.time_to_live = 0xFF;
746 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
748 item.mask = &ipv4_mask;
751 MLX5_ASSERT(attr->ipv6);
752 memset(&ipv6, 0, sizeof(ipv6));
753 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
754 ipv6.hdr.hop_limits = conf->ttl_value;
755 ipv6_mask.hdr.hop_limits = 0xFF;
756 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
758 item.mask = &ipv6_mask;
761 return flow_dv_convert_modify_action(&item, field, NULL, resource,
762 MLX5_MODIFICATION_TYPE_SET, error);
766 * Convert modify-header decrement TTL action to DV specification.
768 * @param[in,out] resource
769 * Pointer to the modify-header resource.
771 * Pointer to action specification.
773 * Pointer to rte_flow_item objects list.
775 * Pointer to flow attributes structure.
776 * @param[in] dev_flow
777 * Pointer to the sub flow.
778 * @param[in] tunnel_decap
779 * Whether action is after tunnel decapsulation.
781 * Pointer to the error structure.
784 * 0 on success, a negative errno value otherwise and rte_errno is set.
787 flow_dv_convert_action_modify_dec_ttl
788 (struct mlx5_flow_dv_modify_hdr_resource *resource,
789 const struct rte_flow_item *items,
790 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
791 bool tunnel_decap, struct rte_flow_error *error)
793 struct rte_flow_item item;
794 struct rte_flow_item_ipv4 ipv4;
795 struct rte_flow_item_ipv4 ipv4_mask;
796 struct rte_flow_item_ipv6 ipv6;
797 struct rte_flow_item_ipv6 ipv6_mask;
798 struct field_modify_info *field;
801 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
803 memset(&ipv4, 0, sizeof(ipv4));
804 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
805 ipv4.hdr.time_to_live = 0xFF;
806 ipv4_mask.hdr.time_to_live = 0xFF;
807 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
809 item.mask = &ipv4_mask;
812 MLX5_ASSERT(attr->ipv6);
813 memset(&ipv6, 0, sizeof(ipv6));
814 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
815 ipv6.hdr.hop_limits = 0xFF;
816 ipv6_mask.hdr.hop_limits = 0xFF;
817 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
819 item.mask = &ipv6_mask;
822 return flow_dv_convert_modify_action(&item, field, NULL, resource,
823 MLX5_MODIFICATION_TYPE_ADD, error);
827 * Convert modify-header increment/decrement TCP Sequence number
828 * to DV specification.
830 * @param[in,out] resource
831 * Pointer to the modify-header resource.
833 * Pointer to action specification.
835 * Pointer to the error structure.
838 * 0 on success, a negative errno value otherwise and rte_errno is set.
841 flow_dv_convert_action_modify_tcp_seq
842 (struct mlx5_flow_dv_modify_hdr_resource *resource,
843 const struct rte_flow_action *action,
844 struct rte_flow_error *error)
846 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
847 uint64_t value = rte_be_to_cpu_32(*conf);
848 struct rte_flow_item item;
849 struct rte_flow_item_tcp tcp;
850 struct rte_flow_item_tcp tcp_mask;
852 memset(&tcp, 0, sizeof(tcp));
853 memset(&tcp_mask, 0, sizeof(tcp_mask));
854 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
856 * The HW has no decrement operation, only increment operation.
857 * To simulate decrement X from Y using increment operation
858 * we need to add UINT32_MAX X times to Y.
859 * Each adding of UINT32_MAX decrements Y by 1.
862 tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
863 tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
864 item.type = RTE_FLOW_ITEM_TYPE_TCP;
866 item.mask = &tcp_mask;
867 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
868 MLX5_MODIFICATION_TYPE_ADD, error);
872 * Convert modify-header increment/decrement TCP Acknowledgment number
873 * to DV specification.
875 * @param[in,out] resource
876 * Pointer to the modify-header resource.
878 * Pointer to action specification.
880 * Pointer to the error structure.
883 * 0 on success, a negative errno value otherwise and rte_errno is set.
886 flow_dv_convert_action_modify_tcp_ack
887 (struct mlx5_flow_dv_modify_hdr_resource *resource,
888 const struct rte_flow_action *action,
889 struct rte_flow_error *error)
891 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
892 uint64_t value = rte_be_to_cpu_32(*conf);
893 struct rte_flow_item item;
894 struct rte_flow_item_tcp tcp;
895 struct rte_flow_item_tcp tcp_mask;
897 memset(&tcp, 0, sizeof(tcp));
898 memset(&tcp_mask, 0, sizeof(tcp_mask));
899 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
901 * The HW has no decrement operation, only increment operation.
902 * To simulate decrement X from Y using increment operation
903 * we need to add UINT32_MAX X times to Y.
904 * Each adding of UINT32_MAX decrements Y by 1.
907 tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
908 tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
909 item.type = RTE_FLOW_ITEM_TYPE_TCP;
911 item.mask = &tcp_mask;
912 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
913 MLX5_MODIFICATION_TYPE_ADD, error);
916 static enum mlx5_modification_field reg_to_field[] = {
917 [REG_NON] = MLX5_MODI_OUT_NONE,
918 [REG_A] = MLX5_MODI_META_DATA_REG_A,
919 [REG_B] = MLX5_MODI_META_DATA_REG_B,
920 [REG_C_0] = MLX5_MODI_META_REG_C_0,
921 [REG_C_1] = MLX5_MODI_META_REG_C_1,
922 [REG_C_2] = MLX5_MODI_META_REG_C_2,
923 [REG_C_3] = MLX5_MODI_META_REG_C_3,
924 [REG_C_4] = MLX5_MODI_META_REG_C_4,
925 [REG_C_5] = MLX5_MODI_META_REG_C_5,
926 [REG_C_6] = MLX5_MODI_META_REG_C_6,
927 [REG_C_7] = MLX5_MODI_META_REG_C_7,
931 * Convert register set to DV specification.
933 * @param[in,out] resource
934 * Pointer to the modify-header resource.
936 * Pointer to action specification.
938 * Pointer to the error structure.
941 * 0 on success, a negative errno value otherwise and rte_errno is set.
944 flow_dv_convert_action_set_reg
945 (struct mlx5_flow_dv_modify_hdr_resource *resource,
946 const struct rte_flow_action *action,
947 struct rte_flow_error *error)
949 const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
950 struct mlx5_modification_cmd *actions = resource->actions;
951 uint32_t i = resource->actions_num;
953 if (i >= MLX5_MAX_MODIFY_NUM)
954 return rte_flow_error_set(error, EINVAL,
955 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
956 "too many items to modify");
957 MLX5_ASSERT(conf->id != REG_NON);
958 MLX5_ASSERT(conf->id < (enum modify_reg)RTE_DIM(reg_to_field));
959 actions[i] = (struct mlx5_modification_cmd) {
960 .action_type = MLX5_MODIFICATION_TYPE_SET,
961 .field = reg_to_field[conf->id],
963 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
964 actions[i].data1 = rte_cpu_to_be_32(conf->data);
966 resource->actions_num = i;
971 * Convert SET_TAG action to DV specification.
974 * Pointer to the rte_eth_dev structure.
975 * @param[in,out] resource
976 * Pointer to the modify-header resource.
978 * Pointer to action specification.
980 * Pointer to the error structure.
983 * 0 on success, a negative errno value otherwise and rte_errno is set.
986 flow_dv_convert_action_set_tag
987 (struct rte_eth_dev *dev,
988 struct mlx5_flow_dv_modify_hdr_resource *resource,
989 const struct rte_flow_action_set_tag *conf,
990 struct rte_flow_error *error)
992 rte_be32_t data = rte_cpu_to_be_32(conf->data);
993 rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
994 struct rte_flow_item item = {
998 struct field_modify_info reg_c_x[] = {
1001 enum mlx5_modification_field reg_type;
1004 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1007 MLX5_ASSERT(ret != REG_NON);
1008 MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1009 reg_type = reg_to_field[ret];
1010 MLX5_ASSERT(reg_type > 0);
1011 reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1012 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1013 MLX5_MODIFICATION_TYPE_SET, error);
1017 * Convert internal COPY_REG action to DV specification.
1020 * Pointer to the rte_eth_dev structure.
1021 * @param[in,out] res
1022 * Pointer to the modify-header resource.
1024 * Pointer to action specification.
1026 * Pointer to the error structure.
1029 * 0 on success, a negative errno value otherwise and rte_errno is set.
1032 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1033 struct mlx5_flow_dv_modify_hdr_resource *res,
1034 const struct rte_flow_action *action,
1035 struct rte_flow_error *error)
1037 const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1038 rte_be32_t mask = RTE_BE32(UINT32_MAX);
1039 struct rte_flow_item item = {
1043 struct field_modify_info reg_src[] = {
1044 {4, 0, reg_to_field[conf->src]},
1047 struct field_modify_info reg_dst = {
1049 .id = reg_to_field[conf->dst],
1051 /* Adjust reg_c[0] usage according to reported mask. */
1052 if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1053 struct mlx5_priv *priv = dev->data->dev_private;
1054 uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1056 MLX5_ASSERT(reg_c0);
1057 MLX5_ASSERT(priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY);
1058 if (conf->dst == REG_C_0) {
1059 /* Copy to reg_c[0], within mask only. */
1060 reg_dst.offset = rte_bsf32(reg_c0);
1062 * Mask is ignoring the enianness, because
1063 * there is no conversion in datapath.
1065 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1066 /* Copy from destination lower bits to reg_c[0]. */
1067 mask = reg_c0 >> reg_dst.offset;
1069 /* Copy from destination upper bits to reg_c[0]. */
1070 mask = reg_c0 << (sizeof(reg_c0) * CHAR_BIT -
1071 rte_fls_u32(reg_c0));
1074 mask = rte_cpu_to_be_32(reg_c0);
1075 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1076 /* Copy from reg_c[0] to destination lower bits. */
1079 /* Copy from reg_c[0] to destination upper bits. */
1080 reg_dst.offset = sizeof(reg_c0) * CHAR_BIT -
1081 (rte_fls_u32(reg_c0) -
1086 return flow_dv_convert_modify_action(&item,
1087 reg_src, ®_dst, res,
1088 MLX5_MODIFICATION_TYPE_COPY,
1093 * Convert MARK action to DV specification. This routine is used
1094 * in extensive metadata only and requires metadata register to be
1095 * handled. In legacy mode hardware tag resource is engaged.
1098 * Pointer to the rte_eth_dev structure.
1100 * Pointer to MARK action specification.
1101 * @param[in,out] resource
1102 * Pointer to the modify-header resource.
1104 * Pointer to the error structure.
1107 * 0 on success, a negative errno value otherwise and rte_errno is set.
1110 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1111 const struct rte_flow_action_mark *conf,
1112 struct mlx5_flow_dv_modify_hdr_resource *resource,
1113 struct rte_flow_error *error)
1115 struct mlx5_priv *priv = dev->data->dev_private;
1116 rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1117 priv->sh->dv_mark_mask);
1118 rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1119 struct rte_flow_item item = {
1123 struct field_modify_info reg_c_x[] = {
1129 return rte_flow_error_set(error, EINVAL,
1130 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1131 NULL, "zero mark action mask");
1132 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1135 MLX5_ASSERT(reg > 0);
1136 if (reg == REG_C_0) {
1137 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1138 uint32_t shl_c0 = rte_bsf32(msk_c0);
1140 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1141 mask = rte_cpu_to_be_32(mask) & msk_c0;
1142 mask = rte_cpu_to_be_32(mask << shl_c0);
1144 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1145 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1146 MLX5_MODIFICATION_TYPE_SET, error);
1150 * Get metadata register index for specified steering domain.
1153 * Pointer to the rte_eth_dev structure.
1155 * Attributes of flow to determine steering domain.
1157 * Pointer to the error structure.
1160 * positive index on success, a negative errno value otherwise
1161 * and rte_errno is set.
1163 static enum modify_reg
1164 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1165 const struct rte_flow_attr *attr,
1166 struct rte_flow_error *error)
1169 mlx5_flow_get_reg_id(dev, attr->transfer ?
1173 MLX5_METADATA_RX, 0, error);
1175 return rte_flow_error_set(error,
1176 ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1177 NULL, "unavailable "
1178 "metadata register");
1183 * Convert SET_META action to DV specification.
1186 * Pointer to the rte_eth_dev structure.
1187 * @param[in,out] resource
1188 * Pointer to the modify-header resource.
1190 * Attributes of flow that includes this item.
1192 * Pointer to action specification.
1194 * Pointer to the error structure.
1197 * 0 on success, a negative errno value otherwise and rte_errno is set.
1200 flow_dv_convert_action_set_meta
1201 (struct rte_eth_dev *dev,
1202 struct mlx5_flow_dv_modify_hdr_resource *resource,
1203 const struct rte_flow_attr *attr,
1204 const struct rte_flow_action_set_meta *conf,
1205 struct rte_flow_error *error)
1207 uint32_t data = conf->data;
1208 uint32_t mask = conf->mask;
1209 struct rte_flow_item item = {
1213 struct field_modify_info reg_c_x[] = {
1216 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1220 MLX5_ASSERT(reg != REG_NON);
1222 * In datapath code there is no endianness
1223 * coversions for perfromance reasons, all
1224 * pattern conversions are done in rte_flow.
1226 if (reg == REG_C_0) {
1227 struct mlx5_priv *priv = dev->data->dev_private;
1228 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1231 MLX5_ASSERT(msk_c0);
1232 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1233 shl_c0 = rte_bsf32(msk_c0);
1235 shl_c0 = sizeof(msk_c0) * CHAR_BIT - rte_fls_u32(msk_c0);
1239 MLX5_ASSERT(!(~msk_c0 & rte_cpu_to_be_32(mask)));
1241 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1242 /* The routine expects parameters in memory as big-endian ones. */
1243 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1244 MLX5_MODIFICATION_TYPE_SET, error);
1248 * Convert modify-header set IPv4 DSCP action to DV specification.
1250 * @param[in,out] resource
1251 * Pointer to the modify-header resource.
1253 * Pointer to action specification.
1255 * Pointer to the error structure.
1258 * 0 on success, a negative errno value otherwise and rte_errno is set.
1261 flow_dv_convert_action_modify_ipv4_dscp
1262 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1263 const struct rte_flow_action *action,
1264 struct rte_flow_error *error)
1266 const struct rte_flow_action_set_dscp *conf =
1267 (const struct rte_flow_action_set_dscp *)(action->conf);
1268 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1269 struct rte_flow_item_ipv4 ipv4;
1270 struct rte_flow_item_ipv4 ipv4_mask;
1272 memset(&ipv4, 0, sizeof(ipv4));
1273 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1274 ipv4.hdr.type_of_service = conf->dscp;
1275 ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1277 item.mask = &ipv4_mask;
1278 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1279 MLX5_MODIFICATION_TYPE_SET, error);
1283 * Convert modify-header set IPv6 DSCP action to DV specification.
1285 * @param[in,out] resource
1286 * Pointer to the modify-header resource.
1288 * Pointer to action specification.
1290 * Pointer to the error structure.
1293 * 0 on success, a negative errno value otherwise and rte_errno is set.
1296 flow_dv_convert_action_modify_ipv6_dscp
1297 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1298 const struct rte_flow_action *action,
1299 struct rte_flow_error *error)
1301 const struct rte_flow_action_set_dscp *conf =
1302 (const struct rte_flow_action_set_dscp *)(action->conf);
1303 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1304 struct rte_flow_item_ipv6 ipv6;
1305 struct rte_flow_item_ipv6 ipv6_mask;
1307 memset(&ipv6, 0, sizeof(ipv6));
1308 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1310 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1311 * rdma-core only accept the DSCP bits byte aligned start from
1312 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1313 * bits in IPv6 case as rdma-core requires byte aligned value.
1315 ipv6.hdr.vtc_flow = conf->dscp;
1316 ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1318 item.mask = &ipv6_mask;
1319 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1320 MLX5_MODIFICATION_TYPE_SET, error);
1324 * Validate MARK item.
1327 * Pointer to the rte_eth_dev structure.
1329 * Item specification.
1331 * Attributes of flow that includes this item.
1333 * Pointer to error structure.
1336 * 0 on success, a negative errno value otherwise and rte_errno is set.
1339 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1340 const struct rte_flow_item *item,
1341 const struct rte_flow_attr *attr __rte_unused,
1342 struct rte_flow_error *error)
1344 struct mlx5_priv *priv = dev->data->dev_private;
1345 struct mlx5_dev_config *config = &priv->config;
1346 const struct rte_flow_item_mark *spec = item->spec;
1347 const struct rte_flow_item_mark *mask = item->mask;
1348 const struct rte_flow_item_mark nic_mask = {
1349 .id = priv->sh->dv_mark_mask,
1353 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1354 return rte_flow_error_set(error, ENOTSUP,
1355 RTE_FLOW_ERROR_TYPE_ITEM, item,
1356 "extended metadata feature"
1358 if (!mlx5_flow_ext_mreg_supported(dev))
1359 return rte_flow_error_set(error, ENOTSUP,
1360 RTE_FLOW_ERROR_TYPE_ITEM, item,
1361 "extended metadata register"
1362 " isn't supported");
1364 return rte_flow_error_set(error, ENOTSUP,
1365 RTE_FLOW_ERROR_TYPE_ITEM, item,
1366 "extended metadata register"
1367 " isn't available");
1368 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1372 return rte_flow_error_set(error, EINVAL,
1373 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1375 "data cannot be empty");
1376 if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1377 return rte_flow_error_set(error, EINVAL,
1378 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1380 "mark id exceeds the limit");
1384 return rte_flow_error_set(error, EINVAL,
1385 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1386 "mask cannot be zero");
1388 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1389 (const uint8_t *)&nic_mask,
1390 sizeof(struct rte_flow_item_mark),
1391 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1398 * Validate META item.
1401 * Pointer to the rte_eth_dev structure.
1403 * Item specification.
1405 * Attributes of flow that includes this item.
1407 * Pointer to error structure.
1410 * 0 on success, a negative errno value otherwise and rte_errno is set.
1413 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1414 const struct rte_flow_item *item,
1415 const struct rte_flow_attr *attr,
1416 struct rte_flow_error *error)
1418 struct mlx5_priv *priv = dev->data->dev_private;
1419 struct mlx5_dev_config *config = &priv->config;
1420 const struct rte_flow_item_meta *spec = item->spec;
1421 const struct rte_flow_item_meta *mask = item->mask;
1422 struct rte_flow_item_meta nic_mask = {
1429 return rte_flow_error_set(error, EINVAL,
1430 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1432 "data cannot be empty");
1433 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1434 if (!mlx5_flow_ext_mreg_supported(dev))
1435 return rte_flow_error_set(error, ENOTSUP,
1436 RTE_FLOW_ERROR_TYPE_ITEM, item,
1437 "extended metadata register"
1438 " isn't supported");
1439 reg = flow_dv_get_metadata_reg(dev, attr, error);
1443 return rte_flow_error_set(error, ENOTSUP,
1444 RTE_FLOW_ERROR_TYPE_ITEM, item,
1445 "unavalable extended metadata register");
1447 return rte_flow_error_set(error, ENOTSUP,
1448 RTE_FLOW_ERROR_TYPE_ITEM, item,
1452 nic_mask.data = priv->sh->dv_meta_mask;
1453 } else if (attr->transfer) {
1454 return rte_flow_error_set(error, ENOTSUP,
1455 RTE_FLOW_ERROR_TYPE_ITEM, item,
1456 "extended metadata feature "
1457 "should be enabled when "
1458 "meta item is requested "
1459 "with e-switch mode ");
1462 mask = &rte_flow_item_meta_mask;
1464 return rte_flow_error_set(error, EINVAL,
1465 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1466 "mask cannot be zero");
1468 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1469 (const uint8_t *)&nic_mask,
1470 sizeof(struct rte_flow_item_meta),
1471 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1476 * Validate TAG item.
1479 * Pointer to the rte_eth_dev structure.
1481 * Item specification.
1483 * Attributes of flow that includes this item.
1485 * Pointer to error structure.
1488 * 0 on success, a negative errno value otherwise and rte_errno is set.
1491 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
1492 const struct rte_flow_item *item,
1493 const struct rte_flow_attr *attr __rte_unused,
1494 struct rte_flow_error *error)
1496 const struct rte_flow_item_tag *spec = item->spec;
1497 const struct rte_flow_item_tag *mask = item->mask;
1498 const struct rte_flow_item_tag nic_mask = {
1499 .data = RTE_BE32(UINT32_MAX),
1504 if (!mlx5_flow_ext_mreg_supported(dev))
1505 return rte_flow_error_set(error, ENOTSUP,
1506 RTE_FLOW_ERROR_TYPE_ITEM, item,
1507 "extensive metadata register"
1508 " isn't supported");
1510 return rte_flow_error_set(error, EINVAL,
1511 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1513 "data cannot be empty");
1515 mask = &rte_flow_item_tag_mask;
1517 return rte_flow_error_set(error, EINVAL,
1518 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1519 "mask cannot be zero");
1521 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1522 (const uint8_t *)&nic_mask,
1523 sizeof(struct rte_flow_item_tag),
1524 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1527 if (mask->index != 0xff)
1528 return rte_flow_error_set(error, EINVAL,
1529 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1530 "partial mask for tag index"
1531 " is not supported");
1532 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
1535 MLX5_ASSERT(ret != REG_NON);
1540 * Validate vport item.
1543 * Pointer to the rte_eth_dev structure.
1545 * Item specification.
1547 * Attributes of flow that includes this item.
1548 * @param[in] item_flags
1549 * Bit-fields that holds the items detected until now.
1551 * Pointer to error structure.
1554 * 0 on success, a negative errno value otherwise and rte_errno is set.
1557 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
1558 const struct rte_flow_item *item,
1559 const struct rte_flow_attr *attr,
1560 uint64_t item_flags,
1561 struct rte_flow_error *error)
1563 const struct rte_flow_item_port_id *spec = item->spec;
1564 const struct rte_flow_item_port_id *mask = item->mask;
1565 const struct rte_flow_item_port_id switch_mask = {
1568 struct mlx5_priv *esw_priv;
1569 struct mlx5_priv *dev_priv;
1572 if (!attr->transfer)
1573 return rte_flow_error_set(error, EINVAL,
1574 RTE_FLOW_ERROR_TYPE_ITEM,
1576 "match on port id is valid only"
1577 " when transfer flag is enabled");
1578 if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
1579 return rte_flow_error_set(error, ENOTSUP,
1580 RTE_FLOW_ERROR_TYPE_ITEM, item,
1581 "multiple source ports are not"
1584 mask = &switch_mask;
1585 if (mask->id != 0xffffffff)
1586 return rte_flow_error_set(error, ENOTSUP,
1587 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1589 "no support for partial mask on"
1591 ret = mlx5_flow_item_acceptable
1592 (item, (const uint8_t *)mask,
1593 (const uint8_t *)&rte_flow_item_port_id_mask,
1594 sizeof(struct rte_flow_item_port_id),
1595 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1600 esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
1602 return rte_flow_error_set(error, rte_errno,
1603 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1604 "failed to obtain E-Switch info for"
1606 dev_priv = mlx5_dev_to_eswitch_info(dev);
1608 return rte_flow_error_set(error, rte_errno,
1609 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1611 "failed to obtain E-Switch info");
1612 if (esw_priv->domain_id != dev_priv->domain_id)
1613 return rte_flow_error_set(error, EINVAL,
1614 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1615 "cannot match on a port from a"
1616 " different E-Switch");
1621 * Validate VLAN item.
1624 * Item specification.
1625 * @param[in] item_flags
1626 * Bit-fields that holds the items detected until now.
1628 * Ethernet device flow is being created on.
1630 * Pointer to error structure.
1633 * 0 on success, a negative errno value otherwise and rte_errno is set.
1636 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
1637 uint64_t item_flags,
1638 struct rte_eth_dev *dev,
1639 struct rte_flow_error *error)
1641 const struct rte_flow_item_vlan *mask = item->mask;
1642 const struct rte_flow_item_vlan nic_mask = {
1643 .tci = RTE_BE16(UINT16_MAX),
1644 .inner_type = RTE_BE16(UINT16_MAX),
1647 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1649 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1650 MLX5_FLOW_LAYER_INNER_L4) :
1651 (MLX5_FLOW_LAYER_OUTER_L3 |
1652 MLX5_FLOW_LAYER_OUTER_L4);
1653 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1654 MLX5_FLOW_LAYER_OUTER_VLAN;
1656 if (item_flags & vlanm)
1657 return rte_flow_error_set(error, EINVAL,
1658 RTE_FLOW_ERROR_TYPE_ITEM, item,
1659 "multiple VLAN layers not supported");
1660 else if ((item_flags & l34m) != 0)
1661 return rte_flow_error_set(error, EINVAL,
1662 RTE_FLOW_ERROR_TYPE_ITEM, item,
1663 "VLAN cannot follow L3/L4 layer");
1665 mask = &rte_flow_item_vlan_mask;
1666 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1667 (const uint8_t *)&nic_mask,
1668 sizeof(struct rte_flow_item_vlan),
1669 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1672 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1673 struct mlx5_priv *priv = dev->data->dev_private;
1675 if (priv->vmwa_context) {
1677 * Non-NULL context means we have a virtual machine
1678 * and SR-IOV enabled, we have to create VLAN interface
1679 * to make hypervisor to setup E-Switch vport
1680 * context correctly. We avoid creating the multiple
1681 * VLAN interfaces, so we cannot support VLAN tag mask.
1683 return rte_flow_error_set(error, EINVAL,
1684 RTE_FLOW_ERROR_TYPE_ITEM,
1686 "VLAN tag mask is not"
1687 " supported in virtual"
1695 * GTP flags are contained in 1 byte of the format:
1696 * -------------------------------------------
1697 * | bit | 0 - 2 | 3 | 4 | 5 | 6 | 7 |
1698 * |-----------------------------------------|
1699 * | value | Version | PT | Res | E | S | PN |
1700 * -------------------------------------------
1702 * Matching is supported only for GTP flags E, S, PN.
1704 #define MLX5_GTP_FLAGS_MASK 0x07
1707 * Validate GTP item.
1710 * Pointer to the rte_eth_dev structure.
1712 * Item specification.
1713 * @param[in] item_flags
1714 * Bit-fields that holds the items detected until now.
1716 * Pointer to error structure.
1719 * 0 on success, a negative errno value otherwise and rte_errno is set.
1722 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1723 const struct rte_flow_item *item,
1724 uint64_t item_flags,
1725 struct rte_flow_error *error)
1727 struct mlx5_priv *priv = dev->data->dev_private;
1728 const struct rte_flow_item_gtp *spec = item->spec;
1729 const struct rte_flow_item_gtp *mask = item->mask;
1730 const struct rte_flow_item_gtp nic_mask = {
1731 .v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
1733 .teid = RTE_BE32(0xffffffff),
1736 if (!priv->config.hca_attr.tunnel_stateless_gtp)
1737 return rte_flow_error_set(error, ENOTSUP,
1738 RTE_FLOW_ERROR_TYPE_ITEM, item,
1739 "GTP support is not enabled");
1740 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1741 return rte_flow_error_set(error, ENOTSUP,
1742 RTE_FLOW_ERROR_TYPE_ITEM, item,
1743 "multiple tunnel layers not"
1745 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1746 return rte_flow_error_set(error, EINVAL,
1747 RTE_FLOW_ERROR_TYPE_ITEM, item,
1748 "no outer UDP layer found");
1750 mask = &rte_flow_item_gtp_mask;
1751 if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
1752 return rte_flow_error_set(error, ENOTSUP,
1753 RTE_FLOW_ERROR_TYPE_ITEM, item,
1754 "Match is supported for GTP"
1756 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1757 (const uint8_t *)&nic_mask,
1758 sizeof(struct rte_flow_item_gtp),
1759 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1763 * Validate GTP PSC item.
1766 * Item specification.
1767 * @param[in] last_item
1768 * Previous validated item in the pattern items.
1769 * @param[in] gtp_item
1770 * Previous GTP item specification.
1772 * Pointer to flow attributes.
1774 * Pointer to error structure.
1777 * 0 on success, a negative errno value otherwise and rte_errno is set.
1780 flow_dv_validate_item_gtp_psc(const struct rte_flow_item *item,
1782 const struct rte_flow_item *gtp_item,
1783 const struct rte_flow_attr *attr,
1784 struct rte_flow_error *error)
1786 const struct rte_flow_item_gtp *gtp_spec;
1787 const struct rte_flow_item_gtp *gtp_mask;
1788 const struct rte_flow_item_gtp_psc *spec;
1789 const struct rte_flow_item_gtp_psc *mask;
1790 const struct rte_flow_item_gtp_psc nic_mask = {
1795 if (!gtp_item || !(last_item & MLX5_FLOW_LAYER_GTP))
1796 return rte_flow_error_set
1797 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
1798 "GTP PSC item must be preceded with GTP item");
1799 gtp_spec = gtp_item->spec;
1800 gtp_mask = gtp_item->mask ? gtp_item->mask : &rte_flow_item_gtp_mask;
1801 /* GTP spec and E flag is requested to match zero. */
1803 (gtp_mask->v_pt_rsv_flags &
1804 ~gtp_spec->v_pt_rsv_flags & MLX5_GTP_EXT_HEADER_FLAG))
1805 return rte_flow_error_set
1806 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
1807 "GTP E flag must be 1 to match GTP PSC");
1808 /* Check the flow is not created in group zero. */
1809 if (!attr->transfer && !attr->group)
1810 return rte_flow_error_set
1811 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1812 "GTP PSC is not supported for group 0");
1813 /* GTP spec is here and E flag is requested to match zero. */
1817 mask = item->mask ? item->mask : &rte_flow_item_gtp_psc_mask;
1818 if (spec->pdu_type > MLX5_GTP_EXT_MAX_PDU_TYPE)
1819 return rte_flow_error_set
1820 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
1821 "PDU type should be smaller than 16");
1822 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1823 (const uint8_t *)&nic_mask,
1824 sizeof(struct rte_flow_item_gtp_psc),
1825 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1829 * Validate IPV4 item.
1830 * Use existing validation function mlx5_flow_validate_item_ipv4(), and
1831 * add specific validation of fragment_offset field,
1834 * Item specification.
1835 * @param[in] item_flags
1836 * Bit-fields that holds the items detected until now.
1838 * Pointer to error structure.
1841 * 0 on success, a negative errno value otherwise and rte_errno is set.
1844 flow_dv_validate_item_ipv4(const struct rte_flow_item *item,
1845 uint64_t item_flags,
1847 uint16_t ether_type,
1848 struct rte_flow_error *error)
1851 const struct rte_flow_item_ipv4 *spec = item->spec;
1852 const struct rte_flow_item_ipv4 *last = item->last;
1853 const struct rte_flow_item_ipv4 *mask = item->mask;
1854 rte_be16_t fragment_offset_spec = 0;
1855 rte_be16_t fragment_offset_last = 0;
1856 const struct rte_flow_item_ipv4 nic_ipv4_mask = {
1858 .src_addr = RTE_BE32(0xffffffff),
1859 .dst_addr = RTE_BE32(0xffffffff),
1860 .type_of_service = 0xff,
1861 .fragment_offset = RTE_BE16(0xffff),
1862 .next_proto_id = 0xff,
1863 .time_to_live = 0xff,
1867 ret = mlx5_flow_validate_item_ipv4(item, item_flags, last_item,
1868 ether_type, &nic_ipv4_mask,
1869 MLX5_ITEM_RANGE_ACCEPTED, error);
1873 fragment_offset_spec = spec->hdr.fragment_offset &
1874 mask->hdr.fragment_offset;
1875 if (!fragment_offset_spec)
1878 * spec and mask are valid, enforce using full mask to make sure the
1879 * complete value is used correctly.
1881 if ((mask->hdr.fragment_offset & RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1882 != RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1883 return rte_flow_error_set(error, EINVAL,
1884 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1885 item, "must use full mask for"
1886 " fragment_offset");
1888 * Match on fragment_offset 0x2000 means MF is 1 and frag-offset is 0,
1889 * indicating this is 1st fragment of fragmented packet.
1890 * This is not yet supported in MLX5, return appropriate error message.
1892 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG))
1893 return rte_flow_error_set(error, ENOTSUP,
1894 RTE_FLOW_ERROR_TYPE_ITEM, item,
1895 "match on first fragment not "
1897 if (fragment_offset_spec && !last)
1898 return rte_flow_error_set(error, ENOTSUP,
1899 RTE_FLOW_ERROR_TYPE_ITEM, item,
1900 "specified value not supported");
1901 /* spec and last are valid, validate the specified range. */
1902 fragment_offset_last = last->hdr.fragment_offset &
1903 mask->hdr.fragment_offset;
1905 * Match on fragment_offset spec 0x2001 and last 0x3fff
1906 * means MF is 1 and frag-offset is > 0.
1907 * This packet is fragment 2nd and onward, excluding last.
1908 * This is not yet supported in MLX5, return appropriate
1911 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG + 1) &&
1912 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
1913 return rte_flow_error_set(error, ENOTSUP,
1914 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1915 last, "match on following "
1916 "fragments not supported");
1918 * Match on fragment_offset spec 0x0001 and last 0x1fff
1919 * means MF is 0 and frag-offset is > 0.
1920 * This packet is last fragment of fragmented packet.
1921 * This is not yet supported in MLX5, return appropriate
1924 if (fragment_offset_spec == RTE_BE16(1) &&
1925 fragment_offset_last == RTE_BE16(RTE_IPV4_HDR_OFFSET_MASK))
1926 return rte_flow_error_set(error, ENOTSUP,
1927 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1928 last, "match on last "
1929 "fragment not supported");
1931 * Match on fragment_offset spec 0x0001 and last 0x3fff
1932 * means MF and/or frag-offset is not 0.
1933 * This is a fragmented packet.
1934 * Other range values are invalid and rejected.
1936 if (!(fragment_offset_spec == RTE_BE16(1) &&
1937 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK)))
1938 return rte_flow_error_set(error, ENOTSUP,
1939 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
1940 "specified range not supported");
1945 * Validate IPV6 fragment extension item.
1948 * Item specification.
1949 * @param[in] item_flags
1950 * Bit-fields that holds the items detected until now.
1952 * Pointer to error structure.
1955 * 0 on success, a negative errno value otherwise and rte_errno is set.
1958 flow_dv_validate_item_ipv6_frag_ext(const struct rte_flow_item *item,
1959 uint64_t item_flags,
1960 struct rte_flow_error *error)
1962 const struct rte_flow_item_ipv6_frag_ext *spec = item->spec;
1963 const struct rte_flow_item_ipv6_frag_ext *last = item->last;
1964 const struct rte_flow_item_ipv6_frag_ext *mask = item->mask;
1965 rte_be16_t frag_data_spec = 0;
1966 rte_be16_t frag_data_last = 0;
1967 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1968 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1969 MLX5_FLOW_LAYER_OUTER_L4;
1971 struct rte_flow_item_ipv6_frag_ext nic_mask = {
1973 .next_header = 0xff,
1974 .frag_data = RTE_BE16(0xffff),
1978 if (item_flags & l4m)
1979 return rte_flow_error_set(error, EINVAL,
1980 RTE_FLOW_ERROR_TYPE_ITEM, item,
1981 "ipv6 fragment extension item cannot "
1983 if ((tunnel && !(item_flags & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
1984 (!tunnel && !(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
1985 return rte_flow_error_set(error, EINVAL,
1986 RTE_FLOW_ERROR_TYPE_ITEM, item,
1987 "ipv6 fragment extension item must "
1988 "follow ipv6 item");
1990 frag_data_spec = spec->hdr.frag_data & mask->hdr.frag_data;
1991 if (!frag_data_spec)
1994 * spec and mask are valid, enforce using full mask to make sure the
1995 * complete value is used correctly.
1997 if ((mask->hdr.frag_data & RTE_BE16(RTE_IPV6_FRAG_USED_MASK)) !=
1998 RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
1999 return rte_flow_error_set(error, EINVAL,
2000 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2001 item, "must use full mask for"
2004 * Match on frag_data 0x00001 means M is 1 and frag-offset is 0.
2005 * This is 1st fragment of fragmented packet.
2007 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_MF_MASK))
2008 return rte_flow_error_set(error, ENOTSUP,
2009 RTE_FLOW_ERROR_TYPE_ITEM, item,
2010 "match on first fragment not "
2012 if (frag_data_spec && !last)
2013 return rte_flow_error_set(error, EINVAL,
2014 RTE_FLOW_ERROR_TYPE_ITEM, item,
2015 "specified value not supported");
2016 ret = mlx5_flow_item_acceptable
2017 (item, (const uint8_t *)mask,
2018 (const uint8_t *)&nic_mask,
2019 sizeof(struct rte_flow_item_ipv6_frag_ext),
2020 MLX5_ITEM_RANGE_ACCEPTED, error);
2023 /* spec and last are valid, validate the specified range. */
2024 frag_data_last = last->hdr.frag_data & mask->hdr.frag_data;
2026 * Match on frag_data spec 0x0009 and last 0xfff9
2027 * means M is 1 and frag-offset is > 0.
2028 * This packet is fragment 2nd and onward, excluding last.
2029 * This is not yet supported in MLX5, return appropriate
2032 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN |
2033 RTE_IPV6_EHDR_MF_MASK) &&
2034 frag_data_last == RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
2035 return rte_flow_error_set(error, ENOTSUP,
2036 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2037 last, "match on following "
2038 "fragments not supported");
2040 * Match on frag_data spec 0x0008 and last 0xfff8
2041 * means M is 0 and frag-offset is > 0.
2042 * This packet is last fragment of fragmented packet.
2043 * This is not yet supported in MLX5, return appropriate
2046 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN) &&
2047 frag_data_last == RTE_BE16(RTE_IPV6_EHDR_FO_MASK))
2048 return rte_flow_error_set(error, ENOTSUP,
2049 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2050 last, "match on last "
2051 "fragment not supported");
2052 /* Other range values are invalid and rejected. */
2053 return rte_flow_error_set(error, EINVAL,
2054 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
2055 "specified range not supported");
2059 * Validate the pop VLAN action.
2062 * Pointer to the rte_eth_dev structure.
2063 * @param[in] action_flags
2064 * Holds the actions detected until now.
2066 * Pointer to the pop vlan action.
2067 * @param[in] item_flags
2068 * The items found in this flow rule.
2070 * Pointer to flow attributes.
2072 * Pointer to error structure.
2075 * 0 on success, a negative errno value otherwise and rte_errno is set.
2078 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
2079 uint64_t action_flags,
2080 const struct rte_flow_action *action,
2081 uint64_t item_flags,
2082 const struct rte_flow_attr *attr,
2083 struct rte_flow_error *error)
2085 const struct mlx5_priv *priv = dev->data->dev_private;
2089 if (!priv->sh->pop_vlan_action)
2090 return rte_flow_error_set(error, ENOTSUP,
2091 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2093 "pop vlan action is not supported");
2095 return rte_flow_error_set(error, ENOTSUP,
2096 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2098 "pop vlan action not supported for "
2100 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
2101 return rte_flow_error_set(error, ENOTSUP,
2102 RTE_FLOW_ERROR_TYPE_ACTION, action,
2103 "no support for multiple VLAN "
2105 /* Pop VLAN with preceding Decap requires inner header with VLAN. */
2106 if ((action_flags & MLX5_FLOW_ACTION_DECAP) &&
2107 !(item_flags & MLX5_FLOW_LAYER_INNER_VLAN))
2108 return rte_flow_error_set(error, ENOTSUP,
2109 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2111 "cannot pop vlan after decap without "
2112 "match on inner vlan in the flow");
2113 /* Pop VLAN without preceding Decap requires outer header with VLAN. */
2114 if (!(action_flags & MLX5_FLOW_ACTION_DECAP) &&
2115 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2116 return rte_flow_error_set(error, ENOTSUP,
2117 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2119 "cannot pop vlan without a "
2120 "match on (outer) vlan in the flow");
2121 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2122 return rte_flow_error_set(error, EINVAL,
2123 RTE_FLOW_ERROR_TYPE_ACTION, action,
2124 "wrong action order, port_id should "
2125 "be after pop VLAN action");
2126 if (!attr->transfer && priv->representor)
2127 return rte_flow_error_set(error, ENOTSUP,
2128 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2129 "pop vlan action for VF representor "
2130 "not supported on NIC table");
2135 * Get VLAN default info from vlan match info.
2138 * the list of item specifications.
2140 * pointer VLAN info to fill to.
2143 * 0 on success, a negative errno value otherwise and rte_errno is set.
2146 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
2147 struct rte_vlan_hdr *vlan)
2149 const struct rte_flow_item_vlan nic_mask = {
2150 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
2151 MLX5DV_FLOW_VLAN_VID_MASK),
2152 .inner_type = RTE_BE16(0xffff),
2157 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2158 int type = items->type;
2160 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
2161 type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
2164 if (items->type != RTE_FLOW_ITEM_TYPE_END) {
2165 const struct rte_flow_item_vlan *vlan_m = items->mask;
2166 const struct rte_flow_item_vlan *vlan_v = items->spec;
2168 /* If VLAN item in pattern doesn't contain data, return here. */
2173 /* Only full match values are accepted */
2174 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
2175 MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
2176 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
2178 rte_be_to_cpu_16(vlan_v->tci &
2179 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
2181 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
2182 MLX5DV_FLOW_VLAN_VID_MASK_BE) {
2183 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
2185 rte_be_to_cpu_16(vlan_v->tci &
2186 MLX5DV_FLOW_VLAN_VID_MASK_BE);
2188 if (vlan_m->inner_type == nic_mask.inner_type)
2189 vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
2190 vlan_m->inner_type);
2195 * Validate the push VLAN action.
2198 * Pointer to the rte_eth_dev structure.
2199 * @param[in] action_flags
2200 * Holds the actions detected until now.
2201 * @param[in] item_flags
2202 * The items found in this flow rule.
2204 * Pointer to the action structure.
2206 * Pointer to flow attributes
2208 * Pointer to error structure.
2211 * 0 on success, a negative errno value otherwise and rte_errno is set.
2214 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
2215 uint64_t action_flags,
2216 const struct rte_flow_item_vlan *vlan_m,
2217 const struct rte_flow_action *action,
2218 const struct rte_flow_attr *attr,
2219 struct rte_flow_error *error)
2221 const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
2222 const struct mlx5_priv *priv = dev->data->dev_private;
2224 if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
2225 push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
2226 return rte_flow_error_set(error, EINVAL,
2227 RTE_FLOW_ERROR_TYPE_ACTION, action,
2228 "invalid vlan ethertype");
2229 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2230 return rte_flow_error_set(error, EINVAL,
2231 RTE_FLOW_ERROR_TYPE_ACTION, action,
2232 "wrong action order, port_id should "
2233 "be after push VLAN");
2234 if (!attr->transfer && priv->representor)
2235 return rte_flow_error_set(error, ENOTSUP,
2236 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2237 "push vlan action for VF representor "
2238 "not supported on NIC table");
2240 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
2241 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
2242 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
2243 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
2244 !(mlx5_flow_find_action
2245 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
2246 return rte_flow_error_set(error, EINVAL,
2247 RTE_FLOW_ERROR_TYPE_ACTION, action,
2248 "not full match mask on VLAN PCP and "
2249 "there is no of_set_vlan_pcp action, "
2250 "push VLAN action cannot figure out "
2253 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
2254 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
2255 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
2256 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
2257 !(mlx5_flow_find_action
2258 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
2259 return rte_flow_error_set(error, EINVAL,
2260 RTE_FLOW_ERROR_TYPE_ACTION, action,
2261 "not full match mask on VLAN VID and "
2262 "there is no of_set_vlan_vid action, "
2263 "push VLAN action cannot figure out "
2270 * Validate the set VLAN PCP.
2272 * @param[in] action_flags
2273 * Holds the actions detected until now.
2274 * @param[in] actions
2275 * Pointer to the list of actions remaining in the flow rule.
2277 * Pointer to error structure.
2280 * 0 on success, a negative errno value otherwise and rte_errno is set.
2283 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
2284 const struct rte_flow_action actions[],
2285 struct rte_flow_error *error)
2287 const struct rte_flow_action *action = actions;
2288 const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2290 if (conf->vlan_pcp > 7)
2291 return rte_flow_error_set(error, EINVAL,
2292 RTE_FLOW_ERROR_TYPE_ACTION, action,
2293 "VLAN PCP value is too big");
2294 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2295 return rte_flow_error_set(error, ENOTSUP,
2296 RTE_FLOW_ERROR_TYPE_ACTION, action,
2297 "set VLAN PCP action must follow "
2298 "the push VLAN action");
2299 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2300 return rte_flow_error_set(error, ENOTSUP,
2301 RTE_FLOW_ERROR_TYPE_ACTION, action,
2302 "Multiple VLAN PCP modification are "
2304 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2305 return rte_flow_error_set(error, EINVAL,
2306 RTE_FLOW_ERROR_TYPE_ACTION, action,
2307 "wrong action order, port_id should "
2308 "be after set VLAN PCP");
2313 * Validate the set VLAN VID.
2315 * @param[in] item_flags
2316 * Holds the items detected in this rule.
2317 * @param[in] action_flags
2318 * Holds the actions detected until now.
2319 * @param[in] actions
2320 * Pointer to the list of actions remaining in the flow rule.
2322 * Pointer to error structure.
2325 * 0 on success, a negative errno value otherwise and rte_errno is set.
2328 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
2329 uint64_t action_flags,
2330 const struct rte_flow_action actions[],
2331 struct rte_flow_error *error)
2333 const struct rte_flow_action *action = actions;
2334 const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
2336 if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
2337 return rte_flow_error_set(error, EINVAL,
2338 RTE_FLOW_ERROR_TYPE_ACTION, action,
2339 "VLAN VID value is too big");
2340 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2341 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2342 return rte_flow_error_set(error, ENOTSUP,
2343 RTE_FLOW_ERROR_TYPE_ACTION, action,
2344 "set VLAN VID action must follow push"
2345 " VLAN action or match on VLAN item");
2346 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
2347 return rte_flow_error_set(error, ENOTSUP,
2348 RTE_FLOW_ERROR_TYPE_ACTION, action,
2349 "Multiple VLAN VID modifications are "
2351 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2352 return rte_flow_error_set(error, EINVAL,
2353 RTE_FLOW_ERROR_TYPE_ACTION, action,
2354 "wrong action order, port_id should "
2355 "be after set VLAN VID");
2360 * Validate the FLAG action.
2363 * Pointer to the rte_eth_dev structure.
2364 * @param[in] action_flags
2365 * Holds the actions detected until now.
2367 * Pointer to flow attributes
2369 * Pointer to error structure.
2372 * 0 on success, a negative errno value otherwise and rte_errno is set.
2375 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2376 uint64_t action_flags,
2377 const struct rte_flow_attr *attr,
2378 struct rte_flow_error *error)
2380 struct mlx5_priv *priv = dev->data->dev_private;
2381 struct mlx5_dev_config *config = &priv->config;
2384 /* Fall back if no extended metadata register support. */
2385 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2386 return mlx5_flow_validate_action_flag(action_flags, attr,
2388 /* Extensive metadata mode requires registers. */
2389 if (!mlx5_flow_ext_mreg_supported(dev))
2390 return rte_flow_error_set(error, ENOTSUP,
2391 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2392 "no metadata registers "
2393 "to support flag action");
2394 if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2395 return rte_flow_error_set(error, ENOTSUP,
2396 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2397 "extended metadata register"
2398 " isn't available");
2399 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2402 MLX5_ASSERT(ret > 0);
2403 if (action_flags & MLX5_FLOW_ACTION_MARK)
2404 return rte_flow_error_set(error, EINVAL,
2405 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2406 "can't mark and flag in same flow");
2407 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2408 return rte_flow_error_set(error, EINVAL,
2409 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2411 " actions in same flow");
2416 * Validate MARK action.
2419 * Pointer to the rte_eth_dev structure.
2421 * Pointer to action.
2422 * @param[in] action_flags
2423 * Holds the actions detected until now.
2425 * Pointer to flow attributes
2427 * Pointer to error structure.
2430 * 0 on success, a negative errno value otherwise and rte_errno is set.
2433 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2434 const struct rte_flow_action *action,
2435 uint64_t action_flags,
2436 const struct rte_flow_attr *attr,
2437 struct rte_flow_error *error)
2439 struct mlx5_priv *priv = dev->data->dev_private;
2440 struct mlx5_dev_config *config = &priv->config;
2441 const struct rte_flow_action_mark *mark = action->conf;
2444 /* Fall back if no extended metadata register support. */
2445 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2446 return mlx5_flow_validate_action_mark(action, action_flags,
2448 /* Extensive metadata mode requires registers. */
2449 if (!mlx5_flow_ext_mreg_supported(dev))
2450 return rte_flow_error_set(error, ENOTSUP,
2451 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2452 "no metadata registers "
2453 "to support mark action");
2454 if (!priv->sh->dv_mark_mask)
2455 return rte_flow_error_set(error, ENOTSUP,
2456 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2457 "extended metadata register"
2458 " isn't available");
2459 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2462 MLX5_ASSERT(ret > 0);
2464 return rte_flow_error_set(error, EINVAL,
2465 RTE_FLOW_ERROR_TYPE_ACTION, action,
2466 "configuration cannot be null");
2467 if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2468 return rte_flow_error_set(error, EINVAL,
2469 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2471 "mark id exceeds the limit");
2472 if (action_flags & MLX5_FLOW_ACTION_FLAG)
2473 return rte_flow_error_set(error, EINVAL,
2474 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2475 "can't flag and mark in same flow");
2476 if (action_flags & MLX5_FLOW_ACTION_MARK)
2477 return rte_flow_error_set(error, EINVAL,
2478 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2479 "can't have 2 mark actions in same"
2485 * Validate SET_META action.
2488 * Pointer to the rte_eth_dev structure.
2490 * Pointer to the action structure.
2491 * @param[in] action_flags
2492 * Holds the actions detected until now.
2494 * Pointer to flow attributes
2496 * Pointer to error structure.
2499 * 0 on success, a negative errno value otherwise and rte_errno is set.
2502 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2503 const struct rte_flow_action *action,
2504 uint64_t action_flags __rte_unused,
2505 const struct rte_flow_attr *attr,
2506 struct rte_flow_error *error)
2508 const struct rte_flow_action_set_meta *conf;
2509 uint32_t nic_mask = UINT32_MAX;
2512 if (!mlx5_flow_ext_mreg_supported(dev))
2513 return rte_flow_error_set(error, ENOTSUP,
2514 RTE_FLOW_ERROR_TYPE_ACTION, action,
2515 "extended metadata register"
2516 " isn't supported");
2517 reg = flow_dv_get_metadata_reg(dev, attr, error);
2521 return rte_flow_error_set(error, ENOTSUP,
2522 RTE_FLOW_ERROR_TYPE_ACTION, action,
2523 "unavalable extended metadata register");
2524 if (reg != REG_A && reg != REG_B) {
2525 struct mlx5_priv *priv = dev->data->dev_private;
2527 nic_mask = priv->sh->dv_meta_mask;
2529 if (!(action->conf))
2530 return rte_flow_error_set(error, EINVAL,
2531 RTE_FLOW_ERROR_TYPE_ACTION, action,
2532 "configuration cannot be null");
2533 conf = (const struct rte_flow_action_set_meta *)action->conf;
2535 return rte_flow_error_set(error, EINVAL,
2536 RTE_FLOW_ERROR_TYPE_ACTION, action,
2537 "zero mask doesn't have any effect");
2538 if (conf->mask & ~nic_mask)
2539 return rte_flow_error_set(error, EINVAL,
2540 RTE_FLOW_ERROR_TYPE_ACTION, action,
2541 "meta data must be within reg C0");
2546 * Validate SET_TAG action.
2549 * Pointer to the rte_eth_dev structure.
2551 * Pointer to the action structure.
2552 * @param[in] action_flags
2553 * Holds the actions detected until now.
2555 * Pointer to flow attributes
2557 * Pointer to error structure.
2560 * 0 on success, a negative errno value otherwise and rte_errno is set.
2563 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2564 const struct rte_flow_action *action,
2565 uint64_t action_flags,
2566 const struct rte_flow_attr *attr,
2567 struct rte_flow_error *error)
2569 const struct rte_flow_action_set_tag *conf;
2570 const uint64_t terminal_action_flags =
2571 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2572 MLX5_FLOW_ACTION_RSS;
2575 if (!mlx5_flow_ext_mreg_supported(dev))
2576 return rte_flow_error_set(error, ENOTSUP,
2577 RTE_FLOW_ERROR_TYPE_ACTION, action,
2578 "extensive metadata register"
2579 " isn't supported");
2580 if (!(action->conf))
2581 return rte_flow_error_set(error, EINVAL,
2582 RTE_FLOW_ERROR_TYPE_ACTION, action,
2583 "configuration cannot be null");
2584 conf = (const struct rte_flow_action_set_tag *)action->conf;
2586 return rte_flow_error_set(error, EINVAL,
2587 RTE_FLOW_ERROR_TYPE_ACTION, action,
2588 "zero mask doesn't have any effect");
2589 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2592 if (!attr->transfer && attr->ingress &&
2593 (action_flags & terminal_action_flags))
2594 return rte_flow_error_set(error, EINVAL,
2595 RTE_FLOW_ERROR_TYPE_ACTION, action,
2596 "set_tag has no effect"
2597 " with terminal actions");
2602 * Validate count action.
2605 * Pointer to rte_eth_dev structure.
2607 * Pointer to error structure.
2610 * 0 on success, a negative errno value otherwise and rte_errno is set.
2613 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2614 struct rte_flow_error *error)
2616 struct mlx5_priv *priv = dev->data->dev_private;
2618 if (!priv->config.devx)
2620 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2624 return rte_flow_error_set
2626 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2628 "count action not supported");
2632 * Validate the L2 encap action.
2635 * Pointer to the rte_eth_dev structure.
2636 * @param[in] action_flags
2637 * Holds the actions detected until now.
2639 * Pointer to the action structure.
2641 * Pointer to flow attributes.
2643 * Pointer to error structure.
2646 * 0 on success, a negative errno value otherwise and rte_errno is set.
2649 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2650 uint64_t action_flags,
2651 const struct rte_flow_action *action,
2652 const struct rte_flow_attr *attr,
2653 struct rte_flow_error *error)
2655 const struct mlx5_priv *priv = dev->data->dev_private;
2657 if (!(action->conf))
2658 return rte_flow_error_set(error, EINVAL,
2659 RTE_FLOW_ERROR_TYPE_ACTION, action,
2660 "configuration cannot be null");
2661 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2662 return rte_flow_error_set(error, EINVAL,
2663 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2664 "can only have a single encap action "
2666 if (!attr->transfer && priv->representor)
2667 return rte_flow_error_set(error, ENOTSUP,
2668 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2669 "encap action for VF representor "
2670 "not supported on NIC table");
2675 * Validate a decap action.
2678 * Pointer to the rte_eth_dev structure.
2679 * @param[in] action_flags
2680 * Holds the actions detected until now.
2682 * Pointer to the action structure.
2683 * @param[in] item_flags
2684 * Holds the items detected.
2686 * Pointer to flow attributes
2688 * Pointer to error structure.
2691 * 0 on success, a negative errno value otherwise and rte_errno is set.
2694 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2695 uint64_t action_flags,
2696 const struct rte_flow_action *action,
2697 const uint64_t item_flags,
2698 const struct rte_flow_attr *attr,
2699 struct rte_flow_error *error)
2701 const struct mlx5_priv *priv = dev->data->dev_private;
2703 if (priv->config.hca_attr.scatter_fcs_w_decap_disable &&
2704 !priv->config.decap_en)
2705 return rte_flow_error_set(error, ENOTSUP,
2706 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2707 "decap is not enabled");
2708 if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2709 return rte_flow_error_set(error, ENOTSUP,
2710 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2712 MLX5_FLOW_ACTION_DECAP ? "can only "
2713 "have a single decap action" : "decap "
2714 "after encap is not supported");
2715 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2716 return rte_flow_error_set(error, EINVAL,
2717 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2718 "can't have decap action after"
2721 return rte_flow_error_set(error, ENOTSUP,
2722 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2724 "decap action not supported for "
2726 if (!attr->transfer && priv->representor)
2727 return rte_flow_error_set(error, ENOTSUP,
2728 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2729 "decap action for VF representor "
2730 "not supported on NIC table");
2731 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_DECAP &&
2732 !(item_flags & MLX5_FLOW_LAYER_VXLAN))
2733 return rte_flow_error_set(error, ENOTSUP,
2734 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2735 "VXLAN item should be present for VXLAN decap");
2739 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2742 * Validate the raw encap and decap actions.
2745 * Pointer to the rte_eth_dev structure.
2747 * Pointer to the decap action.
2749 * Pointer to the encap action.
2751 * Pointer to flow attributes
2752 * @param[in/out] action_flags
2753 * Holds the actions detected until now.
2754 * @param[out] actions_n
2755 * pointer to the number of actions counter.
2757 * Pointer to the action structure.
2758 * @param[in] item_flags
2759 * Holds the items detected.
2761 * Pointer to error structure.
2764 * 0 on success, a negative errno value otherwise and rte_errno is set.
2767 flow_dv_validate_action_raw_encap_decap
2768 (struct rte_eth_dev *dev,
2769 const struct rte_flow_action_raw_decap *decap,
2770 const struct rte_flow_action_raw_encap *encap,
2771 const struct rte_flow_attr *attr, uint64_t *action_flags,
2772 int *actions_n, const struct rte_flow_action *action,
2773 uint64_t item_flags, struct rte_flow_error *error)
2775 const struct mlx5_priv *priv = dev->data->dev_private;
2778 if (encap && (!encap->size || !encap->data))
2779 return rte_flow_error_set(error, EINVAL,
2780 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2781 "raw encap data cannot be empty");
2782 if (decap && encap) {
2783 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2784 encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2787 else if (encap->size <=
2788 MLX5_ENCAPSULATION_DECISION_SIZE &&
2790 MLX5_ENCAPSULATION_DECISION_SIZE)
2793 else if (encap->size >
2794 MLX5_ENCAPSULATION_DECISION_SIZE &&
2796 MLX5_ENCAPSULATION_DECISION_SIZE)
2797 /* 2 L2 actions: encap and decap. */
2800 return rte_flow_error_set(error,
2802 RTE_FLOW_ERROR_TYPE_ACTION,
2803 NULL, "unsupported too small "
2804 "raw decap and too small raw "
2805 "encap combination");
2808 ret = flow_dv_validate_action_decap(dev, *action_flags, action,
2809 item_flags, attr, error);
2812 *action_flags |= MLX5_FLOW_ACTION_DECAP;
2816 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2817 return rte_flow_error_set(error, ENOTSUP,
2818 RTE_FLOW_ERROR_TYPE_ACTION,
2820 "small raw encap size");
2821 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2822 return rte_flow_error_set(error, EINVAL,
2823 RTE_FLOW_ERROR_TYPE_ACTION,
2825 "more than one encap action");
2826 if (!attr->transfer && priv->representor)
2827 return rte_flow_error_set
2829 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2830 "encap action for VF representor "
2831 "not supported on NIC table");
2832 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
2839 * Match encap_decap resource.
2842 * Pointer to the hash list.
2844 * Pointer to exist resource entry object.
2846 * Key of the new entry.
2848 * Pointer to new encap_decap resource.
2851 * 0 on matching, none-zero otherwise.
2854 flow_dv_encap_decap_match_cb(struct mlx5_hlist *list __rte_unused,
2855 struct mlx5_hlist_entry *entry,
2856 uint64_t key __rte_unused, void *cb_ctx)
2858 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
2859 struct mlx5_flow_dv_encap_decap_resource *resource = ctx->data;
2860 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2862 cache_resource = container_of(entry,
2863 struct mlx5_flow_dv_encap_decap_resource,
2865 if (resource->reformat_type == cache_resource->reformat_type &&
2866 resource->ft_type == cache_resource->ft_type &&
2867 resource->flags == cache_resource->flags &&
2868 resource->size == cache_resource->size &&
2869 !memcmp((const void *)resource->buf,
2870 (const void *)cache_resource->buf,
2877 * Allocate encap_decap resource.
2880 * Pointer to the hash list.
2882 * Pointer to exist resource entry object.
2884 * Pointer to new encap_decap resource.
2887 * 0 on matching, none-zero otherwise.
2889 struct mlx5_hlist_entry *
2890 flow_dv_encap_decap_create_cb(struct mlx5_hlist *list,
2891 uint64_t key __rte_unused,
2894 struct mlx5_dev_ctx_shared *sh = list->ctx;
2895 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
2896 struct mlx5dv_dr_domain *domain;
2897 struct mlx5_flow_dv_encap_decap_resource *resource = ctx->data;
2898 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2902 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2903 domain = sh->fdb_domain;
2904 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2905 domain = sh->rx_domain;
2907 domain = sh->tx_domain;
2908 /* Register new encap/decap resource. */
2909 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2911 if (!cache_resource) {
2912 rte_flow_error_set(ctx->error, ENOMEM,
2913 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2914 "cannot allocate resource memory");
2917 *cache_resource = *resource;
2918 cache_resource->idx = idx;
2919 ret = mlx5_flow_os_create_flow_action_packet_reformat
2920 (sh->ctx, domain, cache_resource,
2921 &cache_resource->action);
2923 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
2924 rte_flow_error_set(ctx->error, ENOMEM,
2925 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2926 NULL, "cannot create action");
2930 return &cache_resource->entry;
2934 * Find existing encap/decap resource or create and register a new one.
2936 * @param[in, out] dev
2937 * Pointer to rte_eth_dev structure.
2938 * @param[in, out] resource
2939 * Pointer to encap/decap 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_encap_decap_resource_register
2950 (struct rte_eth_dev *dev,
2951 struct mlx5_flow_dv_encap_decap_resource *resource,
2952 struct mlx5_flow *dev_flow,
2953 struct rte_flow_error *error)
2955 struct mlx5_priv *priv = dev->data->dev_private;
2956 struct mlx5_dev_ctx_shared *sh = priv->sh;
2957 struct mlx5_hlist_entry *entry;
2961 uint32_t refmt_type:8;
2963 * Header reformat actions can be shared between
2964 * non-root tables. One bit to indicate non-root
2968 uint32_t reserve:15;
2971 } encap_decap_key = {
2973 .ft_type = resource->ft_type,
2974 .refmt_type = resource->reformat_type,
2975 .is_root = !!dev_flow->dv.group,
2979 struct mlx5_flow_cb_ctx ctx = {
2985 resource->flags = dev_flow->dv.group ? 0 : 1;
2986 key64 = __rte_raw_cksum(&encap_decap_key.v32,
2987 sizeof(encap_decap_key.v32), 0);
2988 if (resource->reformat_type !=
2989 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2 &&
2991 key64 = __rte_raw_cksum(resource->buf, resource->size, key64);
2992 entry = mlx5_hlist_register(sh->encaps_decaps, key64, &ctx);
2995 resource = container_of(entry, typeof(*resource), entry);
2996 dev_flow->dv.encap_decap = resource;
2997 dev_flow->handle->dvh.rix_encap_decap = resource->idx;
3002 * Find existing table jump resource or create and register a new one.
3004 * @param[in, out] dev
3005 * Pointer to rte_eth_dev structure.
3006 * @param[in, out] tbl
3007 * Pointer to flow table resource.
3008 * @parm[in, out] dev_flow
3009 * Pointer to the dev_flow.
3011 * pointer to error structure.
3014 * 0 on success otherwise -errno and errno is set.
3017 flow_dv_jump_tbl_resource_register
3018 (struct rte_eth_dev *dev __rte_unused,
3019 struct mlx5_flow_tbl_resource *tbl,
3020 struct mlx5_flow *dev_flow,
3021 struct rte_flow_error *error __rte_unused)
3023 struct mlx5_flow_tbl_data_entry *tbl_data =
3024 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
3027 MLX5_ASSERT(tbl_data->jump.action);
3028 dev_flow->handle->rix_jump = tbl_data->idx;
3029 dev_flow->dv.jump = &tbl_data->jump;
3034 flow_dv_port_id_match_cb(struct mlx5_cache_list *list __rte_unused,
3035 struct mlx5_cache_entry *entry, void *cb_ctx)
3037 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3038 struct mlx5_flow_dv_port_id_action_resource *ref = ctx->data;
3039 struct mlx5_flow_dv_port_id_action_resource *res =
3040 container_of(entry, typeof(*res), entry);
3042 return ref->port_id != res->port_id;
3045 struct mlx5_cache_entry *
3046 flow_dv_port_id_create_cb(struct mlx5_cache_list *list,
3047 struct mlx5_cache_entry *entry __rte_unused,
3050 struct mlx5_dev_ctx_shared *sh = list->ctx;
3051 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3052 struct mlx5_flow_dv_port_id_action_resource *ref = ctx->data;
3053 struct mlx5_flow_dv_port_id_action_resource *cache;
3057 /* Register new port id action resource. */
3058 cache = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID], &idx);
3060 rte_flow_error_set(ctx->error, ENOMEM,
3061 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3062 "cannot allocate port_id action cache memory");
3066 ret = mlx5_flow_os_create_flow_action_dest_port(sh->fdb_domain,
3070 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PORT_ID], idx);
3071 rte_flow_error_set(ctx->error, ENOMEM,
3072 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3073 "cannot create action");
3076 return &cache->entry;
3080 * Find existing table port ID resource or create and register a new one.
3082 * @param[in, out] dev
3083 * Pointer to rte_eth_dev structure.
3084 * @param[in, out] resource
3085 * Pointer to port ID action resource.
3086 * @parm[in, out] dev_flow
3087 * Pointer to the dev_flow.
3089 * pointer to error structure.
3092 * 0 on success otherwise -errno and errno is set.
3095 flow_dv_port_id_action_resource_register
3096 (struct rte_eth_dev *dev,
3097 struct mlx5_flow_dv_port_id_action_resource *resource,
3098 struct mlx5_flow *dev_flow,
3099 struct rte_flow_error *error)
3101 struct mlx5_priv *priv = dev->data->dev_private;
3102 struct mlx5_cache_entry *entry;
3103 struct mlx5_flow_dv_port_id_action_resource *cache;
3104 struct mlx5_flow_cb_ctx ctx = {
3109 entry = mlx5_cache_register(&priv->sh->port_id_action_list, &ctx);
3112 cache = container_of(entry, typeof(*cache), entry);
3113 dev_flow->dv.port_id_action = cache;
3114 dev_flow->handle->rix_port_id_action = cache->idx;
3119 flow_dv_push_vlan_match_cb(struct mlx5_cache_list *list __rte_unused,
3120 struct mlx5_cache_entry *entry, void *cb_ctx)
3122 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3123 struct mlx5_flow_dv_push_vlan_action_resource *ref = ctx->data;
3124 struct mlx5_flow_dv_push_vlan_action_resource *res =
3125 container_of(entry, typeof(*res), entry);
3127 return ref->vlan_tag != res->vlan_tag || ref->ft_type != res->ft_type;
3130 struct mlx5_cache_entry *
3131 flow_dv_push_vlan_create_cb(struct mlx5_cache_list *list,
3132 struct mlx5_cache_entry *entry __rte_unused,
3135 struct mlx5_dev_ctx_shared *sh = list->ctx;
3136 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3137 struct mlx5_flow_dv_push_vlan_action_resource *ref = ctx->data;
3138 struct mlx5_flow_dv_push_vlan_action_resource *cache;
3139 struct mlx5dv_dr_domain *domain;
3143 /* Register new port id action resource. */
3144 cache = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN], &idx);
3146 rte_flow_error_set(ctx->error, ENOMEM,
3147 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3148 "cannot allocate push_vlan action cache memory");
3152 if (ref->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3153 domain = sh->fdb_domain;
3154 else if (ref->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
3155 domain = sh->rx_domain;
3157 domain = sh->tx_domain;
3158 ret = mlx5_flow_os_create_flow_action_push_vlan(domain, ref->vlan_tag,
3161 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
3162 rte_flow_error_set(ctx->error, ENOMEM,
3163 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3164 "cannot create push vlan action");
3167 return &cache->entry;
3171 * Find existing push vlan resource or create and register a new one.
3173 * @param [in, out] dev
3174 * Pointer to rte_eth_dev structure.
3175 * @param[in, out] resource
3176 * Pointer to port ID action resource.
3177 * @parm[in, out] dev_flow
3178 * Pointer to the dev_flow.
3180 * pointer to error structure.
3183 * 0 on success otherwise -errno and errno is set.
3186 flow_dv_push_vlan_action_resource_register
3187 (struct rte_eth_dev *dev,
3188 struct mlx5_flow_dv_push_vlan_action_resource *resource,
3189 struct mlx5_flow *dev_flow,
3190 struct rte_flow_error *error)
3192 struct mlx5_priv *priv = dev->data->dev_private;
3193 struct mlx5_flow_dv_push_vlan_action_resource *cache;
3194 struct mlx5_cache_entry *entry;
3195 struct mlx5_flow_cb_ctx ctx = {
3200 entry = mlx5_cache_register(&priv->sh->push_vlan_action_list, &ctx);
3203 cache = container_of(entry, typeof(*cache), entry);
3205 dev_flow->handle->dvh.rix_push_vlan = cache->idx;
3206 dev_flow->dv.push_vlan_res = cache;
3211 * Get the size of specific rte_flow_item_type hdr size
3213 * @param[in] item_type
3214 * Tested rte_flow_item_type.
3217 * sizeof struct item_type, 0 if void or irrelevant.
3220 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
3224 switch (item_type) {
3225 case RTE_FLOW_ITEM_TYPE_ETH:
3226 retval = sizeof(struct rte_ether_hdr);
3228 case RTE_FLOW_ITEM_TYPE_VLAN:
3229 retval = sizeof(struct rte_vlan_hdr);
3231 case RTE_FLOW_ITEM_TYPE_IPV4:
3232 retval = sizeof(struct rte_ipv4_hdr);
3234 case RTE_FLOW_ITEM_TYPE_IPV6:
3235 retval = sizeof(struct rte_ipv6_hdr);
3237 case RTE_FLOW_ITEM_TYPE_UDP:
3238 retval = sizeof(struct rte_udp_hdr);
3240 case RTE_FLOW_ITEM_TYPE_TCP:
3241 retval = sizeof(struct rte_tcp_hdr);
3243 case RTE_FLOW_ITEM_TYPE_VXLAN:
3244 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3245 retval = sizeof(struct rte_vxlan_hdr);
3247 case RTE_FLOW_ITEM_TYPE_GRE:
3248 case RTE_FLOW_ITEM_TYPE_NVGRE:
3249 retval = sizeof(struct rte_gre_hdr);
3251 case RTE_FLOW_ITEM_TYPE_MPLS:
3252 retval = sizeof(struct rte_mpls_hdr);
3254 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
3262 #define MLX5_ENCAP_IPV4_VERSION 0x40
3263 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
3264 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
3265 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
3266 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
3267 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
3268 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
3271 * Convert the encap action data from list of rte_flow_item to raw buffer
3274 * Pointer to rte_flow_item objects list.
3276 * Pointer to the output buffer.
3278 * Pointer to the output buffer size.
3280 * Pointer to the error structure.
3283 * 0 on success, a negative errno value otherwise and rte_errno is set.
3286 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
3287 size_t *size, struct rte_flow_error *error)
3289 struct rte_ether_hdr *eth = NULL;
3290 struct rte_vlan_hdr *vlan = NULL;
3291 struct rte_ipv4_hdr *ipv4 = NULL;
3292 struct rte_ipv6_hdr *ipv6 = NULL;
3293 struct rte_udp_hdr *udp = NULL;
3294 struct rte_vxlan_hdr *vxlan = NULL;
3295 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
3296 struct rte_gre_hdr *gre = NULL;
3298 size_t temp_size = 0;
3301 return rte_flow_error_set(error, EINVAL,
3302 RTE_FLOW_ERROR_TYPE_ACTION,
3303 NULL, "invalid empty data");
3304 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3305 len = flow_dv_get_item_hdr_len(items->type);
3306 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
3307 return rte_flow_error_set(error, EINVAL,
3308 RTE_FLOW_ERROR_TYPE_ACTION,
3309 (void *)items->type,
3310 "items total size is too big"
3311 " for encap action");
3312 rte_memcpy((void *)&buf[temp_size], items->spec, len);
3313 switch (items->type) {
3314 case RTE_FLOW_ITEM_TYPE_ETH:
3315 eth = (struct rte_ether_hdr *)&buf[temp_size];
3317 case RTE_FLOW_ITEM_TYPE_VLAN:
3318 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
3320 return rte_flow_error_set(error, EINVAL,
3321 RTE_FLOW_ERROR_TYPE_ACTION,
3322 (void *)items->type,
3323 "eth header not found");
3324 if (!eth->ether_type)
3325 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
3327 case RTE_FLOW_ITEM_TYPE_IPV4:
3328 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
3330 return rte_flow_error_set(error, EINVAL,
3331 RTE_FLOW_ERROR_TYPE_ACTION,
3332 (void *)items->type,
3333 "neither eth nor vlan"
3335 if (vlan && !vlan->eth_proto)
3336 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3337 else if (eth && !eth->ether_type)
3338 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3339 if (!ipv4->version_ihl)
3340 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
3341 MLX5_ENCAP_IPV4_IHL_MIN;
3342 if (!ipv4->time_to_live)
3343 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
3345 case RTE_FLOW_ITEM_TYPE_IPV6:
3346 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
3348 return rte_flow_error_set(error, EINVAL,
3349 RTE_FLOW_ERROR_TYPE_ACTION,
3350 (void *)items->type,
3351 "neither eth nor vlan"
3353 if (vlan && !vlan->eth_proto)
3354 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3355 else if (eth && !eth->ether_type)
3356 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3357 if (!ipv6->vtc_flow)
3359 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
3360 if (!ipv6->hop_limits)
3361 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
3363 case RTE_FLOW_ITEM_TYPE_UDP:
3364 udp = (struct rte_udp_hdr *)&buf[temp_size];
3366 return rte_flow_error_set(error, EINVAL,
3367 RTE_FLOW_ERROR_TYPE_ACTION,
3368 (void *)items->type,
3369 "ip header not found");
3370 if (ipv4 && !ipv4->next_proto_id)
3371 ipv4->next_proto_id = IPPROTO_UDP;
3372 else if (ipv6 && !ipv6->proto)
3373 ipv6->proto = IPPROTO_UDP;
3375 case RTE_FLOW_ITEM_TYPE_VXLAN:
3376 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
3378 return rte_flow_error_set(error, EINVAL,
3379 RTE_FLOW_ERROR_TYPE_ACTION,
3380 (void *)items->type,
3381 "udp header not found");
3383 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
3384 if (!vxlan->vx_flags)
3386 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
3388 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3389 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
3391 return rte_flow_error_set(error, EINVAL,
3392 RTE_FLOW_ERROR_TYPE_ACTION,
3393 (void *)items->type,
3394 "udp header not found");
3395 if (!vxlan_gpe->proto)
3396 return rte_flow_error_set(error, EINVAL,
3397 RTE_FLOW_ERROR_TYPE_ACTION,
3398 (void *)items->type,
3399 "next protocol not found");
3402 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
3403 if (!vxlan_gpe->vx_flags)
3404 vxlan_gpe->vx_flags =
3405 MLX5_ENCAP_VXLAN_GPE_FLAGS;
3407 case RTE_FLOW_ITEM_TYPE_GRE:
3408 case RTE_FLOW_ITEM_TYPE_NVGRE:
3409 gre = (struct rte_gre_hdr *)&buf[temp_size];
3411 return rte_flow_error_set(error, EINVAL,
3412 RTE_FLOW_ERROR_TYPE_ACTION,
3413 (void *)items->type,
3414 "next protocol not found");
3416 return rte_flow_error_set(error, EINVAL,
3417 RTE_FLOW_ERROR_TYPE_ACTION,
3418 (void *)items->type,
3419 "ip header not found");
3420 if (ipv4 && !ipv4->next_proto_id)
3421 ipv4->next_proto_id = IPPROTO_GRE;
3422 else if (ipv6 && !ipv6->proto)
3423 ipv6->proto = IPPROTO_GRE;
3425 case RTE_FLOW_ITEM_TYPE_VOID:
3428 return rte_flow_error_set(error, EINVAL,
3429 RTE_FLOW_ERROR_TYPE_ACTION,
3430 (void *)items->type,
3431 "unsupported item type");
3441 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
3443 struct rte_ether_hdr *eth = NULL;
3444 struct rte_vlan_hdr *vlan = NULL;
3445 struct rte_ipv6_hdr *ipv6 = NULL;
3446 struct rte_udp_hdr *udp = NULL;
3450 eth = (struct rte_ether_hdr *)data;
3451 next_hdr = (char *)(eth + 1);
3452 proto = RTE_BE16(eth->ether_type);
3455 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
3456 vlan = (struct rte_vlan_hdr *)next_hdr;
3457 proto = RTE_BE16(vlan->eth_proto);
3458 next_hdr += sizeof(struct rte_vlan_hdr);
3461 /* HW calculates IPv4 csum. no need to proceed */
3462 if (proto == RTE_ETHER_TYPE_IPV4)
3465 /* non IPv4/IPv6 header. not supported */
3466 if (proto != RTE_ETHER_TYPE_IPV6) {
3467 return rte_flow_error_set(error, ENOTSUP,
3468 RTE_FLOW_ERROR_TYPE_ACTION,
3469 NULL, "Cannot offload non IPv4/IPv6");
3472 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3474 /* ignore non UDP */
3475 if (ipv6->proto != IPPROTO_UDP)
3478 udp = (struct rte_udp_hdr *)(ipv6 + 1);
3479 udp->dgram_cksum = 0;
3485 * Convert L2 encap action to DV specification.
3488 * Pointer to rte_eth_dev structure.
3490 * Pointer to action structure.
3491 * @param[in, out] dev_flow
3492 * Pointer to the mlx5_flow.
3493 * @param[in] transfer
3494 * Mark if the flow is E-Switch flow.
3496 * Pointer to the error structure.
3499 * 0 on success, a negative errno value otherwise and rte_errno is set.
3502 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3503 const struct rte_flow_action *action,
3504 struct mlx5_flow *dev_flow,
3506 struct rte_flow_error *error)
3508 const struct rte_flow_item *encap_data;
3509 const struct rte_flow_action_raw_encap *raw_encap_data;
3510 struct mlx5_flow_dv_encap_decap_resource res = {
3512 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3513 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3514 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3517 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3519 (const struct rte_flow_action_raw_encap *)action->conf;
3520 res.size = raw_encap_data->size;
3521 memcpy(res.buf, raw_encap_data->data, res.size);
3523 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3525 ((const struct rte_flow_action_vxlan_encap *)
3526 action->conf)->definition;
3529 ((const struct rte_flow_action_nvgre_encap *)
3530 action->conf)->definition;
3531 if (flow_dv_convert_encap_data(encap_data, res.buf,
3535 if (flow_dv_zero_encap_udp_csum(res.buf, error))
3537 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3538 return rte_flow_error_set(error, EINVAL,
3539 RTE_FLOW_ERROR_TYPE_ACTION,
3540 NULL, "can't create L2 encap action");
3545 * Convert L2 decap action to DV specification.
3548 * Pointer to rte_eth_dev structure.
3549 * @param[in, out] dev_flow
3550 * Pointer to the mlx5_flow.
3551 * @param[in] transfer
3552 * Mark if the flow is E-Switch flow.
3554 * Pointer to the error structure.
3557 * 0 on success, a negative errno value otherwise and rte_errno is set.
3560 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3561 struct mlx5_flow *dev_flow,
3563 struct rte_flow_error *error)
3565 struct mlx5_flow_dv_encap_decap_resource res = {
3568 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3569 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3570 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3573 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3574 return rte_flow_error_set(error, EINVAL,
3575 RTE_FLOW_ERROR_TYPE_ACTION,
3576 NULL, "can't create L2 decap action");
3581 * Convert raw decap/encap (L3 tunnel) action to DV specification.
3584 * Pointer to rte_eth_dev structure.
3586 * Pointer to action structure.
3587 * @param[in, out] dev_flow
3588 * Pointer to the mlx5_flow.
3590 * Pointer to the flow attributes.
3592 * Pointer to the error structure.
3595 * 0 on success, a negative errno value otherwise and rte_errno is set.
3598 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3599 const struct rte_flow_action *action,
3600 struct mlx5_flow *dev_flow,
3601 const struct rte_flow_attr *attr,
3602 struct rte_flow_error *error)
3604 const struct rte_flow_action_raw_encap *encap_data;
3605 struct mlx5_flow_dv_encap_decap_resource res;
3607 memset(&res, 0, sizeof(res));
3608 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3609 res.size = encap_data->size;
3610 memcpy(res.buf, encap_data->data, res.size);
3611 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3612 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3613 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3615 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3617 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3618 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3619 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3620 return rte_flow_error_set(error, EINVAL,
3621 RTE_FLOW_ERROR_TYPE_ACTION,
3622 NULL, "can't create encap action");
3627 * Create action push VLAN.
3630 * Pointer to rte_eth_dev structure.
3632 * Pointer to the flow attributes.
3634 * Pointer to the vlan to push to the Ethernet header.
3635 * @param[in, out] dev_flow
3636 * Pointer to the mlx5_flow.
3638 * Pointer to the error structure.
3641 * 0 on success, a negative errno value otherwise and rte_errno is set.
3644 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3645 const struct rte_flow_attr *attr,
3646 const struct rte_vlan_hdr *vlan,
3647 struct mlx5_flow *dev_flow,
3648 struct rte_flow_error *error)
3650 struct mlx5_flow_dv_push_vlan_action_resource res;
3652 memset(&res, 0, sizeof(res));
3654 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3657 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3659 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3660 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3661 return flow_dv_push_vlan_action_resource_register
3662 (dev, &res, dev_flow, error);
3665 static int fdb_mirror;
3668 * Validate the modify-header actions.
3670 * @param[in] action_flags
3671 * Holds the actions detected until now.
3673 * Pointer to the modify action.
3675 * Pointer to error structure.
3678 * 0 on success, a negative errno value otherwise and rte_errno is set.
3681 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3682 const struct rte_flow_action *action,
3683 struct rte_flow_error *error)
3685 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3686 return rte_flow_error_set(error, EINVAL,
3687 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3688 NULL, "action configuration not set");
3689 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3690 return rte_flow_error_set(error, EINVAL,
3691 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3692 "can't have encap action before"
3694 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3695 return rte_flow_error_set(error, EINVAL,
3696 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3697 "can't support sample action before"
3698 " modify action for E-Switch"
3704 * Validate the modify-header MAC address actions.
3706 * @param[in] action_flags
3707 * Holds the actions detected until now.
3709 * Pointer to the modify action.
3710 * @param[in] item_flags
3711 * Holds the items detected.
3713 * Pointer to error structure.
3716 * 0 on success, a negative errno value otherwise and rte_errno is set.
3719 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3720 const struct rte_flow_action *action,
3721 const uint64_t item_flags,
3722 struct rte_flow_error *error)
3726 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3728 if (!(item_flags & MLX5_FLOW_LAYER_L2))
3729 return rte_flow_error_set(error, EINVAL,
3730 RTE_FLOW_ERROR_TYPE_ACTION,
3732 "no L2 item in pattern");
3738 * Validate the modify-header IPv4 address actions.
3740 * @param[in] action_flags
3741 * Holds the actions detected until now.
3743 * Pointer to the modify action.
3744 * @param[in] item_flags
3745 * Holds the items detected.
3747 * Pointer to error structure.
3750 * 0 on success, a negative errno value otherwise and rte_errno is set.
3753 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3754 const struct rte_flow_action *action,
3755 const uint64_t item_flags,
3756 struct rte_flow_error *error)
3761 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3763 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3764 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3765 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3766 if (!(item_flags & layer))
3767 return rte_flow_error_set(error, EINVAL,
3768 RTE_FLOW_ERROR_TYPE_ACTION,
3770 "no ipv4 item in pattern");
3776 * Validate the modify-header IPv6 address actions.
3778 * @param[in] action_flags
3779 * Holds the actions detected until now.
3781 * Pointer to the modify action.
3782 * @param[in] item_flags
3783 * Holds the items detected.
3785 * Pointer to error structure.
3788 * 0 on success, a negative errno value otherwise and rte_errno is set.
3791 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3792 const struct rte_flow_action *action,
3793 const uint64_t item_flags,
3794 struct rte_flow_error *error)
3799 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3801 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3802 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3803 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3804 if (!(item_flags & layer))
3805 return rte_flow_error_set(error, EINVAL,
3806 RTE_FLOW_ERROR_TYPE_ACTION,
3808 "no ipv6 item in pattern");
3814 * Validate the modify-header TP actions.
3816 * @param[in] action_flags
3817 * Holds the actions detected until now.
3819 * Pointer to the modify action.
3820 * @param[in] item_flags
3821 * Holds the items detected.
3823 * Pointer to error structure.
3826 * 0 on success, a negative errno value otherwise and rte_errno is set.
3829 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3830 const struct rte_flow_action *action,
3831 const uint64_t item_flags,
3832 struct rte_flow_error *error)
3837 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3839 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3840 MLX5_FLOW_LAYER_INNER_L4 :
3841 MLX5_FLOW_LAYER_OUTER_L4;
3842 if (!(item_flags & layer))
3843 return rte_flow_error_set(error, EINVAL,
3844 RTE_FLOW_ERROR_TYPE_ACTION,
3845 NULL, "no transport layer "
3852 * Validate the modify-header actions of increment/decrement
3853 * TCP Sequence-number.
3855 * @param[in] action_flags
3856 * Holds the actions detected until now.
3858 * Pointer to the modify action.
3859 * @param[in] item_flags
3860 * Holds the items detected.
3862 * Pointer to error structure.
3865 * 0 on success, a negative errno value otherwise and rte_errno is set.
3868 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3869 const struct rte_flow_action *action,
3870 const uint64_t item_flags,
3871 struct rte_flow_error *error)
3876 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3878 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3879 MLX5_FLOW_LAYER_INNER_L4_TCP :
3880 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3881 if (!(item_flags & layer))
3882 return rte_flow_error_set(error, EINVAL,
3883 RTE_FLOW_ERROR_TYPE_ACTION,
3884 NULL, "no TCP item in"
3886 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3887 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3888 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3889 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3890 return rte_flow_error_set(error, EINVAL,
3891 RTE_FLOW_ERROR_TYPE_ACTION,
3893 "cannot decrease and increase"
3894 " TCP sequence number"
3895 " at the same time");
3901 * Validate the modify-header actions of increment/decrement
3902 * TCP Acknowledgment number.
3904 * @param[in] action_flags
3905 * Holds the actions detected until now.
3907 * Pointer to the modify action.
3908 * @param[in] item_flags
3909 * Holds the items detected.
3911 * Pointer to error structure.
3914 * 0 on success, a negative errno value otherwise and rte_errno is set.
3917 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3918 const struct rte_flow_action *action,
3919 const uint64_t item_flags,
3920 struct rte_flow_error *error)
3925 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3927 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3928 MLX5_FLOW_LAYER_INNER_L4_TCP :
3929 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3930 if (!(item_flags & layer))
3931 return rte_flow_error_set(error, EINVAL,
3932 RTE_FLOW_ERROR_TYPE_ACTION,
3933 NULL, "no TCP item in"
3935 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3936 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3937 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3938 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3939 return rte_flow_error_set(error, EINVAL,
3940 RTE_FLOW_ERROR_TYPE_ACTION,
3942 "cannot decrease and increase"
3943 " TCP acknowledgment number"
3944 " at the same time");
3950 * Validate the modify-header TTL actions.
3952 * @param[in] action_flags
3953 * Holds the actions detected until now.
3955 * Pointer to the modify action.
3956 * @param[in] item_flags
3957 * Holds the items detected.
3959 * Pointer to error structure.
3962 * 0 on success, a negative errno value otherwise and rte_errno is set.
3965 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3966 const struct rte_flow_action *action,
3967 const uint64_t item_flags,
3968 struct rte_flow_error *error)
3973 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3975 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3976 MLX5_FLOW_LAYER_INNER_L3 :
3977 MLX5_FLOW_LAYER_OUTER_L3;
3978 if (!(item_flags & layer))
3979 return rte_flow_error_set(error, EINVAL,
3980 RTE_FLOW_ERROR_TYPE_ACTION,
3982 "no IP protocol in pattern");
3988 * Validate jump action.
3991 * Pointer to the jump action.
3992 * @param[in] action_flags
3993 * Holds the actions detected until now.
3994 * @param[in] attributes
3995 * Pointer to flow attributes
3996 * @param[in] external
3997 * Action belongs to flow rule created by request external to PMD.
3999 * Pointer to error structure.
4002 * 0 on success, a negative errno value otherwise and rte_errno is set.
4005 flow_dv_validate_action_jump(struct rte_eth_dev *dev,
4006 const struct mlx5_flow_tunnel *tunnel,
4007 const struct rte_flow_action *action,
4008 uint64_t action_flags,
4009 const struct rte_flow_attr *attributes,
4010 bool external, struct rte_flow_error *error)
4012 uint32_t target_group, table;
4014 struct flow_grp_info grp_info = {
4015 .external = !!external,
4016 .transfer = !!attributes->transfer,
4020 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
4021 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
4022 return rte_flow_error_set(error, EINVAL,
4023 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4024 "can't have 2 fate actions in"
4026 if (action_flags & MLX5_FLOW_ACTION_METER)
4027 return rte_flow_error_set(error, ENOTSUP,
4028 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4029 "jump with meter not support");
4030 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
4031 return rte_flow_error_set(error, EINVAL,
4032 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4033 "E-Switch mirroring can't support"
4034 " Sample action and jump action in"
4037 return rte_flow_error_set(error, EINVAL,
4038 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4039 NULL, "action configuration not set");
4041 ((const struct rte_flow_action_jump *)action->conf)->group;
4042 ret = mlx5_flow_group_to_table(dev, tunnel, target_group, &table,
4046 if (attributes->group == target_group &&
4047 !(action_flags & (MLX5_FLOW_ACTION_TUNNEL_SET |
4048 MLX5_FLOW_ACTION_TUNNEL_MATCH)))
4049 return rte_flow_error_set(error, EINVAL,
4050 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4051 "target group must be other than"
4052 " the current flow group");
4057 * Validate the port_id action.
4060 * Pointer to rte_eth_dev structure.
4061 * @param[in] action_flags
4062 * Bit-fields that holds the actions detected until now.
4064 * Port_id RTE action structure.
4066 * Attributes of flow that includes this action.
4068 * Pointer to error structure.
4071 * 0 on success, a negative errno value otherwise and rte_errno is set.
4074 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
4075 uint64_t action_flags,
4076 const struct rte_flow_action *action,
4077 const struct rte_flow_attr *attr,
4078 struct rte_flow_error *error)
4080 const struct rte_flow_action_port_id *port_id;
4081 struct mlx5_priv *act_priv;
4082 struct mlx5_priv *dev_priv;
4085 if (!attr->transfer)
4086 return rte_flow_error_set(error, ENOTSUP,
4087 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4089 "port id action is valid in transfer"
4091 if (!action || !action->conf)
4092 return rte_flow_error_set(error, ENOTSUP,
4093 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4095 "port id action parameters must be"
4097 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
4098 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
4099 return rte_flow_error_set(error, EINVAL,
4100 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4101 "can have only one fate actions in"
4103 dev_priv = mlx5_dev_to_eswitch_info(dev);
4105 return rte_flow_error_set(error, rte_errno,
4106 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4108 "failed to obtain E-Switch info");
4109 port_id = action->conf;
4110 port = port_id->original ? dev->data->port_id : port_id->id;
4111 act_priv = mlx5_port_to_eswitch_info(port, false);
4113 return rte_flow_error_set
4115 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
4116 "failed to obtain E-Switch port id for port");
4117 if (act_priv->domain_id != dev_priv->domain_id)
4118 return rte_flow_error_set
4120 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4121 "port does not belong to"
4122 " E-Switch being configured");
4127 * Get the maximum number of modify header actions.
4130 * Pointer to rte_eth_dev structure.
4132 * Flags bits to check if root level.
4135 * Max number of modify header actions device can support.
4137 static inline unsigned int
4138 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
4142 * There's no way to directly query the max capacity from FW.
4143 * The maximal value on root table should be assumed to be supported.
4145 if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
4146 return MLX5_MAX_MODIFY_NUM;
4148 return MLX5_ROOT_TBL_MODIFY_NUM;
4152 * Validate the meter action.
4155 * Pointer to rte_eth_dev structure.
4156 * @param[in] action_flags
4157 * Bit-fields that holds the actions detected until now.
4159 * Pointer to the meter action.
4161 * Attributes of flow that includes this action.
4163 * Pointer to error structure.
4166 * 0 on success, a negative errno value otherwise and rte_ernno is set.
4169 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
4170 uint64_t action_flags,
4171 const struct rte_flow_action *action,
4172 const struct rte_flow_attr *attr,
4173 struct rte_flow_error *error)
4175 struct mlx5_priv *priv = dev->data->dev_private;
4176 const struct rte_flow_action_meter *am = action->conf;
4177 struct mlx5_flow_meter *fm;
4180 return rte_flow_error_set(error, EINVAL,
4181 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4182 "meter action conf is NULL");
4184 if (action_flags & MLX5_FLOW_ACTION_METER)
4185 return rte_flow_error_set(error, ENOTSUP,
4186 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4187 "meter chaining not support");
4188 if (action_flags & MLX5_FLOW_ACTION_JUMP)
4189 return rte_flow_error_set(error, ENOTSUP,
4190 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4191 "meter with jump not support");
4193 return rte_flow_error_set(error, ENOTSUP,
4194 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4196 "meter action not supported");
4197 fm = mlx5_flow_meter_find(priv, am->mtr_id);
4199 return rte_flow_error_set(error, EINVAL,
4200 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4202 if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
4203 (!fm->ingress && !attr->ingress && attr->egress) ||
4204 (!fm->egress && !attr->egress && attr->ingress))))
4205 return rte_flow_error_set(error, EINVAL,
4206 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4207 "Flow attributes are either invalid "
4208 "or have a conflict with current "
4209 "meter attributes");
4214 * Validate the age action.
4216 * @param[in] action_flags
4217 * Holds the actions detected until now.
4219 * Pointer to the age action.
4221 * Pointer to the Ethernet device structure.
4223 * Pointer to error structure.
4226 * 0 on success, a negative errno value otherwise and rte_errno is set.
4229 flow_dv_validate_action_age(uint64_t action_flags,
4230 const struct rte_flow_action *action,
4231 struct rte_eth_dev *dev,
4232 struct rte_flow_error *error)
4234 struct mlx5_priv *priv = dev->data->dev_private;
4235 const struct rte_flow_action_age *age = action->conf;
4237 if (!priv->config.devx || (priv->sh->cmng.counter_fallback &&
4238 !priv->sh->aso_age_mng))
4239 return rte_flow_error_set(error, ENOTSUP,
4240 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4242 "age action not supported");
4243 if (!(action->conf))
4244 return rte_flow_error_set(error, EINVAL,
4245 RTE_FLOW_ERROR_TYPE_ACTION, action,
4246 "configuration cannot be null");
4247 if (!(age->timeout))
4248 return rte_flow_error_set(error, EINVAL,
4249 RTE_FLOW_ERROR_TYPE_ACTION, action,
4250 "invalid timeout value 0");
4251 if (action_flags & MLX5_FLOW_ACTION_AGE)
4252 return rte_flow_error_set(error, EINVAL,
4253 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4254 "duplicate age actions set");
4259 * Validate the modify-header IPv4 DSCP actions.
4261 * @param[in] action_flags
4262 * Holds the actions detected until now.
4264 * Pointer to the modify action.
4265 * @param[in] item_flags
4266 * Holds the items detected.
4268 * Pointer to error structure.
4271 * 0 on success, a negative errno value otherwise and rte_errno is set.
4274 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
4275 const struct rte_flow_action *action,
4276 const uint64_t item_flags,
4277 struct rte_flow_error *error)
4281 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4283 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
4284 return rte_flow_error_set(error, EINVAL,
4285 RTE_FLOW_ERROR_TYPE_ACTION,
4287 "no ipv4 item in pattern");
4293 * Validate the modify-header IPv6 DSCP actions.
4295 * @param[in] action_flags
4296 * Holds the actions detected until now.
4298 * Pointer to the modify action.
4299 * @param[in] item_flags
4300 * Holds the items detected.
4302 * Pointer to error structure.
4305 * 0 on success, a negative errno value otherwise and rte_errno is set.
4308 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
4309 const struct rte_flow_action *action,
4310 const uint64_t item_flags,
4311 struct rte_flow_error *error)
4315 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4317 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
4318 return rte_flow_error_set(error, EINVAL,
4319 RTE_FLOW_ERROR_TYPE_ACTION,
4321 "no ipv6 item in pattern");
4327 * Match modify-header resource.
4330 * Pointer to the hash list.
4332 * Pointer to exist resource entry object.
4334 * Key of the new entry.
4336 * Pointer to new modify-header resource.
4339 * 0 on matching, non-zero otherwise.
4342 flow_dv_modify_match_cb(struct mlx5_hlist *list __rte_unused,
4343 struct mlx5_hlist_entry *entry,
4344 uint64_t key __rte_unused, void *cb_ctx)
4346 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4347 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
4348 struct mlx5_flow_dv_modify_hdr_resource *resource =
4349 container_of(entry, typeof(*resource), entry);
4350 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
4352 key_len += ref->actions_num * sizeof(ref->actions[0]);
4353 return ref->actions_num != resource->actions_num ||
4354 memcmp(&ref->ft_type, &resource->ft_type, key_len);
4357 struct mlx5_hlist_entry *
4358 flow_dv_modify_create_cb(struct mlx5_hlist *list, uint64_t key __rte_unused,
4361 struct mlx5_dev_ctx_shared *sh = list->ctx;
4362 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4363 struct mlx5dv_dr_domain *ns;
4364 struct mlx5_flow_dv_modify_hdr_resource *entry;
4365 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
4367 uint32_t data_len = ref->actions_num * sizeof(ref->actions[0]);
4368 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
4370 entry = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*entry) + data_len, 0,
4373 rte_flow_error_set(ctx->error, ENOMEM,
4374 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4375 "cannot allocate resource memory");
4378 rte_memcpy(&entry->ft_type,
4379 RTE_PTR_ADD(ref, offsetof(typeof(*ref), ft_type)),
4380 key_len + data_len);
4381 if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
4382 ns = sh->fdb_domain;
4383 else if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
4387 ret = mlx5_flow_os_create_flow_action_modify_header
4388 (sh->ctx, ns, entry,
4389 data_len, &entry->action);
4392 rte_flow_error_set(ctx->error, ENOMEM,
4393 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4394 NULL, "cannot create modification action");
4397 return &entry->entry;
4401 * Validate the sample action.
4403 * @param[in] action_flags
4404 * Holds the actions detected until now.
4406 * Pointer to the sample action.
4408 * Pointer to the Ethernet device structure.
4410 * Attributes of flow that includes this action.
4411 * @param[in] item_flags
4412 * Holds the items detected.
4414 * Pointer to error structure.
4417 * 0 on success, a negative errno value otherwise and rte_errno is set.
4420 flow_dv_validate_action_sample(uint64_t action_flags,
4421 const struct rte_flow_action *action,
4422 struct rte_eth_dev *dev,
4423 const struct rte_flow_attr *attr,
4424 const uint64_t item_flags,
4425 struct rte_flow_error *error)
4427 struct mlx5_priv *priv = dev->data->dev_private;
4428 struct mlx5_dev_config *dev_conf = &priv->config;
4429 const struct rte_flow_action_sample *sample = action->conf;
4430 const struct rte_flow_action *act;
4431 uint64_t sub_action_flags = 0;
4432 uint16_t queue_index = 0xFFFF;
4438 return rte_flow_error_set(error, EINVAL,
4439 RTE_FLOW_ERROR_TYPE_ACTION, action,
4440 "configuration cannot be NULL");
4441 if (sample->ratio == 0)
4442 return rte_flow_error_set(error, EINVAL,
4443 RTE_FLOW_ERROR_TYPE_ACTION, action,
4444 "ratio value starts from 1");
4445 if (!priv->config.devx || (sample->ratio > 0 && !priv->sampler_en))
4446 return rte_flow_error_set(error, ENOTSUP,
4447 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4449 "sample action not supported");
4450 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
4451 return rte_flow_error_set(error, EINVAL,
4452 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4453 "Multiple sample actions not "
4455 if (action_flags & MLX5_FLOW_ACTION_METER)
4456 return rte_flow_error_set(error, EINVAL,
4457 RTE_FLOW_ERROR_TYPE_ACTION, action,
4458 "wrong action order, meter should "
4459 "be after sample action");
4460 if (action_flags & MLX5_FLOW_ACTION_JUMP)
4461 return rte_flow_error_set(error, EINVAL,
4462 RTE_FLOW_ERROR_TYPE_ACTION, action,
4463 "wrong action order, jump should "
4464 "be after sample action");
4465 act = sample->actions;
4466 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
4467 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
4468 return rte_flow_error_set(error, ENOTSUP,
4469 RTE_FLOW_ERROR_TYPE_ACTION,
4470 act, "too many actions");
4471 switch (act->type) {
4472 case RTE_FLOW_ACTION_TYPE_QUEUE:
4473 ret = mlx5_flow_validate_action_queue(act,
4479 queue_index = ((const struct rte_flow_action_queue *)
4480 (act->conf))->index;
4481 sub_action_flags |= MLX5_FLOW_ACTION_QUEUE;
4484 case RTE_FLOW_ACTION_TYPE_MARK:
4485 ret = flow_dv_validate_action_mark(dev, act,
4490 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4491 sub_action_flags |= MLX5_FLOW_ACTION_MARK |
4492 MLX5_FLOW_ACTION_MARK_EXT;
4494 sub_action_flags |= MLX5_FLOW_ACTION_MARK;
4497 case RTE_FLOW_ACTION_TYPE_COUNT:
4498 ret = flow_dv_validate_action_count(dev, error);
4501 sub_action_flags |= MLX5_FLOW_ACTION_COUNT;
4504 case RTE_FLOW_ACTION_TYPE_PORT_ID:
4505 ret = flow_dv_validate_action_port_id(dev,
4512 sub_action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4515 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4516 ret = flow_dv_validate_action_raw_encap_decap
4517 (dev, NULL, act->conf, attr, &sub_action_flags,
4518 &actions_n, action, item_flags, error);
4524 return rte_flow_error_set(error, ENOTSUP,
4525 RTE_FLOW_ERROR_TYPE_ACTION,
4527 "Doesn't support optional "
4531 if (attr->ingress && !attr->transfer) {
4532 if (!(sub_action_flags & MLX5_FLOW_ACTION_QUEUE))
4533 return rte_flow_error_set(error, EINVAL,
4534 RTE_FLOW_ERROR_TYPE_ACTION,
4536 "Ingress must has a dest "
4537 "QUEUE for Sample");
4538 } else if (attr->egress && !attr->transfer) {
4539 return rte_flow_error_set(error, ENOTSUP,
4540 RTE_FLOW_ERROR_TYPE_ACTION,
4542 "Sample Only support Ingress "
4544 } else if (sample->actions->type != RTE_FLOW_ACTION_TYPE_END) {
4545 MLX5_ASSERT(attr->transfer);
4546 if (sample->ratio > 1)
4547 return rte_flow_error_set(error, ENOTSUP,
4548 RTE_FLOW_ERROR_TYPE_ACTION,
4550 "E-Switch doesn't support "
4551 "any optional action "
4554 if (sub_action_flags & MLX5_FLOW_ACTION_QUEUE)
4555 return rte_flow_error_set(error, ENOTSUP,
4556 RTE_FLOW_ERROR_TYPE_ACTION,
4558 "unsupported action QUEUE");
4559 if (!(sub_action_flags & MLX5_FLOW_ACTION_PORT_ID))
4560 return rte_flow_error_set(error, EINVAL,
4561 RTE_FLOW_ERROR_TYPE_ACTION,
4563 "E-Switch must has a dest "
4564 "port for mirroring");
4566 /* Continue validation for Xcap actions.*/
4567 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) &&
4568 (queue_index == 0xFFFF ||
4569 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
4570 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
4571 MLX5_FLOW_XCAP_ACTIONS)
4572 return rte_flow_error_set(error, ENOTSUP,
4573 RTE_FLOW_ERROR_TYPE_ACTION,
4574 NULL, "encap and decap "
4575 "combination aren't "
4577 if (!attr->transfer && attr->ingress && (sub_action_flags &
4578 MLX5_FLOW_ACTION_ENCAP))
4579 return rte_flow_error_set(error, ENOTSUP,
4580 RTE_FLOW_ERROR_TYPE_ACTION,
4581 NULL, "encap is not supported"
4582 " for ingress traffic");
4588 * Find existing modify-header resource or create and register a new one.
4590 * @param dev[in, out]
4591 * Pointer to rte_eth_dev structure.
4592 * @param[in, out] resource
4593 * Pointer to modify-header resource.
4594 * @parm[in, out] dev_flow
4595 * Pointer to the dev_flow.
4597 * pointer to error structure.
4600 * 0 on success otherwise -errno and errno is set.
4603 flow_dv_modify_hdr_resource_register
4604 (struct rte_eth_dev *dev,
4605 struct mlx5_flow_dv_modify_hdr_resource *resource,
4606 struct mlx5_flow *dev_flow,
4607 struct rte_flow_error *error)
4609 struct mlx5_priv *priv = dev->data->dev_private;
4610 struct mlx5_dev_ctx_shared *sh = priv->sh;
4611 uint32_t key_len = sizeof(*resource) -
4612 offsetof(typeof(*resource), ft_type) +
4613 resource->actions_num * sizeof(resource->actions[0]);
4614 struct mlx5_hlist_entry *entry;
4615 struct mlx5_flow_cb_ctx ctx = {
4621 resource->flags = dev_flow->dv.group ? 0 :
4622 MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4623 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
4625 return rte_flow_error_set(error, EOVERFLOW,
4626 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4627 "too many modify header items");
4628 key64 = __rte_raw_cksum(&resource->ft_type, key_len, 0);
4629 entry = mlx5_hlist_register(sh->modify_cmds, key64, &ctx);
4632 resource = container_of(entry, typeof(*resource), entry);
4633 dev_flow->handle->dvh.modify_hdr = resource;
4638 * Get DV flow counter by index.
4641 * Pointer to the Ethernet device structure.
4643 * mlx5 flow counter index in the container.
4645 * mlx5 flow counter pool in the container,
4648 * Pointer to the counter, NULL otherwise.
4650 static struct mlx5_flow_counter *
4651 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
4653 struct mlx5_flow_counter_pool **ppool)
4655 struct mlx5_priv *priv = dev->data->dev_private;
4656 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4657 struct mlx5_flow_counter_pool *pool;
4659 /* Decrease to original index and clear shared bit. */
4660 idx = (idx - 1) & (MLX5_CNT_SHARED_OFFSET - 1);
4661 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cmng->n);
4662 pool = cmng->pools[idx / MLX5_COUNTERS_PER_POOL];
4666 return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
4670 * Check the devx counter belongs to the pool.
4673 * Pointer to the counter pool.
4675 * The counter devx ID.
4678 * True if counter belongs to the pool, false otherwise.
4681 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
4683 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
4684 MLX5_COUNTERS_PER_POOL;
4686 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
4692 * Get a pool by devx counter ID.
4695 * Pointer to the counter management.
4697 * The counter devx ID.
4700 * The counter pool pointer if exists, NULL otherwise,
4702 static struct mlx5_flow_counter_pool *
4703 flow_dv_find_pool_by_id(struct mlx5_flow_counter_mng *cmng, int id)
4706 struct mlx5_flow_counter_pool *pool = NULL;
4708 rte_spinlock_lock(&cmng->pool_update_sl);
4709 /* Check last used pool. */
4710 if (cmng->last_pool_idx != POOL_IDX_INVALID &&
4711 flow_dv_is_counter_in_pool(cmng->pools[cmng->last_pool_idx], id)) {
4712 pool = cmng->pools[cmng->last_pool_idx];
4715 /* ID out of range means no suitable pool in the container. */
4716 if (id > cmng->max_id || id < cmng->min_id)
4719 * Find the pool from the end of the container, since mostly counter
4720 * ID is sequence increasing, and the last pool should be the needed
4725 struct mlx5_flow_counter_pool *pool_tmp = cmng->pools[i];
4727 if (flow_dv_is_counter_in_pool(pool_tmp, id)) {
4733 rte_spinlock_unlock(&cmng->pool_update_sl);
4738 * Resize a counter container.
4741 * Pointer to the Ethernet device structure.
4744 * 0 on success, otherwise negative errno value and rte_errno is set.
4747 flow_dv_container_resize(struct rte_eth_dev *dev)
4749 struct mlx5_priv *priv = dev->data->dev_private;
4750 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4751 void *old_pools = cmng->pools;
4752 uint32_t resize = cmng->n + MLX5_CNT_CONTAINER_RESIZE;
4753 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4754 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
4761 memcpy(pools, old_pools, cmng->n *
4762 sizeof(struct mlx5_flow_counter_pool *));
4764 cmng->pools = pools;
4766 mlx5_free(old_pools);
4771 * Query a devx flow counter.
4774 * Pointer to the Ethernet device structure.
4776 * Index to the flow counter.
4778 * The statistics value of packets.
4780 * The statistics value of bytes.
4783 * 0 on success, otherwise a negative errno value and rte_errno is set.
4786 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4789 struct mlx5_priv *priv = dev->data->dev_private;
4790 struct mlx5_flow_counter_pool *pool = NULL;
4791 struct mlx5_flow_counter *cnt;
4794 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4796 if (priv->sh->cmng.counter_fallback)
4797 return mlx5_devx_cmd_flow_counter_query(cnt->dcs_when_active, 0,
4798 0, pkts, bytes, 0, NULL, NULL, 0);
4799 rte_spinlock_lock(&pool->sl);
4804 offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
4805 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4806 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4808 rte_spinlock_unlock(&pool->sl);
4813 * Create and initialize a new counter pool.
4816 * Pointer to the Ethernet device structure.
4818 * The devX counter handle.
4820 * Whether the pool is for counter that was allocated for aging.
4821 * @param[in/out] cont_cur
4822 * Pointer to the container pointer, it will be update in pool resize.
4825 * The pool container pointer on success, NULL otherwise and rte_errno is set.
4827 static struct mlx5_flow_counter_pool *
4828 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4831 struct mlx5_priv *priv = dev->data->dev_private;
4832 struct mlx5_flow_counter_pool *pool;
4833 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4834 bool fallback = priv->sh->cmng.counter_fallback;
4835 uint32_t size = sizeof(*pool);
4837 size += MLX5_COUNTERS_PER_POOL * MLX5_CNT_SIZE;
4838 size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * MLX5_AGE_SIZE);
4839 pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
4845 pool->is_aged = !!age;
4846 pool->query_gen = 0;
4847 pool->min_dcs = dcs;
4848 rte_spinlock_init(&pool->sl);
4849 rte_spinlock_init(&pool->csl);
4850 TAILQ_INIT(&pool->counters[0]);
4851 TAILQ_INIT(&pool->counters[1]);
4852 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
4853 rte_spinlock_lock(&cmng->pool_update_sl);
4854 pool->index = cmng->n_valid;
4855 if (pool->index == cmng->n && flow_dv_container_resize(dev)) {
4857 rte_spinlock_unlock(&cmng->pool_update_sl);
4860 cmng->pools[pool->index] = pool;
4862 if (unlikely(fallback)) {
4863 int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
4865 if (base < cmng->min_id)
4866 cmng->min_id = base;
4867 if (base > cmng->max_id)
4868 cmng->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
4869 cmng->last_pool_idx = pool->index;
4871 rte_spinlock_unlock(&cmng->pool_update_sl);
4876 * Prepare a new counter and/or a new counter pool.
4879 * Pointer to the Ethernet device structure.
4880 * @param[out] cnt_free
4881 * Where to put the pointer of a new counter.
4883 * Whether the pool is for counter that was allocated for aging.
4886 * The counter pool pointer and @p cnt_free is set on success,
4887 * NULL otherwise and rte_errno is set.
4889 static struct mlx5_flow_counter_pool *
4890 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4891 struct mlx5_flow_counter **cnt_free,
4894 struct mlx5_priv *priv = dev->data->dev_private;
4895 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4896 struct mlx5_flow_counter_pool *pool;
4897 struct mlx5_counters tmp_tq;
4898 struct mlx5_devx_obj *dcs = NULL;
4899 struct mlx5_flow_counter *cnt;
4900 enum mlx5_counter_type cnt_type =
4901 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
4902 bool fallback = priv->sh->cmng.counter_fallback;
4906 /* bulk_bitmap must be 0 for single counter allocation. */
4907 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4910 pool = flow_dv_find_pool_by_id(cmng, dcs->id);
4912 pool = flow_dv_pool_create(dev, dcs, age);
4914 mlx5_devx_cmd_destroy(dcs);
4918 i = dcs->id % MLX5_COUNTERS_PER_POOL;
4919 cnt = MLX5_POOL_GET_CNT(pool, i);
4921 cnt->dcs_when_free = dcs;
4925 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4927 rte_errno = ENODATA;
4930 pool = flow_dv_pool_create(dev, dcs, age);
4932 mlx5_devx_cmd_destroy(dcs);
4935 TAILQ_INIT(&tmp_tq);
4936 for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
4937 cnt = MLX5_POOL_GET_CNT(pool, i);
4939 TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
4941 rte_spinlock_lock(&cmng->csl[cnt_type]);
4942 TAILQ_CONCAT(&cmng->counters[cnt_type], &tmp_tq, next);
4943 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4944 *cnt_free = MLX5_POOL_GET_CNT(pool, 0);
4945 (*cnt_free)->pool = pool;
4950 * Allocate a flow counter.
4953 * Pointer to the Ethernet device structure.
4955 * Whether the counter was allocated for aging.
4958 * Index to flow counter on success, 0 otherwise and rte_errno is set.
4961 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t age)
4963 struct mlx5_priv *priv = dev->data->dev_private;
4964 struct mlx5_flow_counter_pool *pool = NULL;
4965 struct mlx5_flow_counter *cnt_free = NULL;
4966 bool fallback = priv->sh->cmng.counter_fallback;
4967 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
4968 enum mlx5_counter_type cnt_type =
4969 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
4972 if (!priv->config.devx) {
4973 rte_errno = ENOTSUP;
4976 /* Get free counters from container. */
4977 rte_spinlock_lock(&cmng->csl[cnt_type]);
4978 cnt_free = TAILQ_FIRST(&cmng->counters[cnt_type]);
4980 TAILQ_REMOVE(&cmng->counters[cnt_type], cnt_free, next);
4981 rte_spinlock_unlock(&cmng->csl[cnt_type]);
4982 if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free, age))
4984 pool = cnt_free->pool;
4986 cnt_free->dcs_when_active = cnt_free->dcs_when_free;
4987 /* Create a DV counter action only in the first time usage. */
4988 if (!cnt_free->action) {
4990 struct mlx5_devx_obj *dcs;
4994 offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
4995 dcs = pool->min_dcs;
4998 dcs = cnt_free->dcs_when_free;
5000 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
5007 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
5008 MLX5_CNT_ARRAY_IDX(pool, cnt_free));
5009 /* Update the counter reset values. */
5010 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
5013 if (!fallback && !priv->sh->cmng.query_thread_on)
5014 /* Start the asynchronous batch query by the host thread. */
5015 mlx5_set_query_alarm(priv->sh);
5019 cnt_free->pool = pool;
5021 cnt_free->dcs_when_free = cnt_free->dcs_when_active;
5022 rte_spinlock_lock(&cmng->csl[cnt_type]);
5023 TAILQ_INSERT_TAIL(&cmng->counters[cnt_type], cnt_free, next);
5024 rte_spinlock_unlock(&cmng->csl[cnt_type]);
5030 * Allocate a shared flow counter.
5033 * Pointer to the shared counter configuration.
5035 * Pointer to save the allocated counter index.
5038 * Index to flow counter on success, 0 otherwise and rte_errno is set.
5042 flow_dv_counter_alloc_shared_cb(void *ctx, union mlx5_l3t_data *data)
5044 struct mlx5_shared_counter_conf *conf = ctx;
5045 struct rte_eth_dev *dev = conf->dev;
5046 struct mlx5_flow_counter *cnt;
5048 data->dword = flow_dv_counter_alloc(dev, 0);
5049 data->dword |= MLX5_CNT_SHARED_OFFSET;
5050 cnt = flow_dv_counter_get_by_idx(dev, data->dword, NULL);
5051 cnt->shared_info.id = conf->id;
5056 * Get a shared flow counter.
5059 * Pointer to the Ethernet device structure.
5061 * Counter identifier.
5064 * Index to flow counter on success, 0 otherwise and rte_errno is set.
5067 flow_dv_counter_get_shared(struct rte_eth_dev *dev, uint32_t id)
5069 struct mlx5_priv *priv = dev->data->dev_private;
5070 struct mlx5_shared_counter_conf conf = {
5074 union mlx5_l3t_data data = {
5078 mlx5_l3t_prepare_entry(priv->sh->cnt_id_tbl, id, &data,
5079 flow_dv_counter_alloc_shared_cb, &conf);
5084 * Get age param from counter index.
5087 * Pointer to the Ethernet device structure.
5088 * @param[in] counter
5089 * Index to the counter handler.
5092 * The aging parameter specified for the counter index.
5094 static struct mlx5_age_param*
5095 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
5098 struct mlx5_flow_counter *cnt;
5099 struct mlx5_flow_counter_pool *pool = NULL;
5101 flow_dv_counter_get_by_idx(dev, counter, &pool);
5102 counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
5103 cnt = MLX5_POOL_GET_CNT(pool, counter);
5104 return MLX5_CNT_TO_AGE(cnt);
5108 * Remove a flow counter from aged counter list.
5111 * Pointer to the Ethernet device structure.
5112 * @param[in] counter
5113 * Index to the counter handler.
5115 * Pointer to the counter handler.
5118 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
5119 uint32_t counter, struct mlx5_flow_counter *cnt)
5121 struct mlx5_age_info *age_info;
5122 struct mlx5_age_param *age_param;
5123 struct mlx5_priv *priv = dev->data->dev_private;
5124 uint16_t expected = AGE_CANDIDATE;
5126 age_info = GET_PORT_AGE_INFO(priv);
5127 age_param = flow_dv_counter_idx_get_age(dev, counter);
5128 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
5129 AGE_FREE, false, __ATOMIC_RELAXED,
5130 __ATOMIC_RELAXED)) {
5132 * We need the lock even it is age timeout,
5133 * since counter may still in process.
5135 rte_spinlock_lock(&age_info->aged_sl);
5136 TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
5137 rte_spinlock_unlock(&age_info->aged_sl);
5138 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
5143 * Release a flow counter.
5146 * Pointer to the Ethernet device structure.
5147 * @param[in] counter
5148 * Index to the counter handler.
5151 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t counter)
5153 struct mlx5_priv *priv = dev->data->dev_private;
5154 struct mlx5_flow_counter_pool *pool = NULL;
5155 struct mlx5_flow_counter *cnt;
5156 enum mlx5_counter_type cnt_type;
5160 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
5162 if (IS_SHARED_CNT(counter) &&
5163 mlx5_l3t_clear_entry(priv->sh->cnt_id_tbl, cnt->shared_info.id))
5166 flow_dv_counter_remove_from_age(dev, counter, cnt);
5169 * Put the counter back to list to be updated in none fallback mode.
5170 * Currently, we are using two list alternately, while one is in query,
5171 * add the freed counter to the other list based on the pool query_gen
5172 * value. After query finishes, add counter the list to the global
5173 * container counter list. The list changes while query starts. In
5174 * this case, lock will not be needed as query callback and release
5175 * function both operate with the different list.
5178 if (!priv->sh->cmng.counter_fallback) {
5179 rte_spinlock_lock(&pool->csl);
5180 TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
5181 rte_spinlock_unlock(&pool->csl);
5183 cnt->dcs_when_free = cnt->dcs_when_active;
5184 cnt_type = pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
5185 MLX5_COUNTER_TYPE_ORIGIN;
5186 rte_spinlock_lock(&priv->sh->cmng.csl[cnt_type]);
5187 TAILQ_INSERT_TAIL(&priv->sh->cmng.counters[cnt_type],
5189 rte_spinlock_unlock(&priv->sh->cmng.csl[cnt_type]);
5194 * Verify the @p attributes will be correctly understood by the NIC and store
5195 * them in the @p flow if everything is correct.
5198 * Pointer to dev struct.
5199 * @param[in] attributes
5200 * Pointer to flow attributes
5201 * @param[in] external
5202 * This flow rule is created by request external to PMD.
5204 * Pointer to error structure.
5207 * - 0 on success and non root table.
5208 * - 1 on success and root table.
5209 * - a negative errno value otherwise and rte_errno is set.
5212 flow_dv_validate_attributes(struct rte_eth_dev *dev,
5213 const struct mlx5_flow_tunnel *tunnel,
5214 const struct rte_flow_attr *attributes,
5215 const struct flow_grp_info *grp_info,
5216 struct rte_flow_error *error)
5218 struct mlx5_priv *priv = dev->data->dev_private;
5219 uint32_t priority_max = priv->config.flow_prio - 1;
5222 #ifndef HAVE_MLX5DV_DR
5223 RTE_SET_USED(tunnel);
5224 RTE_SET_USED(grp_info);
5225 if (attributes->group)
5226 return rte_flow_error_set(error, ENOTSUP,
5227 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5229 "groups are not supported");
5233 ret = mlx5_flow_group_to_table(dev, tunnel, attributes->group, &table,
5238 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
5240 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
5241 attributes->priority >= priority_max)
5242 return rte_flow_error_set(error, ENOTSUP,
5243 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
5245 "priority out of range");
5246 if (attributes->transfer) {
5247 if (!priv->config.dv_esw_en)
5248 return rte_flow_error_set
5250 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5251 "E-Switch dr is not supported");
5252 if (!(priv->representor || priv->master))
5253 return rte_flow_error_set
5254 (error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5255 NULL, "E-Switch configuration can only be"
5256 " done by a master or a representor device");
5257 if (attributes->egress)
5258 return rte_flow_error_set
5260 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
5261 "egress is not supported");
5263 if (!(attributes->egress ^ attributes->ingress))
5264 return rte_flow_error_set(error, ENOTSUP,
5265 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
5266 "must specify exactly one of "
5267 "ingress or egress");
5272 * Internal validation function. For validating both actions and items.
5275 * Pointer to the rte_eth_dev structure.
5277 * Pointer to the flow attributes.
5279 * Pointer to the list of items.
5280 * @param[in] actions
5281 * Pointer to the list of actions.
5282 * @param[in] external
5283 * This flow rule is created by request external to PMD.
5284 * @param[in] hairpin
5285 * Number of hairpin TX actions, 0 means classic flow.
5287 * Pointer to the error structure.
5290 * 0 on success, a negative errno value otherwise and rte_errno is set.
5293 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
5294 const struct rte_flow_item items[],
5295 const struct rte_flow_action actions[],
5296 bool external, int hairpin, struct rte_flow_error *error)
5299 uint64_t action_flags = 0;
5300 uint64_t item_flags = 0;
5301 uint64_t last_item = 0;
5302 uint8_t next_protocol = 0xff;
5303 uint16_t ether_type = 0;
5305 uint8_t item_ipv6_proto = 0;
5306 const struct rte_flow_item *geneve_item = NULL;
5307 const struct rte_flow_item *gre_item = NULL;
5308 const struct rte_flow_item *gtp_item = NULL;
5309 const struct rte_flow_action_raw_decap *decap;
5310 const struct rte_flow_action_raw_encap *encap;
5311 const struct rte_flow_action_rss *rss;
5312 const struct rte_flow_item_tcp nic_tcp_mask = {
5315 .src_port = RTE_BE16(UINT16_MAX),
5316 .dst_port = RTE_BE16(UINT16_MAX),
5319 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
5322 "\xff\xff\xff\xff\xff\xff\xff\xff"
5323 "\xff\xff\xff\xff\xff\xff\xff\xff",
5325 "\xff\xff\xff\xff\xff\xff\xff\xff"
5326 "\xff\xff\xff\xff\xff\xff\xff\xff",
5327 .vtc_flow = RTE_BE32(0xffffffff),
5333 const struct rte_flow_item_ecpri nic_ecpri_mask = {
5337 RTE_BE32(((const struct rte_ecpri_common_hdr) {
5341 .dummy[0] = 0xffffffff,
5344 struct mlx5_priv *priv = dev->data->dev_private;
5345 struct mlx5_dev_config *dev_conf = &priv->config;
5346 uint16_t queue_index = 0xFFFF;
5347 const struct rte_flow_item_vlan *vlan_m = NULL;
5348 int16_t rw_act_num = 0;
5350 const struct mlx5_flow_tunnel *tunnel;
5351 struct flow_grp_info grp_info = {
5352 .external = !!external,
5353 .transfer = !!attr->transfer,
5354 .fdb_def_rule = !!priv->fdb_def_rule,
5356 const struct rte_eth_hairpin_conf *conf;
5360 if (is_flow_tunnel_match_rule(dev, attr, items, actions)) {
5361 tunnel = flow_items_to_tunnel(items);
5362 action_flags |= MLX5_FLOW_ACTION_TUNNEL_MATCH |
5363 MLX5_FLOW_ACTION_DECAP;
5364 } else if (is_flow_tunnel_steer_rule(dev, attr, items, actions)) {
5365 tunnel = flow_actions_to_tunnel(actions);
5366 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
5370 if (tunnel && priv->representor)
5371 return rte_flow_error_set(error, ENOTSUP,
5372 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5373 "decap not supported "
5374 "for VF representor");
5375 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
5376 (dev, tunnel, attr, items, actions);
5377 ret = flow_dv_validate_attributes(dev, tunnel, attr, &grp_info, error);
5380 is_root = (uint64_t)ret;
5381 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5382 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
5383 int type = items->type;
5385 if (!mlx5_flow_os_item_supported(type))
5386 return rte_flow_error_set(error, ENOTSUP,
5387 RTE_FLOW_ERROR_TYPE_ITEM,
5388 NULL, "item not supported");
5390 case MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL:
5391 if (items[0].type != (typeof(items[0].type))
5392 MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL)
5393 return rte_flow_error_set
5395 RTE_FLOW_ERROR_TYPE_ITEM,
5396 NULL, "MLX5 private items "
5397 "must be the first");
5399 case RTE_FLOW_ITEM_TYPE_VOID:
5401 case RTE_FLOW_ITEM_TYPE_PORT_ID:
5402 ret = flow_dv_validate_item_port_id
5403 (dev, items, attr, item_flags, error);
5406 last_item = MLX5_FLOW_ITEM_PORT_ID;
5408 case RTE_FLOW_ITEM_TYPE_ETH:
5409 ret = mlx5_flow_validate_item_eth(items, item_flags,
5413 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
5414 MLX5_FLOW_LAYER_OUTER_L2;
5415 if (items->mask != NULL && items->spec != NULL) {
5417 ((const struct rte_flow_item_eth *)
5420 ((const struct rte_flow_item_eth *)
5422 ether_type = rte_be_to_cpu_16(ether_type);
5427 case RTE_FLOW_ITEM_TYPE_VLAN:
5428 ret = flow_dv_validate_item_vlan(items, item_flags,
5432 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
5433 MLX5_FLOW_LAYER_OUTER_VLAN;
5434 if (items->mask != NULL && items->spec != NULL) {
5436 ((const struct rte_flow_item_vlan *)
5437 items->spec)->inner_type;
5439 ((const struct rte_flow_item_vlan *)
5440 items->mask)->inner_type;
5441 ether_type = rte_be_to_cpu_16(ether_type);
5445 /* Store outer VLAN mask for of_push_vlan action. */
5447 vlan_m = items->mask;
5449 case RTE_FLOW_ITEM_TYPE_IPV4:
5450 mlx5_flow_tunnel_ip_check(items, next_protocol,
5451 &item_flags, &tunnel);
5452 ret = flow_dv_validate_item_ipv4(items, item_flags,
5453 last_item, ether_type,
5457 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
5458 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
5459 if (items->mask != NULL &&
5460 ((const struct rte_flow_item_ipv4 *)
5461 items->mask)->hdr.next_proto_id) {
5463 ((const struct rte_flow_item_ipv4 *)
5464 (items->spec))->hdr.next_proto_id;
5466 ((const struct rte_flow_item_ipv4 *)
5467 (items->mask))->hdr.next_proto_id;
5469 /* Reset for inner layer. */
5470 next_protocol = 0xff;
5473 case RTE_FLOW_ITEM_TYPE_IPV6:
5474 mlx5_flow_tunnel_ip_check(items, next_protocol,
5475 &item_flags, &tunnel);
5476 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
5483 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
5484 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
5485 if (items->mask != NULL &&
5486 ((const struct rte_flow_item_ipv6 *)
5487 items->mask)->hdr.proto) {
5489 ((const struct rte_flow_item_ipv6 *)
5490 items->spec)->hdr.proto;
5492 ((const struct rte_flow_item_ipv6 *)
5493 items->spec)->hdr.proto;
5495 ((const struct rte_flow_item_ipv6 *)
5496 items->mask)->hdr.proto;
5498 /* Reset for inner layer. */
5499 next_protocol = 0xff;
5502 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
5503 ret = flow_dv_validate_item_ipv6_frag_ext(items,
5508 last_item = tunnel ?
5509 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
5510 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
5511 if (items->mask != NULL &&
5512 ((const struct rte_flow_item_ipv6_frag_ext *)
5513 items->mask)->hdr.next_header) {
5515 ((const struct rte_flow_item_ipv6_frag_ext *)
5516 items->spec)->hdr.next_header;
5518 ((const struct rte_flow_item_ipv6_frag_ext *)
5519 items->mask)->hdr.next_header;
5521 /* Reset for inner layer. */
5522 next_protocol = 0xff;
5525 case RTE_FLOW_ITEM_TYPE_TCP:
5526 ret = mlx5_flow_validate_item_tcp
5533 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
5534 MLX5_FLOW_LAYER_OUTER_L4_TCP;
5536 case RTE_FLOW_ITEM_TYPE_UDP:
5537 ret = mlx5_flow_validate_item_udp(items, item_flags,
5542 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
5543 MLX5_FLOW_LAYER_OUTER_L4_UDP;
5545 case RTE_FLOW_ITEM_TYPE_GRE:
5546 ret = mlx5_flow_validate_item_gre(items, item_flags,
5547 next_protocol, error);
5551 last_item = MLX5_FLOW_LAYER_GRE;
5553 case RTE_FLOW_ITEM_TYPE_NVGRE:
5554 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
5559 last_item = MLX5_FLOW_LAYER_NVGRE;
5561 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5562 ret = mlx5_flow_validate_item_gre_key
5563 (items, item_flags, gre_item, error);
5566 last_item = MLX5_FLOW_LAYER_GRE_KEY;
5568 case RTE_FLOW_ITEM_TYPE_VXLAN:
5569 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
5573 last_item = MLX5_FLOW_LAYER_VXLAN;
5575 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5576 ret = mlx5_flow_validate_item_vxlan_gpe(items,
5581 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5583 case RTE_FLOW_ITEM_TYPE_GENEVE:
5584 ret = mlx5_flow_validate_item_geneve(items,
5589 geneve_item = items;
5590 last_item = MLX5_FLOW_LAYER_GENEVE;
5592 case RTE_FLOW_ITEM_TYPE_GENEVE_OPT:
5593 ret = mlx5_flow_validate_item_geneve_opt(items,
5600 last_item = MLX5_FLOW_LAYER_GENEVE_OPT;
5602 case RTE_FLOW_ITEM_TYPE_MPLS:
5603 ret = mlx5_flow_validate_item_mpls(dev, items,
5608 last_item = MLX5_FLOW_LAYER_MPLS;
5611 case RTE_FLOW_ITEM_TYPE_MARK:
5612 ret = flow_dv_validate_item_mark(dev, items, attr,
5616 last_item = MLX5_FLOW_ITEM_MARK;
5618 case RTE_FLOW_ITEM_TYPE_META:
5619 ret = flow_dv_validate_item_meta(dev, items, attr,
5623 last_item = MLX5_FLOW_ITEM_METADATA;
5625 case RTE_FLOW_ITEM_TYPE_ICMP:
5626 ret = mlx5_flow_validate_item_icmp(items, item_flags,
5631 last_item = MLX5_FLOW_LAYER_ICMP;
5633 case RTE_FLOW_ITEM_TYPE_ICMP6:
5634 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
5639 item_ipv6_proto = IPPROTO_ICMPV6;
5640 last_item = MLX5_FLOW_LAYER_ICMP6;
5642 case RTE_FLOW_ITEM_TYPE_TAG:
5643 ret = flow_dv_validate_item_tag(dev, items,
5647 last_item = MLX5_FLOW_ITEM_TAG;
5649 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
5650 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
5652 case RTE_FLOW_ITEM_TYPE_GTP:
5653 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
5658 last_item = MLX5_FLOW_LAYER_GTP;
5660 case RTE_FLOW_ITEM_TYPE_GTP_PSC:
5661 ret = flow_dv_validate_item_gtp_psc(items, last_item,
5666 last_item = MLX5_FLOW_LAYER_GTP_PSC;
5668 case RTE_FLOW_ITEM_TYPE_ECPRI:
5669 /* Capacity will be checked in the translate stage. */
5670 ret = mlx5_flow_validate_item_ecpri(items, item_flags,
5677 last_item = MLX5_FLOW_LAYER_ECPRI;
5680 return rte_flow_error_set(error, ENOTSUP,
5681 RTE_FLOW_ERROR_TYPE_ITEM,
5682 NULL, "item not supported");
5684 item_flags |= last_item;
5686 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5687 int type = actions->type;
5689 if (!mlx5_flow_os_action_supported(type))
5690 return rte_flow_error_set(error, ENOTSUP,
5691 RTE_FLOW_ERROR_TYPE_ACTION,
5693 "action not supported");
5694 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5695 return rte_flow_error_set(error, ENOTSUP,
5696 RTE_FLOW_ERROR_TYPE_ACTION,
5697 actions, "too many actions");
5699 case RTE_FLOW_ACTION_TYPE_VOID:
5701 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5702 ret = flow_dv_validate_action_port_id(dev,
5709 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5712 case RTE_FLOW_ACTION_TYPE_FLAG:
5713 ret = flow_dv_validate_action_flag(dev, action_flags,
5717 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5718 /* Count all modify-header actions as one. */
5719 if (!(action_flags &
5720 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5722 action_flags |= MLX5_FLOW_ACTION_FLAG |
5723 MLX5_FLOW_ACTION_MARK_EXT;
5725 action_flags |= MLX5_FLOW_ACTION_FLAG;
5728 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5730 case RTE_FLOW_ACTION_TYPE_MARK:
5731 ret = flow_dv_validate_action_mark(dev, actions,
5736 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5737 /* Count all modify-header actions as one. */
5738 if (!(action_flags &
5739 MLX5_FLOW_MODIFY_HDR_ACTIONS))
5741 action_flags |= MLX5_FLOW_ACTION_MARK |
5742 MLX5_FLOW_ACTION_MARK_EXT;
5744 action_flags |= MLX5_FLOW_ACTION_MARK;
5747 rw_act_num += MLX5_ACT_NUM_SET_MARK;
5749 case RTE_FLOW_ACTION_TYPE_SET_META:
5750 ret = flow_dv_validate_action_set_meta(dev, actions,
5755 /* Count all modify-header actions as one action. */
5756 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5758 action_flags |= MLX5_FLOW_ACTION_SET_META;
5759 rw_act_num += MLX5_ACT_NUM_SET_META;
5761 case RTE_FLOW_ACTION_TYPE_SET_TAG:
5762 ret = flow_dv_validate_action_set_tag(dev, actions,
5767 /* Count all modify-header actions as one action. */
5768 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5770 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
5771 rw_act_num += MLX5_ACT_NUM_SET_TAG;
5773 case RTE_FLOW_ACTION_TYPE_DROP:
5774 ret = mlx5_flow_validate_action_drop(action_flags,
5778 action_flags |= MLX5_FLOW_ACTION_DROP;
5781 case RTE_FLOW_ACTION_TYPE_QUEUE:
5782 ret = mlx5_flow_validate_action_queue(actions,
5787 queue_index = ((const struct rte_flow_action_queue *)
5788 (actions->conf))->index;
5789 action_flags |= MLX5_FLOW_ACTION_QUEUE;
5792 case RTE_FLOW_ACTION_TYPE_RSS:
5793 rss = actions->conf;
5794 ret = mlx5_flow_validate_action_rss(actions,
5800 if (rss != NULL && rss->queue_num)
5801 queue_index = rss->queue[0];
5802 action_flags |= MLX5_FLOW_ACTION_RSS;
5805 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
5807 mlx5_flow_validate_action_default_miss(action_flags,
5811 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
5814 case RTE_FLOW_ACTION_TYPE_COUNT:
5815 ret = flow_dv_validate_action_count(dev, error);
5818 action_flags |= MLX5_FLOW_ACTION_COUNT;
5821 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5822 if (flow_dv_validate_action_pop_vlan(dev,
5828 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5831 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5832 ret = flow_dv_validate_action_push_vlan(dev,
5839 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5842 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5843 ret = flow_dv_validate_action_set_vlan_pcp
5844 (action_flags, actions, error);
5847 /* Count PCP with push_vlan command. */
5848 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5850 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5851 ret = flow_dv_validate_action_set_vlan_vid
5852 (item_flags, action_flags,
5856 /* Count VID with push_vlan command. */
5857 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5858 rw_act_num += MLX5_ACT_NUM_MDF_VID;
5860 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5861 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5862 ret = flow_dv_validate_action_l2_encap(dev,
5868 action_flags |= MLX5_FLOW_ACTION_ENCAP;
5871 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5872 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5873 ret = flow_dv_validate_action_decap(dev, action_flags,
5874 actions, item_flags,
5878 action_flags |= MLX5_FLOW_ACTION_DECAP;
5881 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5882 ret = flow_dv_validate_action_raw_encap_decap
5883 (dev, NULL, actions->conf, attr, &action_flags,
5884 &actions_n, actions, item_flags, error);
5888 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5889 decap = actions->conf;
5890 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5892 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5896 encap = actions->conf;
5898 ret = flow_dv_validate_action_raw_encap_decap
5900 decap ? decap : &empty_decap, encap,
5901 attr, &action_flags, &actions_n,
5902 actions, item_flags, error);
5906 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5907 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5908 ret = flow_dv_validate_action_modify_mac(action_flags,
5914 /* Count all modify-header actions as one action. */
5915 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5917 action_flags |= actions->type ==
5918 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5919 MLX5_FLOW_ACTION_SET_MAC_SRC :
5920 MLX5_FLOW_ACTION_SET_MAC_DST;
5922 * Even if the source and destination MAC addresses have
5923 * overlap in the header with 4B alignment, the convert
5924 * function will handle them separately and 4 SW actions
5925 * will be created. And 2 actions will be added each
5926 * time no matter how many bytes of address will be set.
5928 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5930 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5931 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5932 ret = flow_dv_validate_action_modify_ipv4(action_flags,
5938 /* Count all modify-header actions as one action. */
5939 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5941 action_flags |= actions->type ==
5942 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5943 MLX5_FLOW_ACTION_SET_IPV4_SRC :
5944 MLX5_FLOW_ACTION_SET_IPV4_DST;
5945 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5947 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5948 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5949 ret = flow_dv_validate_action_modify_ipv6(action_flags,
5955 if (item_ipv6_proto == IPPROTO_ICMPV6)
5956 return rte_flow_error_set(error, ENOTSUP,
5957 RTE_FLOW_ERROR_TYPE_ACTION,
5959 "Can't change header "
5960 "with ICMPv6 proto");
5961 /* Count all modify-header actions as one action. */
5962 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5964 action_flags |= actions->type ==
5965 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5966 MLX5_FLOW_ACTION_SET_IPV6_SRC :
5967 MLX5_FLOW_ACTION_SET_IPV6_DST;
5968 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5970 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5971 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5972 ret = flow_dv_validate_action_modify_tp(action_flags,
5978 /* Count all modify-header actions as one action. */
5979 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5981 action_flags |= actions->type ==
5982 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5983 MLX5_FLOW_ACTION_SET_TP_SRC :
5984 MLX5_FLOW_ACTION_SET_TP_DST;
5985 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5987 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5988 case RTE_FLOW_ACTION_TYPE_SET_TTL:
5989 ret = flow_dv_validate_action_modify_ttl(action_flags,
5995 /* Count all modify-header actions as one action. */
5996 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5998 action_flags |= actions->type ==
5999 RTE_FLOW_ACTION_TYPE_SET_TTL ?
6000 MLX5_FLOW_ACTION_SET_TTL :
6001 MLX5_FLOW_ACTION_DEC_TTL;
6002 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
6004 case RTE_FLOW_ACTION_TYPE_JUMP:
6005 ret = flow_dv_validate_action_jump(dev, tunnel, actions,
6012 action_flags |= MLX5_FLOW_ACTION_JUMP;
6014 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
6015 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
6016 ret = flow_dv_validate_action_modify_tcp_seq
6023 /* Count all modify-header actions as one action. */
6024 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
6026 action_flags |= actions->type ==
6027 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
6028 MLX5_FLOW_ACTION_INC_TCP_SEQ :
6029 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
6030 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
6032 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
6033 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
6034 ret = flow_dv_validate_action_modify_tcp_ack
6041 /* Count all modify-header actions as one action. */
6042 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
6044 action_flags |= actions->type ==
6045 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
6046 MLX5_FLOW_ACTION_INC_TCP_ACK :
6047 MLX5_FLOW_ACTION_DEC_TCP_ACK;
6048 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
6050 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
6052 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
6053 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
6054 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6056 case RTE_FLOW_ACTION_TYPE_METER:
6057 ret = mlx5_flow_validate_action_meter(dev,
6063 action_flags |= MLX5_FLOW_ACTION_METER;
6065 /* Meter action will add one more TAG action. */
6066 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6068 case MLX5_RTE_FLOW_ACTION_TYPE_AGE:
6069 if (!attr->transfer && !attr->group)
6070 return rte_flow_error_set(error, ENOTSUP,
6071 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6073 "Shared ASO age action is not supported for group 0");
6074 action_flags |= MLX5_FLOW_ACTION_AGE;
6077 case RTE_FLOW_ACTION_TYPE_AGE:
6078 ret = flow_dv_validate_action_age(action_flags,
6083 action_flags |= MLX5_FLOW_ACTION_AGE;
6086 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
6087 ret = flow_dv_validate_action_modify_ipv4_dscp
6094 /* Count all modify-header actions as one action. */
6095 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
6097 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
6098 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
6100 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
6101 ret = flow_dv_validate_action_modify_ipv6_dscp
6108 /* Count all modify-header actions as one action. */
6109 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
6111 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
6112 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
6114 case RTE_FLOW_ACTION_TYPE_SAMPLE:
6115 ret = flow_dv_validate_action_sample(action_flags,
6121 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
6124 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
6125 if (actions[0].type != (typeof(actions[0].type))
6126 MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET)
6127 return rte_flow_error_set
6129 RTE_FLOW_ERROR_TYPE_ACTION,
6130 NULL, "MLX5 private action "
6131 "must be the first");
6133 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
6136 return rte_flow_error_set(error, ENOTSUP,
6137 RTE_FLOW_ERROR_TYPE_ACTION,
6139 "action not supported");
6143 * Validate actions in flow rules
6144 * - Explicit decap action is prohibited by the tunnel offload API.
6145 * - Drop action in tunnel steer rule is prohibited by the API.
6146 * - Application cannot use MARK action because it's value can mask
6147 * tunnel default miss nitification.
6148 * - JUMP in tunnel match rule has no support in current PMD
6150 * - TAG & META are reserved for future uses.
6152 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_SET) {
6153 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_DECAP |
6154 MLX5_FLOW_ACTION_MARK |
6155 MLX5_FLOW_ACTION_SET_TAG |
6156 MLX5_FLOW_ACTION_SET_META |
6157 MLX5_FLOW_ACTION_DROP;
6159 if (action_flags & bad_actions_mask)
6160 return rte_flow_error_set
6162 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6163 "Invalid RTE action in tunnel "
6165 if (!(action_flags & MLX5_FLOW_ACTION_JUMP))
6166 return rte_flow_error_set
6168 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6169 "tunnel set decap rule must terminate "
6172 return rte_flow_error_set
6174 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6175 "tunnel flows for ingress traffic only");
6177 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_MATCH) {
6178 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_JUMP |
6179 MLX5_FLOW_ACTION_MARK |
6180 MLX5_FLOW_ACTION_SET_TAG |
6181 MLX5_FLOW_ACTION_SET_META;
6183 if (action_flags & bad_actions_mask)
6184 return rte_flow_error_set
6186 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6187 "Invalid RTE action in tunnel "
6191 * Validate the drop action mutual exclusion with other actions.
6192 * Drop action is mutually-exclusive with any other action, except for
6195 if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
6196 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
6197 return rte_flow_error_set(error, EINVAL,
6198 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6199 "Drop action is mutually-exclusive "
6200 "with any other action, except for "
6202 /* Eswitch has few restrictions on using items and actions */
6203 if (attr->transfer) {
6204 if (!mlx5_flow_ext_mreg_supported(dev) &&
6205 action_flags & MLX5_FLOW_ACTION_FLAG)
6206 return rte_flow_error_set(error, ENOTSUP,
6207 RTE_FLOW_ERROR_TYPE_ACTION,
6209 "unsupported action FLAG");
6210 if (!mlx5_flow_ext_mreg_supported(dev) &&
6211 action_flags & MLX5_FLOW_ACTION_MARK)
6212 return rte_flow_error_set(error, ENOTSUP,
6213 RTE_FLOW_ERROR_TYPE_ACTION,
6215 "unsupported action MARK");
6216 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
6217 return rte_flow_error_set(error, ENOTSUP,
6218 RTE_FLOW_ERROR_TYPE_ACTION,
6220 "unsupported action QUEUE");
6221 if (action_flags & MLX5_FLOW_ACTION_RSS)
6222 return rte_flow_error_set(error, ENOTSUP,
6223 RTE_FLOW_ERROR_TYPE_ACTION,
6225 "unsupported action RSS");
6226 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
6227 return rte_flow_error_set(error, EINVAL,
6228 RTE_FLOW_ERROR_TYPE_ACTION,
6230 "no fate action is found");
6232 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
6233 return rte_flow_error_set(error, EINVAL,
6234 RTE_FLOW_ERROR_TYPE_ACTION,
6236 "no fate action is found");
6239 * Continue validation for Xcap and VLAN actions.
6240 * If hairpin is working in explicit TX rule mode, there is no actions
6241 * splitting and the validation of hairpin ingress flow should be the
6242 * same as other standard flows.
6244 if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
6245 MLX5_FLOW_VLAN_ACTIONS)) &&
6246 (queue_index == 0xFFFF ||
6247 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN ||
6248 ((conf = mlx5_rxq_get_hairpin_conf(dev, queue_index)) != NULL &&
6249 conf->tx_explicit != 0))) {
6250 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
6251 MLX5_FLOW_XCAP_ACTIONS)
6252 return rte_flow_error_set(error, ENOTSUP,
6253 RTE_FLOW_ERROR_TYPE_ACTION,
6254 NULL, "encap and decap "
6255 "combination aren't supported");
6256 if (!attr->transfer && attr->ingress) {
6257 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
6258 return rte_flow_error_set
6260 RTE_FLOW_ERROR_TYPE_ACTION,
6261 NULL, "encap is not supported"
6262 " for ingress traffic");
6263 else if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
6264 return rte_flow_error_set
6266 RTE_FLOW_ERROR_TYPE_ACTION,
6267 NULL, "push VLAN action not "
6268 "supported for ingress");
6269 else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
6270 MLX5_FLOW_VLAN_ACTIONS)
6271 return rte_flow_error_set
6273 RTE_FLOW_ERROR_TYPE_ACTION,
6274 NULL, "no support for "
6275 "multiple VLAN actions");
6279 * Hairpin flow will add one more TAG action in TX implicit mode.
6280 * In TX explicit mode, there will be no hairpin flow ID.
6283 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6284 /* extra metadata enabled: one more TAG action will be add. */
6285 if (dev_conf->dv_flow_en &&
6286 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
6287 mlx5_flow_ext_mreg_supported(dev))
6288 rw_act_num += MLX5_ACT_NUM_SET_TAG;
6289 if ((uint32_t)rw_act_num >
6290 flow_dv_modify_hdr_action_max(dev, is_root)) {
6291 return rte_flow_error_set(error, ENOTSUP,
6292 RTE_FLOW_ERROR_TYPE_ACTION,
6293 NULL, "too many header modify"
6294 " actions to support");
6300 * Internal preparation function. Allocates the DV flow size,
6301 * this size is constant.
6304 * Pointer to the rte_eth_dev structure.
6306 * Pointer to the flow attributes.
6308 * Pointer to the list of items.
6309 * @param[in] actions
6310 * Pointer to the list of actions.
6312 * Pointer to the error structure.
6315 * Pointer to mlx5_flow object on success,
6316 * otherwise NULL and rte_errno is set.
6318 static struct mlx5_flow *
6319 flow_dv_prepare(struct rte_eth_dev *dev,
6320 const struct rte_flow_attr *attr __rte_unused,
6321 const struct rte_flow_item items[] __rte_unused,
6322 const struct rte_flow_action actions[] __rte_unused,
6323 struct rte_flow_error *error)
6325 uint32_t handle_idx = 0;
6326 struct mlx5_flow *dev_flow;
6327 struct mlx5_flow_handle *dev_handle;
6328 struct mlx5_priv *priv = dev->data->dev_private;
6329 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6332 /* In case of corrupting the memory. */
6333 if (wks->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
6334 rte_flow_error_set(error, ENOSPC,
6335 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6336 "not free temporary device flow");
6339 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
6342 rte_flow_error_set(error, ENOMEM,
6343 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6344 "not enough memory to create flow handle");
6347 MLX5_ASSERT(wks->flow_idx < RTE_DIM(wks->flows));
6348 dev_flow = &wks->flows[wks->flow_idx++];
6349 memset(dev_flow, 0, sizeof(*dev_flow));
6350 dev_flow->handle = dev_handle;
6351 dev_flow->handle_idx = handle_idx;
6353 * In some old rdma-core releases, before continuing, a check of the
6354 * length of matching parameter will be done at first. It needs to use
6355 * the length without misc4 param. If the flow has misc4 support, then
6356 * the length needs to be adjusted accordingly. Each param member is
6357 * aligned with a 64B boundary naturally.
6359 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param) -
6360 MLX5_ST_SZ_BYTES(fte_match_set_misc4);
6361 dev_flow->ingress = attr->ingress;
6362 dev_flow->dv.transfer = attr->transfer;
6366 #ifdef RTE_LIBRTE_MLX5_DEBUG
6368 * Sanity check for match mask and value. Similar to check_valid_spec() in
6369 * kernel driver. If unmasked bit is present in value, it returns failure.
6372 * pointer to match mask buffer.
6373 * @param match_value
6374 * pointer to match value buffer.
6377 * 0 if valid, -EINVAL otherwise.
6380 flow_dv_check_valid_spec(void *match_mask, void *match_value)
6382 uint8_t *m = match_mask;
6383 uint8_t *v = match_value;
6386 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
6389 "match_value differs from match_criteria"
6390 " %p[%u] != %p[%u]",
6391 match_value, i, match_mask, i);
6400 * Add match of ip_version.
6404 * @param[in] headers_v
6405 * Values header pointer.
6406 * @param[in] headers_m
6407 * Masks header pointer.
6408 * @param[in] ip_version
6409 * The IP version to set.
6412 flow_dv_set_match_ip_version(uint32_t group,
6418 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
6420 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
6422 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
6423 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
6424 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
6428 * Add Ethernet item to matcher and to the value.
6430 * @param[in, out] matcher
6432 * @param[in, out] key
6433 * Flow matcher value.
6435 * Flow pattern to translate.
6437 * Item is inner pattern.
6440 flow_dv_translate_item_eth(void *matcher, void *key,
6441 const struct rte_flow_item *item, int inner,
6444 const struct rte_flow_item_eth *eth_m = item->mask;
6445 const struct rte_flow_item_eth *eth_v = item->spec;
6446 const struct rte_flow_item_eth nic_mask = {
6447 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6448 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6449 .type = RTE_BE16(0xffff),
6462 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6464 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6466 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6468 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6470 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, dmac_47_16),
6471 ð_m->dst, sizeof(eth_m->dst));
6472 /* The value must be in the range of the mask. */
6473 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, dmac_47_16);
6474 for (i = 0; i < sizeof(eth_m->dst); ++i)
6475 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
6476 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, smac_47_16),
6477 ð_m->src, sizeof(eth_m->src));
6478 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, smac_47_16);
6479 /* The value must be in the range of the mask. */
6480 for (i = 0; i < sizeof(eth_m->dst); ++i)
6481 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
6483 * HW supports match on one Ethertype, the Ethertype following the last
6484 * VLAN tag of the packet (see PRM).
6485 * Set match on ethertype only if ETH header is not followed by VLAN.
6486 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6487 * ethertype, and use ip_version field instead.
6488 * eCPRI over Ether layer will use type value 0xAEFE.
6490 if (eth_m->type == 0xFFFF) {
6491 /* Set cvlan_tag mask for any single\multi\un-tagged case. */
6492 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6493 switch (eth_v->type) {
6494 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
6495 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6497 case RTE_BE16(RTE_ETHER_TYPE_QINQ):
6498 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6499 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6501 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
6502 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
6504 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
6505 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
6511 if (eth_m->has_vlan) {
6512 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6513 if (eth_v->has_vlan) {
6515 * Here, when also has_more_vlan field in VLAN item is
6516 * not set, only single-tagged packets will be matched.
6518 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6522 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
6523 rte_be_to_cpu_16(eth_m->type));
6524 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, ethertype);
6525 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
6529 * Add VLAN item to matcher and to the value.
6531 * @param[in, out] dev_flow
6533 * @param[in, out] matcher
6535 * @param[in, out] key
6536 * Flow matcher value.
6538 * Flow pattern to translate.
6540 * Item is inner pattern.
6543 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
6544 void *matcher, void *key,
6545 const struct rte_flow_item *item,
6546 int inner, uint32_t group)
6548 const struct rte_flow_item_vlan *vlan_m = item->mask;
6549 const struct rte_flow_item_vlan *vlan_v = item->spec;
6556 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6558 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6560 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
6562 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6564 * This is workaround, masks are not supported,
6565 * and pre-validated.
6568 dev_flow->handle->vf_vlan.tag =
6569 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
6572 * When VLAN item exists in flow, mark packet as tagged,
6573 * even if TCI is not specified.
6575 if (!MLX5_GET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag)) {
6576 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
6577 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
6582 vlan_m = &rte_flow_item_vlan_mask;
6583 tci_m = rte_be_to_cpu_16(vlan_m->tci);
6584 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
6585 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_vid, tci_m);
6586 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_vid, tci_v);
6587 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_cfi, tci_m >> 12);
6588 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_cfi, tci_v >> 12);
6589 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_prio, tci_m >> 13);
6590 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_prio, tci_v >> 13);
6592 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6593 * ethertype, and use ip_version field instead.
6595 if (vlan_m->inner_type == 0xFFFF) {
6596 switch (vlan_v->inner_type) {
6597 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
6598 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6599 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6600 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
6602 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
6603 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
6605 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
6606 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
6612 if (vlan_m->has_more_vlan && vlan_v->has_more_vlan) {
6613 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
6614 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
6615 /* Only one vlan_tag bit can be set. */
6616 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
6619 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
6620 rte_be_to_cpu_16(vlan_m->inner_type));
6621 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, ethertype,
6622 rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
6626 * Add IPV4 item to matcher and to the value.
6628 * @param[in, out] matcher
6630 * @param[in, out] key
6631 * Flow matcher value.
6633 * Flow pattern to translate.
6635 * Item is inner pattern.
6637 * The group to insert the rule.
6640 flow_dv_translate_item_ipv4(void *matcher, void *key,
6641 const struct rte_flow_item *item,
6642 int inner, uint32_t group)
6644 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
6645 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
6646 const struct rte_flow_item_ipv4 nic_mask = {
6648 .src_addr = RTE_BE32(0xffffffff),
6649 .dst_addr = RTE_BE32(0xffffffff),
6650 .type_of_service = 0xff,
6651 .next_proto_id = 0xff,
6652 .time_to_live = 0xff,
6662 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6664 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6666 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6668 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6670 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6675 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6676 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6677 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6678 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6679 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
6680 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
6681 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6682 src_ipv4_src_ipv6.ipv4_layout.ipv4);
6683 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6684 src_ipv4_src_ipv6.ipv4_layout.ipv4);
6685 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
6686 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
6687 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
6688 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
6689 ipv4_m->hdr.type_of_service);
6690 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
6691 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
6692 ipv4_m->hdr.type_of_service >> 2);
6693 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
6694 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6695 ipv4_m->hdr.next_proto_id);
6696 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6697 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
6698 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6699 ipv4_m->hdr.time_to_live);
6700 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6701 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
6702 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
6703 !!(ipv4_m->hdr.fragment_offset));
6704 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
6705 !!(ipv4_v->hdr.fragment_offset & ipv4_m->hdr.fragment_offset));
6709 * Add IPV6 item to matcher and to the value.
6711 * @param[in, out] matcher
6713 * @param[in, out] key
6714 * Flow matcher value.
6716 * Flow pattern to translate.
6718 * Item is inner pattern.
6720 * The group to insert the rule.
6723 flow_dv_translate_item_ipv6(void *matcher, void *key,
6724 const struct rte_flow_item *item,
6725 int inner, uint32_t group)
6727 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
6728 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
6729 const struct rte_flow_item_ipv6 nic_mask = {
6732 "\xff\xff\xff\xff\xff\xff\xff\xff"
6733 "\xff\xff\xff\xff\xff\xff\xff\xff",
6735 "\xff\xff\xff\xff\xff\xff\xff\xff"
6736 "\xff\xff\xff\xff\xff\xff\xff\xff",
6737 .vtc_flow = RTE_BE32(0xffffffff),
6744 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6745 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6754 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6756 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6758 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6760 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6762 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6767 size = sizeof(ipv6_m->hdr.dst_addr);
6768 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6769 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6770 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6771 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6772 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
6773 for (i = 0; i < size; ++i)
6774 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
6775 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6776 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6777 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6778 src_ipv4_src_ipv6.ipv6_layout.ipv6);
6779 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
6780 for (i = 0; i < size; ++i)
6781 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
6783 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
6784 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
6785 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
6786 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
6787 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
6788 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
6791 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
6793 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
6796 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
6798 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
6802 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6804 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6805 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
6807 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6808 ipv6_m->hdr.hop_limits);
6809 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6810 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
6811 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
6812 !!(ipv6_m->has_frag_ext));
6813 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
6814 !!(ipv6_v->has_frag_ext & ipv6_m->has_frag_ext));
6818 * Add IPV6 fragment extension item to matcher and to the value.
6820 * @param[in, out] matcher
6822 * @param[in, out] key
6823 * Flow matcher value.
6825 * Flow pattern to translate.
6827 * Item is inner pattern.
6830 flow_dv_translate_item_ipv6_frag_ext(void *matcher, void *key,
6831 const struct rte_flow_item *item,
6834 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_m = item->mask;
6835 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_v = item->spec;
6836 const struct rte_flow_item_ipv6_frag_ext nic_mask = {
6838 .next_header = 0xff,
6839 .frag_data = RTE_BE16(0xffff),
6846 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6848 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6850 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6852 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6854 /* IPv6 fragment extension item exists, so packet is IP fragment. */
6855 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
6856 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 1);
6857 if (!ipv6_frag_ext_v)
6859 if (!ipv6_frag_ext_m)
6860 ipv6_frag_ext_m = &nic_mask;
6861 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6862 ipv6_frag_ext_m->hdr.next_header);
6863 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6864 ipv6_frag_ext_v->hdr.next_header &
6865 ipv6_frag_ext_m->hdr.next_header);
6869 * Add TCP item to matcher and to the value.
6871 * @param[in, out] matcher
6873 * @param[in, out] key
6874 * Flow matcher value.
6876 * Flow pattern to translate.
6878 * Item is inner pattern.
6881 flow_dv_translate_item_tcp(void *matcher, void *key,
6882 const struct rte_flow_item *item,
6885 const struct rte_flow_item_tcp *tcp_m = item->mask;
6886 const struct rte_flow_item_tcp *tcp_v = item->spec;
6891 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6893 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6895 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6897 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6899 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6900 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
6904 tcp_m = &rte_flow_item_tcp_mask;
6905 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
6906 rte_be_to_cpu_16(tcp_m->hdr.src_port));
6907 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
6908 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
6909 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
6910 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
6911 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
6912 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
6913 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
6914 tcp_m->hdr.tcp_flags);
6915 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
6916 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
6920 * Add UDP item to matcher and to the value.
6922 * @param[in, out] matcher
6924 * @param[in, out] key
6925 * Flow matcher value.
6927 * Flow pattern to translate.
6929 * Item is inner pattern.
6932 flow_dv_translate_item_udp(void *matcher, void *key,
6933 const struct rte_flow_item *item,
6936 const struct rte_flow_item_udp *udp_m = item->mask;
6937 const struct rte_flow_item_udp *udp_v = item->spec;
6942 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6944 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6946 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6948 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6950 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6951 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
6955 udp_m = &rte_flow_item_udp_mask;
6956 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
6957 rte_be_to_cpu_16(udp_m->hdr.src_port));
6958 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
6959 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
6960 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
6961 rte_be_to_cpu_16(udp_m->hdr.dst_port));
6962 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6963 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
6967 * Add GRE optional Key item to matcher and to the value.
6969 * @param[in, out] matcher
6971 * @param[in, out] key
6972 * Flow matcher value.
6974 * Flow pattern to translate.
6976 * Item is inner pattern.
6979 flow_dv_translate_item_gre_key(void *matcher, void *key,
6980 const struct rte_flow_item *item)
6982 const rte_be32_t *key_m = item->mask;
6983 const rte_be32_t *key_v = item->spec;
6984 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6985 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6986 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
6988 /* GRE K bit must be on and should already be validated */
6989 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
6990 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
6994 key_m = &gre_key_default_mask;
6995 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
6996 rte_be_to_cpu_32(*key_m) >> 8);
6997 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
6998 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
6999 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
7000 rte_be_to_cpu_32(*key_m) & 0xFF);
7001 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
7002 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
7006 * Add GRE item to matcher and to the value.
7008 * @param[in, out] matcher
7010 * @param[in, out] key
7011 * Flow matcher value.
7013 * Flow pattern to translate.
7015 * Item is inner pattern.
7018 flow_dv_translate_item_gre(void *matcher, void *key,
7019 const struct rte_flow_item *item,
7022 const struct rte_flow_item_gre *gre_m = item->mask;
7023 const struct rte_flow_item_gre *gre_v = item->spec;
7026 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7027 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7034 uint16_t s_present:1;
7035 uint16_t k_present:1;
7036 uint16_t rsvd_bit1:1;
7037 uint16_t c_present:1;
7041 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
7044 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7046 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7048 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7050 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7052 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
7053 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
7057 gre_m = &rte_flow_item_gre_mask;
7058 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
7059 rte_be_to_cpu_16(gre_m->protocol));
7060 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
7061 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
7062 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
7063 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
7064 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
7065 gre_crks_rsvd0_ver_m.c_present);
7066 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
7067 gre_crks_rsvd0_ver_v.c_present &
7068 gre_crks_rsvd0_ver_m.c_present);
7069 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
7070 gre_crks_rsvd0_ver_m.k_present);
7071 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
7072 gre_crks_rsvd0_ver_v.k_present &
7073 gre_crks_rsvd0_ver_m.k_present);
7074 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
7075 gre_crks_rsvd0_ver_m.s_present);
7076 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
7077 gre_crks_rsvd0_ver_v.s_present &
7078 gre_crks_rsvd0_ver_m.s_present);
7082 * Add NVGRE item to matcher and to the value.
7084 * @param[in, out] matcher
7086 * @param[in, out] key
7087 * Flow matcher value.
7089 * Flow pattern to translate.
7091 * Item is inner pattern.
7094 flow_dv_translate_item_nvgre(void *matcher, void *key,
7095 const struct rte_flow_item *item,
7098 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
7099 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
7100 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7101 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7102 const char *tni_flow_id_m;
7103 const char *tni_flow_id_v;
7109 /* For NVGRE, GRE header fields must be set with defined values. */
7110 const struct rte_flow_item_gre gre_spec = {
7111 .c_rsvd0_ver = RTE_BE16(0x2000),
7112 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
7114 const struct rte_flow_item_gre gre_mask = {
7115 .c_rsvd0_ver = RTE_BE16(0xB000),
7116 .protocol = RTE_BE16(UINT16_MAX),
7118 const struct rte_flow_item gre_item = {
7123 flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
7127 nvgre_m = &rte_flow_item_nvgre_mask;
7128 tni_flow_id_m = (const char *)nvgre_m->tni;
7129 tni_flow_id_v = (const char *)nvgre_v->tni;
7130 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
7131 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
7132 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
7133 memcpy(gre_key_m, tni_flow_id_m, size);
7134 for (i = 0; i < size; ++i)
7135 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
7139 * Add VXLAN item to matcher and to the value.
7141 * @param[in, out] matcher
7143 * @param[in, out] key
7144 * Flow matcher value.
7146 * Flow pattern to translate.
7148 * Item is inner pattern.
7151 flow_dv_translate_item_vxlan(void *matcher, void *key,
7152 const struct rte_flow_item *item,
7155 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
7156 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
7159 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7160 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7168 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7170 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7172 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7174 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7176 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
7177 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
7178 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7179 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7180 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7185 vxlan_m = &rte_flow_item_vxlan_mask;
7186 size = sizeof(vxlan_m->vni);
7187 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
7188 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
7189 memcpy(vni_m, vxlan_m->vni, size);
7190 for (i = 0; i < size; ++i)
7191 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7195 * Add VXLAN-GPE item to matcher and to the value.
7197 * @param[in, out] matcher
7199 * @param[in, out] key
7200 * Flow matcher value.
7202 * Flow pattern to translate.
7204 * Item is inner pattern.
7208 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
7209 const struct rte_flow_item *item, int inner)
7211 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
7212 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
7216 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
7218 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7224 uint8_t flags_m = 0xff;
7225 uint8_t flags_v = 0xc;
7228 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7230 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7232 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7234 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7236 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
7237 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
7238 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7239 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7240 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7245 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
7246 size = sizeof(vxlan_m->vni);
7247 vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
7248 vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
7249 memcpy(vni_m, vxlan_m->vni, size);
7250 for (i = 0; i < size; ++i)
7251 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7252 if (vxlan_m->flags) {
7253 flags_m = vxlan_m->flags;
7254 flags_v = vxlan_v->flags;
7256 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
7257 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
7258 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
7260 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
7265 * Add Geneve item to matcher and to the value.
7267 * @param[in, out] matcher
7269 * @param[in, out] key
7270 * Flow matcher value.
7272 * Flow pattern to translate.
7274 * Item is inner pattern.
7278 flow_dv_translate_item_geneve(void *matcher, void *key,
7279 const struct rte_flow_item *item, int inner)
7281 const struct rte_flow_item_geneve *geneve_m = item->mask;
7282 const struct rte_flow_item_geneve *geneve_v = item->spec;
7285 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7286 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7295 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7297 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7299 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7301 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7303 dport = MLX5_UDP_PORT_GENEVE;
7304 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7305 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7306 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7311 geneve_m = &rte_flow_item_geneve_mask;
7312 size = sizeof(geneve_m->vni);
7313 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
7314 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
7315 memcpy(vni_m, geneve_m->vni, size);
7316 for (i = 0; i < size; ++i)
7317 vni_v[i] = vni_m[i] & geneve_v->vni[i];
7318 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
7319 rte_be_to_cpu_16(geneve_m->protocol));
7320 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
7321 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
7322 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
7323 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
7324 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
7325 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7326 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
7327 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7328 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
7329 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7330 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
7331 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
7332 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7336 * Create Geneve TLV option resource.
7338 * @param dev[in, out]
7339 * Pointer to rte_eth_dev structure.
7340 * @param[in, out] tag_be24
7341 * Tag value in big endian then R-shift 8.
7342 * @parm[in, out] dev_flow
7343 * Pointer to the dev_flow.
7345 * pointer to error structure.
7348 * 0 on success otherwise -errno and errno is set.
7352 flow_dev_geneve_tlv_option_resource_register(struct rte_eth_dev *dev,
7353 const struct rte_flow_item *item,
7354 struct rte_flow_error *error)
7356 struct mlx5_priv *priv = dev->data->dev_private;
7357 struct mlx5_dev_ctx_shared *sh = priv->sh;
7358 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource =
7359 sh->geneve_tlv_option_resource;
7360 struct mlx5_devx_obj *obj;
7361 const struct rte_flow_item_geneve_opt *geneve_opt_v = item->spec;
7366 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
7367 if (geneve_opt_resource != NULL) {
7368 if (geneve_opt_resource->option_class ==
7369 geneve_opt_v->option_class &&
7370 geneve_opt_resource->option_type ==
7371 geneve_opt_v->option_type &&
7372 geneve_opt_resource->length ==
7373 geneve_opt_v->option_len) {
7374 /* We already have GENVE TLV option obj allocated. */
7375 __atomic_fetch_add(&geneve_opt_resource->refcnt, 1,
7378 ret = rte_flow_error_set(error, ENOMEM,
7379 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7380 "Only one GENEVE TLV option supported");
7384 /* Create a GENEVE TLV object and resource. */
7385 obj = mlx5_devx_cmd_create_geneve_tlv_option(sh->ctx,
7386 geneve_opt_v->option_class,
7387 geneve_opt_v->option_type,
7388 geneve_opt_v->option_len);
7390 ret = rte_flow_error_set(error, ENODATA,
7391 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7392 "Failed to create GENEVE TLV Devx object");
7395 sh->geneve_tlv_option_resource =
7396 mlx5_malloc(MLX5_MEM_ZERO,
7397 sizeof(*geneve_opt_resource),
7399 if (!sh->geneve_tlv_option_resource) {
7400 claim_zero(mlx5_devx_cmd_destroy(obj));
7401 ret = rte_flow_error_set(error, ENOMEM,
7402 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7403 "GENEVE TLV object memory allocation failed");
7406 geneve_opt_resource = sh->geneve_tlv_option_resource;
7407 geneve_opt_resource->obj = obj;
7408 geneve_opt_resource->option_class = geneve_opt_v->option_class;
7409 geneve_opt_resource->option_type = geneve_opt_v->option_type;
7410 geneve_opt_resource->length = geneve_opt_v->option_len;
7411 __atomic_store_n(&geneve_opt_resource->refcnt, 1,
7415 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
7420 * Add MPLS item to matcher and to the value.
7422 * @param[in, out] matcher
7424 * @param[in, out] key
7425 * Flow matcher value.
7427 * Flow pattern to translate.
7428 * @param[in] prev_layer
7429 * The protocol layer indicated in previous item.
7431 * Item is inner pattern.
7434 flow_dv_translate_item_mpls(void *matcher, void *key,
7435 const struct rte_flow_item *item,
7436 uint64_t prev_layer,
7439 const uint32_t *in_mpls_m = item->mask;
7440 const uint32_t *in_mpls_v = item->spec;
7441 uint32_t *out_mpls_m = 0;
7442 uint32_t *out_mpls_v = 0;
7443 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7444 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7445 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
7447 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7448 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
7449 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7451 switch (prev_layer) {
7452 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7453 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
7454 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
7455 MLX5_UDP_PORT_MPLS);
7457 case MLX5_FLOW_LAYER_GRE:
7458 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
7459 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
7460 RTE_ETHER_TYPE_MPLS);
7463 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
7464 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
7471 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
7472 switch (prev_layer) {
7473 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7475 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7476 outer_first_mpls_over_udp);
7478 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7479 outer_first_mpls_over_udp);
7481 case MLX5_FLOW_LAYER_GRE:
7483 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7484 outer_first_mpls_over_gre);
7486 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7487 outer_first_mpls_over_gre);
7490 /* Inner MPLS not over GRE is not supported. */
7493 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7497 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7503 if (out_mpls_m && out_mpls_v) {
7504 *out_mpls_m = *in_mpls_m;
7505 *out_mpls_v = *in_mpls_v & *in_mpls_m;
7510 * Add metadata register item to matcher
7512 * @param[in, out] matcher
7514 * @param[in, out] key
7515 * Flow matcher value.
7516 * @param[in] reg_type
7517 * Type of device metadata register
7524 flow_dv_match_meta_reg(void *matcher, void *key,
7525 enum modify_reg reg_type,
7526 uint32_t data, uint32_t mask)
7529 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
7531 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7537 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
7538 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
7541 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
7542 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
7546 * The metadata register C0 field might be divided into
7547 * source vport index and META item value, we should set
7548 * this field according to specified mask, not as whole one.
7550 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
7552 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
7553 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
7556 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
7559 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
7560 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
7563 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
7564 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
7567 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
7568 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
7571 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
7572 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
7575 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
7576 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
7579 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
7580 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
7583 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
7584 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
7593 * Add MARK item to matcher
7596 * The device to configure through.
7597 * @param[in, out] matcher
7599 * @param[in, out] key
7600 * Flow matcher value.
7602 * Flow pattern to translate.
7605 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
7606 void *matcher, void *key,
7607 const struct rte_flow_item *item)
7609 struct mlx5_priv *priv = dev->data->dev_private;
7610 const struct rte_flow_item_mark *mark;
7614 mark = item->mask ? (const void *)item->mask :
7615 &rte_flow_item_mark_mask;
7616 mask = mark->id & priv->sh->dv_mark_mask;
7617 mark = (const void *)item->spec;
7619 value = mark->id & priv->sh->dv_mark_mask & mask;
7621 enum modify_reg reg;
7623 /* Get the metadata register index for the mark. */
7624 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
7625 MLX5_ASSERT(reg > 0);
7626 if (reg == REG_C_0) {
7627 struct mlx5_priv *priv = dev->data->dev_private;
7628 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7629 uint32_t shl_c0 = rte_bsf32(msk_c0);
7635 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7640 * Add META item to matcher
7643 * The devich to configure through.
7644 * @param[in, out] matcher
7646 * @param[in, out] key
7647 * Flow matcher value.
7649 * Attributes of flow that includes this item.
7651 * Flow pattern to translate.
7654 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
7655 void *matcher, void *key,
7656 const struct rte_flow_attr *attr,
7657 const struct rte_flow_item *item)
7659 const struct rte_flow_item_meta *meta_m;
7660 const struct rte_flow_item_meta *meta_v;
7662 meta_m = (const void *)item->mask;
7664 meta_m = &rte_flow_item_meta_mask;
7665 meta_v = (const void *)item->spec;
7668 uint32_t value = meta_v->data;
7669 uint32_t mask = meta_m->data;
7671 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
7674 MLX5_ASSERT(reg != REG_NON);
7676 * In datapath code there is no endianness
7677 * coversions for perfromance reasons, all
7678 * pattern conversions are done in rte_flow.
7680 value = rte_cpu_to_be_32(value);
7681 mask = rte_cpu_to_be_32(mask);
7682 if (reg == REG_C_0) {
7683 struct mlx5_priv *priv = dev->data->dev_private;
7684 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7685 uint32_t shl_c0 = rte_bsf32(msk_c0);
7686 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
7687 uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
7694 MLX5_ASSERT(msk_c0);
7695 MLX5_ASSERT(!(~msk_c0 & mask));
7697 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7702 * Add vport metadata Reg C0 item to matcher
7704 * @param[in, out] matcher
7706 * @param[in, out] key
7707 * Flow matcher value.
7709 * Flow pattern to translate.
7712 flow_dv_translate_item_meta_vport(void *matcher, void *key,
7713 uint32_t value, uint32_t mask)
7715 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
7719 * Add tag item to matcher
7722 * The devich to configure through.
7723 * @param[in, out] matcher
7725 * @param[in, out] key
7726 * Flow matcher value.
7728 * Flow pattern to translate.
7731 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
7732 void *matcher, void *key,
7733 const struct rte_flow_item *item)
7735 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
7736 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
7737 uint32_t mask, value;
7740 value = tag_v->data;
7741 mask = tag_m ? tag_m->data : UINT32_MAX;
7742 if (tag_v->id == REG_C_0) {
7743 struct mlx5_priv *priv = dev->data->dev_private;
7744 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7745 uint32_t shl_c0 = rte_bsf32(msk_c0);
7751 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
7755 * Add TAG item to matcher
7758 * The devich to configure through.
7759 * @param[in, out] matcher
7761 * @param[in, out] key
7762 * Flow matcher value.
7764 * Flow pattern to translate.
7767 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
7768 void *matcher, void *key,
7769 const struct rte_flow_item *item)
7771 const struct rte_flow_item_tag *tag_v = item->spec;
7772 const struct rte_flow_item_tag *tag_m = item->mask;
7773 enum modify_reg reg;
7776 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
7777 /* Get the metadata register index for the tag. */
7778 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
7779 MLX5_ASSERT(reg > 0);
7780 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
7784 * Add source vport match to the specified matcher.
7786 * @param[in, out] matcher
7788 * @param[in, out] key
7789 * Flow matcher value.
7791 * Source vport value to match
7796 flow_dv_translate_item_source_vport(void *matcher, void *key,
7797 int16_t port, uint16_t mask)
7799 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7800 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7802 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
7803 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
7807 * Translate port-id item to eswitch match on port-id.
7810 * The devich to configure through.
7811 * @param[in, out] matcher
7813 * @param[in, out] key
7814 * Flow matcher value.
7816 * Flow pattern to translate.
7821 * 0 on success, a negative errno value otherwise.
7824 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
7825 void *key, const struct rte_flow_item *item,
7826 const struct rte_flow_attr *attr)
7828 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
7829 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
7830 struct mlx5_priv *priv;
7833 mask = pid_m ? pid_m->id : 0xffff;
7834 id = pid_v ? pid_v->id : dev->data->port_id;
7835 priv = mlx5_port_to_eswitch_info(id, item == NULL);
7839 * Translate to vport field or to metadata, depending on mode.
7840 * Kernel can use either misc.source_port or half of C0 metadata
7843 if (priv->vport_meta_mask) {
7845 * Provide the hint for SW steering library
7846 * to insert the flow into ingress domain and
7847 * save the extra vport match.
7849 if (mask == 0xffff && priv->vport_id == 0xffff &&
7850 priv->pf_bond < 0 && attr->transfer)
7851 flow_dv_translate_item_source_vport
7852 (matcher, key, priv->vport_id, mask);
7854 flow_dv_translate_item_meta_vport
7856 priv->vport_meta_tag,
7857 priv->vport_meta_mask);
7859 flow_dv_translate_item_source_vport(matcher, key,
7860 priv->vport_id, mask);
7866 * Add ICMP6 item to matcher and to the value.
7868 * @param[in, out] matcher
7870 * @param[in, out] key
7871 * Flow matcher value.
7873 * Flow pattern to translate.
7875 * Item is inner pattern.
7878 flow_dv_translate_item_icmp6(void *matcher, void *key,
7879 const struct rte_flow_item *item,
7882 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
7883 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
7886 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7888 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7890 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7892 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7894 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7896 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7898 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7899 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
7903 icmp6_m = &rte_flow_item_icmp6_mask;
7904 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
7905 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
7906 icmp6_v->type & icmp6_m->type);
7907 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
7908 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
7909 icmp6_v->code & icmp6_m->code);
7913 * Add ICMP item to matcher and to the value.
7915 * @param[in, out] matcher
7917 * @param[in, out] key
7918 * Flow matcher value.
7920 * Flow pattern to translate.
7922 * Item is inner pattern.
7925 flow_dv_translate_item_icmp(void *matcher, void *key,
7926 const struct rte_flow_item *item,
7929 const struct rte_flow_item_icmp *icmp_m = item->mask;
7930 const struct rte_flow_item_icmp *icmp_v = item->spec;
7931 uint32_t icmp_header_data_m = 0;
7932 uint32_t icmp_header_data_v = 0;
7935 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7937 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7939 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7941 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7943 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7945 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7947 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7948 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
7952 icmp_m = &rte_flow_item_icmp_mask;
7953 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
7954 icmp_m->hdr.icmp_type);
7955 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
7956 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
7957 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
7958 icmp_m->hdr.icmp_code);
7959 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
7960 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
7961 icmp_header_data_m = rte_be_to_cpu_16(icmp_m->hdr.icmp_seq_nb);
7962 icmp_header_data_m |= rte_be_to_cpu_16(icmp_m->hdr.icmp_ident) << 16;
7963 if (icmp_header_data_m) {
7964 icmp_header_data_v = rte_be_to_cpu_16(icmp_v->hdr.icmp_seq_nb);
7965 icmp_header_data_v |=
7966 rte_be_to_cpu_16(icmp_v->hdr.icmp_ident) << 16;
7967 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_header_data,
7968 icmp_header_data_m);
7969 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_header_data,
7970 icmp_header_data_v & icmp_header_data_m);
7975 * Add GTP item to matcher and to the value.
7977 * @param[in, out] matcher
7979 * @param[in, out] key
7980 * Flow matcher value.
7982 * Flow pattern to translate.
7984 * Item is inner pattern.
7987 flow_dv_translate_item_gtp(void *matcher, void *key,
7988 const struct rte_flow_item *item, int inner)
7990 const struct rte_flow_item_gtp *gtp_m = item->mask;
7991 const struct rte_flow_item_gtp *gtp_v = item->spec;
7994 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7996 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7997 uint16_t dport = RTE_GTPU_UDP_PORT;
8000 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8002 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8004 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8006 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8008 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
8009 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
8010 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
8015 gtp_m = &rte_flow_item_gtp_mask;
8016 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
8017 gtp_m->v_pt_rsv_flags);
8018 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
8019 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
8020 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
8021 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
8022 gtp_v->msg_type & gtp_m->msg_type);
8023 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
8024 rte_be_to_cpu_32(gtp_m->teid));
8025 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
8026 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
8030 * Add GTP PSC item to matcher.
8032 * @param[in, out] matcher
8034 * @param[in, out] key
8035 * Flow matcher value.
8037 * Flow pattern to translate.
8040 flow_dv_translate_item_gtp_psc(void *matcher, void *key,
8041 const struct rte_flow_item *item)
8043 const struct rte_flow_item_gtp_psc *gtp_psc_m = item->mask;
8044 const struct rte_flow_item_gtp_psc *gtp_psc_v = item->spec;
8045 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
8047 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
8053 uint8_t next_ext_header_type;
8058 /* Always set E-flag match on one, regardless of GTP item settings. */
8059 gtp_flags = MLX5_GET(fte_match_set_misc3, misc3_m, gtpu_msg_flags);
8060 gtp_flags |= MLX5_GTP_EXT_HEADER_FLAG;
8061 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags, gtp_flags);
8062 gtp_flags = MLX5_GET(fte_match_set_misc3, misc3_v, gtpu_msg_flags);
8063 gtp_flags |= MLX5_GTP_EXT_HEADER_FLAG;
8064 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags, gtp_flags);
8065 /*Set next extension header type. */
8068 dw_2.next_ext_header_type = 0xff;
8069 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_dw_2,
8070 rte_cpu_to_be_32(dw_2.w32));
8073 dw_2.next_ext_header_type = 0x85;
8074 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_dw_2,
8075 rte_cpu_to_be_32(dw_2.w32));
8087 /*Set extension header PDU type and Qos. */
8089 gtp_psc_m = &rte_flow_item_gtp_psc_mask;
8091 dw_0.type_flags = MLX5_GTP_PDU_TYPE_SHIFT(gtp_psc_m->pdu_type);
8092 dw_0.qfi = gtp_psc_m->qfi;
8093 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_first_ext_dw_0,
8094 rte_cpu_to_be_32(dw_0.w32));
8096 dw_0.type_flags = MLX5_GTP_PDU_TYPE_SHIFT(gtp_psc_v->pdu_type &
8097 gtp_psc_m->pdu_type);
8098 dw_0.qfi = gtp_psc_v->qfi & gtp_psc_m->qfi;
8099 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_first_ext_dw_0,
8100 rte_cpu_to_be_32(dw_0.w32));
8106 * Add eCPRI item to matcher and to the value.
8109 * The devich to configure through.
8110 * @param[in, out] matcher
8112 * @param[in, out] key
8113 * Flow matcher value.
8115 * Flow pattern to translate.
8116 * @param[in] samples
8117 * Sample IDs to be used in the matching.
8120 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
8121 void *key, const struct rte_flow_item *item)
8123 struct mlx5_priv *priv = dev->data->dev_private;
8124 const struct rte_flow_item_ecpri *ecpri_m = item->mask;
8125 const struct rte_flow_item_ecpri *ecpri_v = item->spec;
8126 struct rte_ecpri_common_hdr common;
8127 void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
8129 void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
8137 ecpri_m = &rte_flow_item_ecpri_mask;
8139 * Maximal four DW samples are supported in a single matching now.
8140 * Two are used now for a eCPRI matching:
8141 * 1. Type: one byte, mask should be 0x00ff0000 in network order
8142 * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
8145 if (!ecpri_m->hdr.common.u32)
8147 samples = priv->sh->fp[MLX5_FLEX_PARSER_ECPRI_0].ids;
8148 /* Need to take the whole DW as the mask to fill the entry. */
8149 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
8150 prog_sample_field_value_0);
8151 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
8152 prog_sample_field_value_0);
8153 /* Already big endian (network order) in the header. */
8154 *(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
8155 *(uint32_t *)dw_v = ecpri_v->hdr.common.u32 & ecpri_m->hdr.common.u32;
8156 /* Sample#0, used for matching type, offset 0. */
8157 MLX5_SET(fte_match_set_misc4, misc4_m,
8158 prog_sample_field_id_0, samples[0]);
8159 /* It makes no sense to set the sample ID in the mask field. */
8160 MLX5_SET(fte_match_set_misc4, misc4_v,
8161 prog_sample_field_id_0, samples[0]);
8163 * Checking if message body part needs to be matched.
8164 * Some wildcard rules only matching type field should be supported.
8166 if (ecpri_m->hdr.dummy[0]) {
8167 common.u32 = rte_be_to_cpu_32(ecpri_v->hdr.common.u32);
8168 switch (common.type) {
8169 case RTE_ECPRI_MSG_TYPE_IQ_DATA:
8170 case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
8171 case RTE_ECPRI_MSG_TYPE_DLY_MSR:
8172 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
8173 prog_sample_field_value_1);
8174 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
8175 prog_sample_field_value_1);
8176 *(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
8177 *(uint32_t *)dw_v = ecpri_v->hdr.dummy[0] &
8178 ecpri_m->hdr.dummy[0];
8179 /* Sample#1, to match message body, offset 4. */
8180 MLX5_SET(fte_match_set_misc4, misc4_m,
8181 prog_sample_field_id_1, samples[1]);
8182 MLX5_SET(fte_match_set_misc4, misc4_v,
8183 prog_sample_field_id_1, samples[1]);
8186 /* Others, do not match any sample ID. */
8192 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
8194 #define HEADER_IS_ZERO(match_criteria, headers) \
8195 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
8196 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
8199 * Calculate flow matcher enable bitmap.
8201 * @param match_criteria
8202 * Pointer to flow matcher criteria.
8205 * Bitmap of enabled fields.
8208 flow_dv_matcher_enable(uint32_t *match_criteria)
8210 uint8_t match_criteria_enable;
8212 match_criteria_enable =
8213 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
8214 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
8215 match_criteria_enable |=
8216 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
8217 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
8218 match_criteria_enable |=
8219 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
8220 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
8221 match_criteria_enable |=
8222 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
8223 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
8224 match_criteria_enable |=
8225 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
8226 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
8227 match_criteria_enable |=
8228 (!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
8229 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
8230 return match_criteria_enable;
8233 struct mlx5_hlist_entry *
8234 flow_dv_tbl_create_cb(struct mlx5_hlist *list, uint64_t key64, void *cb_ctx)
8236 struct mlx5_dev_ctx_shared *sh = list->ctx;
8237 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8238 struct rte_eth_dev *dev = ctx->dev;
8239 struct mlx5_flow_tbl_data_entry *tbl_data;
8240 struct mlx5_flow_tbl_tunnel_prm *tt_prm = ctx->data;
8241 struct rte_flow_error *error = ctx->error;
8242 union mlx5_flow_tbl_key key = { .v64 = key64 };
8243 struct mlx5_flow_tbl_resource *tbl;
8248 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
8250 rte_flow_error_set(error, ENOMEM,
8251 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8253 "cannot allocate flow table data entry");
8256 tbl_data->idx = idx;
8257 tbl_data->tunnel = tt_prm->tunnel;
8258 tbl_data->group_id = tt_prm->group_id;
8259 tbl_data->external = !!tt_prm->external;
8260 tbl_data->tunnel_offload = is_tunnel_offload_active(dev);
8261 tbl_data->is_egress = !!key.direction;
8262 tbl_data->is_transfer = !!key.domain;
8263 tbl_data->dummy = !!key.dummy;
8264 tbl_data->table_id = key.table_id;
8265 tbl = &tbl_data->tbl;
8267 return &tbl_data->entry;
8269 domain = sh->fdb_domain;
8270 else if (key.direction)
8271 domain = sh->tx_domain;
8273 domain = sh->rx_domain;
8274 ret = mlx5_flow_os_create_flow_tbl(domain, key.table_id, &tbl->obj);
8276 rte_flow_error_set(error, ENOMEM,
8277 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8278 NULL, "cannot create flow table object");
8279 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
8283 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
8284 (tbl->obj, &tbl_data->jump.action);
8286 rte_flow_error_set(error, ENOMEM,
8287 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8289 "cannot create flow jump action");
8290 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
8291 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
8295 MKSTR(matcher_name, "%s_%s_%u_matcher_cache",
8296 key.domain ? "FDB" : "NIC", key.direction ? "egress" : "ingress",
8298 mlx5_cache_list_init(&tbl_data->matchers, matcher_name, 0, sh,
8299 flow_dv_matcher_create_cb,
8300 flow_dv_matcher_match_cb,
8301 flow_dv_matcher_remove_cb);
8302 return &tbl_data->entry;
8306 flow_dv_tbl_match_cb(struct mlx5_hlist *list __rte_unused,
8307 struct mlx5_hlist_entry *entry, uint64_t key64,
8308 void *cb_ctx __rte_unused)
8310 struct mlx5_flow_tbl_data_entry *tbl_data =
8311 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
8312 union mlx5_flow_tbl_key key = { .v64 = key64 };
8314 return tbl_data->table_id != key.table_id ||
8315 tbl_data->dummy != key.dummy ||
8316 tbl_data->is_transfer != key.domain ||
8317 tbl_data->is_egress != key.direction;
8323 * @param[in, out] dev
8324 * Pointer to rte_eth_dev structure.
8325 * @param[in] table_id
8328 * Direction of the table.
8329 * @param[in] transfer
8330 * E-Switch or NIC flow.
8332 * Dummy entry for dv API.
8334 * pointer to error structure.
8337 * Returns tables resource based on the index, NULL in case of failed.
8339 struct mlx5_flow_tbl_resource *
8340 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
8341 uint32_t table_id, uint8_t egress,
8344 const struct mlx5_flow_tunnel *tunnel,
8345 uint32_t group_id, uint8_t dummy,
8346 struct rte_flow_error *error)
8348 struct mlx5_priv *priv = dev->data->dev_private;
8349 union mlx5_flow_tbl_key table_key = {
8351 .table_id = table_id,
8353 .domain = !!transfer,
8354 .direction = !!egress,
8357 struct mlx5_flow_tbl_tunnel_prm tt_prm = {
8359 .group_id = group_id,
8360 .external = external,
8362 struct mlx5_flow_cb_ctx ctx = {
8367 struct mlx5_hlist_entry *entry;
8368 struct mlx5_flow_tbl_data_entry *tbl_data;
8370 entry = mlx5_hlist_register(priv->sh->flow_tbls, table_key.v64, &ctx);
8372 rte_flow_error_set(error, ENOMEM,
8373 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8374 "cannot get table");
8377 DRV_LOG(DEBUG, "Table_id %u tunnel %u group %u registered.",
8378 table_id, tunnel ? tunnel->tunnel_id : 0, group_id);
8379 tbl_data = container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
8380 return &tbl_data->tbl;
8384 flow_dv_tbl_remove_cb(struct mlx5_hlist *list,
8385 struct mlx5_hlist_entry *entry)
8387 struct mlx5_dev_ctx_shared *sh = list->ctx;
8388 struct mlx5_flow_tbl_data_entry *tbl_data =
8389 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
8391 MLX5_ASSERT(entry && sh);
8392 if (tbl_data->jump.action)
8393 mlx5_flow_os_destroy_flow_action(tbl_data->jump.action);
8394 if (tbl_data->tbl.obj)
8395 mlx5_flow_os_destroy_flow_tbl(tbl_data->tbl.obj);
8396 if (tbl_data->tunnel_offload && tbl_data->external) {
8397 struct mlx5_hlist_entry *he;
8398 struct mlx5_hlist *tunnel_grp_hash;
8399 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8400 union tunnel_tbl_key tunnel_key = {
8401 .tunnel_id = tbl_data->tunnel ?
8402 tbl_data->tunnel->tunnel_id : 0,
8403 .group = tbl_data->group_id
8405 uint32_t table_id = tbl_data->table_id;
8407 tunnel_grp_hash = tbl_data->tunnel ?
8408 tbl_data->tunnel->groups :
8410 he = mlx5_hlist_lookup(tunnel_grp_hash, tunnel_key.val, NULL);
8412 mlx5_hlist_unregister(tunnel_grp_hash, he);
8414 "Table_id %u tunnel %u group %u released.",
8417 tbl_data->tunnel->tunnel_id : 0,
8418 tbl_data->group_id);
8420 mlx5_cache_list_destroy(&tbl_data->matchers);
8421 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], tbl_data->idx);
8425 * Release a flow table.
8428 * Pointer to device shared structure.
8430 * Table resource to be released.
8433 * Returns 0 if table was released, else return 1;
8436 flow_dv_tbl_resource_release(struct mlx5_dev_ctx_shared *sh,
8437 struct mlx5_flow_tbl_resource *tbl)
8439 struct mlx5_flow_tbl_data_entry *tbl_data =
8440 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
8444 return mlx5_hlist_unregister(sh->flow_tbls, &tbl_data->entry);
8448 flow_dv_matcher_match_cb(struct mlx5_cache_list *list __rte_unused,
8449 struct mlx5_cache_entry *entry, void *cb_ctx)
8451 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8452 struct mlx5_flow_dv_matcher *ref = ctx->data;
8453 struct mlx5_flow_dv_matcher *cur = container_of(entry, typeof(*cur),
8456 return cur->crc != ref->crc ||
8457 cur->priority != ref->priority ||
8458 memcmp((const void *)cur->mask.buf,
8459 (const void *)ref->mask.buf, ref->mask.size);
8462 struct mlx5_cache_entry *
8463 flow_dv_matcher_create_cb(struct mlx5_cache_list *list,
8464 struct mlx5_cache_entry *entry __rte_unused,
8467 struct mlx5_dev_ctx_shared *sh = list->ctx;
8468 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8469 struct mlx5_flow_dv_matcher *ref = ctx->data;
8470 struct mlx5_flow_dv_matcher *cache;
8471 struct mlx5dv_flow_matcher_attr dv_attr = {
8472 .type = IBV_FLOW_ATTR_NORMAL,
8473 .match_mask = (void *)&ref->mask,
8475 struct mlx5_flow_tbl_data_entry *tbl = container_of(ref->tbl,
8479 cache = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*cache), 0, SOCKET_ID_ANY);
8481 rte_flow_error_set(ctx->error, ENOMEM,
8482 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8483 "cannot create matcher");
8487 dv_attr.match_criteria_enable =
8488 flow_dv_matcher_enable(cache->mask.buf);
8489 dv_attr.priority = ref->priority;
8491 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
8492 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->tbl.obj,
8493 &cache->matcher_object);
8496 rte_flow_error_set(ctx->error, ENOMEM,
8497 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8498 "cannot create matcher");
8501 return &cache->entry;
8505 * Register the flow matcher.
8507 * @param[in, out] dev
8508 * Pointer to rte_eth_dev structure.
8509 * @param[in, out] matcher
8510 * Pointer to flow matcher.
8511 * @param[in, out] key
8512 * Pointer to flow table key.
8513 * @parm[in, out] dev_flow
8514 * Pointer to the dev_flow.
8516 * pointer to error structure.
8519 * 0 on success otherwise -errno and errno is set.
8522 flow_dv_matcher_register(struct rte_eth_dev *dev,
8523 struct mlx5_flow_dv_matcher *ref,
8524 union mlx5_flow_tbl_key *key,
8525 struct mlx5_flow *dev_flow,
8526 const struct mlx5_flow_tunnel *tunnel,
8528 struct rte_flow_error *error)
8530 struct mlx5_cache_entry *entry;
8531 struct mlx5_flow_dv_matcher *cache;
8532 struct mlx5_flow_tbl_resource *tbl;
8533 struct mlx5_flow_tbl_data_entry *tbl_data;
8534 struct mlx5_flow_cb_ctx ctx = {
8540 * tunnel offload API requires this registration for cases when
8541 * tunnel match rule was inserted before tunnel set rule.
8543 tbl = flow_dv_tbl_resource_get(dev, key->table_id,
8544 key->direction, key->domain,
8545 dev_flow->external, tunnel,
8546 group_id, 0, error);
8548 return -rte_errno; /* No need to refill the error info */
8549 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
8551 entry = mlx5_cache_register(&tbl_data->matchers, &ctx);
8553 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
8554 return rte_flow_error_set(error, ENOMEM,
8555 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8556 "cannot allocate ref memory");
8558 cache = container_of(entry, typeof(*cache), entry);
8559 dev_flow->handle->dvh.matcher = cache;
8563 struct mlx5_hlist_entry *
8564 flow_dv_tag_create_cb(struct mlx5_hlist *list, uint64_t key, void *ctx)
8566 struct mlx5_dev_ctx_shared *sh = list->ctx;
8567 struct rte_flow_error *error = ctx;
8568 struct mlx5_flow_dv_tag_resource *entry;
8572 entry = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG], &idx);
8574 rte_flow_error_set(error, ENOMEM,
8575 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8576 "cannot allocate resource memory");
8580 entry->tag_id = key;
8581 ret = mlx5_flow_os_create_flow_action_tag(key,
8584 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], idx);
8585 rte_flow_error_set(error, ENOMEM,
8586 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8587 NULL, "cannot create action");
8590 return &entry->entry;
8594 flow_dv_tag_match_cb(struct mlx5_hlist *list __rte_unused,
8595 struct mlx5_hlist_entry *entry, uint64_t key,
8596 void *cb_ctx __rte_unused)
8598 struct mlx5_flow_dv_tag_resource *tag =
8599 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
8601 return key != tag->tag_id;
8605 * Find existing tag resource or create and register a new one.
8607 * @param dev[in, out]
8608 * Pointer to rte_eth_dev structure.
8609 * @param[in, out] tag_be24
8610 * Tag value in big endian then R-shift 8.
8611 * @parm[in, out] dev_flow
8612 * Pointer to the dev_flow.
8614 * pointer to error structure.
8617 * 0 on success otherwise -errno and errno is set.
8620 flow_dv_tag_resource_register
8621 (struct rte_eth_dev *dev,
8623 struct mlx5_flow *dev_flow,
8624 struct rte_flow_error *error)
8626 struct mlx5_priv *priv = dev->data->dev_private;
8627 struct mlx5_flow_dv_tag_resource *cache_resource;
8628 struct mlx5_hlist_entry *entry;
8630 entry = mlx5_hlist_register(priv->sh->tag_table, tag_be24, error);
8632 cache_resource = container_of
8633 (entry, struct mlx5_flow_dv_tag_resource, entry);
8634 dev_flow->handle->dvh.rix_tag = cache_resource->idx;
8635 dev_flow->dv.tag_resource = cache_resource;
8642 flow_dv_tag_remove_cb(struct mlx5_hlist *list,
8643 struct mlx5_hlist_entry *entry)
8645 struct mlx5_dev_ctx_shared *sh = list->ctx;
8646 struct mlx5_flow_dv_tag_resource *tag =
8647 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
8649 MLX5_ASSERT(tag && sh && tag->action);
8650 claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
8651 DRV_LOG(DEBUG, "Tag %p: removed.", (void *)tag);
8652 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], tag->idx);
8659 * Pointer to Ethernet device.
8664 * 1 while a reference on it exists, 0 when freed.
8667 flow_dv_tag_release(struct rte_eth_dev *dev,
8670 struct mlx5_priv *priv = dev->data->dev_private;
8671 struct mlx5_flow_dv_tag_resource *tag;
8673 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
8676 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
8677 dev->data->port_id, (void *)tag, tag->entry.ref_cnt);
8678 return mlx5_hlist_unregister(priv->sh->tag_table, &tag->entry);
8682 * Translate port ID action to vport.
8685 * Pointer to rte_eth_dev structure.
8687 * Pointer to the port ID action.
8688 * @param[out] dst_port_id
8689 * The target port ID.
8691 * Pointer to the error structure.
8694 * 0 on success, a negative errno value otherwise and rte_errno is set.
8697 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
8698 const struct rte_flow_action *action,
8699 uint32_t *dst_port_id,
8700 struct rte_flow_error *error)
8703 struct mlx5_priv *priv;
8704 const struct rte_flow_action_port_id *conf =
8705 (const struct rte_flow_action_port_id *)action->conf;
8707 port = conf->original ? dev->data->port_id : conf->id;
8708 priv = mlx5_port_to_eswitch_info(port, false);
8710 return rte_flow_error_set(error, -rte_errno,
8711 RTE_FLOW_ERROR_TYPE_ACTION,
8713 "No eswitch info was found for port");
8714 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
8716 * This parameter is transferred to
8717 * mlx5dv_dr_action_create_dest_ib_port().
8719 *dst_port_id = priv->dev_port;
8722 * Legacy mode, no LAG configurations is supported.
8723 * This parameter is transferred to
8724 * mlx5dv_dr_action_create_dest_vport().
8726 *dst_port_id = priv->vport_id;
8732 * Create a counter with aging configuration.
8735 * Pointer to rte_eth_dev structure.
8737 * Pointer to the counter action configuration.
8739 * Pointer to the aging action configuration.
8742 * Index to flow counter on success, 0 otherwise.
8745 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
8746 struct mlx5_flow *dev_flow,
8747 const struct rte_flow_action_count *count,
8748 const struct rte_flow_action_age *age)
8751 struct mlx5_age_param *age_param;
8753 if (count && count->shared)
8754 counter = flow_dv_counter_get_shared(dev, count->id);
8756 counter = flow_dv_counter_alloc(dev, !!age);
8757 if (!counter || age == NULL)
8759 age_param = flow_dv_counter_idx_get_age(dev, counter);
8760 age_param->context = age->context ? age->context :
8761 (void *)(uintptr_t)(dev_flow->flow_idx);
8762 age_param->timeout = age->timeout;
8763 age_param->port_id = dev->data->port_id;
8764 __atomic_store_n(&age_param->sec_since_last_hit, 0, __ATOMIC_RELAXED);
8765 __atomic_store_n(&age_param->state, AGE_CANDIDATE, __ATOMIC_RELAXED);
8770 * Add Tx queue matcher
8773 * Pointer to the dev struct.
8774 * @param[in, out] matcher
8776 * @param[in, out] key
8777 * Flow matcher value.
8779 * Flow pattern to translate.
8781 * Item is inner pattern.
8784 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
8785 void *matcher, void *key,
8786 const struct rte_flow_item *item)
8788 const struct mlx5_rte_flow_item_tx_queue *queue_m;
8789 const struct mlx5_rte_flow_item_tx_queue *queue_v;
8791 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8793 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8794 struct mlx5_txq_ctrl *txq;
8798 queue_m = (const void *)item->mask;
8801 queue_v = (const void *)item->spec;
8804 txq = mlx5_txq_get(dev, queue_v->queue);
8807 queue = txq->obj->sq->id;
8808 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
8809 MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
8810 queue & queue_m->queue);
8811 mlx5_txq_release(dev, queue_v->queue);
8815 * Set the hash fields according to the @p flow information.
8817 * @param[in] dev_flow
8818 * Pointer to the mlx5_flow.
8819 * @param[in] rss_desc
8820 * Pointer to the mlx5_flow_rss_desc.
8823 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
8824 struct mlx5_flow_rss_desc *rss_desc)
8826 uint64_t items = dev_flow->handle->layers;
8828 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
8830 dev_flow->hash_fields = 0;
8831 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
8832 if (rss_desc->level >= 2) {
8833 dev_flow->hash_fields |= IBV_RX_HASH_INNER;
8837 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
8838 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
8839 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
8840 if (rss_types & ETH_RSS_L3_SRC_ONLY)
8841 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
8842 else if (rss_types & ETH_RSS_L3_DST_ONLY)
8843 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
8845 dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
8847 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
8848 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
8849 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
8850 if (rss_types & ETH_RSS_L3_SRC_ONLY)
8851 dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
8852 else if (rss_types & ETH_RSS_L3_DST_ONLY)
8853 dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
8855 dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
8858 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
8859 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
8860 if (rss_types & ETH_RSS_UDP) {
8861 if (rss_types & ETH_RSS_L4_SRC_ONLY)
8862 dev_flow->hash_fields |=
8863 IBV_RX_HASH_SRC_PORT_UDP;
8864 else if (rss_types & ETH_RSS_L4_DST_ONLY)
8865 dev_flow->hash_fields |=
8866 IBV_RX_HASH_DST_PORT_UDP;
8868 dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
8870 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
8871 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
8872 if (rss_types & ETH_RSS_TCP) {
8873 if (rss_types & ETH_RSS_L4_SRC_ONLY)
8874 dev_flow->hash_fields |=
8875 IBV_RX_HASH_SRC_PORT_TCP;
8876 else if (rss_types & ETH_RSS_L4_DST_ONLY)
8877 dev_flow->hash_fields |=
8878 IBV_RX_HASH_DST_PORT_TCP;
8880 dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
8886 * Prepare an Rx Hash queue.
8889 * Pointer to Ethernet device.
8890 * @param[in] dev_flow
8891 * Pointer to the mlx5_flow.
8892 * @param[in] rss_desc
8893 * Pointer to the mlx5_flow_rss_desc.
8894 * @param[out] hrxq_idx
8895 * Hash Rx queue index.
8898 * The Verbs/DevX object initialised, NULL otherwise and rte_errno is set.
8900 static struct mlx5_hrxq *
8901 flow_dv_hrxq_prepare(struct rte_eth_dev *dev,
8902 struct mlx5_flow *dev_flow,
8903 struct mlx5_flow_rss_desc *rss_desc,
8906 struct mlx5_priv *priv = dev->data->dev_private;
8907 struct mlx5_flow_handle *dh = dev_flow->handle;
8908 struct mlx5_hrxq *hrxq;
8910 MLX5_ASSERT(rss_desc->queue_num);
8911 rss_desc->key_len = MLX5_RSS_HASH_KEY_LEN;
8912 rss_desc->hash_fields = dev_flow->hash_fields;
8913 rss_desc->tunnel = !!(dh->layers & MLX5_FLOW_LAYER_TUNNEL);
8914 rss_desc->shared_rss = 0;
8915 *hrxq_idx = mlx5_hrxq_get(dev, rss_desc);
8918 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8924 * Release sample sub action resource.
8926 * @param[in, out] dev
8927 * Pointer to rte_eth_dev structure.
8928 * @param[in] act_res
8929 * Pointer to sample sub action resource.
8932 flow_dv_sample_sub_actions_release(struct rte_eth_dev *dev,
8933 struct mlx5_flow_sub_actions_idx *act_res)
8935 if (act_res->rix_hrxq) {
8936 mlx5_hrxq_release(dev, act_res->rix_hrxq);
8937 act_res->rix_hrxq = 0;
8939 if (act_res->rix_encap_decap) {
8940 flow_dv_encap_decap_resource_release(dev,
8941 act_res->rix_encap_decap);
8942 act_res->rix_encap_decap = 0;
8944 if (act_res->rix_port_id_action) {
8945 flow_dv_port_id_action_resource_release(dev,
8946 act_res->rix_port_id_action);
8947 act_res->rix_port_id_action = 0;
8949 if (act_res->rix_tag) {
8950 flow_dv_tag_release(dev, act_res->rix_tag);
8951 act_res->rix_tag = 0;
8954 flow_dv_counter_free(dev, act_res->cnt);
8960 flow_dv_sample_match_cb(struct mlx5_cache_list *list __rte_unused,
8961 struct mlx5_cache_entry *entry, void *cb_ctx)
8963 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8964 struct rte_eth_dev *dev = ctx->dev;
8965 struct mlx5_flow_dv_sample_resource *resource = ctx->data;
8966 struct mlx5_flow_dv_sample_resource *cache_resource =
8967 container_of(entry, typeof(*cache_resource), entry);
8969 if (resource->ratio == cache_resource->ratio &&
8970 resource->ft_type == cache_resource->ft_type &&
8971 resource->ft_id == cache_resource->ft_id &&
8972 resource->set_action == cache_resource->set_action &&
8973 !memcmp((void *)&resource->sample_act,
8974 (void *)&cache_resource->sample_act,
8975 sizeof(struct mlx5_flow_sub_actions_list))) {
8977 * Existing sample action should release the prepared
8978 * sub-actions reference counter.
8980 flow_dv_sample_sub_actions_release(dev,
8981 &resource->sample_idx);
8987 struct mlx5_cache_entry *
8988 flow_dv_sample_create_cb(struct mlx5_cache_list *list __rte_unused,
8989 struct mlx5_cache_entry *entry __rte_unused,
8992 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8993 struct rte_eth_dev *dev = ctx->dev;
8994 struct mlx5_flow_dv_sample_resource *resource = ctx->data;
8995 void **sample_dv_actions = resource->sub_actions;
8996 struct mlx5_flow_dv_sample_resource *cache_resource;
8997 struct mlx5dv_dr_flow_sampler_attr sampler_attr;
8998 struct mlx5_priv *priv = dev->data->dev_private;
8999 struct mlx5_dev_ctx_shared *sh = priv->sh;
9000 struct mlx5_flow_tbl_resource *tbl;
9002 const uint32_t next_ft_step = 1;
9003 uint32_t next_ft_id = resource->ft_id + next_ft_step;
9004 uint8_t is_egress = 0;
9005 uint8_t is_transfer = 0;
9006 struct rte_flow_error *error = ctx->error;
9008 /* Register new sample resource. */
9009 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE], &idx);
9010 if (!cache_resource) {
9011 rte_flow_error_set(error, ENOMEM,
9012 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9014 "cannot allocate resource memory");
9017 *cache_resource = *resource;
9018 /* Create normal path table level */
9019 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
9021 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
9023 tbl = flow_dv_tbl_resource_get(dev, next_ft_id,
9024 is_egress, is_transfer,
9025 true, NULL, 0, 0, error);
9027 rte_flow_error_set(error, ENOMEM,
9028 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9030 "fail to create normal path table "
9036 cache_resource->normal_path_tbl = tbl;
9037 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
9038 ret = mlx5_flow_os_create_flow_action_default_miss
9039 (&cache_resource->default_miss);
9041 rte_flow_error_set(error, ENOMEM,
9042 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9044 "cannot create default miss "
9048 sample_dv_actions[resource->sample_act.actions_num++] =
9049 cache_resource->default_miss;
9051 /* Create a DR sample action */
9052 sampler_attr.sample_ratio = cache_resource->ratio;
9053 sampler_attr.default_next_table = tbl->obj;
9054 sampler_attr.num_sample_actions = resource->sample_act.actions_num;
9055 sampler_attr.sample_actions = (struct mlx5dv_dr_action **)
9056 &sample_dv_actions[0];
9057 sampler_attr.action = cache_resource->set_action;
9058 if (mlx5_os_flow_dr_create_flow_action_sampler
9059 (&sampler_attr, &cache_resource->verbs_action)) {
9060 rte_flow_error_set(error, ENOMEM,
9061 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9062 NULL, "cannot create sample action");
9065 cache_resource->idx = idx;
9066 cache_resource->dev = dev;
9067 return &cache_resource->entry;
9069 if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB &&
9070 cache_resource->default_miss)
9071 claim_zero(mlx5_flow_os_destroy_flow_action
9072 (cache_resource->default_miss));
9074 flow_dv_sample_sub_actions_release(dev,
9075 &cache_resource->sample_idx);
9076 if (cache_resource->normal_path_tbl)
9077 flow_dv_tbl_resource_release(MLX5_SH(dev),
9078 cache_resource->normal_path_tbl);
9079 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_SAMPLE], idx);
9085 * Find existing sample resource or create and register a new one.
9087 * @param[in, out] dev
9088 * Pointer to rte_eth_dev structure.
9089 * @param[in] resource
9090 * Pointer to sample resource.
9091 * @parm[in, out] dev_flow
9092 * Pointer to the dev_flow.
9094 * pointer to error structure.
9097 * 0 on success otherwise -errno and errno is set.
9100 flow_dv_sample_resource_register(struct rte_eth_dev *dev,
9101 struct mlx5_flow_dv_sample_resource *resource,
9102 struct mlx5_flow *dev_flow,
9103 struct rte_flow_error *error)
9105 struct mlx5_flow_dv_sample_resource *cache_resource;
9106 struct mlx5_cache_entry *entry;
9107 struct mlx5_priv *priv = dev->data->dev_private;
9108 struct mlx5_flow_cb_ctx ctx = {
9114 entry = mlx5_cache_register(&priv->sh->sample_action_list, &ctx);
9117 cache_resource = container_of(entry, typeof(*cache_resource), entry);
9118 dev_flow->handle->dvh.rix_sample = cache_resource->idx;
9119 dev_flow->dv.sample_res = cache_resource;
9124 flow_dv_dest_array_match_cb(struct mlx5_cache_list *list __rte_unused,
9125 struct mlx5_cache_entry *entry, void *cb_ctx)
9127 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9128 struct mlx5_flow_dv_dest_array_resource *resource = ctx->data;
9129 struct rte_eth_dev *dev = ctx->dev;
9130 struct mlx5_flow_dv_dest_array_resource *cache_resource =
9131 container_of(entry, typeof(*cache_resource), entry);
9134 if (resource->num_of_dest == cache_resource->num_of_dest &&
9135 resource->ft_type == cache_resource->ft_type &&
9136 !memcmp((void *)cache_resource->sample_act,
9137 (void *)resource->sample_act,
9138 (resource->num_of_dest *
9139 sizeof(struct mlx5_flow_sub_actions_list)))) {
9141 * Existing sample action should release the prepared
9142 * sub-actions reference counter.
9144 for (idx = 0; idx < resource->num_of_dest; idx++)
9145 flow_dv_sample_sub_actions_release(dev,
9146 &resource->sample_idx[idx]);
9152 struct mlx5_cache_entry *
9153 flow_dv_dest_array_create_cb(struct mlx5_cache_list *list __rte_unused,
9154 struct mlx5_cache_entry *entry __rte_unused,
9157 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9158 struct rte_eth_dev *dev = ctx->dev;
9159 struct mlx5_flow_dv_dest_array_resource *cache_resource;
9160 struct mlx5_flow_dv_dest_array_resource *resource = ctx->data;
9161 struct mlx5dv_dr_action_dest_attr *dest_attr[MLX5_MAX_DEST_NUM] = { 0 };
9162 struct mlx5dv_dr_action_dest_reformat dest_reformat[MLX5_MAX_DEST_NUM];
9163 struct mlx5_priv *priv = dev->data->dev_private;
9164 struct mlx5_dev_ctx_shared *sh = priv->sh;
9165 struct mlx5_flow_sub_actions_list *sample_act;
9166 struct mlx5dv_dr_domain *domain;
9167 uint32_t idx = 0, res_idx = 0;
9168 struct rte_flow_error *error = ctx->error;
9171 /* Register new destination array resource. */
9172 cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
9174 if (!cache_resource) {
9175 rte_flow_error_set(error, ENOMEM,
9176 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9178 "cannot allocate resource memory");
9181 *cache_resource = *resource;
9182 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
9183 domain = sh->fdb_domain;
9184 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
9185 domain = sh->rx_domain;
9187 domain = sh->tx_domain;
9188 for (idx = 0; idx < resource->num_of_dest; idx++) {
9189 dest_attr[idx] = (struct mlx5dv_dr_action_dest_attr *)
9190 mlx5_malloc(MLX5_MEM_ZERO,
9191 sizeof(struct mlx5dv_dr_action_dest_attr),
9193 if (!dest_attr[idx]) {
9194 rte_flow_error_set(error, ENOMEM,
9195 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9197 "cannot allocate resource memory");
9200 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST;
9201 sample_act = &resource->sample_act[idx];
9202 if (sample_act->action_flags == MLX5_FLOW_ACTION_QUEUE) {
9203 dest_attr[idx]->dest = sample_act->dr_queue_action;
9204 } else if (sample_act->action_flags ==
9205 (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_ENCAP)) {
9206 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST_REFORMAT;
9207 dest_attr[idx]->dest_reformat = &dest_reformat[idx];
9208 dest_attr[idx]->dest_reformat->reformat =
9209 sample_act->dr_encap_action;
9210 dest_attr[idx]->dest_reformat->dest =
9211 sample_act->dr_port_id_action;
9212 } else if (sample_act->action_flags ==
9213 MLX5_FLOW_ACTION_PORT_ID) {
9214 dest_attr[idx]->dest = sample_act->dr_port_id_action;
9217 /* create a dest array actioin */
9218 ret = mlx5_os_flow_dr_create_flow_action_dest_array
9220 cache_resource->num_of_dest,
9222 &cache_resource->action);
9224 rte_flow_error_set(error, ENOMEM,
9225 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9227 "cannot create destination array action");
9230 cache_resource->idx = res_idx;
9231 cache_resource->dev = dev;
9232 for (idx = 0; idx < resource->num_of_dest; idx++)
9233 mlx5_free(dest_attr[idx]);
9234 return &cache_resource->entry;
9236 for (idx = 0; idx < resource->num_of_dest; idx++) {
9237 struct mlx5_flow_sub_actions_idx *act_res =
9238 &cache_resource->sample_idx[idx];
9239 if (act_res->rix_hrxq &&
9240 !mlx5_hrxq_release(dev,
9242 act_res->rix_hrxq = 0;
9243 if (act_res->rix_encap_decap &&
9244 !flow_dv_encap_decap_resource_release(dev,
9245 act_res->rix_encap_decap))
9246 act_res->rix_encap_decap = 0;
9247 if (act_res->rix_port_id_action &&
9248 !flow_dv_port_id_action_resource_release(dev,
9249 act_res->rix_port_id_action))
9250 act_res->rix_port_id_action = 0;
9252 mlx5_free(dest_attr[idx]);
9255 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DEST_ARRAY], res_idx);
9260 * Find existing destination array resource or create and register a new one.
9262 * @param[in, out] dev
9263 * Pointer to rte_eth_dev structure.
9264 * @param[in] resource
9265 * Pointer to destination array resource.
9266 * @parm[in, out] dev_flow
9267 * Pointer to the dev_flow.
9269 * pointer to error structure.
9272 * 0 on success otherwise -errno and errno is set.
9275 flow_dv_dest_array_resource_register(struct rte_eth_dev *dev,
9276 struct mlx5_flow_dv_dest_array_resource *resource,
9277 struct mlx5_flow *dev_flow,
9278 struct rte_flow_error *error)
9280 struct mlx5_flow_dv_dest_array_resource *cache_resource;
9281 struct mlx5_priv *priv = dev->data->dev_private;
9282 struct mlx5_cache_entry *entry;
9283 struct mlx5_flow_cb_ctx ctx = {
9289 entry = mlx5_cache_register(&priv->sh->dest_array_list, &ctx);
9292 cache_resource = container_of(entry, typeof(*cache_resource), entry);
9293 dev_flow->handle->dvh.rix_dest_array = cache_resource->idx;
9294 dev_flow->dv.dest_array_res = cache_resource;
9299 * Convert Sample action to DV specification.
9302 * Pointer to rte_eth_dev structure.
9304 * Pointer to action structure.
9305 * @param[in, out] dev_flow
9306 * Pointer to the mlx5_flow.
9308 * Pointer to the flow attributes.
9309 * @param[in, out] num_of_dest
9310 * Pointer to the num of destination.
9311 * @param[in, out] sample_actions
9312 * Pointer to sample actions list.
9313 * @param[in, out] res
9314 * Pointer to sample resource.
9316 * Pointer to the error structure.
9319 * 0 on success, a negative errno value otherwise and rte_errno is set.
9322 flow_dv_translate_action_sample(struct rte_eth_dev *dev,
9323 const struct rte_flow_action *action,
9324 struct mlx5_flow *dev_flow,
9325 const struct rte_flow_attr *attr,
9326 uint32_t *num_of_dest,
9327 void **sample_actions,
9328 struct mlx5_flow_dv_sample_resource *res,
9329 struct rte_flow_error *error)
9331 struct mlx5_priv *priv = dev->data->dev_private;
9332 const struct rte_flow_action_sample *sample_action;
9333 const struct rte_flow_action *sub_actions;
9334 const struct rte_flow_action_queue *queue;
9335 struct mlx5_flow_sub_actions_list *sample_act;
9336 struct mlx5_flow_sub_actions_idx *sample_idx;
9337 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
9338 struct mlx5_flow_rss_desc *rss_desc;
9339 uint64_t action_flags = 0;
9342 rss_desc = &wks->rss_desc;
9343 sample_act = &res->sample_act;
9344 sample_idx = &res->sample_idx;
9345 sample_action = (const struct rte_flow_action_sample *)action->conf;
9346 res->ratio = sample_action->ratio;
9347 sub_actions = sample_action->actions;
9348 for (; sub_actions->type != RTE_FLOW_ACTION_TYPE_END; sub_actions++) {
9349 int type = sub_actions->type;
9350 uint32_t pre_rix = 0;
9353 case RTE_FLOW_ACTION_TYPE_QUEUE:
9355 struct mlx5_hrxq *hrxq;
9358 queue = sub_actions->conf;
9359 rss_desc->queue_num = 1;
9360 rss_desc->queue[0] = queue->index;
9361 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
9362 rss_desc, &hrxq_idx);
9364 return rte_flow_error_set
9366 RTE_FLOW_ERROR_TYPE_ACTION,
9368 "cannot create fate queue");
9369 sample_act->dr_queue_action = hrxq->action;
9370 sample_idx->rix_hrxq = hrxq_idx;
9371 sample_actions[sample_act->actions_num++] =
9374 action_flags |= MLX5_FLOW_ACTION_QUEUE;
9375 if (action_flags & MLX5_FLOW_ACTION_MARK)
9376 dev_flow->handle->rix_hrxq = hrxq_idx;
9377 dev_flow->handle->fate_action =
9378 MLX5_FLOW_FATE_QUEUE;
9381 case RTE_FLOW_ACTION_TYPE_MARK:
9383 uint32_t tag_be = mlx5_flow_mark_set
9384 (((const struct rte_flow_action_mark *)
9385 (sub_actions->conf))->id);
9387 dev_flow->handle->mark = 1;
9388 pre_rix = dev_flow->handle->dvh.rix_tag;
9389 /* Save the mark resource before sample */
9390 pre_r = dev_flow->dv.tag_resource;
9391 if (flow_dv_tag_resource_register(dev, tag_be,
9394 MLX5_ASSERT(dev_flow->dv.tag_resource);
9395 sample_act->dr_tag_action =
9396 dev_flow->dv.tag_resource->action;
9397 sample_idx->rix_tag =
9398 dev_flow->handle->dvh.rix_tag;
9399 sample_actions[sample_act->actions_num++] =
9400 sample_act->dr_tag_action;
9401 /* Recover the mark resource after sample */
9402 dev_flow->dv.tag_resource = pre_r;
9403 dev_flow->handle->dvh.rix_tag = pre_rix;
9404 action_flags |= MLX5_FLOW_ACTION_MARK;
9407 case RTE_FLOW_ACTION_TYPE_COUNT:
9411 counter = flow_dv_translate_create_counter(dev,
9412 dev_flow, sub_actions->conf, 0);
9414 return rte_flow_error_set
9416 RTE_FLOW_ERROR_TYPE_ACTION,
9418 "cannot create counter"
9420 sample_idx->cnt = counter;
9421 sample_act->dr_cnt_action =
9422 (flow_dv_counter_get_by_idx(dev,
9423 counter, NULL))->action;
9424 sample_actions[sample_act->actions_num++] =
9425 sample_act->dr_cnt_action;
9426 action_flags |= MLX5_FLOW_ACTION_COUNT;
9429 case RTE_FLOW_ACTION_TYPE_PORT_ID:
9431 struct mlx5_flow_dv_port_id_action_resource
9433 uint32_t port_id = 0;
9435 memset(&port_id_resource, 0, sizeof(port_id_resource));
9436 /* Save the port id resource before sample */
9437 pre_rix = dev_flow->handle->rix_port_id_action;
9438 pre_r = dev_flow->dv.port_id_action;
9439 if (flow_dv_translate_action_port_id(dev, sub_actions,
9442 port_id_resource.port_id = port_id;
9443 if (flow_dv_port_id_action_resource_register
9444 (dev, &port_id_resource, dev_flow, error))
9446 sample_act->dr_port_id_action =
9447 dev_flow->dv.port_id_action->action;
9448 sample_idx->rix_port_id_action =
9449 dev_flow->handle->rix_port_id_action;
9450 sample_actions[sample_act->actions_num++] =
9451 sample_act->dr_port_id_action;
9452 /* Recover the port id resource after sample */
9453 dev_flow->dv.port_id_action = pre_r;
9454 dev_flow->handle->rix_port_id_action = pre_rix;
9456 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9459 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
9460 /* Save the encap resource before sample */
9461 pre_rix = dev_flow->handle->dvh.rix_encap_decap;
9462 pre_r = dev_flow->dv.encap_decap;
9463 if (flow_dv_create_action_l2_encap(dev, sub_actions,
9468 sample_act->dr_encap_action =
9469 dev_flow->dv.encap_decap->action;
9470 sample_idx->rix_encap_decap =
9471 dev_flow->handle->dvh.rix_encap_decap;
9472 sample_actions[sample_act->actions_num++] =
9473 sample_act->dr_encap_action;
9474 /* Recover the encap resource after sample */
9475 dev_flow->dv.encap_decap = pre_r;
9476 dev_flow->handle->dvh.rix_encap_decap = pre_rix;
9477 action_flags |= MLX5_FLOW_ACTION_ENCAP;
9480 return rte_flow_error_set(error, EINVAL,
9481 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9483 "Not support for sampler action");
9486 sample_act->action_flags = action_flags;
9487 res->ft_id = dev_flow->dv.group;
9488 if (attr->transfer) {
9490 uint32_t action_in[MLX5_ST_SZ_DW(set_action_in)];
9491 uint64_t set_action;
9492 } action_ctx = { .set_action = 0 };
9494 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
9495 MLX5_SET(set_action_in, action_ctx.action_in, action_type,
9496 MLX5_MODIFICATION_TYPE_SET);
9497 MLX5_SET(set_action_in, action_ctx.action_in, field,
9498 MLX5_MODI_META_REG_C_0);
9499 MLX5_SET(set_action_in, action_ctx.action_in, data,
9500 priv->vport_meta_tag);
9501 res->set_action = action_ctx.set_action;
9502 } else if (attr->ingress) {
9503 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9505 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_TX;
9511 * Convert Sample action to DV specification.
9514 * Pointer to rte_eth_dev structure.
9515 * @param[in, out] dev_flow
9516 * Pointer to the mlx5_flow.
9517 * @param[in] num_of_dest
9518 * The num of destination.
9519 * @param[in, out] res
9520 * Pointer to sample resource.
9521 * @param[in, out] mdest_res
9522 * Pointer to destination array resource.
9523 * @param[in] sample_actions
9524 * Pointer to sample path actions list.
9525 * @param[in] action_flags
9526 * Holds the actions detected until now.
9528 * Pointer to the error structure.
9531 * 0 on success, a negative errno value otherwise and rte_errno is set.
9534 flow_dv_create_action_sample(struct rte_eth_dev *dev,
9535 struct mlx5_flow *dev_flow,
9536 uint32_t num_of_dest,
9537 struct mlx5_flow_dv_sample_resource *res,
9538 struct mlx5_flow_dv_dest_array_resource *mdest_res,
9539 void **sample_actions,
9540 uint64_t action_flags,
9541 struct rte_flow_error *error)
9543 /* update normal path action resource into last index of array */
9544 uint32_t dest_index = MLX5_MAX_DEST_NUM - 1;
9545 struct mlx5_flow_sub_actions_list *sample_act =
9546 &mdest_res->sample_act[dest_index];
9547 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
9548 struct mlx5_flow_rss_desc *rss_desc;
9549 uint32_t normal_idx = 0;
9550 struct mlx5_hrxq *hrxq;
9554 rss_desc = &wks->rss_desc;
9555 if (num_of_dest > 1) {
9556 if (sample_act->action_flags & MLX5_FLOW_ACTION_QUEUE) {
9557 /* Handle QP action for mirroring */
9558 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
9559 rss_desc, &hrxq_idx);
9561 return rte_flow_error_set
9563 RTE_FLOW_ERROR_TYPE_ACTION,
9565 "cannot create rx queue");
9567 mdest_res->sample_idx[dest_index].rix_hrxq = hrxq_idx;
9568 sample_act->dr_queue_action = hrxq->action;
9569 if (action_flags & MLX5_FLOW_ACTION_MARK)
9570 dev_flow->handle->rix_hrxq = hrxq_idx;
9571 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9573 if (sample_act->action_flags & MLX5_FLOW_ACTION_ENCAP) {
9575 mdest_res->sample_idx[dest_index].rix_encap_decap =
9576 dev_flow->handle->dvh.rix_encap_decap;
9577 sample_act->dr_encap_action =
9578 dev_flow->dv.encap_decap->action;
9580 if (sample_act->action_flags & MLX5_FLOW_ACTION_PORT_ID) {
9582 mdest_res->sample_idx[dest_index].rix_port_id_action =
9583 dev_flow->handle->rix_port_id_action;
9584 sample_act->dr_port_id_action =
9585 dev_flow->dv.port_id_action->action;
9587 sample_act->actions_num = normal_idx;
9588 /* update sample action resource into first index of array */
9589 mdest_res->ft_type = res->ft_type;
9590 memcpy(&mdest_res->sample_idx[0], &res->sample_idx,
9591 sizeof(struct mlx5_flow_sub_actions_idx));
9592 memcpy(&mdest_res->sample_act[0], &res->sample_act,
9593 sizeof(struct mlx5_flow_sub_actions_list));
9594 mdest_res->num_of_dest = num_of_dest;
9595 if (flow_dv_dest_array_resource_register(dev, mdest_res,
9597 return rte_flow_error_set(error, EINVAL,
9598 RTE_FLOW_ERROR_TYPE_ACTION,
9599 NULL, "can't create sample "
9602 res->sub_actions = sample_actions;
9603 if (flow_dv_sample_resource_register(dev, res, dev_flow, error))
9604 return rte_flow_error_set(error, EINVAL,
9605 RTE_FLOW_ERROR_TYPE_ACTION,
9607 "can't create sample action");
9613 * Remove an ASO age action from age actions list.
9616 * Pointer to the Ethernet device structure.
9618 * Pointer to the aso age action handler.
9621 flow_dv_aso_age_remove_from_age(struct rte_eth_dev *dev,
9622 struct mlx5_aso_age_action *age)
9624 struct mlx5_age_info *age_info;
9625 struct mlx5_age_param *age_param = &age->age_params;
9626 struct mlx5_priv *priv = dev->data->dev_private;
9627 uint16_t expected = AGE_CANDIDATE;
9629 age_info = GET_PORT_AGE_INFO(priv);
9630 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
9631 AGE_FREE, false, __ATOMIC_RELAXED,
9632 __ATOMIC_RELAXED)) {
9634 * We need the lock even it is age timeout,
9635 * since age action may still in process.
9637 rte_spinlock_lock(&age_info->aged_sl);
9638 LIST_REMOVE(age, next);
9639 rte_spinlock_unlock(&age_info->aged_sl);
9640 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
9645 * Release an ASO age action.
9648 * Pointer to the Ethernet device structure.
9649 * @param[in] age_idx
9650 * Index of ASO age action to release.
9652 * True if the release operation is during flow destroy operation.
9653 * False if the release operation is during action destroy operation.
9656 * 0 when age action was removed, otherwise the number of references.
9659 flow_dv_aso_age_release(struct rte_eth_dev *dev, uint32_t age_idx)
9661 struct mlx5_priv *priv = dev->data->dev_private;
9662 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
9663 struct mlx5_aso_age_action *age = flow_aso_age_get_by_idx(dev, age_idx);
9664 uint32_t ret = __atomic_sub_fetch(&age->refcnt, 1, __ATOMIC_RELAXED);
9667 flow_dv_aso_age_remove_from_age(dev, age);
9668 rte_spinlock_lock(&mng->free_sl);
9669 LIST_INSERT_HEAD(&mng->free, age, next);
9670 rte_spinlock_unlock(&mng->free_sl);
9676 * Resize the ASO age pools array by MLX5_CNT_CONTAINER_RESIZE pools.
9679 * Pointer to the Ethernet device structure.
9682 * 0 on success, otherwise negative errno value and rte_errno is set.
9685 flow_dv_aso_age_pools_resize(struct rte_eth_dev *dev)
9687 struct mlx5_priv *priv = dev->data->dev_private;
9688 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
9689 void *old_pools = mng->pools;
9690 uint32_t resize = mng->n + MLX5_CNT_CONTAINER_RESIZE;
9691 uint32_t mem_size = sizeof(struct mlx5_aso_age_pool *) * resize;
9692 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
9699 memcpy(pools, old_pools,
9700 mng->n * sizeof(struct mlx5_flow_counter_pool *));
9701 mlx5_free(old_pools);
9703 /* First ASO flow hit allocation - starting ASO data-path. */
9704 int ret = mlx5_aso_queue_start(priv->sh);
9717 * Create and initialize a new ASO aging pool.
9720 * Pointer to the Ethernet device structure.
9721 * @param[out] age_free
9722 * Where to put the pointer of a new age action.
9725 * The age actions pool pointer and @p age_free is set on success,
9726 * NULL otherwise and rte_errno is set.
9728 static struct mlx5_aso_age_pool *
9729 flow_dv_age_pool_create(struct rte_eth_dev *dev,
9730 struct mlx5_aso_age_action **age_free)
9732 struct mlx5_priv *priv = dev->data->dev_private;
9733 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
9734 struct mlx5_aso_age_pool *pool = NULL;
9735 struct mlx5_devx_obj *obj = NULL;
9738 obj = mlx5_devx_cmd_create_flow_hit_aso_obj(priv->sh->ctx,
9741 rte_errno = ENODATA;
9742 DRV_LOG(ERR, "Failed to create flow_hit_aso_obj using DevX.");
9745 pool = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*pool), 0, SOCKET_ID_ANY);
9747 claim_zero(mlx5_devx_cmd_destroy(obj));
9751 pool->flow_hit_aso_obj = obj;
9752 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
9753 rte_spinlock_lock(&mng->resize_sl);
9754 pool->index = mng->next;
9755 /* Resize pools array if there is no room for the new pool in it. */
9756 if (pool->index == mng->n && flow_dv_aso_age_pools_resize(dev)) {
9757 claim_zero(mlx5_devx_cmd_destroy(obj));
9759 rte_spinlock_unlock(&mng->resize_sl);
9762 mng->pools[pool->index] = pool;
9764 rte_spinlock_unlock(&mng->resize_sl);
9765 /* Assign the first action in the new pool, the rest go to free list. */
9766 *age_free = &pool->actions[0];
9767 for (i = 1; i < MLX5_ASO_AGE_ACTIONS_PER_POOL; i++) {
9768 pool->actions[i].offset = i;
9769 LIST_INSERT_HEAD(&mng->free, &pool->actions[i], next);
9775 * Allocate a ASO aging bit.
9778 * Pointer to the Ethernet device structure.
9780 * Pointer to the error structure.
9783 * Index to ASO age action on success, 0 otherwise and rte_errno is set.
9786 flow_dv_aso_age_alloc(struct rte_eth_dev *dev, struct rte_flow_error *error)
9788 struct mlx5_priv *priv = dev->data->dev_private;
9789 const struct mlx5_aso_age_pool *pool;
9790 struct mlx5_aso_age_action *age_free = NULL;
9791 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
9794 /* Try to get the next free age action bit. */
9795 rte_spinlock_lock(&mng->free_sl);
9796 age_free = LIST_FIRST(&mng->free);
9798 LIST_REMOVE(age_free, next);
9799 } else if (!flow_dv_age_pool_create(dev, &age_free)) {
9800 rte_spinlock_unlock(&mng->free_sl);
9801 rte_flow_error_set(error, rte_errno, RTE_FLOW_ERROR_TYPE_ACTION,
9802 NULL, "failed to create ASO age pool");
9803 return 0; /* 0 is an error. */
9805 rte_spinlock_unlock(&mng->free_sl);
9807 ((const struct mlx5_aso_age_action (*)[MLX5_ASO_AGE_ACTIONS_PER_POOL])
9808 (age_free - age_free->offset), const struct mlx5_aso_age_pool,
9810 if (!age_free->dr_action) {
9811 int reg_c = mlx5_flow_get_reg_id(dev, MLX5_ASO_FLOW_HIT, 0,
9815 rte_flow_error_set(error, rte_errno,
9816 RTE_FLOW_ERROR_TYPE_ACTION,
9817 NULL, "failed to get reg_c "
9818 "for ASO flow hit");
9819 return 0; /* 0 is an error. */
9821 #ifdef HAVE_MLX5_DR_CREATE_ACTION_ASO
9822 age_free->dr_action = mlx5_glue->dv_create_flow_action_aso
9823 (priv->sh->rx_domain,
9824 pool->flow_hit_aso_obj->obj, age_free->offset,
9825 MLX5DV_DR_ACTION_FLAGS_ASO_FIRST_HIT_SET,
9827 #endif /* HAVE_MLX5_DR_CREATE_ACTION_ASO */
9828 if (!age_free->dr_action) {
9830 rte_spinlock_lock(&mng->free_sl);
9831 LIST_INSERT_HEAD(&mng->free, age_free, next);
9832 rte_spinlock_unlock(&mng->free_sl);
9833 rte_flow_error_set(error, rte_errno,
9834 RTE_FLOW_ERROR_TYPE_ACTION,
9835 NULL, "failed to create ASO "
9837 return 0; /* 0 is an error. */
9840 __atomic_store_n(&age_free->refcnt, 1, __ATOMIC_RELAXED);
9841 return pool->index | ((age_free->offset + 1) << 16);
9845 * Create a age action using ASO mechanism.
9848 * Pointer to rte_eth_dev structure.
9850 * Pointer to the aging action configuration.
9852 * Pointer to the error structure.
9855 * Index to flow counter on success, 0 otherwise.
9858 flow_dv_translate_create_aso_age(struct rte_eth_dev *dev,
9859 const struct rte_flow_action_age *age,
9860 struct rte_flow_error *error)
9862 uint32_t age_idx = 0;
9863 struct mlx5_aso_age_action *aso_age;
9865 age_idx = flow_dv_aso_age_alloc(dev, error);
9868 aso_age = flow_aso_age_get_by_idx(dev, age_idx);
9869 aso_age->age_params.context = age->context;
9870 aso_age->age_params.timeout = age->timeout;
9871 aso_age->age_params.port_id = dev->data->port_id;
9872 __atomic_store_n(&aso_age->age_params.sec_since_last_hit, 0,
9874 __atomic_store_n(&aso_age->age_params.state, AGE_CANDIDATE,
9880 * Fill the flow with DV spec, lock free
9881 * (mutex should be acquired by caller).
9884 * Pointer to rte_eth_dev structure.
9885 * @param[in, out] dev_flow
9886 * Pointer to the sub flow.
9888 * Pointer to the flow attributes.
9890 * Pointer to the list of items.
9891 * @param[in] actions
9892 * Pointer to the list of actions.
9894 * Pointer to the error structure.
9897 * 0 on success, a negative errno value otherwise and rte_errno is set.
9900 flow_dv_translate(struct rte_eth_dev *dev,
9901 struct mlx5_flow *dev_flow,
9902 const struct rte_flow_attr *attr,
9903 const struct rte_flow_item items[],
9904 const struct rte_flow_action actions[],
9905 struct rte_flow_error *error)
9907 struct mlx5_priv *priv = dev->data->dev_private;
9908 struct mlx5_dev_config *dev_conf = &priv->config;
9909 struct rte_flow *flow = dev_flow->flow;
9910 struct mlx5_flow_handle *handle = dev_flow->handle;
9911 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
9912 struct mlx5_flow_rss_desc *rss_desc;
9913 uint64_t item_flags = 0;
9914 uint64_t last_item = 0;
9915 uint64_t action_flags = 0;
9916 uint64_t priority = attr->priority;
9917 struct mlx5_flow_dv_matcher matcher = {
9919 .size = sizeof(matcher.mask.buf) -
9920 MLX5_ST_SZ_BYTES(fte_match_set_misc4),
9924 bool actions_end = false;
9926 struct mlx5_flow_dv_modify_hdr_resource res;
9927 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
9928 sizeof(struct mlx5_modification_cmd) *
9929 (MLX5_MAX_MODIFY_NUM + 1)];
9931 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
9932 const struct rte_flow_action_count *count = NULL;
9933 const struct rte_flow_action_age *age = NULL;
9934 union flow_dv_attr flow_attr = { .attr = 0 };
9936 union mlx5_flow_tbl_key tbl_key;
9937 uint32_t modify_action_position = UINT32_MAX;
9938 void *match_mask = matcher.mask.buf;
9939 void *match_value = dev_flow->dv.value.buf;
9940 uint8_t next_protocol = 0xff;
9941 struct rte_vlan_hdr vlan = { 0 };
9942 struct mlx5_flow_dv_dest_array_resource mdest_res;
9943 struct mlx5_flow_dv_sample_resource sample_res;
9944 void *sample_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9945 struct mlx5_flow_sub_actions_list *sample_act;
9946 uint32_t sample_act_pos = UINT32_MAX;
9947 uint32_t num_of_dest = 0;
9948 int tmp_actions_n = 0;
9951 const struct mlx5_flow_tunnel *tunnel;
9952 struct flow_grp_info grp_info = {
9953 .external = !!dev_flow->external,
9954 .transfer = !!attr->transfer,
9955 .fdb_def_rule = !!priv->fdb_def_rule,
9956 .skip_scale = !!dev_flow->skip_scale,
9960 return rte_flow_error_set(error, ENOMEM,
9961 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9963 "failed to push flow workspace");
9964 rss_desc = &wks->rss_desc;
9965 memset(&mdest_res, 0, sizeof(struct mlx5_flow_dv_dest_array_resource));
9966 memset(&sample_res, 0, sizeof(struct mlx5_flow_dv_sample_resource));
9967 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
9968 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9969 /* update normal path action resource into last index of array */
9970 sample_act = &mdest_res.sample_act[MLX5_MAX_DEST_NUM - 1];
9971 tunnel = is_flow_tunnel_match_rule(dev, attr, items, actions) ?
9972 flow_items_to_tunnel(items) :
9973 is_flow_tunnel_steer_rule(dev, attr, items, actions) ?
9974 flow_actions_to_tunnel(actions) :
9975 dev_flow->tunnel ? dev_flow->tunnel : NULL;
9976 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
9977 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9978 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
9979 (dev, tunnel, attr, items, actions);
9980 ret = mlx5_flow_group_to_table(dev, tunnel, attr->group, &table,
9984 dev_flow->dv.group = table;
9986 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
9987 if (priority == MLX5_FLOW_PRIO_RSVD)
9988 priority = dev_conf->flow_prio - 1;
9989 /* number of actions must be set to 0 in case of dirty stack. */
9990 mhdr_res->actions_num = 0;
9991 if (is_flow_tunnel_match_rule(dev, attr, items, actions)) {
9993 * do not add decap action if match rule drops packet
9994 * HW rejects rules with decap & drop
9996 * if tunnel match rule was inserted before matching tunnel set
9997 * rule flow table used in the match rule must be registered.
9998 * current implementation handles that in the
9999 * flow_dv_match_register() at the function end.
10001 bool add_decap = true;
10002 const struct rte_flow_action *ptr = actions;
10004 for (; ptr->type != RTE_FLOW_ACTION_TYPE_END; ptr++) {
10005 if (ptr->type == RTE_FLOW_ACTION_TYPE_DROP) {
10011 if (flow_dv_create_action_l2_decap(dev, dev_flow,
10015 dev_flow->dv.actions[actions_n++] =
10016 dev_flow->dv.encap_decap->action;
10017 action_flags |= MLX5_FLOW_ACTION_DECAP;
10020 for (; !actions_end ; actions++) {
10021 const struct rte_flow_action_queue *queue;
10022 const struct rte_flow_action_rss *rss;
10023 const struct rte_flow_action *action = actions;
10024 const uint8_t *rss_key;
10025 const struct rte_flow_action_meter *mtr;
10026 struct mlx5_flow_tbl_resource *tbl;
10027 struct mlx5_aso_age_action *age_act;
10028 uint32_t port_id = 0;
10029 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
10030 int action_type = actions->type;
10031 const struct rte_flow_action *found_action = NULL;
10032 struct mlx5_flow_meter *fm = NULL;
10033 uint32_t jump_group = 0;
10035 if (!mlx5_flow_os_action_supported(action_type))
10036 return rte_flow_error_set(error, ENOTSUP,
10037 RTE_FLOW_ERROR_TYPE_ACTION,
10039 "action not supported");
10040 switch (action_type) {
10041 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
10042 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
10044 case RTE_FLOW_ACTION_TYPE_VOID:
10046 case RTE_FLOW_ACTION_TYPE_PORT_ID:
10047 if (flow_dv_translate_action_port_id(dev, action,
10050 port_id_resource.port_id = port_id;
10051 MLX5_ASSERT(!handle->rix_port_id_action);
10052 if (flow_dv_port_id_action_resource_register
10053 (dev, &port_id_resource, dev_flow, error))
10055 dev_flow->dv.actions[actions_n++] =
10056 dev_flow->dv.port_id_action->action;
10057 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
10058 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
10059 sample_act->action_flags |= MLX5_FLOW_ACTION_PORT_ID;
10062 case RTE_FLOW_ACTION_TYPE_FLAG:
10063 action_flags |= MLX5_FLOW_ACTION_FLAG;
10064 dev_flow->handle->mark = 1;
10065 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
10066 struct rte_flow_action_mark mark = {
10067 .id = MLX5_FLOW_MARK_DEFAULT,
10070 if (flow_dv_convert_action_mark(dev, &mark,
10074 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
10077 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
10079 * Only one FLAG or MARK is supported per device flow
10080 * right now. So the pointer to the tag resource must be
10081 * zero before the register process.
10083 MLX5_ASSERT(!handle->dvh.rix_tag);
10084 if (flow_dv_tag_resource_register(dev, tag_be,
10087 MLX5_ASSERT(dev_flow->dv.tag_resource);
10088 dev_flow->dv.actions[actions_n++] =
10089 dev_flow->dv.tag_resource->action;
10091 case RTE_FLOW_ACTION_TYPE_MARK:
10092 action_flags |= MLX5_FLOW_ACTION_MARK;
10093 dev_flow->handle->mark = 1;
10094 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
10095 const struct rte_flow_action_mark *mark =
10096 (const struct rte_flow_action_mark *)
10099 if (flow_dv_convert_action_mark(dev, mark,
10103 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
10107 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
10108 /* Legacy (non-extensive) MARK action. */
10109 tag_be = mlx5_flow_mark_set
10110 (((const struct rte_flow_action_mark *)
10111 (actions->conf))->id);
10112 MLX5_ASSERT(!handle->dvh.rix_tag);
10113 if (flow_dv_tag_resource_register(dev, tag_be,
10116 MLX5_ASSERT(dev_flow->dv.tag_resource);
10117 dev_flow->dv.actions[actions_n++] =
10118 dev_flow->dv.tag_resource->action;
10120 case RTE_FLOW_ACTION_TYPE_SET_META:
10121 if (flow_dv_convert_action_set_meta
10122 (dev, mhdr_res, attr,
10123 (const struct rte_flow_action_set_meta *)
10124 actions->conf, error))
10126 action_flags |= MLX5_FLOW_ACTION_SET_META;
10128 case RTE_FLOW_ACTION_TYPE_SET_TAG:
10129 if (flow_dv_convert_action_set_tag
10131 (const struct rte_flow_action_set_tag *)
10132 actions->conf, error))
10134 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
10136 case RTE_FLOW_ACTION_TYPE_DROP:
10137 action_flags |= MLX5_FLOW_ACTION_DROP;
10138 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
10140 case RTE_FLOW_ACTION_TYPE_QUEUE:
10141 queue = actions->conf;
10142 rss_desc->queue_num = 1;
10143 rss_desc->queue[0] = queue->index;
10144 action_flags |= MLX5_FLOW_ACTION_QUEUE;
10145 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
10146 sample_act->action_flags |= MLX5_FLOW_ACTION_QUEUE;
10149 case RTE_FLOW_ACTION_TYPE_RSS:
10150 rss = actions->conf;
10151 memcpy(rss_desc->queue, rss->queue,
10152 rss->queue_num * sizeof(uint16_t));
10153 rss_desc->queue_num = rss->queue_num;
10154 /* NULL RSS key indicates default RSS key. */
10155 rss_key = !rss->key ? rss_hash_default_key : rss->key;
10156 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
10158 * rss->level and rss.types should be set in advance
10159 * when expanding items for RSS.
10161 action_flags |= MLX5_FLOW_ACTION_RSS;
10162 dev_flow->handle->fate_action = rss_desc->shared_rss ?
10163 MLX5_FLOW_FATE_SHARED_RSS :
10164 MLX5_FLOW_FATE_QUEUE;
10166 case MLX5_RTE_FLOW_ACTION_TYPE_AGE:
10167 flow->age = (uint32_t)(uintptr_t)(action->conf);
10168 age_act = flow_aso_age_get_by_idx(dev, flow->age);
10169 __atomic_fetch_add(&age_act->refcnt, 1,
10171 dev_flow->dv.actions[actions_n++] = age_act->dr_action;
10172 action_flags |= MLX5_FLOW_ACTION_AGE;
10174 case RTE_FLOW_ACTION_TYPE_AGE:
10175 if (priv->sh->flow_hit_aso_en && attr->group) {
10176 flow->age = flow_dv_translate_create_aso_age
10177 (dev, action->conf, error);
10179 return rte_flow_error_set
10181 RTE_FLOW_ERROR_TYPE_ACTION,
10183 "can't create ASO age action");
10184 dev_flow->dv.actions[actions_n++] =
10185 (flow_aso_age_get_by_idx
10186 (dev, flow->age))->dr_action;
10187 action_flags |= MLX5_FLOW_ACTION_AGE;
10191 case RTE_FLOW_ACTION_TYPE_COUNT:
10192 if (!dev_conf->devx) {
10193 return rte_flow_error_set
10195 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10197 "count action not supported");
10199 /* Save information first, will apply later. */
10200 if (actions->type == RTE_FLOW_ACTION_TYPE_COUNT)
10201 count = action->conf;
10203 age = action->conf;
10204 action_flags |= MLX5_FLOW_ACTION_COUNT;
10206 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
10207 dev_flow->dv.actions[actions_n++] =
10208 priv->sh->pop_vlan_action;
10209 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
10211 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
10212 if (!(action_flags &
10213 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
10214 flow_dev_get_vlan_info_from_items(items, &vlan);
10215 vlan.eth_proto = rte_be_to_cpu_16
10216 ((((const struct rte_flow_action_of_push_vlan *)
10217 actions->conf)->ethertype));
10218 found_action = mlx5_flow_find_action
10220 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
10222 mlx5_update_vlan_vid_pcp(found_action, &vlan);
10223 found_action = mlx5_flow_find_action
10225 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
10227 mlx5_update_vlan_vid_pcp(found_action, &vlan);
10228 if (flow_dv_create_action_push_vlan
10229 (dev, attr, &vlan, dev_flow, error))
10231 dev_flow->dv.actions[actions_n++] =
10232 dev_flow->dv.push_vlan_res->action;
10233 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
10235 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
10236 /* of_vlan_push action handled this action */
10237 MLX5_ASSERT(action_flags &
10238 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
10240 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
10241 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
10243 flow_dev_get_vlan_info_from_items(items, &vlan);
10244 mlx5_update_vlan_vid_pcp(actions, &vlan);
10245 /* If no VLAN push - this is a modify header action */
10246 if (flow_dv_convert_action_modify_vlan_vid
10247 (mhdr_res, actions, error))
10249 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
10251 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
10252 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
10253 if (flow_dv_create_action_l2_encap(dev, actions,
10258 dev_flow->dv.actions[actions_n++] =
10259 dev_flow->dv.encap_decap->action;
10260 action_flags |= MLX5_FLOW_ACTION_ENCAP;
10261 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
10262 sample_act->action_flags |=
10263 MLX5_FLOW_ACTION_ENCAP;
10265 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
10266 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
10267 if (flow_dv_create_action_l2_decap(dev, dev_flow,
10271 dev_flow->dv.actions[actions_n++] =
10272 dev_flow->dv.encap_decap->action;
10273 action_flags |= MLX5_FLOW_ACTION_DECAP;
10275 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
10276 /* Handle encap with preceding decap. */
10277 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
10278 if (flow_dv_create_action_raw_encap
10279 (dev, actions, dev_flow, attr, error))
10281 dev_flow->dv.actions[actions_n++] =
10282 dev_flow->dv.encap_decap->action;
10284 /* Handle encap without preceding decap. */
10285 if (flow_dv_create_action_l2_encap
10286 (dev, actions, dev_flow, attr->transfer,
10289 dev_flow->dv.actions[actions_n++] =
10290 dev_flow->dv.encap_decap->action;
10292 action_flags |= MLX5_FLOW_ACTION_ENCAP;
10293 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
10294 sample_act->action_flags |=
10295 MLX5_FLOW_ACTION_ENCAP;
10297 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
10298 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
10300 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
10301 if (flow_dv_create_action_l2_decap
10302 (dev, dev_flow, attr->transfer, error))
10304 dev_flow->dv.actions[actions_n++] =
10305 dev_flow->dv.encap_decap->action;
10307 /* If decap is followed by encap, handle it at encap. */
10308 action_flags |= MLX5_FLOW_ACTION_DECAP;
10310 case RTE_FLOW_ACTION_TYPE_JUMP:
10311 jump_group = ((const struct rte_flow_action_jump *)
10312 action->conf)->group;
10313 grp_info.std_tbl_fix = 0;
10314 grp_info.skip_scale = 0;
10315 ret = mlx5_flow_group_to_table(dev, tunnel,
10321 tbl = flow_dv_tbl_resource_get(dev, table, attr->egress,
10323 !!dev_flow->external,
10324 tunnel, jump_group, 0,
10327 return rte_flow_error_set
10329 RTE_FLOW_ERROR_TYPE_ACTION,
10331 "cannot create jump action.");
10332 if (flow_dv_jump_tbl_resource_register
10333 (dev, tbl, dev_flow, error)) {
10334 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
10335 return rte_flow_error_set
10337 RTE_FLOW_ERROR_TYPE_ACTION,
10339 "cannot create jump action.");
10341 dev_flow->dv.actions[actions_n++] =
10342 dev_flow->dv.jump->action;
10343 action_flags |= MLX5_FLOW_ACTION_JUMP;
10344 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
10346 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
10347 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
10348 if (flow_dv_convert_action_modify_mac
10349 (mhdr_res, actions, error))
10351 action_flags |= actions->type ==
10352 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
10353 MLX5_FLOW_ACTION_SET_MAC_SRC :
10354 MLX5_FLOW_ACTION_SET_MAC_DST;
10356 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
10357 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
10358 if (flow_dv_convert_action_modify_ipv4
10359 (mhdr_res, actions, error))
10361 action_flags |= actions->type ==
10362 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
10363 MLX5_FLOW_ACTION_SET_IPV4_SRC :
10364 MLX5_FLOW_ACTION_SET_IPV4_DST;
10366 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
10367 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
10368 if (flow_dv_convert_action_modify_ipv6
10369 (mhdr_res, actions, error))
10371 action_flags |= actions->type ==
10372 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
10373 MLX5_FLOW_ACTION_SET_IPV6_SRC :
10374 MLX5_FLOW_ACTION_SET_IPV6_DST;
10376 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
10377 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
10378 if (flow_dv_convert_action_modify_tp
10379 (mhdr_res, actions, items,
10380 &flow_attr, dev_flow, !!(action_flags &
10381 MLX5_FLOW_ACTION_DECAP), error))
10383 action_flags |= actions->type ==
10384 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
10385 MLX5_FLOW_ACTION_SET_TP_SRC :
10386 MLX5_FLOW_ACTION_SET_TP_DST;
10388 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
10389 if (flow_dv_convert_action_modify_dec_ttl
10390 (mhdr_res, items, &flow_attr, dev_flow,
10392 MLX5_FLOW_ACTION_DECAP), error))
10394 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
10396 case RTE_FLOW_ACTION_TYPE_SET_TTL:
10397 if (flow_dv_convert_action_modify_ttl
10398 (mhdr_res, actions, items, &flow_attr,
10399 dev_flow, !!(action_flags &
10400 MLX5_FLOW_ACTION_DECAP), error))
10402 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
10404 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
10405 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
10406 if (flow_dv_convert_action_modify_tcp_seq
10407 (mhdr_res, actions, error))
10409 action_flags |= actions->type ==
10410 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
10411 MLX5_FLOW_ACTION_INC_TCP_SEQ :
10412 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
10415 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
10416 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
10417 if (flow_dv_convert_action_modify_tcp_ack
10418 (mhdr_res, actions, error))
10420 action_flags |= actions->type ==
10421 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
10422 MLX5_FLOW_ACTION_INC_TCP_ACK :
10423 MLX5_FLOW_ACTION_DEC_TCP_ACK;
10425 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
10426 if (flow_dv_convert_action_set_reg
10427 (mhdr_res, actions, error))
10429 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
10431 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
10432 if (flow_dv_convert_action_copy_mreg
10433 (dev, mhdr_res, actions, error))
10435 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
10437 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
10438 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
10439 dev_flow->handle->fate_action =
10440 MLX5_FLOW_FATE_DEFAULT_MISS;
10442 case RTE_FLOW_ACTION_TYPE_METER:
10443 mtr = actions->conf;
10444 if (!flow->meter) {
10445 fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
10448 return rte_flow_error_set(error,
10450 RTE_FLOW_ERROR_TYPE_ACTION,
10453 "or invalid parameters");
10454 flow->meter = fm->idx;
10456 /* Set the meter action. */
10458 fm = mlx5_ipool_get(priv->sh->ipool
10459 [MLX5_IPOOL_MTR], flow->meter);
10461 return rte_flow_error_set(error,
10463 RTE_FLOW_ERROR_TYPE_ACTION,
10466 "or invalid parameters");
10468 dev_flow->dv.actions[actions_n++] =
10469 fm->mfts->meter_action;
10470 action_flags |= MLX5_FLOW_ACTION_METER;
10472 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
10473 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
10476 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
10478 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
10479 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
10482 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
10484 case RTE_FLOW_ACTION_TYPE_SAMPLE:
10485 sample_act_pos = actions_n;
10486 ret = flow_dv_translate_action_sample(dev,
10496 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
10497 /* put encap action into group if work with port id */
10498 if ((action_flags & MLX5_FLOW_ACTION_ENCAP) &&
10499 (action_flags & MLX5_FLOW_ACTION_PORT_ID))
10500 sample_act->action_flags |=
10501 MLX5_FLOW_ACTION_ENCAP;
10503 case RTE_FLOW_ACTION_TYPE_END:
10504 actions_end = true;
10505 if (mhdr_res->actions_num) {
10506 /* create modify action if needed. */
10507 if (flow_dv_modify_hdr_resource_register
10508 (dev, mhdr_res, dev_flow, error))
10510 dev_flow->dv.actions[modify_action_position] =
10511 handle->dvh.modify_hdr->action;
10513 if (action_flags & MLX5_FLOW_ACTION_COUNT) {
10515 flow_dv_translate_create_counter(dev,
10516 dev_flow, count, age);
10518 if (!flow->counter)
10519 return rte_flow_error_set
10521 RTE_FLOW_ERROR_TYPE_ACTION,
10523 "cannot create counter"
10525 dev_flow->dv.actions[actions_n] =
10526 (flow_dv_counter_get_by_idx(dev,
10527 flow->counter, NULL))->action;
10530 if (action_flags & MLX5_FLOW_ACTION_SAMPLE) {
10531 ret = flow_dv_create_action_sample(dev,
10540 return rte_flow_error_set
10542 RTE_FLOW_ERROR_TYPE_ACTION,
10544 "cannot create sample action");
10545 if (num_of_dest > 1) {
10546 dev_flow->dv.actions[sample_act_pos] =
10547 dev_flow->dv.dest_array_res->action;
10549 dev_flow->dv.actions[sample_act_pos] =
10550 dev_flow->dv.sample_res->verbs_action;
10557 if (mhdr_res->actions_num &&
10558 modify_action_position == UINT32_MAX)
10559 modify_action_position = actions_n++;
10562 * For multiple destination (sample action with ratio=1), the encap
10563 * action and port id action will be combined into group action.
10564 * So need remove the original these actions in the flow and only
10565 * use the sample action instead of.
10567 if (num_of_dest > 1 && sample_act->dr_port_id_action) {
10569 void *temp_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
10571 for (i = 0; i < actions_n; i++) {
10572 if ((sample_act->dr_encap_action &&
10573 sample_act->dr_encap_action ==
10574 dev_flow->dv.actions[i]) ||
10575 (sample_act->dr_port_id_action &&
10576 sample_act->dr_port_id_action ==
10577 dev_flow->dv.actions[i]))
10579 temp_actions[tmp_actions_n++] = dev_flow->dv.actions[i];
10581 memcpy((void *)dev_flow->dv.actions,
10582 (void *)temp_actions,
10583 tmp_actions_n * sizeof(void *));
10584 actions_n = tmp_actions_n;
10586 dev_flow->dv.actions_n = actions_n;
10587 dev_flow->act_flags = action_flags;
10588 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
10589 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
10590 int item_type = items->type;
10592 if (!mlx5_flow_os_item_supported(item_type))
10593 return rte_flow_error_set(error, ENOTSUP,
10594 RTE_FLOW_ERROR_TYPE_ITEM,
10595 NULL, "item not supported");
10596 switch (item_type) {
10597 case RTE_FLOW_ITEM_TYPE_PORT_ID:
10598 flow_dv_translate_item_port_id
10599 (dev, match_mask, match_value, items, attr);
10600 last_item = MLX5_FLOW_ITEM_PORT_ID;
10602 case RTE_FLOW_ITEM_TYPE_ETH:
10603 flow_dv_translate_item_eth(match_mask, match_value,
10605 dev_flow->dv.group);
10606 matcher.priority = action_flags &
10607 MLX5_FLOW_ACTION_DEFAULT_MISS &&
10608 !dev_flow->external ?
10609 MLX5_PRIORITY_MAP_L3 :
10610 MLX5_PRIORITY_MAP_L2;
10611 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
10612 MLX5_FLOW_LAYER_OUTER_L2;
10614 case RTE_FLOW_ITEM_TYPE_VLAN:
10615 flow_dv_translate_item_vlan(dev_flow,
10616 match_mask, match_value,
10618 dev_flow->dv.group);
10619 matcher.priority = MLX5_PRIORITY_MAP_L2;
10620 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
10621 MLX5_FLOW_LAYER_INNER_VLAN) :
10622 (MLX5_FLOW_LAYER_OUTER_L2 |
10623 MLX5_FLOW_LAYER_OUTER_VLAN);
10625 case RTE_FLOW_ITEM_TYPE_IPV4:
10626 mlx5_flow_tunnel_ip_check(items, next_protocol,
10627 &item_flags, &tunnel);
10628 flow_dv_translate_item_ipv4(match_mask, match_value,
10630 dev_flow->dv.group);
10631 matcher.priority = MLX5_PRIORITY_MAP_L3;
10632 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
10633 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
10634 if (items->mask != NULL &&
10635 ((const struct rte_flow_item_ipv4 *)
10636 items->mask)->hdr.next_proto_id) {
10638 ((const struct rte_flow_item_ipv4 *)
10639 (items->spec))->hdr.next_proto_id;
10641 ((const struct rte_flow_item_ipv4 *)
10642 (items->mask))->hdr.next_proto_id;
10644 /* Reset for inner layer. */
10645 next_protocol = 0xff;
10648 case RTE_FLOW_ITEM_TYPE_IPV6:
10649 mlx5_flow_tunnel_ip_check(items, next_protocol,
10650 &item_flags, &tunnel);
10651 flow_dv_translate_item_ipv6(match_mask, match_value,
10653 dev_flow->dv.group);
10654 matcher.priority = MLX5_PRIORITY_MAP_L3;
10655 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
10656 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
10657 if (items->mask != NULL &&
10658 ((const struct rte_flow_item_ipv6 *)
10659 items->mask)->hdr.proto) {
10661 ((const struct rte_flow_item_ipv6 *)
10662 items->spec)->hdr.proto;
10664 ((const struct rte_flow_item_ipv6 *)
10665 items->mask)->hdr.proto;
10667 /* Reset for inner layer. */
10668 next_protocol = 0xff;
10671 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
10672 flow_dv_translate_item_ipv6_frag_ext(match_mask,
10675 last_item = tunnel ?
10676 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
10677 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
10678 if (items->mask != NULL &&
10679 ((const struct rte_flow_item_ipv6_frag_ext *)
10680 items->mask)->hdr.next_header) {
10682 ((const struct rte_flow_item_ipv6_frag_ext *)
10683 items->spec)->hdr.next_header;
10685 ((const struct rte_flow_item_ipv6_frag_ext *)
10686 items->mask)->hdr.next_header;
10688 /* Reset for inner layer. */
10689 next_protocol = 0xff;
10692 case RTE_FLOW_ITEM_TYPE_TCP:
10693 flow_dv_translate_item_tcp(match_mask, match_value,
10695 matcher.priority = MLX5_PRIORITY_MAP_L4;
10696 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
10697 MLX5_FLOW_LAYER_OUTER_L4_TCP;
10699 case RTE_FLOW_ITEM_TYPE_UDP:
10700 flow_dv_translate_item_udp(match_mask, match_value,
10702 matcher.priority = MLX5_PRIORITY_MAP_L4;
10703 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
10704 MLX5_FLOW_LAYER_OUTER_L4_UDP;
10706 case RTE_FLOW_ITEM_TYPE_GRE:
10707 flow_dv_translate_item_gre(match_mask, match_value,
10709 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10710 last_item = MLX5_FLOW_LAYER_GRE;
10712 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
10713 flow_dv_translate_item_gre_key(match_mask,
10714 match_value, items);
10715 last_item = MLX5_FLOW_LAYER_GRE_KEY;
10717 case RTE_FLOW_ITEM_TYPE_NVGRE:
10718 flow_dv_translate_item_nvgre(match_mask, match_value,
10720 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10721 last_item = MLX5_FLOW_LAYER_GRE;
10723 case RTE_FLOW_ITEM_TYPE_VXLAN:
10724 flow_dv_translate_item_vxlan(match_mask, match_value,
10726 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10727 last_item = MLX5_FLOW_LAYER_VXLAN;
10729 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
10730 flow_dv_translate_item_vxlan_gpe(match_mask,
10731 match_value, items,
10733 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10734 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
10736 case RTE_FLOW_ITEM_TYPE_GENEVE:
10737 flow_dv_translate_item_geneve(match_mask, match_value,
10739 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10740 last_item = MLX5_FLOW_LAYER_GENEVE;
10742 case RTE_FLOW_ITEM_TYPE_MPLS:
10743 flow_dv_translate_item_mpls(match_mask, match_value,
10744 items, last_item, tunnel);
10745 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10746 last_item = MLX5_FLOW_LAYER_MPLS;
10748 case RTE_FLOW_ITEM_TYPE_MARK:
10749 flow_dv_translate_item_mark(dev, match_mask,
10750 match_value, items);
10751 last_item = MLX5_FLOW_ITEM_MARK;
10753 case RTE_FLOW_ITEM_TYPE_META:
10754 flow_dv_translate_item_meta(dev, match_mask,
10755 match_value, attr, items);
10756 last_item = MLX5_FLOW_ITEM_METADATA;
10758 case RTE_FLOW_ITEM_TYPE_ICMP:
10759 flow_dv_translate_item_icmp(match_mask, match_value,
10761 last_item = MLX5_FLOW_LAYER_ICMP;
10763 case RTE_FLOW_ITEM_TYPE_ICMP6:
10764 flow_dv_translate_item_icmp6(match_mask, match_value,
10766 last_item = MLX5_FLOW_LAYER_ICMP6;
10768 case RTE_FLOW_ITEM_TYPE_TAG:
10769 flow_dv_translate_item_tag(dev, match_mask,
10770 match_value, items);
10771 last_item = MLX5_FLOW_ITEM_TAG;
10773 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
10774 flow_dv_translate_mlx5_item_tag(dev, match_mask,
10775 match_value, items);
10776 last_item = MLX5_FLOW_ITEM_TAG;
10778 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
10779 flow_dv_translate_item_tx_queue(dev, match_mask,
10782 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
10784 case RTE_FLOW_ITEM_TYPE_GTP:
10785 flow_dv_translate_item_gtp(match_mask, match_value,
10787 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
10788 last_item = MLX5_FLOW_LAYER_GTP;
10790 case RTE_FLOW_ITEM_TYPE_GTP_PSC:
10791 ret = flow_dv_translate_item_gtp_psc(match_mask,
10795 return rte_flow_error_set(error, -ret,
10796 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
10797 "cannot create GTP PSC item");
10798 last_item = MLX5_FLOW_LAYER_GTP_PSC;
10800 case RTE_FLOW_ITEM_TYPE_ECPRI:
10801 if (!mlx5_flex_parser_ecpri_exist(dev)) {
10802 /* Create it only the first time to be used. */
10803 ret = mlx5_flex_parser_ecpri_alloc(dev);
10805 return rte_flow_error_set
10807 RTE_FLOW_ERROR_TYPE_ITEM,
10809 "cannot create eCPRI parser");
10811 /* Adjust the length matcher and device flow value. */
10812 matcher.mask.size = MLX5_ST_SZ_BYTES(fte_match_param);
10813 dev_flow->dv.value.size =
10814 MLX5_ST_SZ_BYTES(fte_match_param);
10815 flow_dv_translate_item_ecpri(dev, match_mask,
10816 match_value, items);
10817 /* No other protocol should follow eCPRI layer. */
10818 last_item = MLX5_FLOW_LAYER_ECPRI;
10823 item_flags |= last_item;
10826 * When E-Switch mode is enabled, we have two cases where we need to
10827 * set the source port manually.
10828 * The first one, is in case of Nic steering rule, and the second is
10829 * E-Switch rule where no port_id item was found. In both cases
10830 * the source port is set according the current port in use.
10832 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
10833 (priv->representor || priv->master)) {
10834 if (flow_dv_translate_item_port_id(dev, match_mask,
10835 match_value, NULL, attr))
10838 #ifdef RTE_LIBRTE_MLX5_DEBUG
10839 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
10840 dev_flow->dv.value.buf));
10843 * Layers may be already initialized from prefix flow if this dev_flow
10844 * is the suffix flow.
10846 handle->layers |= item_flags;
10847 if (action_flags & MLX5_FLOW_ACTION_RSS)
10848 flow_dv_hashfields_set(dev_flow, rss_desc);
10849 /* Register matcher. */
10850 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
10851 matcher.mask.size);
10852 matcher.priority = mlx5_os_flow_adjust_priority(dev,
10855 /* reserved field no needs to be set to 0 here. */
10856 tbl_key.domain = attr->transfer;
10857 tbl_key.direction = attr->egress;
10858 tbl_key.table_id = dev_flow->dv.group;
10859 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow,
10860 tunnel, attr->group, error))
10866 * Set hash RX queue by hash fields (see enum ibv_rx_hash_fields)
10869 * @param[in, out] action
10870 * Shred RSS action holding hash RX queue objects.
10871 * @param[in] hash_fields
10872 * Defines combination of packet fields to participate in RX hash.
10873 * @param[in] tunnel
10875 * @param[in] hrxq_idx
10876 * Hash RX queue index to set.
10879 * 0 on success, otherwise negative errno value.
10882 __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
10883 const uint64_t hash_fields,
10887 uint32_t *hrxqs = tunnel ? action->hrxq : action->hrxq_tunnel;
10889 switch (hash_fields & ~IBV_RX_HASH_INNER) {
10890 case MLX5_RSS_HASH_IPV4:
10891 hrxqs[0] = hrxq_idx;
10893 case MLX5_RSS_HASH_IPV4_TCP:
10894 hrxqs[1] = hrxq_idx;
10896 case MLX5_RSS_HASH_IPV4_UDP:
10897 hrxqs[2] = hrxq_idx;
10899 case MLX5_RSS_HASH_IPV6:
10900 hrxqs[3] = hrxq_idx;
10902 case MLX5_RSS_HASH_IPV6_TCP:
10903 hrxqs[4] = hrxq_idx;
10905 case MLX5_RSS_HASH_IPV6_UDP:
10906 hrxqs[5] = hrxq_idx;
10908 case MLX5_RSS_HASH_NONE:
10909 hrxqs[6] = hrxq_idx;
10917 * Look up for hash RX queue by hash fields (see enum ibv_rx_hash_fields)
10921 * Pointer to the Ethernet device structure.
10923 * Shared RSS action ID holding hash RX queue objects.
10924 * @param[in] hash_fields
10925 * Defines combination of packet fields to participate in RX hash.
10926 * @param[in] tunnel
10930 * Valid hash RX queue index, otherwise 0.
10933 __flow_dv_action_rss_hrxq_lookup(struct rte_eth_dev *dev, uint32_t idx,
10934 const uint64_t hash_fields,
10937 struct mlx5_priv *priv = dev->data->dev_private;
10938 struct mlx5_shared_action_rss *shared_rss =
10939 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
10940 const uint32_t *hrxqs = tunnel ? shared_rss->hrxq :
10941 shared_rss->hrxq_tunnel;
10943 switch (hash_fields & ~IBV_RX_HASH_INNER) {
10944 case MLX5_RSS_HASH_IPV4:
10946 case MLX5_RSS_HASH_IPV4_TCP:
10948 case MLX5_RSS_HASH_IPV4_UDP:
10950 case MLX5_RSS_HASH_IPV6:
10952 case MLX5_RSS_HASH_IPV6_TCP:
10954 case MLX5_RSS_HASH_IPV6_UDP:
10956 case MLX5_RSS_HASH_NONE:
10964 * Apply the flow to the NIC, lock free,
10965 * (mutex should be acquired by caller).
10968 * Pointer to the Ethernet device structure.
10969 * @param[in, out] flow
10970 * Pointer to flow structure.
10971 * @param[out] error
10972 * Pointer to error structure.
10975 * 0 on success, a negative errno value otherwise and rte_errno is set.
10978 flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
10979 struct rte_flow_error *error)
10981 struct mlx5_flow_dv_workspace *dv;
10982 struct mlx5_flow_handle *dh;
10983 struct mlx5_flow_handle_dv *dv_h;
10984 struct mlx5_flow *dev_flow;
10985 struct mlx5_priv *priv = dev->data->dev_private;
10986 uint32_t handle_idx;
10990 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
10991 struct mlx5_flow_rss_desc *rss_desc = &wks->rss_desc;
10994 for (idx = wks->flow_idx - 1; idx >= 0; idx--) {
10995 dev_flow = &wks->flows[idx];
10996 dv = &dev_flow->dv;
10997 dh = dev_flow->handle;
11000 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
11001 if (dv->transfer) {
11002 dv->actions[n++] = priv->sh->esw_drop_action;
11004 MLX5_ASSERT(priv->drop_queue.hrxq);
11006 priv->drop_queue.hrxq->action;
11008 } else if ((dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
11009 !dv_h->rix_sample && !dv_h->rix_dest_array)) {
11010 struct mlx5_hrxq *hrxq;
11013 hrxq = flow_dv_hrxq_prepare(dev, dev_flow, rss_desc,
11018 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11019 "cannot get hash queue");
11022 dh->rix_hrxq = hrxq_idx;
11023 dv->actions[n++] = hrxq->action;
11024 } else if (dh->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
11025 struct mlx5_hrxq *hrxq = NULL;
11028 hrxq_idx = __flow_dv_action_rss_hrxq_lookup(dev,
11029 rss_desc->shared_rss,
11030 dev_flow->hash_fields,
11032 MLX5_FLOW_LAYER_TUNNEL));
11034 hrxq = mlx5_ipool_get
11035 (priv->sh->ipool[MLX5_IPOOL_HRXQ],
11040 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11041 "cannot get hash queue");
11044 dh->rix_srss = rss_desc->shared_rss;
11045 dv->actions[n++] = hrxq->action;
11046 } else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
11047 if (!priv->sh->default_miss_action) {
11050 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11051 "default miss action not be created.");
11054 dv->actions[n++] = priv->sh->default_miss_action;
11056 err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
11057 (void *)&dv->value, n,
11058 dv->actions, &dh->drv_flow);
11060 rte_flow_error_set(error, errno,
11061 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11063 "hardware refuses to create flow");
11066 if (priv->vmwa_context &&
11067 dh->vf_vlan.tag && !dh->vf_vlan.created) {
11069 * The rule contains the VLAN pattern.
11070 * For VF we are going to create VLAN
11071 * interface to make hypervisor set correct
11072 * e-Switch vport context.
11074 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
11079 err = rte_errno; /* Save rte_errno before cleanup. */
11080 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
11081 handle_idx, dh, next) {
11082 /* hrxq is union, don't clear it if the flag is not set. */
11083 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE && dh->rix_hrxq) {
11084 mlx5_hrxq_release(dev, dh->rix_hrxq);
11086 } else if (dh->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
11089 if (dh->vf_vlan.tag && dh->vf_vlan.created)
11090 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
11092 rte_errno = err; /* Restore rte_errno. */
11097 flow_dv_matcher_remove_cb(struct mlx5_cache_list *list __rte_unused,
11098 struct mlx5_cache_entry *entry)
11100 struct mlx5_flow_dv_matcher *cache = container_of(entry, typeof(*cache),
11103 claim_zero(mlx5_flow_os_destroy_flow_matcher(cache->matcher_object));
11108 * Release the flow matcher.
11111 * Pointer to Ethernet device.
11113 * Pointer to mlx5_flow_handle.
11116 * 1 while a reference on it exists, 0 when freed.
11119 flow_dv_matcher_release(struct rte_eth_dev *dev,
11120 struct mlx5_flow_handle *handle)
11122 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
11123 struct mlx5_flow_tbl_data_entry *tbl = container_of(matcher->tbl,
11124 typeof(*tbl), tbl);
11127 MLX5_ASSERT(matcher->matcher_object);
11128 ret = mlx5_cache_unregister(&tbl->matchers, &matcher->entry);
11129 flow_dv_tbl_resource_release(MLX5_SH(dev), &tbl->tbl);
11134 * Release encap_decap resource.
11137 * Pointer to the hash list.
11139 * Pointer to exist resource entry object.
11142 flow_dv_encap_decap_remove_cb(struct mlx5_hlist *list,
11143 struct mlx5_hlist_entry *entry)
11145 struct mlx5_dev_ctx_shared *sh = list->ctx;
11146 struct mlx5_flow_dv_encap_decap_resource *res =
11147 container_of(entry, typeof(*res), entry);
11149 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
11150 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], res->idx);
11154 * Release an encap/decap resource.
11157 * Pointer to Ethernet device.
11158 * @param encap_decap_idx
11159 * Index of encap decap resource.
11162 * 1 while a reference on it exists, 0 when freed.
11165 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
11166 uint32_t encap_decap_idx)
11168 struct mlx5_priv *priv = dev->data->dev_private;
11169 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
11171 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
11173 if (!cache_resource)
11175 MLX5_ASSERT(cache_resource->action);
11176 return mlx5_hlist_unregister(priv->sh->encaps_decaps,
11177 &cache_resource->entry);
11181 * Release an jump to table action resource.
11184 * Pointer to Ethernet device.
11186 * Pointer to mlx5_flow_handle.
11189 * 1 while a reference on it exists, 0 when freed.
11192 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
11193 struct mlx5_flow_handle *handle)
11195 struct mlx5_priv *priv = dev->data->dev_private;
11196 struct mlx5_flow_tbl_data_entry *tbl_data;
11198 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
11202 return flow_dv_tbl_resource_release(MLX5_SH(dev), &tbl_data->tbl);
11206 flow_dv_modify_remove_cb(struct mlx5_hlist *list __rte_unused,
11207 struct mlx5_hlist_entry *entry)
11209 struct mlx5_flow_dv_modify_hdr_resource *res =
11210 container_of(entry, typeof(*res), entry);
11212 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
11217 * Release a modify-header resource.
11220 * Pointer to Ethernet device.
11222 * Pointer to mlx5_flow_handle.
11225 * 1 while a reference on it exists, 0 when freed.
11228 flow_dv_modify_hdr_resource_release(struct rte_eth_dev *dev,
11229 struct mlx5_flow_handle *handle)
11231 struct mlx5_priv *priv = dev->data->dev_private;
11232 struct mlx5_flow_dv_modify_hdr_resource *entry = handle->dvh.modify_hdr;
11234 MLX5_ASSERT(entry->action);
11235 return mlx5_hlist_unregister(priv->sh->modify_cmds, &entry->entry);
11239 flow_dv_port_id_remove_cb(struct mlx5_cache_list *list,
11240 struct mlx5_cache_entry *entry)
11242 struct mlx5_dev_ctx_shared *sh = list->ctx;
11243 struct mlx5_flow_dv_port_id_action_resource *cache =
11244 container_of(entry, typeof(*cache), entry);
11246 claim_zero(mlx5_flow_os_destroy_flow_action(cache->action));
11247 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PORT_ID], cache->idx);
11251 * Release port ID action resource.
11254 * Pointer to Ethernet device.
11256 * Pointer to mlx5_flow_handle.
11259 * 1 while a reference on it exists, 0 when freed.
11262 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
11265 struct mlx5_priv *priv = dev->data->dev_private;
11266 struct mlx5_flow_dv_port_id_action_resource *cache;
11268 cache = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID], port_id);
11271 MLX5_ASSERT(cache->action);
11272 return mlx5_cache_unregister(&priv->sh->port_id_action_list,
11277 * Release shared RSS action resource.
11280 * Pointer to Ethernet device.
11282 * Shared RSS action index.
11285 flow_dv_shared_rss_action_release(struct rte_eth_dev *dev, uint32_t srss)
11287 struct mlx5_priv *priv = dev->data->dev_private;
11288 struct mlx5_shared_action_rss *shared_rss;
11290 shared_rss = mlx5_ipool_get
11291 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], srss);
11292 __atomic_sub_fetch(&shared_rss->refcnt, 1, __ATOMIC_RELAXED);
11296 flow_dv_push_vlan_remove_cb(struct mlx5_cache_list *list,
11297 struct mlx5_cache_entry *entry)
11299 struct mlx5_dev_ctx_shared *sh = list->ctx;
11300 struct mlx5_flow_dv_push_vlan_action_resource *cache =
11301 container_of(entry, typeof(*cache), entry);
11303 claim_zero(mlx5_flow_os_destroy_flow_action(cache->action));
11304 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PUSH_VLAN], cache->idx);
11308 * Release push vlan action resource.
11311 * Pointer to Ethernet device.
11313 * Pointer to mlx5_flow_handle.
11316 * 1 while a reference on it exists, 0 when freed.
11319 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
11320 struct mlx5_flow_handle *handle)
11322 struct mlx5_priv *priv = dev->data->dev_private;
11323 struct mlx5_flow_dv_push_vlan_action_resource *cache;
11324 uint32_t idx = handle->dvh.rix_push_vlan;
11326 cache = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
11329 MLX5_ASSERT(cache->action);
11330 return mlx5_cache_unregister(&priv->sh->push_vlan_action_list,
11335 * Release the fate resource.
11338 * Pointer to Ethernet device.
11340 * Pointer to mlx5_flow_handle.
11343 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
11344 struct mlx5_flow_handle *handle)
11346 if (!handle->rix_fate)
11348 switch (handle->fate_action) {
11349 case MLX5_FLOW_FATE_QUEUE:
11350 mlx5_hrxq_release(dev, handle->rix_hrxq);
11352 case MLX5_FLOW_FATE_JUMP:
11353 flow_dv_jump_tbl_resource_release(dev, handle);
11355 case MLX5_FLOW_FATE_PORT_ID:
11356 flow_dv_port_id_action_resource_release(dev,
11357 handle->rix_port_id_action);
11360 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
11363 handle->rix_fate = 0;
11367 flow_dv_sample_remove_cb(struct mlx5_cache_list *list __rte_unused,
11368 struct mlx5_cache_entry *entry)
11370 struct mlx5_flow_dv_sample_resource *cache_resource =
11371 container_of(entry, typeof(*cache_resource), entry);
11372 struct rte_eth_dev *dev = cache_resource->dev;
11373 struct mlx5_priv *priv = dev->data->dev_private;
11375 if (cache_resource->verbs_action)
11376 claim_zero(mlx5_flow_os_destroy_flow_action
11377 (cache_resource->verbs_action));
11378 if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
11379 if (cache_resource->default_miss)
11380 claim_zero(mlx5_flow_os_destroy_flow_action
11381 (cache_resource->default_miss));
11383 if (cache_resource->normal_path_tbl)
11384 flow_dv_tbl_resource_release(MLX5_SH(dev),
11385 cache_resource->normal_path_tbl);
11386 flow_dv_sample_sub_actions_release(dev,
11387 &cache_resource->sample_idx);
11388 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
11389 cache_resource->idx);
11390 DRV_LOG(DEBUG, "sample resource %p: removed",
11391 (void *)cache_resource);
11395 * Release an sample resource.
11398 * Pointer to Ethernet device.
11400 * Pointer to mlx5_flow_handle.
11403 * 1 while a reference on it exists, 0 when freed.
11406 flow_dv_sample_resource_release(struct rte_eth_dev *dev,
11407 struct mlx5_flow_handle *handle)
11409 struct mlx5_priv *priv = dev->data->dev_private;
11410 struct mlx5_flow_dv_sample_resource *cache_resource;
11412 cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
11413 handle->dvh.rix_sample);
11414 if (!cache_resource)
11416 MLX5_ASSERT(cache_resource->verbs_action);
11417 return mlx5_cache_unregister(&priv->sh->sample_action_list,
11418 &cache_resource->entry);
11422 flow_dv_dest_array_remove_cb(struct mlx5_cache_list *list __rte_unused,
11423 struct mlx5_cache_entry *entry)
11425 struct mlx5_flow_dv_dest_array_resource *cache_resource =
11426 container_of(entry, typeof(*cache_resource), entry);
11427 struct rte_eth_dev *dev = cache_resource->dev;
11428 struct mlx5_priv *priv = dev->data->dev_private;
11431 MLX5_ASSERT(cache_resource->action);
11432 if (cache_resource->action)
11433 claim_zero(mlx5_flow_os_destroy_flow_action
11434 (cache_resource->action));
11435 for (; i < cache_resource->num_of_dest; i++)
11436 flow_dv_sample_sub_actions_release(dev,
11437 &cache_resource->sample_idx[i]);
11438 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
11439 cache_resource->idx);
11440 DRV_LOG(DEBUG, "destination array resource %p: removed",
11441 (void *)cache_resource);
11445 * Release an destination array resource.
11448 * Pointer to Ethernet device.
11450 * Pointer to mlx5_flow_handle.
11453 * 1 while a reference on it exists, 0 when freed.
11456 flow_dv_dest_array_resource_release(struct rte_eth_dev *dev,
11457 struct mlx5_flow_handle *handle)
11459 struct mlx5_priv *priv = dev->data->dev_private;
11460 struct mlx5_flow_dv_dest_array_resource *cache;
11462 cache = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
11463 handle->dvh.rix_dest_array);
11466 MLX5_ASSERT(cache->action);
11467 return mlx5_cache_unregister(&priv->sh->dest_array_list,
11472 flow_dv_geneve_tlv_option_resource_release(struct rte_eth_dev *dev)
11474 struct mlx5_priv *priv = dev->data->dev_private;
11475 struct mlx5_dev_ctx_shared *sh = priv->sh;
11476 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource =
11477 sh->geneve_tlv_option_resource;
11478 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
11479 if (geneve_opt_resource) {
11480 if (!(__atomic_sub_fetch(&geneve_opt_resource->refcnt, 1,
11481 __ATOMIC_RELAXED))) {
11482 claim_zero(mlx5_devx_cmd_destroy
11483 (geneve_opt_resource->obj));
11484 mlx5_free(sh->geneve_tlv_option_resource);
11485 sh->geneve_tlv_option_resource = NULL;
11488 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
11492 * Remove the flow from the NIC but keeps it in memory.
11493 * Lock free, (mutex should be acquired by caller).
11496 * Pointer to Ethernet device.
11497 * @param[in, out] flow
11498 * Pointer to flow structure.
11501 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
11503 struct mlx5_flow_handle *dh;
11504 uint32_t handle_idx;
11505 struct mlx5_priv *priv = dev->data->dev_private;
11509 handle_idx = flow->dev_handles;
11510 while (handle_idx) {
11511 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
11515 if (dh->drv_flow) {
11516 claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
11517 dh->drv_flow = NULL;
11519 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE)
11520 flow_dv_fate_resource_release(dev, dh);
11521 if (dh->vf_vlan.tag && dh->vf_vlan.created)
11522 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
11523 handle_idx = dh->next.next;
11528 * Remove the flow from the NIC and the memory.
11529 * Lock free, (mutex should be acquired by caller).
11532 * Pointer to the Ethernet device structure.
11533 * @param[in, out] flow
11534 * Pointer to flow structure.
11537 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
11539 struct mlx5_flow_handle *dev_handle;
11540 struct mlx5_priv *priv = dev->data->dev_private;
11545 flow_dv_remove(dev, flow);
11546 if (flow->counter) {
11547 flow_dv_counter_free(dev, flow->counter);
11551 struct mlx5_flow_meter *fm;
11553 fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
11556 mlx5_flow_meter_detach(fm);
11560 flow_dv_aso_age_release(dev, flow->age);
11561 if (flow->geneve_tlv_option) {
11562 flow_dv_geneve_tlv_option_resource_release(dev);
11563 flow->geneve_tlv_option = 0;
11565 while (flow->dev_handles) {
11566 uint32_t tmp_idx = flow->dev_handles;
11568 dev_handle = mlx5_ipool_get(priv->sh->ipool
11569 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
11572 flow->dev_handles = dev_handle->next.next;
11573 if (dev_handle->dvh.matcher)
11574 flow_dv_matcher_release(dev, dev_handle);
11575 if (dev_handle->dvh.rix_sample)
11576 flow_dv_sample_resource_release(dev, dev_handle);
11577 if (dev_handle->dvh.rix_dest_array)
11578 flow_dv_dest_array_resource_release(dev, dev_handle);
11579 if (dev_handle->dvh.rix_encap_decap)
11580 flow_dv_encap_decap_resource_release(dev,
11581 dev_handle->dvh.rix_encap_decap);
11582 if (dev_handle->dvh.modify_hdr)
11583 flow_dv_modify_hdr_resource_release(dev, dev_handle);
11584 if (dev_handle->dvh.rix_push_vlan)
11585 flow_dv_push_vlan_action_resource_release(dev,
11587 if (dev_handle->dvh.rix_tag)
11588 flow_dv_tag_release(dev,
11589 dev_handle->dvh.rix_tag);
11590 if (dev_handle->fate_action != MLX5_FLOW_FATE_SHARED_RSS)
11591 flow_dv_fate_resource_release(dev, dev_handle);
11593 srss = dev_handle->rix_srss;
11594 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
11598 flow_dv_shared_rss_action_release(dev, srss);
11602 * Release array of hash RX queue objects.
11606 * Pointer to the Ethernet device structure.
11607 * @param[in, out] hrxqs
11608 * Array of hash RX queue objects.
11611 * Total number of references to hash RX queue objects in *hrxqs* array
11612 * after this operation.
11615 __flow_dv_hrxqs_release(struct rte_eth_dev *dev,
11616 uint32_t (*hrxqs)[MLX5_RSS_HASH_FIELDS_LEN])
11621 for (i = 0; i < RTE_DIM(*hrxqs); i++) {
11622 int ret = mlx5_hrxq_release(dev, (*hrxqs)[i]);
11632 * Release all hash RX queue objects representing shared RSS action.
11635 * Pointer to the Ethernet device structure.
11636 * @param[in, out] action
11637 * Shared RSS action to remove hash RX queue objects from.
11640 * Total number of references to hash RX queue objects stored in *action*
11641 * after this operation.
11642 * Expected to be 0 if no external references held.
11645 __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
11646 struct mlx5_shared_action_rss *action)
11648 return __flow_dv_hrxqs_release(dev, &action->hrxq) +
11649 __flow_dv_hrxqs_release(dev, &action->hrxq_tunnel);
11653 * Setup shared RSS action.
11654 * Prepare set of hash RX queue objects sufficient to handle all valid
11655 * hash_fields combinations (see enum ibv_rx_hash_fields).
11658 * Pointer to the Ethernet device structure.
11659 * @param[in] action_idx
11660 * Shared RSS action ipool index.
11661 * @param[in, out] action
11662 * Partially initialized shared RSS action.
11663 * @param[out] error
11664 * Perform verbose error reporting if not NULL. Initialized in case of
11668 * 0 on success, otherwise negative errno value.
11671 __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
11672 uint32_t action_idx,
11673 struct mlx5_shared_action_rss *action,
11674 struct rte_flow_error *error)
11676 struct mlx5_flow_rss_desc rss_desc = { 0 };
11680 if (mlx5_ind_table_obj_setup(dev, action->ind_tbl)) {
11681 return rte_flow_error_set(error, rte_errno,
11682 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11683 "cannot setup indirection table");
11685 memcpy(rss_desc.key, action->origin.key, MLX5_RSS_HASH_KEY_LEN);
11686 rss_desc.key_len = MLX5_RSS_HASH_KEY_LEN;
11687 rss_desc.const_q = action->origin.queue;
11688 rss_desc.queue_num = action->origin.queue_num;
11689 /* Set non-zero value to indicate a shared RSS. */
11690 rss_desc.shared_rss = action_idx;
11691 rss_desc.ind_tbl = action->ind_tbl;
11692 for (i = 0; i < MLX5_RSS_HASH_FIELDS_LEN; i++) {
11694 uint64_t hash_fields = mlx5_rss_hash_fields[i];
11697 for (tunnel = 0; tunnel < 2; tunnel++) {
11698 rss_desc.tunnel = tunnel;
11699 rss_desc.hash_fields = hash_fields;
11700 hrxq_idx = mlx5_hrxq_get(dev, &rss_desc);
11704 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11705 "cannot get hash queue");
11706 goto error_hrxq_new;
11708 err = __flow_dv_action_rss_hrxq_set
11709 (action, hash_fields, tunnel, hrxq_idx);
11716 __flow_dv_action_rss_hrxqs_release(dev, action);
11717 if (!mlx5_ind_table_obj_release(dev, action->ind_tbl, true))
11718 action->ind_tbl = NULL;
11724 * Create shared RSS action.
11727 * Pointer to the Ethernet device structure.
11729 * Shared action configuration.
11731 * RSS action specification used to create shared action.
11732 * @param[out] error
11733 * Perform verbose error reporting if not NULL. Initialized in case of
11737 * A valid shared action ID in case of success, 0 otherwise and
11738 * rte_errno is set.
11741 __flow_dv_action_rss_create(struct rte_eth_dev *dev,
11742 const struct rte_flow_shared_action_conf *conf,
11743 const struct rte_flow_action_rss *rss,
11744 struct rte_flow_error *error)
11746 struct mlx5_priv *priv = dev->data->dev_private;
11747 struct mlx5_shared_action_rss *shared_action = NULL;
11748 void *queue = NULL;
11749 struct rte_flow_action_rss *origin;
11750 const uint8_t *rss_key;
11751 uint32_t queue_size = rss->queue_num * sizeof(uint16_t);
11754 RTE_SET_USED(conf);
11755 queue = mlx5_malloc(0, RTE_ALIGN_CEIL(queue_size, sizeof(void *)),
11757 shared_action = mlx5_ipool_zmalloc
11758 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], &idx);
11759 if (!shared_action || !queue) {
11760 rte_flow_error_set(error, ENOMEM,
11761 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11762 "cannot allocate resource memory");
11763 goto error_rss_init;
11765 if (idx > (1u << MLX5_SHARED_ACTION_TYPE_OFFSET)) {
11766 rte_flow_error_set(error, E2BIG,
11767 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11768 "rss action number out of range");
11769 goto error_rss_init;
11771 shared_action->ind_tbl = mlx5_malloc(MLX5_MEM_ZERO,
11772 sizeof(*shared_action->ind_tbl),
11774 if (!shared_action->ind_tbl) {
11775 rte_flow_error_set(error, ENOMEM,
11776 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
11777 "cannot allocate resource memory");
11778 goto error_rss_init;
11780 memcpy(queue, rss->queue, queue_size);
11781 shared_action->ind_tbl->queues = queue;
11782 shared_action->ind_tbl->queues_n = rss->queue_num;
11783 origin = &shared_action->origin;
11784 origin->func = rss->func;
11785 origin->level = rss->level;
11786 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
11787 origin->types = !rss->types ? ETH_RSS_IP : rss->types;
11788 /* NULL RSS key indicates default RSS key. */
11789 rss_key = !rss->key ? rss_hash_default_key : rss->key;
11790 memcpy(shared_action->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
11791 origin->key = &shared_action->key[0];
11792 origin->key_len = MLX5_RSS_HASH_KEY_LEN;
11793 origin->queue = queue;
11794 origin->queue_num = rss->queue_num;
11795 if (__flow_dv_action_rss_setup(dev, idx, shared_action, error))
11796 goto error_rss_init;
11797 rte_spinlock_init(&shared_action->action_rss_sl);
11798 __atomic_add_fetch(&shared_action->refcnt, 1, __ATOMIC_RELAXED);
11799 rte_spinlock_lock(&priv->shared_act_sl);
11800 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
11801 &priv->rss_shared_actions, idx, shared_action, next);
11802 rte_spinlock_unlock(&priv->shared_act_sl);
11805 if (shared_action) {
11806 if (shared_action->ind_tbl)
11807 mlx5_free(shared_action->ind_tbl);
11808 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
11817 * Destroy the shared RSS action.
11818 * Release related hash RX queue objects.
11821 * Pointer to the Ethernet device structure.
11823 * The shared RSS action object ID to be removed.
11824 * @param[out] error
11825 * Perform verbose error reporting if not NULL. Initialized in case of
11829 * 0 on success, otherwise negative errno value.
11832 __flow_dv_action_rss_release(struct rte_eth_dev *dev, uint32_t idx,
11833 struct rte_flow_error *error)
11835 struct mlx5_priv *priv = dev->data->dev_private;
11836 struct mlx5_shared_action_rss *shared_rss =
11837 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
11838 uint32_t old_refcnt = 1;
11840 uint16_t *queue = NULL;
11843 return rte_flow_error_set(error, EINVAL,
11844 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
11845 "invalid shared action");
11846 remaining = __flow_dv_action_rss_hrxqs_release(dev, shared_rss);
11848 return rte_flow_error_set(error, EBUSY,
11849 RTE_FLOW_ERROR_TYPE_ACTION,
11851 "shared rss hrxq has references");
11852 queue = shared_rss->ind_tbl->queues;
11853 remaining = mlx5_ind_table_obj_release(dev, shared_rss->ind_tbl, true);
11855 return rte_flow_error_set(error, EBUSY,
11856 RTE_FLOW_ERROR_TYPE_ACTION,
11858 "shared rss indirection table has"
11860 if (!__atomic_compare_exchange_n(&shared_rss->refcnt, &old_refcnt,
11861 0, 0, __ATOMIC_ACQUIRE,
11863 return rte_flow_error_set(error, EBUSY,
11864 RTE_FLOW_ERROR_TYPE_ACTION,
11866 "shared rss has references");
11868 rte_spinlock_lock(&priv->shared_act_sl);
11869 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
11870 &priv->rss_shared_actions, idx, shared_rss, next);
11871 rte_spinlock_unlock(&priv->shared_act_sl);
11872 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
11878 * Create shared action, lock free,
11879 * (mutex should be acquired by caller).
11880 * Dispatcher for action type specific call.
11883 * Pointer to the Ethernet device structure.
11885 * Shared action configuration.
11886 * @param[in] action
11887 * Action specification used to create shared action.
11888 * @param[out] error
11889 * Perform verbose error reporting if not NULL. Initialized in case of
11893 * A valid shared action handle in case of success, NULL otherwise and
11894 * rte_errno is set.
11896 static struct rte_flow_shared_action *
11897 flow_dv_action_create(struct rte_eth_dev *dev,
11898 const struct rte_flow_shared_action_conf *conf,
11899 const struct rte_flow_action *action,
11900 struct rte_flow_error *err)
11905 switch (action->type) {
11906 case RTE_FLOW_ACTION_TYPE_RSS:
11907 ret = __flow_dv_action_rss_create(dev, conf, action->conf, err);
11908 idx = (MLX5_SHARED_ACTION_TYPE_RSS <<
11909 MLX5_SHARED_ACTION_TYPE_OFFSET) | ret;
11911 case RTE_FLOW_ACTION_TYPE_AGE:
11912 ret = flow_dv_translate_create_aso_age(dev, action->conf, err);
11913 idx = (MLX5_SHARED_ACTION_TYPE_AGE <<
11914 MLX5_SHARED_ACTION_TYPE_OFFSET) | ret;
11916 struct mlx5_aso_age_action *aso_age =
11917 flow_aso_age_get_by_idx(dev, ret);
11919 if (!aso_age->age_params.context)
11920 aso_age->age_params.context =
11921 (void *)(uintptr_t)idx;
11925 rte_flow_error_set(err, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
11926 NULL, "action type not supported");
11929 return ret ? (struct rte_flow_shared_action *)(uintptr_t)idx : NULL;
11933 * Destroy the shared action.
11934 * Release action related resources on the NIC and the memory.
11935 * Lock free, (mutex should be acquired by caller).
11936 * Dispatcher for action type specific call.
11939 * Pointer to the Ethernet device structure.
11940 * @param[in] action
11941 * The shared action object to be removed.
11942 * @param[out] error
11943 * Perform verbose error reporting if not NULL. Initialized in case of
11947 * 0 on success, otherwise negative errno value.
11950 flow_dv_action_destroy(struct rte_eth_dev *dev,
11951 struct rte_flow_shared_action *action,
11952 struct rte_flow_error *error)
11954 uint32_t act_idx = (uint32_t)(uintptr_t)action;
11955 uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
11956 uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
11960 case MLX5_SHARED_ACTION_TYPE_RSS:
11961 return __flow_dv_action_rss_release(dev, idx, error);
11962 case MLX5_SHARED_ACTION_TYPE_AGE:
11963 ret = flow_dv_aso_age_release(dev, idx);
11966 * In this case, the last flow has a reference will
11967 * actually release the age action.
11969 DRV_LOG(DEBUG, "Shared age action %" PRIu32 " was"
11970 " released with references %d.", idx, ret);
11973 return rte_flow_error_set(error, ENOTSUP,
11974 RTE_FLOW_ERROR_TYPE_ACTION,
11976 "action type not supported");
11981 * Updates in place shared RSS action configuration.
11984 * Pointer to the Ethernet device structure.
11986 * The shared RSS action object ID to be updated.
11987 * @param[in] action_conf
11988 * RSS action specification used to modify *shared_rss*.
11989 * @param[out] error
11990 * Perform verbose error reporting if not NULL. Initialized in case of
11994 * 0 on success, otherwise negative errno value.
11995 * @note: currently only support update of RSS queues.
11998 __flow_dv_action_rss_update(struct rte_eth_dev *dev, uint32_t idx,
11999 const struct rte_flow_action_rss *action_conf,
12000 struct rte_flow_error *error)
12002 struct mlx5_priv *priv = dev->data->dev_private;
12003 struct mlx5_shared_action_rss *shared_rss =
12004 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
12006 void *queue = NULL;
12007 uint16_t *queue_old = NULL;
12008 uint32_t queue_size = action_conf->queue_num * sizeof(uint16_t);
12011 return rte_flow_error_set(error, EINVAL,
12012 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12013 "invalid shared action to update");
12014 queue = mlx5_malloc(MLX5_MEM_ZERO,
12015 RTE_ALIGN_CEIL(queue_size, sizeof(void *)),
12018 return rte_flow_error_set(error, ENOMEM,
12019 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12021 "cannot allocate resource memory");
12022 memcpy(queue, action_conf->queue, queue_size);
12023 MLX5_ASSERT(shared_rss->ind_tbl);
12024 rte_spinlock_lock(&shared_rss->action_rss_sl);
12025 queue_old = shared_rss->ind_tbl->queues;
12026 ret = mlx5_ind_table_obj_modify(dev, shared_rss->ind_tbl,
12027 queue, action_conf->queue_num, true);
12030 ret = rte_flow_error_set(error, rte_errno,
12031 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12032 "cannot update indirection table");
12034 mlx5_free(queue_old);
12035 shared_rss->origin.queue = queue;
12036 shared_rss->origin.queue_num = action_conf->queue_num;
12038 rte_spinlock_unlock(&shared_rss->action_rss_sl);
12043 * Updates in place shared action configuration, lock free,
12044 * (mutex should be acquired by caller).
12047 * Pointer to the Ethernet device structure.
12048 * @param[in] action
12049 * The shared action object to be updated.
12050 * @param[in] action_conf
12051 * Action specification used to modify *action*.
12052 * *action_conf* should be of type correlating with type of the *action*,
12053 * otherwise considered as invalid.
12054 * @param[out] error
12055 * Perform verbose error reporting if not NULL. Initialized in case of
12059 * 0 on success, otherwise negative errno value.
12062 flow_dv_action_update(struct rte_eth_dev *dev,
12063 struct rte_flow_shared_action *action,
12064 const void *action_conf,
12065 struct rte_flow_error *err)
12067 uint32_t act_idx = (uint32_t)(uintptr_t)action;
12068 uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
12069 uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
12072 case MLX5_SHARED_ACTION_TYPE_RSS:
12073 return __flow_dv_action_rss_update(dev, idx, action_conf, err);
12075 return rte_flow_error_set(err, ENOTSUP,
12076 RTE_FLOW_ERROR_TYPE_ACTION,
12078 "action type update not supported");
12083 flow_dv_action_query(struct rte_eth_dev *dev,
12084 const struct rte_flow_shared_action *action, void *data,
12085 struct rte_flow_error *error)
12087 struct mlx5_age_param *age_param;
12088 struct rte_flow_query_age *resp;
12089 uint32_t act_idx = (uint32_t)(uintptr_t)action;
12090 uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
12091 uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
12094 case MLX5_SHARED_ACTION_TYPE_AGE:
12095 age_param = &flow_aso_age_get_by_idx(dev, idx)->age_params;
12097 resp->aged = __atomic_load_n(&age_param->state,
12098 __ATOMIC_RELAXED) == AGE_TMOUT ?
12100 resp->sec_since_last_hit_valid = !resp->aged;
12101 if (resp->sec_since_last_hit_valid)
12102 resp->sec_since_last_hit = __atomic_load_n
12103 (&age_param->sec_since_last_hit, __ATOMIC_RELAXED);
12106 return rte_flow_error_set(error, ENOTSUP,
12107 RTE_FLOW_ERROR_TYPE_ACTION,
12109 "action type query not supported");
12114 * Query a dv flow rule for its statistics via devx.
12117 * Pointer to Ethernet device.
12119 * Pointer to the sub flow.
12121 * data retrieved by the query.
12122 * @param[out] error
12123 * Perform verbose error reporting if not NULL.
12126 * 0 on success, a negative errno value otherwise and rte_errno is set.
12129 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
12130 void *data, struct rte_flow_error *error)
12132 struct mlx5_priv *priv = dev->data->dev_private;
12133 struct rte_flow_query_count *qc = data;
12135 if (!priv->config.devx)
12136 return rte_flow_error_set(error, ENOTSUP,
12137 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12139 "counters are not supported");
12140 if (flow->counter) {
12141 uint64_t pkts, bytes;
12142 struct mlx5_flow_counter *cnt;
12144 cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
12146 int err = _flow_dv_query_count(dev, flow->counter, &pkts,
12150 return rte_flow_error_set(error, -err,
12151 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12152 NULL, "cannot read counters");
12155 qc->hits = pkts - cnt->hits;
12156 qc->bytes = bytes - cnt->bytes;
12159 cnt->bytes = bytes;
12163 return rte_flow_error_set(error, EINVAL,
12164 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12166 "counters are not available");
12170 * Query a flow rule AGE action for aging information.
12173 * Pointer to Ethernet device.
12175 * Pointer to the sub flow.
12177 * data retrieved by the query.
12178 * @param[out] error
12179 * Perform verbose error reporting if not NULL.
12182 * 0 on success, a negative errno value otherwise and rte_errno is set.
12185 flow_dv_query_age(struct rte_eth_dev *dev, struct rte_flow *flow,
12186 void *data, struct rte_flow_error *error)
12188 struct rte_flow_query_age *resp = data;
12189 struct mlx5_age_param *age_param;
12192 struct mlx5_aso_age_action *act =
12193 flow_aso_age_get_by_idx(dev, flow->age);
12195 age_param = &act->age_params;
12196 } else if (flow->counter) {
12197 age_param = flow_dv_counter_idx_get_age(dev, flow->counter);
12199 if (!age_param || !age_param->timeout)
12200 return rte_flow_error_set
12202 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12203 NULL, "cannot read age data");
12205 return rte_flow_error_set(error, EINVAL,
12206 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12207 NULL, "age data not available");
12209 resp->aged = __atomic_load_n(&age_param->state, __ATOMIC_RELAXED) ==
12211 resp->sec_since_last_hit_valid = !resp->aged;
12212 if (resp->sec_since_last_hit_valid)
12213 resp->sec_since_last_hit = __atomic_load_n
12214 (&age_param->sec_since_last_hit, __ATOMIC_RELAXED);
12221 * @see rte_flow_query()
12222 * @see rte_flow_ops
12225 flow_dv_query(struct rte_eth_dev *dev,
12226 struct rte_flow *flow __rte_unused,
12227 const struct rte_flow_action *actions __rte_unused,
12228 void *data __rte_unused,
12229 struct rte_flow_error *error __rte_unused)
12233 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
12234 switch (actions->type) {
12235 case RTE_FLOW_ACTION_TYPE_VOID:
12237 case RTE_FLOW_ACTION_TYPE_COUNT:
12238 ret = flow_dv_query_count(dev, flow, data, error);
12240 case RTE_FLOW_ACTION_TYPE_AGE:
12241 ret = flow_dv_query_age(dev, flow, data, error);
12244 return rte_flow_error_set(error, ENOTSUP,
12245 RTE_FLOW_ERROR_TYPE_ACTION,
12247 "action not supported");
12254 * Destroy the meter table set.
12255 * Lock free, (mutex should be acquired by caller).
12258 * Pointer to Ethernet device.
12260 * Pointer to the meter table set.
12266 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
12267 struct mlx5_meter_domains_infos *tbl)
12269 struct mlx5_priv *priv = dev->data->dev_private;
12270 struct mlx5_meter_domains_infos *mtd =
12271 (struct mlx5_meter_domains_infos *)tbl;
12273 if (!mtd || !priv->config.dv_flow_en)
12275 if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
12276 claim_zero(mlx5_flow_os_destroy_flow
12277 (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
12278 if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
12279 claim_zero(mlx5_flow_os_destroy_flow
12280 (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
12281 if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
12282 claim_zero(mlx5_flow_os_destroy_flow
12283 (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
12284 if (mtd->egress.color_matcher)
12285 claim_zero(mlx5_flow_os_destroy_flow_matcher
12286 (mtd->egress.color_matcher));
12287 if (mtd->egress.any_matcher)
12288 claim_zero(mlx5_flow_os_destroy_flow_matcher
12289 (mtd->egress.any_matcher));
12290 if (mtd->egress.tbl)
12291 flow_dv_tbl_resource_release(MLX5_SH(dev), mtd->egress.tbl);
12292 if (mtd->egress.sfx_tbl)
12293 flow_dv_tbl_resource_release(MLX5_SH(dev), mtd->egress.sfx_tbl);
12294 if (mtd->ingress.color_matcher)
12295 claim_zero(mlx5_flow_os_destroy_flow_matcher
12296 (mtd->ingress.color_matcher));
12297 if (mtd->ingress.any_matcher)
12298 claim_zero(mlx5_flow_os_destroy_flow_matcher
12299 (mtd->ingress.any_matcher));
12300 if (mtd->ingress.tbl)
12301 flow_dv_tbl_resource_release(MLX5_SH(dev), mtd->ingress.tbl);
12302 if (mtd->ingress.sfx_tbl)
12303 flow_dv_tbl_resource_release(MLX5_SH(dev),
12304 mtd->ingress.sfx_tbl);
12305 if (mtd->transfer.color_matcher)
12306 claim_zero(mlx5_flow_os_destroy_flow_matcher
12307 (mtd->transfer.color_matcher));
12308 if (mtd->transfer.any_matcher)
12309 claim_zero(mlx5_flow_os_destroy_flow_matcher
12310 (mtd->transfer.any_matcher));
12311 if (mtd->transfer.tbl)
12312 flow_dv_tbl_resource_release(MLX5_SH(dev), mtd->transfer.tbl);
12313 if (mtd->transfer.sfx_tbl)
12314 flow_dv_tbl_resource_release(MLX5_SH(dev),
12315 mtd->transfer.sfx_tbl);
12316 if (mtd->drop_actn)
12317 claim_zero(mlx5_flow_os_destroy_flow_action(mtd->drop_actn));
12322 /* Number of meter flow actions, count and jump or count and drop. */
12323 #define METER_ACTIONS 2
12326 * Create specify domain meter table and suffix table.
12329 * Pointer to Ethernet device.
12330 * @param[in,out] mtb
12331 * Pointer to DV meter table set.
12332 * @param[in] egress
12334 * @param[in] transfer
12336 * @param[in] color_reg_c_idx
12337 * Reg C index for color match.
12340 * 0 on success, -1 otherwise and rte_errno is set.
12343 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
12344 struct mlx5_meter_domains_infos *mtb,
12345 uint8_t egress, uint8_t transfer,
12346 uint32_t color_reg_c_idx)
12348 struct mlx5_priv *priv = dev->data->dev_private;
12349 struct mlx5_dev_ctx_shared *sh = priv->sh;
12350 struct mlx5_flow_dv_match_params mask = {
12351 .size = sizeof(mask.buf),
12353 struct mlx5_flow_dv_match_params value = {
12354 .size = sizeof(value.buf),
12356 struct mlx5dv_flow_matcher_attr dv_attr = {
12357 .type = IBV_FLOW_ATTR_NORMAL,
12359 .match_criteria_enable = 0,
12360 .match_mask = (void *)&mask,
12362 void *actions[METER_ACTIONS];
12363 struct mlx5_meter_domain_info *dtb;
12364 struct rte_flow_error error;
12369 dtb = &mtb->transfer;
12371 dtb = &mtb->egress;
12373 dtb = &mtb->ingress;
12374 /* Create the meter table with METER level. */
12375 dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
12376 egress, transfer, false, NULL, 0,
12379 DRV_LOG(ERR, "Failed to create meter policer table.");
12382 /* Create the meter suffix table with SUFFIX level. */
12383 dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
12384 MLX5_FLOW_TABLE_LEVEL_SUFFIX,
12385 egress, transfer, false, NULL, 0,
12387 if (!dtb->sfx_tbl) {
12388 DRV_LOG(ERR, "Failed to create meter suffix table.");
12391 /* Create matchers, Any and Color. */
12392 dv_attr.priority = 3;
12393 dv_attr.match_criteria_enable = 0;
12394 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
12395 &dtb->any_matcher);
12397 DRV_LOG(ERR, "Failed to create meter"
12398 " policer default matcher.");
12401 dv_attr.priority = 0;
12402 dv_attr.match_criteria_enable =
12403 1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
12404 flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
12405 rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
12406 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
12407 &dtb->color_matcher);
12409 DRV_LOG(ERR, "Failed to create meter policer color matcher.");
12412 if (mtb->count_actns[RTE_MTR_DROPPED])
12413 actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
12414 actions[i++] = mtb->drop_actn;
12415 /* Default rule: lowest priority, match any, actions: drop. */
12416 ret = mlx5_flow_os_create_flow(dtb->any_matcher, (void *)&value, i,
12418 &dtb->policer_rules[RTE_MTR_DROPPED]);
12420 DRV_LOG(ERR, "Failed to create meter policer drop rule.");
12429 * Create the needed meter and suffix tables.
12430 * Lock free, (mutex should be acquired by caller).
12433 * Pointer to Ethernet device.
12435 * Pointer to the flow meter.
12438 * Pointer to table set on success, NULL otherwise and rte_errno is set.
12440 static struct mlx5_meter_domains_infos *
12441 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
12442 const struct mlx5_flow_meter *fm)
12444 struct mlx5_priv *priv = dev->data->dev_private;
12445 struct mlx5_meter_domains_infos *mtb;
12449 if (!priv->mtr_en) {
12450 rte_errno = ENOTSUP;
12453 mtb = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*mtb), 0, SOCKET_ID_ANY);
12455 DRV_LOG(ERR, "Failed to allocate memory for meter.");
12458 /* Create meter count actions */
12459 for (i = 0; i <= RTE_MTR_DROPPED; i++) {
12460 struct mlx5_flow_counter *cnt;
12461 if (!fm->policer_stats.cnt[i])
12463 cnt = flow_dv_counter_get_by_idx(dev,
12464 fm->policer_stats.cnt[i], NULL);
12465 mtb->count_actns[i] = cnt->action;
12467 /* Create drop action. */
12468 ret = mlx5_flow_os_create_flow_action_drop(&mtb->drop_actn);
12470 DRV_LOG(ERR, "Failed to create drop action.");
12473 /* Egress meter table. */
12474 ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
12476 DRV_LOG(ERR, "Failed to prepare egress meter table.");
12479 /* Ingress meter table. */
12480 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
12482 DRV_LOG(ERR, "Failed to prepare ingress meter table.");
12485 /* FDB meter table. */
12486 if (priv->config.dv_esw_en) {
12487 ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
12488 priv->mtr_color_reg);
12490 DRV_LOG(ERR, "Failed to prepare fdb meter table.");
12496 flow_dv_destroy_mtr_tbl(dev, mtb);
12501 * Destroy domain policer rule.
12504 * Pointer to domain table.
12507 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
12511 for (i = 0; i < RTE_MTR_DROPPED; i++) {
12512 if (dt->policer_rules[i]) {
12513 claim_zero(mlx5_flow_os_destroy_flow
12514 (dt->policer_rules[i]));
12515 dt->policer_rules[i] = NULL;
12518 if (dt->jump_actn) {
12519 claim_zero(mlx5_flow_os_destroy_flow_action(dt->jump_actn));
12520 dt->jump_actn = NULL;
12525 * Destroy policer rules.
12528 * Pointer to Ethernet device.
12530 * Pointer to flow meter structure.
12532 * Pointer to flow attributes.
12538 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
12539 const struct mlx5_flow_meter *fm,
12540 const struct rte_flow_attr *attr)
12542 struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
12547 flow_dv_destroy_domain_policer_rule(&mtb->egress);
12549 flow_dv_destroy_domain_policer_rule(&mtb->ingress);
12550 if (attr->transfer)
12551 flow_dv_destroy_domain_policer_rule(&mtb->transfer);
12556 * Create specify domain meter policer rule.
12559 * Pointer to flow meter structure.
12561 * Pointer to DV meter table set.
12562 * @param[in] mtr_reg_c
12563 * Color match REG_C.
12566 * 0 on success, -1 otherwise.
12569 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
12570 struct mlx5_meter_domain_info *dtb,
12573 struct mlx5_flow_dv_match_params matcher = {
12574 .size = sizeof(matcher.buf),
12576 struct mlx5_flow_dv_match_params value = {
12577 .size = sizeof(value.buf),
12579 struct mlx5_meter_domains_infos *mtb = fm->mfts;
12580 void *actions[METER_ACTIONS];
12584 /* Create jump action. */
12585 if (!dtb->jump_actn)
12586 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
12587 (dtb->sfx_tbl->obj, &dtb->jump_actn);
12589 DRV_LOG(ERR, "Failed to create policer jump action.");
12592 for (i = 0; i < RTE_MTR_DROPPED; i++) {
12595 flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
12596 rte_col_2_mlx5_col(i), UINT8_MAX);
12597 if (mtb->count_actns[i])
12598 actions[j++] = mtb->count_actns[i];
12599 if (fm->action[i] == MTR_POLICER_ACTION_DROP)
12600 actions[j++] = mtb->drop_actn;
12602 actions[j++] = dtb->jump_actn;
12603 ret = mlx5_flow_os_create_flow(dtb->color_matcher,
12604 (void *)&value, j, actions,
12605 &dtb->policer_rules[i]);
12607 DRV_LOG(ERR, "Failed to create policer rule.");
12618 * Create policer rules.
12621 * Pointer to Ethernet device.
12623 * Pointer to flow meter structure.
12625 * Pointer to flow attributes.
12628 * 0 on success, -1 otherwise.
12631 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
12632 struct mlx5_flow_meter *fm,
12633 const struct rte_flow_attr *attr)
12635 struct mlx5_priv *priv = dev->data->dev_private;
12636 struct mlx5_meter_domains_infos *mtb = fm->mfts;
12639 if (attr->egress) {
12640 ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
12641 priv->mtr_color_reg);
12643 DRV_LOG(ERR, "Failed to create egress policer.");
12647 if (attr->ingress) {
12648 ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
12649 priv->mtr_color_reg);
12651 DRV_LOG(ERR, "Failed to create ingress policer.");
12655 if (attr->transfer) {
12656 ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
12657 priv->mtr_color_reg);
12659 DRV_LOG(ERR, "Failed to create transfer policer.");
12665 flow_dv_destroy_policer_rules(dev, fm, attr);
12670 * Validate the batch counter support in root table.
12672 * Create a simple flow with invalid counter and drop action on root table to
12673 * validate if batch counter with offset on root table is supported or not.
12676 * Pointer to rte_eth_dev structure.
12679 * 0 on success, a negative errno value otherwise and rte_errno is set.
12682 mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev)
12684 struct mlx5_priv *priv = dev->data->dev_private;
12685 struct mlx5_dev_ctx_shared *sh = priv->sh;
12686 struct mlx5_flow_dv_match_params mask = {
12687 .size = sizeof(mask.buf),
12689 struct mlx5_flow_dv_match_params value = {
12690 .size = sizeof(value.buf),
12692 struct mlx5dv_flow_matcher_attr dv_attr = {
12693 .type = IBV_FLOW_ATTR_NORMAL,
12695 .match_criteria_enable = 0,
12696 .match_mask = (void *)&mask,
12698 void *actions[2] = { 0 };
12699 struct mlx5_flow_tbl_resource *tbl = NULL;
12700 struct mlx5_devx_obj *dcs = NULL;
12701 void *matcher = NULL;
12705 tbl = flow_dv_tbl_resource_get(dev, 0, 0, 0, false, NULL, 0, 0, NULL);
12708 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
12711 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, UINT16_MAX,
12715 actions[1] = priv->drop_queue.hrxq->action;
12716 dv_attr.match_criteria_enable = flow_dv_matcher_enable(mask.buf);
12717 ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
12721 ret = mlx5_flow_os_create_flow(matcher, (void *)&value, 2,
12725 * If batch counter with offset is not supported, the driver will not
12726 * validate the invalid offset value, flow create should success.
12727 * In this case, it means batch counter is not supported in root table.
12729 * Otherwise, if flow create is failed, counter offset is supported.
12732 DRV_LOG(INFO, "Batch counter is not supported in root "
12733 "table. Switch to fallback mode.");
12734 rte_errno = ENOTSUP;
12736 claim_zero(mlx5_flow_os_destroy_flow(flow));
12738 /* Check matcher to make sure validate fail at flow create. */
12739 if (!matcher || (matcher && errno != EINVAL))
12740 DRV_LOG(ERR, "Unexpected error in counter offset "
12741 "support detection");
12745 claim_zero(mlx5_flow_os_destroy_flow_action(actions[0]));
12747 claim_zero(mlx5_flow_os_destroy_flow_matcher(matcher));
12749 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
12751 claim_zero(mlx5_devx_cmd_destroy(dcs));
12756 * Query a devx counter.
12759 * Pointer to the Ethernet device structure.
12761 * Index to the flow counter.
12763 * Set to clear the counter statistics.
12765 * The statistics value of packets.
12766 * @param[out] bytes
12767 * The statistics value of bytes.
12770 * 0 on success, otherwise return -1.
12773 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
12774 uint64_t *pkts, uint64_t *bytes)
12776 struct mlx5_priv *priv = dev->data->dev_private;
12777 struct mlx5_flow_counter *cnt;
12778 uint64_t inn_pkts, inn_bytes;
12781 if (!priv->config.devx)
12784 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
12787 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
12788 *pkts = inn_pkts - cnt->hits;
12789 *bytes = inn_bytes - cnt->bytes;
12791 cnt->hits = inn_pkts;
12792 cnt->bytes = inn_bytes;
12798 * Get aged-out flows.
12801 * Pointer to the Ethernet device structure.
12802 * @param[in] context
12803 * The address of an array of pointers to the aged-out flows contexts.
12804 * @param[in] nb_contexts
12805 * The length of context array pointers.
12806 * @param[out] error
12807 * Perform verbose error reporting if not NULL. Initialized in case of
12811 * how many contexts get in success, otherwise negative errno value.
12812 * if nb_contexts is 0, return the amount of all aged contexts.
12813 * if nb_contexts is not 0 , return the amount of aged flows reported
12814 * in the context array.
12815 * @note: only stub for now
12818 flow_get_aged_flows(struct rte_eth_dev *dev,
12820 uint32_t nb_contexts,
12821 struct rte_flow_error *error)
12823 struct mlx5_priv *priv = dev->data->dev_private;
12824 struct mlx5_age_info *age_info;
12825 struct mlx5_age_param *age_param;
12826 struct mlx5_flow_counter *counter;
12827 struct mlx5_aso_age_action *act;
12830 if (nb_contexts && !context)
12831 return rte_flow_error_set(error, EINVAL,
12832 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12833 NULL, "empty context");
12834 age_info = GET_PORT_AGE_INFO(priv);
12835 rte_spinlock_lock(&age_info->aged_sl);
12836 LIST_FOREACH(act, &age_info->aged_aso, next) {
12839 context[nb_flows - 1] =
12840 act->age_params.context;
12841 if (!(--nb_contexts))
12845 TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
12848 age_param = MLX5_CNT_TO_AGE(counter);
12849 context[nb_flows - 1] = age_param->context;
12850 if (!(--nb_contexts))
12854 rte_spinlock_unlock(&age_info->aged_sl);
12855 MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
12860 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
12863 flow_dv_counter_allocate(struct rte_eth_dev *dev)
12865 return flow_dv_counter_alloc(dev, 0);
12869 * Validate shared action.
12870 * Dispatcher for action type specific validation.
12873 * Pointer to the Ethernet device structure.
12875 * Shared action configuration.
12876 * @param[in] action
12877 * The shared action object to validate.
12878 * @param[out] error
12879 * Perform verbose error reporting if not NULL. Initialized in case of
12883 * 0 on success, otherwise negative errno value.
12886 flow_dv_action_validate(struct rte_eth_dev *dev,
12887 const struct rte_flow_shared_action_conf *conf,
12888 const struct rte_flow_action *action,
12889 struct rte_flow_error *err)
12891 struct mlx5_priv *priv = dev->data->dev_private;
12893 RTE_SET_USED(conf);
12894 switch (action->type) {
12895 case RTE_FLOW_ACTION_TYPE_RSS:
12896 return mlx5_validate_action_rss(dev, action, err);
12897 case RTE_FLOW_ACTION_TYPE_AGE:
12898 if (!priv->sh->aso_age_mng)
12899 return rte_flow_error_set(err, ENOTSUP,
12900 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12902 "shared age action not supported");
12903 return flow_dv_validate_action_age(0, action, dev, err);
12905 return rte_flow_error_set(err, ENOTSUP,
12906 RTE_FLOW_ERROR_TYPE_ACTION,
12908 "action type not supported");
12913 flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
12915 struct mlx5_priv *priv = dev->data->dev_private;
12918 if ((domains & MLX5_DOMAIN_BIT_NIC_RX) && priv->sh->rx_domain != NULL) {
12919 ret = mlx5_os_flow_dr_sync_domain(priv->sh->rx_domain,
12924 if ((domains & MLX5_DOMAIN_BIT_NIC_TX) && priv->sh->tx_domain != NULL) {
12925 ret = mlx5_os_flow_dr_sync_domain(priv->sh->tx_domain, flags);
12929 if ((domains & MLX5_DOMAIN_BIT_FDB) && priv->sh->fdb_domain != NULL) {
12930 ret = mlx5_os_flow_dr_sync_domain(priv->sh->fdb_domain, flags);
12937 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
12938 .validate = flow_dv_validate,
12939 .prepare = flow_dv_prepare,
12940 .translate = flow_dv_translate,
12941 .apply = flow_dv_apply,
12942 .remove = flow_dv_remove,
12943 .destroy = flow_dv_destroy,
12944 .query = flow_dv_query,
12945 .create_mtr_tbls = flow_dv_create_mtr_tbl,
12946 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
12947 .create_policer_rules = flow_dv_create_policer_rules,
12948 .destroy_policer_rules = flow_dv_destroy_policer_rules,
12949 .counter_alloc = flow_dv_counter_allocate,
12950 .counter_free = flow_dv_counter_free,
12951 .counter_query = flow_dv_counter_query,
12952 .get_aged_flows = flow_get_aged_flows,
12953 .action_validate = flow_dv_action_validate,
12954 .action_create = flow_dv_action_create,
12955 .action_destroy = flow_dv_action_destroy,
12956 .action_update = flow_dv_action_update,
12957 .action_query = flow_dv_action_query,
12958 .sync_domain = flow_dv_sync_domain,
12961 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */