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 <ethdev_driver.h>
15 #include <rte_flow_driver.h>
16 #include <rte_malloc.h>
17 #include <rte_cycles.h>
18 #include <rte_bus_pci.h>
21 #include <rte_vxlan.h>
23 #include <rte_eal_paging.h>
26 #include <rte_mtr_driver.h>
27 #include <rte_tailq.h>
29 #include <mlx5_glue.h>
30 #include <mlx5_devx_cmds.h>
32 #include <mlx5_malloc.h>
34 #include "mlx5_defs.h"
36 #include "mlx5_common_os.h"
37 #include "mlx5_flow.h"
38 #include "mlx5_flow_os.h"
41 #include "rte_pmd_mlx5.h"
43 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
45 #ifndef HAVE_IBV_FLOW_DEVX_COUNTERS
46 #define MLX5DV_FLOW_ACTION_COUNTERS_DEVX 0
49 #ifndef HAVE_MLX5DV_DR_ESWITCH
50 #ifndef MLX5DV_FLOW_TABLE_TYPE_FDB
51 #define MLX5DV_FLOW_TABLE_TYPE_FDB 0
55 #ifndef HAVE_MLX5DV_DR
56 #define MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL 1
59 /* VLAN header definitions */
60 #define MLX5DV_FLOW_VLAN_PCP_SHIFT 13
61 #define MLX5DV_FLOW_VLAN_PCP_MASK (0x7 << MLX5DV_FLOW_VLAN_PCP_SHIFT)
62 #define MLX5DV_FLOW_VLAN_VID_MASK 0x0fff
63 #define MLX5DV_FLOW_VLAN_PCP_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK)
64 #define MLX5DV_FLOW_VLAN_VID_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_VID_MASK)
79 flow_dv_tbl_resource_release(struct mlx5_dev_ctx_shared *sh,
80 struct mlx5_flow_tbl_resource *tbl);
83 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
84 uint32_t encap_decap_idx);
87 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
90 flow_dv_shared_rss_action_release(struct rte_eth_dev *dev, uint32_t srss);
93 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
97 flow_dv_get_esw_manager_vport_id(struct rte_eth_dev *dev)
99 struct mlx5_priv *priv = dev->data->dev_private;
100 struct mlx5_common_device *cdev = priv->sh->cdev;
102 if (cdev->config.hca_attr.esw_mgr_vport_id_valid)
103 return (int16_t)cdev->config.hca_attr.esw_mgr_vport_id;
105 if (priv->pci_dev == NULL)
107 switch (priv->pci_dev->id.device_id) {
108 case PCI_DEVICE_ID_MELLANOX_CONNECTX5BF:
109 case PCI_DEVICE_ID_MELLANOX_CONNECTX6DXBF:
110 case PCI_DEVICE_ID_MELLANOX_CONNECTX7BF:
111 return (int16_t)0xfffe;
118 * Initialize flow attributes structure according to flow items' types.
120 * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
121 * mode. For tunnel mode, the items to be modified are the outermost ones.
124 * Pointer to item specification.
126 * Pointer to flow attributes structure.
127 * @param[in] dev_flow
128 * Pointer to the sub flow.
129 * @param[in] tunnel_decap
130 * Whether action is after tunnel decapsulation.
133 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
134 struct mlx5_flow *dev_flow, bool tunnel_decap)
136 uint64_t layers = dev_flow->handle->layers;
139 * If layers is already initialized, it means this dev_flow is the
140 * suffix flow, the layers flags is set by the prefix flow. Need to
141 * use the layer flags from prefix flow as the suffix flow may not
142 * have the user defined items as the flow is split.
145 if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
147 else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
149 if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
151 else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
156 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
157 uint8_t next_protocol = 0xff;
158 switch (item->type) {
159 case RTE_FLOW_ITEM_TYPE_GRE:
160 case RTE_FLOW_ITEM_TYPE_NVGRE:
161 case RTE_FLOW_ITEM_TYPE_VXLAN:
162 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
163 case RTE_FLOW_ITEM_TYPE_GENEVE:
164 case RTE_FLOW_ITEM_TYPE_MPLS:
168 case RTE_FLOW_ITEM_TYPE_IPV4:
171 if (item->mask != NULL &&
172 ((const struct rte_flow_item_ipv4 *)
173 item->mask)->hdr.next_proto_id)
175 ((const struct rte_flow_item_ipv4 *)
176 (item->spec))->hdr.next_proto_id &
177 ((const struct rte_flow_item_ipv4 *)
178 (item->mask))->hdr.next_proto_id;
179 if ((next_protocol == IPPROTO_IPIP ||
180 next_protocol == IPPROTO_IPV6) && tunnel_decap)
183 case RTE_FLOW_ITEM_TYPE_IPV6:
186 if (item->mask != NULL &&
187 ((const struct rte_flow_item_ipv6 *)
188 item->mask)->hdr.proto)
190 ((const struct rte_flow_item_ipv6 *)
191 (item->spec))->hdr.proto &
192 ((const struct rte_flow_item_ipv6 *)
193 (item->mask))->hdr.proto;
194 if ((next_protocol == IPPROTO_IPIP ||
195 next_protocol == IPPROTO_IPV6) && tunnel_decap)
198 case RTE_FLOW_ITEM_TYPE_UDP:
202 case RTE_FLOW_ITEM_TYPE_TCP:
214 * Convert rte_mtr_color to mlx5 color.
223 rte_col_2_mlx5_col(enum rte_color rcol)
226 case RTE_COLOR_GREEN:
227 return MLX5_FLOW_COLOR_GREEN;
228 case RTE_COLOR_YELLOW:
229 return MLX5_FLOW_COLOR_YELLOW;
231 return MLX5_FLOW_COLOR_RED;
235 return MLX5_FLOW_COLOR_UNDEFINED;
238 struct field_modify_info {
239 uint32_t size; /* Size of field in protocol header, in bytes. */
240 uint32_t offset; /* Offset of field in protocol header, in bytes. */
241 enum mlx5_modification_field id;
244 struct field_modify_info modify_eth[] = {
245 {4, 0, MLX5_MODI_OUT_DMAC_47_16},
246 {2, 4, MLX5_MODI_OUT_DMAC_15_0},
247 {4, 6, MLX5_MODI_OUT_SMAC_47_16},
248 {2, 10, MLX5_MODI_OUT_SMAC_15_0},
252 struct field_modify_info modify_vlan_out_first_vid[] = {
253 /* Size in bits !!! */
254 {12, 0, MLX5_MODI_OUT_FIRST_VID},
258 struct field_modify_info modify_ipv4[] = {
259 {1, 1, MLX5_MODI_OUT_IP_DSCP},
260 {1, 8, MLX5_MODI_OUT_IPV4_TTL},
261 {4, 12, MLX5_MODI_OUT_SIPV4},
262 {4, 16, MLX5_MODI_OUT_DIPV4},
266 struct field_modify_info modify_ipv6[] = {
267 {1, 0, MLX5_MODI_OUT_IP_DSCP},
268 {1, 7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
269 {4, 8, MLX5_MODI_OUT_SIPV6_127_96},
270 {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
271 {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
272 {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
273 {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
274 {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
275 {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
276 {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
280 struct field_modify_info modify_udp[] = {
281 {2, 0, MLX5_MODI_OUT_UDP_SPORT},
282 {2, 2, MLX5_MODI_OUT_UDP_DPORT},
286 struct field_modify_info modify_tcp[] = {
287 {2, 0, MLX5_MODI_OUT_TCP_SPORT},
288 {2, 2, MLX5_MODI_OUT_TCP_DPORT},
289 {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
290 {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
295 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
296 uint8_t next_protocol, uint64_t *item_flags,
299 MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
300 item->type == RTE_FLOW_ITEM_TYPE_IPV6);
301 if (next_protocol == IPPROTO_IPIP) {
302 *item_flags |= MLX5_FLOW_LAYER_IPIP;
305 if (next_protocol == IPPROTO_IPV6) {
306 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
311 static inline struct mlx5_hlist *
312 flow_dv_hlist_prepare(struct mlx5_dev_ctx_shared *sh, struct mlx5_hlist **phl,
313 const char *name, uint32_t size, bool direct_key,
314 bool lcores_share, void *ctx,
315 mlx5_list_create_cb cb_create,
316 mlx5_list_match_cb cb_match,
317 mlx5_list_remove_cb cb_remove,
318 mlx5_list_clone_cb cb_clone,
319 mlx5_list_clone_free_cb cb_clone_free,
320 struct rte_flow_error *error)
322 struct mlx5_hlist *hl;
323 struct mlx5_hlist *expected = NULL;
324 char s[MLX5_NAME_SIZE];
326 hl = __atomic_load_n(phl, __ATOMIC_SEQ_CST);
329 snprintf(s, sizeof(s), "%s_%s", sh->ibdev_name, name);
330 hl = mlx5_hlist_create(s, size, direct_key, lcores_share,
331 ctx, cb_create, cb_match, cb_remove, cb_clone,
334 DRV_LOG(ERR, "%s hash creation failed", name);
335 rte_flow_error_set(error, ENOMEM,
336 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
337 "cannot allocate resource memory");
340 if (!__atomic_compare_exchange_n(phl, &expected, hl, false,
343 mlx5_hlist_destroy(hl);
344 hl = __atomic_load_n(phl, __ATOMIC_SEQ_CST);
349 /* Update VLAN's VID/PCP based on input rte_flow_action.
352 * Pointer to struct rte_flow_action.
354 * Pointer to struct rte_vlan_hdr.
357 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
358 struct rte_vlan_hdr *vlan)
361 if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
363 ((const struct rte_flow_action_of_set_vlan_pcp *)
364 action->conf)->vlan_pcp;
365 vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
366 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
367 vlan->vlan_tci |= vlan_tci;
368 } else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
369 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
370 vlan->vlan_tci |= rte_be_to_cpu_16
371 (((const struct rte_flow_action_of_set_vlan_vid *)
372 action->conf)->vlan_vid);
377 * Fetch 1, 2, 3 or 4 byte field from the byte array
378 * and return as unsigned integer in host-endian format.
381 * Pointer to data array.
383 * Size of field to extract.
386 * converted field in host endian format.
388 static inline uint32_t
389 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
398 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
401 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
402 ret = (ret << 8) | *(data + sizeof(uint16_t));
405 ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
416 * Convert modify-header action to DV specification.
418 * Data length of each action is determined by provided field description
419 * and the item mask. Data bit offset and width of each action is determined
420 * by provided item mask.
423 * Pointer to item specification.
425 * Pointer to field modification information.
426 * For MLX5_MODIFICATION_TYPE_SET specifies destination field.
427 * For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
428 * For MLX5_MODIFICATION_TYPE_COPY specifies source field.
430 * Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
431 * Negative offset value sets the same offset as source offset.
432 * size field is ignored, value is taken from source field.
433 * @param[in,out] resource
434 * Pointer to the modify-header resource.
436 * Type of modification.
438 * Pointer to the error structure.
441 * 0 on success, a negative errno value otherwise and rte_errno is set.
444 flow_dv_convert_modify_action(struct rte_flow_item *item,
445 struct field_modify_info *field,
446 struct field_modify_info *dcopy,
447 struct mlx5_flow_dv_modify_hdr_resource *resource,
448 uint32_t type, struct rte_flow_error *error)
450 uint32_t i = resource->actions_num;
451 struct mlx5_modification_cmd *actions = resource->actions;
452 uint32_t carry_b = 0;
455 * The item and mask are provided in big-endian format.
456 * The fields should be presented as in big-endian format either.
457 * Mask must be always present, it defines the actual field width.
459 MLX5_ASSERT(item->mask);
460 MLX5_ASSERT(field->size);
466 bool next_field = true;
467 bool next_dcopy = true;
469 if (i >= MLX5_MAX_MODIFY_NUM)
470 return rte_flow_error_set(error, EINVAL,
471 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
472 "too many items to modify");
473 /* Fetch variable byte size mask from the array. */
474 mask = flow_dv_fetch_field((const uint8_t *)item->mask +
475 field->offset, field->size);
480 /* Deduce actual data width in bits from mask value. */
481 off_b = rte_bsf32(mask) + carry_b;
482 size_b = sizeof(uint32_t) * CHAR_BIT -
483 off_b - __builtin_clz(mask);
485 actions[i] = (struct mlx5_modification_cmd) {
489 .length = (size_b == sizeof(uint32_t) * CHAR_BIT) ?
492 if (type == MLX5_MODIFICATION_TYPE_COPY) {
494 actions[i].dst_field = dcopy->id;
495 actions[i].dst_offset =
496 (int)dcopy->offset < 0 ? off_b : dcopy->offset;
497 /* Convert entire record to big-endian format. */
498 actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
500 * Destination field overflow. Copy leftovers of
501 * a source field to the next destination field.
504 if ((size_b > dcopy->size * CHAR_BIT - dcopy->offset) &&
507 dcopy->size * CHAR_BIT - dcopy->offset;
508 carry_b = actions[i].length;
512 * Not enough bits in a source filed to fill a
513 * destination field. Switch to the next source.
515 if ((size_b < dcopy->size * CHAR_BIT - dcopy->offset) &&
516 (size_b == field->size * CHAR_BIT - off_b)) {
518 field->size * CHAR_BIT - off_b;
519 dcopy->offset += actions[i].length;
525 MLX5_ASSERT(item->spec);
526 data = flow_dv_fetch_field((const uint8_t *)item->spec +
527 field->offset, field->size);
528 /* Shift out the trailing masked bits from data. */
529 data = (data & mask) >> off_b;
530 actions[i].data1 = rte_cpu_to_be_32(data);
532 /* Convert entire record to expected big-endian format. */
533 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
537 } while (field->size);
538 if (resource->actions_num == i)
539 return rte_flow_error_set(error, EINVAL,
540 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
541 "invalid modification flow item");
542 resource->actions_num = i;
547 * Convert modify-header set IPv4 address action to DV specification.
549 * @param[in,out] resource
550 * Pointer to the modify-header resource.
552 * Pointer to action specification.
554 * Pointer to the error structure.
557 * 0 on success, a negative errno value otherwise and rte_errno is set.
560 flow_dv_convert_action_modify_ipv4
561 (struct mlx5_flow_dv_modify_hdr_resource *resource,
562 const struct rte_flow_action *action,
563 struct rte_flow_error *error)
565 const struct rte_flow_action_set_ipv4 *conf =
566 (const struct rte_flow_action_set_ipv4 *)(action->conf);
567 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
568 struct rte_flow_item_ipv4 ipv4;
569 struct rte_flow_item_ipv4 ipv4_mask;
571 memset(&ipv4, 0, sizeof(ipv4));
572 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
573 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
574 ipv4.hdr.src_addr = conf->ipv4_addr;
575 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
577 ipv4.hdr.dst_addr = conf->ipv4_addr;
578 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
581 item.mask = &ipv4_mask;
582 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
583 MLX5_MODIFICATION_TYPE_SET, error);
587 * Convert modify-header set IPv6 address action to DV specification.
589 * @param[in,out] resource
590 * Pointer to the modify-header resource.
592 * Pointer to action specification.
594 * Pointer to the error structure.
597 * 0 on success, a negative errno value otherwise and rte_errno is set.
600 flow_dv_convert_action_modify_ipv6
601 (struct mlx5_flow_dv_modify_hdr_resource *resource,
602 const struct rte_flow_action *action,
603 struct rte_flow_error *error)
605 const struct rte_flow_action_set_ipv6 *conf =
606 (const struct rte_flow_action_set_ipv6 *)(action->conf);
607 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
608 struct rte_flow_item_ipv6 ipv6;
609 struct rte_flow_item_ipv6 ipv6_mask;
611 memset(&ipv6, 0, sizeof(ipv6));
612 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
613 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
614 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
615 sizeof(ipv6.hdr.src_addr));
616 memcpy(&ipv6_mask.hdr.src_addr,
617 &rte_flow_item_ipv6_mask.hdr.src_addr,
618 sizeof(ipv6.hdr.src_addr));
620 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
621 sizeof(ipv6.hdr.dst_addr));
622 memcpy(&ipv6_mask.hdr.dst_addr,
623 &rte_flow_item_ipv6_mask.hdr.dst_addr,
624 sizeof(ipv6.hdr.dst_addr));
627 item.mask = &ipv6_mask;
628 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
629 MLX5_MODIFICATION_TYPE_SET, error);
633 * Convert modify-header set MAC address action to DV specification.
635 * @param[in,out] resource
636 * Pointer to the modify-header resource.
638 * Pointer to action specification.
640 * Pointer to the error structure.
643 * 0 on success, a negative errno value otherwise and rte_errno is set.
646 flow_dv_convert_action_modify_mac
647 (struct mlx5_flow_dv_modify_hdr_resource *resource,
648 const struct rte_flow_action *action,
649 struct rte_flow_error *error)
651 const struct rte_flow_action_set_mac *conf =
652 (const struct rte_flow_action_set_mac *)(action->conf);
653 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
654 struct rte_flow_item_eth eth;
655 struct rte_flow_item_eth eth_mask;
657 memset(ð, 0, sizeof(eth));
658 memset(ð_mask, 0, sizeof(eth_mask));
659 if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
660 memcpy(ð.src.addr_bytes, &conf->mac_addr,
661 sizeof(eth.src.addr_bytes));
662 memcpy(ð_mask.src.addr_bytes,
663 &rte_flow_item_eth_mask.src.addr_bytes,
664 sizeof(eth_mask.src.addr_bytes));
666 memcpy(ð.dst.addr_bytes, &conf->mac_addr,
667 sizeof(eth.dst.addr_bytes));
668 memcpy(ð_mask.dst.addr_bytes,
669 &rte_flow_item_eth_mask.dst.addr_bytes,
670 sizeof(eth_mask.dst.addr_bytes));
673 item.mask = ð_mask;
674 return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
675 MLX5_MODIFICATION_TYPE_SET, error);
679 * Convert modify-header set VLAN VID action to DV specification.
681 * @param[in,out] resource
682 * Pointer to the modify-header resource.
684 * Pointer to action specification.
686 * Pointer to the error structure.
689 * 0 on success, a negative errno value otherwise and rte_errno is set.
692 flow_dv_convert_action_modify_vlan_vid
693 (struct mlx5_flow_dv_modify_hdr_resource *resource,
694 const struct rte_flow_action *action,
695 struct rte_flow_error *error)
697 const struct rte_flow_action_of_set_vlan_vid *conf =
698 (const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
699 int i = resource->actions_num;
700 struct mlx5_modification_cmd *actions = resource->actions;
701 struct field_modify_info *field = modify_vlan_out_first_vid;
703 if (i >= MLX5_MAX_MODIFY_NUM)
704 return rte_flow_error_set(error, EINVAL,
705 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
706 "too many items to modify");
707 actions[i] = (struct mlx5_modification_cmd) {
708 .action_type = MLX5_MODIFICATION_TYPE_SET,
710 .length = field->size,
711 .offset = field->offset,
713 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
714 actions[i].data1 = conf->vlan_vid;
715 actions[i].data1 = actions[i].data1 << 16;
716 resource->actions_num = ++i;
721 * Convert modify-header set TP action to DV specification.
723 * @param[in,out] resource
724 * Pointer to the modify-header resource.
726 * Pointer to action specification.
728 * Pointer to rte_flow_item objects list.
730 * Pointer to flow attributes structure.
731 * @param[in] dev_flow
732 * Pointer to the sub flow.
733 * @param[in] tunnel_decap
734 * Whether action is after tunnel decapsulation.
736 * Pointer to the error structure.
739 * 0 on success, a negative errno value otherwise and rte_errno is set.
742 flow_dv_convert_action_modify_tp
743 (struct mlx5_flow_dv_modify_hdr_resource *resource,
744 const struct rte_flow_action *action,
745 const struct rte_flow_item *items,
746 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
747 bool tunnel_decap, struct rte_flow_error *error)
749 const struct rte_flow_action_set_tp *conf =
750 (const struct rte_flow_action_set_tp *)(action->conf);
751 struct rte_flow_item item;
752 struct rte_flow_item_udp udp;
753 struct rte_flow_item_udp udp_mask;
754 struct rte_flow_item_tcp tcp;
755 struct rte_flow_item_tcp tcp_mask;
756 struct field_modify_info *field;
759 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
761 memset(&udp, 0, sizeof(udp));
762 memset(&udp_mask, 0, sizeof(udp_mask));
763 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
764 udp.hdr.src_port = conf->port;
765 udp_mask.hdr.src_port =
766 rte_flow_item_udp_mask.hdr.src_port;
768 udp.hdr.dst_port = conf->port;
769 udp_mask.hdr.dst_port =
770 rte_flow_item_udp_mask.hdr.dst_port;
772 item.type = RTE_FLOW_ITEM_TYPE_UDP;
774 item.mask = &udp_mask;
777 MLX5_ASSERT(attr->tcp);
778 memset(&tcp, 0, sizeof(tcp));
779 memset(&tcp_mask, 0, sizeof(tcp_mask));
780 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
781 tcp.hdr.src_port = conf->port;
782 tcp_mask.hdr.src_port =
783 rte_flow_item_tcp_mask.hdr.src_port;
785 tcp.hdr.dst_port = conf->port;
786 tcp_mask.hdr.dst_port =
787 rte_flow_item_tcp_mask.hdr.dst_port;
789 item.type = RTE_FLOW_ITEM_TYPE_TCP;
791 item.mask = &tcp_mask;
794 return flow_dv_convert_modify_action(&item, field, NULL, resource,
795 MLX5_MODIFICATION_TYPE_SET, error);
799 * Convert modify-header set TTL action to DV specification.
801 * @param[in,out] resource
802 * Pointer to the modify-header resource.
804 * Pointer to action specification.
806 * Pointer to rte_flow_item objects list.
808 * Pointer to flow attributes structure.
809 * @param[in] dev_flow
810 * Pointer to the sub flow.
811 * @param[in] tunnel_decap
812 * Whether action is after tunnel decapsulation.
814 * Pointer to the error structure.
817 * 0 on success, a negative errno value otherwise and rte_errno is set.
820 flow_dv_convert_action_modify_ttl
821 (struct mlx5_flow_dv_modify_hdr_resource *resource,
822 const struct rte_flow_action *action,
823 const struct rte_flow_item *items,
824 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
825 bool tunnel_decap, struct rte_flow_error *error)
827 const struct rte_flow_action_set_ttl *conf =
828 (const struct rte_flow_action_set_ttl *)(action->conf);
829 struct rte_flow_item item;
830 struct rte_flow_item_ipv4 ipv4;
831 struct rte_flow_item_ipv4 ipv4_mask;
832 struct rte_flow_item_ipv6 ipv6;
833 struct rte_flow_item_ipv6 ipv6_mask;
834 struct field_modify_info *field;
837 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
839 memset(&ipv4, 0, sizeof(ipv4));
840 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
841 ipv4.hdr.time_to_live = conf->ttl_value;
842 ipv4_mask.hdr.time_to_live = 0xFF;
843 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
845 item.mask = &ipv4_mask;
848 MLX5_ASSERT(attr->ipv6);
849 memset(&ipv6, 0, sizeof(ipv6));
850 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
851 ipv6.hdr.hop_limits = conf->ttl_value;
852 ipv6_mask.hdr.hop_limits = 0xFF;
853 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
855 item.mask = &ipv6_mask;
858 return flow_dv_convert_modify_action(&item, field, NULL, resource,
859 MLX5_MODIFICATION_TYPE_SET, error);
863 * Convert modify-header decrement TTL action to DV specification.
865 * @param[in,out] resource
866 * Pointer to the modify-header resource.
868 * Pointer to action specification.
870 * Pointer to rte_flow_item objects list.
872 * Pointer to flow attributes structure.
873 * @param[in] dev_flow
874 * Pointer to the sub flow.
875 * @param[in] tunnel_decap
876 * Whether action is after tunnel decapsulation.
878 * Pointer to the error structure.
881 * 0 on success, a negative errno value otherwise and rte_errno is set.
884 flow_dv_convert_action_modify_dec_ttl
885 (struct mlx5_flow_dv_modify_hdr_resource *resource,
886 const struct rte_flow_item *items,
887 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
888 bool tunnel_decap, struct rte_flow_error *error)
890 struct rte_flow_item item;
891 struct rte_flow_item_ipv4 ipv4;
892 struct rte_flow_item_ipv4 ipv4_mask;
893 struct rte_flow_item_ipv6 ipv6;
894 struct rte_flow_item_ipv6 ipv6_mask;
895 struct field_modify_info *field;
898 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
900 memset(&ipv4, 0, sizeof(ipv4));
901 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
902 ipv4.hdr.time_to_live = 0xFF;
903 ipv4_mask.hdr.time_to_live = 0xFF;
904 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
906 item.mask = &ipv4_mask;
909 MLX5_ASSERT(attr->ipv6);
910 memset(&ipv6, 0, sizeof(ipv6));
911 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
912 ipv6.hdr.hop_limits = 0xFF;
913 ipv6_mask.hdr.hop_limits = 0xFF;
914 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
916 item.mask = &ipv6_mask;
919 return flow_dv_convert_modify_action(&item, field, NULL, resource,
920 MLX5_MODIFICATION_TYPE_ADD, error);
924 * Convert modify-header increment/decrement TCP Sequence number
925 * to DV specification.
927 * @param[in,out] resource
928 * Pointer to the modify-header resource.
930 * Pointer to action specification.
932 * Pointer to the error structure.
935 * 0 on success, a negative errno value otherwise and rte_errno is set.
938 flow_dv_convert_action_modify_tcp_seq
939 (struct mlx5_flow_dv_modify_hdr_resource *resource,
940 const struct rte_flow_action *action,
941 struct rte_flow_error *error)
943 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
944 uint64_t value = rte_be_to_cpu_32(*conf);
945 struct rte_flow_item item;
946 struct rte_flow_item_tcp tcp;
947 struct rte_flow_item_tcp tcp_mask;
949 memset(&tcp, 0, sizeof(tcp));
950 memset(&tcp_mask, 0, sizeof(tcp_mask));
951 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
953 * The HW has no decrement operation, only increment operation.
954 * To simulate decrement X from Y using increment operation
955 * we need to add UINT32_MAX X times to Y.
956 * Each adding of UINT32_MAX decrements Y by 1.
959 tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
960 tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
961 item.type = RTE_FLOW_ITEM_TYPE_TCP;
963 item.mask = &tcp_mask;
964 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
965 MLX5_MODIFICATION_TYPE_ADD, error);
969 * Convert modify-header increment/decrement TCP Acknowledgment number
970 * to DV specification.
972 * @param[in,out] resource
973 * Pointer to the modify-header resource.
975 * Pointer to action specification.
977 * Pointer to the error structure.
980 * 0 on success, a negative errno value otherwise and rte_errno is set.
983 flow_dv_convert_action_modify_tcp_ack
984 (struct mlx5_flow_dv_modify_hdr_resource *resource,
985 const struct rte_flow_action *action,
986 struct rte_flow_error *error)
988 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
989 uint64_t value = rte_be_to_cpu_32(*conf);
990 struct rte_flow_item item;
991 struct rte_flow_item_tcp tcp;
992 struct rte_flow_item_tcp tcp_mask;
994 memset(&tcp, 0, sizeof(tcp));
995 memset(&tcp_mask, 0, sizeof(tcp_mask));
996 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
998 * The HW has no decrement operation, only increment operation.
999 * To simulate decrement X from Y using increment operation
1000 * we need to add UINT32_MAX X times to Y.
1001 * Each adding of UINT32_MAX decrements Y by 1.
1003 value *= UINT32_MAX;
1004 tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
1005 tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
1006 item.type = RTE_FLOW_ITEM_TYPE_TCP;
1008 item.mask = &tcp_mask;
1009 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
1010 MLX5_MODIFICATION_TYPE_ADD, error);
1013 static enum mlx5_modification_field reg_to_field[] = {
1014 [REG_NON] = MLX5_MODI_OUT_NONE,
1015 [REG_A] = MLX5_MODI_META_DATA_REG_A,
1016 [REG_B] = MLX5_MODI_META_DATA_REG_B,
1017 [REG_C_0] = MLX5_MODI_META_REG_C_0,
1018 [REG_C_1] = MLX5_MODI_META_REG_C_1,
1019 [REG_C_2] = MLX5_MODI_META_REG_C_2,
1020 [REG_C_3] = MLX5_MODI_META_REG_C_3,
1021 [REG_C_4] = MLX5_MODI_META_REG_C_4,
1022 [REG_C_5] = MLX5_MODI_META_REG_C_5,
1023 [REG_C_6] = MLX5_MODI_META_REG_C_6,
1024 [REG_C_7] = MLX5_MODI_META_REG_C_7,
1028 * Convert register set to DV specification.
1030 * @param[in,out] resource
1031 * Pointer to the modify-header resource.
1033 * Pointer to action specification.
1035 * Pointer to the error structure.
1038 * 0 on success, a negative errno value otherwise and rte_errno is set.
1041 flow_dv_convert_action_set_reg
1042 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1043 const struct rte_flow_action *action,
1044 struct rte_flow_error *error)
1046 const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
1047 struct mlx5_modification_cmd *actions = resource->actions;
1048 uint32_t i = resource->actions_num;
1050 if (i >= MLX5_MAX_MODIFY_NUM)
1051 return rte_flow_error_set(error, EINVAL,
1052 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1053 "too many items to modify");
1054 MLX5_ASSERT(conf->id != REG_NON);
1055 MLX5_ASSERT(conf->id < (enum modify_reg)RTE_DIM(reg_to_field));
1056 actions[i] = (struct mlx5_modification_cmd) {
1057 .action_type = MLX5_MODIFICATION_TYPE_SET,
1058 .field = reg_to_field[conf->id],
1059 .offset = conf->offset,
1060 .length = conf->length,
1062 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
1063 actions[i].data1 = rte_cpu_to_be_32(conf->data);
1065 resource->actions_num = i;
1070 * Convert SET_TAG action to DV specification.
1073 * Pointer to the rte_eth_dev structure.
1074 * @param[in,out] resource
1075 * Pointer to the modify-header resource.
1077 * Pointer to action specification.
1079 * Pointer to the error structure.
1082 * 0 on success, a negative errno value otherwise and rte_errno is set.
1085 flow_dv_convert_action_set_tag
1086 (struct rte_eth_dev *dev,
1087 struct mlx5_flow_dv_modify_hdr_resource *resource,
1088 const struct rte_flow_action_set_tag *conf,
1089 struct rte_flow_error *error)
1091 rte_be32_t data = rte_cpu_to_be_32(conf->data);
1092 rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1093 struct rte_flow_item item = {
1097 struct field_modify_info reg_c_x[] = {
1100 enum mlx5_modification_field reg_type;
1103 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1106 MLX5_ASSERT(ret != REG_NON);
1107 MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1108 reg_type = reg_to_field[ret];
1109 MLX5_ASSERT(reg_type > 0);
1110 reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1111 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1112 MLX5_MODIFICATION_TYPE_SET, error);
1116 * Convert internal COPY_REG action to DV specification.
1119 * Pointer to the rte_eth_dev structure.
1120 * @param[in,out] res
1121 * Pointer to the modify-header resource.
1123 * Pointer to action specification.
1125 * Pointer to the error structure.
1128 * 0 on success, a negative errno value otherwise and rte_errno is set.
1131 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1132 struct mlx5_flow_dv_modify_hdr_resource *res,
1133 const struct rte_flow_action *action,
1134 struct rte_flow_error *error)
1136 const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1137 rte_be32_t mask = RTE_BE32(UINT32_MAX);
1138 struct rte_flow_item item = {
1142 struct field_modify_info reg_src[] = {
1143 {4, 0, reg_to_field[conf->src]},
1146 struct field_modify_info reg_dst = {
1148 .id = reg_to_field[conf->dst],
1150 /* Adjust reg_c[0] usage according to reported mask. */
1151 if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1152 struct mlx5_priv *priv = dev->data->dev_private;
1153 uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1155 MLX5_ASSERT(reg_c0);
1156 MLX5_ASSERT(priv->sh->config.dv_xmeta_en !=
1157 MLX5_XMETA_MODE_LEGACY);
1158 if (conf->dst == REG_C_0) {
1159 /* Copy to reg_c[0], within mask only. */
1160 reg_dst.offset = rte_bsf32(reg_c0);
1161 mask = rte_cpu_to_be_32(reg_c0 >> reg_dst.offset);
1164 mask = rte_cpu_to_be_32(reg_c0);
1167 return flow_dv_convert_modify_action(&item,
1168 reg_src, ®_dst, res,
1169 MLX5_MODIFICATION_TYPE_COPY,
1174 * Convert MARK action to DV specification. This routine is used
1175 * in extensive metadata only and requires metadata register to be
1176 * handled. In legacy mode hardware tag resource is engaged.
1179 * Pointer to the rte_eth_dev structure.
1181 * Pointer to MARK action specification.
1182 * @param[in,out] resource
1183 * Pointer to the modify-header resource.
1185 * Pointer to the error structure.
1188 * 0 on success, a negative errno value otherwise and rte_errno is set.
1191 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1192 const struct rte_flow_action_mark *conf,
1193 struct mlx5_flow_dv_modify_hdr_resource *resource,
1194 struct rte_flow_error *error)
1196 struct mlx5_priv *priv = dev->data->dev_private;
1197 rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1198 priv->sh->dv_mark_mask);
1199 rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1200 struct rte_flow_item item = {
1204 struct field_modify_info reg_c_x[] = {
1210 return rte_flow_error_set(error, EINVAL,
1211 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1212 NULL, "zero mark action mask");
1213 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1216 MLX5_ASSERT(reg > 0);
1217 if (reg == REG_C_0) {
1218 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1219 uint32_t shl_c0 = rte_bsf32(msk_c0);
1221 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1222 mask = rte_cpu_to_be_32(mask) & msk_c0;
1223 mask = rte_cpu_to_be_32(mask << shl_c0);
1225 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1226 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1227 MLX5_MODIFICATION_TYPE_SET, error);
1231 * Get metadata register index for specified steering domain.
1234 * Pointer to the rte_eth_dev structure.
1236 * Attributes of flow to determine steering domain.
1238 * Pointer to the error structure.
1241 * positive index on success, a negative errno value otherwise
1242 * and rte_errno is set.
1244 static enum modify_reg
1245 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1246 const struct rte_flow_attr *attr,
1247 struct rte_flow_error *error)
1250 mlx5_flow_get_reg_id(dev, attr->transfer ?
1254 MLX5_METADATA_RX, 0, error);
1256 return rte_flow_error_set(error,
1257 ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1258 NULL, "unavailable "
1259 "metadata register");
1264 * Convert SET_META action to DV specification.
1267 * Pointer to the rte_eth_dev structure.
1268 * @param[in,out] resource
1269 * Pointer to the modify-header resource.
1271 * Attributes of flow that includes this item.
1273 * Pointer to action specification.
1275 * Pointer to the error structure.
1278 * 0 on success, a negative errno value otherwise and rte_errno is set.
1281 flow_dv_convert_action_set_meta
1282 (struct rte_eth_dev *dev,
1283 struct mlx5_flow_dv_modify_hdr_resource *resource,
1284 const struct rte_flow_attr *attr,
1285 const struct rte_flow_action_set_meta *conf,
1286 struct rte_flow_error *error)
1288 uint32_t mask = rte_cpu_to_be_32(conf->mask);
1289 uint32_t data = rte_cpu_to_be_32(conf->data) & mask;
1290 struct rte_flow_item item = {
1294 struct field_modify_info reg_c_x[] = {
1297 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1301 MLX5_ASSERT(reg != REG_NON);
1302 if (reg == REG_C_0) {
1303 struct mlx5_priv *priv = dev->data->dev_private;
1304 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1305 uint32_t shl_c0 = rte_bsf32(msk_c0);
1307 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1308 mask = rte_cpu_to_be_32(mask) & msk_c0;
1309 mask = rte_cpu_to_be_32(mask << shl_c0);
1311 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1312 /* The routine expects parameters in memory as big-endian ones. */
1313 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1314 MLX5_MODIFICATION_TYPE_SET, error);
1318 * Convert modify-header set IPv4 DSCP action to DV specification.
1320 * @param[in,out] resource
1321 * Pointer to the modify-header resource.
1323 * Pointer to action specification.
1325 * Pointer to the error structure.
1328 * 0 on success, a negative errno value otherwise and rte_errno is set.
1331 flow_dv_convert_action_modify_ipv4_dscp
1332 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1333 const struct rte_flow_action *action,
1334 struct rte_flow_error *error)
1336 const struct rte_flow_action_set_dscp *conf =
1337 (const struct rte_flow_action_set_dscp *)(action->conf);
1338 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1339 struct rte_flow_item_ipv4 ipv4;
1340 struct rte_flow_item_ipv4 ipv4_mask;
1342 memset(&ipv4, 0, sizeof(ipv4));
1343 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1344 ipv4.hdr.type_of_service = conf->dscp;
1345 ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1347 item.mask = &ipv4_mask;
1348 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1349 MLX5_MODIFICATION_TYPE_SET, error);
1353 * Convert modify-header set IPv6 DSCP action to DV specification.
1355 * @param[in,out] resource
1356 * Pointer to the modify-header resource.
1358 * Pointer to action specification.
1360 * Pointer to the error structure.
1363 * 0 on success, a negative errno value otherwise and rte_errno is set.
1366 flow_dv_convert_action_modify_ipv6_dscp
1367 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1368 const struct rte_flow_action *action,
1369 struct rte_flow_error *error)
1371 const struct rte_flow_action_set_dscp *conf =
1372 (const struct rte_flow_action_set_dscp *)(action->conf);
1373 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1374 struct rte_flow_item_ipv6 ipv6;
1375 struct rte_flow_item_ipv6 ipv6_mask;
1377 memset(&ipv6, 0, sizeof(ipv6));
1378 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1380 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1381 * rdma-core only accept the DSCP bits byte aligned start from
1382 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1383 * bits in IPv6 case as rdma-core requires byte aligned value.
1385 ipv6.hdr.vtc_flow = conf->dscp;
1386 ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1388 item.mask = &ipv6_mask;
1389 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1390 MLX5_MODIFICATION_TYPE_SET, error);
1394 mlx5_flow_item_field_width(struct rte_eth_dev *dev,
1395 enum rte_flow_field_id field, int inherit,
1396 const struct rte_flow_attr *attr,
1397 struct rte_flow_error *error)
1399 struct mlx5_priv *priv = dev->data->dev_private;
1402 case RTE_FLOW_FIELD_START:
1404 case RTE_FLOW_FIELD_MAC_DST:
1405 case RTE_FLOW_FIELD_MAC_SRC:
1407 case RTE_FLOW_FIELD_VLAN_TYPE:
1409 case RTE_FLOW_FIELD_VLAN_ID:
1411 case RTE_FLOW_FIELD_MAC_TYPE:
1413 case RTE_FLOW_FIELD_IPV4_DSCP:
1415 case RTE_FLOW_FIELD_IPV4_TTL:
1417 case RTE_FLOW_FIELD_IPV4_SRC:
1418 case RTE_FLOW_FIELD_IPV4_DST:
1420 case RTE_FLOW_FIELD_IPV6_DSCP:
1422 case RTE_FLOW_FIELD_IPV6_HOPLIMIT:
1424 case RTE_FLOW_FIELD_IPV6_SRC:
1425 case RTE_FLOW_FIELD_IPV6_DST:
1427 case RTE_FLOW_FIELD_TCP_PORT_SRC:
1428 case RTE_FLOW_FIELD_TCP_PORT_DST:
1430 case RTE_FLOW_FIELD_TCP_SEQ_NUM:
1431 case RTE_FLOW_FIELD_TCP_ACK_NUM:
1433 case RTE_FLOW_FIELD_TCP_FLAGS:
1435 case RTE_FLOW_FIELD_UDP_PORT_SRC:
1436 case RTE_FLOW_FIELD_UDP_PORT_DST:
1438 case RTE_FLOW_FIELD_VXLAN_VNI:
1439 case RTE_FLOW_FIELD_GENEVE_VNI:
1441 case RTE_FLOW_FIELD_GTP_TEID:
1442 case RTE_FLOW_FIELD_TAG:
1444 case RTE_FLOW_FIELD_MARK:
1445 return __builtin_popcount(priv->sh->dv_mark_mask);
1446 case RTE_FLOW_FIELD_META:
1447 return (flow_dv_get_metadata_reg(dev, attr, error) == REG_C_0) ?
1448 __builtin_popcount(priv->sh->dv_meta_mask) : 32;
1449 case RTE_FLOW_FIELD_POINTER:
1450 case RTE_FLOW_FIELD_VALUE:
1451 return inherit < 0 ? 0 : inherit;
1459 mlx5_flow_field_id_to_modify_info
1460 (const struct rte_flow_action_modify_data *data,
1461 struct field_modify_info *info, uint32_t *mask,
1462 uint32_t width, struct rte_eth_dev *dev,
1463 const struct rte_flow_attr *attr, struct rte_flow_error *error)
1465 struct mlx5_priv *priv = dev->data->dev_private;
1469 switch (data->field) {
1470 case RTE_FLOW_FIELD_START:
1471 /* not supported yet */
1474 case RTE_FLOW_FIELD_MAC_DST:
1475 off = data->offset > 16 ? data->offset - 16 : 0;
1477 if (data->offset < 16) {
1478 info[idx] = (struct field_modify_info){2, 4,
1479 MLX5_MODI_OUT_DMAC_15_0};
1481 mask[1] = rte_cpu_to_be_16(0xffff >>
1485 mask[1] = RTE_BE16(0xffff);
1492 info[idx] = (struct field_modify_info){4, 0,
1493 MLX5_MODI_OUT_DMAC_47_16};
1494 mask[0] = rte_cpu_to_be_32((0xffffffff >>
1495 (32 - width)) << off);
1497 if (data->offset < 16)
1498 info[idx++] = (struct field_modify_info){2, 0,
1499 MLX5_MODI_OUT_DMAC_15_0};
1500 info[idx] = (struct field_modify_info){4, off,
1501 MLX5_MODI_OUT_DMAC_47_16};
1504 case RTE_FLOW_FIELD_MAC_SRC:
1505 off = data->offset > 16 ? data->offset - 16 : 0;
1507 if (data->offset < 16) {
1508 info[idx] = (struct field_modify_info){2, 4,
1509 MLX5_MODI_OUT_SMAC_15_0};
1511 mask[1] = rte_cpu_to_be_16(0xffff >>
1515 mask[1] = RTE_BE16(0xffff);
1522 info[idx] = (struct field_modify_info){4, 0,
1523 MLX5_MODI_OUT_SMAC_47_16};
1524 mask[0] = rte_cpu_to_be_32((0xffffffff >>
1525 (32 - width)) << off);
1527 if (data->offset < 16)
1528 info[idx++] = (struct field_modify_info){2, 0,
1529 MLX5_MODI_OUT_SMAC_15_0};
1530 info[idx] = (struct field_modify_info){4, off,
1531 MLX5_MODI_OUT_SMAC_47_16};
1534 case RTE_FLOW_FIELD_VLAN_TYPE:
1535 /* not supported yet */
1537 case RTE_FLOW_FIELD_VLAN_ID:
1538 info[idx] = (struct field_modify_info){2, 0,
1539 MLX5_MODI_OUT_FIRST_VID};
1541 mask[idx] = rte_cpu_to_be_16(0x0fff >> (12 - width));
1543 case RTE_FLOW_FIELD_MAC_TYPE:
1544 info[idx] = (struct field_modify_info){2, 0,
1545 MLX5_MODI_OUT_ETHERTYPE};
1547 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1549 case RTE_FLOW_FIELD_IPV4_DSCP:
1550 info[idx] = (struct field_modify_info){1, 0,
1551 MLX5_MODI_OUT_IP_DSCP};
1553 mask[idx] = 0x3f >> (6 - width);
1555 case RTE_FLOW_FIELD_IPV4_TTL:
1556 info[idx] = (struct field_modify_info){1, 0,
1557 MLX5_MODI_OUT_IPV4_TTL};
1559 mask[idx] = 0xff >> (8 - width);
1561 case RTE_FLOW_FIELD_IPV4_SRC:
1562 info[idx] = (struct field_modify_info){4, 0,
1563 MLX5_MODI_OUT_SIPV4};
1565 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1568 case RTE_FLOW_FIELD_IPV4_DST:
1569 info[idx] = (struct field_modify_info){4, 0,
1570 MLX5_MODI_OUT_DIPV4};
1572 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1575 case RTE_FLOW_FIELD_IPV6_DSCP:
1576 info[idx] = (struct field_modify_info){1, 0,
1577 MLX5_MODI_OUT_IP_DSCP};
1579 mask[idx] = 0x3f >> (6 - width);
1581 case RTE_FLOW_FIELD_IPV6_HOPLIMIT:
1582 info[idx] = (struct field_modify_info){1, 0,
1583 MLX5_MODI_OUT_IPV6_HOPLIMIT};
1585 mask[idx] = 0xff >> (8 - width);
1587 case RTE_FLOW_FIELD_IPV6_SRC:
1589 if (data->offset < 32) {
1590 info[idx] = (struct field_modify_info){4, 12,
1591 MLX5_MODI_OUT_SIPV6_31_0};
1594 rte_cpu_to_be_32(0xffffffff >>
1598 mask[3] = RTE_BE32(0xffffffff);
1605 if (data->offset < 64) {
1606 info[idx] = (struct field_modify_info){4, 8,
1607 MLX5_MODI_OUT_SIPV6_63_32};
1610 rte_cpu_to_be_32(0xffffffff >>
1614 mask[2] = RTE_BE32(0xffffffff);
1621 if (data->offset < 96) {
1622 info[idx] = (struct field_modify_info){4, 4,
1623 MLX5_MODI_OUT_SIPV6_95_64};
1626 rte_cpu_to_be_32(0xffffffff >>
1630 mask[1] = RTE_BE32(0xffffffff);
1637 info[idx] = (struct field_modify_info){4, 0,
1638 MLX5_MODI_OUT_SIPV6_127_96};
1639 mask[0] = rte_cpu_to_be_32(0xffffffff >> (32 - width));
1641 if (data->offset < 32)
1642 info[idx++] = (struct field_modify_info){4, 0,
1643 MLX5_MODI_OUT_SIPV6_31_0};
1644 if (data->offset < 64)
1645 info[idx++] = (struct field_modify_info){4, 0,
1646 MLX5_MODI_OUT_SIPV6_63_32};
1647 if (data->offset < 96)
1648 info[idx++] = (struct field_modify_info){4, 0,
1649 MLX5_MODI_OUT_SIPV6_95_64};
1650 if (data->offset < 128)
1651 info[idx++] = (struct field_modify_info){4, 0,
1652 MLX5_MODI_OUT_SIPV6_127_96};
1655 case RTE_FLOW_FIELD_IPV6_DST:
1657 if (data->offset < 32) {
1658 info[idx] = (struct field_modify_info){4, 12,
1659 MLX5_MODI_OUT_DIPV6_31_0};
1662 rte_cpu_to_be_32(0xffffffff >>
1666 mask[3] = RTE_BE32(0xffffffff);
1673 if (data->offset < 64) {
1674 info[idx] = (struct field_modify_info){4, 8,
1675 MLX5_MODI_OUT_DIPV6_63_32};
1678 rte_cpu_to_be_32(0xffffffff >>
1682 mask[2] = RTE_BE32(0xffffffff);
1689 if (data->offset < 96) {
1690 info[idx] = (struct field_modify_info){4, 4,
1691 MLX5_MODI_OUT_DIPV6_95_64};
1694 rte_cpu_to_be_32(0xffffffff >>
1698 mask[1] = RTE_BE32(0xffffffff);
1705 info[idx] = (struct field_modify_info){4, 0,
1706 MLX5_MODI_OUT_DIPV6_127_96};
1707 mask[0] = rte_cpu_to_be_32(0xffffffff >> (32 - width));
1709 if (data->offset < 32)
1710 info[idx++] = (struct field_modify_info){4, 0,
1711 MLX5_MODI_OUT_DIPV6_31_0};
1712 if (data->offset < 64)
1713 info[idx++] = (struct field_modify_info){4, 0,
1714 MLX5_MODI_OUT_DIPV6_63_32};
1715 if (data->offset < 96)
1716 info[idx++] = (struct field_modify_info){4, 0,
1717 MLX5_MODI_OUT_DIPV6_95_64};
1718 if (data->offset < 128)
1719 info[idx++] = (struct field_modify_info){4, 0,
1720 MLX5_MODI_OUT_DIPV6_127_96};
1723 case RTE_FLOW_FIELD_TCP_PORT_SRC:
1724 info[idx] = (struct field_modify_info){2, 0,
1725 MLX5_MODI_OUT_TCP_SPORT};
1727 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1729 case RTE_FLOW_FIELD_TCP_PORT_DST:
1730 info[idx] = (struct field_modify_info){2, 0,
1731 MLX5_MODI_OUT_TCP_DPORT};
1733 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1735 case RTE_FLOW_FIELD_TCP_SEQ_NUM:
1736 info[idx] = (struct field_modify_info){4, 0,
1737 MLX5_MODI_OUT_TCP_SEQ_NUM};
1739 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1742 case RTE_FLOW_FIELD_TCP_ACK_NUM:
1743 info[idx] = (struct field_modify_info){4, 0,
1744 MLX5_MODI_OUT_TCP_ACK_NUM};
1746 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1749 case RTE_FLOW_FIELD_TCP_FLAGS:
1750 info[idx] = (struct field_modify_info){2, 0,
1751 MLX5_MODI_OUT_TCP_FLAGS};
1753 mask[idx] = rte_cpu_to_be_16(0x1ff >> (9 - width));
1755 case RTE_FLOW_FIELD_UDP_PORT_SRC:
1756 info[idx] = (struct field_modify_info){2, 0,
1757 MLX5_MODI_OUT_UDP_SPORT};
1759 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1761 case RTE_FLOW_FIELD_UDP_PORT_DST:
1762 info[idx] = (struct field_modify_info){2, 0,
1763 MLX5_MODI_OUT_UDP_DPORT};
1765 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1767 case RTE_FLOW_FIELD_VXLAN_VNI:
1768 /* not supported yet */
1770 case RTE_FLOW_FIELD_GENEVE_VNI:
1771 /* not supported yet*/
1773 case RTE_FLOW_FIELD_GTP_TEID:
1774 info[idx] = (struct field_modify_info){4, 0,
1775 MLX5_MODI_GTP_TEID};
1777 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1780 case RTE_FLOW_FIELD_TAG:
1782 int reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG,
1783 data->level, error);
1786 MLX5_ASSERT(reg != REG_NON);
1787 MLX5_ASSERT((unsigned int)reg < RTE_DIM(reg_to_field));
1788 info[idx] = (struct field_modify_info){4, 0,
1792 rte_cpu_to_be_32(0xffffffff >>
1796 case RTE_FLOW_FIELD_MARK:
1798 uint32_t mark_mask = priv->sh->dv_mark_mask;
1799 uint32_t mark_count = __builtin_popcount(mark_mask);
1800 int reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK,
1804 MLX5_ASSERT(reg != REG_NON);
1805 MLX5_ASSERT((unsigned int)reg < RTE_DIM(reg_to_field));
1806 info[idx] = (struct field_modify_info){4, 0,
1809 mask[idx] = rte_cpu_to_be_32((mark_mask >>
1810 (mark_count - width)) & mark_mask);
1813 case RTE_FLOW_FIELD_META:
1815 uint32_t meta_mask = priv->sh->dv_meta_mask;
1816 uint32_t meta_count = __builtin_popcount(meta_mask);
1817 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1820 MLX5_ASSERT(reg != REG_NON);
1821 MLX5_ASSERT((unsigned int)reg < RTE_DIM(reg_to_field));
1822 info[idx] = (struct field_modify_info){4, 0,
1825 mask[idx] = rte_cpu_to_be_32((meta_mask >>
1826 (meta_count - width)) & meta_mask);
1829 case RTE_FLOW_FIELD_POINTER:
1830 case RTE_FLOW_FIELD_VALUE:
1838 * Convert modify_field action to DV specification.
1841 * Pointer to the rte_eth_dev structure.
1842 * @param[in,out] resource
1843 * Pointer to the modify-header resource.
1845 * Pointer to action specification.
1847 * Attributes of flow that includes this item.
1849 * Pointer to the error structure.
1852 * 0 on success, a negative errno value otherwise and rte_errno is set.
1855 flow_dv_convert_action_modify_field
1856 (struct rte_eth_dev *dev,
1857 struct mlx5_flow_dv_modify_hdr_resource *resource,
1858 const struct rte_flow_action *action,
1859 const struct rte_flow_attr *attr,
1860 struct rte_flow_error *error)
1862 const struct rte_flow_action_modify_field *conf =
1863 (const struct rte_flow_action_modify_field *)(action->conf);
1864 struct rte_flow_item item = {
1868 struct field_modify_info field[MLX5_ACT_MAX_MOD_FIELDS] = {
1870 struct field_modify_info dcopy[MLX5_ACT_MAX_MOD_FIELDS] = {
1872 uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS] = {0, 0, 0, 0, 0};
1873 uint32_t type, meta = 0;
1875 if (conf->src.field == RTE_FLOW_FIELD_POINTER ||
1876 conf->src.field == RTE_FLOW_FIELD_VALUE) {
1877 type = MLX5_MODIFICATION_TYPE_SET;
1878 /** For SET fill the destination field (field) first. */
1879 mlx5_flow_field_id_to_modify_info(&conf->dst, field, mask,
1882 item.spec = conf->src.field == RTE_FLOW_FIELD_POINTER ?
1883 (void *)(uintptr_t)conf->src.pvalue :
1884 (void *)(uintptr_t)&conf->src.value;
1885 if (conf->dst.field == RTE_FLOW_FIELD_META) {
1886 meta = *(const unaligned_uint32_t *)item.spec;
1887 meta = rte_cpu_to_be_32(meta);
1891 type = MLX5_MODIFICATION_TYPE_COPY;
1892 /** For COPY fill the destination field (dcopy) without mask. */
1893 mlx5_flow_field_id_to_modify_info(&conf->dst, dcopy, NULL,
1896 /** Then construct the source field (field) with mask. */
1897 mlx5_flow_field_id_to_modify_info(&conf->src, field, mask,
1902 return flow_dv_convert_modify_action(&item,
1903 field, dcopy, resource, type, error);
1907 * Validate MARK item.
1910 * Pointer to the rte_eth_dev structure.
1912 * Item specification.
1914 * Attributes of flow that includes this item.
1916 * Pointer to error structure.
1919 * 0 on success, a negative errno value otherwise and rte_errno is set.
1922 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1923 const struct rte_flow_item *item,
1924 const struct rte_flow_attr *attr __rte_unused,
1925 struct rte_flow_error *error)
1927 struct mlx5_priv *priv = dev->data->dev_private;
1928 struct mlx5_sh_config *config = &priv->sh->config;
1929 const struct rte_flow_item_mark *spec = item->spec;
1930 const struct rte_flow_item_mark *mask = item->mask;
1931 const struct rte_flow_item_mark nic_mask = {
1932 .id = priv->sh->dv_mark_mask,
1936 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1937 return rte_flow_error_set(error, ENOTSUP,
1938 RTE_FLOW_ERROR_TYPE_ITEM, item,
1939 "extended metadata feature"
1941 if (!mlx5_flow_ext_mreg_supported(dev))
1942 return rte_flow_error_set(error, ENOTSUP,
1943 RTE_FLOW_ERROR_TYPE_ITEM, item,
1944 "extended metadata register"
1945 " isn't supported");
1947 return rte_flow_error_set(error, ENOTSUP,
1948 RTE_FLOW_ERROR_TYPE_ITEM, item,
1949 "extended metadata register"
1950 " isn't available");
1951 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1955 return rte_flow_error_set(error, EINVAL,
1956 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1958 "data cannot be empty");
1959 if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1960 return rte_flow_error_set(error, EINVAL,
1961 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1963 "mark id exceeds the limit");
1967 return rte_flow_error_set(error, EINVAL,
1968 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1969 "mask cannot be zero");
1971 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1972 (const uint8_t *)&nic_mask,
1973 sizeof(struct rte_flow_item_mark),
1974 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1981 * Validate META item.
1984 * Pointer to the rte_eth_dev structure.
1986 * Item specification.
1988 * Attributes of flow that includes this item.
1990 * Pointer to error structure.
1993 * 0 on success, a negative errno value otherwise and rte_errno is set.
1996 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1997 const struct rte_flow_item *item,
1998 const struct rte_flow_attr *attr,
1999 struct rte_flow_error *error)
2001 struct mlx5_priv *priv = dev->data->dev_private;
2002 struct mlx5_sh_config *config = &priv->sh->config;
2003 const struct rte_flow_item_meta *spec = item->spec;
2004 const struct rte_flow_item_meta *mask = item->mask;
2005 struct rte_flow_item_meta nic_mask = {
2012 return rte_flow_error_set(error, EINVAL,
2013 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2015 "data cannot be empty");
2016 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
2017 if (!mlx5_flow_ext_mreg_supported(dev))
2018 return rte_flow_error_set(error, ENOTSUP,
2019 RTE_FLOW_ERROR_TYPE_ITEM, item,
2020 "extended metadata register"
2021 " isn't supported");
2022 reg = flow_dv_get_metadata_reg(dev, attr, error);
2026 return rte_flow_error_set(error, ENOTSUP,
2027 RTE_FLOW_ERROR_TYPE_ITEM, item,
2028 "unavailable extended metadata register");
2030 return rte_flow_error_set(error, ENOTSUP,
2031 RTE_FLOW_ERROR_TYPE_ITEM, item,
2035 nic_mask.data = priv->sh->dv_meta_mask;
2038 return rte_flow_error_set(error, ENOTSUP,
2039 RTE_FLOW_ERROR_TYPE_ITEM, item,
2040 "extended metadata feature "
2041 "should be enabled when "
2042 "meta item is requested "
2043 "with e-switch mode ");
2045 return rte_flow_error_set(error, ENOTSUP,
2046 RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 "match on metadata for ingress "
2048 "is not supported in legacy "
2052 mask = &rte_flow_item_meta_mask;
2054 return rte_flow_error_set(error, EINVAL,
2055 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
2056 "mask cannot be zero");
2058 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2059 (const uint8_t *)&nic_mask,
2060 sizeof(struct rte_flow_item_meta),
2061 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2066 * Validate TAG item.
2069 * Pointer to the rte_eth_dev structure.
2071 * Item specification.
2073 * Attributes of flow that includes this item.
2075 * Pointer to error structure.
2078 * 0 on success, a negative errno value otherwise and rte_errno is set.
2081 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
2082 const struct rte_flow_item *item,
2083 const struct rte_flow_attr *attr __rte_unused,
2084 struct rte_flow_error *error)
2086 const struct rte_flow_item_tag *spec = item->spec;
2087 const struct rte_flow_item_tag *mask = item->mask;
2088 const struct rte_flow_item_tag nic_mask = {
2089 .data = RTE_BE32(UINT32_MAX),
2094 if (!mlx5_flow_ext_mreg_supported(dev))
2095 return rte_flow_error_set(error, ENOTSUP,
2096 RTE_FLOW_ERROR_TYPE_ITEM, item,
2097 "extensive metadata register"
2098 " isn't supported");
2100 return rte_flow_error_set(error, EINVAL,
2101 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2103 "data cannot be empty");
2105 mask = &rte_flow_item_tag_mask;
2107 return rte_flow_error_set(error, EINVAL,
2108 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
2109 "mask cannot be zero");
2111 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2112 (const uint8_t *)&nic_mask,
2113 sizeof(struct rte_flow_item_tag),
2114 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2117 if (mask->index != 0xff)
2118 return rte_flow_error_set(error, EINVAL,
2119 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
2120 "partial mask for tag index"
2121 " is not supported");
2122 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
2125 MLX5_ASSERT(ret != REG_NON);
2130 * Validate vport item.
2133 * Pointer to the rte_eth_dev structure.
2135 * Item specification.
2137 * Attributes of flow that includes this item.
2138 * @param[in] item_flags
2139 * Bit-fields that holds the items detected until now.
2141 * Pointer to error structure.
2144 * 0 on success, a negative errno value otherwise and rte_errno is set.
2147 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
2148 const struct rte_flow_item *item,
2149 const struct rte_flow_attr *attr,
2150 uint64_t item_flags,
2151 struct rte_flow_error *error)
2153 const struct rte_flow_item_port_id *spec = item->spec;
2154 const struct rte_flow_item_port_id *mask = item->mask;
2155 const struct rte_flow_item_port_id switch_mask = {
2158 struct mlx5_priv *esw_priv;
2159 struct mlx5_priv *dev_priv;
2162 if (!attr->transfer)
2163 return rte_flow_error_set(error, EINVAL,
2164 RTE_FLOW_ERROR_TYPE_ITEM,
2166 "match on port id is valid only"
2167 " when transfer flag is enabled");
2168 if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
2169 return rte_flow_error_set(error, ENOTSUP,
2170 RTE_FLOW_ERROR_TYPE_ITEM, item,
2171 "multiple source ports are not"
2174 mask = &switch_mask;
2175 if (mask->id != 0xffffffff)
2176 return rte_flow_error_set(error, ENOTSUP,
2177 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2179 "no support for partial mask on"
2181 ret = mlx5_flow_item_acceptable
2182 (item, (const uint8_t *)mask,
2183 (const uint8_t *)&rte_flow_item_port_id_mask,
2184 sizeof(struct rte_flow_item_port_id),
2185 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2190 if (spec->id == MLX5_PORT_ESW_MGR)
2192 esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
2194 return rte_flow_error_set(error, rte_errno,
2195 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
2196 "failed to obtain E-Switch info for"
2198 dev_priv = mlx5_dev_to_eswitch_info(dev);
2200 return rte_flow_error_set(error, rte_errno,
2201 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2203 "failed to obtain E-Switch info");
2204 if (esw_priv->domain_id != dev_priv->domain_id)
2205 return rte_flow_error_set(error, EINVAL,
2206 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
2207 "cannot match on a port from a"
2208 " different E-Switch");
2213 * Validate VLAN item.
2216 * Item specification.
2217 * @param[in] item_flags
2218 * Bit-fields that holds the items detected until now.
2220 * Ethernet device flow is being created on.
2222 * Pointer to error structure.
2225 * 0 on success, a negative errno value otherwise and rte_errno is set.
2228 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
2229 uint64_t item_flags,
2230 struct rte_eth_dev *dev,
2231 struct rte_flow_error *error)
2233 const struct rte_flow_item_vlan *mask = item->mask;
2234 const struct rte_flow_item_vlan nic_mask = {
2235 .tci = RTE_BE16(UINT16_MAX),
2236 .inner_type = RTE_BE16(UINT16_MAX),
2239 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2241 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2242 MLX5_FLOW_LAYER_INNER_L4) :
2243 (MLX5_FLOW_LAYER_OUTER_L3 |
2244 MLX5_FLOW_LAYER_OUTER_L4);
2245 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2246 MLX5_FLOW_LAYER_OUTER_VLAN;
2248 if (item_flags & vlanm)
2249 return rte_flow_error_set(error, EINVAL,
2250 RTE_FLOW_ERROR_TYPE_ITEM, item,
2251 "multiple VLAN layers not supported");
2252 else if ((item_flags & l34m) != 0)
2253 return rte_flow_error_set(error, EINVAL,
2254 RTE_FLOW_ERROR_TYPE_ITEM, item,
2255 "VLAN cannot follow L3/L4 layer");
2257 mask = &rte_flow_item_vlan_mask;
2258 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2259 (const uint8_t *)&nic_mask,
2260 sizeof(struct rte_flow_item_vlan),
2261 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2264 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2265 struct mlx5_priv *priv = dev->data->dev_private;
2267 if (priv->vmwa_context) {
2269 * Non-NULL context means we have a virtual machine
2270 * and SR-IOV enabled, we have to create VLAN interface
2271 * to make hypervisor to setup E-Switch vport
2272 * context correctly. We avoid creating the multiple
2273 * VLAN interfaces, so we cannot support VLAN tag mask.
2275 return rte_flow_error_set(error, EINVAL,
2276 RTE_FLOW_ERROR_TYPE_ITEM,
2278 "VLAN tag mask is not"
2279 " supported in virtual"
2287 * GTP flags are contained in 1 byte of the format:
2288 * -------------------------------------------
2289 * | bit | 0 - 2 | 3 | 4 | 5 | 6 | 7 |
2290 * |-----------------------------------------|
2291 * | value | Version | PT | Res | E | S | PN |
2292 * -------------------------------------------
2294 * Matching is supported only for GTP flags E, S, PN.
2296 #define MLX5_GTP_FLAGS_MASK 0x07
2299 * Validate GTP item.
2302 * Pointer to the rte_eth_dev structure.
2304 * Item specification.
2305 * @param[in] item_flags
2306 * Bit-fields that holds the items detected until now.
2308 * Pointer to error structure.
2311 * 0 on success, a negative errno value otherwise and rte_errno is set.
2314 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
2315 const struct rte_flow_item *item,
2316 uint64_t item_flags,
2317 struct rte_flow_error *error)
2319 struct mlx5_priv *priv = dev->data->dev_private;
2320 const struct rte_flow_item_gtp *spec = item->spec;
2321 const struct rte_flow_item_gtp *mask = item->mask;
2322 const struct rte_flow_item_gtp nic_mask = {
2323 .v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
2325 .teid = RTE_BE32(0xffffffff),
2328 if (!priv->sh->cdev->config.hca_attr.tunnel_stateless_gtp)
2329 return rte_flow_error_set(error, ENOTSUP,
2330 RTE_FLOW_ERROR_TYPE_ITEM, item,
2331 "GTP support is not enabled");
2332 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2333 return rte_flow_error_set(error, ENOTSUP,
2334 RTE_FLOW_ERROR_TYPE_ITEM, item,
2335 "multiple tunnel layers not"
2337 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2338 return rte_flow_error_set(error, EINVAL,
2339 RTE_FLOW_ERROR_TYPE_ITEM, item,
2340 "no outer UDP layer found");
2342 mask = &rte_flow_item_gtp_mask;
2343 if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
2344 return rte_flow_error_set(error, ENOTSUP,
2345 RTE_FLOW_ERROR_TYPE_ITEM, item,
2346 "Match is supported for GTP"
2348 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2349 (const uint8_t *)&nic_mask,
2350 sizeof(struct rte_flow_item_gtp),
2351 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2355 * Validate GTP PSC item.
2358 * Item specification.
2359 * @param[in] last_item
2360 * Previous validated item in the pattern items.
2361 * @param[in] gtp_item
2362 * Previous GTP item specification.
2364 * Pointer to flow attributes.
2366 * Pointer to error structure.
2369 * 0 on success, a negative errno value otherwise and rte_errno is set.
2372 flow_dv_validate_item_gtp_psc(const struct rte_flow_item *item,
2374 const struct rte_flow_item *gtp_item,
2375 const struct rte_flow_attr *attr,
2376 struct rte_flow_error *error)
2378 const struct rte_flow_item_gtp *gtp_spec;
2379 const struct rte_flow_item_gtp *gtp_mask;
2380 const struct rte_flow_item_gtp_psc *mask;
2381 const struct rte_flow_item_gtp_psc nic_mask = {
2386 if (!gtp_item || !(last_item & MLX5_FLOW_LAYER_GTP))
2387 return rte_flow_error_set
2388 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2389 "GTP PSC item must be preceded with GTP item");
2390 gtp_spec = gtp_item->spec;
2391 gtp_mask = gtp_item->mask ? gtp_item->mask : &rte_flow_item_gtp_mask;
2392 /* GTP spec and E flag is requested to match zero. */
2394 (gtp_mask->v_pt_rsv_flags &
2395 ~gtp_spec->v_pt_rsv_flags & MLX5_GTP_EXT_HEADER_FLAG))
2396 return rte_flow_error_set
2397 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2398 "GTP E flag must be 1 to match GTP PSC");
2399 /* Check the flow is not created in group zero. */
2400 if (!attr->transfer && !attr->group)
2401 return rte_flow_error_set
2402 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2403 "GTP PSC is not supported for group 0");
2404 /* GTP spec is here and E flag is requested to match zero. */
2407 mask = item->mask ? item->mask : &rte_flow_item_gtp_psc_mask;
2408 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2409 (const uint8_t *)&nic_mask,
2410 sizeof(struct rte_flow_item_gtp_psc),
2411 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2415 * Validate IPV4 item.
2416 * Use existing validation function mlx5_flow_validate_item_ipv4(), and
2417 * add specific validation of fragment_offset field,
2420 * Item specification.
2421 * @param[in] item_flags
2422 * Bit-fields that holds the items detected until now.
2424 * Pointer to error structure.
2427 * 0 on success, a negative errno value otherwise and rte_errno is set.
2430 flow_dv_validate_item_ipv4(struct rte_eth_dev *dev,
2431 const struct rte_flow_item *item,
2432 uint64_t item_flags, uint64_t last_item,
2433 uint16_t ether_type, struct rte_flow_error *error)
2436 struct mlx5_priv *priv = dev->data->dev_private;
2437 struct mlx5_hca_attr *attr = &priv->sh->cdev->config.hca_attr;
2438 const struct rte_flow_item_ipv4 *spec = item->spec;
2439 const struct rte_flow_item_ipv4 *last = item->last;
2440 const struct rte_flow_item_ipv4 *mask = item->mask;
2441 rte_be16_t fragment_offset_spec = 0;
2442 rte_be16_t fragment_offset_last = 0;
2443 struct rte_flow_item_ipv4 nic_ipv4_mask = {
2445 .src_addr = RTE_BE32(0xffffffff),
2446 .dst_addr = RTE_BE32(0xffffffff),
2447 .type_of_service = 0xff,
2448 .fragment_offset = RTE_BE16(0xffff),
2449 .next_proto_id = 0xff,
2450 .time_to_live = 0xff,
2454 if (mask && (mask->hdr.version_ihl & RTE_IPV4_HDR_IHL_MASK)) {
2455 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2456 bool ihl_cap = !tunnel ?
2457 attr->outer_ipv4_ihl : attr->inner_ipv4_ihl;
2459 return rte_flow_error_set(error, ENOTSUP,
2460 RTE_FLOW_ERROR_TYPE_ITEM,
2462 "IPV4 ihl offload not supported");
2463 nic_ipv4_mask.hdr.version_ihl = mask->hdr.version_ihl;
2465 ret = mlx5_flow_validate_item_ipv4(item, item_flags, last_item,
2466 ether_type, &nic_ipv4_mask,
2467 MLX5_ITEM_RANGE_ACCEPTED, error);
2471 fragment_offset_spec = spec->hdr.fragment_offset &
2472 mask->hdr.fragment_offset;
2473 if (!fragment_offset_spec)
2476 * spec and mask are valid, enforce using full mask to make sure the
2477 * complete value is used correctly.
2479 if ((mask->hdr.fragment_offset & RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
2480 != RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
2481 return rte_flow_error_set(error, EINVAL,
2482 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2483 item, "must use full mask for"
2484 " fragment_offset");
2486 * Match on fragment_offset 0x2000 means MF is 1 and frag-offset is 0,
2487 * indicating this is 1st fragment of fragmented packet.
2488 * This is not yet supported in MLX5, return appropriate error message.
2490 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG))
2491 return rte_flow_error_set(error, ENOTSUP,
2492 RTE_FLOW_ERROR_TYPE_ITEM, item,
2493 "match on first fragment not "
2495 if (fragment_offset_spec && !last)
2496 return rte_flow_error_set(error, ENOTSUP,
2497 RTE_FLOW_ERROR_TYPE_ITEM, item,
2498 "specified value not supported");
2499 /* spec and last are valid, validate the specified range. */
2500 fragment_offset_last = last->hdr.fragment_offset &
2501 mask->hdr.fragment_offset;
2503 * Match on fragment_offset spec 0x2001 and last 0x3fff
2504 * means MF is 1 and frag-offset is > 0.
2505 * This packet is fragment 2nd and onward, excluding last.
2506 * This is not yet supported in MLX5, return appropriate
2509 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG + 1) &&
2510 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
2511 return rte_flow_error_set(error, ENOTSUP,
2512 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2513 last, "match on following "
2514 "fragments not supported");
2516 * Match on fragment_offset spec 0x0001 and last 0x1fff
2517 * means MF is 0 and frag-offset is > 0.
2518 * This packet is last fragment of fragmented packet.
2519 * This is not yet supported in MLX5, return appropriate
2522 if (fragment_offset_spec == RTE_BE16(1) &&
2523 fragment_offset_last == RTE_BE16(RTE_IPV4_HDR_OFFSET_MASK))
2524 return rte_flow_error_set(error, ENOTSUP,
2525 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2526 last, "match on last "
2527 "fragment not supported");
2529 * Match on fragment_offset spec 0x0001 and last 0x3fff
2530 * means MF and/or frag-offset is not 0.
2531 * This is a fragmented packet.
2532 * Other range values are invalid and rejected.
2534 if (!(fragment_offset_spec == RTE_BE16(1) &&
2535 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK)))
2536 return rte_flow_error_set(error, ENOTSUP,
2537 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
2538 "specified range not supported");
2543 * Validate IPV6 fragment extension item.
2546 * Item specification.
2547 * @param[in] item_flags
2548 * Bit-fields that holds the items detected until now.
2550 * Pointer to error structure.
2553 * 0 on success, a negative errno value otherwise and rte_errno is set.
2556 flow_dv_validate_item_ipv6_frag_ext(const struct rte_flow_item *item,
2557 uint64_t item_flags,
2558 struct rte_flow_error *error)
2560 const struct rte_flow_item_ipv6_frag_ext *spec = item->spec;
2561 const struct rte_flow_item_ipv6_frag_ext *last = item->last;
2562 const struct rte_flow_item_ipv6_frag_ext *mask = item->mask;
2563 rte_be16_t frag_data_spec = 0;
2564 rte_be16_t frag_data_last = 0;
2565 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2566 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2567 MLX5_FLOW_LAYER_OUTER_L4;
2569 struct rte_flow_item_ipv6_frag_ext nic_mask = {
2571 .next_header = 0xff,
2572 .frag_data = RTE_BE16(0xffff),
2576 if (item_flags & l4m)
2577 return rte_flow_error_set(error, EINVAL,
2578 RTE_FLOW_ERROR_TYPE_ITEM, item,
2579 "ipv6 fragment extension item cannot "
2581 if ((tunnel && !(item_flags & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
2582 (!tunnel && !(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
2583 return rte_flow_error_set(error, EINVAL,
2584 RTE_FLOW_ERROR_TYPE_ITEM, item,
2585 "ipv6 fragment extension item must "
2586 "follow ipv6 item");
2588 frag_data_spec = spec->hdr.frag_data & mask->hdr.frag_data;
2589 if (!frag_data_spec)
2592 * spec and mask are valid, enforce using full mask to make sure the
2593 * complete value is used correctly.
2595 if ((mask->hdr.frag_data & RTE_BE16(RTE_IPV6_FRAG_USED_MASK)) !=
2596 RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
2597 return rte_flow_error_set(error, EINVAL,
2598 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2599 item, "must use full mask for"
2602 * Match on frag_data 0x00001 means M is 1 and frag-offset is 0.
2603 * This is 1st fragment of fragmented packet.
2605 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_MF_MASK))
2606 return rte_flow_error_set(error, ENOTSUP,
2607 RTE_FLOW_ERROR_TYPE_ITEM, item,
2608 "match on first fragment not "
2610 if (frag_data_spec && !last)
2611 return rte_flow_error_set(error, EINVAL,
2612 RTE_FLOW_ERROR_TYPE_ITEM, item,
2613 "specified value not supported");
2614 ret = mlx5_flow_item_acceptable
2615 (item, (const uint8_t *)mask,
2616 (const uint8_t *)&nic_mask,
2617 sizeof(struct rte_flow_item_ipv6_frag_ext),
2618 MLX5_ITEM_RANGE_ACCEPTED, error);
2621 /* spec and last are valid, validate the specified range. */
2622 frag_data_last = last->hdr.frag_data & mask->hdr.frag_data;
2624 * Match on frag_data spec 0x0009 and last 0xfff9
2625 * means M is 1 and frag-offset is > 0.
2626 * This packet is fragment 2nd and onward, excluding last.
2627 * This is not yet supported in MLX5, return appropriate
2630 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN |
2631 RTE_IPV6_EHDR_MF_MASK) &&
2632 frag_data_last == RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
2633 return rte_flow_error_set(error, ENOTSUP,
2634 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2635 last, "match on following "
2636 "fragments not supported");
2638 * Match on frag_data spec 0x0008 and last 0xfff8
2639 * means M is 0 and frag-offset is > 0.
2640 * This packet is last fragment of fragmented packet.
2641 * This is not yet supported in MLX5, return appropriate
2644 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN) &&
2645 frag_data_last == RTE_BE16(RTE_IPV6_EHDR_FO_MASK))
2646 return rte_flow_error_set(error, ENOTSUP,
2647 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2648 last, "match on last "
2649 "fragment not supported");
2650 /* Other range values are invalid and rejected. */
2651 return rte_flow_error_set(error, EINVAL,
2652 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
2653 "specified range not supported");
2657 * Validate ASO CT item.
2660 * Pointer to the rte_eth_dev structure.
2662 * Item specification.
2663 * @param[in] item_flags
2664 * Pointer to bit-fields that holds the items detected until now.
2666 * Pointer to error structure.
2669 * 0 on success, a negative errno value otherwise and rte_errno is set.
2672 flow_dv_validate_item_aso_ct(struct rte_eth_dev *dev,
2673 const struct rte_flow_item *item,
2674 uint64_t *item_flags,
2675 struct rte_flow_error *error)
2677 const struct rte_flow_item_conntrack *spec = item->spec;
2678 const struct rte_flow_item_conntrack *mask = item->mask;
2682 if (*item_flags & MLX5_FLOW_LAYER_ASO_CT)
2683 return rte_flow_error_set(error, EINVAL,
2684 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
2685 "Only one CT is supported");
2687 mask = &rte_flow_item_conntrack_mask;
2688 flags = spec->flags & mask->flags;
2689 if ((flags & RTE_FLOW_CONNTRACK_PKT_STATE_VALID) &&
2690 ((flags & RTE_FLOW_CONNTRACK_PKT_STATE_INVALID) ||
2691 (flags & RTE_FLOW_CONNTRACK_PKT_STATE_BAD) ||
2692 (flags & RTE_FLOW_CONNTRACK_PKT_STATE_DISABLED)))
2693 return rte_flow_error_set(error, EINVAL,
2694 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
2695 "Conflict status bits");
2696 /* State change also needs to be considered. */
2697 *item_flags |= MLX5_FLOW_LAYER_ASO_CT;
2702 * Validate the pop VLAN action.
2705 * Pointer to the rte_eth_dev structure.
2706 * @param[in] action_flags
2707 * Holds the actions detected until now.
2709 * Pointer to the pop vlan action.
2710 * @param[in] item_flags
2711 * The items found in this flow rule.
2713 * Pointer to flow attributes.
2715 * Pointer to error structure.
2718 * 0 on success, a negative errno value otherwise and rte_errno is set.
2721 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
2722 uint64_t action_flags,
2723 const struct rte_flow_action *action,
2724 uint64_t item_flags,
2725 const struct rte_flow_attr *attr,
2726 struct rte_flow_error *error)
2728 const struct mlx5_priv *priv = dev->data->dev_private;
2729 struct mlx5_dev_ctx_shared *sh = priv->sh;
2730 bool direction_error = false;
2732 if (!priv->sh->pop_vlan_action)
2733 return rte_flow_error_set(error, ENOTSUP,
2734 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2736 "pop vlan action is not supported");
2737 /* Pop VLAN is not supported in egress except for CX6 FDB mode. */
2738 if (attr->transfer) {
2739 bool fdb_tx = priv->representor_id != UINT16_MAX;
2740 bool is_cx5 = sh->steering_format_version ==
2741 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5;
2743 if (fdb_tx && is_cx5)
2744 direction_error = true;
2745 } else if (attr->egress) {
2746 direction_error = true;
2748 if (direction_error)
2749 return rte_flow_error_set(error, ENOTSUP,
2750 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2752 "pop vlan action not supported for egress");
2753 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
2754 return rte_flow_error_set(error, ENOTSUP,
2755 RTE_FLOW_ERROR_TYPE_ACTION, action,
2756 "no support for multiple VLAN "
2758 /* Pop VLAN with preceding Decap requires inner header with VLAN. */
2759 if ((action_flags & MLX5_FLOW_ACTION_DECAP) &&
2760 !(item_flags & MLX5_FLOW_LAYER_INNER_VLAN))
2761 return rte_flow_error_set(error, ENOTSUP,
2762 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2764 "cannot pop vlan after decap without "
2765 "match on inner vlan in the flow");
2766 /* Pop VLAN without preceding Decap requires outer header with VLAN. */
2767 if (!(action_flags & MLX5_FLOW_ACTION_DECAP) &&
2768 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2769 return rte_flow_error_set(error, ENOTSUP,
2770 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2772 "cannot pop vlan without a "
2773 "match on (outer) vlan in the flow");
2774 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2775 return rte_flow_error_set(error, EINVAL,
2776 RTE_FLOW_ERROR_TYPE_ACTION, action,
2777 "wrong action order, port_id should "
2778 "be after pop VLAN action");
2779 if (!attr->transfer && priv->representor)
2780 return rte_flow_error_set(error, ENOTSUP,
2781 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2782 "pop vlan action for VF representor "
2783 "not supported on NIC table");
2788 * Get VLAN default info from vlan match info.
2791 * the list of item specifications.
2793 * pointer VLAN info to fill to.
2796 * 0 on success, a negative errno value otherwise and rte_errno is set.
2799 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
2800 struct rte_vlan_hdr *vlan)
2802 const struct rte_flow_item_vlan nic_mask = {
2803 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
2804 MLX5DV_FLOW_VLAN_VID_MASK),
2805 .inner_type = RTE_BE16(0xffff),
2810 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2811 int type = items->type;
2813 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
2814 type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
2817 if (items->type != RTE_FLOW_ITEM_TYPE_END) {
2818 const struct rte_flow_item_vlan *vlan_m = items->mask;
2819 const struct rte_flow_item_vlan *vlan_v = items->spec;
2821 /* If VLAN item in pattern doesn't contain data, return here. */
2826 /* Only full match values are accepted */
2827 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
2828 MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
2829 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
2831 rte_be_to_cpu_16(vlan_v->tci &
2832 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
2834 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
2835 MLX5DV_FLOW_VLAN_VID_MASK_BE) {
2836 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
2838 rte_be_to_cpu_16(vlan_v->tci &
2839 MLX5DV_FLOW_VLAN_VID_MASK_BE);
2841 if (vlan_m->inner_type == nic_mask.inner_type)
2842 vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
2843 vlan_m->inner_type);
2848 * Validate the push VLAN action.
2851 * Pointer to the rte_eth_dev structure.
2852 * @param[in] action_flags
2853 * Holds the actions detected until now.
2854 * @param[in] item_flags
2855 * The items found in this flow rule.
2857 * Pointer to the action structure.
2859 * Pointer to flow attributes
2861 * Pointer to error structure.
2864 * 0 on success, a negative errno value otherwise and rte_errno is set.
2867 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
2868 uint64_t action_flags,
2869 const struct rte_flow_item_vlan *vlan_m,
2870 const struct rte_flow_action *action,
2871 const struct rte_flow_attr *attr,
2872 struct rte_flow_error *error)
2874 const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
2875 const struct mlx5_priv *priv = dev->data->dev_private;
2876 struct mlx5_dev_ctx_shared *sh = priv->sh;
2877 bool direction_error = false;
2879 if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
2880 push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
2881 return rte_flow_error_set(error, EINVAL,
2882 RTE_FLOW_ERROR_TYPE_ACTION, action,
2883 "invalid vlan ethertype");
2884 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2885 return rte_flow_error_set(error, EINVAL,
2886 RTE_FLOW_ERROR_TYPE_ACTION, action,
2887 "wrong action order, port_id should "
2888 "be after push VLAN");
2889 /* Push VLAN is not supported in ingress except for CX6 FDB mode. */
2890 if (attr->transfer) {
2891 bool fdb_tx = priv->representor_id != UINT16_MAX;
2892 bool is_cx5 = sh->steering_format_version ==
2893 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5;
2895 if (!fdb_tx && is_cx5)
2896 direction_error = true;
2897 } else if (attr->ingress) {
2898 direction_error = true;
2900 if (direction_error)
2901 return rte_flow_error_set(error, ENOTSUP,
2902 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
2904 "push vlan action not supported for ingress");
2905 if (!attr->transfer && priv->representor)
2906 return rte_flow_error_set(error, ENOTSUP,
2907 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2908 "push vlan action for VF representor "
2909 "not supported on NIC table");
2911 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
2912 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
2913 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
2914 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
2915 !(mlx5_flow_find_action
2916 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
2917 return rte_flow_error_set(error, EINVAL,
2918 RTE_FLOW_ERROR_TYPE_ACTION, action,
2919 "not full match mask on VLAN PCP and "
2920 "there is no of_set_vlan_pcp action, "
2921 "push VLAN action cannot figure out "
2924 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
2925 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
2926 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
2927 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
2928 !(mlx5_flow_find_action
2929 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
2930 return rte_flow_error_set(error, EINVAL,
2931 RTE_FLOW_ERROR_TYPE_ACTION, action,
2932 "not full match mask on VLAN VID and "
2933 "there is no of_set_vlan_vid action, "
2934 "push VLAN action cannot figure out "
2941 * Validate the set VLAN PCP.
2943 * @param[in] action_flags
2944 * Holds the actions detected until now.
2945 * @param[in] actions
2946 * Pointer to the list of actions remaining in the flow rule.
2948 * Pointer to error structure.
2951 * 0 on success, a negative errno value otherwise and rte_errno is set.
2954 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
2955 const struct rte_flow_action actions[],
2956 struct rte_flow_error *error)
2958 const struct rte_flow_action *action = actions;
2959 const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2961 if (conf->vlan_pcp > 7)
2962 return rte_flow_error_set(error, EINVAL,
2963 RTE_FLOW_ERROR_TYPE_ACTION, action,
2964 "VLAN PCP value is too big");
2965 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2966 return rte_flow_error_set(error, ENOTSUP,
2967 RTE_FLOW_ERROR_TYPE_ACTION, action,
2968 "set VLAN PCP action must follow "
2969 "the push VLAN action");
2970 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2971 return rte_flow_error_set(error, ENOTSUP,
2972 RTE_FLOW_ERROR_TYPE_ACTION, action,
2973 "Multiple VLAN PCP modification are "
2975 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2976 return rte_flow_error_set(error, EINVAL,
2977 RTE_FLOW_ERROR_TYPE_ACTION, action,
2978 "wrong action order, port_id should "
2979 "be after set VLAN PCP");
2984 * Validate the set VLAN VID.
2986 * @param[in] item_flags
2987 * Holds the items detected in this rule.
2988 * @param[in] action_flags
2989 * Holds the actions detected until now.
2990 * @param[in] actions
2991 * Pointer to the list of actions remaining in the flow rule.
2993 * Pointer to error structure.
2996 * 0 on success, a negative errno value otherwise and rte_errno is set.
2999 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
3000 uint64_t action_flags,
3001 const struct rte_flow_action actions[],
3002 struct rte_flow_error *error)
3004 const struct rte_flow_action *action = actions;
3005 const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
3007 if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
3008 return rte_flow_error_set(error, EINVAL,
3009 RTE_FLOW_ERROR_TYPE_ACTION, action,
3010 "VLAN VID value is too big");
3011 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
3012 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
3013 return rte_flow_error_set(error, ENOTSUP,
3014 RTE_FLOW_ERROR_TYPE_ACTION, action,
3015 "set VLAN VID action must follow push"
3016 " VLAN action or match on VLAN item");
3017 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
3018 return rte_flow_error_set(error, ENOTSUP,
3019 RTE_FLOW_ERROR_TYPE_ACTION, action,
3020 "Multiple VLAN VID modifications are "
3022 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
3023 return rte_flow_error_set(error, EINVAL,
3024 RTE_FLOW_ERROR_TYPE_ACTION, action,
3025 "wrong action order, port_id should "
3026 "be after set VLAN VID");
3031 * Validate the FLAG action.
3034 * Pointer to the rte_eth_dev structure.
3035 * @param[in] action_flags
3036 * Holds the actions detected until now.
3038 * Pointer to flow attributes
3040 * Pointer to error structure.
3043 * 0 on success, a negative errno value otherwise and rte_errno is set.
3046 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
3047 uint64_t action_flags,
3048 const struct rte_flow_attr *attr,
3049 struct rte_flow_error *error)
3051 struct mlx5_priv *priv = dev->data->dev_private;
3052 struct mlx5_sh_config *config = &priv->sh->config;
3055 /* Fall back if no extended metadata register support. */
3056 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
3057 return mlx5_flow_validate_action_flag(action_flags, attr,
3059 /* Extensive metadata mode requires registers. */
3060 if (!mlx5_flow_ext_mreg_supported(dev))
3061 return rte_flow_error_set(error, ENOTSUP,
3062 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3063 "no metadata registers "
3064 "to support flag action");
3065 if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
3066 return rte_flow_error_set(error, ENOTSUP,
3067 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3068 "extended metadata register"
3069 " isn't available");
3070 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3073 MLX5_ASSERT(ret > 0);
3074 if (action_flags & MLX5_FLOW_ACTION_MARK)
3075 return rte_flow_error_set(error, EINVAL,
3076 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3077 "can't mark and flag in same flow");
3078 if (action_flags & MLX5_FLOW_ACTION_FLAG)
3079 return rte_flow_error_set(error, EINVAL,
3080 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3082 " actions in same flow");
3087 * Validate MARK action.
3090 * Pointer to the rte_eth_dev structure.
3092 * Pointer to action.
3093 * @param[in] action_flags
3094 * Holds the actions detected until now.
3096 * Pointer to flow attributes
3098 * Pointer to error structure.
3101 * 0 on success, a negative errno value otherwise and rte_errno is set.
3104 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
3105 const struct rte_flow_action *action,
3106 uint64_t action_flags,
3107 const struct rte_flow_attr *attr,
3108 struct rte_flow_error *error)
3110 struct mlx5_priv *priv = dev->data->dev_private;
3111 struct mlx5_sh_config *config = &priv->sh->config;
3112 const struct rte_flow_action_mark *mark = action->conf;
3115 if (is_tunnel_offload_active(dev))
3116 return rte_flow_error_set(error, ENOTSUP,
3117 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3119 "if tunnel offload active");
3120 /* Fall back if no extended metadata register support. */
3121 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
3122 return mlx5_flow_validate_action_mark(action, action_flags,
3124 /* Extensive metadata mode requires registers. */
3125 if (!mlx5_flow_ext_mreg_supported(dev))
3126 return rte_flow_error_set(error, ENOTSUP,
3127 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3128 "no metadata registers "
3129 "to support mark action");
3130 if (!priv->sh->dv_mark_mask)
3131 return rte_flow_error_set(error, ENOTSUP,
3132 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3133 "extended metadata register"
3134 " isn't available");
3135 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3138 MLX5_ASSERT(ret > 0);
3140 return rte_flow_error_set(error, EINVAL,
3141 RTE_FLOW_ERROR_TYPE_ACTION, action,
3142 "configuration cannot be null");
3143 if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
3144 return rte_flow_error_set(error, EINVAL,
3145 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3147 "mark id exceeds the limit");
3148 if (action_flags & MLX5_FLOW_ACTION_FLAG)
3149 return rte_flow_error_set(error, EINVAL,
3150 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3151 "can't flag and mark in same flow");
3152 if (action_flags & MLX5_FLOW_ACTION_MARK)
3153 return rte_flow_error_set(error, EINVAL,
3154 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3155 "can't have 2 mark actions in same"
3161 * Validate SET_META action.
3164 * Pointer to the rte_eth_dev structure.
3166 * Pointer to the action structure.
3167 * @param[in] action_flags
3168 * Holds the actions detected until now.
3170 * Pointer to flow attributes
3172 * Pointer to error structure.
3175 * 0 on success, a negative errno value otherwise and rte_errno is set.
3178 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
3179 const struct rte_flow_action *action,
3180 uint64_t action_flags __rte_unused,
3181 const struct rte_flow_attr *attr,
3182 struct rte_flow_error *error)
3184 struct mlx5_priv *priv = dev->data->dev_private;
3185 struct mlx5_sh_config *config = &priv->sh->config;
3186 const struct rte_flow_action_set_meta *conf;
3187 uint32_t nic_mask = UINT32_MAX;
3190 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
3191 !mlx5_flow_ext_mreg_supported(dev))
3192 return rte_flow_error_set(error, ENOTSUP,
3193 RTE_FLOW_ERROR_TYPE_ACTION, action,
3194 "extended metadata register"
3195 " isn't supported");
3196 reg = flow_dv_get_metadata_reg(dev, attr, error);
3200 return rte_flow_error_set(error, ENOTSUP,
3201 RTE_FLOW_ERROR_TYPE_ACTION, action,
3202 "unavailable extended metadata register");
3203 if (reg != REG_A && reg != REG_B) {
3204 struct mlx5_priv *priv = dev->data->dev_private;
3206 nic_mask = priv->sh->dv_meta_mask;
3208 if (!(action->conf))
3209 return rte_flow_error_set(error, EINVAL,
3210 RTE_FLOW_ERROR_TYPE_ACTION, action,
3211 "configuration cannot be null");
3212 conf = (const struct rte_flow_action_set_meta *)action->conf;
3214 return rte_flow_error_set(error, EINVAL,
3215 RTE_FLOW_ERROR_TYPE_ACTION, action,
3216 "zero mask doesn't have any effect");
3217 if (conf->mask & ~nic_mask)
3218 return rte_flow_error_set(error, EINVAL,
3219 RTE_FLOW_ERROR_TYPE_ACTION, action,
3220 "meta data must be within reg C0");
3225 * Validate SET_TAG action.
3228 * Pointer to the rte_eth_dev structure.
3230 * Pointer to the action structure.
3231 * @param[in] action_flags
3232 * Holds the actions detected until now.
3234 * Pointer to flow attributes
3236 * Pointer to error structure.
3239 * 0 on success, a negative errno value otherwise and rte_errno is set.
3242 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
3243 const struct rte_flow_action *action,
3244 uint64_t action_flags,
3245 const struct rte_flow_attr *attr,
3246 struct rte_flow_error *error)
3248 const struct rte_flow_action_set_tag *conf;
3249 const uint64_t terminal_action_flags =
3250 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
3251 MLX5_FLOW_ACTION_RSS;
3254 if (!mlx5_flow_ext_mreg_supported(dev))
3255 return rte_flow_error_set(error, ENOTSUP,
3256 RTE_FLOW_ERROR_TYPE_ACTION, action,
3257 "extensive metadata register"
3258 " isn't supported");
3259 if (!(action->conf))
3260 return rte_flow_error_set(error, EINVAL,
3261 RTE_FLOW_ERROR_TYPE_ACTION, action,
3262 "configuration cannot be null");
3263 conf = (const struct rte_flow_action_set_tag *)action->conf;
3265 return rte_flow_error_set(error, EINVAL,
3266 RTE_FLOW_ERROR_TYPE_ACTION, action,
3267 "zero mask doesn't have any effect");
3268 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
3271 if (!attr->transfer && attr->ingress &&
3272 (action_flags & terminal_action_flags))
3273 return rte_flow_error_set(error, EINVAL,
3274 RTE_FLOW_ERROR_TYPE_ACTION, action,
3275 "set_tag has no effect"
3276 " with terminal actions");
3281 * Validate count action.
3284 * Pointer to rte_eth_dev structure.
3286 * Indicator if action is shared.
3287 * @param[in] action_flags
3288 * Holds the actions detected until now.
3290 * Pointer to error structure.
3293 * 0 on success, a negative errno value otherwise and rte_errno is set.
3296 flow_dv_validate_action_count(struct rte_eth_dev *dev, bool shared,
3297 uint64_t action_flags,
3298 struct rte_flow_error *error)
3300 struct mlx5_priv *priv = dev->data->dev_private;
3302 if (!priv->sh->cdev->config.devx)
3304 if (action_flags & MLX5_FLOW_ACTION_COUNT)
3305 return rte_flow_error_set(error, EINVAL,
3306 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3307 "duplicate count actions set");
3308 if (shared && (action_flags & MLX5_FLOW_ACTION_AGE) &&
3309 !priv->sh->flow_hit_aso_en)
3310 return rte_flow_error_set(error, EINVAL,
3311 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3312 "old age and shared count combination is not supported");
3313 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
3317 return rte_flow_error_set
3319 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3321 "count action not supported");
3325 * Validate the L2 encap action.
3328 * Pointer to the rte_eth_dev structure.
3329 * @param[in] action_flags
3330 * Holds the actions detected until now.
3332 * Pointer to the action structure.
3334 * Pointer to flow attributes.
3336 * Pointer to error structure.
3339 * 0 on success, a negative errno value otherwise and rte_errno is set.
3342 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
3343 uint64_t action_flags,
3344 const struct rte_flow_action *action,
3345 const struct rte_flow_attr *attr,
3346 struct rte_flow_error *error)
3348 const struct mlx5_priv *priv = dev->data->dev_private;
3350 if (!(action->conf))
3351 return rte_flow_error_set(error, EINVAL,
3352 RTE_FLOW_ERROR_TYPE_ACTION, action,
3353 "configuration cannot be null");
3354 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3355 return rte_flow_error_set(error, EINVAL,
3356 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3357 "can only have a single encap action "
3359 if (!attr->transfer && priv->representor)
3360 return rte_flow_error_set(error, ENOTSUP,
3361 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3362 "encap action for VF representor "
3363 "not supported on NIC table");
3368 * Validate a decap action.
3371 * Pointer to the rte_eth_dev structure.
3372 * @param[in] action_flags
3373 * Holds the actions detected until now.
3375 * Pointer to the action structure.
3376 * @param[in] item_flags
3377 * Holds the items detected.
3379 * Pointer to flow attributes
3381 * Pointer to error structure.
3384 * 0 on success, a negative errno value otherwise and rte_errno is set.
3387 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
3388 uint64_t action_flags,
3389 const struct rte_flow_action *action,
3390 const uint64_t item_flags,
3391 const struct rte_flow_attr *attr,
3392 struct rte_flow_error *error)
3394 const struct mlx5_priv *priv = dev->data->dev_private;
3396 if (priv->sh->cdev->config.hca_attr.scatter_fcs_w_decap_disable &&
3397 !priv->sh->config.decap_en)
3398 return rte_flow_error_set(error, ENOTSUP,
3399 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3400 "decap is not enabled");
3401 if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
3402 return rte_flow_error_set(error, ENOTSUP,
3403 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3405 MLX5_FLOW_ACTION_DECAP ? "can only "
3406 "have a single decap action" : "decap "
3407 "after encap is not supported");
3408 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
3409 return rte_flow_error_set(error, EINVAL,
3410 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3411 "can't have decap action after"
3414 return rte_flow_error_set(error, ENOTSUP,
3415 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
3417 "decap action not supported for "
3419 if (!attr->transfer && priv->representor)
3420 return rte_flow_error_set(error, ENOTSUP,
3421 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3422 "decap action for VF representor "
3423 "not supported on NIC table");
3424 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_DECAP &&
3425 !(item_flags & MLX5_FLOW_LAYER_VXLAN))
3426 return rte_flow_error_set(error, ENOTSUP,
3427 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3428 "VXLAN item should be present for VXLAN decap");
3432 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
3435 * Validate the raw encap and decap actions.
3438 * Pointer to the rte_eth_dev structure.
3440 * Pointer to the decap action.
3442 * Pointer to the encap action.
3444 * Pointer to flow attributes
3445 * @param[in/out] action_flags
3446 * Holds the actions detected until now.
3447 * @param[out] actions_n
3448 * pointer to the number of actions counter.
3450 * Pointer to the action structure.
3451 * @param[in] item_flags
3452 * Holds the items detected.
3454 * Pointer to error structure.
3457 * 0 on success, a negative errno value otherwise and rte_errno is set.
3460 flow_dv_validate_action_raw_encap_decap
3461 (struct rte_eth_dev *dev,
3462 const struct rte_flow_action_raw_decap *decap,
3463 const struct rte_flow_action_raw_encap *encap,
3464 const struct rte_flow_attr *attr, uint64_t *action_flags,
3465 int *actions_n, const struct rte_flow_action *action,
3466 uint64_t item_flags, struct rte_flow_error *error)
3468 const struct mlx5_priv *priv = dev->data->dev_private;
3471 if (encap && (!encap->size || !encap->data))
3472 return rte_flow_error_set(error, EINVAL,
3473 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3474 "raw encap data cannot be empty");
3475 if (decap && encap) {
3476 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
3477 encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3480 else if (encap->size <=
3481 MLX5_ENCAPSULATION_DECISION_SIZE &&
3483 MLX5_ENCAPSULATION_DECISION_SIZE)
3486 else if (encap->size >
3487 MLX5_ENCAPSULATION_DECISION_SIZE &&
3489 MLX5_ENCAPSULATION_DECISION_SIZE)
3490 /* 2 L2 actions: encap and decap. */
3493 return rte_flow_error_set(error,
3495 RTE_FLOW_ERROR_TYPE_ACTION,
3496 NULL, "unsupported too small "
3497 "raw decap and too small raw "
3498 "encap combination");
3501 ret = flow_dv_validate_action_decap(dev, *action_flags, action,
3502 item_flags, attr, error);
3505 *action_flags |= MLX5_FLOW_ACTION_DECAP;
3509 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
3510 return rte_flow_error_set(error, ENOTSUP,
3511 RTE_FLOW_ERROR_TYPE_ACTION,
3513 "small raw encap size");
3514 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
3515 return rte_flow_error_set(error, EINVAL,
3516 RTE_FLOW_ERROR_TYPE_ACTION,
3518 "more than one encap action");
3519 if (!attr->transfer && priv->representor)
3520 return rte_flow_error_set
3522 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3523 "encap action for VF representor "
3524 "not supported on NIC table");
3525 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
3532 * Validate the ASO CT action.
3535 * Pointer to the rte_eth_dev structure.
3536 * @param[in] action_flags
3537 * Holds the actions detected until now.
3538 * @param[in] item_flags
3539 * The items found in this flow rule.
3541 * Pointer to flow attributes.
3543 * Pointer to error structure.
3546 * 0 on success, a negative errno value otherwise and rte_errno is set.
3549 flow_dv_validate_action_aso_ct(struct rte_eth_dev *dev,
3550 uint64_t action_flags,
3551 uint64_t item_flags,
3552 const struct rte_flow_attr *attr,
3553 struct rte_flow_error *error)
3557 if (attr->group == 0 && !attr->transfer)
3558 return rte_flow_error_set(error, ENOTSUP,
3559 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3561 "Only support non-root table");
3562 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
3563 return rte_flow_error_set(error, ENOTSUP,
3564 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3565 "CT cannot follow a fate action");
3566 if ((action_flags & MLX5_FLOW_ACTION_METER) ||
3567 (action_flags & MLX5_FLOW_ACTION_AGE))
3568 return rte_flow_error_set(error, EINVAL,
3569 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3570 "Only one ASO action is supported");
3571 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3572 return rte_flow_error_set(error, EINVAL,
3573 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3574 "Encap cannot exist before CT");
3575 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP))
3576 return rte_flow_error_set(error, EINVAL,
3577 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3578 "Not a outer TCP packet");
3583 flow_dv_encap_decap_match_cb(void *tool_ctx __rte_unused,
3584 struct mlx5_list_entry *entry, void *cb_ctx)
3586 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3587 struct mlx5_flow_dv_encap_decap_resource *ctx_resource = ctx->data;
3588 struct mlx5_flow_dv_encap_decap_resource *resource;
3590 resource = container_of(entry, struct mlx5_flow_dv_encap_decap_resource,
3592 if (resource->reformat_type == ctx_resource->reformat_type &&
3593 resource->ft_type == ctx_resource->ft_type &&
3594 resource->flags == ctx_resource->flags &&
3595 resource->size == ctx_resource->size &&
3596 !memcmp((const void *)resource->buf,
3597 (const void *)ctx_resource->buf,
3603 struct mlx5_list_entry *
3604 flow_dv_encap_decap_create_cb(void *tool_ctx, void *cb_ctx)
3606 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3607 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3608 struct mlx5dv_dr_domain *domain;
3609 struct mlx5_flow_dv_encap_decap_resource *ctx_resource = ctx->data;
3610 struct mlx5_flow_dv_encap_decap_resource *resource;
3614 if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3615 domain = sh->fdb_domain;
3616 else if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
3617 domain = sh->rx_domain;
3619 domain = sh->tx_domain;
3620 /* Register new encap/decap resource. */
3621 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], &idx);
3623 rte_flow_error_set(ctx->error, ENOMEM,
3624 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3625 "cannot allocate resource memory");
3628 *resource = *ctx_resource;
3629 resource->idx = idx;
3630 ret = mlx5_flow_os_create_flow_action_packet_reformat(sh->cdev->ctx,
3634 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
3635 rte_flow_error_set(ctx->error, ENOMEM,
3636 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3637 NULL, "cannot create action");
3641 return &resource->entry;
3644 struct mlx5_list_entry *
3645 flow_dv_encap_decap_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
3648 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3649 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3650 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
3653 cache_resource = mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
3655 if (!cache_resource) {
3656 rte_flow_error_set(ctx->error, ENOMEM,
3657 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3658 "cannot allocate resource memory");
3661 memcpy(cache_resource, oentry, sizeof(*cache_resource));
3662 cache_resource->idx = idx;
3663 return &cache_resource->entry;
3667 flow_dv_encap_decap_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
3669 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3670 struct mlx5_flow_dv_encap_decap_resource *res =
3671 container_of(entry, typeof(*res), entry);
3673 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], res->idx);
3677 * Find existing encap/decap resource or create and register a new one.
3679 * @param[in, out] dev
3680 * Pointer to rte_eth_dev structure.
3681 * @param[in, out] resource
3682 * Pointer to encap/decap resource.
3683 * @parm[in, out] dev_flow
3684 * Pointer to the dev_flow.
3686 * pointer to error structure.
3689 * 0 on success otherwise -errno and errno is set.
3692 flow_dv_encap_decap_resource_register
3693 (struct rte_eth_dev *dev,
3694 struct mlx5_flow_dv_encap_decap_resource *resource,
3695 struct mlx5_flow *dev_flow,
3696 struct rte_flow_error *error)
3698 struct mlx5_priv *priv = dev->data->dev_private;
3699 struct mlx5_dev_ctx_shared *sh = priv->sh;
3700 struct mlx5_list_entry *entry;
3704 uint32_t refmt_type:8;
3706 * Header reformat actions can be shared between
3707 * non-root tables. One bit to indicate non-root
3711 uint32_t reserve:15;
3714 } encap_decap_key = {
3716 .ft_type = resource->ft_type,
3717 .refmt_type = resource->reformat_type,
3718 .is_root = !!dev_flow->dv.group,
3722 struct mlx5_flow_cb_ctx ctx = {
3726 struct mlx5_hlist *encaps_decaps;
3729 encaps_decaps = flow_dv_hlist_prepare(sh, &sh->encaps_decaps,
3731 MLX5_FLOW_ENCAP_DECAP_HTABLE_SZ,
3733 flow_dv_encap_decap_create_cb,
3734 flow_dv_encap_decap_match_cb,
3735 flow_dv_encap_decap_remove_cb,
3736 flow_dv_encap_decap_clone_cb,
3737 flow_dv_encap_decap_clone_free_cb,
3739 if (unlikely(!encaps_decaps))
3741 resource->flags = dev_flow->dv.group ? 0 : 1;
3742 key64 = __rte_raw_cksum(&encap_decap_key.v32,
3743 sizeof(encap_decap_key.v32), 0);
3744 if (resource->reformat_type !=
3745 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2 &&
3747 key64 = __rte_raw_cksum(resource->buf, resource->size, key64);
3748 entry = mlx5_hlist_register(encaps_decaps, key64, &ctx);
3751 resource = container_of(entry, typeof(*resource), entry);
3752 dev_flow->dv.encap_decap = resource;
3753 dev_flow->handle->dvh.rix_encap_decap = resource->idx;
3758 * Find existing table jump resource or create and register a new one.
3760 * @param[in, out] dev
3761 * Pointer to rte_eth_dev structure.
3762 * @param[in, out] tbl
3763 * Pointer to flow table resource.
3764 * @parm[in, out] dev_flow
3765 * Pointer to the dev_flow.
3767 * pointer to error structure.
3770 * 0 on success otherwise -errno and errno is set.
3773 flow_dv_jump_tbl_resource_register
3774 (struct rte_eth_dev *dev __rte_unused,
3775 struct mlx5_flow_tbl_resource *tbl,
3776 struct mlx5_flow *dev_flow,
3777 struct rte_flow_error *error __rte_unused)
3779 struct mlx5_flow_tbl_data_entry *tbl_data =
3780 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
3783 MLX5_ASSERT(tbl_data->jump.action);
3784 dev_flow->handle->rix_jump = tbl_data->idx;
3785 dev_flow->dv.jump = &tbl_data->jump;
3790 flow_dv_port_id_match_cb(void *tool_ctx __rte_unused,
3791 struct mlx5_list_entry *entry, void *cb_ctx)
3793 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3794 struct mlx5_flow_dv_port_id_action_resource *ref = ctx->data;
3795 struct mlx5_flow_dv_port_id_action_resource *res =
3796 container_of(entry, typeof(*res), entry);
3798 return ref->port_id != res->port_id;
3801 struct mlx5_list_entry *
3802 flow_dv_port_id_create_cb(void *tool_ctx, void *cb_ctx)
3804 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3805 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3806 struct mlx5_flow_dv_port_id_action_resource *ref = ctx->data;
3807 struct mlx5_flow_dv_port_id_action_resource *resource;
3811 /* Register new port id action resource. */
3812 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID], &idx);
3814 rte_flow_error_set(ctx->error, ENOMEM,
3815 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3816 "cannot allocate port_id action memory");
3820 ret = mlx5_flow_os_create_flow_action_dest_port(sh->fdb_domain,
3824 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PORT_ID], idx);
3825 rte_flow_error_set(ctx->error, ENOMEM,
3826 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3827 "cannot create action");
3830 resource->idx = idx;
3831 return &resource->entry;
3834 struct mlx5_list_entry *
3835 flow_dv_port_id_clone_cb(void *tool_ctx,
3836 struct mlx5_list_entry *entry __rte_unused,
3839 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3840 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3841 struct mlx5_flow_dv_port_id_action_resource *resource;
3844 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID], &idx);
3846 rte_flow_error_set(ctx->error, ENOMEM,
3847 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3848 "cannot allocate port_id action memory");
3851 memcpy(resource, entry, sizeof(*resource));
3852 resource->idx = idx;
3853 return &resource->entry;
3857 flow_dv_port_id_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
3859 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3860 struct mlx5_flow_dv_port_id_action_resource *resource =
3861 container_of(entry, typeof(*resource), entry);
3863 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PORT_ID], resource->idx);
3867 * Find existing table port ID resource or create and register a new one.
3869 * @param[in, out] dev
3870 * Pointer to rte_eth_dev structure.
3871 * @param[in, out] ref
3872 * Pointer to port ID action resource reference.
3873 * @parm[in, out] dev_flow
3874 * Pointer to the dev_flow.
3876 * pointer to error structure.
3879 * 0 on success otherwise -errno and errno is set.
3882 flow_dv_port_id_action_resource_register
3883 (struct rte_eth_dev *dev,
3884 struct mlx5_flow_dv_port_id_action_resource *ref,
3885 struct mlx5_flow *dev_flow,
3886 struct rte_flow_error *error)
3888 struct mlx5_priv *priv = dev->data->dev_private;
3889 struct mlx5_list_entry *entry;
3890 struct mlx5_flow_dv_port_id_action_resource *resource;
3891 struct mlx5_flow_cb_ctx ctx = {
3896 entry = mlx5_list_register(priv->sh->port_id_action_list, &ctx);
3899 resource = container_of(entry, typeof(*resource), entry);
3900 dev_flow->dv.port_id_action = resource;
3901 dev_flow->handle->rix_port_id_action = resource->idx;
3906 flow_dv_push_vlan_match_cb(void *tool_ctx __rte_unused,
3907 struct mlx5_list_entry *entry, void *cb_ctx)
3909 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3910 struct mlx5_flow_dv_push_vlan_action_resource *ref = ctx->data;
3911 struct mlx5_flow_dv_push_vlan_action_resource *res =
3912 container_of(entry, typeof(*res), entry);
3914 return ref->vlan_tag != res->vlan_tag || ref->ft_type != res->ft_type;
3917 struct mlx5_list_entry *
3918 flow_dv_push_vlan_create_cb(void *tool_ctx, void *cb_ctx)
3920 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3921 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3922 struct mlx5_flow_dv_push_vlan_action_resource *ref = ctx->data;
3923 struct mlx5_flow_dv_push_vlan_action_resource *resource;
3924 struct mlx5dv_dr_domain *domain;
3928 /* Register new port id action resource. */
3929 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN], &idx);
3931 rte_flow_error_set(ctx->error, ENOMEM,
3932 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3933 "cannot allocate push_vlan action memory");
3937 if (ref->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3938 domain = sh->fdb_domain;
3939 else if (ref->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
3940 domain = sh->rx_domain;
3942 domain = sh->tx_domain;
3943 ret = mlx5_flow_os_create_flow_action_push_vlan(domain, ref->vlan_tag,
3946 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
3947 rte_flow_error_set(ctx->error, ENOMEM,
3948 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3949 "cannot create push vlan action");
3952 resource->idx = idx;
3953 return &resource->entry;
3956 struct mlx5_list_entry *
3957 flow_dv_push_vlan_clone_cb(void *tool_ctx,
3958 struct mlx5_list_entry *entry __rte_unused,
3961 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3962 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3963 struct mlx5_flow_dv_push_vlan_action_resource *resource;
3966 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN], &idx);
3968 rte_flow_error_set(ctx->error, ENOMEM,
3969 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3970 "cannot allocate push_vlan action memory");
3973 memcpy(resource, entry, sizeof(*resource));
3974 resource->idx = idx;
3975 return &resource->entry;
3979 flow_dv_push_vlan_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
3981 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3982 struct mlx5_flow_dv_push_vlan_action_resource *resource =
3983 container_of(entry, typeof(*resource), entry);
3985 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PUSH_VLAN], resource->idx);
3989 * Find existing push vlan resource or create and register a new one.
3991 * @param [in, out] dev
3992 * Pointer to rte_eth_dev structure.
3993 * @param[in, out] ref
3994 * Pointer to port ID action resource reference.
3995 * @parm[in, out] dev_flow
3996 * Pointer to the dev_flow.
3998 * pointer to error structure.
4001 * 0 on success otherwise -errno and errno is set.
4004 flow_dv_push_vlan_action_resource_register
4005 (struct rte_eth_dev *dev,
4006 struct mlx5_flow_dv_push_vlan_action_resource *ref,
4007 struct mlx5_flow *dev_flow,
4008 struct rte_flow_error *error)
4010 struct mlx5_priv *priv = dev->data->dev_private;
4011 struct mlx5_flow_dv_push_vlan_action_resource *resource;
4012 struct mlx5_list_entry *entry;
4013 struct mlx5_flow_cb_ctx ctx = {
4018 entry = mlx5_list_register(priv->sh->push_vlan_action_list, &ctx);
4021 resource = container_of(entry, typeof(*resource), entry);
4023 dev_flow->handle->dvh.rix_push_vlan = resource->idx;
4024 dev_flow->dv.push_vlan_res = resource;
4029 * Get the size of specific rte_flow_item_type hdr size
4031 * @param[in] item_type
4032 * Tested rte_flow_item_type.
4035 * sizeof struct item_type, 0 if void or irrelevant.
4038 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
4042 switch (item_type) {
4043 case RTE_FLOW_ITEM_TYPE_ETH:
4044 retval = sizeof(struct rte_ether_hdr);
4046 case RTE_FLOW_ITEM_TYPE_VLAN:
4047 retval = sizeof(struct rte_vlan_hdr);
4049 case RTE_FLOW_ITEM_TYPE_IPV4:
4050 retval = sizeof(struct rte_ipv4_hdr);
4052 case RTE_FLOW_ITEM_TYPE_IPV6:
4053 retval = sizeof(struct rte_ipv6_hdr);
4055 case RTE_FLOW_ITEM_TYPE_UDP:
4056 retval = sizeof(struct rte_udp_hdr);
4058 case RTE_FLOW_ITEM_TYPE_TCP:
4059 retval = sizeof(struct rte_tcp_hdr);
4061 case RTE_FLOW_ITEM_TYPE_VXLAN:
4062 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
4063 retval = sizeof(struct rte_vxlan_hdr);
4065 case RTE_FLOW_ITEM_TYPE_GRE:
4066 case RTE_FLOW_ITEM_TYPE_NVGRE:
4067 retval = sizeof(struct rte_gre_hdr);
4069 case RTE_FLOW_ITEM_TYPE_MPLS:
4070 retval = sizeof(struct rte_mpls_hdr);
4072 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
4080 #define MLX5_ENCAP_IPV4_VERSION 0x40
4081 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
4082 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
4083 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
4084 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
4085 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
4086 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
4089 * Convert the encap action data from list of rte_flow_item to raw buffer
4092 * Pointer to rte_flow_item objects list.
4094 * Pointer to the output buffer.
4096 * Pointer to the output buffer size.
4098 * Pointer to the error structure.
4101 * 0 on success, a negative errno value otherwise and rte_errno is set.
4104 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
4105 size_t *size, struct rte_flow_error *error)
4107 struct rte_ether_hdr *eth = NULL;
4108 struct rte_vlan_hdr *vlan = NULL;
4109 struct rte_ipv4_hdr *ipv4 = NULL;
4110 struct rte_ipv6_hdr *ipv6 = NULL;
4111 struct rte_udp_hdr *udp = NULL;
4112 struct rte_vxlan_hdr *vxlan = NULL;
4113 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
4114 struct rte_gre_hdr *gre = NULL;
4116 size_t temp_size = 0;
4119 return rte_flow_error_set(error, EINVAL,
4120 RTE_FLOW_ERROR_TYPE_ACTION,
4121 NULL, "invalid empty data");
4122 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4123 len = flow_dv_get_item_hdr_len(items->type);
4124 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
4125 return rte_flow_error_set(error, EINVAL,
4126 RTE_FLOW_ERROR_TYPE_ACTION,
4127 (void *)items->type,
4128 "items total size is too big"
4129 " for encap action");
4130 rte_memcpy((void *)&buf[temp_size], items->spec, len);
4131 switch (items->type) {
4132 case RTE_FLOW_ITEM_TYPE_ETH:
4133 eth = (struct rte_ether_hdr *)&buf[temp_size];
4135 case RTE_FLOW_ITEM_TYPE_VLAN:
4136 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
4138 return rte_flow_error_set(error, EINVAL,
4139 RTE_FLOW_ERROR_TYPE_ACTION,
4140 (void *)items->type,
4141 "eth header not found");
4142 if (!eth->ether_type)
4143 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
4145 case RTE_FLOW_ITEM_TYPE_IPV4:
4146 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
4148 return rte_flow_error_set(error, EINVAL,
4149 RTE_FLOW_ERROR_TYPE_ACTION,
4150 (void *)items->type,
4151 "neither eth nor vlan"
4153 if (vlan && !vlan->eth_proto)
4154 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
4155 else if (eth && !eth->ether_type)
4156 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
4157 if (!ipv4->version_ihl)
4158 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
4159 MLX5_ENCAP_IPV4_IHL_MIN;
4160 if (!ipv4->time_to_live)
4161 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
4163 case RTE_FLOW_ITEM_TYPE_IPV6:
4164 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
4166 return rte_flow_error_set(error, EINVAL,
4167 RTE_FLOW_ERROR_TYPE_ACTION,
4168 (void *)items->type,
4169 "neither eth nor vlan"
4171 if (vlan && !vlan->eth_proto)
4172 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
4173 else if (eth && !eth->ether_type)
4174 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
4175 if (!ipv6->vtc_flow)
4177 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
4178 if (!ipv6->hop_limits)
4179 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
4181 case RTE_FLOW_ITEM_TYPE_UDP:
4182 udp = (struct rte_udp_hdr *)&buf[temp_size];
4184 return rte_flow_error_set(error, EINVAL,
4185 RTE_FLOW_ERROR_TYPE_ACTION,
4186 (void *)items->type,
4187 "ip header not found");
4188 if (ipv4 && !ipv4->next_proto_id)
4189 ipv4->next_proto_id = IPPROTO_UDP;
4190 else if (ipv6 && !ipv6->proto)
4191 ipv6->proto = IPPROTO_UDP;
4193 case RTE_FLOW_ITEM_TYPE_VXLAN:
4194 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
4196 return rte_flow_error_set(error, EINVAL,
4197 RTE_FLOW_ERROR_TYPE_ACTION,
4198 (void *)items->type,
4199 "udp header not found");
4201 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
4202 if (!vxlan->vx_flags)
4204 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
4206 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
4207 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
4209 return rte_flow_error_set(error, EINVAL,
4210 RTE_FLOW_ERROR_TYPE_ACTION,
4211 (void *)items->type,
4212 "udp header not found");
4213 if (!vxlan_gpe->proto)
4214 return rte_flow_error_set(error, EINVAL,
4215 RTE_FLOW_ERROR_TYPE_ACTION,
4216 (void *)items->type,
4217 "next protocol not found");
4220 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
4221 if (!vxlan_gpe->vx_flags)
4222 vxlan_gpe->vx_flags =
4223 MLX5_ENCAP_VXLAN_GPE_FLAGS;
4225 case RTE_FLOW_ITEM_TYPE_GRE:
4226 case RTE_FLOW_ITEM_TYPE_NVGRE:
4227 gre = (struct rte_gre_hdr *)&buf[temp_size];
4229 return rte_flow_error_set(error, EINVAL,
4230 RTE_FLOW_ERROR_TYPE_ACTION,
4231 (void *)items->type,
4232 "next protocol not found");
4234 return rte_flow_error_set(error, EINVAL,
4235 RTE_FLOW_ERROR_TYPE_ACTION,
4236 (void *)items->type,
4237 "ip header not found");
4238 if (ipv4 && !ipv4->next_proto_id)
4239 ipv4->next_proto_id = IPPROTO_GRE;
4240 else if (ipv6 && !ipv6->proto)
4241 ipv6->proto = IPPROTO_GRE;
4243 case RTE_FLOW_ITEM_TYPE_VOID:
4246 return rte_flow_error_set(error, EINVAL,
4247 RTE_FLOW_ERROR_TYPE_ACTION,
4248 (void *)items->type,
4249 "unsupported item type");
4259 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
4261 struct rte_ether_hdr *eth = NULL;
4262 struct rte_vlan_hdr *vlan = NULL;
4263 struct rte_ipv6_hdr *ipv6 = NULL;
4264 struct rte_udp_hdr *udp = NULL;
4268 eth = (struct rte_ether_hdr *)data;
4269 next_hdr = (char *)(eth + 1);
4270 proto = RTE_BE16(eth->ether_type);
4273 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
4274 vlan = (struct rte_vlan_hdr *)next_hdr;
4275 proto = RTE_BE16(vlan->eth_proto);
4276 next_hdr += sizeof(struct rte_vlan_hdr);
4279 /* HW calculates IPv4 csum. no need to proceed */
4280 if (proto == RTE_ETHER_TYPE_IPV4)
4283 /* non IPv4/IPv6 header. not supported */
4284 if (proto != RTE_ETHER_TYPE_IPV6) {
4285 return rte_flow_error_set(error, ENOTSUP,
4286 RTE_FLOW_ERROR_TYPE_ACTION,
4287 NULL, "Cannot offload non IPv4/IPv6");
4290 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
4292 /* ignore non UDP */
4293 if (ipv6->proto != IPPROTO_UDP)
4296 udp = (struct rte_udp_hdr *)(ipv6 + 1);
4297 udp->dgram_cksum = 0;
4303 * Convert L2 encap action to DV specification.
4306 * Pointer to rte_eth_dev structure.
4308 * Pointer to action structure.
4309 * @param[in, out] dev_flow
4310 * Pointer to the mlx5_flow.
4311 * @param[in] transfer
4312 * Mark if the flow is E-Switch flow.
4314 * Pointer to the error structure.
4317 * 0 on success, a negative errno value otherwise and rte_errno is set.
4320 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
4321 const struct rte_flow_action *action,
4322 struct mlx5_flow *dev_flow,
4324 struct rte_flow_error *error)
4326 const struct rte_flow_item *encap_data;
4327 const struct rte_flow_action_raw_encap *raw_encap_data;
4328 struct mlx5_flow_dv_encap_decap_resource res = {
4330 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
4331 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
4332 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
4335 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
4337 (const struct rte_flow_action_raw_encap *)action->conf;
4338 res.size = raw_encap_data->size;
4339 memcpy(res.buf, raw_encap_data->data, res.size);
4341 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
4343 ((const struct rte_flow_action_vxlan_encap *)
4344 action->conf)->definition;
4347 ((const struct rte_flow_action_nvgre_encap *)
4348 action->conf)->definition;
4349 if (flow_dv_convert_encap_data(encap_data, res.buf,
4353 if (flow_dv_zero_encap_udp_csum(res.buf, error))
4355 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
4356 return rte_flow_error_set(error, EINVAL,
4357 RTE_FLOW_ERROR_TYPE_ACTION,
4358 NULL, "can't create L2 encap action");
4363 * Convert L2 decap action to DV specification.
4366 * Pointer to rte_eth_dev structure.
4367 * @param[in, out] dev_flow
4368 * Pointer to the mlx5_flow.
4369 * @param[in] transfer
4370 * Mark if the flow is E-Switch flow.
4372 * Pointer to the error structure.
4375 * 0 on success, a negative errno value otherwise and rte_errno is set.
4378 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
4379 struct mlx5_flow *dev_flow,
4381 struct rte_flow_error *error)
4383 struct mlx5_flow_dv_encap_decap_resource res = {
4386 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
4387 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
4388 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
4391 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
4392 return rte_flow_error_set(error, EINVAL,
4393 RTE_FLOW_ERROR_TYPE_ACTION,
4394 NULL, "can't create L2 decap action");
4399 * Convert raw decap/encap (L3 tunnel) action to DV specification.
4402 * Pointer to rte_eth_dev structure.
4404 * Pointer to action structure.
4405 * @param[in, out] dev_flow
4406 * Pointer to the mlx5_flow.
4408 * Pointer to the flow attributes.
4410 * Pointer to the error structure.
4413 * 0 on success, a negative errno value otherwise and rte_errno is set.
4416 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
4417 const struct rte_flow_action *action,
4418 struct mlx5_flow *dev_flow,
4419 const struct rte_flow_attr *attr,
4420 struct rte_flow_error *error)
4422 const struct rte_flow_action_raw_encap *encap_data;
4423 struct mlx5_flow_dv_encap_decap_resource res;
4425 memset(&res, 0, sizeof(res));
4426 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
4427 res.size = encap_data->size;
4428 memcpy(res.buf, encap_data->data, res.size);
4429 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
4430 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
4431 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
4433 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
4435 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
4436 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
4437 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
4438 return rte_flow_error_set(error, EINVAL,
4439 RTE_FLOW_ERROR_TYPE_ACTION,
4440 NULL, "can't create encap action");
4445 * Create action push VLAN.
4448 * Pointer to rte_eth_dev structure.
4450 * Pointer to the flow attributes.
4452 * Pointer to the vlan to push to the Ethernet header.
4453 * @param[in, out] dev_flow
4454 * Pointer to the mlx5_flow.
4456 * Pointer to the error structure.
4459 * 0 on success, a negative errno value otherwise and rte_errno is set.
4462 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
4463 const struct rte_flow_attr *attr,
4464 const struct rte_vlan_hdr *vlan,
4465 struct mlx5_flow *dev_flow,
4466 struct rte_flow_error *error)
4468 struct mlx5_flow_dv_push_vlan_action_resource res;
4470 memset(&res, 0, sizeof(res));
4472 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
4475 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
4477 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
4478 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
4479 return flow_dv_push_vlan_action_resource_register
4480 (dev, &res, dev_flow, error);
4484 * Validate the modify-header actions.
4486 * @param[in] action_flags
4487 * Holds the actions detected until now.
4489 * Pointer to the modify action.
4491 * Pointer to error structure.
4494 * 0 on success, a negative errno value otherwise and rte_errno is set.
4497 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
4498 const struct rte_flow_action *action,
4499 struct rte_flow_error *error)
4501 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
4502 return rte_flow_error_set(error, EINVAL,
4503 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4504 NULL, "action configuration not set");
4505 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
4506 return rte_flow_error_set(error, EINVAL,
4507 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4508 "can't have encap action before"
4514 * Validate the modify-header MAC address actions.
4516 * @param[in] action_flags
4517 * Holds the actions detected until now.
4519 * Pointer to the modify action.
4520 * @param[in] item_flags
4521 * Holds the items detected.
4523 * Pointer to error structure.
4526 * 0 on success, a negative errno value otherwise and rte_errno is set.
4529 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
4530 const struct rte_flow_action *action,
4531 const uint64_t item_flags,
4532 struct rte_flow_error *error)
4536 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4538 if (!(item_flags & MLX5_FLOW_LAYER_L2))
4539 return rte_flow_error_set(error, EINVAL,
4540 RTE_FLOW_ERROR_TYPE_ACTION,
4542 "no L2 item in pattern");
4548 * Validate the modify-header IPv4 address actions.
4550 * @param[in] action_flags
4551 * Holds the actions detected until now.
4553 * Pointer to the modify action.
4554 * @param[in] item_flags
4555 * Holds the items detected.
4557 * Pointer to error structure.
4560 * 0 on success, a negative errno value otherwise and rte_errno is set.
4563 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
4564 const struct rte_flow_action *action,
4565 const uint64_t item_flags,
4566 struct rte_flow_error *error)
4571 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4573 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4574 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
4575 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4576 if (!(item_flags & layer))
4577 return rte_flow_error_set(error, EINVAL,
4578 RTE_FLOW_ERROR_TYPE_ACTION,
4580 "no ipv4 item in pattern");
4586 * Validate the modify-header IPv6 address actions.
4588 * @param[in] action_flags
4589 * Holds the actions detected until now.
4591 * Pointer to the modify action.
4592 * @param[in] item_flags
4593 * Holds the items detected.
4595 * Pointer to error structure.
4598 * 0 on success, a negative errno value otherwise and rte_errno is set.
4601 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
4602 const struct rte_flow_action *action,
4603 const uint64_t item_flags,
4604 struct rte_flow_error *error)
4609 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4611 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4612 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
4613 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4614 if (!(item_flags & layer))
4615 return rte_flow_error_set(error, EINVAL,
4616 RTE_FLOW_ERROR_TYPE_ACTION,
4618 "no ipv6 item in pattern");
4624 * Validate the modify-header TP actions.
4626 * @param[in] action_flags
4627 * Holds the actions detected until now.
4629 * Pointer to the modify action.
4630 * @param[in] item_flags
4631 * Holds the items detected.
4633 * Pointer to error structure.
4636 * 0 on success, a negative errno value otherwise and rte_errno is set.
4639 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
4640 const struct rte_flow_action *action,
4641 const uint64_t item_flags,
4642 struct rte_flow_error *error)
4647 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4649 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4650 MLX5_FLOW_LAYER_INNER_L4 :
4651 MLX5_FLOW_LAYER_OUTER_L4;
4652 if (!(item_flags & layer))
4653 return rte_flow_error_set(error, EINVAL,
4654 RTE_FLOW_ERROR_TYPE_ACTION,
4655 NULL, "no transport layer "
4662 * Validate the modify-header actions of increment/decrement
4663 * TCP Sequence-number.
4665 * @param[in] action_flags
4666 * Holds the actions detected until now.
4668 * Pointer to the modify action.
4669 * @param[in] item_flags
4670 * Holds the items detected.
4672 * Pointer to error structure.
4675 * 0 on success, a negative errno value otherwise and rte_errno is set.
4678 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
4679 const struct rte_flow_action *action,
4680 const uint64_t item_flags,
4681 struct rte_flow_error *error)
4686 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4688 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4689 MLX5_FLOW_LAYER_INNER_L4_TCP :
4690 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4691 if (!(item_flags & layer))
4692 return rte_flow_error_set(error, EINVAL,
4693 RTE_FLOW_ERROR_TYPE_ACTION,
4694 NULL, "no TCP item in"
4696 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
4697 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
4698 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
4699 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
4700 return rte_flow_error_set(error, EINVAL,
4701 RTE_FLOW_ERROR_TYPE_ACTION,
4703 "cannot decrease and increase"
4704 " TCP sequence number"
4705 " at the same time");
4711 * Validate the modify-header actions of increment/decrement
4712 * TCP Acknowledgment number.
4714 * @param[in] action_flags
4715 * Holds the actions detected until now.
4717 * Pointer to the modify action.
4718 * @param[in] item_flags
4719 * Holds the items detected.
4721 * Pointer to error structure.
4724 * 0 on success, a negative errno value otherwise and rte_errno is set.
4727 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
4728 const struct rte_flow_action *action,
4729 const uint64_t item_flags,
4730 struct rte_flow_error *error)
4735 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4737 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4738 MLX5_FLOW_LAYER_INNER_L4_TCP :
4739 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4740 if (!(item_flags & layer))
4741 return rte_flow_error_set(error, EINVAL,
4742 RTE_FLOW_ERROR_TYPE_ACTION,
4743 NULL, "no TCP item in"
4745 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
4746 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
4747 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
4748 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
4749 return rte_flow_error_set(error, EINVAL,
4750 RTE_FLOW_ERROR_TYPE_ACTION,
4752 "cannot decrease and increase"
4753 " TCP acknowledgment number"
4754 " at the same time");
4760 * Validate the modify-header TTL actions.
4762 * @param[in] action_flags
4763 * Holds the actions detected until now.
4765 * Pointer to the modify action.
4766 * @param[in] item_flags
4767 * Holds the items detected.
4769 * Pointer to error structure.
4772 * 0 on success, a negative errno value otherwise and rte_errno is set.
4775 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
4776 const struct rte_flow_action *action,
4777 const uint64_t item_flags,
4778 struct rte_flow_error *error)
4783 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4785 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4786 MLX5_FLOW_LAYER_INNER_L3 :
4787 MLX5_FLOW_LAYER_OUTER_L3;
4788 if (!(item_flags & layer))
4789 return rte_flow_error_set(error, EINVAL,
4790 RTE_FLOW_ERROR_TYPE_ACTION,
4792 "no IP protocol in pattern");
4798 * Validate the generic modify field actions.
4800 * Pointer to the rte_eth_dev structure.
4801 * @param[in] action_flags
4802 * Holds the actions detected until now.
4804 * Pointer to the modify action.
4806 * Pointer to the flow attributes.
4808 * Pointer to error structure.
4811 * Number of header fields to modify (0 or more) on success,
4812 * a negative errno value otherwise and rte_errno is set.
4815 flow_dv_validate_action_modify_field(struct rte_eth_dev *dev,
4816 const uint64_t action_flags,
4817 const struct rte_flow_action *action,
4818 const struct rte_flow_attr *attr,
4819 struct rte_flow_error *error)
4822 struct mlx5_priv *priv = dev->data->dev_private;
4823 struct mlx5_sh_config *config = &priv->sh->config;
4824 const struct rte_flow_action_modify_field *action_modify_field =
4826 uint32_t dst_width = mlx5_flow_item_field_width(dev,
4827 action_modify_field->dst.field,
4829 uint32_t src_width = mlx5_flow_item_field_width(dev,
4830 action_modify_field->src.field,
4831 dst_width, attr, error);
4833 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4837 if (action_modify_field->width == 0)
4838 return rte_flow_error_set(error, EINVAL,
4839 RTE_FLOW_ERROR_TYPE_ACTION, action,
4840 "no bits are requested to be modified");
4841 else if (action_modify_field->width > dst_width ||
4842 action_modify_field->width > src_width)
4843 return rte_flow_error_set(error, EINVAL,
4844 RTE_FLOW_ERROR_TYPE_ACTION, action,
4845 "cannot modify more bits than"
4846 " the width of a field");
4847 if (action_modify_field->dst.field != RTE_FLOW_FIELD_VALUE &&
4848 action_modify_field->dst.field != RTE_FLOW_FIELD_POINTER) {
4849 if ((action_modify_field->dst.offset +
4850 action_modify_field->width > dst_width) ||
4851 (action_modify_field->dst.offset % 32))
4852 return rte_flow_error_set(error, EINVAL,
4853 RTE_FLOW_ERROR_TYPE_ACTION, action,
4854 "destination offset is too big"
4855 " or not aligned to 4 bytes");
4856 if (action_modify_field->dst.level &&
4857 action_modify_field->dst.field != RTE_FLOW_FIELD_TAG)
4858 return rte_flow_error_set(error, ENOTSUP,
4859 RTE_FLOW_ERROR_TYPE_ACTION, action,
4860 "inner header fields modification"
4861 " is not supported");
4863 if (action_modify_field->src.field != RTE_FLOW_FIELD_VALUE &&
4864 action_modify_field->src.field != RTE_FLOW_FIELD_POINTER) {
4865 if (!attr->transfer && !attr->group)
4866 return rte_flow_error_set(error, ENOTSUP,
4867 RTE_FLOW_ERROR_TYPE_ACTION, action,
4868 "modify field action is not"
4869 " supported for group 0");
4870 if ((action_modify_field->src.offset +
4871 action_modify_field->width > src_width) ||
4872 (action_modify_field->src.offset % 32))
4873 return rte_flow_error_set(error, EINVAL,
4874 RTE_FLOW_ERROR_TYPE_ACTION, action,
4875 "source offset is too big"
4876 " or not aligned to 4 bytes");
4877 if (action_modify_field->src.level &&
4878 action_modify_field->src.field != RTE_FLOW_FIELD_TAG)
4879 return rte_flow_error_set(error, ENOTSUP,
4880 RTE_FLOW_ERROR_TYPE_ACTION, action,
4881 "inner header fields modification"
4882 " is not supported");
4884 if ((action_modify_field->dst.field ==
4885 action_modify_field->src.field) &&
4886 (action_modify_field->dst.level ==
4887 action_modify_field->src.level))
4888 return rte_flow_error_set(error, EINVAL,
4889 RTE_FLOW_ERROR_TYPE_ACTION, action,
4890 "source and destination fields"
4891 " cannot be the same");
4892 if (action_modify_field->dst.field == RTE_FLOW_FIELD_VALUE ||
4893 action_modify_field->dst.field == RTE_FLOW_FIELD_POINTER ||
4894 action_modify_field->dst.field == RTE_FLOW_FIELD_MARK)
4895 return rte_flow_error_set(error, EINVAL,
4896 RTE_FLOW_ERROR_TYPE_ACTION, action,
4897 "mark, immediate value or a pointer to it"
4898 " cannot be used as a destination");
4899 if (action_modify_field->dst.field == RTE_FLOW_FIELD_START ||
4900 action_modify_field->src.field == RTE_FLOW_FIELD_START)
4901 return rte_flow_error_set(error, ENOTSUP,
4902 RTE_FLOW_ERROR_TYPE_ACTION, action,
4903 "modifications of an arbitrary"
4904 " place in a packet is not supported");
4905 if (action_modify_field->dst.field == RTE_FLOW_FIELD_VLAN_TYPE ||
4906 action_modify_field->src.field == RTE_FLOW_FIELD_VLAN_TYPE)
4907 return rte_flow_error_set(error, ENOTSUP,
4908 RTE_FLOW_ERROR_TYPE_ACTION, action,
4909 "modifications of the 802.1Q Tag"
4910 " Identifier is not supported");
4911 if (action_modify_field->dst.field == RTE_FLOW_FIELD_VXLAN_VNI ||
4912 action_modify_field->src.field == RTE_FLOW_FIELD_VXLAN_VNI)
4913 return rte_flow_error_set(error, ENOTSUP,
4914 RTE_FLOW_ERROR_TYPE_ACTION, action,
4915 "modifications of the VXLAN Network"
4916 " Identifier is not supported");
4917 if (action_modify_field->dst.field == RTE_FLOW_FIELD_GENEVE_VNI ||
4918 action_modify_field->src.field == RTE_FLOW_FIELD_GENEVE_VNI)
4919 return rte_flow_error_set(error, ENOTSUP,
4920 RTE_FLOW_ERROR_TYPE_ACTION, action,
4921 "modifications of the GENEVE Network"
4922 " Identifier is not supported");
4923 if (action_modify_field->dst.field == RTE_FLOW_FIELD_MARK ||
4924 action_modify_field->src.field == RTE_FLOW_FIELD_MARK)
4925 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4926 !mlx5_flow_ext_mreg_supported(dev))
4927 return rte_flow_error_set(error, ENOTSUP,
4928 RTE_FLOW_ERROR_TYPE_ACTION, action,
4929 "cannot modify mark in legacy mode"
4930 " or without extensive registers");
4931 if (action_modify_field->dst.field == RTE_FLOW_FIELD_META ||
4932 action_modify_field->src.field == RTE_FLOW_FIELD_META) {
4933 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
4934 !mlx5_flow_ext_mreg_supported(dev))
4935 return rte_flow_error_set(error, ENOTSUP,
4936 RTE_FLOW_ERROR_TYPE_ACTION, action,
4937 "cannot modify meta without"
4938 " extensive registers support");
4939 ret = flow_dv_get_metadata_reg(dev, attr, error);
4940 if (ret < 0 || ret == REG_NON)
4941 return rte_flow_error_set(error, ENOTSUP,
4942 RTE_FLOW_ERROR_TYPE_ACTION, action,
4943 "cannot modify meta without"
4944 " extensive registers available");
4946 if (action_modify_field->operation != RTE_FLOW_MODIFY_SET)
4947 return rte_flow_error_set(error, ENOTSUP,
4948 RTE_FLOW_ERROR_TYPE_ACTION, action,
4949 "add and sub operations"
4950 " are not supported");
4951 return (action_modify_field->width / 32) +
4952 !!(action_modify_field->width % 32);
4956 * Validate jump action.
4959 * Pointer to the jump action.
4960 * @param[in] action_flags
4961 * Holds the actions detected until now.
4962 * @param[in] attributes
4963 * Pointer to flow attributes
4964 * @param[in] external
4965 * Action belongs to flow rule created by request external to PMD.
4967 * Pointer to error structure.
4970 * 0 on success, a negative errno value otherwise and rte_errno is set.
4973 flow_dv_validate_action_jump(struct rte_eth_dev *dev,
4974 const struct mlx5_flow_tunnel *tunnel,
4975 const struct rte_flow_action *action,
4976 uint64_t action_flags,
4977 const struct rte_flow_attr *attributes,
4978 bool external, struct rte_flow_error *error)
4980 uint32_t target_group, table = 0;
4982 struct flow_grp_info grp_info = {
4983 .external = !!external,
4984 .transfer = !!attributes->transfer,
4988 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
4989 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
4990 return rte_flow_error_set(error, EINVAL,
4991 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4992 "can't have 2 fate actions in"
4995 return rte_flow_error_set(error, EINVAL,
4996 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4997 NULL, "action configuration not set");
4999 ((const struct rte_flow_action_jump *)action->conf)->group;
5000 ret = mlx5_flow_group_to_table(dev, tunnel, target_group, &table,
5004 if (attributes->group == target_group &&
5005 !(action_flags & (MLX5_FLOW_ACTION_TUNNEL_SET |
5006 MLX5_FLOW_ACTION_TUNNEL_MATCH)))
5007 return rte_flow_error_set(error, EINVAL,
5008 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5009 "target group must be other than"
5010 " the current flow group");
5012 return rte_flow_error_set(error, EINVAL,
5013 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
5014 NULL, "root table shouldn't be destination");
5019 * Validate action PORT_ID / REPRESENTED_PORT.
5022 * Pointer to rte_eth_dev structure.
5023 * @param[in] action_flags
5024 * Bit-fields that holds the actions detected until now.
5026 * PORT_ID / REPRESENTED_PORT action structure.
5028 * Attributes of flow that includes this action.
5030 * Pointer to error structure.
5033 * 0 on success, a negative errno value otherwise and rte_errno is set.
5036 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
5037 uint64_t action_flags,
5038 const struct rte_flow_action *action,
5039 const struct rte_flow_attr *attr,
5040 struct rte_flow_error *error)
5042 const struct rte_flow_action_port_id *port_id;
5043 const struct rte_flow_action_ethdev *ethdev;
5044 struct mlx5_priv *act_priv;
5045 struct mlx5_priv *dev_priv;
5048 if (!attr->transfer)
5049 return rte_flow_error_set(error, ENOTSUP,
5050 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5052 "port action is valid in transfer"
5054 if (!action || !action->conf)
5055 return rte_flow_error_set(error, ENOTSUP,
5056 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
5058 "port action parameters must be"
5060 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
5061 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
5062 return rte_flow_error_set(error, EINVAL,
5063 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5064 "can have only one fate actions in"
5066 dev_priv = mlx5_dev_to_eswitch_info(dev);
5068 return rte_flow_error_set(error, rte_errno,
5069 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5071 "failed to obtain E-Switch info");
5072 switch (action->type) {
5073 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5074 port_id = action->conf;
5075 port = port_id->original ? dev->data->port_id : port_id->id;
5077 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
5078 ethdev = action->conf;
5079 port = ethdev->port_id;
5083 return rte_flow_error_set
5085 RTE_FLOW_ERROR_TYPE_ACTION, action,
5086 "unknown E-Switch action");
5088 act_priv = mlx5_port_to_eswitch_info(port, false);
5090 return rte_flow_error_set
5092 RTE_FLOW_ERROR_TYPE_ACTION_CONF, action->conf,
5093 "failed to obtain E-Switch port id for port");
5094 if (act_priv->domain_id != dev_priv->domain_id)
5095 return rte_flow_error_set
5097 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5098 "port does not belong to"
5099 " E-Switch being configured");
5104 * Get the maximum number of modify header actions.
5107 * Pointer to rte_eth_dev structure.
5109 * Whether action is on root table.
5112 * Max number of modify header actions device can support.
5114 static inline unsigned int
5115 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
5119 * There's no way to directly query the max capacity from FW.
5120 * The maximal value on root table should be assumed to be supported.
5123 return MLX5_MAX_MODIFY_NUM;
5125 return MLX5_ROOT_TBL_MODIFY_NUM;
5129 * Validate the meter action.
5132 * Pointer to rte_eth_dev structure.
5133 * @param[in] action_flags
5134 * Bit-fields that holds the actions detected until now.
5135 * @param[in] item_flags
5136 * Holds the items detected.
5138 * Pointer to the meter action.
5140 * Attributes of flow that includes this action.
5141 * @param[in] port_id_item
5142 * Pointer to item indicating port id.
5144 * Pointer to error structure.
5147 * 0 on success, a negative errno value otherwise and rte_errno is set.
5150 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
5151 uint64_t action_flags, uint64_t item_flags,
5152 const struct rte_flow_action *action,
5153 const struct rte_flow_attr *attr,
5154 const struct rte_flow_item *port_id_item,
5156 struct rte_flow_error *error)
5158 struct mlx5_priv *priv = dev->data->dev_private;
5159 const struct rte_flow_action_meter *am = action->conf;
5160 struct mlx5_flow_meter_info *fm;
5161 struct mlx5_flow_meter_policy *mtr_policy;
5162 struct mlx5_flow_mtr_mng *mtrmng = priv->sh->mtrmng;
5165 return rte_flow_error_set(error, EINVAL,
5166 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5167 "meter action conf is NULL");
5169 if (action_flags & MLX5_FLOW_ACTION_METER)
5170 return rte_flow_error_set(error, ENOTSUP,
5171 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5172 "meter chaining not support");
5173 if (action_flags & MLX5_FLOW_ACTION_JUMP)
5174 return rte_flow_error_set(error, ENOTSUP,
5175 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5176 "meter with jump not support");
5178 return rte_flow_error_set(error, ENOTSUP,
5179 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5181 "meter action not supported");
5182 fm = mlx5_flow_meter_find(priv, am->mtr_id, NULL);
5184 return rte_flow_error_set(error, EINVAL,
5185 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5187 /* aso meter can always be shared by different domains */
5188 if (fm->ref_cnt && !priv->sh->meter_aso_en &&
5189 !(fm->transfer == attr->transfer ||
5190 (!fm->ingress && !attr->ingress && attr->egress) ||
5191 (!fm->egress && !attr->egress && attr->ingress)))
5192 return rte_flow_error_set(error, EINVAL,
5193 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5194 "Flow attributes domain are either invalid "
5195 "or have a domain conflict with current "
5196 "meter attributes");
5197 if (fm->def_policy) {
5198 if (!((attr->transfer &&
5199 mtrmng->def_policy[MLX5_MTR_DOMAIN_TRANSFER]) ||
5201 mtrmng->def_policy[MLX5_MTR_DOMAIN_EGRESS]) ||
5203 mtrmng->def_policy[MLX5_MTR_DOMAIN_INGRESS])))
5204 return rte_flow_error_set(error, EINVAL,
5205 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5206 "Flow attributes domain "
5207 "have a conflict with current "
5208 "meter domain attributes");
5211 mtr_policy = mlx5_flow_meter_policy_find(dev,
5212 fm->policy_id, NULL);
5214 return rte_flow_error_set(error, EINVAL,
5215 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5216 "Invalid policy id for meter ");
5217 if (!((attr->transfer && mtr_policy->transfer) ||
5218 (attr->egress && mtr_policy->egress) ||
5219 (attr->ingress && mtr_policy->ingress)))
5220 return rte_flow_error_set(error, EINVAL,
5221 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5222 "Flow attributes domain "
5223 "have a conflict with current "
5224 "meter domain attributes");
5225 if (attr->transfer && mtr_policy->dev) {
5227 * When policy has fate action of port_id,
5228 * the flow should have the same src port as policy.
5230 struct mlx5_priv *policy_port_priv =
5231 mtr_policy->dev->data->dev_private;
5232 int32_t flow_src_port = priv->representor_id;
5235 const struct rte_flow_item_port_id *spec =
5237 struct mlx5_priv *port_priv =
5238 mlx5_port_to_eswitch_info(spec->id,
5241 return rte_flow_error_set(error,
5243 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5245 "Failed to get port info.");
5246 flow_src_port = port_priv->representor_id;
5248 if (flow_src_port != policy_port_priv->representor_id)
5249 return rte_flow_error_set(error,
5251 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5253 "Flow and meter policy "
5254 "have different src port.");
5255 } else if (mtr_policy->is_rss) {
5256 struct mlx5_flow_meter_policy *fp;
5257 struct mlx5_meter_policy_action_container *acg;
5258 struct mlx5_meter_policy_action_container *acy;
5259 const struct rte_flow_action *rss_act;
5262 fp = mlx5_flow_meter_hierarchy_get_final_policy(dev,
5265 return rte_flow_error_set(error, EINVAL,
5266 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5267 "Unable to get the final "
5268 "policy in the hierarchy");
5269 acg = &fp->act_cnt[RTE_COLOR_GREEN];
5270 acy = &fp->act_cnt[RTE_COLOR_YELLOW];
5271 MLX5_ASSERT(acg->fate_action ==
5272 MLX5_FLOW_FATE_SHARED_RSS ||
5274 MLX5_FLOW_FATE_SHARED_RSS);
5275 if (acg->fate_action == MLX5_FLOW_FATE_SHARED_RSS)
5279 ret = mlx5_flow_validate_action_rss(rss_act,
5280 action_flags, dev, attr,
5285 *def_policy = false;
5291 * Validate the age action.
5293 * @param[in] action_flags
5294 * Holds the actions detected until now.
5296 * Pointer to the age action.
5298 * Pointer to the Ethernet device structure.
5300 * Pointer to error structure.
5303 * 0 on success, a negative errno value otherwise and rte_errno is set.
5306 flow_dv_validate_action_age(uint64_t action_flags,
5307 const struct rte_flow_action *action,
5308 struct rte_eth_dev *dev,
5309 struct rte_flow_error *error)
5311 struct mlx5_priv *priv = dev->data->dev_private;
5312 const struct rte_flow_action_age *age = action->conf;
5314 if (!priv->sh->cdev->config.devx ||
5315 (priv->sh->cmng.counter_fallback && !priv->sh->aso_age_mng))
5316 return rte_flow_error_set(error, ENOTSUP,
5317 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5319 "age action not supported");
5320 if (!(action->conf))
5321 return rte_flow_error_set(error, EINVAL,
5322 RTE_FLOW_ERROR_TYPE_ACTION, action,
5323 "configuration cannot be null");
5324 if (!(age->timeout))
5325 return rte_flow_error_set(error, EINVAL,
5326 RTE_FLOW_ERROR_TYPE_ACTION, action,
5327 "invalid timeout value 0");
5328 if (action_flags & MLX5_FLOW_ACTION_AGE)
5329 return rte_flow_error_set(error, EINVAL,
5330 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5331 "duplicate age actions set");
5336 * Validate the modify-header IPv4 DSCP actions.
5338 * @param[in] action_flags
5339 * Holds the actions detected until now.
5341 * Pointer to the modify action.
5342 * @param[in] item_flags
5343 * Holds the items detected.
5345 * Pointer to error structure.
5348 * 0 on success, a negative errno value otherwise and rte_errno is set.
5351 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
5352 const struct rte_flow_action *action,
5353 const uint64_t item_flags,
5354 struct rte_flow_error *error)
5358 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
5360 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
5361 return rte_flow_error_set(error, EINVAL,
5362 RTE_FLOW_ERROR_TYPE_ACTION,
5364 "no ipv4 item in pattern");
5370 * Validate the modify-header IPv6 DSCP actions.
5372 * @param[in] action_flags
5373 * Holds the actions detected until now.
5375 * Pointer to the modify action.
5376 * @param[in] item_flags
5377 * Holds the items detected.
5379 * Pointer to error structure.
5382 * 0 on success, a negative errno value otherwise and rte_errno is set.
5385 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
5386 const struct rte_flow_action *action,
5387 const uint64_t item_flags,
5388 struct rte_flow_error *error)
5392 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
5394 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
5395 return rte_flow_error_set(error, EINVAL,
5396 RTE_FLOW_ERROR_TYPE_ACTION,
5398 "no ipv6 item in pattern");
5404 flow_dv_modify_match_cb(void *tool_ctx __rte_unused,
5405 struct mlx5_list_entry *entry, void *cb_ctx)
5407 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
5408 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
5409 struct mlx5_flow_dv_modify_hdr_resource *resource =
5410 container_of(entry, typeof(*resource), entry);
5411 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
5413 key_len += ref->actions_num * sizeof(ref->actions[0]);
5414 return ref->actions_num != resource->actions_num ||
5415 memcmp(&ref->ft_type, &resource->ft_type, key_len);
5418 static struct mlx5_indexed_pool *
5419 flow_dv_modify_ipool_get(struct mlx5_dev_ctx_shared *sh, uint8_t index)
5421 struct mlx5_indexed_pool *ipool = __atomic_load_n
5422 (&sh->mdh_ipools[index], __ATOMIC_SEQ_CST);
5425 struct mlx5_indexed_pool *expected = NULL;
5426 struct mlx5_indexed_pool_config cfg =
5427 (struct mlx5_indexed_pool_config) {
5428 .size = sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
5430 sizeof(struct mlx5_modification_cmd),
5435 .release_mem_en = !!sh->config.reclaim_mode,
5437 sh->config.reclaim_mode ? 0 : (1 << 16),
5438 .malloc = mlx5_malloc,
5440 .type = "mlx5_modify_action_resource",
5443 cfg.size = RTE_ALIGN(cfg.size, sizeof(ipool));
5444 ipool = mlx5_ipool_create(&cfg);
5447 if (!__atomic_compare_exchange_n(&sh->mdh_ipools[index],
5448 &expected, ipool, false,
5450 __ATOMIC_SEQ_CST)) {
5451 mlx5_ipool_destroy(ipool);
5452 ipool = __atomic_load_n(&sh->mdh_ipools[index],
5459 struct mlx5_list_entry *
5460 flow_dv_modify_create_cb(void *tool_ctx, void *cb_ctx)
5462 struct mlx5_dev_ctx_shared *sh = tool_ctx;
5463 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
5464 struct mlx5dv_dr_domain *ns;
5465 struct mlx5_flow_dv_modify_hdr_resource *entry;
5466 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
5467 struct mlx5_indexed_pool *ipool = flow_dv_modify_ipool_get(sh,
5468 ref->actions_num - 1);
5470 uint32_t data_len = ref->actions_num * sizeof(ref->actions[0]);
5471 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
5474 if (unlikely(!ipool)) {
5475 rte_flow_error_set(ctx->error, ENOMEM,
5476 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5477 NULL, "cannot allocate modify ipool");
5480 entry = mlx5_ipool_zmalloc(ipool, &idx);
5482 rte_flow_error_set(ctx->error, ENOMEM,
5483 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5484 "cannot allocate resource memory");
5487 rte_memcpy(&entry->ft_type,
5488 RTE_PTR_ADD(ref, offsetof(typeof(*ref), ft_type)),
5489 key_len + data_len);
5490 if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
5491 ns = sh->fdb_domain;
5492 else if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
5496 ret = mlx5_flow_os_create_flow_action_modify_header
5497 (sh->cdev->ctx, ns, entry,
5498 data_len, &entry->action);
5500 mlx5_ipool_free(sh->mdh_ipools[ref->actions_num - 1], idx);
5501 rte_flow_error_set(ctx->error, ENOMEM,
5502 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5503 NULL, "cannot create modification action");
5507 return &entry->entry;
5510 struct mlx5_list_entry *
5511 flow_dv_modify_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
5514 struct mlx5_dev_ctx_shared *sh = tool_ctx;
5515 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
5516 struct mlx5_flow_dv_modify_hdr_resource *entry;
5517 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
5518 uint32_t data_len = ref->actions_num * sizeof(ref->actions[0]);
5521 entry = mlx5_ipool_malloc(sh->mdh_ipools[ref->actions_num - 1],
5524 rte_flow_error_set(ctx->error, ENOMEM,
5525 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5526 "cannot allocate resource memory");
5529 memcpy(entry, oentry, sizeof(*entry) + data_len);
5531 return &entry->entry;
5535 flow_dv_modify_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
5537 struct mlx5_dev_ctx_shared *sh = tool_ctx;
5538 struct mlx5_flow_dv_modify_hdr_resource *res =
5539 container_of(entry, typeof(*res), entry);
5541 mlx5_ipool_free(sh->mdh_ipools[res->actions_num - 1], res->idx);
5545 * Validate the sample action.
5547 * @param[in, out] action_flags
5548 * Holds the actions detected until now.
5550 * Pointer to the sample action.
5552 * Pointer to the Ethernet device structure.
5554 * Attributes of flow that includes this action.
5555 * @param[in] item_flags
5556 * Holds the items detected.
5558 * Pointer to the RSS action.
5559 * @param[out] sample_rss
5560 * Pointer to the RSS action in sample action list.
5562 * Pointer to the COUNT action in sample action list.
5563 * @param[out] fdb_mirror_limit
5564 * Pointer to the FDB mirror limitation flag.
5566 * Pointer to error structure.
5569 * 0 on success, a negative errno value otherwise and rte_errno is set.
5572 flow_dv_validate_action_sample(uint64_t *action_flags,
5573 const struct rte_flow_action *action,
5574 struct rte_eth_dev *dev,
5575 const struct rte_flow_attr *attr,
5576 uint64_t item_flags,
5577 const struct rte_flow_action_rss *rss,
5578 const struct rte_flow_action_rss **sample_rss,
5579 const struct rte_flow_action_count **count,
5580 int *fdb_mirror_limit,
5581 struct rte_flow_error *error)
5583 struct mlx5_priv *priv = dev->data->dev_private;
5584 struct mlx5_sh_config *dev_conf = &priv->sh->config;
5585 const struct rte_flow_action_sample *sample = action->conf;
5586 const struct rte_flow_action *act;
5587 uint64_t sub_action_flags = 0;
5588 uint16_t queue_index = 0xFFFF;
5593 return rte_flow_error_set(error, EINVAL,
5594 RTE_FLOW_ERROR_TYPE_ACTION, action,
5595 "configuration cannot be NULL");
5596 if (sample->ratio == 0)
5597 return rte_flow_error_set(error, EINVAL,
5598 RTE_FLOW_ERROR_TYPE_ACTION, action,
5599 "ratio value starts from 1");
5600 if (!priv->sh->cdev->config.devx ||
5601 (sample->ratio > 0 && !priv->sampler_en))
5602 return rte_flow_error_set(error, ENOTSUP,
5603 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5605 "sample action not supported");
5606 if (*action_flags & MLX5_FLOW_ACTION_SAMPLE)
5607 return rte_flow_error_set(error, EINVAL,
5608 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5609 "Multiple sample actions not "
5611 if (*action_flags & MLX5_FLOW_ACTION_METER)
5612 return rte_flow_error_set(error, EINVAL,
5613 RTE_FLOW_ERROR_TYPE_ACTION, action,
5614 "wrong action order, meter should "
5615 "be after sample action");
5616 if (*action_flags & MLX5_FLOW_ACTION_JUMP)
5617 return rte_flow_error_set(error, EINVAL,
5618 RTE_FLOW_ERROR_TYPE_ACTION, action,
5619 "wrong action order, jump should "
5620 "be after sample action");
5621 if (*action_flags & MLX5_FLOW_ACTION_CT)
5622 return rte_flow_error_set(error, EINVAL,
5623 RTE_FLOW_ERROR_TYPE_ACTION, action,
5624 "Sample after CT not supported");
5625 act = sample->actions;
5626 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
5627 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5628 return rte_flow_error_set(error, ENOTSUP,
5629 RTE_FLOW_ERROR_TYPE_ACTION,
5630 act, "too many actions");
5631 switch (act->type) {
5632 case RTE_FLOW_ACTION_TYPE_QUEUE:
5633 ret = mlx5_flow_validate_action_queue(act,
5639 queue_index = ((const struct rte_flow_action_queue *)
5640 (act->conf))->index;
5641 sub_action_flags |= MLX5_FLOW_ACTION_QUEUE;
5644 case RTE_FLOW_ACTION_TYPE_RSS:
5645 *sample_rss = act->conf;
5646 ret = mlx5_flow_validate_action_rss(act,
5653 if (rss && *sample_rss &&
5654 ((*sample_rss)->level != rss->level ||
5655 (*sample_rss)->types != rss->types))
5656 return rte_flow_error_set(error, ENOTSUP,
5657 RTE_FLOW_ERROR_TYPE_ACTION,
5659 "Can't use the different RSS types "
5660 "or level in the same flow");
5661 if (*sample_rss != NULL && (*sample_rss)->queue_num)
5662 queue_index = (*sample_rss)->queue[0];
5663 sub_action_flags |= MLX5_FLOW_ACTION_RSS;
5666 case RTE_FLOW_ACTION_TYPE_MARK:
5667 ret = flow_dv_validate_action_mark(dev, act,
5672 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
5673 sub_action_flags |= MLX5_FLOW_ACTION_MARK |
5674 MLX5_FLOW_ACTION_MARK_EXT;
5676 sub_action_flags |= MLX5_FLOW_ACTION_MARK;
5679 case RTE_FLOW_ACTION_TYPE_COUNT:
5680 ret = flow_dv_validate_action_count
5681 (dev, false, *action_flags | sub_action_flags,
5686 sub_action_flags |= MLX5_FLOW_ACTION_COUNT;
5687 *action_flags |= MLX5_FLOW_ACTION_COUNT;
5690 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5691 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
5692 ret = flow_dv_validate_action_port_id(dev,
5699 sub_action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5702 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5703 ret = flow_dv_validate_action_raw_encap_decap
5704 (dev, NULL, act->conf, attr, &sub_action_flags,
5705 &actions_n, action, item_flags, error);
5710 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5711 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5712 ret = flow_dv_validate_action_l2_encap(dev,
5718 sub_action_flags |= MLX5_FLOW_ACTION_ENCAP;
5722 return rte_flow_error_set(error, ENOTSUP,
5723 RTE_FLOW_ERROR_TYPE_ACTION,
5725 "Doesn't support optional "
5729 if (attr->ingress && !attr->transfer) {
5730 if (!(sub_action_flags & (MLX5_FLOW_ACTION_QUEUE |
5731 MLX5_FLOW_ACTION_RSS)))
5732 return rte_flow_error_set(error, EINVAL,
5733 RTE_FLOW_ERROR_TYPE_ACTION,
5735 "Ingress must has a dest "
5736 "QUEUE for Sample");
5737 } else if (attr->egress && !attr->transfer) {
5738 return rte_flow_error_set(error, ENOTSUP,
5739 RTE_FLOW_ERROR_TYPE_ACTION,
5741 "Sample Only support Ingress "
5743 } else if (sample->actions->type != RTE_FLOW_ACTION_TYPE_END) {
5744 MLX5_ASSERT(attr->transfer);
5745 if (sample->ratio > 1)
5746 return rte_flow_error_set(error, ENOTSUP,
5747 RTE_FLOW_ERROR_TYPE_ACTION,
5749 "E-Switch doesn't support "
5750 "any optional action "
5752 if (sub_action_flags & MLX5_FLOW_ACTION_QUEUE)
5753 return rte_flow_error_set(error, ENOTSUP,
5754 RTE_FLOW_ERROR_TYPE_ACTION,
5756 "unsupported action QUEUE");
5757 if (sub_action_flags & MLX5_FLOW_ACTION_RSS)
5758 return rte_flow_error_set(error, ENOTSUP,
5759 RTE_FLOW_ERROR_TYPE_ACTION,
5761 "unsupported action QUEUE");
5762 if (!(sub_action_flags & MLX5_FLOW_ACTION_PORT_ID))
5763 return rte_flow_error_set(error, EINVAL,
5764 RTE_FLOW_ERROR_TYPE_ACTION,
5766 "E-Switch must has a dest "
5767 "port for mirroring");
5768 if (!priv->sh->cdev->config.hca_attr.reg_c_preserve &&
5769 priv->representor_id != UINT16_MAX)
5770 *fdb_mirror_limit = 1;
5772 /* Continue validation for Xcap actions.*/
5773 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) &&
5774 (queue_index == 0xFFFF ||
5775 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
5776 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
5777 MLX5_FLOW_XCAP_ACTIONS)
5778 return rte_flow_error_set(error, ENOTSUP,
5779 RTE_FLOW_ERROR_TYPE_ACTION,
5780 NULL, "encap and decap "
5781 "combination aren't "
5783 if (!attr->transfer && attr->ingress && (sub_action_flags &
5784 MLX5_FLOW_ACTION_ENCAP))
5785 return rte_flow_error_set(error, ENOTSUP,
5786 RTE_FLOW_ERROR_TYPE_ACTION,
5787 NULL, "encap is not supported"
5788 " for ingress traffic");
5794 * Find existing modify-header resource or create and register a new one.
5796 * @param dev[in, out]
5797 * Pointer to rte_eth_dev structure.
5798 * @param[in, out] resource
5799 * Pointer to modify-header resource.
5800 * @parm[in, out] dev_flow
5801 * Pointer to the dev_flow.
5803 * pointer to error structure.
5806 * 0 on success otherwise -errno and errno is set.
5809 flow_dv_modify_hdr_resource_register
5810 (struct rte_eth_dev *dev,
5811 struct mlx5_flow_dv_modify_hdr_resource *resource,
5812 struct mlx5_flow *dev_flow,
5813 struct rte_flow_error *error)
5815 struct mlx5_priv *priv = dev->data->dev_private;
5816 struct mlx5_dev_ctx_shared *sh = priv->sh;
5817 uint32_t key_len = sizeof(*resource) -
5818 offsetof(typeof(*resource), ft_type) +
5819 resource->actions_num * sizeof(resource->actions[0]);
5820 struct mlx5_list_entry *entry;
5821 struct mlx5_flow_cb_ctx ctx = {
5825 struct mlx5_hlist *modify_cmds;
5828 modify_cmds = flow_dv_hlist_prepare(sh, &sh->modify_cmds,
5830 MLX5_FLOW_HDR_MODIFY_HTABLE_SZ,
5832 flow_dv_modify_create_cb,
5833 flow_dv_modify_match_cb,
5834 flow_dv_modify_remove_cb,
5835 flow_dv_modify_clone_cb,
5836 flow_dv_modify_clone_free_cb,
5838 if (unlikely(!modify_cmds))
5840 resource->root = !dev_flow->dv.group;
5841 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
5843 return rte_flow_error_set(error, EOVERFLOW,
5844 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5845 "too many modify header items");
5846 key64 = __rte_raw_cksum(&resource->ft_type, key_len, 0);
5847 entry = mlx5_hlist_register(modify_cmds, key64, &ctx);
5850 resource = container_of(entry, typeof(*resource), entry);
5851 dev_flow->handle->dvh.modify_hdr = resource;
5856 * Get DV flow counter by index.
5859 * Pointer to the Ethernet device structure.
5861 * mlx5 flow counter index in the container.
5863 * mlx5 flow counter pool in the container.
5866 * Pointer to the counter, NULL otherwise.
5868 static struct mlx5_flow_counter *
5869 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
5871 struct mlx5_flow_counter_pool **ppool)
5873 struct mlx5_priv *priv = dev->data->dev_private;
5874 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
5875 struct mlx5_flow_counter_pool *pool;
5877 /* Decrease to original index and clear shared bit. */
5878 idx = (idx - 1) & (MLX5_CNT_SHARED_OFFSET - 1);
5879 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cmng->n);
5880 pool = cmng->pools[idx / MLX5_COUNTERS_PER_POOL];
5884 return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
5888 * Check the devx counter belongs to the pool.
5891 * Pointer to the counter pool.
5893 * The counter devx ID.
5896 * True if counter belongs to the pool, false otherwise.
5899 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
5901 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
5902 MLX5_COUNTERS_PER_POOL;
5904 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
5910 * Get a pool by devx counter ID.
5913 * Pointer to the counter management.
5915 * The counter devx ID.
5918 * The counter pool pointer if exists, NULL otherwise,
5920 static struct mlx5_flow_counter_pool *
5921 flow_dv_find_pool_by_id(struct mlx5_flow_counter_mng *cmng, int id)
5924 struct mlx5_flow_counter_pool *pool = NULL;
5926 rte_spinlock_lock(&cmng->pool_update_sl);
5927 /* Check last used pool. */
5928 if (cmng->last_pool_idx != POOL_IDX_INVALID &&
5929 flow_dv_is_counter_in_pool(cmng->pools[cmng->last_pool_idx], id)) {
5930 pool = cmng->pools[cmng->last_pool_idx];
5933 /* ID out of range means no suitable pool in the container. */
5934 if (id > cmng->max_id || id < cmng->min_id)
5937 * Find the pool from the end of the container, since mostly counter
5938 * ID is sequence increasing, and the last pool should be the needed
5943 struct mlx5_flow_counter_pool *pool_tmp = cmng->pools[i];
5945 if (flow_dv_is_counter_in_pool(pool_tmp, id)) {
5951 rte_spinlock_unlock(&cmng->pool_update_sl);
5956 * Resize a counter container.
5959 * Pointer to the Ethernet device structure.
5962 * 0 on success, otherwise negative errno value and rte_errno is set.
5965 flow_dv_container_resize(struct rte_eth_dev *dev)
5967 struct mlx5_priv *priv = dev->data->dev_private;
5968 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
5969 void *old_pools = cmng->pools;
5970 uint32_t resize = cmng->n + MLX5_CNT_CONTAINER_RESIZE;
5971 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
5972 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
5979 memcpy(pools, old_pools, cmng->n *
5980 sizeof(struct mlx5_flow_counter_pool *));
5982 cmng->pools = pools;
5984 mlx5_free(old_pools);
5989 * Query a devx flow counter.
5992 * Pointer to the Ethernet device structure.
5993 * @param[in] counter
5994 * Index to the flow counter.
5996 * The statistics value of packets.
5998 * The statistics value of bytes.
6001 * 0 on success, otherwise a negative errno value and rte_errno is set.
6004 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
6007 struct mlx5_priv *priv = dev->data->dev_private;
6008 struct mlx5_flow_counter_pool *pool = NULL;
6009 struct mlx5_flow_counter *cnt;
6012 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
6014 if (priv->sh->cmng.counter_fallback)
6015 return mlx5_devx_cmd_flow_counter_query(cnt->dcs_when_active, 0,
6016 0, pkts, bytes, 0, NULL, NULL, 0);
6017 rte_spinlock_lock(&pool->sl);
6022 offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
6023 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
6024 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
6026 rte_spinlock_unlock(&pool->sl);
6031 * Create and initialize a new counter pool.
6034 * Pointer to the Ethernet device structure.
6036 * The devX counter handle.
6038 * Whether the pool is for counter that was allocated for aging.
6039 * @param[in/out] cont_cur
6040 * Pointer to the container pointer, it will be update in pool resize.
6043 * The pool container pointer on success, NULL otherwise and rte_errno is set.
6045 static struct mlx5_flow_counter_pool *
6046 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
6049 struct mlx5_priv *priv = dev->data->dev_private;
6050 struct mlx5_flow_counter_pool *pool;
6051 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
6052 bool fallback = priv->sh->cmng.counter_fallback;
6053 uint32_t size = sizeof(*pool);
6055 size += MLX5_COUNTERS_PER_POOL * MLX5_CNT_SIZE;
6056 size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * MLX5_AGE_SIZE);
6057 pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
6063 pool->is_aged = !!age;
6064 pool->query_gen = 0;
6065 pool->min_dcs = dcs;
6066 rte_spinlock_init(&pool->sl);
6067 rte_spinlock_init(&pool->csl);
6068 TAILQ_INIT(&pool->counters[0]);
6069 TAILQ_INIT(&pool->counters[1]);
6070 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
6071 rte_spinlock_lock(&cmng->pool_update_sl);
6072 pool->index = cmng->n_valid;
6073 if (pool->index == cmng->n && flow_dv_container_resize(dev)) {
6075 rte_spinlock_unlock(&cmng->pool_update_sl);
6078 cmng->pools[pool->index] = pool;
6080 if (unlikely(fallback)) {
6081 int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
6083 if (base < cmng->min_id)
6084 cmng->min_id = base;
6085 if (base > cmng->max_id)
6086 cmng->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
6087 cmng->last_pool_idx = pool->index;
6089 rte_spinlock_unlock(&cmng->pool_update_sl);
6094 * Prepare a new counter and/or a new counter pool.
6097 * Pointer to the Ethernet device structure.
6098 * @param[out] cnt_free
6099 * Where to put the pointer of a new counter.
6101 * Whether the pool is for counter that was allocated for aging.
6104 * The counter pool pointer and @p cnt_free is set on success,
6105 * NULL otherwise and rte_errno is set.
6107 static struct mlx5_flow_counter_pool *
6108 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
6109 struct mlx5_flow_counter **cnt_free,
6112 struct mlx5_priv *priv = dev->data->dev_private;
6113 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
6114 struct mlx5_flow_counter_pool *pool;
6115 struct mlx5_counters tmp_tq;
6116 struct mlx5_devx_obj *dcs = NULL;
6117 struct mlx5_flow_counter *cnt;
6118 enum mlx5_counter_type cnt_type =
6119 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
6120 bool fallback = priv->sh->cmng.counter_fallback;
6124 /* bulk_bitmap must be 0 for single counter allocation. */
6125 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->cdev->ctx, 0);
6128 pool = flow_dv_find_pool_by_id(cmng, dcs->id);
6130 pool = flow_dv_pool_create(dev, dcs, age);
6132 mlx5_devx_cmd_destroy(dcs);
6136 i = dcs->id % MLX5_COUNTERS_PER_POOL;
6137 cnt = MLX5_POOL_GET_CNT(pool, i);
6139 cnt->dcs_when_free = dcs;
6143 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->cdev->ctx, 0x4);
6145 rte_errno = ENODATA;
6148 pool = flow_dv_pool_create(dev, dcs, age);
6150 mlx5_devx_cmd_destroy(dcs);
6153 TAILQ_INIT(&tmp_tq);
6154 for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
6155 cnt = MLX5_POOL_GET_CNT(pool, i);
6157 TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
6159 rte_spinlock_lock(&cmng->csl[cnt_type]);
6160 TAILQ_CONCAT(&cmng->counters[cnt_type], &tmp_tq, next);
6161 rte_spinlock_unlock(&cmng->csl[cnt_type]);
6162 *cnt_free = MLX5_POOL_GET_CNT(pool, 0);
6163 (*cnt_free)->pool = pool;
6168 * Allocate a flow counter.
6171 * Pointer to the Ethernet device structure.
6173 * Whether the counter was allocated for aging.
6176 * Index to flow counter on success, 0 otherwise and rte_errno is set.
6179 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t age)
6181 struct mlx5_priv *priv = dev->data->dev_private;
6182 struct mlx5_flow_counter_pool *pool = NULL;
6183 struct mlx5_flow_counter *cnt_free = NULL;
6184 bool fallback = priv->sh->cmng.counter_fallback;
6185 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
6186 enum mlx5_counter_type cnt_type =
6187 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
6190 if (!priv->sh->cdev->config.devx) {
6191 rte_errno = ENOTSUP;
6194 /* Get free counters from container. */
6195 rte_spinlock_lock(&cmng->csl[cnt_type]);
6196 cnt_free = TAILQ_FIRST(&cmng->counters[cnt_type]);
6198 TAILQ_REMOVE(&cmng->counters[cnt_type], cnt_free, next);
6199 rte_spinlock_unlock(&cmng->csl[cnt_type]);
6200 if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free, age))
6202 pool = cnt_free->pool;
6204 cnt_free->dcs_when_active = cnt_free->dcs_when_free;
6205 /* Create a DV counter action only in the first time usage. */
6206 if (!cnt_free->action) {
6208 struct mlx5_devx_obj *dcs;
6212 offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
6213 dcs = pool->min_dcs;
6216 dcs = cnt_free->dcs_when_free;
6218 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
6225 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
6226 MLX5_CNT_ARRAY_IDX(pool, cnt_free));
6227 /* Update the counter reset values. */
6228 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
6231 if (!fallback && !priv->sh->cmng.query_thread_on)
6232 /* Start the asynchronous batch query by the host thread. */
6233 mlx5_set_query_alarm(priv->sh);
6235 * When the count action isn't shared (by ID), shared_info field is
6236 * used for indirect action API's refcnt.
6237 * When the counter action is not shared neither by ID nor by indirect
6238 * action API, shared info must be 1.
6240 cnt_free->shared_info.refcnt = 1;
6244 cnt_free->pool = pool;
6246 cnt_free->dcs_when_free = cnt_free->dcs_when_active;
6247 rte_spinlock_lock(&cmng->csl[cnt_type]);
6248 TAILQ_INSERT_TAIL(&cmng->counters[cnt_type], cnt_free, next);
6249 rte_spinlock_unlock(&cmng->csl[cnt_type]);
6255 * Get age param from counter index.
6258 * Pointer to the Ethernet device structure.
6259 * @param[in] counter
6260 * Index to the counter handler.
6263 * The aging parameter specified for the counter index.
6265 static struct mlx5_age_param*
6266 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
6269 struct mlx5_flow_counter *cnt;
6270 struct mlx5_flow_counter_pool *pool = NULL;
6272 flow_dv_counter_get_by_idx(dev, counter, &pool);
6273 counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
6274 cnt = MLX5_POOL_GET_CNT(pool, counter);
6275 return MLX5_CNT_TO_AGE(cnt);
6279 * Remove a flow counter from aged counter list.
6282 * Pointer to the Ethernet device structure.
6283 * @param[in] counter
6284 * Index to the counter handler.
6286 * Pointer to the counter handler.
6289 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
6290 uint32_t counter, struct mlx5_flow_counter *cnt)
6292 struct mlx5_age_info *age_info;
6293 struct mlx5_age_param *age_param;
6294 struct mlx5_priv *priv = dev->data->dev_private;
6295 uint16_t expected = AGE_CANDIDATE;
6297 age_info = GET_PORT_AGE_INFO(priv);
6298 age_param = flow_dv_counter_idx_get_age(dev, counter);
6299 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
6300 AGE_FREE, false, __ATOMIC_RELAXED,
6301 __ATOMIC_RELAXED)) {
6303 * We need the lock even it is age timeout,
6304 * since counter may still in process.
6306 rte_spinlock_lock(&age_info->aged_sl);
6307 TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
6308 rte_spinlock_unlock(&age_info->aged_sl);
6309 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
6314 * Release a flow counter.
6317 * Pointer to the Ethernet device structure.
6318 * @param[in] counter
6319 * Index to the counter handler.
6322 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t counter)
6324 struct mlx5_priv *priv = dev->data->dev_private;
6325 struct mlx5_flow_counter_pool *pool = NULL;
6326 struct mlx5_flow_counter *cnt;
6327 enum mlx5_counter_type cnt_type;
6331 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
6333 if (pool->is_aged) {
6334 flow_dv_counter_remove_from_age(dev, counter, cnt);
6337 * If the counter action is shared by indirect action API,
6338 * the atomic function reduces its references counter.
6339 * If after the reduction the action is still referenced, the
6340 * function returns here and does not release it.
6341 * When the counter action is not shared by
6342 * indirect action API, shared info is 1 before the reduction,
6343 * so this condition is failed and function doesn't return here.
6345 if (__atomic_sub_fetch(&cnt->shared_info.refcnt, 1,
6351 * Put the counter back to list to be updated in none fallback mode.
6352 * Currently, we are using two list alternately, while one is in query,
6353 * add the freed counter to the other list based on the pool query_gen
6354 * value. After query finishes, add counter the list to the global
6355 * container counter list. The list changes while query starts. In
6356 * this case, lock will not be needed as query callback and release
6357 * function both operate with the different list.
6359 if (!priv->sh->cmng.counter_fallback) {
6360 rte_spinlock_lock(&pool->csl);
6361 TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
6362 rte_spinlock_unlock(&pool->csl);
6364 cnt->dcs_when_free = cnt->dcs_when_active;
6365 cnt_type = pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6366 MLX5_COUNTER_TYPE_ORIGIN;
6367 rte_spinlock_lock(&priv->sh->cmng.csl[cnt_type]);
6368 TAILQ_INSERT_TAIL(&priv->sh->cmng.counters[cnt_type],
6370 rte_spinlock_unlock(&priv->sh->cmng.csl[cnt_type]);
6375 * Resize a meter id container.
6378 * Pointer to the Ethernet device structure.
6381 * 0 on success, otherwise negative errno value and rte_errno is set.
6384 flow_dv_mtr_container_resize(struct rte_eth_dev *dev)
6386 struct mlx5_priv *priv = dev->data->dev_private;
6387 struct mlx5_aso_mtr_pools_mng *pools_mng =
6388 &priv->sh->mtrmng->pools_mng;
6389 void *old_pools = pools_mng->pools;
6390 uint32_t resize = pools_mng->n + MLX5_MTRS_CONTAINER_RESIZE;
6391 uint32_t mem_size = sizeof(struct mlx5_aso_mtr_pool *) * resize;
6392 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
6399 if (mlx5_aso_queue_init(priv->sh, ASO_OPC_MOD_POLICER)) {
6404 memcpy(pools, old_pools, pools_mng->n *
6405 sizeof(struct mlx5_aso_mtr_pool *));
6406 pools_mng->n = resize;
6407 pools_mng->pools = pools;
6409 mlx5_free(old_pools);
6414 * Prepare a new meter and/or a new meter pool.
6417 * Pointer to the Ethernet device structure.
6418 * @param[out] mtr_free
6419 * Where to put the pointer of a new meter.g.
6422 * The meter pool pointer and @mtr_free is set on success,
6423 * NULL otherwise and rte_errno is set.
6425 static struct mlx5_aso_mtr_pool *
6426 flow_dv_mtr_pool_create(struct rte_eth_dev *dev, struct mlx5_aso_mtr **mtr_free)
6428 struct mlx5_priv *priv = dev->data->dev_private;
6429 struct mlx5_aso_mtr_pools_mng *pools_mng = &priv->sh->mtrmng->pools_mng;
6430 struct mlx5_aso_mtr_pool *pool = NULL;
6431 struct mlx5_devx_obj *dcs = NULL;
6433 uint32_t log_obj_size;
6435 log_obj_size = rte_log2_u32(MLX5_ASO_MTRS_PER_POOL >> 1);
6436 dcs = mlx5_devx_cmd_create_flow_meter_aso_obj(priv->sh->cdev->ctx,
6437 priv->sh->cdev->pdn,
6440 rte_errno = ENODATA;
6443 pool = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*pool), 0, SOCKET_ID_ANY);
6446 claim_zero(mlx5_devx_cmd_destroy(dcs));
6449 pool->devx_obj = dcs;
6450 rte_rwlock_write_lock(&pools_mng->resize_mtrwl);
6451 pool->index = pools_mng->n_valid;
6452 if (pool->index == pools_mng->n && flow_dv_mtr_container_resize(dev)) {
6454 claim_zero(mlx5_devx_cmd_destroy(dcs));
6455 rte_rwlock_write_unlock(&pools_mng->resize_mtrwl);
6458 pools_mng->pools[pool->index] = pool;
6459 pools_mng->n_valid++;
6460 rte_rwlock_write_unlock(&pools_mng->resize_mtrwl);
6461 for (i = 1; i < MLX5_ASO_MTRS_PER_POOL; ++i) {
6462 pool->mtrs[i].offset = i;
6463 LIST_INSERT_HEAD(&pools_mng->meters, &pool->mtrs[i], next);
6465 pool->mtrs[0].offset = 0;
6466 *mtr_free = &pool->mtrs[0];
6471 * Release a flow meter into pool.
6474 * Pointer to the Ethernet device structure.
6475 * @param[in] mtr_idx
6476 * Index to aso flow meter.
6479 flow_dv_aso_mtr_release_to_pool(struct rte_eth_dev *dev, uint32_t mtr_idx)
6481 struct mlx5_priv *priv = dev->data->dev_private;
6482 struct mlx5_aso_mtr_pools_mng *pools_mng =
6483 &priv->sh->mtrmng->pools_mng;
6484 struct mlx5_aso_mtr *aso_mtr = mlx5_aso_meter_by_idx(priv, mtr_idx);
6486 MLX5_ASSERT(aso_mtr);
6487 rte_spinlock_lock(&pools_mng->mtrsl);
6488 memset(&aso_mtr->fm, 0, sizeof(struct mlx5_flow_meter_info));
6489 aso_mtr->state = ASO_METER_FREE;
6490 LIST_INSERT_HEAD(&pools_mng->meters, aso_mtr, next);
6491 rte_spinlock_unlock(&pools_mng->mtrsl);
6495 * Allocate a aso flow meter.
6498 * Pointer to the Ethernet device structure.
6501 * Index to aso flow meter on success, 0 otherwise and rte_errno is set.
6504 flow_dv_mtr_alloc(struct rte_eth_dev *dev)
6506 struct mlx5_priv *priv = dev->data->dev_private;
6507 struct mlx5_aso_mtr *mtr_free = NULL;
6508 struct mlx5_aso_mtr_pools_mng *pools_mng =
6509 &priv->sh->mtrmng->pools_mng;
6510 struct mlx5_aso_mtr_pool *pool;
6511 uint32_t mtr_idx = 0;
6513 if (!priv->sh->cdev->config.devx) {
6514 rte_errno = ENOTSUP;
6517 /* Allocate the flow meter memory. */
6518 /* Get free meters from management. */
6519 rte_spinlock_lock(&pools_mng->mtrsl);
6520 mtr_free = LIST_FIRST(&pools_mng->meters);
6522 LIST_REMOVE(mtr_free, next);
6523 if (!mtr_free && !flow_dv_mtr_pool_create(dev, &mtr_free)) {
6524 rte_spinlock_unlock(&pools_mng->mtrsl);
6527 mtr_free->state = ASO_METER_WAIT;
6528 rte_spinlock_unlock(&pools_mng->mtrsl);
6529 pool = container_of(mtr_free,
6530 struct mlx5_aso_mtr_pool,
6531 mtrs[mtr_free->offset]);
6532 mtr_idx = MLX5_MAKE_MTR_IDX(pool->index, mtr_free->offset);
6533 if (!mtr_free->fm.meter_action) {
6534 #ifdef HAVE_MLX5_DR_CREATE_ACTION_ASO
6535 struct rte_flow_error error;
6538 reg_id = mlx5_flow_get_reg_id(dev, MLX5_MTR_COLOR, 0, &error);
6539 mtr_free->fm.meter_action =
6540 mlx5_glue->dv_create_flow_action_aso
6541 (priv->sh->rx_domain,
6542 pool->devx_obj->obj,
6544 (1 << MLX5_FLOW_COLOR_GREEN),
6546 #endif /* HAVE_MLX5_DR_CREATE_ACTION_ASO */
6547 if (!mtr_free->fm.meter_action) {
6548 flow_dv_aso_mtr_release_to_pool(dev, mtr_idx);
6556 * Verify the @p attributes will be correctly understood by the NIC and store
6557 * them in the @p flow if everything is correct.
6560 * Pointer to dev struct.
6561 * @param[in] attributes
6562 * Pointer to flow attributes
6563 * @param[in] external
6564 * This flow rule is created by request external to PMD.
6566 * Pointer to error structure.
6569 * - 0 on success and non root table.
6570 * - 1 on success and root table.
6571 * - a negative errno value otherwise and rte_errno is set.
6574 flow_dv_validate_attributes(struct rte_eth_dev *dev,
6575 const struct mlx5_flow_tunnel *tunnel,
6576 const struct rte_flow_attr *attributes,
6577 const struct flow_grp_info *grp_info,
6578 struct rte_flow_error *error)
6580 struct mlx5_priv *priv = dev->data->dev_private;
6581 uint32_t lowest_priority = mlx5_get_lowest_priority(dev, attributes);
6584 #ifndef HAVE_MLX5DV_DR
6585 RTE_SET_USED(tunnel);
6586 RTE_SET_USED(grp_info);
6587 if (attributes->group)
6588 return rte_flow_error_set(error, ENOTSUP,
6589 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6591 "groups are not supported");
6595 ret = mlx5_flow_group_to_table(dev, tunnel, attributes->group, &table,
6600 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
6602 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
6603 attributes->priority > lowest_priority)
6604 return rte_flow_error_set(error, ENOTSUP,
6605 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
6607 "priority out of range");
6608 if (attributes->transfer) {
6609 if (!priv->sh->config.dv_esw_en)
6610 return rte_flow_error_set
6612 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6613 "E-Switch dr is not supported");
6614 if (attributes->egress)
6615 return rte_flow_error_set
6617 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
6618 "egress is not supported");
6620 if (!(attributes->egress ^ attributes->ingress))
6621 return rte_flow_error_set(error, ENOTSUP,
6622 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
6623 "must specify exactly one of "
6624 "ingress or egress");
6629 validate_integrity_bits(const struct rte_flow_item_integrity *mask,
6630 int64_t pattern_flags, uint64_t l3_flags,
6631 uint64_t l4_flags, uint64_t ip4_flag,
6632 struct rte_flow_error *error)
6634 if (mask->l3_ok && !(pattern_flags & l3_flags))
6635 return rte_flow_error_set(error, EINVAL,
6636 RTE_FLOW_ERROR_TYPE_ITEM,
6637 NULL, "missing L3 protocol");
6639 if (mask->ipv4_csum_ok && !(pattern_flags & ip4_flag))
6640 return rte_flow_error_set(error, EINVAL,
6641 RTE_FLOW_ERROR_TYPE_ITEM,
6642 NULL, "missing IPv4 protocol");
6644 if ((mask->l4_ok || mask->l4_csum_ok) && !(pattern_flags & l4_flags))
6645 return rte_flow_error_set(error, EINVAL,
6646 RTE_FLOW_ERROR_TYPE_ITEM,
6647 NULL, "missing L4 protocol");
6653 flow_dv_validate_item_integrity_post(const struct
6654 rte_flow_item *integrity_items[2],
6655 int64_t pattern_flags,
6656 struct rte_flow_error *error)
6658 const struct rte_flow_item_integrity *mask;
6661 if (pattern_flags & MLX5_FLOW_ITEM_OUTER_INTEGRITY) {
6662 mask = (typeof(mask))integrity_items[0]->mask;
6663 ret = validate_integrity_bits(mask, pattern_flags,
6664 MLX5_FLOW_LAYER_OUTER_L3,
6665 MLX5_FLOW_LAYER_OUTER_L4,
6666 MLX5_FLOW_LAYER_OUTER_L3_IPV4,
6671 if (pattern_flags & MLX5_FLOW_ITEM_INNER_INTEGRITY) {
6672 mask = (typeof(mask))integrity_items[1]->mask;
6673 ret = validate_integrity_bits(mask, pattern_flags,
6674 MLX5_FLOW_LAYER_INNER_L3,
6675 MLX5_FLOW_LAYER_INNER_L4,
6676 MLX5_FLOW_LAYER_INNER_L3_IPV4,
6685 flow_dv_validate_item_integrity(struct rte_eth_dev *dev,
6686 const struct rte_flow_item *integrity_item,
6687 uint64_t pattern_flags, uint64_t *last_item,
6688 const struct rte_flow_item *integrity_items[2],
6689 struct rte_flow_error *error)
6691 struct mlx5_priv *priv = dev->data->dev_private;
6692 const struct rte_flow_item_integrity *mask = (typeof(mask))
6693 integrity_item->mask;
6694 const struct rte_flow_item_integrity *spec = (typeof(spec))
6695 integrity_item->spec;
6697 if (!priv->sh->cdev->config.hca_attr.pkt_integrity_match)
6698 return rte_flow_error_set(error, ENOTSUP,
6699 RTE_FLOW_ERROR_TYPE_ITEM,
6701 "packet integrity integrity_item not supported");
6703 return rte_flow_error_set(error, ENOTSUP,
6704 RTE_FLOW_ERROR_TYPE_ITEM,
6706 "no spec for integrity item");
6708 mask = &rte_flow_item_integrity_mask;
6709 if (!mlx5_validate_integrity_item(mask))
6710 return rte_flow_error_set(error, ENOTSUP,
6711 RTE_FLOW_ERROR_TYPE_ITEM,
6713 "unsupported integrity filter");
6714 if (spec->level > 1) {
6715 if (pattern_flags & MLX5_FLOW_ITEM_INNER_INTEGRITY)
6716 return rte_flow_error_set
6718 RTE_FLOW_ERROR_TYPE_ITEM,
6719 NULL, "multiple inner integrity items not supported");
6720 integrity_items[1] = integrity_item;
6721 *last_item |= MLX5_FLOW_ITEM_INNER_INTEGRITY;
6723 if (pattern_flags & MLX5_FLOW_ITEM_OUTER_INTEGRITY)
6724 return rte_flow_error_set
6726 RTE_FLOW_ERROR_TYPE_ITEM,
6727 NULL, "multiple outer integrity items not supported");
6728 integrity_items[0] = integrity_item;
6729 *last_item |= MLX5_FLOW_ITEM_OUTER_INTEGRITY;
6735 flow_dv_validate_item_flex(struct rte_eth_dev *dev,
6736 const struct rte_flow_item *item,
6737 uint64_t item_flags,
6738 uint64_t *last_item,
6740 struct rte_flow_error *error)
6742 const struct rte_flow_item_flex *flow_spec = item->spec;
6743 const struct rte_flow_item_flex *flow_mask = item->mask;
6744 struct mlx5_flex_item *flex;
6747 return rte_flow_error_set(error, EINVAL,
6748 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
6749 "flex flow item spec cannot be NULL");
6751 return rte_flow_error_set(error, EINVAL,
6752 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
6753 "flex flow item mask cannot be NULL");
6755 return rte_flow_error_set(error, ENOTSUP,
6756 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
6757 "flex flow item last not supported");
6758 if (mlx5_flex_acquire_index(dev, flow_spec->handle, false) < 0)
6759 return rte_flow_error_set(error, EINVAL,
6760 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
6761 "invalid flex flow item handle");
6762 flex = (struct mlx5_flex_item *)flow_spec->handle;
6763 switch (flex->tunnel_mode) {
6764 case FLEX_TUNNEL_MODE_SINGLE:
6766 (MLX5_FLOW_ITEM_OUTER_FLEX | MLX5_FLOW_ITEM_INNER_FLEX))
6767 rte_flow_error_set(error, EINVAL,
6768 RTE_FLOW_ERROR_TYPE_ITEM,
6769 NULL, "multiple flex items not supported");
6771 case FLEX_TUNNEL_MODE_OUTER:
6773 rte_flow_error_set(error, EINVAL,
6774 RTE_FLOW_ERROR_TYPE_ITEM,
6775 NULL, "inner flex item was not configured");
6776 if (item_flags & MLX5_FLOW_ITEM_OUTER_FLEX)
6777 rte_flow_error_set(error, ENOTSUP,
6778 RTE_FLOW_ERROR_TYPE_ITEM,
6779 NULL, "multiple flex items not supported");
6781 case FLEX_TUNNEL_MODE_INNER:
6783 rte_flow_error_set(error, EINVAL,
6784 RTE_FLOW_ERROR_TYPE_ITEM,
6785 NULL, "outer flex item was not configured");
6786 if (item_flags & MLX5_FLOW_ITEM_INNER_FLEX)
6787 rte_flow_error_set(error, EINVAL,
6788 RTE_FLOW_ERROR_TYPE_ITEM,
6789 NULL, "multiple flex items not supported");
6791 case FLEX_TUNNEL_MODE_MULTI:
6792 if ((is_inner && (item_flags & MLX5_FLOW_ITEM_INNER_FLEX)) ||
6793 (!is_inner && (item_flags & MLX5_FLOW_ITEM_OUTER_FLEX))) {
6794 rte_flow_error_set(error, EINVAL,
6795 RTE_FLOW_ERROR_TYPE_ITEM,
6796 NULL, "multiple flex items not supported");
6799 case FLEX_TUNNEL_MODE_TUNNEL:
6800 if (is_inner || (item_flags & MLX5_FLOW_ITEM_FLEX_TUNNEL))
6801 rte_flow_error_set(error, EINVAL,
6802 RTE_FLOW_ERROR_TYPE_ITEM,
6803 NULL, "multiple flex tunnel items not supported");
6806 rte_flow_error_set(error, EINVAL,
6807 RTE_FLOW_ERROR_TYPE_ITEM,
6808 NULL, "invalid flex item configuration");
6810 *last_item = flex->tunnel_mode == FLEX_TUNNEL_MODE_TUNNEL ?
6811 MLX5_FLOW_ITEM_FLEX_TUNNEL : is_inner ?
6812 MLX5_FLOW_ITEM_INNER_FLEX : MLX5_FLOW_ITEM_OUTER_FLEX;
6817 * Internal validation function. For validating both actions and items.
6820 * Pointer to the rte_eth_dev structure.
6822 * Pointer to the flow attributes.
6824 * Pointer to the list of items.
6825 * @param[in] actions
6826 * Pointer to the list of actions.
6827 * @param[in] external
6828 * This flow rule is created by request external to PMD.
6829 * @param[in] hairpin
6830 * Number of hairpin TX actions, 0 means classic flow.
6832 * Pointer to the error structure.
6835 * 0 on success, a negative errno value otherwise and rte_errno is set.
6838 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
6839 const struct rte_flow_item items[],
6840 const struct rte_flow_action actions[],
6841 bool external, int hairpin, struct rte_flow_error *error)
6844 uint64_t action_flags = 0;
6845 uint64_t item_flags = 0;
6846 uint64_t last_item = 0;
6847 uint8_t next_protocol = 0xff;
6848 uint16_t ether_type = 0;
6850 uint8_t item_ipv6_proto = 0;
6851 int fdb_mirror_limit = 0;
6852 int modify_after_mirror = 0;
6853 const struct rte_flow_item *geneve_item = NULL;
6854 const struct rte_flow_item *gre_item = NULL;
6855 const struct rte_flow_item *gtp_item = NULL;
6856 const struct rte_flow_action_raw_decap *decap;
6857 const struct rte_flow_action_raw_encap *encap;
6858 const struct rte_flow_action_rss *rss = NULL;
6859 const struct rte_flow_action_rss *sample_rss = NULL;
6860 const struct rte_flow_action_count *sample_count = NULL;
6861 const struct rte_flow_item_tcp nic_tcp_mask = {
6864 .src_port = RTE_BE16(UINT16_MAX),
6865 .dst_port = RTE_BE16(UINT16_MAX),
6868 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
6871 "\xff\xff\xff\xff\xff\xff\xff\xff"
6872 "\xff\xff\xff\xff\xff\xff\xff\xff",
6874 "\xff\xff\xff\xff\xff\xff\xff\xff"
6875 "\xff\xff\xff\xff\xff\xff\xff\xff",
6876 .vtc_flow = RTE_BE32(0xffffffff),
6882 const struct rte_flow_item_ecpri nic_ecpri_mask = {
6886 RTE_BE32(((const struct rte_ecpri_common_hdr) {
6890 .dummy[0] = 0xffffffff,
6893 struct mlx5_priv *priv = dev->data->dev_private;
6894 struct mlx5_sh_config *dev_conf = &priv->sh->config;
6895 uint16_t queue_index = 0xFFFF;
6896 const struct rte_flow_item_vlan *vlan_m = NULL;
6897 uint32_t rw_act_num = 0;
6899 const struct mlx5_flow_tunnel *tunnel;
6900 enum mlx5_tof_rule_type tof_rule_type;
6901 struct flow_grp_info grp_info = {
6902 .external = !!external,
6903 .transfer = !!attr->transfer,
6904 .fdb_def_rule = !!priv->fdb_def_rule,
6905 .std_tbl_fix = true,
6907 const struct rte_eth_hairpin_conf *conf;
6908 const struct rte_flow_item *integrity_items[2] = {NULL, NULL};
6909 const struct rte_flow_item *port_id_item = NULL;
6910 bool def_policy = false;
6911 uint16_t udp_dport = 0;
6915 tunnel = is_tunnel_offload_active(dev) ?
6916 mlx5_get_tof(items, actions, &tof_rule_type) : NULL;
6918 if (!dev_conf->dv_flow_en)
6919 return rte_flow_error_set
6921 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6922 NULL, "tunnel offload requires DV flow interface");
6923 if (priv->representor)
6924 return rte_flow_error_set
6926 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6927 NULL, "decap not supported for VF representor");
6928 if (tof_rule_type == MLX5_TUNNEL_OFFLOAD_SET_RULE)
6929 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
6930 else if (tof_rule_type == MLX5_TUNNEL_OFFLOAD_MATCH_RULE)
6931 action_flags |= MLX5_FLOW_ACTION_TUNNEL_MATCH |
6932 MLX5_FLOW_ACTION_DECAP;
6933 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
6934 (dev, attr, tunnel, tof_rule_type);
6936 ret = flow_dv_validate_attributes(dev, tunnel, attr, &grp_info, error);
6939 is_root = (uint64_t)ret;
6940 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
6941 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
6942 int type = items->type;
6944 if (!mlx5_flow_os_item_supported(type))
6945 return rte_flow_error_set(error, ENOTSUP,
6946 RTE_FLOW_ERROR_TYPE_ITEM,
6947 NULL, "item not supported");
6949 case RTE_FLOW_ITEM_TYPE_VOID:
6951 case RTE_FLOW_ITEM_TYPE_PORT_ID:
6952 ret = flow_dv_validate_item_port_id
6953 (dev, items, attr, item_flags, error);
6956 last_item = MLX5_FLOW_ITEM_PORT_ID;
6957 port_id_item = items;
6959 case RTE_FLOW_ITEM_TYPE_ETH:
6960 ret = mlx5_flow_validate_item_eth(items, item_flags,
6964 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
6965 MLX5_FLOW_LAYER_OUTER_L2;
6966 if (items->mask != NULL && items->spec != NULL) {
6968 ((const struct rte_flow_item_eth *)
6971 ((const struct rte_flow_item_eth *)
6973 ether_type = rte_be_to_cpu_16(ether_type);
6978 case RTE_FLOW_ITEM_TYPE_VLAN:
6979 ret = flow_dv_validate_item_vlan(items, item_flags,
6983 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
6984 MLX5_FLOW_LAYER_OUTER_VLAN;
6985 if (items->mask != NULL && items->spec != NULL) {
6987 ((const struct rte_flow_item_vlan *)
6988 items->spec)->inner_type;
6990 ((const struct rte_flow_item_vlan *)
6991 items->mask)->inner_type;
6992 ether_type = rte_be_to_cpu_16(ether_type);
6996 /* Store outer VLAN mask for of_push_vlan action. */
6998 vlan_m = items->mask;
7000 case RTE_FLOW_ITEM_TYPE_IPV4:
7001 mlx5_flow_tunnel_ip_check(items, next_protocol,
7002 &item_flags, &tunnel);
7003 ret = flow_dv_validate_item_ipv4(dev, items, item_flags,
7004 last_item, ether_type,
7008 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
7009 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
7010 if (items->mask != NULL &&
7011 ((const struct rte_flow_item_ipv4 *)
7012 items->mask)->hdr.next_proto_id) {
7014 ((const struct rte_flow_item_ipv4 *)
7015 (items->spec))->hdr.next_proto_id;
7017 ((const struct rte_flow_item_ipv4 *)
7018 (items->mask))->hdr.next_proto_id;
7020 /* Reset for inner layer. */
7021 next_protocol = 0xff;
7024 case RTE_FLOW_ITEM_TYPE_IPV6:
7025 mlx5_flow_tunnel_ip_check(items, next_protocol,
7026 &item_flags, &tunnel);
7027 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
7034 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
7035 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
7036 if (items->mask != NULL &&
7037 ((const struct rte_flow_item_ipv6 *)
7038 items->mask)->hdr.proto) {
7040 ((const struct rte_flow_item_ipv6 *)
7041 items->spec)->hdr.proto;
7043 ((const struct rte_flow_item_ipv6 *)
7044 items->spec)->hdr.proto;
7046 ((const struct rte_flow_item_ipv6 *)
7047 items->mask)->hdr.proto;
7049 /* Reset for inner layer. */
7050 next_protocol = 0xff;
7053 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
7054 ret = flow_dv_validate_item_ipv6_frag_ext(items,
7059 last_item = tunnel ?
7060 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
7061 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
7062 if (items->mask != NULL &&
7063 ((const struct rte_flow_item_ipv6_frag_ext *)
7064 items->mask)->hdr.next_header) {
7066 ((const struct rte_flow_item_ipv6_frag_ext *)
7067 items->spec)->hdr.next_header;
7069 ((const struct rte_flow_item_ipv6_frag_ext *)
7070 items->mask)->hdr.next_header;
7072 /* Reset for inner layer. */
7073 next_protocol = 0xff;
7076 case RTE_FLOW_ITEM_TYPE_TCP:
7077 ret = mlx5_flow_validate_item_tcp
7084 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
7085 MLX5_FLOW_LAYER_OUTER_L4_TCP;
7087 case RTE_FLOW_ITEM_TYPE_UDP:
7088 ret = mlx5_flow_validate_item_udp(items, item_flags,
7091 const struct rte_flow_item_udp *spec = items->spec;
7092 const struct rte_flow_item_udp *mask = items->mask;
7094 mask = &rte_flow_item_udp_mask;
7096 udp_dport = rte_be_to_cpu_16
7097 (spec->hdr.dst_port &
7098 mask->hdr.dst_port);
7101 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
7102 MLX5_FLOW_LAYER_OUTER_L4_UDP;
7104 case RTE_FLOW_ITEM_TYPE_GRE:
7105 ret = mlx5_flow_validate_item_gre(items, item_flags,
7106 next_protocol, error);
7110 last_item = MLX5_FLOW_LAYER_GRE;
7112 case RTE_FLOW_ITEM_TYPE_NVGRE:
7113 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
7118 last_item = MLX5_FLOW_LAYER_NVGRE;
7120 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
7121 ret = mlx5_flow_validate_item_gre_key
7122 (items, item_flags, gre_item, error);
7125 last_item = MLX5_FLOW_LAYER_GRE_KEY;
7127 case RTE_FLOW_ITEM_TYPE_VXLAN:
7128 ret = mlx5_flow_validate_item_vxlan(dev, udp_dport,
7133 last_item = MLX5_FLOW_LAYER_VXLAN;
7135 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
7136 ret = mlx5_flow_validate_item_vxlan_gpe(items,
7141 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
7143 case RTE_FLOW_ITEM_TYPE_GENEVE:
7144 ret = mlx5_flow_validate_item_geneve(items,
7149 geneve_item = items;
7150 last_item = MLX5_FLOW_LAYER_GENEVE;
7152 case RTE_FLOW_ITEM_TYPE_GENEVE_OPT:
7153 ret = mlx5_flow_validate_item_geneve_opt(items,
7160 last_item = MLX5_FLOW_LAYER_GENEVE_OPT;
7162 case RTE_FLOW_ITEM_TYPE_MPLS:
7163 ret = mlx5_flow_validate_item_mpls(dev, items,
7168 last_item = MLX5_FLOW_LAYER_MPLS;
7171 case RTE_FLOW_ITEM_TYPE_MARK:
7172 ret = flow_dv_validate_item_mark(dev, items, attr,
7176 last_item = MLX5_FLOW_ITEM_MARK;
7178 case RTE_FLOW_ITEM_TYPE_META:
7179 ret = flow_dv_validate_item_meta(dev, items, attr,
7183 last_item = MLX5_FLOW_ITEM_METADATA;
7185 case RTE_FLOW_ITEM_TYPE_ICMP:
7186 ret = mlx5_flow_validate_item_icmp(items, item_flags,
7191 last_item = MLX5_FLOW_LAYER_ICMP;
7193 case RTE_FLOW_ITEM_TYPE_ICMP6:
7194 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
7199 item_ipv6_proto = IPPROTO_ICMPV6;
7200 last_item = MLX5_FLOW_LAYER_ICMP6;
7202 case RTE_FLOW_ITEM_TYPE_TAG:
7203 ret = flow_dv_validate_item_tag(dev, items,
7207 last_item = MLX5_FLOW_ITEM_TAG;
7209 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
7210 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
7212 case RTE_FLOW_ITEM_TYPE_GTP:
7213 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
7218 last_item = MLX5_FLOW_LAYER_GTP;
7220 case RTE_FLOW_ITEM_TYPE_GTP_PSC:
7221 ret = flow_dv_validate_item_gtp_psc(items, last_item,
7226 last_item = MLX5_FLOW_LAYER_GTP_PSC;
7228 case RTE_FLOW_ITEM_TYPE_ECPRI:
7229 /* Capacity will be checked in the translate stage. */
7230 ret = mlx5_flow_validate_item_ecpri(items, item_flags,
7237 last_item = MLX5_FLOW_LAYER_ECPRI;
7239 case RTE_FLOW_ITEM_TYPE_INTEGRITY:
7240 ret = flow_dv_validate_item_integrity(dev, items,
7248 case RTE_FLOW_ITEM_TYPE_CONNTRACK:
7249 ret = flow_dv_validate_item_aso_ct(dev, items,
7250 &item_flags, error);
7254 case MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL:
7255 /* tunnel offload item was processed before
7256 * list it here as a supported type
7259 case RTE_FLOW_ITEM_TYPE_FLEX:
7260 ret = flow_dv_validate_item_flex(dev, items, item_flags,
7262 tunnel != 0, error);
7267 return rte_flow_error_set(error, ENOTSUP,
7268 RTE_FLOW_ERROR_TYPE_ITEM,
7269 NULL, "item not supported");
7271 item_flags |= last_item;
7273 if (item_flags & MLX5_FLOW_ITEM_INTEGRITY) {
7274 ret = flow_dv_validate_item_integrity_post(integrity_items,
7279 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
7280 int type = actions->type;
7281 bool shared_count = false;
7283 if (!mlx5_flow_os_action_supported(type))
7284 return rte_flow_error_set(error, ENOTSUP,
7285 RTE_FLOW_ERROR_TYPE_ACTION,
7287 "action not supported");
7288 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
7289 return rte_flow_error_set(error, ENOTSUP,
7290 RTE_FLOW_ERROR_TYPE_ACTION,
7291 actions, "too many actions");
7293 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY)
7294 return rte_flow_error_set(error, ENOTSUP,
7295 RTE_FLOW_ERROR_TYPE_ACTION,
7296 NULL, "meter action with policy "
7297 "must be the last action");
7299 case RTE_FLOW_ACTION_TYPE_VOID:
7301 case RTE_FLOW_ACTION_TYPE_PORT_ID:
7302 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
7303 ret = flow_dv_validate_action_port_id(dev,
7310 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
7313 case RTE_FLOW_ACTION_TYPE_FLAG:
7314 ret = flow_dv_validate_action_flag(dev, action_flags,
7318 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7319 /* Count all modify-header actions as one. */
7320 if (!(action_flags &
7321 MLX5_FLOW_MODIFY_HDR_ACTIONS))
7323 action_flags |= MLX5_FLOW_ACTION_FLAG |
7324 MLX5_FLOW_ACTION_MARK_EXT;
7325 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7326 modify_after_mirror = 1;
7329 action_flags |= MLX5_FLOW_ACTION_FLAG;
7332 rw_act_num += MLX5_ACT_NUM_SET_MARK;
7334 case RTE_FLOW_ACTION_TYPE_MARK:
7335 ret = flow_dv_validate_action_mark(dev, actions,
7340 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7341 /* Count all modify-header actions as one. */
7342 if (!(action_flags &
7343 MLX5_FLOW_MODIFY_HDR_ACTIONS))
7345 action_flags |= MLX5_FLOW_ACTION_MARK |
7346 MLX5_FLOW_ACTION_MARK_EXT;
7347 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7348 modify_after_mirror = 1;
7350 action_flags |= MLX5_FLOW_ACTION_MARK;
7353 rw_act_num += MLX5_ACT_NUM_SET_MARK;
7355 case RTE_FLOW_ACTION_TYPE_SET_META:
7356 ret = flow_dv_validate_action_set_meta(dev, actions,
7361 /* Count all modify-header actions as one action. */
7362 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7364 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7365 modify_after_mirror = 1;
7366 action_flags |= MLX5_FLOW_ACTION_SET_META;
7367 rw_act_num += MLX5_ACT_NUM_SET_META;
7369 case RTE_FLOW_ACTION_TYPE_SET_TAG:
7370 ret = flow_dv_validate_action_set_tag(dev, actions,
7375 /* Count all modify-header actions as one action. */
7376 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7378 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7379 modify_after_mirror = 1;
7380 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7381 rw_act_num += MLX5_ACT_NUM_SET_TAG;
7383 case RTE_FLOW_ACTION_TYPE_DROP:
7384 ret = mlx5_flow_validate_action_drop(action_flags,
7388 action_flags |= MLX5_FLOW_ACTION_DROP;
7391 case RTE_FLOW_ACTION_TYPE_QUEUE:
7392 ret = mlx5_flow_validate_action_queue(actions,
7397 queue_index = ((const struct rte_flow_action_queue *)
7398 (actions->conf))->index;
7399 action_flags |= MLX5_FLOW_ACTION_QUEUE;
7402 case RTE_FLOW_ACTION_TYPE_RSS:
7403 rss = actions->conf;
7404 ret = mlx5_flow_validate_action_rss(actions,
7410 if (rss && sample_rss &&
7411 (sample_rss->level != rss->level ||
7412 sample_rss->types != rss->types))
7413 return rte_flow_error_set(error, ENOTSUP,
7414 RTE_FLOW_ERROR_TYPE_ACTION,
7416 "Can't use the different RSS types "
7417 "or level in the same flow");
7418 if (rss != NULL && rss->queue_num)
7419 queue_index = rss->queue[0];
7420 action_flags |= MLX5_FLOW_ACTION_RSS;
7423 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
7425 mlx5_flow_validate_action_default_miss(action_flags,
7429 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
7432 case MLX5_RTE_FLOW_ACTION_TYPE_COUNT:
7433 shared_count = true;
7435 case RTE_FLOW_ACTION_TYPE_COUNT:
7436 ret = flow_dv_validate_action_count(dev, shared_count,
7441 action_flags |= MLX5_FLOW_ACTION_COUNT;
7444 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
7445 if (flow_dv_validate_action_pop_vlan(dev,
7451 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7452 modify_after_mirror = 1;
7453 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
7456 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
7457 ret = flow_dv_validate_action_push_vlan(dev,
7464 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7465 modify_after_mirror = 1;
7466 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
7469 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
7470 ret = flow_dv_validate_action_set_vlan_pcp
7471 (action_flags, actions, error);
7474 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7475 modify_after_mirror = 1;
7476 /* Count PCP with push_vlan command. */
7477 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
7479 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
7480 ret = flow_dv_validate_action_set_vlan_vid
7481 (item_flags, action_flags,
7485 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7486 modify_after_mirror = 1;
7487 /* Count VID with push_vlan command. */
7488 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
7489 rw_act_num += MLX5_ACT_NUM_MDF_VID;
7491 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
7492 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
7493 ret = flow_dv_validate_action_l2_encap(dev,
7499 action_flags |= MLX5_FLOW_ACTION_ENCAP;
7502 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
7503 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
7504 ret = flow_dv_validate_action_decap(dev, action_flags,
7505 actions, item_flags,
7509 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7510 modify_after_mirror = 1;
7511 action_flags |= MLX5_FLOW_ACTION_DECAP;
7514 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
7515 ret = flow_dv_validate_action_raw_encap_decap
7516 (dev, NULL, actions->conf, attr, &action_flags,
7517 &actions_n, actions, item_flags, error);
7521 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
7522 decap = actions->conf;
7523 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
7525 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
7529 encap = actions->conf;
7531 ret = flow_dv_validate_action_raw_encap_decap
7533 decap ? decap : &empty_decap, encap,
7534 attr, &action_flags, &actions_n,
7535 actions, item_flags, error);
7538 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) &&
7539 (action_flags & MLX5_FLOW_ACTION_DECAP))
7540 modify_after_mirror = 1;
7542 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
7543 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
7544 ret = flow_dv_validate_action_modify_mac(action_flags,
7550 /* Count all modify-header actions as one action. */
7551 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7553 action_flags |= actions->type ==
7554 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
7555 MLX5_FLOW_ACTION_SET_MAC_SRC :
7556 MLX5_FLOW_ACTION_SET_MAC_DST;
7557 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7558 modify_after_mirror = 1;
7560 * Even if the source and destination MAC addresses have
7561 * overlap in the header with 4B alignment, the convert
7562 * function will handle them separately and 4 SW actions
7563 * will be created. And 2 actions will be added each
7564 * time no matter how many bytes of address will be set.
7566 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
7568 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
7569 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
7570 ret = flow_dv_validate_action_modify_ipv4(action_flags,
7576 /* Count all modify-header actions as one action. */
7577 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7579 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7580 modify_after_mirror = 1;
7581 action_flags |= actions->type ==
7582 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
7583 MLX5_FLOW_ACTION_SET_IPV4_SRC :
7584 MLX5_FLOW_ACTION_SET_IPV4_DST;
7585 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
7587 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
7588 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
7589 ret = flow_dv_validate_action_modify_ipv6(action_flags,
7595 if (item_ipv6_proto == IPPROTO_ICMPV6)
7596 return rte_flow_error_set(error, ENOTSUP,
7597 RTE_FLOW_ERROR_TYPE_ACTION,
7599 "Can't change header "
7600 "with ICMPv6 proto");
7601 /* Count all modify-header actions as one action. */
7602 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7604 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7605 modify_after_mirror = 1;
7606 action_flags |= actions->type ==
7607 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
7608 MLX5_FLOW_ACTION_SET_IPV6_SRC :
7609 MLX5_FLOW_ACTION_SET_IPV6_DST;
7610 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
7612 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
7613 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
7614 ret = flow_dv_validate_action_modify_tp(action_flags,
7620 /* Count all modify-header actions as one action. */
7621 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7623 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7624 modify_after_mirror = 1;
7625 action_flags |= actions->type ==
7626 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
7627 MLX5_FLOW_ACTION_SET_TP_SRC :
7628 MLX5_FLOW_ACTION_SET_TP_DST;
7629 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
7631 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
7632 case RTE_FLOW_ACTION_TYPE_SET_TTL:
7633 ret = flow_dv_validate_action_modify_ttl(action_flags,
7639 /* Count all modify-header actions as one action. */
7640 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7642 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7643 modify_after_mirror = 1;
7644 action_flags |= actions->type ==
7645 RTE_FLOW_ACTION_TYPE_SET_TTL ?
7646 MLX5_FLOW_ACTION_SET_TTL :
7647 MLX5_FLOW_ACTION_DEC_TTL;
7648 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
7650 case RTE_FLOW_ACTION_TYPE_JUMP:
7651 ret = flow_dv_validate_action_jump(dev, tunnel, actions,
7657 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) &&
7659 return rte_flow_error_set(error, EINVAL,
7660 RTE_FLOW_ERROR_TYPE_ACTION,
7662 "sample and jump action combination is not supported");
7664 action_flags |= MLX5_FLOW_ACTION_JUMP;
7666 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
7667 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
7668 ret = flow_dv_validate_action_modify_tcp_seq
7675 /* Count all modify-header actions as one action. */
7676 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7678 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7679 modify_after_mirror = 1;
7680 action_flags |= actions->type ==
7681 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
7682 MLX5_FLOW_ACTION_INC_TCP_SEQ :
7683 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
7684 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
7686 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
7687 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
7688 ret = flow_dv_validate_action_modify_tcp_ack
7695 /* Count all modify-header actions as one action. */
7696 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7698 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7699 modify_after_mirror = 1;
7700 action_flags |= actions->type ==
7701 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
7702 MLX5_FLOW_ACTION_INC_TCP_ACK :
7703 MLX5_FLOW_ACTION_DEC_TCP_ACK;
7704 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
7706 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
7708 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
7709 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
7710 rw_act_num += MLX5_ACT_NUM_SET_TAG;
7712 case RTE_FLOW_ACTION_TYPE_METER:
7713 ret = mlx5_flow_validate_action_meter(dev,
7722 action_flags |= MLX5_FLOW_ACTION_METER;
7725 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY;
7727 /* Meter action will add one more TAG action. */
7728 rw_act_num += MLX5_ACT_NUM_SET_TAG;
7730 case MLX5_RTE_FLOW_ACTION_TYPE_AGE:
7731 if (!attr->transfer && !attr->group)
7732 return rte_flow_error_set(error, ENOTSUP,
7733 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7735 "Shared ASO age action is not supported for group 0");
7736 if (action_flags & MLX5_FLOW_ACTION_AGE)
7737 return rte_flow_error_set
7739 RTE_FLOW_ERROR_TYPE_ACTION,
7741 "duplicate age actions set");
7742 action_flags |= MLX5_FLOW_ACTION_AGE;
7745 case RTE_FLOW_ACTION_TYPE_AGE:
7746 ret = flow_dv_validate_action_age(action_flags,
7752 * Validate the regular AGE action (using counter)
7753 * mutual exclusion with share counter actions.
7755 if (!priv->sh->flow_hit_aso_en) {
7757 return rte_flow_error_set
7759 RTE_FLOW_ERROR_TYPE_ACTION,
7761 "old age and shared count combination is not supported");
7763 return rte_flow_error_set
7765 RTE_FLOW_ERROR_TYPE_ACTION,
7767 "old age action and count must be in the same sub flow");
7769 action_flags |= MLX5_FLOW_ACTION_AGE;
7772 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
7773 ret = flow_dv_validate_action_modify_ipv4_dscp
7780 /* Count all modify-header actions as one action. */
7781 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7783 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7784 modify_after_mirror = 1;
7785 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
7786 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
7788 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
7789 ret = flow_dv_validate_action_modify_ipv6_dscp
7796 /* Count all modify-header actions as one action. */
7797 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7799 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7800 modify_after_mirror = 1;
7801 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
7802 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
7804 case RTE_FLOW_ACTION_TYPE_SAMPLE:
7805 ret = flow_dv_validate_action_sample(&action_flags,
7814 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
7817 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
7818 ret = flow_dv_validate_action_modify_field(dev,
7825 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7826 modify_after_mirror = 1;
7827 /* Count all modify-header actions as one action. */
7828 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7830 action_flags |= MLX5_FLOW_ACTION_MODIFY_FIELD;
7833 case RTE_FLOW_ACTION_TYPE_CONNTRACK:
7834 ret = flow_dv_validate_action_aso_ct(dev, action_flags,
7839 action_flags |= MLX5_FLOW_ACTION_CT;
7841 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
7842 /* tunnel offload action was processed before
7843 * list it here as a supported type
7847 return rte_flow_error_set(error, ENOTSUP,
7848 RTE_FLOW_ERROR_TYPE_ACTION,
7850 "action not supported");
7854 * Validate actions in flow rules
7855 * - Explicit decap action is prohibited by the tunnel offload API.
7856 * - Drop action in tunnel steer rule is prohibited by the API.
7857 * - Application cannot use MARK action because it's value can mask
7858 * tunnel default miss notification.
7859 * - JUMP in tunnel match rule has no support in current PMD
7861 * - TAG & META are reserved for future uses.
7863 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_SET) {
7864 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_DECAP |
7865 MLX5_FLOW_ACTION_MARK |
7866 MLX5_FLOW_ACTION_SET_TAG |
7867 MLX5_FLOW_ACTION_SET_META |
7868 MLX5_FLOW_ACTION_DROP;
7870 if (action_flags & bad_actions_mask)
7871 return rte_flow_error_set
7873 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7874 "Invalid RTE action in tunnel "
7876 if (!(action_flags & MLX5_FLOW_ACTION_JUMP))
7877 return rte_flow_error_set
7879 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7880 "tunnel set decap rule must terminate "
7883 return rte_flow_error_set
7885 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7886 "tunnel flows for ingress traffic only");
7888 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_MATCH) {
7889 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_JUMP |
7890 MLX5_FLOW_ACTION_MARK |
7891 MLX5_FLOW_ACTION_SET_TAG |
7892 MLX5_FLOW_ACTION_SET_META;
7894 if (action_flags & bad_actions_mask)
7895 return rte_flow_error_set
7897 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7898 "Invalid RTE action in tunnel "
7902 * Validate the drop action mutual exclusion with other actions.
7903 * Drop action is mutually-exclusive with any other action, except for
7905 * Drop action compatibility with tunnel offload was already validated.
7907 if (action_flags & (MLX5_FLOW_ACTION_TUNNEL_MATCH |
7908 MLX5_FLOW_ACTION_TUNNEL_MATCH));
7909 else if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
7910 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
7911 return rte_flow_error_set(error, EINVAL,
7912 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7913 "Drop action is mutually-exclusive "
7914 "with any other action, except for "
7916 /* Eswitch has few restrictions on using items and actions */
7917 if (attr->transfer) {
7918 if (!mlx5_flow_ext_mreg_supported(dev) &&
7919 action_flags & MLX5_FLOW_ACTION_FLAG)
7920 return rte_flow_error_set(error, ENOTSUP,
7921 RTE_FLOW_ERROR_TYPE_ACTION,
7923 "unsupported action FLAG");
7924 if (!mlx5_flow_ext_mreg_supported(dev) &&
7925 action_flags & MLX5_FLOW_ACTION_MARK)
7926 return rte_flow_error_set(error, ENOTSUP,
7927 RTE_FLOW_ERROR_TYPE_ACTION,
7929 "unsupported action MARK");
7930 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
7931 return rte_flow_error_set(error, ENOTSUP,
7932 RTE_FLOW_ERROR_TYPE_ACTION,
7934 "unsupported action QUEUE");
7935 if (action_flags & MLX5_FLOW_ACTION_RSS)
7936 return rte_flow_error_set(error, ENOTSUP,
7937 RTE_FLOW_ERROR_TYPE_ACTION,
7939 "unsupported action RSS");
7940 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
7941 return rte_flow_error_set(error, EINVAL,
7942 RTE_FLOW_ERROR_TYPE_ACTION,
7944 "no fate action is found");
7946 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
7947 return rte_flow_error_set(error, EINVAL,
7948 RTE_FLOW_ERROR_TYPE_ACTION,
7950 "no fate action is found");
7953 * Continue validation for Xcap and VLAN actions.
7954 * If hairpin is working in explicit TX rule mode, there is no actions
7955 * splitting and the validation of hairpin ingress flow should be the
7956 * same as other standard flows.
7958 if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
7959 MLX5_FLOW_VLAN_ACTIONS)) &&
7960 (queue_index == 0xFFFF ||
7961 mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN ||
7962 ((conf = mlx5_rxq_get_hairpin_conf(dev, queue_index)) != NULL &&
7963 conf->tx_explicit != 0))) {
7964 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
7965 MLX5_FLOW_XCAP_ACTIONS)
7966 return rte_flow_error_set(error, ENOTSUP,
7967 RTE_FLOW_ERROR_TYPE_ACTION,
7968 NULL, "encap and decap "
7969 "combination aren't supported");
7970 if (!attr->transfer && attr->ingress) {
7971 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
7972 return rte_flow_error_set
7974 RTE_FLOW_ERROR_TYPE_ACTION,
7975 NULL, "encap is not supported"
7976 " for ingress traffic");
7977 else if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
7978 return rte_flow_error_set
7980 RTE_FLOW_ERROR_TYPE_ACTION,
7981 NULL, "push VLAN action not "
7982 "supported for ingress");
7983 else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
7984 MLX5_FLOW_VLAN_ACTIONS)
7985 return rte_flow_error_set
7987 RTE_FLOW_ERROR_TYPE_ACTION,
7988 NULL, "no support for "
7989 "multiple VLAN actions");
7992 if (action_flags & MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY) {
7993 if ((action_flags & (MLX5_FLOW_FATE_ACTIONS &
7994 ~MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY)) &&
7996 return rte_flow_error_set
7998 RTE_FLOW_ERROR_TYPE_ACTION,
7999 NULL, "fate action not supported for "
8000 "meter with policy");
8002 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
8003 return rte_flow_error_set
8005 RTE_FLOW_ERROR_TYPE_ACTION,
8006 NULL, "modify header action in egress "
8007 "cannot be done before meter action");
8008 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
8009 return rte_flow_error_set
8011 RTE_FLOW_ERROR_TYPE_ACTION,
8012 NULL, "encap action in egress "
8013 "cannot be done before meter action");
8014 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
8015 return rte_flow_error_set
8017 RTE_FLOW_ERROR_TYPE_ACTION,
8018 NULL, "push vlan action in egress "
8019 "cannot be done before meter action");
8023 * Hairpin flow will add one more TAG action in TX implicit mode.
8024 * In TX explicit mode, there will be no hairpin flow ID.
8027 rw_act_num += MLX5_ACT_NUM_SET_TAG;
8028 /* extra metadata enabled: one more TAG action will be add. */
8029 if (dev_conf->dv_flow_en &&
8030 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
8031 mlx5_flow_ext_mreg_supported(dev))
8032 rw_act_num += MLX5_ACT_NUM_SET_TAG;
8034 flow_dv_modify_hdr_action_max(dev, is_root)) {
8035 return rte_flow_error_set(error, ENOTSUP,
8036 RTE_FLOW_ERROR_TYPE_ACTION,
8037 NULL, "too many header modify"
8038 " actions to support");
8040 /* Eswitch egress mirror and modify flow has limitation on CX5 */
8041 if (fdb_mirror_limit && modify_after_mirror)
8042 return rte_flow_error_set(error, EINVAL,
8043 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
8044 "sample before modify action is not supported");
8049 * Internal preparation function. Allocates the DV flow size,
8050 * this size is constant.
8053 * Pointer to the rte_eth_dev structure.
8055 * Pointer to the flow attributes.
8057 * Pointer to the list of items.
8058 * @param[in] actions
8059 * Pointer to the list of actions.
8061 * Pointer to the error structure.
8064 * Pointer to mlx5_flow object on success,
8065 * otherwise NULL and rte_errno is set.
8067 static struct mlx5_flow *
8068 flow_dv_prepare(struct rte_eth_dev *dev,
8069 const struct rte_flow_attr *attr __rte_unused,
8070 const struct rte_flow_item items[] __rte_unused,
8071 const struct rte_flow_action actions[] __rte_unused,
8072 struct rte_flow_error *error)
8074 uint32_t handle_idx = 0;
8075 struct mlx5_flow *dev_flow;
8076 struct mlx5_flow_handle *dev_handle;
8077 struct mlx5_priv *priv = dev->data->dev_private;
8078 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
8081 wks->skip_matcher_reg = 0;
8083 wks->final_policy = NULL;
8084 /* In case of corrupting the memory. */
8085 if (wks->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
8086 rte_flow_error_set(error, ENOSPC,
8087 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8088 "not free temporary device flow");
8091 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8094 rte_flow_error_set(error, ENOMEM,
8095 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8096 "not enough memory to create flow handle");
8099 MLX5_ASSERT(wks->flow_idx < RTE_DIM(wks->flows));
8100 dev_flow = &wks->flows[wks->flow_idx++];
8101 memset(dev_flow, 0, sizeof(*dev_flow));
8102 dev_flow->handle = dev_handle;
8103 dev_flow->handle_idx = handle_idx;
8104 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
8105 dev_flow->ingress = attr->ingress;
8106 dev_flow->dv.transfer = attr->transfer;
8110 #ifdef RTE_LIBRTE_MLX5_DEBUG
8112 * Sanity check for match mask and value. Similar to check_valid_spec() in
8113 * kernel driver. If unmasked bit is present in value, it returns failure.
8116 * pointer to match mask buffer.
8117 * @param match_value
8118 * pointer to match value buffer.
8121 * 0 if valid, -EINVAL otherwise.
8124 flow_dv_check_valid_spec(void *match_mask, void *match_value)
8126 uint8_t *m = match_mask;
8127 uint8_t *v = match_value;
8130 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
8133 "match_value differs from match_criteria"
8134 " %p[%u] != %p[%u]",
8135 match_value, i, match_mask, i);
8144 * Add match of ip_version.
8148 * @param[in] headers_v
8149 * Values header pointer.
8150 * @param[in] headers_m
8151 * Masks header pointer.
8152 * @param[in] ip_version
8153 * The IP version to set.
8156 flow_dv_set_match_ip_version(uint32_t group,
8162 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
8164 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
8166 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
8167 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
8168 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
8172 * Add Ethernet item to matcher and to the value.
8174 * @param[in, out] matcher
8176 * @param[in, out] key
8177 * Flow matcher value.
8179 * Flow pattern to translate.
8181 * Item is inner pattern.
8184 flow_dv_translate_item_eth(void *matcher, void *key,
8185 const struct rte_flow_item *item, int inner,
8188 const struct rte_flow_item_eth *eth_m = item->mask;
8189 const struct rte_flow_item_eth *eth_v = item->spec;
8190 const struct rte_flow_item_eth nic_mask = {
8191 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
8192 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
8193 .type = RTE_BE16(0xffff),
8206 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
8208 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8210 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
8212 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8214 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, dmac_47_16),
8215 ð_m->dst, sizeof(eth_m->dst));
8216 /* The value must be in the range of the mask. */
8217 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, dmac_47_16);
8218 for (i = 0; i < sizeof(eth_m->dst); ++i)
8219 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
8220 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, smac_47_16),
8221 ð_m->src, sizeof(eth_m->src));
8222 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, smac_47_16);
8223 /* The value must be in the range of the mask. */
8224 for (i = 0; i < sizeof(eth_m->dst); ++i)
8225 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
8227 * HW supports match on one Ethertype, the Ethertype following the last
8228 * VLAN tag of the packet (see PRM).
8229 * Set match on ethertype only if ETH header is not followed by VLAN.
8230 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
8231 * ethertype, and use ip_version field instead.
8232 * eCPRI over Ether layer will use type value 0xAEFE.
8234 if (eth_m->type == 0xFFFF) {
8235 /* Set cvlan_tag mask for any single\multi\un-tagged case. */
8236 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
8237 switch (eth_v->type) {
8238 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
8239 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
8241 case RTE_BE16(RTE_ETHER_TYPE_QINQ):
8242 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
8243 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
8245 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
8246 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
8248 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
8249 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
8255 if (eth_m->has_vlan) {
8256 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
8257 if (eth_v->has_vlan) {
8259 * Here, when also has_more_vlan field in VLAN item is
8260 * not set, only single-tagged packets will be matched.
8262 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
8266 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
8267 rte_be_to_cpu_16(eth_m->type));
8268 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, ethertype);
8269 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
8273 * Add VLAN item to matcher and to the value.
8275 * @param[in, out] dev_flow
8277 * @param[in, out] matcher
8279 * @param[in, out] key
8280 * Flow matcher value.
8282 * Flow pattern to translate.
8284 * Item is inner pattern.
8287 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
8288 void *matcher, void *key,
8289 const struct rte_flow_item *item,
8290 int inner, uint32_t group)
8292 const struct rte_flow_item_vlan *vlan_m = item->mask;
8293 const struct rte_flow_item_vlan *vlan_v = item->spec;
8300 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
8302 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8304 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
8306 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8308 * This is workaround, masks are not supported,
8309 * and pre-validated.
8312 dev_flow->handle->vf_vlan.tag =
8313 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
8316 * When VLAN item exists in flow, mark packet as tagged,
8317 * even if TCI is not specified.
8319 if (!MLX5_GET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag)) {
8320 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
8321 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
8326 vlan_m = &rte_flow_item_vlan_mask;
8327 tci_m = rte_be_to_cpu_16(vlan_m->tci);
8328 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
8329 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_vid, tci_m);
8330 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_vid, tci_v);
8331 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_cfi, tci_m >> 12);
8332 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_cfi, tci_v >> 12);
8333 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_prio, tci_m >> 13);
8334 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_prio, tci_v >> 13);
8336 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
8337 * ethertype, and use ip_version field instead.
8339 if (vlan_m->inner_type == 0xFFFF) {
8340 switch (vlan_v->inner_type) {
8341 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
8342 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
8343 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
8344 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
8346 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
8347 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
8349 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
8350 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
8356 if (vlan_m->has_more_vlan && vlan_v->has_more_vlan) {
8357 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
8358 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
8359 /* Only one vlan_tag bit can be set. */
8360 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
8363 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
8364 rte_be_to_cpu_16(vlan_m->inner_type));
8365 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, ethertype,
8366 rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
8370 * Add IPV4 item to matcher and to the value.
8372 * @param[in, out] matcher
8374 * @param[in, out] key
8375 * Flow matcher value.
8377 * Flow pattern to translate.
8379 * Item is inner pattern.
8381 * The group to insert the rule.
8384 flow_dv_translate_item_ipv4(void *matcher, void *key,
8385 const struct rte_flow_item *item,
8386 int inner, uint32_t group)
8388 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
8389 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
8390 const struct rte_flow_item_ipv4 nic_mask = {
8392 .src_addr = RTE_BE32(0xffffffff),
8393 .dst_addr = RTE_BE32(0xffffffff),
8394 .type_of_service = 0xff,
8395 .next_proto_id = 0xff,
8396 .time_to_live = 0xff,
8403 uint8_t tos, ihl_m, ihl_v;
8406 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8408 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8410 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8412 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8414 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
8419 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
8420 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
8421 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
8422 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
8423 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
8424 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
8425 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
8426 src_ipv4_src_ipv6.ipv4_layout.ipv4);
8427 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
8428 src_ipv4_src_ipv6.ipv4_layout.ipv4);
8429 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
8430 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
8431 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
8432 ihl_m = ipv4_m->hdr.version_ihl & RTE_IPV4_HDR_IHL_MASK;
8433 ihl_v = ipv4_v->hdr.version_ihl & RTE_IPV4_HDR_IHL_MASK;
8434 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ipv4_ihl, ihl_m);
8435 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ipv4_ihl, ihl_m & ihl_v);
8436 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
8437 ipv4_m->hdr.type_of_service);
8438 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
8439 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
8440 ipv4_m->hdr.type_of_service >> 2);
8441 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
8442 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
8443 ipv4_m->hdr.next_proto_id);
8444 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
8445 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
8446 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
8447 ipv4_m->hdr.time_to_live);
8448 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
8449 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
8450 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
8451 !!(ipv4_m->hdr.fragment_offset));
8452 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
8453 !!(ipv4_v->hdr.fragment_offset & ipv4_m->hdr.fragment_offset));
8457 * Add IPV6 item to matcher and to the value.
8459 * @param[in, out] matcher
8461 * @param[in, out] key
8462 * Flow matcher value.
8464 * Flow pattern to translate.
8466 * Item is inner pattern.
8468 * The group to insert the rule.
8471 flow_dv_translate_item_ipv6(void *matcher, void *key,
8472 const struct rte_flow_item *item,
8473 int inner, uint32_t group)
8475 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
8476 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
8477 const struct rte_flow_item_ipv6 nic_mask = {
8480 "\xff\xff\xff\xff\xff\xff\xff\xff"
8481 "\xff\xff\xff\xff\xff\xff\xff\xff",
8483 "\xff\xff\xff\xff\xff\xff\xff\xff"
8484 "\xff\xff\xff\xff\xff\xff\xff\xff",
8485 .vtc_flow = RTE_BE32(0xffffffff),
8492 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8493 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8502 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8504 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8506 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8508 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8510 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
8515 size = sizeof(ipv6_m->hdr.dst_addr);
8516 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
8517 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
8518 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
8519 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
8520 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
8521 for (i = 0; i < size; ++i)
8522 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
8523 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
8524 src_ipv4_src_ipv6.ipv6_layout.ipv6);
8525 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
8526 src_ipv4_src_ipv6.ipv6_layout.ipv6);
8527 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
8528 for (i = 0; i < size; ++i)
8529 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
8531 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
8532 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
8533 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
8534 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
8535 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
8536 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
8539 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
8541 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
8544 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
8546 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
8550 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
8552 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
8553 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
8555 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
8556 ipv6_m->hdr.hop_limits);
8557 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
8558 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
8559 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
8560 !!(ipv6_m->has_frag_ext));
8561 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
8562 !!(ipv6_v->has_frag_ext & ipv6_m->has_frag_ext));
8566 * Add IPV6 fragment extension item to matcher and to the value.
8568 * @param[in, out] matcher
8570 * @param[in, out] key
8571 * Flow matcher value.
8573 * Flow pattern to translate.
8575 * Item is inner pattern.
8578 flow_dv_translate_item_ipv6_frag_ext(void *matcher, void *key,
8579 const struct rte_flow_item *item,
8582 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_m = item->mask;
8583 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_v = item->spec;
8584 const struct rte_flow_item_ipv6_frag_ext nic_mask = {
8586 .next_header = 0xff,
8587 .frag_data = RTE_BE16(0xffff),
8594 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8596 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8598 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8600 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8602 /* IPv6 fragment extension item exists, so packet is IP fragment. */
8603 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
8604 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 1);
8605 if (!ipv6_frag_ext_v)
8607 if (!ipv6_frag_ext_m)
8608 ipv6_frag_ext_m = &nic_mask;
8609 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
8610 ipv6_frag_ext_m->hdr.next_header);
8611 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
8612 ipv6_frag_ext_v->hdr.next_header &
8613 ipv6_frag_ext_m->hdr.next_header);
8617 * Add TCP item to matcher and to the value.
8619 * @param[in, out] matcher
8621 * @param[in, out] key
8622 * Flow matcher value.
8624 * Flow pattern to translate.
8626 * Item is inner pattern.
8629 flow_dv_translate_item_tcp(void *matcher, void *key,
8630 const struct rte_flow_item *item,
8633 const struct rte_flow_item_tcp *tcp_m = item->mask;
8634 const struct rte_flow_item_tcp *tcp_v = item->spec;
8639 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8641 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8643 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8645 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8647 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
8648 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
8652 tcp_m = &rte_flow_item_tcp_mask;
8653 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
8654 rte_be_to_cpu_16(tcp_m->hdr.src_port));
8655 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
8656 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
8657 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
8658 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
8659 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
8660 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
8661 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
8662 tcp_m->hdr.tcp_flags);
8663 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
8664 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
8668 * Add UDP item to matcher and to the value.
8670 * @param[in, out] matcher
8672 * @param[in, out] key
8673 * Flow matcher value.
8675 * Flow pattern to translate.
8677 * Item is inner pattern.
8680 flow_dv_translate_item_udp(void *matcher, void *key,
8681 const struct rte_flow_item *item,
8684 const struct rte_flow_item_udp *udp_m = item->mask;
8685 const struct rte_flow_item_udp *udp_v = item->spec;
8690 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8692 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8694 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8696 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8698 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
8699 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
8703 udp_m = &rte_flow_item_udp_mask;
8704 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
8705 rte_be_to_cpu_16(udp_m->hdr.src_port));
8706 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
8707 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
8708 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
8709 rte_be_to_cpu_16(udp_m->hdr.dst_port));
8710 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
8711 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
8715 * Add GRE optional Key item to matcher and to the value.
8717 * @param[in, out] matcher
8719 * @param[in, out] key
8720 * Flow matcher value.
8722 * Flow pattern to translate.
8724 * Item is inner pattern.
8727 flow_dv_translate_item_gre_key(void *matcher, void *key,
8728 const struct rte_flow_item *item)
8730 const rte_be32_t *key_m = item->mask;
8731 const rte_be32_t *key_v = item->spec;
8732 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8733 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8734 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
8736 /* GRE K bit must be on and should already be validated */
8737 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
8738 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
8742 key_m = &gre_key_default_mask;
8743 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
8744 rte_be_to_cpu_32(*key_m) >> 8);
8745 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
8746 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
8747 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
8748 rte_be_to_cpu_32(*key_m) & 0xFF);
8749 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
8750 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
8754 * Add GRE item to matcher and to the value.
8756 * @param[in, out] matcher
8758 * @param[in, out] key
8759 * Flow matcher value.
8761 * Flow pattern to translate.
8762 * @param[in] pattern_flags
8763 * Accumulated pattern flags.
8766 flow_dv_translate_item_gre(void *matcher, void *key,
8767 const struct rte_flow_item *item,
8768 uint64_t pattern_flags)
8770 static const struct rte_flow_item_gre empty_gre = {0,};
8771 const struct rte_flow_item_gre *gre_m = item->mask;
8772 const struct rte_flow_item_gre *gre_v = item->spec;
8773 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
8774 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8775 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8776 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8783 uint16_t s_present:1;
8784 uint16_t k_present:1;
8785 uint16_t rsvd_bit1:1;
8786 uint16_t c_present:1;
8790 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
8791 uint16_t protocol_m, protocol_v;
8793 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
8794 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
8800 gre_m = &rte_flow_item_gre_mask;
8802 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
8803 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
8804 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
8805 gre_crks_rsvd0_ver_m.c_present);
8806 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
8807 gre_crks_rsvd0_ver_v.c_present &
8808 gre_crks_rsvd0_ver_m.c_present);
8809 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
8810 gre_crks_rsvd0_ver_m.k_present);
8811 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
8812 gre_crks_rsvd0_ver_v.k_present &
8813 gre_crks_rsvd0_ver_m.k_present);
8814 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
8815 gre_crks_rsvd0_ver_m.s_present);
8816 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
8817 gre_crks_rsvd0_ver_v.s_present &
8818 gre_crks_rsvd0_ver_m.s_present);
8819 protocol_m = rte_be_to_cpu_16(gre_m->protocol);
8820 protocol_v = rte_be_to_cpu_16(gre_v->protocol);
8822 /* Force next protocol to prevent matchers duplication */
8823 protocol_v = mlx5_translate_tunnel_etypes(pattern_flags);
8825 protocol_m = 0xFFFF;
8827 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, protocol_m);
8828 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
8829 protocol_m & protocol_v);
8833 * Add NVGRE item to matcher and to the value.
8835 * @param[in, out] matcher
8837 * @param[in, out] key
8838 * Flow matcher value.
8840 * Flow pattern to translate.
8841 * @param[in] pattern_flags
8842 * Accumulated pattern flags.
8845 flow_dv_translate_item_nvgre(void *matcher, void *key,
8846 const struct rte_flow_item *item,
8847 unsigned long pattern_flags)
8849 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
8850 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
8851 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8852 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8853 const char *tni_flow_id_m;
8854 const char *tni_flow_id_v;
8860 /* For NVGRE, GRE header fields must be set with defined values. */
8861 const struct rte_flow_item_gre gre_spec = {
8862 .c_rsvd0_ver = RTE_BE16(0x2000),
8863 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
8865 const struct rte_flow_item_gre gre_mask = {
8866 .c_rsvd0_ver = RTE_BE16(0xB000),
8867 .protocol = RTE_BE16(UINT16_MAX),
8869 const struct rte_flow_item gre_item = {
8874 flow_dv_translate_item_gre(matcher, key, &gre_item, pattern_flags);
8878 nvgre_m = &rte_flow_item_nvgre_mask;
8879 tni_flow_id_m = (const char *)nvgre_m->tni;
8880 tni_flow_id_v = (const char *)nvgre_v->tni;
8881 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
8882 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
8883 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
8884 memcpy(gre_key_m, tni_flow_id_m, size);
8885 for (i = 0; i < size; ++i)
8886 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
8890 * Add VXLAN item to matcher and to the value.
8893 * Pointer to the Ethernet device structure.
8895 * Flow rule attributes.
8896 * @param[in, out] matcher
8898 * @param[in, out] key
8899 * Flow matcher value.
8901 * Flow pattern to translate.
8903 * Item is inner pattern.
8906 flow_dv_translate_item_vxlan(struct rte_eth_dev *dev,
8907 const struct rte_flow_attr *attr,
8908 void *matcher, void *key,
8909 const struct rte_flow_item *item,
8912 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
8913 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
8918 uint32_t *tunnel_header_v;
8919 uint32_t *tunnel_header_m;
8921 struct mlx5_priv *priv = dev->data->dev_private;
8922 const struct rte_flow_item_vxlan nic_mask = {
8923 .vni = "\xff\xff\xff",
8928 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8930 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8932 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8934 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8936 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
8937 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
8938 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
8939 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
8940 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
8942 dport = MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport);
8946 if ((!attr->group && !priv->sh->tunnel_header_0_1) ||
8947 (attr->group && !priv->sh->misc5_cap))
8948 vxlan_m = &rte_flow_item_vxlan_mask;
8950 vxlan_m = &nic_mask;
8952 if ((priv->sh->steering_format_version ==
8953 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 &&
8954 dport != MLX5_UDP_PORT_VXLAN) ||
8955 (!attr->group && !attr->transfer && !priv->sh->tunnel_header_0_1) ||
8956 ((attr->group || attr->transfer) && !priv->sh->misc5_cap)) {
8963 misc_m = MLX5_ADDR_OF(fte_match_param,
8964 matcher, misc_parameters);
8965 misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8966 size = sizeof(vxlan_m->vni);
8967 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
8968 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
8969 memcpy(vni_m, vxlan_m->vni, size);
8970 for (i = 0; i < size; ++i)
8971 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
8974 misc5_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_5);
8975 misc5_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_5);
8976 tunnel_header_v = (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc5,
8979 tunnel_header_m = (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc5,
8982 *tunnel_header_v = (vxlan_v->vni[0] & vxlan_m->vni[0]) |
8983 (vxlan_v->vni[1] & vxlan_m->vni[1]) << 8 |
8984 (vxlan_v->vni[2] & vxlan_m->vni[2]) << 16;
8985 if (*tunnel_header_v)
8986 *tunnel_header_m = vxlan_m->vni[0] |
8987 vxlan_m->vni[1] << 8 |
8988 vxlan_m->vni[2] << 16;
8990 *tunnel_header_m = 0x0;
8991 *tunnel_header_v |= (vxlan_v->rsvd1 & vxlan_m->rsvd1) << 24;
8992 if (vxlan_v->rsvd1 & vxlan_m->rsvd1)
8993 *tunnel_header_m |= vxlan_m->rsvd1 << 24;
8997 * Add VXLAN-GPE item to matcher and to the value.
8999 * @param[in, out] matcher
9001 * @param[in, out] key
9002 * Flow matcher value.
9004 * Flow pattern to translate.
9006 * Item is inner pattern.
9010 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
9011 const struct rte_flow_item *item,
9012 const uint64_t pattern_flags)
9014 static const struct rte_flow_item_vxlan_gpe dummy_vxlan_gpe_hdr = {0, };
9015 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
9016 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
9017 /* The item was validated to be on the outer side */
9018 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
9019 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9021 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
9023 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9025 MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
9027 MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
9028 int i, size = sizeof(vxlan_m->vni);
9029 uint8_t flags_m = 0xff;
9030 uint8_t flags_v = 0xc;
9031 uint8_t m_protocol, v_protocol;
9033 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
9034 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
9035 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
9036 MLX5_UDP_PORT_VXLAN_GPE);
9039 vxlan_v = &dummy_vxlan_gpe_hdr;
9040 vxlan_m = &dummy_vxlan_gpe_hdr;
9043 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
9045 memcpy(vni_m, vxlan_m->vni, size);
9046 for (i = 0; i < size; ++i)
9047 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
9048 if (vxlan_m->flags) {
9049 flags_m = vxlan_m->flags;
9050 flags_v = vxlan_v->flags;
9052 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
9053 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
9054 m_protocol = vxlan_m->protocol;
9055 v_protocol = vxlan_v->protocol;
9057 /* Force next protocol to ensure next headers parsing. */
9058 if (pattern_flags & MLX5_FLOW_LAYER_INNER_L2)
9059 v_protocol = RTE_VXLAN_GPE_TYPE_ETH;
9060 else if (pattern_flags & MLX5_FLOW_LAYER_INNER_L3_IPV4)
9061 v_protocol = RTE_VXLAN_GPE_TYPE_IPV4;
9062 else if (pattern_flags & MLX5_FLOW_LAYER_INNER_L3_IPV6)
9063 v_protocol = RTE_VXLAN_GPE_TYPE_IPV6;
9067 MLX5_SET(fte_match_set_misc3, misc_m,
9068 outer_vxlan_gpe_next_protocol, m_protocol);
9069 MLX5_SET(fte_match_set_misc3, misc_v,
9070 outer_vxlan_gpe_next_protocol, m_protocol & v_protocol);
9074 * Add Geneve item to matcher and to the value.
9076 * @param[in, out] matcher
9078 * @param[in, out] key
9079 * Flow matcher value.
9081 * Flow pattern to translate.
9083 * Item is inner pattern.
9087 flow_dv_translate_item_geneve(void *matcher, void *key,
9088 const struct rte_flow_item *item,
9089 uint64_t pattern_flags)
9091 static const struct rte_flow_item_geneve empty_geneve = {0,};
9092 const struct rte_flow_item_geneve *geneve_m = item->mask;
9093 const struct rte_flow_item_geneve *geneve_v = item->spec;
9094 /* GENEVE flow item validation allows single tunnel item */
9095 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
9096 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9097 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9098 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9101 char *vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
9102 char *vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
9103 size_t size = sizeof(geneve_m->vni), i;
9104 uint16_t protocol_m, protocol_v;
9106 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
9107 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
9108 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
9109 MLX5_UDP_PORT_GENEVE);
9112 geneve_v = &empty_geneve;
9113 geneve_m = &empty_geneve;
9116 geneve_m = &rte_flow_item_geneve_mask;
9118 memcpy(vni_m, geneve_m->vni, size);
9119 for (i = 0; i < size; ++i)
9120 vni_v[i] = vni_m[i] & geneve_v->vni[i];
9121 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
9122 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
9123 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
9124 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
9125 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
9126 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
9127 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
9128 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
9129 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
9130 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
9131 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
9132 protocol_m = rte_be_to_cpu_16(geneve_m->protocol);
9133 protocol_v = rte_be_to_cpu_16(geneve_v->protocol);
9135 /* Force next protocol to prevent matchers duplication */
9136 protocol_v = mlx5_translate_tunnel_etypes(pattern_flags);
9138 protocol_m = 0xFFFF;
9140 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type, protocol_m);
9141 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
9142 protocol_m & protocol_v);
9146 * Create Geneve TLV option resource.
9148 * @param dev[in, out]
9149 * Pointer to rte_eth_dev structure.
9150 * @param[in, out] tag_be24
9151 * Tag value in big endian then R-shift 8.
9152 * @parm[in, out] dev_flow
9153 * Pointer to the dev_flow.
9155 * pointer to error structure.
9158 * 0 on success otherwise -errno and errno is set.
9162 flow_dev_geneve_tlv_option_resource_register(struct rte_eth_dev *dev,
9163 const struct rte_flow_item *item,
9164 struct rte_flow_error *error)
9166 struct mlx5_priv *priv = dev->data->dev_private;
9167 struct mlx5_dev_ctx_shared *sh = priv->sh;
9168 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource =
9169 sh->geneve_tlv_option_resource;
9170 struct mlx5_devx_obj *obj;
9171 const struct rte_flow_item_geneve_opt *geneve_opt_v = item->spec;
9176 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
9177 if (geneve_opt_resource != NULL) {
9178 if (geneve_opt_resource->option_class ==
9179 geneve_opt_v->option_class &&
9180 geneve_opt_resource->option_type ==
9181 geneve_opt_v->option_type &&
9182 geneve_opt_resource->length ==
9183 geneve_opt_v->option_len) {
9184 /* We already have GENEVE TLV option obj allocated. */
9185 __atomic_fetch_add(&geneve_opt_resource->refcnt, 1,
9188 ret = rte_flow_error_set(error, ENOMEM,
9189 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9190 "Only one GENEVE TLV option supported");
9194 /* Create a GENEVE TLV object and resource. */
9195 obj = mlx5_devx_cmd_create_geneve_tlv_option(sh->cdev->ctx,
9196 geneve_opt_v->option_class,
9197 geneve_opt_v->option_type,
9198 geneve_opt_v->option_len);
9200 ret = rte_flow_error_set(error, ENODATA,
9201 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9202 "Failed to create GENEVE TLV Devx object");
9205 sh->geneve_tlv_option_resource =
9206 mlx5_malloc(MLX5_MEM_ZERO,
9207 sizeof(*geneve_opt_resource),
9209 if (!sh->geneve_tlv_option_resource) {
9210 claim_zero(mlx5_devx_cmd_destroy(obj));
9211 ret = rte_flow_error_set(error, ENOMEM,
9212 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9213 "GENEVE TLV object memory allocation failed");
9216 geneve_opt_resource = sh->geneve_tlv_option_resource;
9217 geneve_opt_resource->obj = obj;
9218 geneve_opt_resource->option_class = geneve_opt_v->option_class;
9219 geneve_opt_resource->option_type = geneve_opt_v->option_type;
9220 geneve_opt_resource->length = geneve_opt_v->option_len;
9221 __atomic_store_n(&geneve_opt_resource->refcnt, 1,
9225 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
9230 * Add Geneve TLV option item to matcher.
9232 * @param[in, out] dev
9233 * Pointer to rte_eth_dev structure.
9234 * @param[in, out] matcher
9236 * @param[in, out] key
9237 * Flow matcher value.
9239 * Flow pattern to translate.
9241 * Pointer to error structure.
9244 flow_dv_translate_item_geneve_opt(struct rte_eth_dev *dev, void *matcher,
9245 void *key, const struct rte_flow_item *item,
9246 struct rte_flow_error *error)
9248 const struct rte_flow_item_geneve_opt *geneve_opt_m = item->mask;
9249 const struct rte_flow_item_geneve_opt *geneve_opt_v = item->spec;
9250 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9251 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9252 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
9254 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9255 rte_be32_t opt_data_key = 0, opt_data_mask = 0;
9261 geneve_opt_m = &rte_flow_item_geneve_opt_mask;
9262 ret = flow_dev_geneve_tlv_option_resource_register(dev, item,
9265 DRV_LOG(ERR, "Failed to create geneve_tlv_obj");
9269 * Set the option length in GENEVE header if not requested.
9270 * The GENEVE TLV option length is expressed by the option length field
9271 * in the GENEVE header.
9272 * If the option length was not requested but the GENEVE TLV option item
9273 * is present we set the option length field implicitly.
9275 if (!MLX5_GET16(fte_match_set_misc, misc_m, geneve_opt_len)) {
9276 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
9277 MLX5_GENEVE_OPTLEN_MASK);
9278 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
9279 geneve_opt_v->option_len + 1);
9281 MLX5_SET(fte_match_set_misc, misc_m, geneve_tlv_option_0_exist, 1);
9282 MLX5_SET(fte_match_set_misc, misc_v, geneve_tlv_option_0_exist, 1);
9284 if (geneve_opt_v->data) {
9285 memcpy(&opt_data_key, geneve_opt_v->data,
9286 RTE_MIN((uint32_t)(geneve_opt_v->option_len * 4),
9287 sizeof(opt_data_key)));
9288 MLX5_ASSERT((uint32_t)(geneve_opt_v->option_len * 4) <=
9289 sizeof(opt_data_key));
9290 memcpy(&opt_data_mask, geneve_opt_m->data,
9291 RTE_MIN((uint32_t)(geneve_opt_v->option_len * 4),
9292 sizeof(opt_data_mask)));
9293 MLX5_ASSERT((uint32_t)(geneve_opt_v->option_len * 4) <=
9294 sizeof(opt_data_mask));
9295 MLX5_SET(fte_match_set_misc3, misc3_m,
9296 geneve_tlv_option_0_data,
9297 rte_be_to_cpu_32(opt_data_mask));
9298 MLX5_SET(fte_match_set_misc3, misc3_v,
9299 geneve_tlv_option_0_data,
9300 rte_be_to_cpu_32(opt_data_key & opt_data_mask));
9306 * Add MPLS item to matcher and to the value.
9308 * @param[in, out] matcher
9310 * @param[in, out] key
9311 * Flow matcher value.
9313 * Flow pattern to translate.
9314 * @param[in] prev_layer
9315 * The protocol layer indicated in previous item.
9317 * Item is inner pattern.
9320 flow_dv_translate_item_mpls(void *matcher, void *key,
9321 const struct rte_flow_item *item,
9322 uint64_t prev_layer,
9325 const uint32_t *in_mpls_m = item->mask;
9326 const uint32_t *in_mpls_v = item->spec;
9327 uint32_t *out_mpls_m = 0;
9328 uint32_t *out_mpls_v = 0;
9329 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9330 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9331 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
9333 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
9334 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
9335 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9337 switch (prev_layer) {
9338 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
9339 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
9340 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
9342 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
9343 MLX5_UDP_PORT_MPLS);
9346 case MLX5_FLOW_LAYER_GRE:
9348 case MLX5_FLOW_LAYER_GRE_KEY:
9349 if (!MLX5_GET16(fte_match_set_misc, misc_v, gre_protocol)) {
9350 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
9352 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
9353 RTE_ETHER_TYPE_MPLS);
9362 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
9363 switch (prev_layer) {
9364 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
9366 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
9367 outer_first_mpls_over_udp);
9369 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
9370 outer_first_mpls_over_udp);
9372 case MLX5_FLOW_LAYER_GRE:
9374 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
9375 outer_first_mpls_over_gre);
9377 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
9378 outer_first_mpls_over_gre);
9381 /* Inner MPLS not over GRE is not supported. */
9384 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
9388 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
9394 if (out_mpls_m && out_mpls_v) {
9395 *out_mpls_m = *in_mpls_m;
9396 *out_mpls_v = *in_mpls_v & *in_mpls_m;
9401 * Add metadata register item to matcher
9403 * @param[in, out] matcher
9405 * @param[in, out] key
9406 * Flow matcher value.
9407 * @param[in] reg_type
9408 * Type of device metadata register
9415 flow_dv_match_meta_reg(void *matcher, void *key,
9416 enum modify_reg reg_type,
9417 uint32_t data, uint32_t mask)
9420 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
9422 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
9428 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
9429 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
9432 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
9433 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
9437 * The metadata register C0 field might be divided into
9438 * source vport index and META item value, we should set
9439 * this field according to specified mask, not as whole one.
9441 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
9443 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
9444 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
9447 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
9450 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
9451 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
9454 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
9455 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
9458 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
9459 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
9462 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
9463 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
9466 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
9467 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
9470 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
9471 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
9474 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
9475 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
9484 * Add MARK item to matcher
9487 * The device to configure through.
9488 * @param[in, out] matcher
9490 * @param[in, out] key
9491 * Flow matcher value.
9493 * Flow pattern to translate.
9496 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
9497 void *matcher, void *key,
9498 const struct rte_flow_item *item)
9500 struct mlx5_priv *priv = dev->data->dev_private;
9501 const struct rte_flow_item_mark *mark;
9505 mark = item->mask ? (const void *)item->mask :
9506 &rte_flow_item_mark_mask;
9507 mask = mark->id & priv->sh->dv_mark_mask;
9508 mark = (const void *)item->spec;
9510 value = mark->id & priv->sh->dv_mark_mask & mask;
9512 enum modify_reg reg;
9514 /* Get the metadata register index for the mark. */
9515 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
9516 MLX5_ASSERT(reg > 0);
9517 if (reg == REG_C_0) {
9518 struct mlx5_priv *priv = dev->data->dev_private;
9519 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
9520 uint32_t shl_c0 = rte_bsf32(msk_c0);
9526 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
9531 * Add META item to matcher
9534 * The devich to configure through.
9535 * @param[in, out] matcher
9537 * @param[in, out] key
9538 * Flow matcher value.
9540 * Attributes of flow that includes this item.
9542 * Flow pattern to translate.
9545 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
9546 void *matcher, void *key,
9547 const struct rte_flow_attr *attr,
9548 const struct rte_flow_item *item)
9550 const struct rte_flow_item_meta *meta_m;
9551 const struct rte_flow_item_meta *meta_v;
9553 meta_m = (const void *)item->mask;
9555 meta_m = &rte_flow_item_meta_mask;
9556 meta_v = (const void *)item->spec;
9559 uint32_t value = meta_v->data;
9560 uint32_t mask = meta_m->data;
9562 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
9565 MLX5_ASSERT(reg != REG_NON);
9566 if (reg == REG_C_0) {
9567 struct mlx5_priv *priv = dev->data->dev_private;
9568 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
9569 uint32_t shl_c0 = rte_bsf32(msk_c0);
9575 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
9580 * Add vport metadata Reg C0 item to matcher
9582 * @param[in, out] matcher
9584 * @param[in, out] key
9585 * Flow matcher value.
9587 * Flow pattern to translate.
9590 flow_dv_translate_item_meta_vport(void *matcher, void *key,
9591 uint32_t value, uint32_t mask)
9593 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
9597 * Add tag item to matcher
9600 * The devich to configure through.
9601 * @param[in, out] matcher
9603 * @param[in, out] key
9604 * Flow matcher value.
9606 * Flow pattern to translate.
9609 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
9610 void *matcher, void *key,
9611 const struct rte_flow_item *item)
9613 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
9614 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
9615 uint32_t mask, value;
9618 value = tag_v->data;
9619 mask = tag_m ? tag_m->data : UINT32_MAX;
9620 if (tag_v->id == REG_C_0) {
9621 struct mlx5_priv *priv = dev->data->dev_private;
9622 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
9623 uint32_t shl_c0 = rte_bsf32(msk_c0);
9629 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
9633 * Add TAG item to matcher
9636 * The devich to configure through.
9637 * @param[in, out] matcher
9639 * @param[in, out] key
9640 * Flow matcher value.
9642 * Flow pattern to translate.
9645 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
9646 void *matcher, void *key,
9647 const struct rte_flow_item *item)
9649 const struct rte_flow_item_tag *tag_v = item->spec;
9650 const struct rte_flow_item_tag *tag_m = item->mask;
9651 enum modify_reg reg;
9654 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
9655 /* Get the metadata register index for the tag. */
9656 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
9657 MLX5_ASSERT(reg > 0);
9658 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
9662 * Add source vport match to the specified matcher.
9664 * @param[in, out] matcher
9666 * @param[in, out] key
9667 * Flow matcher value.
9669 * Source vport value to match
9674 flow_dv_translate_item_source_vport(void *matcher, void *key,
9675 int16_t port, uint16_t mask)
9677 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9678 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9680 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
9681 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
9685 * Translate port-id item to eswitch match on port-id.
9688 * The devich to configure through.
9689 * @param[in, out] matcher
9691 * @param[in, out] key
9692 * Flow matcher value.
9694 * Flow pattern to translate.
9699 * 0 on success, a negative errno value otherwise.
9702 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
9703 void *key, const struct rte_flow_item *item,
9704 const struct rte_flow_attr *attr)
9706 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
9707 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
9708 struct mlx5_priv *priv;
9711 if (pid_v && pid_v->id == MLX5_PORT_ESW_MGR) {
9712 flow_dv_translate_item_source_vport(matcher, key,
9713 flow_dv_get_esw_manager_vport_id(dev), 0xffff);
9716 mask = pid_m ? pid_m->id : 0xffff;
9717 id = pid_v ? pid_v->id : dev->data->port_id;
9718 priv = mlx5_port_to_eswitch_info(id, item == NULL);
9722 * Translate to vport field or to metadata, depending on mode.
9723 * Kernel can use either misc.source_port or half of C0 metadata
9726 if (priv->vport_meta_mask) {
9728 * Provide the hint for SW steering library
9729 * to insert the flow into ingress domain and
9730 * save the extra vport match.
9732 if (mask == 0xffff && priv->vport_id == 0xffff &&
9733 priv->pf_bond < 0 && attr->transfer)
9734 flow_dv_translate_item_source_vport
9735 (matcher, key, priv->vport_id, mask);
9737 * We should always set the vport metadata register,
9738 * otherwise the SW steering library can drop
9739 * the rule if wire vport metadata value is not zero,
9740 * it depends on kernel configuration.
9742 flow_dv_translate_item_meta_vport(matcher, key,
9743 priv->vport_meta_tag,
9744 priv->vport_meta_mask);
9746 flow_dv_translate_item_source_vport(matcher, key,
9747 priv->vport_id, mask);
9753 * Add ICMP6 item to matcher and to the value.
9755 * @param[in, out] matcher
9757 * @param[in, out] key
9758 * Flow matcher value.
9760 * Flow pattern to translate.
9762 * Item is inner pattern.
9765 flow_dv_translate_item_icmp6(void *matcher, void *key,
9766 const struct rte_flow_item *item,
9769 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
9770 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
9773 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
9775 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9777 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9779 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
9781 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9783 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9785 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
9786 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
9790 icmp6_m = &rte_flow_item_icmp6_mask;
9791 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
9792 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
9793 icmp6_v->type & icmp6_m->type);
9794 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
9795 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
9796 icmp6_v->code & icmp6_m->code);
9800 * Add ICMP item to matcher and to the value.
9802 * @param[in, out] matcher
9804 * @param[in, out] key
9805 * Flow matcher value.
9807 * Flow pattern to translate.
9809 * Item is inner pattern.
9812 flow_dv_translate_item_icmp(void *matcher, void *key,
9813 const struct rte_flow_item *item,
9816 const struct rte_flow_item_icmp *icmp_m = item->mask;
9817 const struct rte_flow_item_icmp *icmp_v = item->spec;
9818 uint32_t icmp_header_data_m = 0;
9819 uint32_t icmp_header_data_v = 0;
9822 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
9824 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9826 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9828 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
9830 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9832 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9834 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
9835 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
9839 icmp_m = &rte_flow_item_icmp_mask;
9840 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
9841 icmp_m->hdr.icmp_type);
9842 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
9843 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
9844 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
9845 icmp_m->hdr.icmp_code);
9846 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
9847 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
9848 icmp_header_data_m = rte_be_to_cpu_16(icmp_m->hdr.icmp_seq_nb);
9849 icmp_header_data_m |= rte_be_to_cpu_16(icmp_m->hdr.icmp_ident) << 16;
9850 if (icmp_header_data_m) {
9851 icmp_header_data_v = rte_be_to_cpu_16(icmp_v->hdr.icmp_seq_nb);
9852 icmp_header_data_v |=
9853 rte_be_to_cpu_16(icmp_v->hdr.icmp_ident) << 16;
9854 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_header_data,
9855 icmp_header_data_m);
9856 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_header_data,
9857 icmp_header_data_v & icmp_header_data_m);
9862 * Add GTP item to matcher and to the value.
9864 * @param[in, out] matcher
9866 * @param[in, out] key
9867 * Flow matcher value.
9869 * Flow pattern to translate.
9871 * Item is inner pattern.
9874 flow_dv_translate_item_gtp(void *matcher, void *key,
9875 const struct rte_flow_item *item, int inner)
9877 const struct rte_flow_item_gtp *gtp_m = item->mask;
9878 const struct rte_flow_item_gtp *gtp_v = item->spec;
9881 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
9883 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9884 uint16_t dport = RTE_GTPU_UDP_PORT;
9887 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9889 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
9891 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9893 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9895 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
9896 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
9897 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
9902 gtp_m = &rte_flow_item_gtp_mask;
9903 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
9904 gtp_m->v_pt_rsv_flags);
9905 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
9906 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
9907 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
9908 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
9909 gtp_v->msg_type & gtp_m->msg_type);
9910 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
9911 rte_be_to_cpu_32(gtp_m->teid));
9912 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
9913 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
9917 * Add GTP PSC item to matcher.
9919 * @param[in, out] matcher
9921 * @param[in, out] key
9922 * Flow matcher value.
9924 * Flow pattern to translate.
9927 flow_dv_translate_item_gtp_psc(void *matcher, void *key,
9928 const struct rte_flow_item *item)
9930 const struct rte_flow_item_gtp_psc *gtp_psc_m = item->mask;
9931 const struct rte_flow_item_gtp_psc *gtp_psc_v = item->spec;
9932 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
9934 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9940 uint8_t next_ext_header_type;
9945 /* Always set E-flag match on one, regardless of GTP item settings. */
9946 gtp_flags = MLX5_GET(fte_match_set_misc3, misc3_m, gtpu_msg_flags);
9947 gtp_flags |= MLX5_GTP_EXT_HEADER_FLAG;
9948 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags, gtp_flags);
9949 gtp_flags = MLX5_GET(fte_match_set_misc3, misc3_v, gtpu_msg_flags);
9950 gtp_flags |= MLX5_GTP_EXT_HEADER_FLAG;
9951 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags, gtp_flags);
9952 /*Set next extension header type. */
9955 dw_2.next_ext_header_type = 0xff;
9956 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_dw_2,
9957 rte_cpu_to_be_32(dw_2.w32));
9960 dw_2.next_ext_header_type = 0x85;
9961 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_dw_2,
9962 rte_cpu_to_be_32(dw_2.w32));
9974 /*Set extension header PDU type and Qos. */
9976 gtp_psc_m = &rte_flow_item_gtp_psc_mask;
9978 dw_0.type_flags = MLX5_GTP_PDU_TYPE_SHIFT(gtp_psc_m->hdr.type);
9979 dw_0.qfi = gtp_psc_m->hdr.qfi;
9980 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_first_ext_dw_0,
9981 rte_cpu_to_be_32(dw_0.w32));
9983 dw_0.type_flags = MLX5_GTP_PDU_TYPE_SHIFT(gtp_psc_v->hdr.type &
9984 gtp_psc_m->hdr.type);
9985 dw_0.qfi = gtp_psc_v->hdr.qfi & gtp_psc_m->hdr.qfi;
9986 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_first_ext_dw_0,
9987 rte_cpu_to_be_32(dw_0.w32));
9993 * Add eCPRI item to matcher and to the value.
9996 * The devich to configure through.
9997 * @param[in, out] matcher
9999 * @param[in, out] key
10000 * Flow matcher value.
10002 * Flow pattern to translate.
10003 * @param[in] last_item
10007 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
10008 void *key, const struct rte_flow_item *item,
10009 uint64_t last_item)
10011 struct mlx5_priv *priv = dev->data->dev_private;
10012 const struct rte_flow_item_ecpri *ecpri_m = item->mask;
10013 const struct rte_flow_item_ecpri *ecpri_v = item->spec;
10014 struct rte_ecpri_common_hdr common;
10015 void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
10016 misc_parameters_4);
10017 void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
10023 * In case of eCPRI over Ethernet, if EtherType is not specified,
10024 * match on eCPRI EtherType implicitly.
10026 if (last_item & MLX5_FLOW_LAYER_OUTER_L2) {
10027 void *hdrs_m, *hdrs_v, *l2m, *l2v;
10029 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
10030 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
10031 l2m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, ethertype);
10032 l2v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, ethertype);
10033 if (*(uint16_t *)l2m == 0 && *(uint16_t *)l2v == 0) {
10034 *(uint16_t *)l2m = UINT16_MAX;
10035 *(uint16_t *)l2v = RTE_BE16(RTE_ETHER_TYPE_ECPRI);
10041 ecpri_m = &rte_flow_item_ecpri_mask;
10043 * Maximal four DW samples are supported in a single matching now.
10044 * Two are used now for a eCPRI matching:
10045 * 1. Type: one byte, mask should be 0x00ff0000 in network order
10046 * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
10049 if (!ecpri_m->hdr.common.u32)
10051 samples = priv->sh->ecpri_parser.ids;
10052 /* Need to take the whole DW as the mask to fill the entry. */
10053 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
10054 prog_sample_field_value_0);
10055 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
10056 prog_sample_field_value_0);
10057 /* Already big endian (network order) in the header. */
10058 *(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
10059 *(uint32_t *)dw_v = ecpri_v->hdr.common.u32 & ecpri_m->hdr.common.u32;
10060 /* Sample#0, used for matching type, offset 0. */
10061 MLX5_SET(fte_match_set_misc4, misc4_m,
10062 prog_sample_field_id_0, samples[0]);
10063 /* It makes no sense to set the sample ID in the mask field. */
10064 MLX5_SET(fte_match_set_misc4, misc4_v,
10065 prog_sample_field_id_0, samples[0]);
10067 * Checking if message body part needs to be matched.
10068 * Some wildcard rules only matching type field should be supported.
10070 if (ecpri_m->hdr.dummy[0]) {
10071 common.u32 = rte_be_to_cpu_32(ecpri_v->hdr.common.u32);
10072 switch (common.type) {
10073 case RTE_ECPRI_MSG_TYPE_IQ_DATA:
10074 case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
10075 case RTE_ECPRI_MSG_TYPE_DLY_MSR:
10076 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
10077 prog_sample_field_value_1);
10078 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
10079 prog_sample_field_value_1);
10080 *(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
10081 *(uint32_t *)dw_v = ecpri_v->hdr.dummy[0] &
10082 ecpri_m->hdr.dummy[0];
10083 /* Sample#1, to match message body, offset 4. */
10084 MLX5_SET(fte_match_set_misc4, misc4_m,
10085 prog_sample_field_id_1, samples[1]);
10086 MLX5_SET(fte_match_set_misc4, misc4_v,
10087 prog_sample_field_id_1, samples[1]);
10090 /* Others, do not match any sample ID. */
10097 * Add connection tracking status item to matcher
10100 * The devich to configure through.
10101 * @param[in, out] matcher
10103 * @param[in, out] key
10104 * Flow matcher value.
10106 * Flow pattern to translate.
10109 flow_dv_translate_item_aso_ct(struct rte_eth_dev *dev,
10110 void *matcher, void *key,
10111 const struct rte_flow_item *item)
10113 uint32_t reg_value = 0;
10115 /* 8LSB 0b 11/0000/11, middle 4 bits are reserved. */
10116 uint32_t reg_mask = 0;
10117 const struct rte_flow_item_conntrack *spec = item->spec;
10118 const struct rte_flow_item_conntrack *mask = item->mask;
10120 struct rte_flow_error error;
10123 mask = &rte_flow_item_conntrack_mask;
10124 if (!spec || !mask->flags)
10126 flags = spec->flags & mask->flags;
10127 /* The conflict should be checked in the validation. */
10128 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_VALID)
10129 reg_value |= MLX5_CT_SYNDROME_VALID;
10130 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_CHANGED)
10131 reg_value |= MLX5_CT_SYNDROME_STATE_CHANGE;
10132 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_INVALID)
10133 reg_value |= MLX5_CT_SYNDROME_INVALID;
10134 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_DISABLED)
10135 reg_value |= MLX5_CT_SYNDROME_TRAP;
10136 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_BAD)
10137 reg_value |= MLX5_CT_SYNDROME_BAD_PACKET;
10138 if (mask->flags & (RTE_FLOW_CONNTRACK_PKT_STATE_VALID |
10139 RTE_FLOW_CONNTRACK_PKT_STATE_INVALID |
10140 RTE_FLOW_CONNTRACK_PKT_STATE_DISABLED))
10142 if (mask->flags & RTE_FLOW_CONNTRACK_PKT_STATE_CHANGED)
10143 reg_mask |= MLX5_CT_SYNDROME_STATE_CHANGE;
10144 if (mask->flags & RTE_FLOW_CONNTRACK_PKT_STATE_BAD)
10145 reg_mask |= MLX5_CT_SYNDROME_BAD_PACKET;
10146 /* The REG_C_x value could be saved during startup. */
10147 reg_id = mlx5_flow_get_reg_id(dev, MLX5_ASO_CONNTRACK, 0, &error);
10148 if (reg_id == REG_NON)
10150 flow_dv_match_meta_reg(matcher, key, (enum modify_reg)reg_id,
10151 reg_value, reg_mask);
10155 flow_dv_translate_item_flex(struct rte_eth_dev *dev, void *matcher, void *key,
10156 const struct rte_flow_item *item,
10157 struct mlx5_flow *dev_flow, bool is_inner)
10159 const struct rte_flow_item_flex *spec =
10160 (const struct rte_flow_item_flex *)item->spec;
10161 int index = mlx5_flex_acquire_index(dev, spec->handle, false);
10163 MLX5_ASSERT(index >= 0 && index <= (int)(sizeof(uint32_t) * CHAR_BIT));
10166 if (!(dev_flow->handle->flex_item & RTE_BIT32(index))) {
10167 /* Don't count both inner and outer flex items in one rule. */
10168 if (mlx5_flex_acquire_index(dev, spec->handle, true) != index)
10169 MLX5_ASSERT(false);
10170 dev_flow->handle->flex_item |= RTE_BIT32(index);
10172 mlx5_flex_flow_translate_item(dev, matcher, key, item, is_inner);
10175 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
10177 #define HEADER_IS_ZERO(match_criteria, headers) \
10178 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
10179 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
10182 * Calculate flow matcher enable bitmap.
10184 * @param match_criteria
10185 * Pointer to flow matcher criteria.
10188 * Bitmap of enabled fields.
10191 flow_dv_matcher_enable(uint32_t *match_criteria)
10193 uint8_t match_criteria_enable;
10195 match_criteria_enable =
10196 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
10197 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
10198 match_criteria_enable |=
10199 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
10200 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
10201 match_criteria_enable |=
10202 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
10203 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
10204 match_criteria_enable |=
10205 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
10206 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
10207 match_criteria_enable |=
10208 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
10209 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
10210 match_criteria_enable |=
10211 (!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
10212 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
10213 match_criteria_enable |=
10214 (!HEADER_IS_ZERO(match_criteria, misc_parameters_5)) <<
10215 MLX5_MATCH_CRITERIA_ENABLE_MISC5_BIT;
10216 return match_criteria_enable;
10220 __flow_dv_adjust_buf_size(size_t *size, uint8_t match_criteria)
10223 * Check flow matching criteria first, subtract misc5/4 length if flow
10224 * doesn't own misc5/4 parameters. In some old rdma-core releases,
10225 * misc5/4 are not supported, and matcher creation failure is expected
10226 * w/o subtraction. If misc5 is provided, misc4 must be counted in since
10227 * misc5 is right after misc4.
10229 if (!(match_criteria & (1 << MLX5_MATCH_CRITERIA_ENABLE_MISC5_BIT))) {
10230 *size = MLX5_ST_SZ_BYTES(fte_match_param) -
10231 MLX5_ST_SZ_BYTES(fte_match_set_misc5);
10232 if (!(match_criteria & (1 <<
10233 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT))) {
10234 *size -= MLX5_ST_SZ_BYTES(fte_match_set_misc4);
10239 static struct mlx5_list_entry *
10240 flow_dv_matcher_clone_cb(void *tool_ctx __rte_unused,
10241 struct mlx5_list_entry *entry, void *cb_ctx)
10243 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10244 struct mlx5_flow_dv_matcher *ref = ctx->data;
10245 struct mlx5_flow_tbl_data_entry *tbl = container_of(ref->tbl,
10246 typeof(*tbl), tbl);
10247 struct mlx5_flow_dv_matcher *resource = mlx5_malloc(MLX5_MEM_ANY,
10252 rte_flow_error_set(ctx->error, ENOMEM,
10253 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10254 "cannot create matcher");
10257 memcpy(resource, entry, sizeof(*resource));
10258 resource->tbl = &tbl->tbl;
10259 return &resource->entry;
10263 flow_dv_matcher_clone_free_cb(void *tool_ctx __rte_unused,
10264 struct mlx5_list_entry *entry)
10269 struct mlx5_list_entry *
10270 flow_dv_tbl_create_cb(void *tool_ctx, void *cb_ctx)
10272 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10273 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10274 struct rte_eth_dev *dev = ctx->dev;
10275 struct mlx5_flow_tbl_data_entry *tbl_data;
10276 struct mlx5_flow_tbl_tunnel_prm *tt_prm = ctx->data2;
10277 struct rte_flow_error *error = ctx->error;
10278 union mlx5_flow_tbl_key key = { .v64 = *(uint64_t *)(ctx->data) };
10279 struct mlx5_flow_tbl_resource *tbl;
10284 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
10286 rte_flow_error_set(error, ENOMEM,
10287 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10289 "cannot allocate flow table data entry");
10292 tbl_data->idx = idx;
10293 tbl_data->tunnel = tt_prm->tunnel;
10294 tbl_data->group_id = tt_prm->group_id;
10295 tbl_data->external = !!tt_prm->external;
10296 tbl_data->tunnel_offload = is_tunnel_offload_active(dev);
10297 tbl_data->is_egress = !!key.is_egress;
10298 tbl_data->is_transfer = !!key.is_fdb;
10299 tbl_data->dummy = !!key.dummy;
10300 tbl_data->level = key.level;
10301 tbl_data->id = key.id;
10302 tbl = &tbl_data->tbl;
10304 return &tbl_data->entry;
10306 domain = sh->fdb_domain;
10307 else if (key.is_egress)
10308 domain = sh->tx_domain;
10310 domain = sh->rx_domain;
10311 ret = mlx5_flow_os_create_flow_tbl(domain, key.level, &tbl->obj);
10313 rte_flow_error_set(error, ENOMEM,
10314 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10315 NULL, "cannot create flow table object");
10316 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
10319 if (key.level != 0) {
10320 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
10321 (tbl->obj, &tbl_data->jump.action);
10323 rte_flow_error_set(error, ENOMEM,
10324 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10326 "cannot create flow jump action");
10327 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
10328 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
10332 MKSTR(matcher_name, "%s_%s_%u_%u_matcher_list",
10333 key.is_fdb ? "FDB" : "NIC", key.is_egress ? "egress" : "ingress",
10334 key.level, key.id);
10335 tbl_data->matchers = mlx5_list_create(matcher_name, sh, true,
10336 flow_dv_matcher_create_cb,
10337 flow_dv_matcher_match_cb,
10338 flow_dv_matcher_remove_cb,
10339 flow_dv_matcher_clone_cb,
10340 flow_dv_matcher_clone_free_cb);
10341 if (!tbl_data->matchers) {
10342 rte_flow_error_set(error, ENOMEM,
10343 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10345 "cannot create tbl matcher list");
10346 mlx5_flow_os_destroy_flow_action(tbl_data->jump.action);
10347 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
10348 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
10351 return &tbl_data->entry;
10355 flow_dv_tbl_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
10358 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10359 struct mlx5_flow_tbl_data_entry *tbl_data =
10360 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
10361 union mlx5_flow_tbl_key key = { .v64 = *(uint64_t *)(ctx->data) };
10363 return tbl_data->level != key.level ||
10364 tbl_data->id != key.id ||
10365 tbl_data->dummy != key.dummy ||
10366 tbl_data->is_transfer != !!key.is_fdb ||
10367 tbl_data->is_egress != !!key.is_egress;
10370 struct mlx5_list_entry *
10371 flow_dv_tbl_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
10374 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10375 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10376 struct mlx5_flow_tbl_data_entry *tbl_data;
10377 struct rte_flow_error *error = ctx->error;
10380 tbl_data = mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
10382 rte_flow_error_set(error, ENOMEM,
10383 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10385 "cannot allocate flow table data entry");
10388 memcpy(tbl_data, oentry, sizeof(*tbl_data));
10389 tbl_data->idx = idx;
10390 return &tbl_data->entry;
10394 flow_dv_tbl_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
10396 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10397 struct mlx5_flow_tbl_data_entry *tbl_data =
10398 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
10400 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], tbl_data->idx);
10404 * Get a flow table.
10406 * @param[in, out] dev
10407 * Pointer to rte_eth_dev structure.
10408 * @param[in] table_level
10409 * Table level to use.
10410 * @param[in] egress
10411 * Direction of the table.
10412 * @param[in] transfer
10413 * E-Switch or NIC flow.
10415 * Dummy entry for dv API.
10416 * @param[in] table_id
10418 * @param[out] error
10419 * pointer to error structure.
10422 * Returns tables resource based on the index, NULL in case of failed.
10424 struct mlx5_flow_tbl_resource *
10425 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
10426 uint32_t table_level, uint8_t egress,
10429 const struct mlx5_flow_tunnel *tunnel,
10430 uint32_t group_id, uint8_t dummy,
10432 struct rte_flow_error *error)
10434 struct mlx5_priv *priv = dev->data->dev_private;
10435 union mlx5_flow_tbl_key table_key = {
10437 .level = table_level,
10441 .is_fdb = !!transfer,
10442 .is_egress = !!egress,
10445 struct mlx5_flow_tbl_tunnel_prm tt_prm = {
10447 .group_id = group_id,
10448 .external = external,
10450 struct mlx5_flow_cb_ctx ctx = {
10453 .data = &table_key.v64,
10456 struct mlx5_list_entry *entry;
10457 struct mlx5_flow_tbl_data_entry *tbl_data;
10459 entry = mlx5_hlist_register(priv->sh->flow_tbls, table_key.v64, &ctx);
10461 rte_flow_error_set(error, ENOMEM,
10462 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10463 "cannot get table");
10466 DRV_LOG(DEBUG, "table_level %u table_id %u "
10467 "tunnel %u group %u registered.",
10468 table_level, table_id,
10469 tunnel ? tunnel->tunnel_id : 0, group_id);
10470 tbl_data = container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
10471 return &tbl_data->tbl;
10475 flow_dv_tbl_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
10477 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10478 struct mlx5_flow_tbl_data_entry *tbl_data =
10479 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
10481 MLX5_ASSERT(entry && sh);
10482 if (tbl_data->jump.action)
10483 mlx5_flow_os_destroy_flow_action(tbl_data->jump.action);
10484 if (tbl_data->tbl.obj)
10485 mlx5_flow_os_destroy_flow_tbl(tbl_data->tbl.obj);
10486 if (tbl_data->tunnel_offload && tbl_data->external) {
10487 struct mlx5_list_entry *he;
10488 struct mlx5_hlist *tunnel_grp_hash;
10489 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
10490 union tunnel_tbl_key tunnel_key = {
10491 .tunnel_id = tbl_data->tunnel ?
10492 tbl_data->tunnel->tunnel_id : 0,
10493 .group = tbl_data->group_id
10495 uint32_t table_level = tbl_data->level;
10496 struct mlx5_flow_cb_ctx ctx = {
10497 .data = (void *)&tunnel_key.val,
10500 tunnel_grp_hash = tbl_data->tunnel ?
10501 tbl_data->tunnel->groups :
10503 he = mlx5_hlist_lookup(tunnel_grp_hash, tunnel_key.val, &ctx);
10505 mlx5_hlist_unregister(tunnel_grp_hash, he);
10507 "table_level %u id %u tunnel %u group %u released.",
10511 tbl_data->tunnel->tunnel_id : 0,
10512 tbl_data->group_id);
10514 mlx5_list_destroy(tbl_data->matchers);
10515 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], tbl_data->idx);
10519 * Release a flow table.
10522 * Pointer to device shared structure.
10524 * Table resource to be released.
10527 * Returns 0 if table was released, else return 1;
10530 flow_dv_tbl_resource_release(struct mlx5_dev_ctx_shared *sh,
10531 struct mlx5_flow_tbl_resource *tbl)
10533 struct mlx5_flow_tbl_data_entry *tbl_data =
10534 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
10538 return mlx5_hlist_unregister(sh->flow_tbls, &tbl_data->entry);
10542 flow_dv_matcher_match_cb(void *tool_ctx __rte_unused,
10543 struct mlx5_list_entry *entry, void *cb_ctx)
10545 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10546 struct mlx5_flow_dv_matcher *ref = ctx->data;
10547 struct mlx5_flow_dv_matcher *cur = container_of(entry, typeof(*cur),
10550 return cur->crc != ref->crc ||
10551 cur->priority != ref->priority ||
10552 memcmp((const void *)cur->mask.buf,
10553 (const void *)ref->mask.buf, ref->mask.size);
10556 struct mlx5_list_entry *
10557 flow_dv_matcher_create_cb(void *tool_ctx, void *cb_ctx)
10559 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10560 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10561 struct mlx5_flow_dv_matcher *ref = ctx->data;
10562 struct mlx5_flow_dv_matcher *resource;
10563 struct mlx5dv_flow_matcher_attr dv_attr = {
10564 .type = IBV_FLOW_ATTR_NORMAL,
10565 .match_mask = (void *)&ref->mask,
10567 struct mlx5_flow_tbl_data_entry *tbl = container_of(ref->tbl,
10568 typeof(*tbl), tbl);
10571 resource = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*resource), 0,
10574 rte_flow_error_set(ctx->error, ENOMEM,
10575 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10576 "cannot create matcher");
10580 dv_attr.match_criteria_enable =
10581 flow_dv_matcher_enable(resource->mask.buf);
10582 __flow_dv_adjust_buf_size(&ref->mask.size,
10583 dv_attr.match_criteria_enable);
10584 dv_attr.priority = ref->priority;
10585 if (tbl->is_egress)
10586 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
10587 ret = mlx5_flow_os_create_flow_matcher(sh->cdev->ctx, &dv_attr,
10589 &resource->matcher_object);
10591 mlx5_free(resource);
10592 rte_flow_error_set(ctx->error, ENOMEM,
10593 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10594 "cannot create matcher");
10597 return &resource->entry;
10601 * Register the flow matcher.
10603 * @param[in, out] dev
10604 * Pointer to rte_eth_dev structure.
10605 * @param[in, out] matcher
10606 * Pointer to flow matcher.
10607 * @param[in, out] key
10608 * Pointer to flow table key.
10609 * @parm[in, out] dev_flow
10610 * Pointer to the dev_flow.
10611 * @param[out] error
10612 * pointer to error structure.
10615 * 0 on success otherwise -errno and errno is set.
10618 flow_dv_matcher_register(struct rte_eth_dev *dev,
10619 struct mlx5_flow_dv_matcher *ref,
10620 union mlx5_flow_tbl_key *key,
10621 struct mlx5_flow *dev_flow,
10622 const struct mlx5_flow_tunnel *tunnel,
10624 struct rte_flow_error *error)
10626 struct mlx5_list_entry *entry;
10627 struct mlx5_flow_dv_matcher *resource;
10628 struct mlx5_flow_tbl_resource *tbl;
10629 struct mlx5_flow_tbl_data_entry *tbl_data;
10630 struct mlx5_flow_cb_ctx ctx = {
10635 * tunnel offload API requires this registration for cases when
10636 * tunnel match rule was inserted before tunnel set rule.
10638 tbl = flow_dv_tbl_resource_get(dev, key->level,
10639 key->is_egress, key->is_fdb,
10640 dev_flow->external, tunnel,
10641 group_id, 0, key->id, error);
10643 return -rte_errno; /* No need to refill the error info */
10644 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
10646 entry = mlx5_list_register(tbl_data->matchers, &ctx);
10648 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
10649 return rte_flow_error_set(error, ENOMEM,
10650 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10651 "cannot allocate ref memory");
10653 resource = container_of(entry, typeof(*resource), entry);
10654 dev_flow->handle->dvh.matcher = resource;
10658 struct mlx5_list_entry *
10659 flow_dv_tag_create_cb(void *tool_ctx, void *cb_ctx)
10661 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10662 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10663 struct mlx5_flow_dv_tag_resource *entry;
10667 entry = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG], &idx);
10669 rte_flow_error_set(ctx->error, ENOMEM,
10670 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10671 "cannot allocate resource memory");
10675 entry->tag_id = *(uint32_t *)(ctx->data);
10676 ret = mlx5_flow_os_create_flow_action_tag(entry->tag_id,
10679 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], idx);
10680 rte_flow_error_set(ctx->error, ENOMEM,
10681 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10682 NULL, "cannot create action");
10685 return &entry->entry;
10689 flow_dv_tag_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
10692 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10693 struct mlx5_flow_dv_tag_resource *tag =
10694 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
10696 return *(uint32_t *)(ctx->data) != tag->tag_id;
10699 struct mlx5_list_entry *
10700 flow_dv_tag_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
10703 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10704 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10705 struct mlx5_flow_dv_tag_resource *entry;
10708 entry = mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TAG], &idx);
10710 rte_flow_error_set(ctx->error, ENOMEM,
10711 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10712 "cannot allocate tag resource memory");
10715 memcpy(entry, oentry, sizeof(*entry));
10717 return &entry->entry;
10721 flow_dv_tag_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
10723 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10724 struct mlx5_flow_dv_tag_resource *tag =
10725 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
10727 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], tag->idx);
10731 * Find existing tag resource or create and register a new one.
10733 * @param dev[in, out]
10734 * Pointer to rte_eth_dev structure.
10735 * @param[in, out] tag_be24
10736 * Tag value in big endian then R-shift 8.
10737 * @parm[in, out] dev_flow
10738 * Pointer to the dev_flow.
10739 * @param[out] error
10740 * pointer to error structure.
10743 * 0 on success otherwise -errno and errno is set.
10746 flow_dv_tag_resource_register
10747 (struct rte_eth_dev *dev,
10749 struct mlx5_flow *dev_flow,
10750 struct rte_flow_error *error)
10752 struct mlx5_priv *priv = dev->data->dev_private;
10753 struct mlx5_flow_dv_tag_resource *resource;
10754 struct mlx5_list_entry *entry;
10755 struct mlx5_flow_cb_ctx ctx = {
10759 struct mlx5_hlist *tag_table;
10761 tag_table = flow_dv_hlist_prepare(priv->sh, &priv->sh->tag_table,
10763 MLX5_TAGS_HLIST_ARRAY_SIZE,
10764 false, false, priv->sh,
10765 flow_dv_tag_create_cb,
10766 flow_dv_tag_match_cb,
10767 flow_dv_tag_remove_cb,
10768 flow_dv_tag_clone_cb,
10769 flow_dv_tag_clone_free_cb,
10771 if (unlikely(!tag_table))
10773 entry = mlx5_hlist_register(tag_table, tag_be24, &ctx);
10775 resource = container_of(entry, struct mlx5_flow_dv_tag_resource,
10777 dev_flow->handle->dvh.rix_tag = resource->idx;
10778 dev_flow->dv.tag_resource = resource;
10785 flow_dv_tag_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
10787 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10788 struct mlx5_flow_dv_tag_resource *tag =
10789 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
10791 MLX5_ASSERT(tag && sh && tag->action);
10792 claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
10793 DRV_LOG(DEBUG, "Tag %p: removed.", (void *)tag);
10794 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], tag->idx);
10801 * Pointer to Ethernet device.
10806 * 1 while a reference on it exists, 0 when freed.
10809 flow_dv_tag_release(struct rte_eth_dev *dev,
10812 struct mlx5_priv *priv = dev->data->dev_private;
10813 struct mlx5_flow_dv_tag_resource *tag;
10815 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
10818 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
10819 dev->data->port_id, (void *)tag, tag->entry.ref_cnt);
10820 return mlx5_hlist_unregister(priv->sh->tag_table, &tag->entry);
10824 * Translate action PORT_ID / REPRESENTED_PORT to vport.
10827 * Pointer to rte_eth_dev structure.
10828 * @param[in] action
10829 * Pointer to action PORT_ID / REPRESENTED_PORT.
10830 * @param[out] dst_port_id
10831 * The target port ID.
10832 * @param[out] error
10833 * Pointer to the error structure.
10836 * 0 on success, a negative errno value otherwise and rte_errno is set.
10839 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
10840 const struct rte_flow_action *action,
10841 uint32_t *dst_port_id,
10842 struct rte_flow_error *error)
10845 struct mlx5_priv *priv;
10847 switch (action->type) {
10848 case RTE_FLOW_ACTION_TYPE_PORT_ID: {
10849 const struct rte_flow_action_port_id *conf;
10851 conf = (const struct rte_flow_action_port_id *)action->conf;
10852 port = conf->original ? dev->data->port_id : conf->id;
10855 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT: {
10856 const struct rte_flow_action_ethdev *ethdev;
10858 ethdev = (const struct rte_flow_action_ethdev *)action->conf;
10859 port = ethdev->port_id;
10863 MLX5_ASSERT(false);
10864 return rte_flow_error_set(error, EINVAL,
10865 RTE_FLOW_ERROR_TYPE_ACTION, action,
10866 "unknown E-Switch action");
10869 priv = mlx5_port_to_eswitch_info(port, false);
10871 return rte_flow_error_set(error, -rte_errno,
10872 RTE_FLOW_ERROR_TYPE_ACTION,
10874 "No eswitch info was found for port");
10875 #ifdef HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT
10877 * This parameter is transferred to
10878 * mlx5dv_dr_action_create_dest_ib_port().
10880 *dst_port_id = priv->dev_port;
10883 * Legacy mode, no LAG configurations is supported.
10884 * This parameter is transferred to
10885 * mlx5dv_dr_action_create_dest_vport().
10887 *dst_port_id = priv->vport_id;
10893 * Create a counter with aging configuration.
10896 * Pointer to rte_eth_dev structure.
10897 * @param[in] dev_flow
10898 * Pointer to the mlx5_flow.
10899 * @param[out] count
10900 * Pointer to the counter action configuration.
10902 * Pointer to the aging action configuration.
10905 * Index to flow counter on success, 0 otherwise.
10908 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
10909 struct mlx5_flow *dev_flow,
10910 const struct rte_flow_action_count *count
10912 const struct rte_flow_action_age *age)
10915 struct mlx5_age_param *age_param;
10917 counter = flow_dv_counter_alloc(dev, !!age);
10918 if (!counter || age == NULL)
10920 age_param = flow_dv_counter_idx_get_age(dev, counter);
10921 age_param->context = age->context ? age->context :
10922 (void *)(uintptr_t)(dev_flow->flow_idx);
10923 age_param->timeout = age->timeout;
10924 age_param->port_id = dev->data->port_id;
10925 __atomic_store_n(&age_param->sec_since_last_hit, 0, __ATOMIC_RELAXED);
10926 __atomic_store_n(&age_param->state, AGE_CANDIDATE, __ATOMIC_RELAXED);
10931 * Add Tx queue matcher
10934 * Pointer to the dev struct.
10935 * @param[in, out] matcher
10937 * @param[in, out] key
10938 * Flow matcher value.
10940 * Flow pattern to translate.
10942 * Item is inner pattern.
10945 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
10946 void *matcher, void *key,
10947 const struct rte_flow_item *item)
10949 const struct mlx5_rte_flow_item_tx_queue *queue_m;
10950 const struct mlx5_rte_flow_item_tx_queue *queue_v;
10952 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
10954 MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
10955 struct mlx5_txq_ctrl *txq;
10956 uint32_t queue, mask;
10958 queue_m = (const void *)item->mask;
10959 queue_v = (const void *)item->spec;
10962 txq = mlx5_txq_get(dev, queue_v->queue);
10965 if (txq->type == MLX5_TXQ_TYPE_HAIRPIN)
10966 queue = txq->obj->sq->id;
10968 queue = txq->obj->sq_obj.sq->id;
10969 mask = queue_m == NULL ? UINT32_MAX : queue_m->queue;
10970 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, mask);
10971 MLX5_SET(fte_match_set_misc, misc_v, source_sqn, queue & mask);
10972 mlx5_txq_release(dev, queue_v->queue);
10976 * Set the hash fields according to the @p flow information.
10978 * @param[in] item_flags
10979 * The match pattern item flags.
10980 * @param[in] rss_desc
10981 * Pointer to the mlx5_flow_rss_desc.
10982 * @param[out] hash_fields
10983 * Pointer to the RSS hash fields.
10986 flow_dv_hashfields_set(uint64_t item_flags,
10987 struct mlx5_flow_rss_desc *rss_desc,
10988 uint64_t *hash_fields)
10990 uint64_t items = item_flags;
10991 uint64_t fields = 0;
10993 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
10996 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
10997 if (rss_desc->level >= 2)
11000 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
11001 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4)) ||
11003 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
11004 if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
11005 fields |= IBV_RX_HASH_SRC_IPV4;
11006 else if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
11007 fields |= IBV_RX_HASH_DST_IPV4;
11009 fields |= MLX5_IPV4_IBV_RX_HASH;
11011 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
11012 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6)) ||
11014 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
11015 if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
11016 fields |= IBV_RX_HASH_SRC_IPV6;
11017 else if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
11018 fields |= IBV_RX_HASH_DST_IPV6;
11020 fields |= MLX5_IPV6_IBV_RX_HASH;
11025 * There is no match between the RSS types and the
11026 * L3 protocol (IPv4/IPv6) defined in the flow rule.
11029 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
11030 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP)) ||
11032 if (rss_types & RTE_ETH_RSS_UDP) {
11033 if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
11034 fields |= IBV_RX_HASH_SRC_PORT_UDP;
11035 else if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
11036 fields |= IBV_RX_HASH_DST_PORT_UDP;
11038 fields |= MLX5_UDP_IBV_RX_HASH;
11040 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
11041 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP)) ||
11043 if (rss_types & RTE_ETH_RSS_TCP) {
11044 if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
11045 fields |= IBV_RX_HASH_SRC_PORT_TCP;
11046 else if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
11047 fields |= IBV_RX_HASH_DST_PORT_TCP;
11049 fields |= MLX5_TCP_IBV_RX_HASH;
11053 fields |= IBV_RX_HASH_INNER;
11054 *hash_fields = fields;
11058 * Prepare an Rx Hash queue.
11061 * Pointer to Ethernet device.
11062 * @param[in] dev_flow
11063 * Pointer to the mlx5_flow.
11064 * @param[in] rss_desc
11065 * Pointer to the mlx5_flow_rss_desc.
11066 * @param[out] hrxq_idx
11067 * Hash Rx queue index.
11070 * The Verbs/DevX object initialised, NULL otherwise and rte_errno is set.
11072 static struct mlx5_hrxq *
11073 flow_dv_hrxq_prepare(struct rte_eth_dev *dev,
11074 struct mlx5_flow *dev_flow,
11075 struct mlx5_flow_rss_desc *rss_desc,
11076 uint32_t *hrxq_idx)
11078 struct mlx5_flow_handle *dh = dev_flow->handle;
11079 struct mlx5_hrxq *hrxq;
11081 MLX5_ASSERT(rss_desc->queue_num);
11082 rss_desc->key_len = MLX5_RSS_HASH_KEY_LEN;
11083 rss_desc->hash_fields = dev_flow->hash_fields;
11084 rss_desc->tunnel = !!(dh->layers & MLX5_FLOW_LAYER_TUNNEL);
11085 rss_desc->shared_rss = 0;
11086 if (rss_desc->hash_fields == 0)
11087 rss_desc->queue_num = 1;
11088 hrxq = mlx5_hrxq_get(dev, rss_desc);
11089 *hrxq_idx = hrxq ? hrxq->idx : 0;
11094 * Release sample sub action resource.
11096 * @param[in, out] dev
11097 * Pointer to rte_eth_dev structure.
11098 * @param[in] act_res
11099 * Pointer to sample sub action resource.
11102 flow_dv_sample_sub_actions_release(struct rte_eth_dev *dev,
11103 struct mlx5_flow_sub_actions_idx *act_res)
11105 if (act_res->rix_hrxq) {
11106 mlx5_hrxq_release(dev, act_res->rix_hrxq);
11107 act_res->rix_hrxq = 0;
11109 if (act_res->rix_encap_decap) {
11110 flow_dv_encap_decap_resource_release(dev,
11111 act_res->rix_encap_decap);
11112 act_res->rix_encap_decap = 0;
11114 if (act_res->rix_port_id_action) {
11115 flow_dv_port_id_action_resource_release(dev,
11116 act_res->rix_port_id_action);
11117 act_res->rix_port_id_action = 0;
11119 if (act_res->rix_tag) {
11120 flow_dv_tag_release(dev, act_res->rix_tag);
11121 act_res->rix_tag = 0;
11123 if (act_res->rix_jump) {
11124 flow_dv_jump_tbl_resource_release(dev, act_res->rix_jump);
11125 act_res->rix_jump = 0;
11130 flow_dv_sample_match_cb(void *tool_ctx __rte_unused,
11131 struct mlx5_list_entry *entry, void *cb_ctx)
11133 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11134 struct rte_eth_dev *dev = ctx->dev;
11135 struct mlx5_flow_dv_sample_resource *ctx_resource = ctx->data;
11136 struct mlx5_flow_dv_sample_resource *resource = container_of(entry,
11140 if (ctx_resource->ratio == resource->ratio &&
11141 ctx_resource->ft_type == resource->ft_type &&
11142 ctx_resource->ft_id == resource->ft_id &&
11143 ctx_resource->set_action == resource->set_action &&
11144 !memcmp((void *)&ctx_resource->sample_act,
11145 (void *)&resource->sample_act,
11146 sizeof(struct mlx5_flow_sub_actions_list))) {
11148 * Existing sample action should release the prepared
11149 * sub-actions reference counter.
11151 flow_dv_sample_sub_actions_release(dev,
11152 &ctx_resource->sample_idx);
11158 struct mlx5_list_entry *
11159 flow_dv_sample_create_cb(void *tool_ctx __rte_unused, void *cb_ctx)
11161 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11162 struct rte_eth_dev *dev = ctx->dev;
11163 struct mlx5_flow_dv_sample_resource *ctx_resource = ctx->data;
11164 void **sample_dv_actions = ctx_resource->sub_actions;
11165 struct mlx5_flow_dv_sample_resource *resource;
11166 struct mlx5dv_dr_flow_sampler_attr sampler_attr;
11167 struct mlx5_priv *priv = dev->data->dev_private;
11168 struct mlx5_dev_ctx_shared *sh = priv->sh;
11169 struct mlx5_flow_tbl_resource *tbl;
11171 const uint32_t next_ft_step = 1;
11172 uint32_t next_ft_id = ctx_resource->ft_id + next_ft_step;
11173 uint8_t is_egress = 0;
11174 uint8_t is_transfer = 0;
11175 struct rte_flow_error *error = ctx->error;
11177 /* Register new sample resource. */
11178 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE], &idx);
11180 rte_flow_error_set(error, ENOMEM,
11181 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11183 "cannot allocate resource memory");
11186 *resource = *ctx_resource;
11187 /* Create normal path table level */
11188 if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
11190 else if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
11192 tbl = flow_dv_tbl_resource_get(dev, next_ft_id,
11193 is_egress, is_transfer,
11194 true, NULL, 0, 0, 0, error);
11196 rte_flow_error_set(error, ENOMEM,
11197 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11199 "fail to create normal path table "
11203 resource->normal_path_tbl = tbl;
11204 if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
11205 if (!sh->default_miss_action) {
11206 rte_flow_error_set(error, ENOMEM,
11207 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11209 "default miss action was not "
11213 sample_dv_actions[ctx_resource->sample_act.actions_num++] =
11214 sh->default_miss_action;
11216 /* Create a DR sample action */
11217 sampler_attr.sample_ratio = resource->ratio;
11218 sampler_attr.default_next_table = tbl->obj;
11219 sampler_attr.num_sample_actions = ctx_resource->sample_act.actions_num;
11220 sampler_attr.sample_actions = (struct mlx5dv_dr_action **)
11221 &sample_dv_actions[0];
11222 sampler_attr.action = resource->set_action;
11223 if (mlx5_os_flow_dr_create_flow_action_sampler
11224 (&sampler_attr, &resource->verbs_action)) {
11225 rte_flow_error_set(error, ENOMEM,
11226 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11227 NULL, "cannot create sample action");
11230 resource->idx = idx;
11231 resource->dev = dev;
11232 return &resource->entry;
11234 if (resource->ft_type != MLX5DV_FLOW_TABLE_TYPE_FDB)
11235 flow_dv_sample_sub_actions_release(dev,
11236 &resource->sample_idx);
11237 if (resource->normal_path_tbl)
11238 flow_dv_tbl_resource_release(MLX5_SH(dev),
11239 resource->normal_path_tbl);
11240 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_SAMPLE], idx);
11245 struct mlx5_list_entry *
11246 flow_dv_sample_clone_cb(void *tool_ctx __rte_unused,
11247 struct mlx5_list_entry *entry __rte_unused,
11250 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11251 struct rte_eth_dev *dev = ctx->dev;
11252 struct mlx5_flow_dv_sample_resource *resource;
11253 struct mlx5_priv *priv = dev->data->dev_private;
11254 struct mlx5_dev_ctx_shared *sh = priv->sh;
11257 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE], &idx);
11259 rte_flow_error_set(ctx->error, ENOMEM,
11260 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11262 "cannot allocate resource memory");
11265 memcpy(resource, entry, sizeof(*resource));
11266 resource->idx = idx;
11267 resource->dev = dev;
11268 return &resource->entry;
11272 flow_dv_sample_clone_free_cb(void *tool_ctx __rte_unused,
11273 struct mlx5_list_entry *entry)
11275 struct mlx5_flow_dv_sample_resource *resource =
11276 container_of(entry, typeof(*resource), entry);
11277 struct rte_eth_dev *dev = resource->dev;
11278 struct mlx5_priv *priv = dev->data->dev_private;
11280 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE], resource->idx);
11284 * Find existing sample resource or create and register a new one.
11286 * @param[in, out] dev
11287 * Pointer to rte_eth_dev structure.
11289 * Pointer to sample resource reference.
11290 * @parm[in, out] dev_flow
11291 * Pointer to the dev_flow.
11292 * @param[out] error
11293 * pointer to error structure.
11296 * 0 on success otherwise -errno and errno is set.
11299 flow_dv_sample_resource_register(struct rte_eth_dev *dev,
11300 struct mlx5_flow_dv_sample_resource *ref,
11301 struct mlx5_flow *dev_flow,
11302 struct rte_flow_error *error)
11304 struct mlx5_flow_dv_sample_resource *resource;
11305 struct mlx5_list_entry *entry;
11306 struct mlx5_priv *priv = dev->data->dev_private;
11307 struct mlx5_flow_cb_ctx ctx = {
11313 entry = mlx5_list_register(priv->sh->sample_action_list, &ctx);
11316 resource = container_of(entry, typeof(*resource), entry);
11317 dev_flow->handle->dvh.rix_sample = resource->idx;
11318 dev_flow->dv.sample_res = resource;
11323 flow_dv_dest_array_match_cb(void *tool_ctx __rte_unused,
11324 struct mlx5_list_entry *entry, void *cb_ctx)
11326 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11327 struct mlx5_flow_dv_dest_array_resource *ctx_resource = ctx->data;
11328 struct rte_eth_dev *dev = ctx->dev;
11329 struct mlx5_flow_dv_dest_array_resource *resource =
11330 container_of(entry, typeof(*resource), entry);
11333 if (ctx_resource->num_of_dest == resource->num_of_dest &&
11334 ctx_resource->ft_type == resource->ft_type &&
11335 !memcmp((void *)resource->sample_act,
11336 (void *)ctx_resource->sample_act,
11337 (ctx_resource->num_of_dest *
11338 sizeof(struct mlx5_flow_sub_actions_list)))) {
11340 * Existing sample action should release the prepared
11341 * sub-actions reference counter.
11343 for (idx = 0; idx < ctx_resource->num_of_dest; idx++)
11344 flow_dv_sample_sub_actions_release(dev,
11345 &ctx_resource->sample_idx[idx]);
11351 struct mlx5_list_entry *
11352 flow_dv_dest_array_create_cb(void *tool_ctx __rte_unused, void *cb_ctx)
11354 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11355 struct rte_eth_dev *dev = ctx->dev;
11356 struct mlx5_flow_dv_dest_array_resource *resource;
11357 struct mlx5_flow_dv_dest_array_resource *ctx_resource = ctx->data;
11358 struct mlx5dv_dr_action_dest_attr *dest_attr[MLX5_MAX_DEST_NUM] = { 0 };
11359 struct mlx5dv_dr_action_dest_reformat dest_reformat[MLX5_MAX_DEST_NUM];
11360 struct mlx5_priv *priv = dev->data->dev_private;
11361 struct mlx5_dev_ctx_shared *sh = priv->sh;
11362 struct mlx5_flow_sub_actions_list *sample_act;
11363 struct mlx5dv_dr_domain *domain;
11364 uint32_t idx = 0, res_idx = 0;
11365 struct rte_flow_error *error = ctx->error;
11366 uint64_t action_flags;
11369 /* Register new destination array resource. */
11370 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
11373 rte_flow_error_set(error, ENOMEM,
11374 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11376 "cannot allocate resource memory");
11379 *resource = *ctx_resource;
11380 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
11381 domain = sh->fdb_domain;
11382 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
11383 domain = sh->rx_domain;
11385 domain = sh->tx_domain;
11386 for (idx = 0; idx < ctx_resource->num_of_dest; idx++) {
11387 dest_attr[idx] = (struct mlx5dv_dr_action_dest_attr *)
11388 mlx5_malloc(MLX5_MEM_ZERO,
11389 sizeof(struct mlx5dv_dr_action_dest_attr),
11391 if (!dest_attr[idx]) {
11392 rte_flow_error_set(error, ENOMEM,
11393 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11395 "cannot allocate resource memory");
11398 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST;
11399 sample_act = &ctx_resource->sample_act[idx];
11400 action_flags = sample_act->action_flags;
11401 switch (action_flags) {
11402 case MLX5_FLOW_ACTION_QUEUE:
11403 dest_attr[idx]->dest = sample_act->dr_queue_action;
11405 case (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_ENCAP):
11406 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST_REFORMAT;
11407 dest_attr[idx]->dest_reformat = &dest_reformat[idx];
11408 dest_attr[idx]->dest_reformat->reformat =
11409 sample_act->dr_encap_action;
11410 dest_attr[idx]->dest_reformat->dest =
11411 sample_act->dr_port_id_action;
11413 case MLX5_FLOW_ACTION_PORT_ID:
11414 dest_attr[idx]->dest = sample_act->dr_port_id_action;
11416 case MLX5_FLOW_ACTION_JUMP:
11417 dest_attr[idx]->dest = sample_act->dr_jump_action;
11420 rte_flow_error_set(error, EINVAL,
11421 RTE_FLOW_ERROR_TYPE_ACTION,
11423 "unsupported actions type");
11427 /* create a dest array action */
11428 ret = mlx5_os_flow_dr_create_flow_action_dest_array
11430 resource->num_of_dest,
11432 &resource->action);
11434 rte_flow_error_set(error, ENOMEM,
11435 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11437 "cannot create destination array action");
11440 resource->idx = res_idx;
11441 resource->dev = dev;
11442 for (idx = 0; idx < ctx_resource->num_of_dest; idx++)
11443 mlx5_free(dest_attr[idx]);
11444 return &resource->entry;
11446 for (idx = 0; idx < ctx_resource->num_of_dest; idx++) {
11447 flow_dv_sample_sub_actions_release(dev,
11448 &resource->sample_idx[idx]);
11449 if (dest_attr[idx])
11450 mlx5_free(dest_attr[idx]);
11452 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DEST_ARRAY], res_idx);
11456 struct mlx5_list_entry *
11457 flow_dv_dest_array_clone_cb(void *tool_ctx __rte_unused,
11458 struct mlx5_list_entry *entry __rte_unused,
11461 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11462 struct rte_eth_dev *dev = ctx->dev;
11463 struct mlx5_flow_dv_dest_array_resource *resource;
11464 struct mlx5_priv *priv = dev->data->dev_private;
11465 struct mlx5_dev_ctx_shared *sh = priv->sh;
11466 uint32_t res_idx = 0;
11467 struct rte_flow_error *error = ctx->error;
11469 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
11472 rte_flow_error_set(error, ENOMEM,
11473 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11475 "cannot allocate dest-array memory");
11478 memcpy(resource, entry, sizeof(*resource));
11479 resource->idx = res_idx;
11480 resource->dev = dev;
11481 return &resource->entry;
11485 flow_dv_dest_array_clone_free_cb(void *tool_ctx __rte_unused,
11486 struct mlx5_list_entry *entry)
11488 struct mlx5_flow_dv_dest_array_resource *resource =
11489 container_of(entry, typeof(*resource), entry);
11490 struct rte_eth_dev *dev = resource->dev;
11491 struct mlx5_priv *priv = dev->data->dev_private;
11493 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY], resource->idx);
11497 * Find existing destination array resource or create and register a new one.
11499 * @param[in, out] dev
11500 * Pointer to rte_eth_dev structure.
11502 * Pointer to destination array resource reference.
11503 * @parm[in, out] dev_flow
11504 * Pointer to the dev_flow.
11505 * @param[out] error
11506 * pointer to error structure.
11509 * 0 on success otherwise -errno and errno is set.
11512 flow_dv_dest_array_resource_register(struct rte_eth_dev *dev,
11513 struct mlx5_flow_dv_dest_array_resource *ref,
11514 struct mlx5_flow *dev_flow,
11515 struct rte_flow_error *error)
11517 struct mlx5_flow_dv_dest_array_resource *resource;
11518 struct mlx5_priv *priv = dev->data->dev_private;
11519 struct mlx5_list_entry *entry;
11520 struct mlx5_flow_cb_ctx ctx = {
11526 entry = mlx5_list_register(priv->sh->dest_array_list, &ctx);
11529 resource = container_of(entry, typeof(*resource), entry);
11530 dev_flow->handle->dvh.rix_dest_array = resource->idx;
11531 dev_flow->dv.dest_array_res = resource;
11536 * Convert Sample action to DV specification.
11539 * Pointer to rte_eth_dev structure.
11540 * @param[in] action
11541 * Pointer to sample action structure.
11542 * @param[in, out] dev_flow
11543 * Pointer to the mlx5_flow.
11545 * Pointer to the flow attributes.
11546 * @param[in, out] num_of_dest
11547 * Pointer to the num of destination.
11548 * @param[in, out] sample_actions
11549 * Pointer to sample actions list.
11550 * @param[in, out] res
11551 * Pointer to sample resource.
11552 * @param[out] error
11553 * Pointer to the error structure.
11556 * 0 on success, a negative errno value otherwise and rte_errno is set.
11559 flow_dv_translate_action_sample(struct rte_eth_dev *dev,
11560 const struct rte_flow_action_sample *action,
11561 struct mlx5_flow *dev_flow,
11562 const struct rte_flow_attr *attr,
11563 uint32_t *num_of_dest,
11564 void **sample_actions,
11565 struct mlx5_flow_dv_sample_resource *res,
11566 struct rte_flow_error *error)
11568 struct mlx5_priv *priv = dev->data->dev_private;
11569 const struct rte_flow_action *sub_actions;
11570 struct mlx5_flow_sub_actions_list *sample_act;
11571 struct mlx5_flow_sub_actions_idx *sample_idx;
11572 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
11573 struct rte_flow *flow = dev_flow->flow;
11574 struct mlx5_flow_rss_desc *rss_desc;
11575 uint64_t action_flags = 0;
11578 rss_desc = &wks->rss_desc;
11579 sample_act = &res->sample_act;
11580 sample_idx = &res->sample_idx;
11581 res->ratio = action->ratio;
11582 sub_actions = action->actions;
11583 for (; sub_actions->type != RTE_FLOW_ACTION_TYPE_END; sub_actions++) {
11584 int type = sub_actions->type;
11585 uint32_t pre_rix = 0;
11588 case RTE_FLOW_ACTION_TYPE_QUEUE:
11590 const struct rte_flow_action_queue *queue;
11591 struct mlx5_hrxq *hrxq;
11594 queue = sub_actions->conf;
11595 rss_desc->queue_num = 1;
11596 rss_desc->queue[0] = queue->index;
11597 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
11598 rss_desc, &hrxq_idx);
11600 return rte_flow_error_set
11602 RTE_FLOW_ERROR_TYPE_ACTION,
11604 "cannot create fate queue");
11605 sample_act->dr_queue_action = hrxq->action;
11606 sample_idx->rix_hrxq = hrxq_idx;
11607 sample_actions[sample_act->actions_num++] =
11610 action_flags |= MLX5_FLOW_ACTION_QUEUE;
11611 if (action_flags & MLX5_FLOW_ACTION_MARK)
11612 dev_flow->handle->rix_hrxq = hrxq_idx;
11613 dev_flow->handle->fate_action =
11614 MLX5_FLOW_FATE_QUEUE;
11617 case RTE_FLOW_ACTION_TYPE_RSS:
11619 struct mlx5_hrxq *hrxq;
11621 const struct rte_flow_action_rss *rss;
11622 const uint8_t *rss_key;
11624 rss = sub_actions->conf;
11625 memcpy(rss_desc->queue, rss->queue,
11626 rss->queue_num * sizeof(uint16_t));
11627 rss_desc->queue_num = rss->queue_num;
11628 /* NULL RSS key indicates default RSS key. */
11629 rss_key = !rss->key ? rss_hash_default_key : rss->key;
11630 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
11632 * rss->level and rss.types should be set in advance
11633 * when expanding items for RSS.
11635 flow_dv_hashfields_set(dev_flow->handle->layers,
11637 &dev_flow->hash_fields);
11638 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
11639 rss_desc, &hrxq_idx);
11641 return rte_flow_error_set
11643 RTE_FLOW_ERROR_TYPE_ACTION,
11645 "cannot create fate queue");
11646 sample_act->dr_queue_action = hrxq->action;
11647 sample_idx->rix_hrxq = hrxq_idx;
11648 sample_actions[sample_act->actions_num++] =
11651 action_flags |= MLX5_FLOW_ACTION_RSS;
11652 if (action_flags & MLX5_FLOW_ACTION_MARK)
11653 dev_flow->handle->rix_hrxq = hrxq_idx;
11654 dev_flow->handle->fate_action =
11655 MLX5_FLOW_FATE_QUEUE;
11658 case RTE_FLOW_ACTION_TYPE_MARK:
11660 uint32_t tag_be = mlx5_flow_mark_set
11661 (((const struct rte_flow_action_mark *)
11662 (sub_actions->conf))->id);
11665 pre_rix = dev_flow->handle->dvh.rix_tag;
11666 /* Save the mark resource before sample */
11667 pre_r = dev_flow->dv.tag_resource;
11668 if (flow_dv_tag_resource_register(dev, tag_be,
11671 MLX5_ASSERT(dev_flow->dv.tag_resource);
11672 sample_act->dr_tag_action =
11673 dev_flow->dv.tag_resource->action;
11674 sample_idx->rix_tag =
11675 dev_flow->handle->dvh.rix_tag;
11676 sample_actions[sample_act->actions_num++] =
11677 sample_act->dr_tag_action;
11678 /* Recover the mark resource after sample */
11679 dev_flow->dv.tag_resource = pre_r;
11680 dev_flow->handle->dvh.rix_tag = pre_rix;
11681 action_flags |= MLX5_FLOW_ACTION_MARK;
11684 case RTE_FLOW_ACTION_TYPE_COUNT:
11686 if (!flow->counter) {
11688 flow_dv_translate_create_counter(dev,
11689 dev_flow, sub_actions->conf,
11691 if (!flow->counter)
11692 return rte_flow_error_set
11694 RTE_FLOW_ERROR_TYPE_ACTION,
11696 "cannot create counter"
11699 sample_act->dr_cnt_action =
11700 (flow_dv_counter_get_by_idx(dev,
11701 flow->counter, NULL))->action;
11702 sample_actions[sample_act->actions_num++] =
11703 sample_act->dr_cnt_action;
11704 action_flags |= MLX5_FLOW_ACTION_COUNT;
11707 case RTE_FLOW_ACTION_TYPE_PORT_ID:
11708 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
11710 struct mlx5_flow_dv_port_id_action_resource
11712 uint32_t port_id = 0;
11714 memset(&port_id_resource, 0, sizeof(port_id_resource));
11715 /* Save the port id resource before sample */
11716 pre_rix = dev_flow->handle->rix_port_id_action;
11717 pre_r = dev_flow->dv.port_id_action;
11718 if (flow_dv_translate_action_port_id(dev, sub_actions,
11721 port_id_resource.port_id = port_id;
11722 if (flow_dv_port_id_action_resource_register
11723 (dev, &port_id_resource, dev_flow, error))
11725 sample_act->dr_port_id_action =
11726 dev_flow->dv.port_id_action->action;
11727 sample_idx->rix_port_id_action =
11728 dev_flow->handle->rix_port_id_action;
11729 sample_actions[sample_act->actions_num++] =
11730 sample_act->dr_port_id_action;
11731 /* Recover the port id resource after sample */
11732 dev_flow->dv.port_id_action = pre_r;
11733 dev_flow->handle->rix_port_id_action = pre_rix;
11735 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
11738 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
11739 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
11740 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
11741 /* Save the encap resource before sample */
11742 pre_rix = dev_flow->handle->dvh.rix_encap_decap;
11743 pre_r = dev_flow->dv.encap_decap;
11744 if (flow_dv_create_action_l2_encap(dev, sub_actions,
11749 sample_act->dr_encap_action =
11750 dev_flow->dv.encap_decap->action;
11751 sample_idx->rix_encap_decap =
11752 dev_flow->handle->dvh.rix_encap_decap;
11753 sample_actions[sample_act->actions_num++] =
11754 sample_act->dr_encap_action;
11755 /* Recover the encap resource after sample */
11756 dev_flow->dv.encap_decap = pre_r;
11757 dev_flow->handle->dvh.rix_encap_decap = pre_rix;
11758 action_flags |= MLX5_FLOW_ACTION_ENCAP;
11761 return rte_flow_error_set(error, EINVAL,
11762 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11764 "Not support for sampler action");
11767 sample_act->action_flags = action_flags;
11768 res->ft_id = dev_flow->dv.group;
11769 if (attr->transfer) {
11771 uint32_t action_in[MLX5_ST_SZ_DW(set_action_in)];
11772 uint64_t set_action;
11773 } action_ctx = { .set_action = 0 };
11775 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
11776 MLX5_SET(set_action_in, action_ctx.action_in, action_type,
11777 MLX5_MODIFICATION_TYPE_SET);
11778 MLX5_SET(set_action_in, action_ctx.action_in, field,
11779 MLX5_MODI_META_REG_C_0);
11780 MLX5_SET(set_action_in, action_ctx.action_in, data,
11781 priv->vport_meta_tag);
11782 res->set_action = action_ctx.set_action;
11783 } else if (attr->ingress) {
11784 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
11786 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_TX;
11792 * Convert Sample action to DV specification.
11795 * Pointer to rte_eth_dev structure.
11796 * @param[in, out] dev_flow
11797 * Pointer to the mlx5_flow.
11798 * @param[in] num_of_dest
11799 * The num of destination.
11800 * @param[in, out] res
11801 * Pointer to sample resource.
11802 * @param[in, out] mdest_res
11803 * Pointer to destination array resource.
11804 * @param[in] sample_actions
11805 * Pointer to sample path actions list.
11806 * @param[in] action_flags
11807 * Holds the actions detected until now.
11808 * @param[out] error
11809 * Pointer to the error structure.
11812 * 0 on success, a negative errno value otherwise and rte_errno is set.
11815 flow_dv_create_action_sample(struct rte_eth_dev *dev,
11816 struct mlx5_flow *dev_flow,
11817 uint32_t num_of_dest,
11818 struct mlx5_flow_dv_sample_resource *res,
11819 struct mlx5_flow_dv_dest_array_resource *mdest_res,
11820 void **sample_actions,
11821 uint64_t action_flags,
11822 struct rte_flow_error *error)
11824 /* update normal path action resource into last index of array */
11825 uint32_t dest_index = MLX5_MAX_DEST_NUM - 1;
11826 struct mlx5_flow_sub_actions_list *sample_act =
11827 &mdest_res->sample_act[dest_index];
11828 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
11829 struct mlx5_flow_rss_desc *rss_desc;
11830 uint32_t normal_idx = 0;
11831 struct mlx5_hrxq *hrxq;
11835 rss_desc = &wks->rss_desc;
11836 if (num_of_dest > 1) {
11837 if (sample_act->action_flags & MLX5_FLOW_ACTION_QUEUE) {
11838 /* Handle QP action for mirroring */
11839 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
11840 rss_desc, &hrxq_idx);
11842 return rte_flow_error_set
11844 RTE_FLOW_ERROR_TYPE_ACTION,
11846 "cannot create rx queue");
11848 mdest_res->sample_idx[dest_index].rix_hrxq = hrxq_idx;
11849 sample_act->dr_queue_action = hrxq->action;
11850 if (action_flags & MLX5_FLOW_ACTION_MARK)
11851 dev_flow->handle->rix_hrxq = hrxq_idx;
11852 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
11854 if (sample_act->action_flags & MLX5_FLOW_ACTION_ENCAP) {
11856 mdest_res->sample_idx[dest_index].rix_encap_decap =
11857 dev_flow->handle->dvh.rix_encap_decap;
11858 sample_act->dr_encap_action =
11859 dev_flow->dv.encap_decap->action;
11860 dev_flow->handle->dvh.rix_encap_decap = 0;
11862 if (sample_act->action_flags & MLX5_FLOW_ACTION_PORT_ID) {
11864 mdest_res->sample_idx[dest_index].rix_port_id_action =
11865 dev_flow->handle->rix_port_id_action;
11866 sample_act->dr_port_id_action =
11867 dev_flow->dv.port_id_action->action;
11868 dev_flow->handle->rix_port_id_action = 0;
11870 if (sample_act->action_flags & MLX5_FLOW_ACTION_JUMP) {
11872 mdest_res->sample_idx[dest_index].rix_jump =
11873 dev_flow->handle->rix_jump;
11874 sample_act->dr_jump_action =
11875 dev_flow->dv.jump->action;
11876 dev_flow->handle->rix_jump = 0;
11878 sample_act->actions_num = normal_idx;
11879 /* update sample action resource into first index of array */
11880 mdest_res->ft_type = res->ft_type;
11881 memcpy(&mdest_res->sample_idx[0], &res->sample_idx,
11882 sizeof(struct mlx5_flow_sub_actions_idx));
11883 memcpy(&mdest_res->sample_act[0], &res->sample_act,
11884 sizeof(struct mlx5_flow_sub_actions_list));
11885 mdest_res->num_of_dest = num_of_dest;
11886 if (flow_dv_dest_array_resource_register(dev, mdest_res,
11888 return rte_flow_error_set(error, EINVAL,
11889 RTE_FLOW_ERROR_TYPE_ACTION,
11890 NULL, "can't create sample "
11893 res->sub_actions = sample_actions;
11894 if (flow_dv_sample_resource_register(dev, res, dev_flow, error))
11895 return rte_flow_error_set(error, EINVAL,
11896 RTE_FLOW_ERROR_TYPE_ACTION,
11898 "can't create sample action");
11904 * Remove an ASO age action from age actions list.
11907 * Pointer to the Ethernet device structure.
11909 * Pointer to the aso age action handler.
11912 flow_dv_aso_age_remove_from_age(struct rte_eth_dev *dev,
11913 struct mlx5_aso_age_action *age)
11915 struct mlx5_age_info *age_info;
11916 struct mlx5_age_param *age_param = &age->age_params;
11917 struct mlx5_priv *priv = dev->data->dev_private;
11918 uint16_t expected = AGE_CANDIDATE;
11920 age_info = GET_PORT_AGE_INFO(priv);
11921 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
11922 AGE_FREE, false, __ATOMIC_RELAXED,
11923 __ATOMIC_RELAXED)) {
11925 * We need the lock even it is age timeout,
11926 * since age action may still in process.
11928 rte_spinlock_lock(&age_info->aged_sl);
11929 LIST_REMOVE(age, next);
11930 rte_spinlock_unlock(&age_info->aged_sl);
11931 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
11936 * Release an ASO age action.
11939 * Pointer to the Ethernet device structure.
11940 * @param[in] age_idx
11941 * Index of ASO age action to release.
11943 * True if the release operation is during flow destroy operation.
11944 * False if the release operation is during action destroy operation.
11947 * 0 when age action was removed, otherwise the number of references.
11950 flow_dv_aso_age_release(struct rte_eth_dev *dev, uint32_t age_idx)
11952 struct mlx5_priv *priv = dev->data->dev_private;
11953 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
11954 struct mlx5_aso_age_action *age = flow_aso_age_get_by_idx(dev, age_idx);
11955 uint32_t ret = __atomic_sub_fetch(&age->refcnt, 1, __ATOMIC_RELAXED);
11958 flow_dv_aso_age_remove_from_age(dev, age);
11959 rte_spinlock_lock(&mng->free_sl);
11960 LIST_INSERT_HEAD(&mng->free, age, next);
11961 rte_spinlock_unlock(&mng->free_sl);
11967 * Resize the ASO age pools array by MLX5_CNT_CONTAINER_RESIZE pools.
11970 * Pointer to the Ethernet device structure.
11973 * 0 on success, otherwise negative errno value and rte_errno is set.
11976 flow_dv_aso_age_pools_resize(struct rte_eth_dev *dev)
11978 struct mlx5_priv *priv = dev->data->dev_private;
11979 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
11980 void *old_pools = mng->pools;
11981 uint32_t resize = mng->n + MLX5_CNT_CONTAINER_RESIZE;
11982 uint32_t mem_size = sizeof(struct mlx5_aso_age_pool *) * resize;
11983 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
11986 rte_errno = ENOMEM;
11990 memcpy(pools, old_pools,
11991 mng->n * sizeof(struct mlx5_flow_counter_pool *));
11992 mlx5_free(old_pools);
11994 /* First ASO flow hit allocation - starting ASO data-path. */
11995 int ret = mlx5_aso_flow_hit_queue_poll_start(priv->sh);
12003 mng->pools = pools;
12008 * Create and initialize a new ASO aging pool.
12011 * Pointer to the Ethernet device structure.
12012 * @param[out] age_free
12013 * Where to put the pointer of a new age action.
12016 * The age actions pool pointer and @p age_free is set on success,
12017 * NULL otherwise and rte_errno is set.
12019 static struct mlx5_aso_age_pool *
12020 flow_dv_age_pool_create(struct rte_eth_dev *dev,
12021 struct mlx5_aso_age_action **age_free)
12023 struct mlx5_priv *priv = dev->data->dev_private;
12024 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
12025 struct mlx5_aso_age_pool *pool = NULL;
12026 struct mlx5_devx_obj *obj = NULL;
12029 obj = mlx5_devx_cmd_create_flow_hit_aso_obj(priv->sh->cdev->ctx,
12030 priv->sh->cdev->pdn);
12032 rte_errno = ENODATA;
12033 DRV_LOG(ERR, "Failed to create flow_hit_aso_obj using DevX.");
12036 pool = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*pool), 0, SOCKET_ID_ANY);
12038 claim_zero(mlx5_devx_cmd_destroy(obj));
12039 rte_errno = ENOMEM;
12042 pool->flow_hit_aso_obj = obj;
12043 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
12044 rte_rwlock_write_lock(&mng->resize_rwl);
12045 pool->index = mng->next;
12046 /* Resize pools array if there is no room for the new pool in it. */
12047 if (pool->index == mng->n && flow_dv_aso_age_pools_resize(dev)) {
12048 claim_zero(mlx5_devx_cmd_destroy(obj));
12050 rte_rwlock_write_unlock(&mng->resize_rwl);
12053 mng->pools[pool->index] = pool;
12055 rte_rwlock_write_unlock(&mng->resize_rwl);
12056 /* Assign the first action in the new pool, the rest go to free list. */
12057 *age_free = &pool->actions[0];
12058 for (i = 1; i < MLX5_ASO_AGE_ACTIONS_PER_POOL; i++) {
12059 pool->actions[i].offset = i;
12060 LIST_INSERT_HEAD(&mng->free, &pool->actions[i], next);
12066 * Allocate a ASO aging bit.
12069 * Pointer to the Ethernet device structure.
12070 * @param[out] error
12071 * Pointer to the error structure.
12074 * Index to ASO age action on success, 0 otherwise and rte_errno is set.
12077 flow_dv_aso_age_alloc(struct rte_eth_dev *dev, struct rte_flow_error *error)
12079 struct mlx5_priv *priv = dev->data->dev_private;
12080 const struct mlx5_aso_age_pool *pool;
12081 struct mlx5_aso_age_action *age_free = NULL;
12082 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
12085 /* Try to get the next free age action bit. */
12086 rte_spinlock_lock(&mng->free_sl);
12087 age_free = LIST_FIRST(&mng->free);
12089 LIST_REMOVE(age_free, next);
12090 } else if (!flow_dv_age_pool_create(dev, &age_free)) {
12091 rte_spinlock_unlock(&mng->free_sl);
12092 rte_flow_error_set(error, rte_errno, RTE_FLOW_ERROR_TYPE_ACTION,
12093 NULL, "failed to create ASO age pool");
12094 return 0; /* 0 is an error. */
12096 rte_spinlock_unlock(&mng->free_sl);
12097 pool = container_of
12098 ((const struct mlx5_aso_age_action (*)[MLX5_ASO_AGE_ACTIONS_PER_POOL])
12099 (age_free - age_free->offset), const struct mlx5_aso_age_pool,
12101 if (!age_free->dr_action) {
12102 int reg_c = mlx5_flow_get_reg_id(dev, MLX5_ASO_FLOW_HIT, 0,
12106 rte_flow_error_set(error, rte_errno,
12107 RTE_FLOW_ERROR_TYPE_ACTION,
12108 NULL, "failed to get reg_c "
12109 "for ASO flow hit");
12110 return 0; /* 0 is an error. */
12112 #ifdef HAVE_MLX5_DR_CREATE_ACTION_ASO
12113 age_free->dr_action = mlx5_glue->dv_create_flow_action_aso
12114 (priv->sh->rx_domain,
12115 pool->flow_hit_aso_obj->obj, age_free->offset,
12116 MLX5DV_DR_ACTION_FLAGS_ASO_FIRST_HIT_SET,
12117 (reg_c - REG_C_0));
12118 #endif /* HAVE_MLX5_DR_CREATE_ACTION_ASO */
12119 if (!age_free->dr_action) {
12121 rte_spinlock_lock(&mng->free_sl);
12122 LIST_INSERT_HEAD(&mng->free, age_free, next);
12123 rte_spinlock_unlock(&mng->free_sl);
12124 rte_flow_error_set(error, rte_errno,
12125 RTE_FLOW_ERROR_TYPE_ACTION,
12126 NULL, "failed to create ASO "
12127 "flow hit action");
12128 return 0; /* 0 is an error. */
12131 __atomic_store_n(&age_free->refcnt, 1, __ATOMIC_RELAXED);
12132 return pool->index | ((age_free->offset + 1) << 16);
12136 * Initialize flow ASO age parameters.
12139 * Pointer to rte_eth_dev structure.
12140 * @param[in] age_idx
12141 * Index of ASO age action.
12142 * @param[in] context
12143 * Pointer to flow counter age context.
12144 * @param[in] timeout
12145 * Aging timeout in seconds.
12149 flow_dv_aso_age_params_init(struct rte_eth_dev *dev,
12154 struct mlx5_aso_age_action *aso_age;
12156 aso_age = flow_aso_age_get_by_idx(dev, age_idx);
12157 MLX5_ASSERT(aso_age);
12158 aso_age->age_params.context = context;
12159 aso_age->age_params.timeout = timeout;
12160 aso_age->age_params.port_id = dev->data->port_id;
12161 __atomic_store_n(&aso_age->age_params.sec_since_last_hit, 0,
12163 __atomic_store_n(&aso_age->age_params.state, AGE_CANDIDATE,
12168 flow_dv_translate_integrity_l4(const struct rte_flow_item_integrity *mask,
12169 const struct rte_flow_item_integrity *value,
12170 void *headers_m, void *headers_v)
12173 /* RTE l4_ok filter aggregates hardware l4_ok and
12174 * l4_checksum_ok filters.
12175 * Positive RTE l4_ok match requires hardware match on both L4
12176 * hardware integrity bits.
12177 * For negative match, check hardware l4_checksum_ok bit only,
12178 * because hardware sets that bit to 0 for all packets
12181 if (value->l4_ok) {
12182 MLX5_SET(fte_match_set_lyr_2_4, headers_m, l4_ok, 1);
12183 MLX5_SET(fte_match_set_lyr_2_4, headers_v, l4_ok, 1);
12185 MLX5_SET(fte_match_set_lyr_2_4, headers_m, l4_checksum_ok, 1);
12186 MLX5_SET(fte_match_set_lyr_2_4, headers_v, l4_checksum_ok,
12189 if (mask->l4_csum_ok) {
12190 MLX5_SET(fte_match_set_lyr_2_4, headers_m, l4_checksum_ok, 1);
12191 MLX5_SET(fte_match_set_lyr_2_4, headers_v, l4_checksum_ok,
12192 value->l4_csum_ok);
12197 flow_dv_translate_integrity_l3(const struct rte_flow_item_integrity *mask,
12198 const struct rte_flow_item_integrity *value,
12199 void *headers_m, void *headers_v, bool is_ipv4)
12202 /* RTE l3_ok filter aggregates for IPv4 hardware l3_ok and
12203 * ipv4_csum_ok filters.
12204 * Positive RTE l3_ok match requires hardware match on both L3
12205 * hardware integrity bits.
12206 * For negative match, check hardware l3_csum_ok bit only,
12207 * because hardware sets that bit to 0 for all packets
12211 if (value->l3_ok) {
12212 MLX5_SET(fte_match_set_lyr_2_4, headers_m,
12214 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
12217 MLX5_SET(fte_match_set_lyr_2_4, headers_m,
12218 ipv4_checksum_ok, 1);
12219 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
12220 ipv4_checksum_ok, !!value->l3_ok);
12222 MLX5_SET(fte_match_set_lyr_2_4, headers_m, l3_ok, 1);
12223 MLX5_SET(fte_match_set_lyr_2_4, headers_v, l3_ok,
12227 if (mask->ipv4_csum_ok) {
12228 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ipv4_checksum_ok, 1);
12229 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ipv4_checksum_ok,
12230 value->ipv4_csum_ok);
12235 set_integrity_bits(void *headers_m, void *headers_v,
12236 const struct rte_flow_item *integrity_item, bool is_l3_ip4)
12238 const struct rte_flow_item_integrity *spec = integrity_item->spec;
12239 const struct rte_flow_item_integrity *mask = integrity_item->mask;
12241 /* Integrity bits validation cleared spec pointer */
12242 MLX5_ASSERT(spec != NULL);
12244 mask = &rte_flow_item_integrity_mask;
12245 flow_dv_translate_integrity_l3(mask, spec, headers_m, headers_v,
12247 flow_dv_translate_integrity_l4(mask, spec, headers_m, headers_v);
12251 flow_dv_translate_item_integrity_post(void *matcher, void *key,
12253 struct rte_flow_item *integrity_items[2],
12254 uint64_t pattern_flags)
12256 void *headers_m, *headers_v;
12259 if (pattern_flags & MLX5_FLOW_ITEM_INNER_INTEGRITY) {
12260 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
12262 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
12263 is_l3_ip4 = (pattern_flags & MLX5_FLOW_LAYER_INNER_L3_IPV4) !=
12265 set_integrity_bits(headers_m, headers_v,
12266 integrity_items[1], is_l3_ip4);
12268 if (pattern_flags & MLX5_FLOW_ITEM_OUTER_INTEGRITY) {
12269 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
12271 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
12272 is_l3_ip4 = (pattern_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4) !=
12274 set_integrity_bits(headers_m, headers_v,
12275 integrity_items[0], is_l3_ip4);
12280 flow_dv_translate_item_integrity(const struct rte_flow_item *item,
12281 const struct rte_flow_item *integrity_items[2],
12282 uint64_t *last_item)
12284 const struct rte_flow_item_integrity *spec = (typeof(spec))item->spec;
12286 /* integrity bits validation cleared spec pointer */
12287 MLX5_ASSERT(spec != NULL);
12288 if (spec->level > 1) {
12289 integrity_items[1] = item;
12290 *last_item |= MLX5_FLOW_ITEM_INNER_INTEGRITY;
12292 integrity_items[0] = item;
12293 *last_item |= MLX5_FLOW_ITEM_OUTER_INTEGRITY;
12298 * Prepares DV flow counter with aging configuration.
12299 * Gets it by index when exists, creates a new one when doesn't.
12302 * Pointer to rte_eth_dev structure.
12303 * @param[in] dev_flow
12304 * Pointer to the mlx5_flow.
12305 * @param[in, out] flow
12306 * Pointer to the sub flow.
12308 * Pointer to the counter action configuration.
12310 * Pointer to the aging action configuration.
12311 * @param[out] error
12312 * Pointer to the error structure.
12315 * Pointer to the counter, NULL otherwise.
12317 static struct mlx5_flow_counter *
12318 flow_dv_prepare_counter(struct rte_eth_dev *dev,
12319 struct mlx5_flow *dev_flow,
12320 struct rte_flow *flow,
12321 const struct rte_flow_action_count *count,
12322 const struct rte_flow_action_age *age,
12323 struct rte_flow_error *error)
12325 if (!flow->counter) {
12326 flow->counter = flow_dv_translate_create_counter(dev, dev_flow,
12328 if (!flow->counter) {
12329 rte_flow_error_set(error, rte_errno,
12330 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12331 "cannot create counter object.");
12335 return flow_dv_counter_get_by_idx(dev, flow->counter, NULL);
12339 * Release an ASO CT action by its own device.
12342 * Pointer to the Ethernet device structure.
12344 * Index of ASO CT action to release.
12347 * 0 when CT action was removed, otherwise the number of references.
12350 flow_dv_aso_ct_dev_release(struct rte_eth_dev *dev, uint32_t idx)
12352 struct mlx5_priv *priv = dev->data->dev_private;
12353 struct mlx5_aso_ct_pools_mng *mng = priv->sh->ct_mng;
12355 struct mlx5_aso_ct_action *ct = flow_aso_ct_get_by_dev_idx(dev, idx);
12356 enum mlx5_aso_ct_state state =
12357 __atomic_load_n(&ct->state, __ATOMIC_RELAXED);
12359 /* Cannot release when CT is in the ASO SQ. */
12360 if (state == ASO_CONNTRACK_WAIT || state == ASO_CONNTRACK_QUERY)
12362 ret = __atomic_sub_fetch(&ct->refcnt, 1, __ATOMIC_RELAXED);
12364 if (ct->dr_action_orig) {
12365 #ifdef HAVE_MLX5_DR_ACTION_ASO_CT
12366 claim_zero(mlx5_glue->destroy_flow_action
12367 (ct->dr_action_orig));
12369 ct->dr_action_orig = NULL;
12371 if (ct->dr_action_rply) {
12372 #ifdef HAVE_MLX5_DR_ACTION_ASO_CT
12373 claim_zero(mlx5_glue->destroy_flow_action
12374 (ct->dr_action_rply));
12376 ct->dr_action_rply = NULL;
12378 /* Clear the state to free, no need in 1st allocation. */
12379 MLX5_ASO_CT_UPDATE_STATE(ct, ASO_CONNTRACK_FREE);
12380 rte_spinlock_lock(&mng->ct_sl);
12381 LIST_INSERT_HEAD(&mng->free_cts, ct, next);
12382 rte_spinlock_unlock(&mng->ct_sl);
12388 flow_dv_aso_ct_release(struct rte_eth_dev *dev, uint32_t own_idx,
12389 struct rte_flow_error *error)
12391 uint16_t owner = (uint16_t)MLX5_INDIRECT_ACT_CT_GET_OWNER(own_idx);
12392 uint32_t idx = MLX5_INDIRECT_ACT_CT_GET_IDX(own_idx);
12393 struct rte_eth_dev *owndev = &rte_eth_devices[owner];
12396 MLX5_ASSERT(owner < RTE_MAX_ETHPORTS);
12397 if (dev->data->dev_started != 1)
12398 return rte_flow_error_set(error, EAGAIN,
12399 RTE_FLOW_ERROR_TYPE_ACTION,
12401 "Indirect CT action cannot be destroyed when the port is stopped");
12402 ret = flow_dv_aso_ct_dev_release(owndev, idx);
12404 return rte_flow_error_set(error, EAGAIN,
12405 RTE_FLOW_ERROR_TYPE_ACTION,
12407 "Current state prevents indirect CT action from being destroyed");
12412 * Resize the ASO CT pools array by 64 pools.
12415 * Pointer to the Ethernet device structure.
12418 * 0 on success, otherwise negative errno value and rte_errno is set.
12421 flow_dv_aso_ct_pools_resize(struct rte_eth_dev *dev)
12423 struct mlx5_priv *priv = dev->data->dev_private;
12424 struct mlx5_aso_ct_pools_mng *mng = priv->sh->ct_mng;
12425 void *old_pools = mng->pools;
12426 /* Magic number now, need a macro. */
12427 uint32_t resize = mng->n + 64;
12428 uint32_t mem_size = sizeof(struct mlx5_aso_ct_pool *) * resize;
12429 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
12432 rte_errno = ENOMEM;
12435 rte_rwlock_write_lock(&mng->resize_rwl);
12436 /* ASO SQ/QP was already initialized in the startup. */
12438 /* Realloc could be an alternative choice. */
12439 rte_memcpy(pools, old_pools,
12440 mng->n * sizeof(struct mlx5_aso_ct_pool *));
12441 mlx5_free(old_pools);
12444 mng->pools = pools;
12445 rte_rwlock_write_unlock(&mng->resize_rwl);
12450 * Create and initialize a new ASO CT pool.
12453 * Pointer to the Ethernet device structure.
12454 * @param[out] ct_free
12455 * Where to put the pointer of a new CT action.
12458 * The CT actions pool pointer and @p ct_free is set on success,
12459 * NULL otherwise and rte_errno is set.
12461 static struct mlx5_aso_ct_pool *
12462 flow_dv_ct_pool_create(struct rte_eth_dev *dev,
12463 struct mlx5_aso_ct_action **ct_free)
12465 struct mlx5_priv *priv = dev->data->dev_private;
12466 struct mlx5_aso_ct_pools_mng *mng = priv->sh->ct_mng;
12467 struct mlx5_aso_ct_pool *pool = NULL;
12468 struct mlx5_devx_obj *obj = NULL;
12470 uint32_t log_obj_size = rte_log2_u32(MLX5_ASO_CT_ACTIONS_PER_POOL);
12472 obj = mlx5_devx_cmd_create_conn_track_offload_obj(priv->sh->cdev->ctx,
12473 priv->sh->cdev->pdn,
12476 rte_errno = ENODATA;
12477 DRV_LOG(ERR, "Failed to create conn_track_offload_obj using DevX.");
12480 pool = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*pool), 0, SOCKET_ID_ANY);
12482 rte_errno = ENOMEM;
12483 claim_zero(mlx5_devx_cmd_destroy(obj));
12486 pool->devx_obj = obj;
12487 pool->index = mng->next;
12488 /* Resize pools array if there is no room for the new pool in it. */
12489 if (pool->index == mng->n && flow_dv_aso_ct_pools_resize(dev)) {
12490 claim_zero(mlx5_devx_cmd_destroy(obj));
12494 mng->pools[pool->index] = pool;
12496 /* Assign the first action in the new pool, the rest go to free list. */
12497 *ct_free = &pool->actions[0];
12498 /* Lock outside, the list operation is safe here. */
12499 for (i = 1; i < MLX5_ASO_CT_ACTIONS_PER_POOL; i++) {
12500 /* refcnt is 0 when allocating the memory. */
12501 pool->actions[i].offset = i;
12502 LIST_INSERT_HEAD(&mng->free_cts, &pool->actions[i], next);
12508 * Allocate a ASO CT action from free list.
12511 * Pointer to the Ethernet device structure.
12512 * @param[out] error
12513 * Pointer to the error structure.
12516 * Index to ASO CT action on success, 0 otherwise and rte_errno is set.
12519 flow_dv_aso_ct_alloc(struct rte_eth_dev *dev, struct rte_flow_error *error)
12521 struct mlx5_priv *priv = dev->data->dev_private;
12522 struct mlx5_aso_ct_pools_mng *mng = priv->sh->ct_mng;
12523 struct mlx5_aso_ct_action *ct = NULL;
12524 struct mlx5_aso_ct_pool *pool;
12529 if (!priv->sh->cdev->config.devx) {
12530 rte_errno = ENOTSUP;
12533 /* Get a free CT action, if no, a new pool will be created. */
12534 rte_spinlock_lock(&mng->ct_sl);
12535 ct = LIST_FIRST(&mng->free_cts);
12537 LIST_REMOVE(ct, next);
12538 } else if (!flow_dv_ct_pool_create(dev, &ct)) {
12539 rte_spinlock_unlock(&mng->ct_sl);
12540 rte_flow_error_set(error, rte_errno, RTE_FLOW_ERROR_TYPE_ACTION,
12541 NULL, "failed to create ASO CT pool");
12544 rte_spinlock_unlock(&mng->ct_sl);
12545 pool = container_of(ct, struct mlx5_aso_ct_pool, actions[ct->offset]);
12546 ct_idx = MLX5_MAKE_CT_IDX(pool->index, ct->offset);
12547 /* 0: inactive, 1: created, 2+: used by flows. */
12548 __atomic_store_n(&ct->refcnt, 1, __ATOMIC_RELAXED);
12549 reg_c = mlx5_flow_get_reg_id(dev, MLX5_ASO_CONNTRACK, 0, error);
12550 if (!ct->dr_action_orig) {
12551 #ifdef HAVE_MLX5_DR_ACTION_ASO_CT
12552 ct->dr_action_orig = mlx5_glue->dv_create_flow_action_aso
12553 (priv->sh->rx_domain, pool->devx_obj->obj,
12555 MLX5DV_DR_ACTION_FLAGS_ASO_CT_DIRECTION_INITIATOR,
12558 RTE_SET_USED(reg_c);
12560 if (!ct->dr_action_orig) {
12561 flow_dv_aso_ct_dev_release(dev, ct_idx);
12562 rte_flow_error_set(error, rte_errno,
12563 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12564 "failed to create ASO CT action");
12568 if (!ct->dr_action_rply) {
12569 #ifdef HAVE_MLX5_DR_ACTION_ASO_CT
12570 ct->dr_action_rply = mlx5_glue->dv_create_flow_action_aso
12571 (priv->sh->rx_domain, pool->devx_obj->obj,
12573 MLX5DV_DR_ACTION_FLAGS_ASO_CT_DIRECTION_RESPONDER,
12576 if (!ct->dr_action_rply) {
12577 flow_dv_aso_ct_dev_release(dev, ct_idx);
12578 rte_flow_error_set(error, rte_errno,
12579 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12580 "failed to create ASO CT action");
12588 * Create a conntrack object with context and actions by using ASO mechanism.
12591 * Pointer to rte_eth_dev structure.
12593 * Pointer to conntrack information profile.
12594 * @param[out] error
12595 * Pointer to the error structure.
12598 * Index to conntrack object on success, 0 otherwise.
12601 flow_dv_translate_create_conntrack(struct rte_eth_dev *dev,
12602 const struct rte_flow_action_conntrack *pro,
12603 struct rte_flow_error *error)
12605 struct mlx5_priv *priv = dev->data->dev_private;
12606 struct mlx5_dev_ctx_shared *sh = priv->sh;
12607 struct mlx5_aso_ct_action *ct;
12610 if (!sh->ct_aso_en)
12611 return rte_flow_error_set(error, ENOTSUP,
12612 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12613 "Connection is not supported");
12614 idx = flow_dv_aso_ct_alloc(dev, error);
12616 return rte_flow_error_set(error, rte_errno,
12617 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12618 "Failed to allocate CT object");
12619 ct = flow_aso_ct_get_by_dev_idx(dev, idx);
12620 if (mlx5_aso_ct_update_by_wqe(sh, ct, pro))
12621 return rte_flow_error_set(error, EBUSY,
12622 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12623 "Failed to update CT");
12624 ct->is_original = !!pro->is_original_dir;
12625 ct->peer = pro->peer_port;
12630 * Fill the flow with DV spec, lock free
12631 * (mutex should be acquired by caller).
12634 * Pointer to rte_eth_dev structure.
12635 * @param[in, out] dev_flow
12636 * Pointer to the sub flow.
12638 * Pointer to the flow attributes.
12640 * Pointer to the list of items.
12641 * @param[in] actions
12642 * Pointer to the list of actions.
12643 * @param[out] error
12644 * Pointer to the error structure.
12647 * 0 on success, a negative errno value otherwise and rte_errno is set.
12650 flow_dv_translate(struct rte_eth_dev *dev,
12651 struct mlx5_flow *dev_flow,
12652 const struct rte_flow_attr *attr,
12653 const struct rte_flow_item items[],
12654 const struct rte_flow_action actions[],
12655 struct rte_flow_error *error)
12657 struct mlx5_priv *priv = dev->data->dev_private;
12658 struct mlx5_sh_config *dev_conf = &priv->sh->config;
12659 struct rte_flow *flow = dev_flow->flow;
12660 struct mlx5_flow_handle *handle = dev_flow->handle;
12661 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
12662 struct mlx5_flow_rss_desc *rss_desc;
12663 uint64_t item_flags = 0;
12664 uint64_t last_item = 0;
12665 uint64_t action_flags = 0;
12666 struct mlx5_flow_dv_matcher matcher = {
12668 .size = sizeof(matcher.mask.buf),
12672 bool actions_end = false;
12674 struct mlx5_flow_dv_modify_hdr_resource res;
12675 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
12676 sizeof(struct mlx5_modification_cmd) *
12677 (MLX5_MAX_MODIFY_NUM + 1)];
12679 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
12680 const struct rte_flow_action_count *count = NULL;
12681 const struct rte_flow_action_age *non_shared_age = NULL;
12682 union flow_dv_attr flow_attr = { .attr = 0 };
12684 union mlx5_flow_tbl_key tbl_key;
12685 uint32_t modify_action_position = UINT32_MAX;
12686 void *match_mask = matcher.mask.buf;
12687 void *match_value = dev_flow->dv.value.buf;
12688 uint8_t next_protocol = 0xff;
12689 struct rte_vlan_hdr vlan = { 0 };
12690 struct mlx5_flow_dv_dest_array_resource mdest_res;
12691 struct mlx5_flow_dv_sample_resource sample_res;
12692 void *sample_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
12693 const struct rte_flow_action_sample *sample = NULL;
12694 struct mlx5_flow_sub_actions_list *sample_act;
12695 uint32_t sample_act_pos = UINT32_MAX;
12696 uint32_t age_act_pos = UINT32_MAX;
12697 uint32_t num_of_dest = 0;
12698 int tmp_actions_n = 0;
12701 const struct mlx5_flow_tunnel *tunnel = NULL;
12702 struct flow_grp_info grp_info = {
12703 .external = !!dev_flow->external,
12704 .transfer = !!attr->transfer,
12705 .fdb_def_rule = !!priv->fdb_def_rule,
12706 .skip_scale = dev_flow->skip_scale &
12707 (1 << MLX5_SCALE_FLOW_GROUP_BIT),
12708 .std_tbl_fix = true,
12710 const struct rte_flow_item *integrity_items[2] = {NULL, NULL};
12711 const struct rte_flow_item *tunnel_item = NULL;
12714 return rte_flow_error_set(error, ENOMEM,
12715 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12717 "failed to push flow workspace");
12718 rss_desc = &wks->rss_desc;
12719 memset(&mdest_res, 0, sizeof(struct mlx5_flow_dv_dest_array_resource));
12720 memset(&sample_res, 0, sizeof(struct mlx5_flow_dv_sample_resource));
12721 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
12722 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
12723 /* update normal path action resource into last index of array */
12724 sample_act = &mdest_res.sample_act[MLX5_MAX_DEST_NUM - 1];
12725 if (is_tunnel_offload_active(dev)) {
12726 if (dev_flow->tunnel) {
12727 RTE_VERIFY(dev_flow->tof_type ==
12728 MLX5_TUNNEL_OFFLOAD_MISS_RULE);
12729 tunnel = dev_flow->tunnel;
12731 tunnel = mlx5_get_tof(items, actions,
12732 &dev_flow->tof_type);
12733 dev_flow->tunnel = tunnel;
12735 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
12736 (dev, attr, tunnel, dev_flow->tof_type);
12738 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
12739 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
12740 ret = mlx5_flow_group_to_table(dev, tunnel, attr->group, &table,
12744 dev_flow->dv.group = table;
12745 if (attr->transfer)
12746 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
12747 /* number of actions must be set to 0 in case of dirty stack. */
12748 mhdr_res->actions_num = 0;
12749 if (is_flow_tunnel_match_rule(dev_flow->tof_type)) {
12751 * do not add decap action if match rule drops packet
12752 * HW rejects rules with decap & drop
12754 * if tunnel match rule was inserted before matching tunnel set
12755 * rule flow table used in the match rule must be registered.
12756 * current implementation handles that in the
12757 * flow_dv_match_register() at the function end.
12759 bool add_decap = true;
12760 const struct rte_flow_action *ptr = actions;
12762 for (; ptr->type != RTE_FLOW_ACTION_TYPE_END; ptr++) {
12763 if (ptr->type == RTE_FLOW_ACTION_TYPE_DROP) {
12769 if (flow_dv_create_action_l2_decap(dev, dev_flow,
12773 dev_flow->dv.actions[actions_n++] =
12774 dev_flow->dv.encap_decap->action;
12775 action_flags |= MLX5_FLOW_ACTION_DECAP;
12778 for (; !actions_end ; actions++) {
12779 const struct rte_flow_action_queue *queue;
12780 const struct rte_flow_action_rss *rss;
12781 const struct rte_flow_action *action = actions;
12782 const uint8_t *rss_key;
12783 struct mlx5_flow_tbl_resource *tbl;
12784 struct mlx5_aso_age_action *age_act;
12785 struct mlx5_flow_counter *cnt_act;
12786 uint32_t port_id = 0;
12787 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
12788 int action_type = actions->type;
12789 const struct rte_flow_action *found_action = NULL;
12790 uint32_t jump_group = 0;
12791 uint32_t owner_idx;
12792 struct mlx5_aso_ct_action *ct;
12794 if (!mlx5_flow_os_action_supported(action_type))
12795 return rte_flow_error_set(error, ENOTSUP,
12796 RTE_FLOW_ERROR_TYPE_ACTION,
12798 "action not supported");
12799 switch (action_type) {
12800 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
12801 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
12803 case RTE_FLOW_ACTION_TYPE_VOID:
12805 case RTE_FLOW_ACTION_TYPE_PORT_ID:
12806 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
12807 if (flow_dv_translate_action_port_id(dev, action,
12810 port_id_resource.port_id = port_id;
12811 MLX5_ASSERT(!handle->rix_port_id_action);
12812 if (flow_dv_port_id_action_resource_register
12813 (dev, &port_id_resource, dev_flow, error))
12815 dev_flow->dv.actions[actions_n++] =
12816 dev_flow->dv.port_id_action->action;
12817 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
12818 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
12819 sample_act->action_flags |= MLX5_FLOW_ACTION_PORT_ID;
12822 case RTE_FLOW_ACTION_TYPE_FLAG:
12823 action_flags |= MLX5_FLOW_ACTION_FLAG;
12825 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
12826 struct rte_flow_action_mark mark = {
12827 .id = MLX5_FLOW_MARK_DEFAULT,
12830 if (flow_dv_convert_action_mark(dev, &mark,
12834 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
12837 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
12839 * Only one FLAG or MARK is supported per device flow
12840 * right now. So the pointer to the tag resource must be
12841 * zero before the register process.
12843 MLX5_ASSERT(!handle->dvh.rix_tag);
12844 if (flow_dv_tag_resource_register(dev, tag_be,
12847 MLX5_ASSERT(dev_flow->dv.tag_resource);
12848 dev_flow->dv.actions[actions_n++] =
12849 dev_flow->dv.tag_resource->action;
12851 case RTE_FLOW_ACTION_TYPE_MARK:
12852 action_flags |= MLX5_FLOW_ACTION_MARK;
12854 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
12855 const struct rte_flow_action_mark *mark =
12856 (const struct rte_flow_action_mark *)
12859 if (flow_dv_convert_action_mark(dev, mark,
12863 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
12867 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
12868 /* Legacy (non-extensive) MARK action. */
12869 tag_be = mlx5_flow_mark_set
12870 (((const struct rte_flow_action_mark *)
12871 (actions->conf))->id);
12872 MLX5_ASSERT(!handle->dvh.rix_tag);
12873 if (flow_dv_tag_resource_register(dev, tag_be,
12876 MLX5_ASSERT(dev_flow->dv.tag_resource);
12877 dev_flow->dv.actions[actions_n++] =
12878 dev_flow->dv.tag_resource->action;
12880 case RTE_FLOW_ACTION_TYPE_SET_META:
12881 if (flow_dv_convert_action_set_meta
12882 (dev, mhdr_res, attr,
12883 (const struct rte_flow_action_set_meta *)
12884 actions->conf, error))
12886 action_flags |= MLX5_FLOW_ACTION_SET_META;
12888 case RTE_FLOW_ACTION_TYPE_SET_TAG:
12889 if (flow_dv_convert_action_set_tag
12891 (const struct rte_flow_action_set_tag *)
12892 actions->conf, error))
12894 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
12896 case RTE_FLOW_ACTION_TYPE_DROP:
12897 action_flags |= MLX5_FLOW_ACTION_DROP;
12898 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
12900 case RTE_FLOW_ACTION_TYPE_QUEUE:
12901 queue = actions->conf;
12902 rss_desc->queue_num = 1;
12903 rss_desc->queue[0] = queue->index;
12904 action_flags |= MLX5_FLOW_ACTION_QUEUE;
12905 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
12906 sample_act->action_flags |= MLX5_FLOW_ACTION_QUEUE;
12909 case RTE_FLOW_ACTION_TYPE_RSS:
12910 rss = actions->conf;
12911 memcpy(rss_desc->queue, rss->queue,
12912 rss->queue_num * sizeof(uint16_t));
12913 rss_desc->queue_num = rss->queue_num;
12914 /* NULL RSS key indicates default RSS key. */
12915 rss_key = !rss->key ? rss_hash_default_key : rss->key;
12916 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
12918 * rss->level and rss.types should be set in advance
12919 * when expanding items for RSS.
12921 action_flags |= MLX5_FLOW_ACTION_RSS;
12922 dev_flow->handle->fate_action = rss_desc->shared_rss ?
12923 MLX5_FLOW_FATE_SHARED_RSS :
12924 MLX5_FLOW_FATE_QUEUE;
12926 case MLX5_RTE_FLOW_ACTION_TYPE_AGE:
12927 owner_idx = (uint32_t)(uintptr_t)action->conf;
12928 age_act = flow_aso_age_get_by_idx(dev, owner_idx);
12929 if (flow->age == 0) {
12930 flow->age = owner_idx;
12931 __atomic_fetch_add(&age_act->refcnt, 1,
12934 age_act_pos = actions_n++;
12935 action_flags |= MLX5_FLOW_ACTION_AGE;
12937 case RTE_FLOW_ACTION_TYPE_AGE:
12938 non_shared_age = action->conf;
12939 age_act_pos = actions_n++;
12940 action_flags |= MLX5_FLOW_ACTION_AGE;
12942 case MLX5_RTE_FLOW_ACTION_TYPE_COUNT:
12943 owner_idx = (uint32_t)(uintptr_t)action->conf;
12944 cnt_act = flow_dv_counter_get_by_idx(dev, owner_idx,
12946 MLX5_ASSERT(cnt_act != NULL);
12948 * When creating meter drop flow in drop table, the
12949 * counter should not overwrite the rte flow counter.
12951 if (attr->group == MLX5_FLOW_TABLE_LEVEL_METER &&
12952 dev_flow->dv.table_id == MLX5_MTR_TABLE_ID_DROP) {
12953 dev_flow->dv.actions[actions_n++] =
12956 if (flow->counter == 0) {
12957 flow->counter = owner_idx;
12959 (&cnt_act->shared_info.refcnt,
12960 1, __ATOMIC_RELAXED);
12962 /* Save information first, will apply later. */
12963 action_flags |= MLX5_FLOW_ACTION_COUNT;
12966 case RTE_FLOW_ACTION_TYPE_COUNT:
12967 if (!priv->sh->cdev->config.devx) {
12968 return rte_flow_error_set
12970 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12972 "count action not supported");
12974 /* Save information first, will apply later. */
12975 count = action->conf;
12976 action_flags |= MLX5_FLOW_ACTION_COUNT;
12978 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
12979 dev_flow->dv.actions[actions_n++] =
12980 priv->sh->pop_vlan_action;
12981 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
12983 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
12984 if (!(action_flags &
12985 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
12986 flow_dev_get_vlan_info_from_items(items, &vlan);
12987 vlan.eth_proto = rte_be_to_cpu_16
12988 ((((const struct rte_flow_action_of_push_vlan *)
12989 actions->conf)->ethertype));
12990 found_action = mlx5_flow_find_action
12992 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
12994 mlx5_update_vlan_vid_pcp(found_action, &vlan);
12995 found_action = mlx5_flow_find_action
12997 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
12999 mlx5_update_vlan_vid_pcp(found_action, &vlan);
13000 if (flow_dv_create_action_push_vlan
13001 (dev, attr, &vlan, dev_flow, error))
13003 dev_flow->dv.actions[actions_n++] =
13004 dev_flow->dv.push_vlan_res->action;
13005 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
13007 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
13008 /* of_vlan_push action handled this action */
13009 MLX5_ASSERT(action_flags &
13010 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
13012 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
13013 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
13015 flow_dev_get_vlan_info_from_items(items, &vlan);
13016 mlx5_update_vlan_vid_pcp(actions, &vlan);
13017 /* If no VLAN push - this is a modify header action */
13018 if (flow_dv_convert_action_modify_vlan_vid
13019 (mhdr_res, actions, error))
13021 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
13023 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
13024 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
13025 if (flow_dv_create_action_l2_encap(dev, actions,
13030 dev_flow->dv.actions[actions_n++] =
13031 dev_flow->dv.encap_decap->action;
13032 action_flags |= MLX5_FLOW_ACTION_ENCAP;
13033 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
13034 sample_act->action_flags |=
13035 MLX5_FLOW_ACTION_ENCAP;
13037 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
13038 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
13039 if (flow_dv_create_action_l2_decap(dev, dev_flow,
13043 dev_flow->dv.actions[actions_n++] =
13044 dev_flow->dv.encap_decap->action;
13045 action_flags |= MLX5_FLOW_ACTION_DECAP;
13047 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
13048 /* Handle encap with preceding decap. */
13049 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
13050 if (flow_dv_create_action_raw_encap
13051 (dev, actions, dev_flow, attr, error))
13053 dev_flow->dv.actions[actions_n++] =
13054 dev_flow->dv.encap_decap->action;
13056 /* Handle encap without preceding decap. */
13057 if (flow_dv_create_action_l2_encap
13058 (dev, actions, dev_flow, attr->transfer,
13061 dev_flow->dv.actions[actions_n++] =
13062 dev_flow->dv.encap_decap->action;
13064 action_flags |= MLX5_FLOW_ACTION_ENCAP;
13065 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
13066 sample_act->action_flags |=
13067 MLX5_FLOW_ACTION_ENCAP;
13069 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
13070 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
13072 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
13073 if (flow_dv_create_action_l2_decap
13074 (dev, dev_flow, attr->transfer, error))
13076 dev_flow->dv.actions[actions_n++] =
13077 dev_flow->dv.encap_decap->action;
13079 /* If decap is followed by encap, handle it at encap. */
13080 action_flags |= MLX5_FLOW_ACTION_DECAP;
13082 case MLX5_RTE_FLOW_ACTION_TYPE_JUMP:
13083 dev_flow->dv.actions[actions_n++] =
13084 (void *)(uintptr_t)action->conf;
13085 action_flags |= MLX5_FLOW_ACTION_JUMP;
13087 case RTE_FLOW_ACTION_TYPE_JUMP:
13088 jump_group = ((const struct rte_flow_action_jump *)
13089 action->conf)->group;
13090 grp_info.std_tbl_fix = 0;
13091 if (dev_flow->skip_scale &
13092 (1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT))
13093 grp_info.skip_scale = 1;
13095 grp_info.skip_scale = 0;
13096 ret = mlx5_flow_group_to_table(dev, tunnel,
13102 tbl = flow_dv_tbl_resource_get(dev, table, attr->egress,
13104 !!dev_flow->external,
13105 tunnel, jump_group, 0,
13108 return rte_flow_error_set
13110 RTE_FLOW_ERROR_TYPE_ACTION,
13112 "cannot create jump action.");
13113 if (flow_dv_jump_tbl_resource_register
13114 (dev, tbl, dev_flow, error)) {
13115 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
13116 return rte_flow_error_set
13118 RTE_FLOW_ERROR_TYPE_ACTION,
13120 "cannot create jump action.");
13122 dev_flow->dv.actions[actions_n++] =
13123 dev_flow->dv.jump->action;
13124 action_flags |= MLX5_FLOW_ACTION_JUMP;
13125 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
13126 sample_act->action_flags |= MLX5_FLOW_ACTION_JUMP;
13129 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
13130 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
13131 if (flow_dv_convert_action_modify_mac
13132 (mhdr_res, actions, error))
13134 action_flags |= actions->type ==
13135 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
13136 MLX5_FLOW_ACTION_SET_MAC_SRC :
13137 MLX5_FLOW_ACTION_SET_MAC_DST;
13139 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
13140 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
13141 if (flow_dv_convert_action_modify_ipv4
13142 (mhdr_res, actions, error))
13144 action_flags |= actions->type ==
13145 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
13146 MLX5_FLOW_ACTION_SET_IPV4_SRC :
13147 MLX5_FLOW_ACTION_SET_IPV4_DST;
13149 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
13150 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
13151 if (flow_dv_convert_action_modify_ipv6
13152 (mhdr_res, actions, error))
13154 action_flags |= actions->type ==
13155 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
13156 MLX5_FLOW_ACTION_SET_IPV6_SRC :
13157 MLX5_FLOW_ACTION_SET_IPV6_DST;
13159 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
13160 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
13161 if (flow_dv_convert_action_modify_tp
13162 (mhdr_res, actions, items,
13163 &flow_attr, dev_flow, !!(action_flags &
13164 MLX5_FLOW_ACTION_DECAP), error))
13166 action_flags |= actions->type ==
13167 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
13168 MLX5_FLOW_ACTION_SET_TP_SRC :
13169 MLX5_FLOW_ACTION_SET_TP_DST;
13171 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
13172 if (flow_dv_convert_action_modify_dec_ttl
13173 (mhdr_res, items, &flow_attr, dev_flow,
13175 MLX5_FLOW_ACTION_DECAP), error))
13177 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
13179 case RTE_FLOW_ACTION_TYPE_SET_TTL:
13180 if (flow_dv_convert_action_modify_ttl
13181 (mhdr_res, actions, items, &flow_attr,
13182 dev_flow, !!(action_flags &
13183 MLX5_FLOW_ACTION_DECAP), error))
13185 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
13187 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
13188 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
13189 if (flow_dv_convert_action_modify_tcp_seq
13190 (mhdr_res, actions, error))
13192 action_flags |= actions->type ==
13193 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
13194 MLX5_FLOW_ACTION_INC_TCP_SEQ :
13195 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
13198 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
13199 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
13200 if (flow_dv_convert_action_modify_tcp_ack
13201 (mhdr_res, actions, error))
13203 action_flags |= actions->type ==
13204 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
13205 MLX5_FLOW_ACTION_INC_TCP_ACK :
13206 MLX5_FLOW_ACTION_DEC_TCP_ACK;
13208 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
13209 if (flow_dv_convert_action_set_reg
13210 (mhdr_res, actions, error))
13212 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
13214 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
13215 if (flow_dv_convert_action_copy_mreg
13216 (dev, mhdr_res, actions, error))
13218 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
13220 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
13221 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
13222 dev_flow->handle->fate_action =
13223 MLX5_FLOW_FATE_DEFAULT_MISS;
13225 case RTE_FLOW_ACTION_TYPE_METER:
13227 return rte_flow_error_set(error, rte_errno,
13228 RTE_FLOW_ERROR_TYPE_ACTION,
13229 NULL, "Failed to get meter in flow.");
13230 /* Set the meter action. */
13231 dev_flow->dv.actions[actions_n++] =
13232 wks->fm->meter_action;
13233 action_flags |= MLX5_FLOW_ACTION_METER;
13235 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
13236 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
13239 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
13241 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
13242 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
13245 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
13247 case RTE_FLOW_ACTION_TYPE_SAMPLE:
13248 sample_act_pos = actions_n;
13249 sample = (const struct rte_flow_action_sample *)
13252 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
13253 /* put encap action into group if work with port id */
13254 if ((action_flags & MLX5_FLOW_ACTION_ENCAP) &&
13255 (action_flags & MLX5_FLOW_ACTION_PORT_ID))
13256 sample_act->action_flags |=
13257 MLX5_FLOW_ACTION_ENCAP;
13259 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
13260 if (flow_dv_convert_action_modify_field
13261 (dev, mhdr_res, actions, attr, error))
13263 action_flags |= MLX5_FLOW_ACTION_MODIFY_FIELD;
13265 case RTE_FLOW_ACTION_TYPE_CONNTRACK:
13266 owner_idx = (uint32_t)(uintptr_t)action->conf;
13267 ct = flow_aso_ct_get_by_idx(dev, owner_idx);
13269 return rte_flow_error_set(error, EINVAL,
13270 RTE_FLOW_ERROR_TYPE_ACTION,
13272 "Failed to get CT object.");
13273 if (mlx5_aso_ct_available(priv->sh, ct))
13274 return rte_flow_error_set(error, rte_errno,
13275 RTE_FLOW_ERROR_TYPE_ACTION,
13277 "CT is unavailable.");
13278 if (ct->is_original)
13279 dev_flow->dv.actions[actions_n] =
13280 ct->dr_action_orig;
13282 dev_flow->dv.actions[actions_n] =
13283 ct->dr_action_rply;
13284 if (flow->ct == 0) {
13285 flow->indirect_type =
13286 MLX5_INDIRECT_ACTION_TYPE_CT;
13287 flow->ct = owner_idx;
13288 __atomic_fetch_add(&ct->refcnt, 1,
13292 action_flags |= MLX5_FLOW_ACTION_CT;
13294 case RTE_FLOW_ACTION_TYPE_END:
13295 actions_end = true;
13296 if (mhdr_res->actions_num) {
13297 /* create modify action if needed. */
13298 if (flow_dv_modify_hdr_resource_register
13299 (dev, mhdr_res, dev_flow, error))
13301 dev_flow->dv.actions[modify_action_position] =
13302 handle->dvh.modify_hdr->action;
13305 * Handle AGE and COUNT action by single HW counter
13306 * when they are not shared.
13308 if (action_flags & MLX5_FLOW_ACTION_AGE) {
13309 if ((non_shared_age && count) ||
13310 !(priv->sh->flow_hit_aso_en &&
13311 (attr->group || attr->transfer))) {
13312 /* Creates age by counters. */
13313 cnt_act = flow_dv_prepare_counter
13320 dev_flow->dv.actions[age_act_pos] =
13324 if (!flow->age && non_shared_age) {
13325 flow->age = flow_dv_aso_age_alloc
13329 flow_dv_aso_age_params_init
13331 non_shared_age->context ?
13332 non_shared_age->context :
13333 (void *)(uintptr_t)
13334 (dev_flow->flow_idx),
13335 non_shared_age->timeout);
13337 age_act = flow_aso_age_get_by_idx(dev,
13339 dev_flow->dv.actions[age_act_pos] =
13340 age_act->dr_action;
13342 if (action_flags & MLX5_FLOW_ACTION_COUNT) {
13344 * Create one count action, to be used
13345 * by all sub-flows.
13347 cnt_act = flow_dv_prepare_counter(dev, dev_flow,
13352 dev_flow->dv.actions[actions_n++] =
13358 if (mhdr_res->actions_num &&
13359 modify_action_position == UINT32_MAX)
13360 modify_action_position = actions_n++;
13362 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
13363 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
13364 int item_type = items->type;
13366 if (!mlx5_flow_os_item_supported(item_type))
13367 return rte_flow_error_set(error, ENOTSUP,
13368 RTE_FLOW_ERROR_TYPE_ITEM,
13369 NULL, "item not supported");
13370 switch (item_type) {
13371 case RTE_FLOW_ITEM_TYPE_PORT_ID:
13372 flow_dv_translate_item_port_id
13373 (dev, match_mask, match_value, items, attr);
13374 last_item = MLX5_FLOW_ITEM_PORT_ID;
13376 case RTE_FLOW_ITEM_TYPE_ETH:
13377 flow_dv_translate_item_eth(match_mask, match_value,
13379 dev_flow->dv.group);
13380 matcher.priority = action_flags &
13381 MLX5_FLOW_ACTION_DEFAULT_MISS &&
13382 !dev_flow->external ?
13383 MLX5_PRIORITY_MAP_L3 :
13384 MLX5_PRIORITY_MAP_L2;
13385 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
13386 MLX5_FLOW_LAYER_OUTER_L2;
13388 case RTE_FLOW_ITEM_TYPE_VLAN:
13389 flow_dv_translate_item_vlan(dev_flow,
13390 match_mask, match_value,
13392 dev_flow->dv.group);
13393 matcher.priority = MLX5_PRIORITY_MAP_L2;
13394 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
13395 MLX5_FLOW_LAYER_INNER_VLAN) :
13396 (MLX5_FLOW_LAYER_OUTER_L2 |
13397 MLX5_FLOW_LAYER_OUTER_VLAN);
13399 case RTE_FLOW_ITEM_TYPE_IPV4:
13400 mlx5_flow_tunnel_ip_check(items, next_protocol,
13401 &item_flags, &tunnel);
13402 flow_dv_translate_item_ipv4(match_mask, match_value,
13404 dev_flow->dv.group);
13405 matcher.priority = MLX5_PRIORITY_MAP_L3;
13406 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
13407 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
13408 if (items->mask != NULL &&
13409 ((const struct rte_flow_item_ipv4 *)
13410 items->mask)->hdr.next_proto_id) {
13412 ((const struct rte_flow_item_ipv4 *)
13413 (items->spec))->hdr.next_proto_id;
13415 ((const struct rte_flow_item_ipv4 *)
13416 (items->mask))->hdr.next_proto_id;
13418 /* Reset for inner layer. */
13419 next_protocol = 0xff;
13422 case RTE_FLOW_ITEM_TYPE_IPV6:
13423 mlx5_flow_tunnel_ip_check(items, next_protocol,
13424 &item_flags, &tunnel);
13425 flow_dv_translate_item_ipv6(match_mask, match_value,
13427 dev_flow->dv.group);
13428 matcher.priority = MLX5_PRIORITY_MAP_L3;
13429 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
13430 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
13431 if (items->mask != NULL &&
13432 ((const struct rte_flow_item_ipv6 *)
13433 items->mask)->hdr.proto) {
13435 ((const struct rte_flow_item_ipv6 *)
13436 items->spec)->hdr.proto;
13438 ((const struct rte_flow_item_ipv6 *)
13439 items->mask)->hdr.proto;
13441 /* Reset for inner layer. */
13442 next_protocol = 0xff;
13445 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
13446 flow_dv_translate_item_ipv6_frag_ext(match_mask,
13449 last_item = tunnel ?
13450 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
13451 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
13452 if (items->mask != NULL &&
13453 ((const struct rte_flow_item_ipv6_frag_ext *)
13454 items->mask)->hdr.next_header) {
13456 ((const struct rte_flow_item_ipv6_frag_ext *)
13457 items->spec)->hdr.next_header;
13459 ((const struct rte_flow_item_ipv6_frag_ext *)
13460 items->mask)->hdr.next_header;
13462 /* Reset for inner layer. */
13463 next_protocol = 0xff;
13466 case RTE_FLOW_ITEM_TYPE_TCP:
13467 flow_dv_translate_item_tcp(match_mask, match_value,
13469 matcher.priority = MLX5_PRIORITY_MAP_L4;
13470 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
13471 MLX5_FLOW_LAYER_OUTER_L4_TCP;
13473 case RTE_FLOW_ITEM_TYPE_UDP:
13474 flow_dv_translate_item_udp(match_mask, match_value,
13476 matcher.priority = MLX5_PRIORITY_MAP_L4;
13477 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
13478 MLX5_FLOW_LAYER_OUTER_L4_UDP;
13480 case RTE_FLOW_ITEM_TYPE_GRE:
13481 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13482 last_item = MLX5_FLOW_LAYER_GRE;
13483 tunnel_item = items;
13485 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
13486 flow_dv_translate_item_gre_key(match_mask,
13487 match_value, items);
13488 last_item = MLX5_FLOW_LAYER_GRE_KEY;
13490 case RTE_FLOW_ITEM_TYPE_NVGRE:
13491 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13492 last_item = MLX5_FLOW_LAYER_GRE;
13493 tunnel_item = items;
13495 case RTE_FLOW_ITEM_TYPE_VXLAN:
13496 flow_dv_translate_item_vxlan(dev, attr,
13497 match_mask, match_value,
13499 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13500 last_item = MLX5_FLOW_LAYER_VXLAN;
13502 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
13503 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13504 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
13505 tunnel_item = items;
13507 case RTE_FLOW_ITEM_TYPE_GENEVE:
13508 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13509 last_item = MLX5_FLOW_LAYER_GENEVE;
13510 tunnel_item = items;
13512 case RTE_FLOW_ITEM_TYPE_GENEVE_OPT:
13513 ret = flow_dv_translate_item_geneve_opt(dev, match_mask,
13517 return rte_flow_error_set(error, -ret,
13518 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
13519 "cannot create GENEVE TLV option");
13520 flow->geneve_tlv_option = 1;
13521 last_item = MLX5_FLOW_LAYER_GENEVE_OPT;
13523 case RTE_FLOW_ITEM_TYPE_MPLS:
13524 flow_dv_translate_item_mpls(match_mask, match_value,
13525 items, last_item, tunnel);
13526 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13527 last_item = MLX5_FLOW_LAYER_MPLS;
13529 case RTE_FLOW_ITEM_TYPE_MARK:
13530 flow_dv_translate_item_mark(dev, match_mask,
13531 match_value, items);
13532 last_item = MLX5_FLOW_ITEM_MARK;
13534 case RTE_FLOW_ITEM_TYPE_META:
13535 flow_dv_translate_item_meta(dev, match_mask,
13536 match_value, attr, items);
13537 last_item = MLX5_FLOW_ITEM_METADATA;
13539 case RTE_FLOW_ITEM_TYPE_ICMP:
13540 flow_dv_translate_item_icmp(match_mask, match_value,
13542 last_item = MLX5_FLOW_LAYER_ICMP;
13544 case RTE_FLOW_ITEM_TYPE_ICMP6:
13545 flow_dv_translate_item_icmp6(match_mask, match_value,
13547 last_item = MLX5_FLOW_LAYER_ICMP6;
13549 case RTE_FLOW_ITEM_TYPE_TAG:
13550 flow_dv_translate_item_tag(dev, match_mask,
13551 match_value, items);
13552 last_item = MLX5_FLOW_ITEM_TAG;
13554 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
13555 flow_dv_translate_mlx5_item_tag(dev, match_mask,
13556 match_value, items);
13557 last_item = MLX5_FLOW_ITEM_TAG;
13559 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
13560 flow_dv_translate_item_tx_queue(dev, match_mask,
13563 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
13565 case RTE_FLOW_ITEM_TYPE_GTP:
13566 flow_dv_translate_item_gtp(match_mask, match_value,
13568 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13569 last_item = MLX5_FLOW_LAYER_GTP;
13571 case RTE_FLOW_ITEM_TYPE_GTP_PSC:
13572 ret = flow_dv_translate_item_gtp_psc(match_mask,
13576 return rte_flow_error_set(error, -ret,
13577 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
13578 "cannot create GTP PSC item");
13579 last_item = MLX5_FLOW_LAYER_GTP_PSC;
13581 case RTE_FLOW_ITEM_TYPE_ECPRI:
13582 if (!mlx5_flex_parser_ecpri_exist(dev)) {
13583 /* Create it only the first time to be used. */
13584 ret = mlx5_flex_parser_ecpri_alloc(dev);
13586 return rte_flow_error_set
13588 RTE_FLOW_ERROR_TYPE_ITEM,
13590 "cannot create eCPRI parser");
13592 flow_dv_translate_item_ecpri(dev, match_mask,
13593 match_value, items,
13595 /* No other protocol should follow eCPRI layer. */
13596 last_item = MLX5_FLOW_LAYER_ECPRI;
13598 case RTE_FLOW_ITEM_TYPE_INTEGRITY:
13599 flow_dv_translate_item_integrity(items, integrity_items,
13602 case RTE_FLOW_ITEM_TYPE_CONNTRACK:
13603 flow_dv_translate_item_aso_ct(dev, match_mask,
13604 match_value, items);
13606 case RTE_FLOW_ITEM_TYPE_FLEX:
13607 flow_dv_translate_item_flex(dev, match_mask,
13608 match_value, items,
13609 dev_flow, tunnel != 0);
13610 last_item = tunnel ? MLX5_FLOW_ITEM_INNER_FLEX :
13611 MLX5_FLOW_ITEM_OUTER_FLEX;
13616 item_flags |= last_item;
13619 * When E-Switch mode is enabled, we have two cases where we need to
13620 * set the source port manually.
13621 * The first one, is in case of Nic steering rule, and the second is
13622 * E-Switch rule where no port_id item was found. In both cases
13623 * the source port is set according the current port in use.
13625 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) && priv->sh->esw_mode) {
13626 if (flow_dv_translate_item_port_id(dev, match_mask,
13627 match_value, NULL, attr))
13630 if (item_flags & MLX5_FLOW_ITEM_INTEGRITY) {
13631 flow_dv_translate_item_integrity_post(match_mask, match_value,
13635 if (item_flags & MLX5_FLOW_LAYER_VXLAN_GPE)
13636 flow_dv_translate_item_vxlan_gpe(match_mask, match_value,
13637 tunnel_item, item_flags);
13638 else if (item_flags & MLX5_FLOW_LAYER_GENEVE)
13639 flow_dv_translate_item_geneve(match_mask, match_value,
13640 tunnel_item, item_flags);
13641 else if (item_flags & MLX5_FLOW_LAYER_GRE) {
13642 if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_GRE)
13643 flow_dv_translate_item_gre(match_mask, match_value,
13644 tunnel_item, item_flags);
13645 else if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_NVGRE)
13646 flow_dv_translate_item_nvgre(match_mask, match_value,
13647 tunnel_item, item_flags);
13649 MLX5_ASSERT(false);
13651 #ifdef RTE_LIBRTE_MLX5_DEBUG
13652 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
13653 dev_flow->dv.value.buf));
13656 * Layers may be already initialized from prefix flow if this dev_flow
13657 * is the suffix flow.
13659 handle->layers |= item_flags;
13660 if (action_flags & MLX5_FLOW_ACTION_RSS)
13661 flow_dv_hashfields_set(dev_flow->handle->layers,
13663 &dev_flow->hash_fields);
13664 /* If has RSS action in the sample action, the Sample/Mirror resource
13665 * should be registered after the hash filed be update.
13667 if (action_flags & MLX5_FLOW_ACTION_SAMPLE) {
13668 ret = flow_dv_translate_action_sample(dev,
13677 ret = flow_dv_create_action_sample(dev,
13686 return rte_flow_error_set
13688 RTE_FLOW_ERROR_TYPE_ACTION,
13690 "cannot create sample action");
13691 if (num_of_dest > 1) {
13692 dev_flow->dv.actions[sample_act_pos] =
13693 dev_flow->dv.dest_array_res->action;
13695 dev_flow->dv.actions[sample_act_pos] =
13696 dev_flow->dv.sample_res->verbs_action;
13700 * For multiple destination (sample action with ratio=1), the encap
13701 * action and port id action will be combined into group action.
13702 * So need remove the original these actions in the flow and only
13703 * use the sample action instead of.
13705 if (num_of_dest > 1 &&
13706 (sample_act->dr_port_id_action || sample_act->dr_jump_action)) {
13708 void *temp_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
13710 for (i = 0; i < actions_n; i++) {
13711 if ((sample_act->dr_encap_action &&
13712 sample_act->dr_encap_action ==
13713 dev_flow->dv.actions[i]) ||
13714 (sample_act->dr_port_id_action &&
13715 sample_act->dr_port_id_action ==
13716 dev_flow->dv.actions[i]) ||
13717 (sample_act->dr_jump_action &&
13718 sample_act->dr_jump_action ==
13719 dev_flow->dv.actions[i]))
13721 temp_actions[tmp_actions_n++] = dev_flow->dv.actions[i];
13723 memcpy((void *)dev_flow->dv.actions,
13724 (void *)temp_actions,
13725 tmp_actions_n * sizeof(void *));
13726 actions_n = tmp_actions_n;
13728 dev_flow->dv.actions_n = actions_n;
13729 dev_flow->act_flags = action_flags;
13730 if (wks->skip_matcher_reg)
13732 /* Register matcher. */
13733 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
13734 matcher.mask.size);
13735 matcher.priority = mlx5_get_matcher_priority(dev, attr,
13737 dev_flow->external);
13739 * When creating meter drop flow in drop table, using original
13740 * 5-tuple match, the matcher priority should be lower than
13743 if (attr->group == MLX5_FLOW_TABLE_LEVEL_METER &&
13744 dev_flow->dv.table_id == MLX5_MTR_TABLE_ID_DROP &&
13745 matcher.priority <= MLX5_REG_BITS)
13746 matcher.priority += MLX5_REG_BITS;
13747 /* reserved field no needs to be set to 0 here. */
13748 tbl_key.is_fdb = attr->transfer;
13749 tbl_key.is_egress = attr->egress;
13750 tbl_key.level = dev_flow->dv.group;
13751 tbl_key.id = dev_flow->dv.table_id;
13752 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow,
13753 tunnel, attr->group, error))
13759 * Set hash RX queue by hash fields (see enum ibv_rx_hash_fields)
13762 * @param[in, out] action
13763 * Shred RSS action holding hash RX queue objects.
13764 * @param[in] hash_fields
13765 * Defines combination of packet fields to participate in RX hash.
13766 * @param[in] tunnel
13768 * @param[in] hrxq_idx
13769 * Hash RX queue index to set.
13772 * 0 on success, otherwise negative errno value.
13775 __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
13776 const uint64_t hash_fields,
13779 uint32_t *hrxqs = action->hrxq;
13781 switch (hash_fields & ~IBV_RX_HASH_INNER) {
13782 case MLX5_RSS_HASH_IPV4:
13783 /* fall-through. */
13784 case MLX5_RSS_HASH_IPV4_DST_ONLY:
13785 /* fall-through. */
13786 case MLX5_RSS_HASH_IPV4_SRC_ONLY:
13787 hrxqs[0] = hrxq_idx;
13789 case MLX5_RSS_HASH_IPV4_TCP:
13790 /* fall-through. */
13791 case MLX5_RSS_HASH_IPV4_TCP_DST_ONLY:
13792 /* fall-through. */
13793 case MLX5_RSS_HASH_IPV4_TCP_SRC_ONLY:
13794 hrxqs[1] = hrxq_idx;
13796 case MLX5_RSS_HASH_IPV4_UDP:
13797 /* fall-through. */
13798 case MLX5_RSS_HASH_IPV4_UDP_DST_ONLY:
13799 /* fall-through. */
13800 case MLX5_RSS_HASH_IPV4_UDP_SRC_ONLY:
13801 hrxqs[2] = hrxq_idx;
13803 case MLX5_RSS_HASH_IPV6:
13804 /* fall-through. */
13805 case MLX5_RSS_HASH_IPV6_DST_ONLY:
13806 /* fall-through. */
13807 case MLX5_RSS_HASH_IPV6_SRC_ONLY:
13808 hrxqs[3] = hrxq_idx;
13810 case MLX5_RSS_HASH_IPV6_TCP:
13811 /* fall-through. */
13812 case MLX5_RSS_HASH_IPV6_TCP_DST_ONLY:
13813 /* fall-through. */
13814 case MLX5_RSS_HASH_IPV6_TCP_SRC_ONLY:
13815 hrxqs[4] = hrxq_idx;
13817 case MLX5_RSS_HASH_IPV6_UDP:
13818 /* fall-through. */
13819 case MLX5_RSS_HASH_IPV6_UDP_DST_ONLY:
13820 /* fall-through. */
13821 case MLX5_RSS_HASH_IPV6_UDP_SRC_ONLY:
13822 hrxqs[5] = hrxq_idx;
13824 case MLX5_RSS_HASH_NONE:
13825 hrxqs[6] = hrxq_idx;
13833 * Look up for hash RX queue by hash fields (see enum ibv_rx_hash_fields)
13837 * Pointer to the Ethernet device structure.
13839 * Shared RSS action ID holding hash RX queue objects.
13840 * @param[in] hash_fields
13841 * Defines combination of packet fields to participate in RX hash.
13842 * @param[in] tunnel
13846 * Valid hash RX queue index, otherwise 0.
13849 flow_dv_action_rss_hrxq_lookup(struct rte_eth_dev *dev, uint32_t idx,
13850 const uint64_t hash_fields)
13852 struct mlx5_priv *priv = dev->data->dev_private;
13853 struct mlx5_shared_action_rss *shared_rss =
13854 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
13855 const uint32_t *hrxqs = shared_rss->hrxq;
13857 switch (hash_fields & ~IBV_RX_HASH_INNER) {
13858 case MLX5_RSS_HASH_IPV4:
13859 /* fall-through. */
13860 case MLX5_RSS_HASH_IPV4_DST_ONLY:
13861 /* fall-through. */
13862 case MLX5_RSS_HASH_IPV4_SRC_ONLY:
13864 case MLX5_RSS_HASH_IPV4_TCP:
13865 /* fall-through. */
13866 case MLX5_RSS_HASH_IPV4_TCP_DST_ONLY:
13867 /* fall-through. */
13868 case MLX5_RSS_HASH_IPV4_TCP_SRC_ONLY:
13870 case MLX5_RSS_HASH_IPV4_UDP:
13871 /* fall-through. */
13872 case MLX5_RSS_HASH_IPV4_UDP_DST_ONLY:
13873 /* fall-through. */
13874 case MLX5_RSS_HASH_IPV4_UDP_SRC_ONLY:
13876 case MLX5_RSS_HASH_IPV6:
13877 /* fall-through. */
13878 case MLX5_RSS_HASH_IPV6_DST_ONLY:
13879 /* fall-through. */
13880 case MLX5_RSS_HASH_IPV6_SRC_ONLY:
13882 case MLX5_RSS_HASH_IPV6_TCP:
13883 /* fall-through. */
13884 case MLX5_RSS_HASH_IPV6_TCP_DST_ONLY:
13885 /* fall-through. */
13886 case MLX5_RSS_HASH_IPV6_TCP_SRC_ONLY:
13888 case MLX5_RSS_HASH_IPV6_UDP:
13889 /* fall-through. */
13890 case MLX5_RSS_HASH_IPV6_UDP_DST_ONLY:
13891 /* fall-through. */
13892 case MLX5_RSS_HASH_IPV6_UDP_SRC_ONLY:
13894 case MLX5_RSS_HASH_NONE:
13903 * Apply the flow to the NIC, lock free,
13904 * (mutex should be acquired by caller).
13907 * Pointer to the Ethernet device structure.
13908 * @param[in, out] flow
13909 * Pointer to flow structure.
13910 * @param[out] error
13911 * Pointer to error structure.
13914 * 0 on success, a negative errno value otherwise and rte_errno is set.
13917 flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
13918 struct rte_flow_error *error)
13920 struct mlx5_flow_dv_workspace *dv;
13921 struct mlx5_flow_handle *dh;
13922 struct mlx5_flow_handle_dv *dv_h;
13923 struct mlx5_flow *dev_flow;
13924 struct mlx5_priv *priv = dev->data->dev_private;
13925 uint32_t handle_idx;
13929 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
13930 struct mlx5_flow_rss_desc *rss_desc = &wks->rss_desc;
13934 for (idx = wks->flow_idx - 1; idx >= 0; idx--) {
13935 dev_flow = &wks->flows[idx];
13936 dv = &dev_flow->dv;
13937 dh = dev_flow->handle;
13940 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
13941 if (dv->transfer) {
13942 MLX5_ASSERT(priv->sh->dr_drop_action);
13943 dv->actions[n++] = priv->sh->dr_drop_action;
13945 #ifdef HAVE_MLX5DV_DR
13946 /* DR supports drop action placeholder. */
13947 MLX5_ASSERT(priv->sh->dr_drop_action);
13948 dv->actions[n++] = dv->group ?
13949 priv->sh->dr_drop_action :
13950 priv->root_drop_action;
13952 /* For DV we use the explicit drop queue. */
13953 MLX5_ASSERT(priv->drop_queue.hrxq);
13955 priv->drop_queue.hrxq->action;
13958 } else if ((dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
13959 !dv_h->rix_sample && !dv_h->rix_dest_array)) {
13960 struct mlx5_hrxq *hrxq;
13963 hrxq = flow_dv_hrxq_prepare(dev, dev_flow, rss_desc,
13968 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
13969 "cannot get hash queue");
13972 dh->rix_hrxq = hrxq_idx;
13973 dv->actions[n++] = hrxq->action;
13974 } else if (dh->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
13975 struct mlx5_hrxq *hrxq = NULL;
13978 hrxq_idx = flow_dv_action_rss_hrxq_lookup(dev,
13979 rss_desc->shared_rss,
13980 dev_flow->hash_fields);
13982 hrxq = mlx5_ipool_get
13983 (priv->sh->ipool[MLX5_IPOOL_HRXQ],
13988 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
13989 "cannot get hash queue");
13992 dh->rix_srss = rss_desc->shared_rss;
13993 dv->actions[n++] = hrxq->action;
13994 } else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
13995 if (!priv->sh->default_miss_action) {
13998 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
13999 "default miss action not be created.");
14002 dv->actions[n++] = priv->sh->default_miss_action;
14004 misc_mask = flow_dv_matcher_enable(dv->value.buf);
14005 __flow_dv_adjust_buf_size(&dv->value.size, misc_mask);
14006 err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
14007 (void *)&dv->value, n,
14008 dv->actions, &dh->drv_flow);
14012 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
14014 (!priv->sh->config.allow_duplicate_pattern &&
14016 "duplicating pattern is not allowed" :
14017 "hardware refuses to create flow");
14020 if (priv->vmwa_context &&
14021 dh->vf_vlan.tag && !dh->vf_vlan.created) {
14023 * The rule contains the VLAN pattern.
14024 * For VF we are going to create VLAN
14025 * interface to make hypervisor set correct
14026 * e-Switch vport context.
14028 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
14033 err = rte_errno; /* Save rte_errno before cleanup. */
14034 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
14035 handle_idx, dh, next) {
14036 /* hrxq is union, don't clear it if the flag is not set. */
14037 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE && dh->rix_hrxq) {
14038 mlx5_hrxq_release(dev, dh->rix_hrxq);
14040 } else if (dh->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
14043 if (dh->vf_vlan.tag && dh->vf_vlan.created)
14044 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
14046 rte_errno = err; /* Restore rte_errno. */
14051 flow_dv_matcher_remove_cb(void *tool_ctx __rte_unused,
14052 struct mlx5_list_entry *entry)
14054 struct mlx5_flow_dv_matcher *resource = container_of(entry,
14058 claim_zero(mlx5_flow_os_destroy_flow_matcher(resource->matcher_object));
14059 mlx5_free(resource);
14063 * Release the flow matcher.
14066 * Pointer to Ethernet device.
14068 * Index to port ID action resource.
14071 * 1 while a reference on it exists, 0 when freed.
14074 flow_dv_matcher_release(struct rte_eth_dev *dev,
14075 struct mlx5_flow_handle *handle)
14077 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
14078 struct mlx5_flow_tbl_data_entry *tbl = container_of(matcher->tbl,
14079 typeof(*tbl), tbl);
14082 MLX5_ASSERT(matcher->matcher_object);
14083 ret = mlx5_list_unregister(tbl->matchers, &matcher->entry);
14084 flow_dv_tbl_resource_release(MLX5_SH(dev), &tbl->tbl);
14089 flow_dv_encap_decap_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
14091 struct mlx5_dev_ctx_shared *sh = tool_ctx;
14092 struct mlx5_flow_dv_encap_decap_resource *res =
14093 container_of(entry, typeof(*res), entry);
14095 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
14096 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], res->idx);
14100 * Release an encap/decap resource.
14103 * Pointer to Ethernet device.
14104 * @param encap_decap_idx
14105 * Index of encap decap resource.
14108 * 1 while a reference on it exists, 0 when freed.
14111 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
14112 uint32_t encap_decap_idx)
14114 struct mlx5_priv *priv = dev->data->dev_private;
14115 struct mlx5_flow_dv_encap_decap_resource *resource;
14117 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
14121 MLX5_ASSERT(resource->action);
14122 return mlx5_hlist_unregister(priv->sh->encaps_decaps, &resource->entry);
14126 * Release an jump to table action resource.
14129 * Pointer to Ethernet device.
14131 * Index to the jump action resource.
14134 * 1 while a reference on it exists, 0 when freed.
14137 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
14140 struct mlx5_priv *priv = dev->data->dev_private;
14141 struct mlx5_flow_tbl_data_entry *tbl_data;
14143 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
14147 return flow_dv_tbl_resource_release(MLX5_SH(dev), &tbl_data->tbl);
14151 flow_dv_modify_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
14153 struct mlx5_flow_dv_modify_hdr_resource *res =
14154 container_of(entry, typeof(*res), entry);
14155 struct mlx5_dev_ctx_shared *sh = tool_ctx;
14157 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
14158 mlx5_ipool_free(sh->mdh_ipools[res->actions_num - 1], res->idx);
14162 * Release a modify-header resource.
14165 * Pointer to Ethernet device.
14167 * Pointer to mlx5_flow_handle.
14170 * 1 while a reference on it exists, 0 when freed.
14173 flow_dv_modify_hdr_resource_release(struct rte_eth_dev *dev,
14174 struct mlx5_flow_handle *handle)
14176 struct mlx5_priv *priv = dev->data->dev_private;
14177 struct mlx5_flow_dv_modify_hdr_resource *entry = handle->dvh.modify_hdr;
14179 MLX5_ASSERT(entry->action);
14180 return mlx5_hlist_unregister(priv->sh->modify_cmds, &entry->entry);
14184 flow_dv_port_id_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
14186 struct mlx5_dev_ctx_shared *sh = tool_ctx;
14187 struct mlx5_flow_dv_port_id_action_resource *resource =
14188 container_of(entry, typeof(*resource), entry);
14190 claim_zero(mlx5_flow_os_destroy_flow_action(resource->action));
14191 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PORT_ID], resource->idx);
14195 * Release port ID action resource.
14198 * Pointer to Ethernet device.
14200 * Pointer to mlx5_flow_handle.
14203 * 1 while a reference on it exists, 0 when freed.
14206 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
14209 struct mlx5_priv *priv = dev->data->dev_private;
14210 struct mlx5_flow_dv_port_id_action_resource *resource;
14212 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID], port_id);
14215 MLX5_ASSERT(resource->action);
14216 return mlx5_list_unregister(priv->sh->port_id_action_list,
14221 * Release shared RSS action resource.
14224 * Pointer to Ethernet device.
14226 * Shared RSS action index.
14229 flow_dv_shared_rss_action_release(struct rte_eth_dev *dev, uint32_t srss)
14231 struct mlx5_priv *priv = dev->data->dev_private;
14232 struct mlx5_shared_action_rss *shared_rss;
14234 shared_rss = mlx5_ipool_get
14235 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], srss);
14236 __atomic_sub_fetch(&shared_rss->refcnt, 1, __ATOMIC_RELAXED);
14240 flow_dv_push_vlan_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
14242 struct mlx5_dev_ctx_shared *sh = tool_ctx;
14243 struct mlx5_flow_dv_push_vlan_action_resource *resource =
14244 container_of(entry, typeof(*resource), entry);
14246 claim_zero(mlx5_flow_os_destroy_flow_action(resource->action));
14247 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PUSH_VLAN], resource->idx);
14251 * Release push vlan action resource.
14254 * Pointer to Ethernet device.
14256 * Pointer to mlx5_flow_handle.
14259 * 1 while a reference on it exists, 0 when freed.
14262 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
14263 struct mlx5_flow_handle *handle)
14265 struct mlx5_priv *priv = dev->data->dev_private;
14266 struct mlx5_flow_dv_push_vlan_action_resource *resource;
14267 uint32_t idx = handle->dvh.rix_push_vlan;
14269 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
14272 MLX5_ASSERT(resource->action);
14273 return mlx5_list_unregister(priv->sh->push_vlan_action_list,
14278 * Release the fate resource.
14281 * Pointer to Ethernet device.
14283 * Pointer to mlx5_flow_handle.
14286 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
14287 struct mlx5_flow_handle *handle)
14289 if (!handle->rix_fate)
14291 switch (handle->fate_action) {
14292 case MLX5_FLOW_FATE_QUEUE:
14293 if (!handle->dvh.rix_sample && !handle->dvh.rix_dest_array)
14294 mlx5_hrxq_release(dev, handle->rix_hrxq);
14296 case MLX5_FLOW_FATE_JUMP:
14297 flow_dv_jump_tbl_resource_release(dev, handle->rix_jump);
14299 case MLX5_FLOW_FATE_PORT_ID:
14300 flow_dv_port_id_action_resource_release(dev,
14301 handle->rix_port_id_action);
14304 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
14307 handle->rix_fate = 0;
14311 flow_dv_sample_remove_cb(void *tool_ctx __rte_unused,
14312 struct mlx5_list_entry *entry)
14314 struct mlx5_flow_dv_sample_resource *resource = container_of(entry,
14317 struct rte_eth_dev *dev = resource->dev;
14318 struct mlx5_priv *priv = dev->data->dev_private;
14320 if (resource->verbs_action)
14321 claim_zero(mlx5_flow_os_destroy_flow_action
14322 (resource->verbs_action));
14323 if (resource->normal_path_tbl)
14324 flow_dv_tbl_resource_release(MLX5_SH(dev),
14325 resource->normal_path_tbl);
14326 flow_dv_sample_sub_actions_release(dev, &resource->sample_idx);
14327 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE], resource->idx);
14328 DRV_LOG(DEBUG, "sample resource %p: removed", (void *)resource);
14332 * Release an sample resource.
14335 * Pointer to Ethernet device.
14337 * Pointer to mlx5_flow_handle.
14340 * 1 while a reference on it exists, 0 when freed.
14343 flow_dv_sample_resource_release(struct rte_eth_dev *dev,
14344 struct mlx5_flow_handle *handle)
14346 struct mlx5_priv *priv = dev->data->dev_private;
14347 struct mlx5_flow_dv_sample_resource *resource;
14349 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
14350 handle->dvh.rix_sample);
14353 MLX5_ASSERT(resource->verbs_action);
14354 return mlx5_list_unregister(priv->sh->sample_action_list,
14359 flow_dv_dest_array_remove_cb(void *tool_ctx __rte_unused,
14360 struct mlx5_list_entry *entry)
14362 struct mlx5_flow_dv_dest_array_resource *resource =
14363 container_of(entry, typeof(*resource), entry);
14364 struct rte_eth_dev *dev = resource->dev;
14365 struct mlx5_priv *priv = dev->data->dev_private;
14368 MLX5_ASSERT(resource->action);
14369 if (resource->action)
14370 claim_zero(mlx5_flow_os_destroy_flow_action(resource->action));
14371 for (; i < resource->num_of_dest; i++)
14372 flow_dv_sample_sub_actions_release(dev,
14373 &resource->sample_idx[i]);
14374 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY], resource->idx);
14375 DRV_LOG(DEBUG, "destination array resource %p: removed",
14380 * Release an destination array resource.
14383 * Pointer to Ethernet device.
14385 * Pointer to mlx5_flow_handle.
14388 * 1 while a reference on it exists, 0 when freed.
14391 flow_dv_dest_array_resource_release(struct rte_eth_dev *dev,
14392 struct mlx5_flow_handle *handle)
14394 struct mlx5_priv *priv = dev->data->dev_private;
14395 struct mlx5_flow_dv_dest_array_resource *resource;
14397 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
14398 handle->dvh.rix_dest_array);
14401 MLX5_ASSERT(resource->action);
14402 return mlx5_list_unregister(priv->sh->dest_array_list,
14407 flow_dv_geneve_tlv_option_resource_release(struct rte_eth_dev *dev)
14409 struct mlx5_priv *priv = dev->data->dev_private;
14410 struct mlx5_dev_ctx_shared *sh = priv->sh;
14411 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource =
14412 sh->geneve_tlv_option_resource;
14413 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
14414 if (geneve_opt_resource) {
14415 if (!(__atomic_sub_fetch(&geneve_opt_resource->refcnt, 1,
14416 __ATOMIC_RELAXED))) {
14417 claim_zero(mlx5_devx_cmd_destroy
14418 (geneve_opt_resource->obj));
14419 mlx5_free(sh->geneve_tlv_option_resource);
14420 sh->geneve_tlv_option_resource = NULL;
14423 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
14427 * Remove the flow from the NIC but keeps it in memory.
14428 * Lock free, (mutex should be acquired by caller).
14431 * Pointer to Ethernet device.
14432 * @param[in, out] flow
14433 * Pointer to flow structure.
14436 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
14438 struct mlx5_flow_handle *dh;
14439 uint32_t handle_idx;
14440 struct mlx5_priv *priv = dev->data->dev_private;
14444 handle_idx = flow->dev_handles;
14445 while (handle_idx) {
14446 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
14450 if (dh->drv_flow) {
14451 claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
14452 dh->drv_flow = NULL;
14454 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE)
14455 flow_dv_fate_resource_release(dev, dh);
14456 if (dh->vf_vlan.tag && dh->vf_vlan.created)
14457 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
14458 handle_idx = dh->next.next;
14463 * Remove the flow from the NIC and the memory.
14464 * Lock free, (mutex should be acquired by caller).
14467 * Pointer to the Ethernet device structure.
14468 * @param[in, out] flow
14469 * Pointer to flow structure.
14472 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
14474 struct mlx5_flow_handle *dev_handle;
14475 struct mlx5_priv *priv = dev->data->dev_private;
14476 struct mlx5_flow_meter_info *fm = NULL;
14481 flow_dv_remove(dev, flow);
14482 if (flow->counter) {
14483 flow_dv_counter_free(dev, flow->counter);
14487 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
14489 mlx5_flow_meter_detach(priv, fm);
14492 /* Keep the current age handling by default. */
14493 if (flow->indirect_type == MLX5_INDIRECT_ACTION_TYPE_CT && flow->ct)
14494 flow_dv_aso_ct_release(dev, flow->ct, NULL);
14495 else if (flow->age)
14496 flow_dv_aso_age_release(dev, flow->age);
14497 if (flow->geneve_tlv_option) {
14498 flow_dv_geneve_tlv_option_resource_release(dev);
14499 flow->geneve_tlv_option = 0;
14501 while (flow->dev_handles) {
14502 uint32_t tmp_idx = flow->dev_handles;
14504 dev_handle = mlx5_ipool_get(priv->sh->ipool
14505 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
14508 flow->dev_handles = dev_handle->next.next;
14509 while (dev_handle->flex_item) {
14510 int index = rte_bsf32(dev_handle->flex_item);
14512 mlx5_flex_release_index(dev, index);
14513 dev_handle->flex_item &= ~RTE_BIT32(index);
14515 if (dev_handle->dvh.matcher)
14516 flow_dv_matcher_release(dev, dev_handle);
14517 if (dev_handle->dvh.rix_sample)
14518 flow_dv_sample_resource_release(dev, dev_handle);
14519 if (dev_handle->dvh.rix_dest_array)
14520 flow_dv_dest_array_resource_release(dev, dev_handle);
14521 if (dev_handle->dvh.rix_encap_decap)
14522 flow_dv_encap_decap_resource_release(dev,
14523 dev_handle->dvh.rix_encap_decap);
14524 if (dev_handle->dvh.modify_hdr)
14525 flow_dv_modify_hdr_resource_release(dev, dev_handle);
14526 if (dev_handle->dvh.rix_push_vlan)
14527 flow_dv_push_vlan_action_resource_release(dev,
14529 if (dev_handle->dvh.rix_tag)
14530 flow_dv_tag_release(dev,
14531 dev_handle->dvh.rix_tag);
14532 if (dev_handle->fate_action != MLX5_FLOW_FATE_SHARED_RSS)
14533 flow_dv_fate_resource_release(dev, dev_handle);
14535 srss = dev_handle->rix_srss;
14536 if (fm && dev_handle->is_meter_flow_id &&
14537 dev_handle->split_flow_id)
14538 mlx5_ipool_free(fm->flow_ipool,
14539 dev_handle->split_flow_id);
14540 else if (dev_handle->split_flow_id &&
14541 !dev_handle->is_meter_flow_id)
14542 mlx5_ipool_free(priv->sh->ipool
14543 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
14544 dev_handle->split_flow_id);
14545 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
14549 flow_dv_shared_rss_action_release(dev, srss);
14553 * Release array of hash RX queue objects.
14557 * Pointer to the Ethernet device structure.
14558 * @param[in, out] hrxqs
14559 * Array of hash RX queue objects.
14562 * Total number of references to hash RX queue objects in *hrxqs* array
14563 * after this operation.
14566 __flow_dv_hrxqs_release(struct rte_eth_dev *dev,
14567 uint32_t (*hrxqs)[MLX5_RSS_HASH_FIELDS_LEN])
14572 for (i = 0; i < RTE_DIM(*hrxqs); i++) {
14573 int ret = mlx5_hrxq_release(dev, (*hrxqs)[i]);
14583 * Release all hash RX queue objects representing shared RSS action.
14586 * Pointer to the Ethernet device structure.
14587 * @param[in, out] action
14588 * Shared RSS action to remove hash RX queue objects from.
14591 * Total number of references to hash RX queue objects stored in *action*
14592 * after this operation.
14593 * Expected to be 0 if no external references held.
14596 __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
14597 struct mlx5_shared_action_rss *shared_rss)
14599 return __flow_dv_hrxqs_release(dev, &shared_rss->hrxq);
14603 * Adjust L3/L4 hash value of pre-created shared RSS hrxq according to
14606 * Only one hash value is available for one L3+L4 combination:
14608 * MLX5_RSS_HASH_IPV4, MLX5_RSS_HASH_IPV4_SRC_ONLY, and
14609 * MLX5_RSS_HASH_IPV4_DST_ONLY are mutually exclusive so they can share
14610 * same slot in mlx5_rss_hash_fields.
14612 * @param[in] rss_types
14614 * @param[in, out] hash_field
14615 * hash_field variable needed to be adjusted.
14621 flow_dv_action_rss_l34_hash_adjust(uint64_t rss_types,
14622 uint64_t *hash_field)
14624 switch (*hash_field & ~IBV_RX_HASH_INNER) {
14625 case MLX5_RSS_HASH_IPV4:
14626 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
14627 *hash_field &= ~MLX5_RSS_HASH_IPV4;
14628 if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
14629 *hash_field |= IBV_RX_HASH_DST_IPV4;
14630 else if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
14631 *hash_field |= IBV_RX_HASH_SRC_IPV4;
14633 *hash_field |= MLX5_RSS_HASH_IPV4;
14636 case MLX5_RSS_HASH_IPV6:
14637 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
14638 *hash_field &= ~MLX5_RSS_HASH_IPV6;
14639 if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
14640 *hash_field |= IBV_RX_HASH_DST_IPV6;
14641 else if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
14642 *hash_field |= IBV_RX_HASH_SRC_IPV6;
14644 *hash_field |= MLX5_RSS_HASH_IPV6;
14647 case MLX5_RSS_HASH_IPV4_UDP:
14648 /* fall-through. */
14649 case MLX5_RSS_HASH_IPV6_UDP:
14650 if (rss_types & RTE_ETH_RSS_UDP) {
14651 *hash_field &= ~MLX5_UDP_IBV_RX_HASH;
14652 if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
14653 *hash_field |= IBV_RX_HASH_DST_PORT_UDP;
14654 else if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
14655 *hash_field |= IBV_RX_HASH_SRC_PORT_UDP;
14657 *hash_field |= MLX5_UDP_IBV_RX_HASH;
14660 case MLX5_RSS_HASH_IPV4_TCP:
14661 /* fall-through. */
14662 case MLX5_RSS_HASH_IPV6_TCP:
14663 if (rss_types & RTE_ETH_RSS_TCP) {
14664 *hash_field &= ~MLX5_TCP_IBV_RX_HASH;
14665 if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
14666 *hash_field |= IBV_RX_HASH_DST_PORT_TCP;
14667 else if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
14668 *hash_field |= IBV_RX_HASH_SRC_PORT_TCP;
14670 *hash_field |= MLX5_TCP_IBV_RX_HASH;
14679 * Setup shared RSS action.
14680 * Prepare set of hash RX queue objects sufficient to handle all valid
14681 * hash_fields combinations (see enum ibv_rx_hash_fields).
14684 * Pointer to the Ethernet device structure.
14685 * @param[in] action_idx
14686 * Shared RSS action ipool index.
14687 * @param[in, out] action
14688 * Partially initialized shared RSS action.
14689 * @param[out] error
14690 * Perform verbose error reporting if not NULL. Initialized in case of
14694 * 0 on success, otherwise negative errno value.
14697 __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
14698 uint32_t action_idx,
14699 struct mlx5_shared_action_rss *shared_rss,
14700 struct rte_flow_error *error)
14702 struct mlx5_priv *priv = dev->data->dev_private;
14703 struct mlx5_flow_rss_desc rss_desc = { 0 };
14707 shared_rss->ind_tbl = mlx5_ind_table_obj_new
14708 (dev, shared_rss->origin.queue,
14709 shared_rss->origin.queue_num,
14711 !!dev->data->dev_started);
14712 if (!shared_rss->ind_tbl)
14713 return rte_flow_error_set(error, rte_errno,
14714 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14715 "cannot setup indirection table");
14716 memcpy(rss_desc.key, shared_rss->origin.key, MLX5_RSS_HASH_KEY_LEN);
14717 rss_desc.key_len = MLX5_RSS_HASH_KEY_LEN;
14718 rss_desc.const_q = shared_rss->origin.queue;
14719 rss_desc.queue_num = shared_rss->origin.queue_num;
14720 /* Set non-zero value to indicate a shared RSS. */
14721 rss_desc.shared_rss = action_idx;
14722 rss_desc.ind_tbl = shared_rss->ind_tbl;
14723 if (priv->sh->config.dv_flow_en == 2)
14724 rss_desc.hws_flags = MLX5DR_ACTION_FLAG_HWS_RX;
14725 for (i = 0; i < MLX5_RSS_HASH_FIELDS_LEN; i++) {
14726 struct mlx5_hrxq *hrxq;
14727 uint64_t hash_fields = mlx5_rss_hash_fields[i];
14730 flow_dv_action_rss_l34_hash_adjust(shared_rss->origin.types,
14732 if (shared_rss->origin.level > 1) {
14733 hash_fields |= IBV_RX_HASH_INNER;
14736 rss_desc.tunnel = tunnel;
14737 rss_desc.hash_fields = hash_fields;
14738 hrxq = mlx5_hrxq_get(dev, &rss_desc);
14742 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14743 "cannot get hash queue");
14744 goto error_hrxq_new;
14746 err = __flow_dv_action_rss_hrxq_set
14747 (shared_rss, hash_fields, hrxq->idx);
14753 __flow_dv_action_rss_hrxqs_release(dev, shared_rss);
14754 if (!mlx5_ind_table_obj_release(dev, shared_rss->ind_tbl, true))
14755 shared_rss->ind_tbl = NULL;
14761 * Create shared RSS action.
14764 * Pointer to the Ethernet device structure.
14766 * Shared action configuration.
14768 * RSS action specification used to create shared action.
14769 * @param[out] error
14770 * Perform verbose error reporting if not NULL. Initialized in case of
14774 * A valid shared action ID in case of success, 0 otherwise and
14775 * rte_errno is set.
14778 __flow_dv_action_rss_create(struct rte_eth_dev *dev,
14779 const struct rte_flow_indir_action_conf *conf,
14780 const struct rte_flow_action_rss *rss,
14781 struct rte_flow_error *error)
14783 struct mlx5_priv *priv = dev->data->dev_private;
14784 struct mlx5_shared_action_rss *shared_rss = NULL;
14785 struct rte_flow_action_rss *origin;
14786 const uint8_t *rss_key;
14789 RTE_SET_USED(conf);
14790 shared_rss = mlx5_ipool_zmalloc
14791 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], &idx);
14793 rte_flow_error_set(error, ENOMEM,
14794 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14795 "cannot allocate resource memory");
14796 goto error_rss_init;
14798 if (idx > (1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET)) {
14799 rte_flow_error_set(error, E2BIG,
14800 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14801 "rss action number out of range");
14802 goto error_rss_init;
14804 origin = &shared_rss->origin;
14805 origin->func = rss->func;
14806 origin->level = rss->level;
14807 /* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
14808 origin->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
14809 /* NULL RSS key indicates default RSS key. */
14810 rss_key = !rss->key ? rss_hash_default_key : rss->key;
14811 memcpy(shared_rss->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
14812 origin->key = &shared_rss->key[0];
14813 origin->key_len = MLX5_RSS_HASH_KEY_LEN;
14814 origin->queue = rss->queue;
14815 origin->queue_num = rss->queue_num;
14816 if (__flow_dv_action_rss_setup(dev, idx, shared_rss, error))
14817 goto error_rss_init;
14818 /* Update queue with indirect table queue memoyr. */
14819 origin->queue = shared_rss->ind_tbl->queues;
14820 rte_spinlock_init(&shared_rss->action_rss_sl);
14821 __atomic_add_fetch(&shared_rss->refcnt, 1, __ATOMIC_RELAXED);
14822 rte_spinlock_lock(&priv->shared_act_sl);
14823 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
14824 &priv->rss_shared_actions, idx, shared_rss, next);
14825 rte_spinlock_unlock(&priv->shared_act_sl);
14829 if (shared_rss->ind_tbl)
14830 mlx5_ind_table_obj_release(dev, shared_rss->ind_tbl,
14831 !!dev->data->dev_started);
14832 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
14839 * Destroy the shared RSS action.
14840 * Release related hash RX queue objects.
14843 * Pointer to the Ethernet device structure.
14845 * The shared RSS action object ID to be removed.
14846 * @param[out] error
14847 * Perform verbose error reporting if not NULL. Initialized in case of
14851 * 0 on success, otherwise negative errno value.
14854 __flow_dv_action_rss_release(struct rte_eth_dev *dev, uint32_t idx,
14855 struct rte_flow_error *error)
14857 struct mlx5_priv *priv = dev->data->dev_private;
14858 struct mlx5_shared_action_rss *shared_rss =
14859 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
14860 uint32_t old_refcnt = 1;
14864 return rte_flow_error_set(error, EINVAL,
14865 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
14866 "invalid shared action");
14867 if (!__atomic_compare_exchange_n(&shared_rss->refcnt, &old_refcnt,
14868 0, 0, __ATOMIC_ACQUIRE,
14870 return rte_flow_error_set(error, EBUSY,
14871 RTE_FLOW_ERROR_TYPE_ACTION,
14873 "shared rss has references");
14874 remaining = __flow_dv_action_rss_hrxqs_release(dev, shared_rss);
14876 return rte_flow_error_set(error, EBUSY,
14877 RTE_FLOW_ERROR_TYPE_ACTION,
14879 "shared rss hrxq has references");
14880 remaining = mlx5_ind_table_obj_release(dev, shared_rss->ind_tbl,
14881 !!dev->data->dev_started);
14883 return rte_flow_error_set(error, EBUSY,
14884 RTE_FLOW_ERROR_TYPE_ACTION,
14886 "shared rss indirection table has"
14888 rte_spinlock_lock(&priv->shared_act_sl);
14889 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
14890 &priv->rss_shared_actions, idx, shared_rss, next);
14891 rte_spinlock_unlock(&priv->shared_act_sl);
14892 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
14898 * Create indirect action, lock free,
14899 * (mutex should be acquired by caller).
14900 * Dispatcher for action type specific call.
14903 * Pointer to the Ethernet device structure.
14905 * Shared action configuration.
14906 * @param[in] action
14907 * Action specification used to create indirect action.
14908 * @param[out] error
14909 * Perform verbose error reporting if not NULL. Initialized in case of
14913 * A valid shared action handle in case of success, NULL otherwise and
14914 * rte_errno is set.
14916 struct rte_flow_action_handle *
14917 flow_dv_action_create(struct rte_eth_dev *dev,
14918 const struct rte_flow_indir_action_conf *conf,
14919 const struct rte_flow_action *action,
14920 struct rte_flow_error *err)
14922 struct mlx5_priv *priv = dev->data->dev_private;
14923 uint32_t age_idx = 0;
14927 switch (action->type) {
14928 case RTE_FLOW_ACTION_TYPE_RSS:
14929 ret = __flow_dv_action_rss_create(dev, conf, action->conf, err);
14930 idx = (MLX5_INDIRECT_ACTION_TYPE_RSS <<
14931 MLX5_INDIRECT_ACTION_TYPE_OFFSET) | ret;
14933 case RTE_FLOW_ACTION_TYPE_AGE:
14934 age_idx = flow_dv_aso_age_alloc(dev, err);
14939 idx = (MLX5_INDIRECT_ACTION_TYPE_AGE <<
14940 MLX5_INDIRECT_ACTION_TYPE_OFFSET) | age_idx;
14941 flow_dv_aso_age_params_init(dev, age_idx,
14942 ((const struct rte_flow_action_age *)
14943 action->conf)->context ?
14944 ((const struct rte_flow_action_age *)
14945 action->conf)->context :
14946 (void *)(uintptr_t)idx,
14947 ((const struct rte_flow_action_age *)
14948 action->conf)->timeout);
14951 case RTE_FLOW_ACTION_TYPE_COUNT:
14952 ret = flow_dv_translate_create_counter(dev, NULL, NULL, NULL);
14953 idx = (MLX5_INDIRECT_ACTION_TYPE_COUNT <<
14954 MLX5_INDIRECT_ACTION_TYPE_OFFSET) | ret;
14956 case RTE_FLOW_ACTION_TYPE_CONNTRACK:
14957 ret = flow_dv_translate_create_conntrack(dev, action->conf,
14959 idx = MLX5_INDIRECT_ACT_CT_GEN_IDX(PORT_ID(priv), ret);
14962 rte_flow_error_set(err, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
14963 NULL, "action type not supported");
14966 return ret ? (struct rte_flow_action_handle *)(uintptr_t)idx : NULL;
14970 * Destroy the indirect action.
14971 * Release action related resources on the NIC and the memory.
14972 * Lock free, (mutex should be acquired by caller).
14973 * Dispatcher for action type specific call.
14976 * Pointer to the Ethernet device structure.
14977 * @param[in] handle
14978 * The indirect action object handle to be removed.
14979 * @param[out] error
14980 * Perform verbose error reporting if not NULL. Initialized in case of
14984 * 0 on success, otherwise negative errno value.
14987 flow_dv_action_destroy(struct rte_eth_dev *dev,
14988 struct rte_flow_action_handle *handle,
14989 struct rte_flow_error *error)
14991 uint32_t act_idx = (uint32_t)(uintptr_t)handle;
14992 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
14993 uint32_t idx = act_idx & ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
14994 struct mlx5_flow_counter *cnt;
14995 uint32_t no_flow_refcnt = 1;
14999 case MLX5_INDIRECT_ACTION_TYPE_RSS:
15000 return __flow_dv_action_rss_release(dev, idx, error);
15001 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
15002 cnt = flow_dv_counter_get_by_idx(dev, idx, NULL);
15003 if (!__atomic_compare_exchange_n(&cnt->shared_info.refcnt,
15004 &no_flow_refcnt, 1, false,
15007 return rte_flow_error_set(error, EBUSY,
15008 RTE_FLOW_ERROR_TYPE_ACTION,
15010 "Indirect count action has references");
15011 flow_dv_counter_free(dev, idx);
15013 case MLX5_INDIRECT_ACTION_TYPE_AGE:
15014 ret = flow_dv_aso_age_release(dev, idx);
15017 * In this case, the last flow has a reference will
15018 * actually release the age action.
15020 DRV_LOG(DEBUG, "Indirect age action %" PRIu32 " was"
15021 " released with references %d.", idx, ret);
15023 case MLX5_INDIRECT_ACTION_TYPE_CT:
15024 ret = flow_dv_aso_ct_release(dev, idx, error);
15028 DRV_LOG(DEBUG, "Connection tracking object %u still "
15029 "has references %d.", idx, ret);
15032 return rte_flow_error_set(error, ENOTSUP,
15033 RTE_FLOW_ERROR_TYPE_ACTION,
15035 "action type not supported");
15040 * Updates in place shared RSS action configuration.
15043 * Pointer to the Ethernet device structure.
15045 * The shared RSS action object ID to be updated.
15046 * @param[in] action_conf
15047 * RSS action specification used to modify *shared_rss*.
15048 * @param[out] error
15049 * Perform verbose error reporting if not NULL. Initialized in case of
15053 * 0 on success, otherwise negative errno value.
15054 * @note: currently only support update of RSS queues.
15057 __flow_dv_action_rss_update(struct rte_eth_dev *dev, uint32_t idx,
15058 const struct rte_flow_action_rss *action_conf,
15059 struct rte_flow_error *error)
15061 struct mlx5_priv *priv = dev->data->dev_private;
15062 struct mlx5_shared_action_rss *shared_rss =
15063 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
15065 void *queue = NULL;
15066 void *queue_i = NULL;
15067 uint32_t queue_size = action_conf->queue_num * sizeof(uint16_t);
15068 bool dev_started = !!dev->data->dev_started;
15071 return rte_flow_error_set(error, EINVAL,
15072 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
15073 "invalid shared action to update");
15074 if (priv->obj_ops.ind_table_modify == NULL)
15075 return rte_flow_error_set(error, ENOTSUP,
15076 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
15077 "cannot modify indirection table");
15078 queue = mlx5_malloc(MLX5_MEM_ZERO,
15079 RTE_ALIGN_CEIL(queue_size, sizeof(void *)),
15082 return rte_flow_error_set(error, ENOMEM,
15083 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15085 "cannot allocate resource memory");
15086 memcpy(queue, action_conf->queue, queue_size);
15087 MLX5_ASSERT(shared_rss->ind_tbl);
15088 rte_spinlock_lock(&shared_rss->action_rss_sl);
15089 queue_i = shared_rss->ind_tbl->queues;
15090 ret = mlx5_ind_table_obj_modify(dev, shared_rss->ind_tbl,
15091 queue, action_conf->queue_num,
15092 true /* standalone */,
15093 dev_started /* ref_new_qs */,
15094 dev_started /* deref_old_qs */);
15096 ret = rte_flow_error_set(error, rte_errno,
15097 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
15098 "cannot update indirection table");
15100 /* Restore the queue to indirect table internal queue. */
15101 memcpy(queue_i, queue, queue_size);
15102 shared_rss->ind_tbl->queues = queue_i;
15103 shared_rss->origin.queue_num = action_conf->queue_num;
15106 rte_spinlock_unlock(&shared_rss->action_rss_sl);
15111 * Updates in place conntrack context or direction.
15112 * Context update should be synchronized.
15115 * Pointer to the Ethernet device structure.
15117 * The conntrack object ID to be updated.
15118 * @param[in] update
15119 * Pointer to the structure of information to update.
15120 * @param[out] error
15121 * Perform verbose error reporting if not NULL. Initialized in case of
15125 * 0 on success, otherwise negative errno value.
15128 __flow_dv_action_ct_update(struct rte_eth_dev *dev, uint32_t idx,
15129 const struct rte_flow_modify_conntrack *update,
15130 struct rte_flow_error *error)
15132 struct mlx5_priv *priv = dev->data->dev_private;
15133 struct mlx5_aso_ct_action *ct;
15134 const struct rte_flow_action_conntrack *new_prf;
15136 uint16_t owner = (uint16_t)MLX5_INDIRECT_ACT_CT_GET_OWNER(idx);
15139 if (PORT_ID(priv) != owner)
15140 return rte_flow_error_set(error, EACCES,
15141 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15143 "CT object owned by another port");
15144 dev_idx = MLX5_INDIRECT_ACT_CT_GET_IDX(idx);
15145 ct = flow_aso_ct_get_by_dev_idx(dev, dev_idx);
15147 return rte_flow_error_set(error, ENOMEM,
15148 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15150 "CT object is inactive");
15151 new_prf = &update->new_ct;
15152 if (update->direction)
15153 ct->is_original = !!new_prf->is_original_dir;
15154 if (update->state) {
15155 /* Only validate the profile when it needs to be updated. */
15156 ret = mlx5_validate_action_ct(dev, new_prf, error);
15159 ret = mlx5_aso_ct_update_by_wqe(priv->sh, ct, new_prf);
15161 return rte_flow_error_set(error, EIO,
15162 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15164 "Failed to send CT context update WQE");
15165 /* Block until ready or a failure. */
15166 ret = mlx5_aso_ct_available(priv->sh, ct);
15168 rte_flow_error_set(error, rte_errno,
15169 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15171 "Timeout to get the CT update");
15177 * Updates in place shared action configuration, lock free,
15178 * (mutex should be acquired by caller).
15181 * Pointer to the Ethernet device structure.
15182 * @param[in] handle
15183 * The indirect action object handle to be updated.
15184 * @param[in] update
15185 * Action specification used to modify the action pointed by *handle*.
15186 * *update* could be of same type with the action pointed by the *handle*
15187 * handle argument, or some other structures like a wrapper, depending on
15188 * the indirect action type.
15189 * @param[out] error
15190 * Perform verbose error reporting if not NULL. Initialized in case of
15194 * 0 on success, otherwise negative errno value.
15197 flow_dv_action_update(struct rte_eth_dev *dev,
15198 struct rte_flow_action_handle *handle,
15199 const void *update,
15200 struct rte_flow_error *err)
15202 uint32_t act_idx = (uint32_t)(uintptr_t)handle;
15203 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
15204 uint32_t idx = act_idx & ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
15205 const void *action_conf;
15208 case MLX5_INDIRECT_ACTION_TYPE_RSS:
15209 action_conf = ((const struct rte_flow_action *)update)->conf;
15210 return __flow_dv_action_rss_update(dev, idx, action_conf, err);
15211 case MLX5_INDIRECT_ACTION_TYPE_CT:
15212 return __flow_dv_action_ct_update(dev, idx, update, err);
15214 return rte_flow_error_set(err, ENOTSUP,
15215 RTE_FLOW_ERROR_TYPE_ACTION,
15217 "action type update not supported");
15222 * Destroy the meter sub policy table rules.
15223 * Lock free, (mutex should be acquired by caller).
15226 * Pointer to Ethernet device.
15227 * @param[in] sub_policy
15228 * Pointer to meter sub policy table.
15231 __flow_dv_destroy_sub_policy_rules(struct rte_eth_dev *dev,
15232 struct mlx5_flow_meter_sub_policy *sub_policy)
15234 struct mlx5_priv *priv = dev->data->dev_private;
15235 struct mlx5_flow_tbl_data_entry *tbl;
15236 struct mlx5_flow_meter_policy *policy = sub_policy->main_policy;
15237 struct mlx5_flow_meter_info *next_fm;
15238 struct mlx5_sub_policy_color_rule *color_rule;
15242 for (i = 0; i < RTE_COLORS; i++) {
15244 if (i == RTE_COLOR_GREEN && policy &&
15245 policy->act_cnt[i].fate_action == MLX5_FLOW_FATE_MTR)
15246 next_fm = mlx5_flow_meter_find(priv,
15247 policy->act_cnt[i].next_mtr_id, NULL);
15248 RTE_TAILQ_FOREACH_SAFE(color_rule, &sub_policy->color_rules[i],
15250 claim_zero(mlx5_flow_os_destroy_flow(color_rule->rule));
15251 tbl = container_of(color_rule->matcher->tbl,
15252 typeof(*tbl), tbl);
15253 mlx5_list_unregister(tbl->matchers,
15254 &color_rule->matcher->entry);
15255 TAILQ_REMOVE(&sub_policy->color_rules[i],
15256 color_rule, next_port);
15257 mlx5_free(color_rule);
15259 mlx5_flow_meter_detach(priv, next_fm);
15262 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
15263 if (sub_policy->rix_hrxq[i]) {
15264 if (policy && !policy->is_hierarchy)
15265 mlx5_hrxq_release(dev, sub_policy->rix_hrxq[i]);
15266 sub_policy->rix_hrxq[i] = 0;
15268 if (sub_policy->jump_tbl[i]) {
15269 flow_dv_tbl_resource_release(MLX5_SH(dev),
15270 sub_policy->jump_tbl[i]);
15271 sub_policy->jump_tbl[i] = NULL;
15274 if (sub_policy->tbl_rsc) {
15275 flow_dv_tbl_resource_release(MLX5_SH(dev),
15276 sub_policy->tbl_rsc);
15277 sub_policy->tbl_rsc = NULL;
15282 * Destroy policy rules, lock free,
15283 * (mutex should be acquired by caller).
15284 * Dispatcher for action type specific call.
15287 * Pointer to the Ethernet device structure.
15288 * @param[in] mtr_policy
15289 * Meter policy struct.
15292 flow_dv_destroy_policy_rules(struct rte_eth_dev *dev,
15293 struct mlx5_flow_meter_policy *mtr_policy)
15296 struct mlx5_flow_meter_sub_policy *sub_policy;
15297 uint16_t sub_policy_num;
15299 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
15300 sub_policy_num = (mtr_policy->sub_policy_num >>
15301 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i)) &
15302 MLX5_MTR_SUB_POLICY_NUM_MASK;
15303 for (j = 0; j < sub_policy_num; j++) {
15304 sub_policy = mtr_policy->sub_policys[i][j];
15306 __flow_dv_destroy_sub_policy_rules(dev,
15313 * Destroy policy action, lock free,
15314 * (mutex should be acquired by caller).
15315 * Dispatcher for action type specific call.
15318 * Pointer to the Ethernet device structure.
15319 * @param[in] mtr_policy
15320 * Meter policy struct.
15323 flow_dv_destroy_mtr_policy_acts(struct rte_eth_dev *dev,
15324 struct mlx5_flow_meter_policy *mtr_policy)
15326 struct rte_flow_action *rss_action;
15327 struct mlx5_flow_handle dev_handle;
15330 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
15331 if (mtr_policy->act_cnt[i].rix_mark) {
15332 flow_dv_tag_release(dev,
15333 mtr_policy->act_cnt[i].rix_mark);
15334 mtr_policy->act_cnt[i].rix_mark = 0;
15336 if (mtr_policy->act_cnt[i].modify_hdr) {
15337 dev_handle.dvh.modify_hdr =
15338 mtr_policy->act_cnt[i].modify_hdr;
15339 flow_dv_modify_hdr_resource_release(dev, &dev_handle);
15341 switch (mtr_policy->act_cnt[i].fate_action) {
15342 case MLX5_FLOW_FATE_SHARED_RSS:
15343 rss_action = mtr_policy->act_cnt[i].rss;
15344 mlx5_free(rss_action);
15346 case MLX5_FLOW_FATE_PORT_ID:
15347 if (mtr_policy->act_cnt[i].rix_port_id_action) {
15348 flow_dv_port_id_action_resource_release(dev,
15349 mtr_policy->act_cnt[i].rix_port_id_action);
15350 mtr_policy->act_cnt[i].rix_port_id_action = 0;
15353 case MLX5_FLOW_FATE_DROP:
15354 case MLX5_FLOW_FATE_JUMP:
15355 for (j = 0; j < MLX5_MTR_DOMAIN_MAX; j++)
15356 mtr_policy->act_cnt[i].dr_jump_action[j] =
15360 /*Queue action do nothing*/
15364 for (j = 0; j < MLX5_MTR_DOMAIN_MAX; j++)
15365 mtr_policy->dr_drop_action[j] = NULL;
15369 * Create policy action per domain, lock free,
15370 * (mutex should be acquired by caller).
15371 * Dispatcher for action type specific call.
15374 * Pointer to the Ethernet device structure.
15375 * @param[in] mtr_policy
15376 * Meter policy struct.
15377 * @param[in] action
15378 * Action specification used to create meter actions.
15379 * @param[out] error
15380 * Perform verbose error reporting if not NULL. Initialized in case of
15384 * 0 on success, otherwise negative errno value.
15387 __flow_dv_create_domain_policy_acts(struct rte_eth_dev *dev,
15388 struct mlx5_flow_meter_policy *mtr_policy,
15389 const struct rte_flow_action *actions[RTE_COLORS],
15390 enum mlx5_meter_domain domain,
15391 struct rte_mtr_error *error)
15393 struct mlx5_priv *priv = dev->data->dev_private;
15394 struct rte_flow_error flow_err;
15395 const struct rte_flow_action *act;
15396 uint64_t action_flags;
15397 struct mlx5_flow_handle dh;
15398 struct mlx5_flow dev_flow;
15399 struct mlx5_flow_dv_port_id_action_resource port_id_action;
15401 uint8_t egress, transfer;
15402 struct mlx5_meter_policy_action_container *act_cnt = NULL;
15404 struct mlx5_flow_dv_modify_hdr_resource res;
15405 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
15406 sizeof(struct mlx5_modification_cmd) *
15407 (MLX5_MAX_MODIFY_NUM + 1)];
15409 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
15410 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
15413 egress = (domain == MLX5_MTR_DOMAIN_EGRESS) ? 1 : 0;
15414 transfer = (domain == MLX5_MTR_DOMAIN_TRANSFER) ? 1 : 0;
15415 memset(&dh, 0, sizeof(struct mlx5_flow_handle));
15416 memset(&dev_flow, 0, sizeof(struct mlx5_flow));
15417 memset(&port_id_action, 0,
15418 sizeof(struct mlx5_flow_dv_port_id_action_resource));
15419 memset(mhdr_res, 0, sizeof(*mhdr_res));
15420 mhdr_res->ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
15421 (egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
15422 MLX5DV_FLOW_TABLE_TYPE_NIC_RX);
15423 dev_flow.handle = &dh;
15424 dev_flow.dv.port_id_action = &port_id_action;
15425 dev_flow.external = true;
15426 for (i = 0; i < RTE_COLORS; i++) {
15427 if (i < MLX5_MTR_RTE_COLORS)
15428 act_cnt = &mtr_policy->act_cnt[i];
15429 /* Skip the color policy actions creation. */
15430 if ((i == RTE_COLOR_YELLOW && mtr_policy->skip_y) ||
15431 (i == RTE_COLOR_GREEN && mtr_policy->skip_g))
15434 for (act = actions[i];
15435 act && act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
15436 switch (act->type) {
15437 case RTE_FLOW_ACTION_TYPE_MARK:
15439 uint32_t tag_be = mlx5_flow_mark_set
15440 (((const struct rte_flow_action_mark *)
15443 if (i >= MLX5_MTR_RTE_COLORS)
15444 return -rte_mtr_error_set(error,
15446 RTE_MTR_ERROR_TYPE_METER_POLICY,
15448 "cannot create policy "
15449 "mark action for this color");
15451 if (flow_dv_tag_resource_register(dev, tag_be,
15452 &dev_flow, &flow_err))
15453 return -rte_mtr_error_set(error,
15455 RTE_MTR_ERROR_TYPE_METER_POLICY,
15457 "cannot setup policy mark action");
15458 MLX5_ASSERT(dev_flow.dv.tag_resource);
15459 act_cnt->rix_mark =
15460 dev_flow.handle->dvh.rix_tag;
15461 action_flags |= MLX5_FLOW_ACTION_MARK;
15464 case RTE_FLOW_ACTION_TYPE_SET_TAG:
15465 if (i >= MLX5_MTR_RTE_COLORS)
15466 return -rte_mtr_error_set(error,
15468 RTE_MTR_ERROR_TYPE_METER_POLICY,
15470 "cannot create policy "
15471 "set tag action for this color");
15472 if (flow_dv_convert_action_set_tag
15474 (const struct rte_flow_action_set_tag *)
15475 act->conf, &flow_err))
15476 return -rte_mtr_error_set(error,
15478 RTE_MTR_ERROR_TYPE_METER_POLICY,
15479 NULL, "cannot convert policy "
15481 if (!mhdr_res->actions_num)
15482 return -rte_mtr_error_set(error,
15484 RTE_MTR_ERROR_TYPE_METER_POLICY,
15485 NULL, "cannot find policy "
15487 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
15489 case RTE_FLOW_ACTION_TYPE_DROP:
15491 struct mlx5_flow_mtr_mng *mtrmng =
15493 struct mlx5_flow_tbl_data_entry *tbl_data;
15496 * Create the drop table with
15497 * METER DROP level.
15499 if (!mtrmng->drop_tbl[domain]) {
15500 mtrmng->drop_tbl[domain] =
15501 flow_dv_tbl_resource_get(dev,
15502 MLX5_FLOW_TABLE_LEVEL_METER,
15503 egress, transfer, false, NULL, 0,
15504 0, MLX5_MTR_TABLE_ID_DROP, &flow_err);
15505 if (!mtrmng->drop_tbl[domain])
15506 return -rte_mtr_error_set
15508 RTE_MTR_ERROR_TYPE_METER_POLICY,
15510 "Failed to create meter drop table");
15512 tbl_data = container_of
15513 (mtrmng->drop_tbl[domain],
15514 struct mlx5_flow_tbl_data_entry, tbl);
15515 if (i < MLX5_MTR_RTE_COLORS) {
15516 act_cnt->dr_jump_action[domain] =
15517 tbl_data->jump.action;
15518 act_cnt->fate_action =
15519 MLX5_FLOW_FATE_DROP;
15521 if (i == RTE_COLOR_RED)
15522 mtr_policy->dr_drop_action[domain] =
15523 tbl_data->jump.action;
15524 action_flags |= MLX5_FLOW_ACTION_DROP;
15527 case RTE_FLOW_ACTION_TYPE_QUEUE:
15529 if (i >= MLX5_MTR_RTE_COLORS)
15530 return -rte_mtr_error_set(error,
15532 RTE_MTR_ERROR_TYPE_METER_POLICY,
15533 NULL, "cannot create policy "
15534 "fate queue for this color");
15536 ((const struct rte_flow_action_queue *)
15537 (act->conf))->index;
15538 act_cnt->fate_action =
15539 MLX5_FLOW_FATE_QUEUE;
15540 dev_flow.handle->fate_action =
15541 MLX5_FLOW_FATE_QUEUE;
15542 mtr_policy->is_queue = 1;
15543 action_flags |= MLX5_FLOW_ACTION_QUEUE;
15546 case RTE_FLOW_ACTION_TYPE_RSS:
15550 if (i >= MLX5_MTR_RTE_COLORS)
15551 return -rte_mtr_error_set(error,
15553 RTE_MTR_ERROR_TYPE_METER_POLICY,
15555 "cannot create policy "
15556 "rss action for this color");
15558 * Save RSS conf into policy struct
15559 * for translate stage.
15561 rss_size = (int)rte_flow_conv
15562 (RTE_FLOW_CONV_OP_ACTION,
15563 NULL, 0, act, &flow_err);
15565 return -rte_mtr_error_set(error,
15567 RTE_MTR_ERROR_TYPE_METER_POLICY,
15568 NULL, "Get the wrong "
15569 "rss action struct size");
15570 act_cnt->rss = mlx5_malloc(MLX5_MEM_ZERO,
15571 rss_size, 0, SOCKET_ID_ANY);
15573 return -rte_mtr_error_set(error,
15575 RTE_MTR_ERROR_TYPE_METER_POLICY,
15577 "Fail to malloc rss action memory");
15578 ret = rte_flow_conv(RTE_FLOW_CONV_OP_ACTION,
15579 act_cnt->rss, rss_size,
15582 return -rte_mtr_error_set(error,
15584 RTE_MTR_ERROR_TYPE_METER_POLICY,
15585 NULL, "Fail to save "
15586 "rss action into policy struct");
15587 act_cnt->fate_action =
15588 MLX5_FLOW_FATE_SHARED_RSS;
15589 action_flags |= MLX5_FLOW_ACTION_RSS;
15592 case RTE_FLOW_ACTION_TYPE_PORT_ID:
15593 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
15595 struct mlx5_flow_dv_port_id_action_resource
15597 uint32_t port_id = 0;
15599 if (i >= MLX5_MTR_RTE_COLORS)
15600 return -rte_mtr_error_set(error,
15602 RTE_MTR_ERROR_TYPE_METER_POLICY,
15603 NULL, "cannot create policy "
15604 "port action for this color");
15605 memset(&port_id_resource, 0,
15606 sizeof(port_id_resource));
15607 if (flow_dv_translate_action_port_id(dev, act,
15608 &port_id, &flow_err))
15609 return -rte_mtr_error_set(error,
15611 RTE_MTR_ERROR_TYPE_METER_POLICY,
15612 NULL, "cannot translate "
15613 "policy port action");
15614 port_id_resource.port_id = port_id;
15615 if (flow_dv_port_id_action_resource_register
15616 (dev, &port_id_resource,
15617 &dev_flow, &flow_err))
15618 return -rte_mtr_error_set(error,
15620 RTE_MTR_ERROR_TYPE_METER_POLICY,
15621 NULL, "cannot setup "
15622 "policy port action");
15623 act_cnt->rix_port_id_action =
15624 dev_flow.handle->rix_port_id_action;
15625 act_cnt->fate_action =
15626 MLX5_FLOW_FATE_PORT_ID;
15627 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
15630 case RTE_FLOW_ACTION_TYPE_JUMP:
15632 uint32_t jump_group = 0;
15633 uint32_t table = 0;
15634 struct mlx5_flow_tbl_data_entry *tbl_data;
15635 struct flow_grp_info grp_info = {
15636 .external = !!dev_flow.external,
15637 .transfer = !!transfer,
15638 .fdb_def_rule = !!priv->fdb_def_rule,
15640 .skip_scale = dev_flow.skip_scale &
15641 (1 << MLX5_SCALE_FLOW_GROUP_BIT),
15643 struct mlx5_flow_meter_sub_policy *sub_policy =
15644 mtr_policy->sub_policys[domain][0];
15646 if (i >= MLX5_MTR_RTE_COLORS)
15647 return -rte_mtr_error_set(error,
15649 RTE_MTR_ERROR_TYPE_METER_POLICY,
15651 "cannot create policy "
15652 "jump action for this color");
15654 ((const struct rte_flow_action_jump *)
15656 if (mlx5_flow_group_to_table(dev, NULL,
15659 &grp_info, &flow_err))
15660 return -rte_mtr_error_set(error,
15662 RTE_MTR_ERROR_TYPE_METER_POLICY,
15663 NULL, "cannot setup "
15664 "policy jump action");
15665 sub_policy->jump_tbl[i] =
15666 flow_dv_tbl_resource_get(dev,
15669 !!dev_flow.external,
15670 NULL, jump_group, 0,
15673 (!sub_policy->jump_tbl[i])
15674 return -rte_mtr_error_set(error,
15676 RTE_MTR_ERROR_TYPE_METER_POLICY,
15677 NULL, "cannot create jump action.");
15678 tbl_data = container_of
15679 (sub_policy->jump_tbl[i],
15680 struct mlx5_flow_tbl_data_entry, tbl);
15681 act_cnt->dr_jump_action[domain] =
15682 tbl_data->jump.action;
15683 act_cnt->fate_action =
15684 MLX5_FLOW_FATE_JUMP;
15685 action_flags |= MLX5_FLOW_ACTION_JUMP;
15689 * No need to check meter hierarchy for Y or R colors
15690 * here since it is done in the validation stage.
15692 case RTE_FLOW_ACTION_TYPE_METER:
15694 const struct rte_flow_action_meter *mtr;
15695 struct mlx5_flow_meter_info *next_fm;
15696 struct mlx5_flow_meter_policy *next_policy;
15697 struct rte_flow_action tag_action;
15698 struct mlx5_rte_flow_action_set_tag set_tag;
15699 uint32_t next_mtr_idx = 0;
15702 next_fm = mlx5_flow_meter_find(priv,
15706 return -rte_mtr_error_set(error, EINVAL,
15707 RTE_MTR_ERROR_TYPE_MTR_ID, NULL,
15708 "Fail to find next meter.");
15709 if (next_fm->def_policy)
15710 return -rte_mtr_error_set(error, EINVAL,
15711 RTE_MTR_ERROR_TYPE_MTR_ID, NULL,
15712 "Hierarchy only supports termination meter.");
15713 next_policy = mlx5_flow_meter_policy_find(dev,
15714 next_fm->policy_id, NULL);
15715 MLX5_ASSERT(next_policy);
15716 if (next_fm->drop_cnt) {
15719 mlx5_flow_get_reg_id(dev,
15722 (struct rte_flow_error *)error);
15723 set_tag.offset = (priv->mtr_reg_share ?
15724 MLX5_MTR_COLOR_BITS : 0);
15725 set_tag.length = (priv->mtr_reg_share ?
15726 MLX5_MTR_IDLE_BITS_IN_COLOR_REG :
15728 set_tag.data = next_mtr_idx;
15730 (enum rte_flow_action_type)
15731 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
15732 tag_action.conf = &set_tag;
15733 if (flow_dv_convert_action_set_reg
15734 (mhdr_res, &tag_action,
15735 (struct rte_flow_error *)error))
15738 MLX5_FLOW_ACTION_SET_TAG;
15740 act_cnt->fate_action = MLX5_FLOW_FATE_MTR;
15741 act_cnt->next_mtr_id = next_fm->meter_id;
15742 act_cnt->next_sub_policy = NULL;
15743 mtr_policy->is_hierarchy = 1;
15744 mtr_policy->dev = next_policy->dev;
15746 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY;
15750 return -rte_mtr_error_set(error, ENOTSUP,
15751 RTE_MTR_ERROR_TYPE_METER_POLICY,
15752 NULL, "action type not supported");
15754 if (action_flags & MLX5_FLOW_ACTION_SET_TAG) {
15755 /* create modify action if needed. */
15756 dev_flow.dv.group = 1;
15757 if (flow_dv_modify_hdr_resource_register
15758 (dev, mhdr_res, &dev_flow, &flow_err))
15759 return -rte_mtr_error_set(error,
15761 RTE_MTR_ERROR_TYPE_METER_POLICY,
15762 NULL, "cannot register policy "
15764 act_cnt->modify_hdr =
15765 dev_flow.handle->dvh.modify_hdr;
15773 * Create policy action per domain, lock free,
15774 * (mutex should be acquired by caller).
15775 * Dispatcher for action type specific call.
15778 * Pointer to the Ethernet device structure.
15779 * @param[in] mtr_policy
15780 * Meter policy struct.
15781 * @param[in] action
15782 * Action specification used to create meter actions.
15783 * @param[out] error
15784 * Perform verbose error reporting if not NULL. Initialized in case of
15788 * 0 on success, otherwise negative errno value.
15791 flow_dv_create_mtr_policy_acts(struct rte_eth_dev *dev,
15792 struct mlx5_flow_meter_policy *mtr_policy,
15793 const struct rte_flow_action *actions[RTE_COLORS],
15794 struct rte_mtr_error *error)
15797 uint16_t sub_policy_num;
15799 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
15800 sub_policy_num = (mtr_policy->sub_policy_num >>
15801 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i)) &
15802 MLX5_MTR_SUB_POLICY_NUM_MASK;
15803 if (sub_policy_num) {
15804 ret = __flow_dv_create_domain_policy_acts(dev,
15805 mtr_policy, actions,
15806 (enum mlx5_meter_domain)i, error);
15807 /* Cleaning resource is done in the caller level. */
15816 * Query a DV flow rule for its statistics via DevX.
15819 * Pointer to Ethernet device.
15820 * @param[in] cnt_idx
15821 * Index to the flow counter.
15823 * Data retrieved by the query.
15824 * @param[out] error
15825 * Perform verbose error reporting if not NULL.
15828 * 0 on success, a negative errno value otherwise and rte_errno is set.
15831 flow_dv_query_count(struct rte_eth_dev *dev, uint32_t cnt_idx, void *data,
15832 struct rte_flow_error *error)
15834 struct mlx5_priv *priv = dev->data->dev_private;
15835 struct rte_flow_query_count *qc = data;
15837 if (!priv->sh->cdev->config.devx)
15838 return rte_flow_error_set(error, ENOTSUP,
15839 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15841 "counters are not supported");
15843 uint64_t pkts, bytes;
15844 struct mlx5_flow_counter *cnt;
15845 int err = _flow_dv_query_count(dev, cnt_idx, &pkts, &bytes);
15848 return rte_flow_error_set(error, -err,
15849 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15850 NULL, "cannot read counters");
15851 cnt = flow_dv_counter_get_by_idx(dev, cnt_idx, NULL);
15854 qc->hits = pkts - cnt->hits;
15855 qc->bytes = bytes - cnt->bytes;
15858 cnt->bytes = bytes;
15862 return rte_flow_error_set(error, EINVAL,
15863 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15865 "counters are not available");
15869 flow_dv_action_query(struct rte_eth_dev *dev,
15870 const struct rte_flow_action_handle *handle, void *data,
15871 struct rte_flow_error *error)
15873 struct mlx5_age_param *age_param;
15874 struct rte_flow_query_age *resp;
15875 uint32_t act_idx = (uint32_t)(uintptr_t)handle;
15876 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
15877 uint32_t idx = act_idx & ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
15878 struct mlx5_priv *priv = dev->data->dev_private;
15879 struct mlx5_aso_ct_action *ct;
15884 case MLX5_INDIRECT_ACTION_TYPE_AGE:
15885 age_param = &flow_aso_age_get_by_idx(dev, idx)->age_params;
15887 resp->aged = __atomic_load_n(&age_param->state,
15888 __ATOMIC_RELAXED) == AGE_TMOUT ?
15890 resp->sec_since_last_hit_valid = !resp->aged;
15891 if (resp->sec_since_last_hit_valid)
15892 resp->sec_since_last_hit = __atomic_load_n
15893 (&age_param->sec_since_last_hit, __ATOMIC_RELAXED);
15895 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
15896 return flow_dv_query_count(dev, idx, data, error);
15897 case MLX5_INDIRECT_ACTION_TYPE_CT:
15898 owner = (uint16_t)MLX5_INDIRECT_ACT_CT_GET_OWNER(idx);
15899 if (owner != PORT_ID(priv))
15900 return rte_flow_error_set(error, EACCES,
15901 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15903 "CT object owned by another port");
15904 dev_idx = MLX5_INDIRECT_ACT_CT_GET_IDX(idx);
15905 ct = flow_aso_ct_get_by_dev_idx(dev, dev_idx);
15908 return rte_flow_error_set(error, EFAULT,
15909 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15911 "CT object is inactive");
15912 ((struct rte_flow_action_conntrack *)data)->peer_port =
15914 ((struct rte_flow_action_conntrack *)data)->is_original_dir =
15916 if (mlx5_aso_ct_query_by_wqe(priv->sh, ct, data))
15917 return rte_flow_error_set(error, EIO,
15918 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15920 "Failed to query CT context");
15923 return rte_flow_error_set(error, ENOTSUP,
15924 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
15925 "action type query not supported");
15930 * Query a flow rule AGE action for aging information.
15933 * Pointer to Ethernet device.
15935 * Pointer to the sub flow.
15937 * data retrieved by the query.
15938 * @param[out] error
15939 * Perform verbose error reporting if not NULL.
15942 * 0 on success, a negative errno value otherwise and rte_errno is set.
15945 flow_dv_query_age(struct rte_eth_dev *dev, struct rte_flow *flow,
15946 void *data, struct rte_flow_error *error)
15948 struct rte_flow_query_age *resp = data;
15949 struct mlx5_age_param *age_param;
15952 struct mlx5_aso_age_action *act =
15953 flow_aso_age_get_by_idx(dev, flow->age);
15955 age_param = &act->age_params;
15956 } else if (flow->counter) {
15957 age_param = flow_dv_counter_idx_get_age(dev, flow->counter);
15959 if (!age_param || !age_param->timeout)
15960 return rte_flow_error_set
15962 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15963 NULL, "cannot read age data");
15965 return rte_flow_error_set(error, EINVAL,
15966 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15967 NULL, "age data not available");
15969 resp->aged = __atomic_load_n(&age_param->state, __ATOMIC_RELAXED) ==
15971 resp->sec_since_last_hit_valid = !resp->aged;
15972 if (resp->sec_since_last_hit_valid)
15973 resp->sec_since_last_hit = __atomic_load_n
15974 (&age_param->sec_since_last_hit, __ATOMIC_RELAXED);
15981 * @see rte_flow_query()
15982 * @see rte_flow_ops
15985 flow_dv_query(struct rte_eth_dev *dev,
15986 struct rte_flow *flow __rte_unused,
15987 const struct rte_flow_action *actions __rte_unused,
15988 void *data __rte_unused,
15989 struct rte_flow_error *error __rte_unused)
15993 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
15994 switch (actions->type) {
15995 case RTE_FLOW_ACTION_TYPE_VOID:
15997 case RTE_FLOW_ACTION_TYPE_COUNT:
15998 ret = flow_dv_query_count(dev, flow->counter, data,
16001 case RTE_FLOW_ACTION_TYPE_AGE:
16002 ret = flow_dv_query_age(dev, flow, data, error);
16005 return rte_flow_error_set(error, ENOTSUP,
16006 RTE_FLOW_ERROR_TYPE_ACTION,
16008 "action not supported");
16015 * Destroy the meter table set.
16016 * Lock free, (mutex should be acquired by caller).
16019 * Pointer to Ethernet device.
16021 * Meter information table.
16024 flow_dv_destroy_mtr_tbls(struct rte_eth_dev *dev,
16025 struct mlx5_flow_meter_info *fm)
16027 struct mlx5_priv *priv = dev->data->dev_private;
16030 if (!fm || !priv->sh->config.dv_flow_en)
16032 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16033 if (fm->drop_rule[i]) {
16034 claim_zero(mlx5_flow_os_destroy_flow(fm->drop_rule[i]));
16035 fm->drop_rule[i] = NULL;
16041 flow_dv_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
16043 struct mlx5_priv *priv = dev->data->dev_private;
16044 struct mlx5_flow_mtr_mng *mtrmng = priv->sh->mtrmng;
16045 struct mlx5_flow_tbl_data_entry *tbl;
16048 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16049 if (mtrmng->def_rule[i]) {
16050 claim_zero(mlx5_flow_os_destroy_flow
16051 (mtrmng->def_rule[i]));
16052 mtrmng->def_rule[i] = NULL;
16054 if (mtrmng->def_matcher[i]) {
16055 tbl = container_of(mtrmng->def_matcher[i]->tbl,
16056 struct mlx5_flow_tbl_data_entry, tbl);
16057 mlx5_list_unregister(tbl->matchers,
16058 &mtrmng->def_matcher[i]->entry);
16059 mtrmng->def_matcher[i] = NULL;
16061 for (j = 0; j < MLX5_REG_BITS; j++) {
16062 if (mtrmng->drop_matcher[i][j]) {
16064 container_of(mtrmng->drop_matcher[i][j]->tbl,
16065 struct mlx5_flow_tbl_data_entry,
16067 mlx5_list_unregister(tbl->matchers,
16068 &mtrmng->drop_matcher[i][j]->entry);
16069 mtrmng->drop_matcher[i][j] = NULL;
16072 if (mtrmng->drop_tbl[i]) {
16073 flow_dv_tbl_resource_release(MLX5_SH(dev),
16074 mtrmng->drop_tbl[i]);
16075 mtrmng->drop_tbl[i] = NULL;
16080 /* Number of meter flow actions, count and jump or count and drop. */
16081 #define METER_ACTIONS 2
16084 __flow_dv_destroy_domain_def_policy(struct rte_eth_dev *dev,
16085 enum mlx5_meter_domain domain)
16087 struct mlx5_priv *priv = dev->data->dev_private;
16088 struct mlx5_flow_meter_def_policy *def_policy =
16089 priv->sh->mtrmng->def_policy[domain];
16091 __flow_dv_destroy_sub_policy_rules(dev, &def_policy->sub_policy);
16092 mlx5_free(def_policy);
16093 priv->sh->mtrmng->def_policy[domain] = NULL;
16097 * Destroy the default policy table set.
16100 * Pointer to Ethernet device.
16103 flow_dv_destroy_def_policy(struct rte_eth_dev *dev)
16105 struct mlx5_priv *priv = dev->data->dev_private;
16108 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++)
16109 if (priv->sh->mtrmng->def_policy[i])
16110 __flow_dv_destroy_domain_def_policy(dev,
16111 (enum mlx5_meter_domain)i);
16112 priv->sh->mtrmng->def_policy_id = MLX5_INVALID_POLICY_ID;
16116 __flow_dv_create_policy_flow(struct rte_eth_dev *dev,
16117 uint32_t color_reg_c_idx,
16118 enum rte_color color, void *matcher_object,
16119 int actions_n, void *actions,
16120 bool match_src_port, const struct rte_flow_item *item,
16121 void **rule, const struct rte_flow_attr *attr)
16124 struct mlx5_flow_dv_match_params value = {
16125 .size = sizeof(value.buf),
16127 struct mlx5_flow_dv_match_params matcher = {
16128 .size = sizeof(matcher.buf),
16130 struct mlx5_priv *priv = dev->data->dev_private;
16133 if (match_src_port && priv->sh->esw_mode) {
16134 if (flow_dv_translate_item_port_id(dev, matcher.buf,
16135 value.buf, item, attr)) {
16136 DRV_LOG(ERR, "Failed to create meter policy%d flow's"
16137 " value with port.", color);
16141 flow_dv_match_meta_reg(matcher.buf, value.buf,
16142 (enum modify_reg)color_reg_c_idx,
16143 rte_col_2_mlx5_col(color), UINT32_MAX);
16144 misc_mask = flow_dv_matcher_enable(value.buf);
16145 __flow_dv_adjust_buf_size(&value.size, misc_mask);
16146 ret = mlx5_flow_os_create_flow(matcher_object, (void *)&value,
16147 actions_n, actions, rule);
16149 DRV_LOG(ERR, "Failed to create meter policy%d flow.", color);
16156 __flow_dv_create_policy_matcher(struct rte_eth_dev *dev,
16157 uint32_t color_reg_c_idx,
16159 struct mlx5_flow_meter_sub_policy *sub_policy,
16160 const struct rte_flow_attr *attr,
16161 bool match_src_port,
16162 const struct rte_flow_item *item,
16163 struct mlx5_flow_dv_matcher **policy_matcher,
16164 struct rte_flow_error *error)
16166 struct mlx5_list_entry *entry;
16167 struct mlx5_flow_tbl_resource *tbl_rsc = sub_policy->tbl_rsc;
16168 struct mlx5_flow_dv_matcher matcher = {
16170 .size = sizeof(matcher.mask.buf),
16174 struct mlx5_flow_dv_match_params value = {
16175 .size = sizeof(value.buf),
16177 struct mlx5_flow_cb_ctx ctx = {
16181 struct mlx5_flow_tbl_data_entry *tbl_data;
16182 struct mlx5_priv *priv = dev->data->dev_private;
16183 const uint32_t color_mask = (UINT32_C(1) << MLX5_MTR_COLOR_BITS) - 1;
16185 if (match_src_port && priv->sh->esw_mode) {
16186 if (flow_dv_translate_item_port_id(dev, matcher.mask.buf,
16187 value.buf, item, attr)) {
16188 DRV_LOG(ERR, "Failed to register meter policy%d matcher"
16189 " with port.", priority);
16193 tbl_data = container_of(tbl_rsc, struct mlx5_flow_tbl_data_entry, tbl);
16194 if (priority < RTE_COLOR_RED)
16195 flow_dv_match_meta_reg(matcher.mask.buf, value.buf,
16196 (enum modify_reg)color_reg_c_idx, 0, color_mask);
16197 matcher.priority = priority;
16198 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
16199 matcher.mask.size);
16200 entry = mlx5_list_register(tbl_data->matchers, &ctx);
16202 DRV_LOG(ERR, "Failed to register meter drop matcher.");
16206 container_of(entry, struct mlx5_flow_dv_matcher, entry);
16211 * Create the policy rules per domain.
16214 * Pointer to Ethernet device.
16215 * @param[in] sub_policy
16216 * Pointer to sub policy table..
16217 * @param[in] egress
16218 * Direction of the table.
16219 * @param[in] transfer
16220 * E-Switch or NIC flow.
16222 * Pointer to policy action list per color.
16225 * 0 on success, -1 otherwise.
16228 __flow_dv_create_domain_policy_rules(struct rte_eth_dev *dev,
16229 struct mlx5_flow_meter_sub_policy *sub_policy,
16230 uint8_t egress, uint8_t transfer, bool match_src_port,
16231 struct mlx5_meter_policy_acts acts[RTE_COLORS])
16233 struct mlx5_priv *priv = dev->data->dev_private;
16234 struct rte_flow_error flow_err;
16235 uint32_t color_reg_c_idx;
16236 struct rte_flow_attr attr = {
16237 .group = MLX5_FLOW_TABLE_LEVEL_POLICY,
16240 .egress = !!egress,
16241 .transfer = !!transfer,
16245 int ret = mlx5_flow_get_reg_id(dev, MLX5_MTR_COLOR, 0, &flow_err);
16246 struct mlx5_sub_policy_color_rule *color_rule;
16248 struct mlx5_sub_policy_color_rule *tmp_rules[RTE_COLORS] = {NULL};
16252 /* Create policy table with POLICY level. */
16253 if (!sub_policy->tbl_rsc)
16254 sub_policy->tbl_rsc = flow_dv_tbl_resource_get(dev,
16255 MLX5_FLOW_TABLE_LEVEL_POLICY,
16256 egress, transfer, false, NULL, 0, 0,
16257 sub_policy->idx, &flow_err);
16258 if (!sub_policy->tbl_rsc) {
16260 "Failed to create meter sub policy table.");
16263 /* Prepare matchers. */
16264 color_reg_c_idx = ret;
16265 for (i = 0; i < RTE_COLORS; i++) {
16266 TAILQ_INIT(&sub_policy->color_rules[i]);
16267 if (!acts[i].actions_n)
16269 color_rule = mlx5_malloc(MLX5_MEM_ZERO,
16270 sizeof(struct mlx5_sub_policy_color_rule),
16273 DRV_LOG(ERR, "No memory to create color rule.");
16276 tmp_rules[i] = color_rule;
16277 TAILQ_INSERT_TAIL(&sub_policy->color_rules[i],
16278 color_rule, next_port);
16279 color_rule->src_port = priv->representor_id;
16282 /* Create matchers for colors. */
16283 svport_match = (i != RTE_COLOR_RED) ? match_src_port : false;
16284 if (__flow_dv_create_policy_matcher(dev, color_reg_c_idx,
16285 MLX5_MTR_POLICY_MATCHER_PRIO, sub_policy,
16286 &attr, svport_match, NULL,
16287 &color_rule->matcher, &flow_err)) {
16288 DRV_LOG(ERR, "Failed to create color%u matcher.", i);
16291 /* Create flow, matching color. */
16292 if (__flow_dv_create_policy_flow(dev,
16293 color_reg_c_idx, (enum rte_color)i,
16294 color_rule->matcher->matcher_object,
16295 acts[i].actions_n, acts[i].dv_actions,
16296 svport_match, NULL, &color_rule->rule,
16298 DRV_LOG(ERR, "Failed to create color%u rule.", i);
16304 /* All the policy rules will be cleared. */
16306 color_rule = tmp_rules[i];
16308 if (color_rule->rule)
16309 mlx5_flow_os_destroy_flow(color_rule->rule);
16310 if (color_rule->matcher) {
16311 struct mlx5_flow_tbl_data_entry *tbl =
16312 container_of(color_rule->matcher->tbl,
16313 typeof(*tbl), tbl);
16314 mlx5_list_unregister(tbl->matchers,
16315 &color_rule->matcher->entry);
16317 TAILQ_REMOVE(&sub_policy->color_rules[i],
16318 color_rule, next_port);
16319 mlx5_free(color_rule);
16326 __flow_dv_create_policy_acts_rules(struct rte_eth_dev *dev,
16327 struct mlx5_flow_meter_policy *mtr_policy,
16328 struct mlx5_flow_meter_sub_policy *sub_policy,
16331 struct mlx5_priv *priv = dev->data->dev_private;
16332 struct mlx5_meter_policy_acts acts[RTE_COLORS];
16333 struct mlx5_flow_dv_tag_resource *tag;
16334 struct mlx5_flow_dv_port_id_action_resource *port_action;
16335 struct mlx5_hrxq *hrxq;
16336 struct mlx5_flow_meter_info *next_fm = NULL;
16337 struct mlx5_flow_meter_policy *next_policy;
16338 struct mlx5_flow_meter_sub_policy *next_sub_policy;
16339 struct mlx5_flow_tbl_data_entry *tbl_data;
16340 struct rte_flow_error error;
16341 uint8_t egress = (domain == MLX5_MTR_DOMAIN_EGRESS) ? 1 : 0;
16342 uint8_t transfer = (domain == MLX5_MTR_DOMAIN_TRANSFER) ? 1 : 0;
16343 bool mtr_first = egress || (transfer && priv->representor_id != UINT16_MAX);
16344 bool match_src_port = false;
16347 /* If RSS or Queue, no previous actions / rules is created. */
16348 for (i = 0; i < RTE_COLORS; i++) {
16349 acts[i].actions_n = 0;
16350 if (i == RTE_COLOR_RED) {
16351 /* Only support drop on red. */
16352 acts[i].dv_actions[0] =
16353 mtr_policy->dr_drop_action[domain];
16354 acts[i].actions_n = 1;
16357 if (i == RTE_COLOR_GREEN &&
16358 mtr_policy->act_cnt[i].fate_action == MLX5_FLOW_FATE_MTR) {
16359 struct rte_flow_attr attr = {
16360 .transfer = transfer
16363 next_fm = mlx5_flow_meter_find(priv,
16364 mtr_policy->act_cnt[i].next_mtr_id,
16368 "Failed to get next hierarchy meter.");
16371 if (mlx5_flow_meter_attach(priv, next_fm,
16373 DRV_LOG(ERR, "%s", error.message);
16377 /* Meter action must be the first for TX. */
16379 acts[i].dv_actions[acts[i].actions_n] =
16380 next_fm->meter_action;
16381 acts[i].actions_n++;
16384 if (mtr_policy->act_cnt[i].rix_mark) {
16385 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG],
16386 mtr_policy->act_cnt[i].rix_mark);
16388 DRV_LOG(ERR, "Failed to find "
16389 "mark action for policy.");
16392 acts[i].dv_actions[acts[i].actions_n] = tag->action;
16393 acts[i].actions_n++;
16395 if (mtr_policy->act_cnt[i].modify_hdr) {
16396 acts[i].dv_actions[acts[i].actions_n] =
16397 mtr_policy->act_cnt[i].modify_hdr->action;
16398 acts[i].actions_n++;
16400 if (mtr_policy->act_cnt[i].fate_action) {
16401 switch (mtr_policy->act_cnt[i].fate_action) {
16402 case MLX5_FLOW_FATE_PORT_ID:
16403 port_action = mlx5_ipool_get
16404 (priv->sh->ipool[MLX5_IPOOL_PORT_ID],
16405 mtr_policy->act_cnt[i].rix_port_id_action);
16406 if (!port_action) {
16407 DRV_LOG(ERR, "Failed to find "
16408 "port action for policy.");
16411 acts[i].dv_actions[acts[i].actions_n] =
16412 port_action->action;
16413 acts[i].actions_n++;
16414 mtr_policy->dev = dev;
16415 match_src_port = true;
16417 case MLX5_FLOW_FATE_DROP:
16418 case MLX5_FLOW_FATE_JUMP:
16419 acts[i].dv_actions[acts[i].actions_n] =
16420 mtr_policy->act_cnt[i].dr_jump_action[domain];
16421 acts[i].actions_n++;
16423 case MLX5_FLOW_FATE_SHARED_RSS:
16424 case MLX5_FLOW_FATE_QUEUE:
16425 hrxq = mlx5_ipool_get
16426 (priv->sh->ipool[MLX5_IPOOL_HRXQ],
16427 sub_policy->rix_hrxq[i]);
16429 DRV_LOG(ERR, "Failed to find "
16430 "queue action for policy.");
16433 acts[i].dv_actions[acts[i].actions_n] =
16435 acts[i].actions_n++;
16437 case MLX5_FLOW_FATE_MTR:
16440 "No next hierarchy meter.");
16444 acts[i].dv_actions[acts[i].actions_n] =
16445 next_fm->meter_action;
16446 acts[i].actions_n++;
16448 if (mtr_policy->act_cnt[i].next_sub_policy) {
16450 mtr_policy->act_cnt[i].next_sub_policy;
16453 mlx5_flow_meter_policy_find(dev,
16454 next_fm->policy_id, NULL);
16455 MLX5_ASSERT(next_policy);
16457 next_policy->sub_policys[domain][0];
16460 container_of(next_sub_policy->tbl_rsc,
16461 struct mlx5_flow_tbl_data_entry, tbl);
16462 acts[i].dv_actions[acts[i].actions_n++] =
16463 tbl_data->jump.action;
16464 if (mtr_policy->act_cnt[i].modify_hdr)
16465 match_src_port = !!transfer;
16468 /*Queue action do nothing*/
16473 if (__flow_dv_create_domain_policy_rules(dev, sub_policy,
16474 egress, transfer, match_src_port, acts)) {
16476 "Failed to create policy rules per domain.");
16482 mlx5_flow_meter_detach(priv, next_fm);
16487 * Create the policy rules.
16490 * Pointer to Ethernet device.
16491 * @param[in,out] mtr_policy
16492 * Pointer to meter policy table.
16495 * 0 on success, -1 otherwise.
16498 flow_dv_create_policy_rules(struct rte_eth_dev *dev,
16499 struct mlx5_flow_meter_policy *mtr_policy)
16502 uint16_t sub_policy_num;
16504 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16505 sub_policy_num = (mtr_policy->sub_policy_num >>
16506 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i)) &
16507 MLX5_MTR_SUB_POLICY_NUM_MASK;
16508 if (!sub_policy_num)
16510 /* Prepare actions list and create policy rules. */
16511 if (__flow_dv_create_policy_acts_rules(dev, mtr_policy,
16512 mtr_policy->sub_policys[i][0], i)) {
16513 DRV_LOG(ERR, "Failed to create policy action "
16514 "list per domain.");
16522 __flow_dv_create_domain_def_policy(struct rte_eth_dev *dev, uint32_t domain)
16524 struct mlx5_priv *priv = dev->data->dev_private;
16525 struct mlx5_flow_mtr_mng *mtrmng = priv->sh->mtrmng;
16526 struct mlx5_flow_meter_def_policy *def_policy;
16527 struct mlx5_flow_tbl_resource *jump_tbl;
16528 struct mlx5_flow_tbl_data_entry *tbl_data;
16529 uint8_t egress, transfer;
16530 struct rte_flow_error error;
16531 struct mlx5_meter_policy_acts acts[RTE_COLORS];
16534 egress = (domain == MLX5_MTR_DOMAIN_EGRESS) ? 1 : 0;
16535 transfer = (domain == MLX5_MTR_DOMAIN_TRANSFER) ? 1 : 0;
16536 def_policy = mtrmng->def_policy[domain];
16538 def_policy = mlx5_malloc(MLX5_MEM_ZERO,
16539 sizeof(struct mlx5_flow_meter_def_policy),
16540 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
16542 DRV_LOG(ERR, "Failed to alloc default policy table.");
16543 goto def_policy_error;
16545 mtrmng->def_policy[domain] = def_policy;
16546 /* Create the meter suffix table with SUFFIX level. */
16547 jump_tbl = flow_dv_tbl_resource_get(dev,
16548 MLX5_FLOW_TABLE_LEVEL_METER,
16549 egress, transfer, false, NULL, 0,
16550 0, MLX5_MTR_TABLE_ID_SUFFIX, &error);
16553 "Failed to create meter suffix table.");
16554 goto def_policy_error;
16556 def_policy->sub_policy.jump_tbl[RTE_COLOR_GREEN] = jump_tbl;
16557 tbl_data = container_of(jump_tbl,
16558 struct mlx5_flow_tbl_data_entry, tbl);
16559 def_policy->dr_jump_action[RTE_COLOR_GREEN] =
16560 tbl_data->jump.action;
16561 acts[RTE_COLOR_GREEN].dv_actions[0] = tbl_data->jump.action;
16562 acts[RTE_COLOR_GREEN].actions_n = 1;
16564 * YELLOW has the same default policy as GREEN does.
16565 * G & Y share the same table and action. The 2nd time of table
16566 * resource getting is just to update the reference count for
16567 * the releasing stage.
16569 jump_tbl = flow_dv_tbl_resource_get(dev,
16570 MLX5_FLOW_TABLE_LEVEL_METER,
16571 egress, transfer, false, NULL, 0,
16572 0, MLX5_MTR_TABLE_ID_SUFFIX, &error);
16575 "Failed to get meter suffix table.");
16576 goto def_policy_error;
16578 def_policy->sub_policy.jump_tbl[RTE_COLOR_YELLOW] = jump_tbl;
16579 tbl_data = container_of(jump_tbl,
16580 struct mlx5_flow_tbl_data_entry, tbl);
16581 def_policy->dr_jump_action[RTE_COLOR_YELLOW] =
16582 tbl_data->jump.action;
16583 acts[RTE_COLOR_YELLOW].dv_actions[0] = tbl_data->jump.action;
16584 acts[RTE_COLOR_YELLOW].actions_n = 1;
16585 /* Create jump action to the drop table. */
16586 if (!mtrmng->drop_tbl[domain]) {
16587 mtrmng->drop_tbl[domain] = flow_dv_tbl_resource_get
16588 (dev, MLX5_FLOW_TABLE_LEVEL_METER,
16589 egress, transfer, false, NULL, 0,
16590 0, MLX5_MTR_TABLE_ID_DROP, &error);
16591 if (!mtrmng->drop_tbl[domain]) {
16592 DRV_LOG(ERR, "Failed to create meter "
16593 "drop table for default policy.");
16594 goto def_policy_error;
16597 /* all RED: unique Drop table for jump action. */
16598 tbl_data = container_of(mtrmng->drop_tbl[domain],
16599 struct mlx5_flow_tbl_data_entry, tbl);
16600 def_policy->dr_jump_action[RTE_COLOR_RED] =
16601 tbl_data->jump.action;
16602 acts[RTE_COLOR_RED].dv_actions[0] = tbl_data->jump.action;
16603 acts[RTE_COLOR_RED].actions_n = 1;
16604 /* Create default policy rules. */
16605 ret = __flow_dv_create_domain_policy_rules(dev,
16606 &def_policy->sub_policy,
16607 egress, transfer, false, acts);
16609 DRV_LOG(ERR, "Failed to create default policy rules.");
16610 goto def_policy_error;
16615 __flow_dv_destroy_domain_def_policy(dev,
16616 (enum mlx5_meter_domain)domain);
16621 * Create the default policy table set.
16624 * Pointer to Ethernet device.
16626 * 0 on success, -1 otherwise.
16629 flow_dv_create_def_policy(struct rte_eth_dev *dev)
16631 struct mlx5_priv *priv = dev->data->dev_private;
16634 /* Non-termination policy table. */
16635 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16636 if (!priv->sh->config.dv_esw_en &&
16637 i == MLX5_MTR_DOMAIN_TRANSFER)
16639 if (__flow_dv_create_domain_def_policy(dev, i)) {
16640 DRV_LOG(ERR, "Failed to create default policy");
16641 /* Rollback the created default policies for others. */
16642 flow_dv_destroy_def_policy(dev);
16650 * Create the needed meter tables.
16651 * Lock free, (mutex should be acquired by caller).
16654 * Pointer to Ethernet device.
16656 * Meter information table.
16657 * @param[in] mtr_idx
16659 * @param[in] domain_bitmap
16662 * 0 on success, -1 otherwise.
16665 flow_dv_create_mtr_tbls(struct rte_eth_dev *dev,
16666 struct mlx5_flow_meter_info *fm,
16668 uint8_t domain_bitmap)
16670 struct mlx5_priv *priv = dev->data->dev_private;
16671 struct mlx5_flow_mtr_mng *mtrmng = priv->sh->mtrmng;
16672 struct rte_flow_error error;
16673 struct mlx5_flow_tbl_data_entry *tbl_data;
16674 uint8_t egress, transfer;
16675 void *actions[METER_ACTIONS];
16676 int domain, ret, i;
16677 struct mlx5_flow_counter *cnt;
16678 struct mlx5_flow_dv_match_params value = {
16679 .size = sizeof(value.buf),
16681 struct mlx5_flow_dv_match_params matcher_para = {
16682 .size = sizeof(matcher_para.buf),
16684 int mtr_id_reg_c = mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
16686 uint32_t mtr_id_mask = (UINT32_C(1) << mtrmng->max_mtr_bits) - 1;
16687 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
16688 struct mlx5_list_entry *entry;
16689 struct mlx5_flow_dv_matcher matcher = {
16691 .size = sizeof(matcher.mask.buf),
16694 struct mlx5_flow_dv_matcher *drop_matcher;
16695 struct mlx5_flow_cb_ctx ctx = {
16701 if (!priv->mtr_en || mtr_id_reg_c < 0) {
16702 rte_errno = ENOTSUP;
16705 for (domain = 0; domain < MLX5_MTR_DOMAIN_MAX; domain++) {
16706 if (!(domain_bitmap & (1 << domain)) ||
16707 (mtrmng->def_rule[domain] && !fm->drop_cnt))
16709 egress = (domain == MLX5_MTR_DOMAIN_EGRESS) ? 1 : 0;
16710 transfer = (domain == MLX5_MTR_DOMAIN_TRANSFER) ? 1 : 0;
16711 /* Create the drop table with METER DROP level. */
16712 if (!mtrmng->drop_tbl[domain]) {
16713 mtrmng->drop_tbl[domain] = flow_dv_tbl_resource_get(dev,
16714 MLX5_FLOW_TABLE_LEVEL_METER,
16715 egress, transfer, false, NULL, 0,
16716 0, MLX5_MTR_TABLE_ID_DROP, &error);
16717 if (!mtrmng->drop_tbl[domain]) {
16718 DRV_LOG(ERR, "Failed to create meter drop table.");
16722 /* Create default matcher in drop table. */
16723 matcher.tbl = mtrmng->drop_tbl[domain],
16724 tbl_data = container_of(mtrmng->drop_tbl[domain],
16725 struct mlx5_flow_tbl_data_entry, tbl);
16726 if (!mtrmng->def_matcher[domain]) {
16727 flow_dv_match_meta_reg(matcher.mask.buf, value.buf,
16728 (enum modify_reg)mtr_id_reg_c,
16730 matcher.priority = MLX5_MTRS_DEFAULT_RULE_PRIORITY;
16731 matcher.crc = rte_raw_cksum
16732 ((const void *)matcher.mask.buf,
16733 matcher.mask.size);
16734 entry = mlx5_list_register(tbl_data->matchers, &ctx);
16736 DRV_LOG(ERR, "Failed to register meter "
16737 "drop default matcher.");
16740 mtrmng->def_matcher[domain] = container_of(entry,
16741 struct mlx5_flow_dv_matcher, entry);
16743 /* Create default rule in drop table. */
16744 if (!mtrmng->def_rule[domain]) {
16746 actions[i++] = priv->sh->dr_drop_action;
16747 flow_dv_match_meta_reg(matcher_para.buf, value.buf,
16748 (enum modify_reg)mtr_id_reg_c, 0, 0);
16749 misc_mask = flow_dv_matcher_enable(value.buf);
16750 __flow_dv_adjust_buf_size(&value.size, misc_mask);
16751 ret = mlx5_flow_os_create_flow
16752 (mtrmng->def_matcher[domain]->matcher_object,
16753 (void *)&value, i, actions,
16754 &mtrmng->def_rule[domain]);
16756 DRV_LOG(ERR, "Failed to create meter "
16757 "default drop rule for drop table.");
16763 MLX5_ASSERT(mtrmng->max_mtr_bits);
16764 if (!mtrmng->drop_matcher[domain][mtrmng->max_mtr_bits - 1]) {
16765 /* Create matchers for Drop. */
16766 flow_dv_match_meta_reg(matcher.mask.buf, value.buf,
16767 (enum modify_reg)mtr_id_reg_c, 0,
16768 (mtr_id_mask << mtr_id_offset));
16769 matcher.priority = MLX5_REG_BITS - mtrmng->max_mtr_bits;
16770 matcher.crc = rte_raw_cksum
16771 ((const void *)matcher.mask.buf,
16772 matcher.mask.size);
16773 entry = mlx5_list_register(tbl_data->matchers, &ctx);
16776 "Failed to register meter drop matcher.");
16779 mtrmng->drop_matcher[domain][mtrmng->max_mtr_bits - 1] =
16780 container_of(entry, struct mlx5_flow_dv_matcher,
16784 mtrmng->drop_matcher[domain][mtrmng->max_mtr_bits - 1];
16785 /* Create drop rule, matching meter_id only. */
16786 flow_dv_match_meta_reg(matcher_para.buf, value.buf,
16787 (enum modify_reg)mtr_id_reg_c,
16788 (mtr_idx << mtr_id_offset), UINT32_MAX);
16790 cnt = flow_dv_counter_get_by_idx(dev,
16791 fm->drop_cnt, NULL);
16792 actions[i++] = cnt->action;
16793 actions[i++] = priv->sh->dr_drop_action;
16794 misc_mask = flow_dv_matcher_enable(value.buf);
16795 __flow_dv_adjust_buf_size(&value.size, misc_mask);
16796 ret = mlx5_flow_os_create_flow(drop_matcher->matcher_object,
16797 (void *)&value, i, actions,
16798 &fm->drop_rule[domain]);
16800 DRV_LOG(ERR, "Failed to create meter "
16801 "drop rule for drop table.");
16807 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16808 if (fm->drop_rule[i]) {
16809 claim_zero(mlx5_flow_os_destroy_flow
16810 (fm->drop_rule[i]));
16811 fm->drop_rule[i] = NULL;
16817 static struct mlx5_flow_meter_sub_policy *
16818 __flow_dv_meter_get_rss_sub_policy(struct rte_eth_dev *dev,
16819 struct mlx5_flow_meter_policy *mtr_policy,
16820 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS],
16821 struct mlx5_flow_meter_sub_policy *next_sub_policy,
16824 struct mlx5_priv *priv = dev->data->dev_private;
16825 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
16826 uint32_t sub_policy_idx = 0;
16827 uint32_t hrxq_idx[MLX5_MTR_RTE_COLORS] = {0};
16829 struct mlx5_hrxq *hrxq;
16830 struct mlx5_flow_handle dh;
16831 struct mlx5_meter_policy_action_container *act_cnt;
16832 uint32_t domain = MLX5_MTR_DOMAIN_INGRESS;
16833 uint16_t sub_policy_num;
16834 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
16837 rte_spinlock_lock(&mtr_policy->sl);
16838 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
16841 hrxq = mlx5_hrxq_get(dev, rss_desc[i]);
16843 rte_spinlock_unlock(&mtr_policy->sl);
16846 hrxq_idx[i] = hrxq->idx;
16848 sub_policy_num = (mtr_policy->sub_policy_num >>
16849 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
16850 MLX5_MTR_SUB_POLICY_NUM_MASK;
16851 for (j = 0; j < sub_policy_num; j++) {
16852 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
16855 mtr_policy->sub_policys[domain][j]->rix_hrxq[i])
16858 if (i >= MLX5_MTR_RTE_COLORS) {
16860 * Found the sub policy table with
16861 * the same queue per color.
16863 rte_spinlock_unlock(&mtr_policy->sl);
16864 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++)
16865 mlx5_hrxq_release(dev, hrxq_idx[i]);
16867 return mtr_policy->sub_policys[domain][j];
16870 /* Create sub policy. */
16871 if (!mtr_policy->sub_policys[domain][0]->rix_hrxq[0]) {
16872 /* Reuse the first pre-allocated sub_policy. */
16873 sub_policy = mtr_policy->sub_policys[domain][0];
16874 sub_policy_idx = sub_policy->idx;
16876 sub_policy = mlx5_ipool_zmalloc
16877 (priv->sh->ipool[MLX5_IPOOL_MTR_POLICY],
16880 sub_policy_idx > MLX5_MAX_SUB_POLICY_TBL_NUM) {
16881 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++)
16882 mlx5_hrxq_release(dev, hrxq_idx[i]);
16883 goto rss_sub_policy_error;
16885 sub_policy->idx = sub_policy_idx;
16886 sub_policy->main_policy = mtr_policy;
16888 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
16891 sub_policy->rix_hrxq[i] = hrxq_idx[i];
16892 if (mtr_policy->is_hierarchy) {
16893 act_cnt = &mtr_policy->act_cnt[i];
16894 act_cnt->next_sub_policy = next_sub_policy;
16895 mlx5_hrxq_release(dev, hrxq_idx[i]);
16898 * Overwrite the last action from
16899 * RSS action to Queue action.
16901 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
16904 DRV_LOG(ERR, "Failed to get policy hrxq");
16905 goto rss_sub_policy_error;
16907 act_cnt = &mtr_policy->act_cnt[i];
16908 if (act_cnt->rix_mark || act_cnt->modify_hdr) {
16909 memset(&dh, 0, sizeof(struct mlx5_flow_handle));
16910 if (act_cnt->rix_mark)
16912 dh.fate_action = MLX5_FLOW_FATE_QUEUE;
16913 dh.rix_hrxq = hrxq_idx[i];
16914 flow_drv_rxq_flags_set(dev, &dh);
16918 if (__flow_dv_create_policy_acts_rules(dev, mtr_policy,
16919 sub_policy, domain)) {
16920 DRV_LOG(ERR, "Failed to create policy "
16921 "rules for ingress domain.");
16922 goto rss_sub_policy_error;
16924 if (sub_policy != mtr_policy->sub_policys[domain][0]) {
16925 i = (mtr_policy->sub_policy_num >>
16926 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
16927 MLX5_MTR_SUB_POLICY_NUM_MASK;
16928 if (i >= MLX5_MTR_RSS_MAX_SUB_POLICY) {
16929 DRV_LOG(ERR, "No free sub-policy slot.");
16930 goto rss_sub_policy_error;
16932 mtr_policy->sub_policys[domain][i] = sub_policy;
16934 mtr_policy->sub_policy_num &= ~(MLX5_MTR_SUB_POLICY_NUM_MASK <<
16935 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain));
16936 mtr_policy->sub_policy_num |=
16937 (i & MLX5_MTR_SUB_POLICY_NUM_MASK) <<
16938 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain);
16940 rte_spinlock_unlock(&mtr_policy->sl);
16943 rss_sub_policy_error:
16945 __flow_dv_destroy_sub_policy_rules(dev, sub_policy);
16946 if (sub_policy != mtr_policy->sub_policys[domain][0]) {
16947 i = (mtr_policy->sub_policy_num >>
16948 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
16949 MLX5_MTR_SUB_POLICY_NUM_MASK;
16950 mtr_policy->sub_policys[domain][i] = NULL;
16951 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MTR_POLICY],
16955 rte_spinlock_unlock(&mtr_policy->sl);
16960 * Find the policy table for prefix table with RSS.
16963 * Pointer to Ethernet device.
16964 * @param[in] mtr_policy
16965 * Pointer to meter policy table.
16966 * @param[in] rss_desc
16967 * Pointer to rss_desc
16969 * Pointer to table set on success, NULL otherwise and rte_errno is set.
16971 static struct mlx5_flow_meter_sub_policy *
16972 flow_dv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
16973 struct mlx5_flow_meter_policy *mtr_policy,
16974 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
16976 struct mlx5_priv *priv = dev->data->dev_private;
16977 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
16978 struct mlx5_flow_meter_info *next_fm;
16979 struct mlx5_flow_meter_policy *next_policy;
16980 struct mlx5_flow_meter_sub_policy *next_sub_policy = NULL;
16981 struct mlx5_flow_meter_policy *policies[MLX5_MTR_CHAIN_MAX_NUM];
16982 struct mlx5_flow_meter_sub_policy *sub_policies[MLX5_MTR_CHAIN_MAX_NUM];
16983 uint32_t domain = MLX5_MTR_DOMAIN_INGRESS;
16984 bool reuse_sub_policy;
16989 /* Iterate hierarchy to get all policies in this hierarchy. */
16990 policies[i++] = mtr_policy;
16991 if (!mtr_policy->is_hierarchy)
16993 if (i >= MLX5_MTR_CHAIN_MAX_NUM) {
16994 DRV_LOG(ERR, "Exceed max meter number in hierarchy.");
16997 next_fm = mlx5_flow_meter_find(priv,
16998 mtr_policy->act_cnt[RTE_COLOR_GREEN].next_mtr_id, NULL);
17000 DRV_LOG(ERR, "Failed to get next meter in hierarchy.");
17004 mlx5_flow_meter_policy_find(dev, next_fm->policy_id,
17006 MLX5_ASSERT(next_policy);
17007 mtr_policy = next_policy;
17011 * From last policy to the first one in hierarchy,
17012 * create / get the sub policy for each of them.
17014 sub_policy = __flow_dv_meter_get_rss_sub_policy(dev,
17018 &reuse_sub_policy);
17020 DRV_LOG(ERR, "Failed to get the sub policy.");
17023 if (!reuse_sub_policy)
17024 sub_policies[j++] = sub_policy;
17025 next_sub_policy = sub_policy;
17030 uint16_t sub_policy_num;
17032 sub_policy = sub_policies[--j];
17033 mtr_policy = sub_policy->main_policy;
17034 __flow_dv_destroy_sub_policy_rules(dev, sub_policy);
17035 if (sub_policy != mtr_policy->sub_policys[domain][0]) {
17036 sub_policy_num = (mtr_policy->sub_policy_num >>
17037 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
17038 MLX5_MTR_SUB_POLICY_NUM_MASK;
17039 mtr_policy->sub_policys[domain][sub_policy_num - 1] =
17042 mtr_policy->sub_policy_num &=
17043 ~(MLX5_MTR_SUB_POLICY_NUM_MASK <<
17044 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i));
17045 mtr_policy->sub_policy_num |=
17046 (sub_policy_num & MLX5_MTR_SUB_POLICY_NUM_MASK) <<
17047 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i);
17048 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MTR_POLICY],
17056 * Create the sub policy tag rule for all meters in hierarchy.
17059 * Pointer to Ethernet device.
17061 * Meter information table.
17062 * @param[in] src_port
17063 * The src port this extra rule should use.
17065 * The src port match item.
17066 * @param[out] error
17067 * Perform verbose error reporting if not NULL.
17069 * 0 on success, a negative errno value otherwise and rte_errno is set.
17072 flow_dv_meter_hierarchy_rule_create(struct rte_eth_dev *dev,
17073 struct mlx5_flow_meter_info *fm,
17075 const struct rte_flow_item *item,
17076 struct rte_flow_error *error)
17078 struct mlx5_priv *priv = dev->data->dev_private;
17079 struct mlx5_flow_meter_policy *mtr_policy;
17080 struct mlx5_flow_meter_sub_policy *sub_policy;
17081 struct mlx5_flow_meter_info *next_fm = NULL;
17082 struct mlx5_flow_meter_policy *next_policy;
17083 struct mlx5_flow_meter_sub_policy *next_sub_policy;
17084 struct mlx5_flow_tbl_data_entry *tbl_data;
17085 struct mlx5_sub_policy_color_rule *color_rule;
17086 struct mlx5_meter_policy_acts acts;
17087 uint32_t color_reg_c_idx;
17088 bool mtr_first = (src_port != UINT16_MAX) ? true : false;
17089 struct rte_flow_attr attr = {
17090 .group = MLX5_FLOW_TABLE_LEVEL_POLICY,
17097 uint32_t domain = MLX5_MTR_DOMAIN_TRANSFER;
17100 mtr_policy = mlx5_flow_meter_policy_find(dev, fm->policy_id, NULL);
17101 MLX5_ASSERT(mtr_policy);
17102 if (!mtr_policy->is_hierarchy)
17104 next_fm = mlx5_flow_meter_find(priv,
17105 mtr_policy->act_cnt[RTE_COLOR_GREEN].next_mtr_id, NULL);
17107 return rte_flow_error_set(error, EINVAL,
17108 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
17109 "Failed to find next meter in hierarchy.");
17111 if (!next_fm->drop_cnt)
17113 color_reg_c_idx = mlx5_flow_get_reg_id(dev, MLX5_MTR_COLOR, 0, error);
17114 sub_policy = mtr_policy->sub_policys[domain][0];
17115 for (i = 0; i < RTE_COLORS; i++) {
17116 bool rule_exist = false;
17117 struct mlx5_meter_policy_action_container *act_cnt;
17119 if (i >= RTE_COLOR_YELLOW)
17121 TAILQ_FOREACH(color_rule,
17122 &sub_policy->color_rules[i], next_port)
17123 if (color_rule->src_port == src_port) {
17129 color_rule = mlx5_malloc(MLX5_MEM_ZERO,
17130 sizeof(struct mlx5_sub_policy_color_rule),
17133 return rte_flow_error_set(error, ENOMEM,
17134 RTE_FLOW_ERROR_TYPE_ACTION,
17135 NULL, "No memory to create tag color rule.");
17136 color_rule->src_port = src_port;
17138 next_policy = mlx5_flow_meter_policy_find(dev,
17139 next_fm->policy_id, NULL);
17140 MLX5_ASSERT(next_policy);
17141 next_sub_policy = next_policy->sub_policys[domain][0];
17142 tbl_data = container_of(next_sub_policy->tbl_rsc,
17143 struct mlx5_flow_tbl_data_entry, tbl);
17144 act_cnt = &mtr_policy->act_cnt[i];
17146 acts.dv_actions[0] = next_fm->meter_action;
17147 acts.dv_actions[1] = act_cnt->modify_hdr->action;
17149 acts.dv_actions[0] = act_cnt->modify_hdr->action;
17150 acts.dv_actions[1] = next_fm->meter_action;
17152 acts.dv_actions[2] = tbl_data->jump.action;
17153 acts.actions_n = 3;
17154 if (mlx5_flow_meter_attach(priv, next_fm, &attr, error)) {
17158 if (__flow_dv_create_policy_matcher(dev, color_reg_c_idx,
17159 MLX5_MTR_POLICY_MATCHER_PRIO, sub_policy,
17161 &color_rule->matcher, error)) {
17162 rte_flow_error_set(error, errno,
17163 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
17164 "Failed to create hierarchy meter matcher.");
17167 if (__flow_dv_create_policy_flow(dev, color_reg_c_idx,
17169 color_rule->matcher->matcher_object,
17170 acts.actions_n, acts.dv_actions,
17172 &color_rule->rule, &attr)) {
17173 rte_flow_error_set(error, errno,
17174 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
17175 "Failed to create hierarchy meter rule.");
17178 TAILQ_INSERT_TAIL(&sub_policy->color_rules[i],
17179 color_rule, next_port);
17183 * Recursive call to iterate all meters in hierarchy and
17184 * create needed rules.
17186 return flow_dv_meter_hierarchy_rule_create(dev, next_fm,
17187 src_port, item, error);
17190 if (color_rule->rule)
17191 mlx5_flow_os_destroy_flow(color_rule->rule);
17192 if (color_rule->matcher) {
17193 struct mlx5_flow_tbl_data_entry *tbl =
17194 container_of(color_rule->matcher->tbl,
17195 typeof(*tbl), tbl);
17196 mlx5_list_unregister(tbl->matchers,
17197 &color_rule->matcher->entry);
17199 mlx5_free(color_rule);
17202 mlx5_flow_meter_detach(priv, next_fm);
17207 * Destroy the sub policy table with RX queue.
17210 * Pointer to Ethernet device.
17211 * @param[in] mtr_policy
17212 * Pointer to meter policy table.
17215 flow_dv_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
17216 struct mlx5_flow_meter_policy *mtr_policy)
17218 struct mlx5_priv *priv = dev->data->dev_private;
17219 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
17220 uint32_t domain = MLX5_MTR_DOMAIN_INGRESS;
17222 uint16_t sub_policy_num, new_policy_num;
17224 rte_spinlock_lock(&mtr_policy->sl);
17225 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
17226 switch (mtr_policy->act_cnt[i].fate_action) {
17227 case MLX5_FLOW_FATE_SHARED_RSS:
17228 sub_policy_num = (mtr_policy->sub_policy_num >>
17229 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
17230 MLX5_MTR_SUB_POLICY_NUM_MASK;
17231 new_policy_num = sub_policy_num;
17232 for (j = 0; j < sub_policy_num; j++) {
17234 mtr_policy->sub_policys[domain][j];
17236 __flow_dv_destroy_sub_policy_rules(dev,
17239 mtr_policy->sub_policys[domain][0]) {
17240 mtr_policy->sub_policys[domain][j] =
17243 (priv->sh->ipool[MLX5_IPOOL_MTR_POLICY],
17249 if (new_policy_num != sub_policy_num) {
17250 mtr_policy->sub_policy_num &=
17251 ~(MLX5_MTR_SUB_POLICY_NUM_MASK <<
17252 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain));
17253 mtr_policy->sub_policy_num |=
17255 MLX5_MTR_SUB_POLICY_NUM_MASK) <<
17256 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain);
17259 case MLX5_FLOW_FATE_QUEUE:
17260 sub_policy = mtr_policy->sub_policys[domain][0];
17261 __flow_dv_destroy_sub_policy_rules(dev,
17265 /*Other actions without queue and do nothing*/
17269 rte_spinlock_unlock(&mtr_policy->sl);
17272 * Check whether the DR drop action is supported on the root table or not.
17274 * Create a simple flow with DR drop action on root table to validate
17275 * if DR drop action on root table is supported or not.
17278 * Pointer to rte_eth_dev structure.
17281 * 0 on success, a negative errno value otherwise and rte_errno is set.
17284 mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev)
17286 struct mlx5_priv *priv = dev->data->dev_private;
17287 struct mlx5_dev_ctx_shared *sh = priv->sh;
17288 struct mlx5_flow_dv_match_params mask = {
17289 .size = sizeof(mask.buf),
17291 struct mlx5_flow_dv_match_params value = {
17292 .size = sizeof(value.buf),
17294 struct mlx5dv_flow_matcher_attr dv_attr = {
17295 .type = IBV_FLOW_ATTR_NORMAL,
17297 .match_criteria_enable = 0,
17298 .match_mask = (void *)&mask,
17300 struct mlx5_flow_tbl_resource *tbl = NULL;
17301 void *matcher = NULL;
17305 tbl = flow_dv_tbl_resource_get(dev, 0, 0, 0, false, NULL,
17309 dv_attr.match_criteria_enable = flow_dv_matcher_enable(mask.buf);
17310 __flow_dv_adjust_buf_size(&mask.size, dv_attr.match_criteria_enable);
17311 ret = mlx5_flow_os_create_flow_matcher(sh->cdev->ctx, &dv_attr,
17312 tbl->obj, &matcher);
17315 __flow_dv_adjust_buf_size(&value.size, dv_attr.match_criteria_enable);
17316 ret = mlx5_flow_os_create_flow(matcher, (void *)&value, 1,
17317 &sh->dr_drop_action, &flow);
17320 * If DR drop action is not supported on root table, flow create will
17321 * be failed with EOPNOTSUPP or EPROTONOSUPPORT.
17325 (errno == EPROTONOSUPPORT || errno == EOPNOTSUPP))
17326 DRV_LOG(INFO, "DR drop action is not supported in root table.");
17328 DRV_LOG(ERR, "Unexpected error in DR drop action support detection");
17331 claim_zero(mlx5_flow_os_destroy_flow(flow));
17334 claim_zero(mlx5_flow_os_destroy_flow_matcher(matcher));
17336 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
17341 * Validate the batch counter support in root table.
17343 * Create a simple flow with invalid counter and drop action on root table to
17344 * validate if batch counter with offset on root table is supported or not.
17347 * Pointer to rte_eth_dev structure.
17350 * 0 on success, a negative errno value otherwise and rte_errno is set.
17353 mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev)
17355 struct mlx5_priv *priv = dev->data->dev_private;
17356 struct mlx5_dev_ctx_shared *sh = priv->sh;
17357 struct mlx5_flow_dv_match_params mask = {
17358 .size = sizeof(mask.buf),
17360 struct mlx5_flow_dv_match_params value = {
17361 .size = sizeof(value.buf),
17363 struct mlx5dv_flow_matcher_attr dv_attr = {
17364 .type = IBV_FLOW_ATTR_NORMAL | IBV_FLOW_ATTR_FLAGS_EGRESS,
17366 .match_criteria_enable = 0,
17367 .match_mask = (void *)&mask,
17369 void *actions[2] = { 0 };
17370 struct mlx5_flow_tbl_resource *tbl = NULL;
17371 struct mlx5_devx_obj *dcs = NULL;
17372 void *matcher = NULL;
17376 tbl = flow_dv_tbl_resource_get(dev, 0, 1, 0, false, NULL,
17380 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->cdev->ctx, 0x4);
17383 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, UINT16_MAX,
17387 dv_attr.match_criteria_enable = flow_dv_matcher_enable(mask.buf);
17388 __flow_dv_adjust_buf_size(&mask.size, dv_attr.match_criteria_enable);
17389 ret = mlx5_flow_os_create_flow_matcher(sh->cdev->ctx, &dv_attr,
17390 tbl->obj, &matcher);
17393 __flow_dv_adjust_buf_size(&value.size, dv_attr.match_criteria_enable);
17394 ret = mlx5_flow_os_create_flow(matcher, (void *)&value, 1,
17398 * If batch counter with offset is not supported, the driver will not
17399 * validate the invalid offset value, flow create should success.
17400 * In this case, it means batch counter is not supported in root table.
17402 * Otherwise, if flow create is failed, counter offset is supported.
17405 DRV_LOG(INFO, "Batch counter is not supported in root "
17406 "table. Switch to fallback mode.");
17407 rte_errno = ENOTSUP;
17409 claim_zero(mlx5_flow_os_destroy_flow(flow));
17411 /* Check matcher to make sure validate fail at flow create. */
17412 if (!matcher || (matcher && errno != EINVAL))
17413 DRV_LOG(ERR, "Unexpected error in counter offset "
17414 "support detection");
17418 claim_zero(mlx5_flow_os_destroy_flow_action(actions[0]));
17420 claim_zero(mlx5_flow_os_destroy_flow_matcher(matcher));
17422 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
17424 claim_zero(mlx5_devx_cmd_destroy(dcs));
17429 * Query a devx counter.
17432 * Pointer to the Ethernet device structure.
17434 * Index to the flow counter.
17436 * Set to clear the counter statistics.
17438 * The statistics value of packets.
17439 * @param[out] bytes
17440 * The statistics value of bytes.
17443 * 0 on success, otherwise return -1.
17446 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
17447 uint64_t *pkts, uint64_t *bytes, void **action)
17449 struct mlx5_priv *priv = dev->data->dev_private;
17450 struct mlx5_flow_counter *cnt;
17451 uint64_t inn_pkts, inn_bytes;
17454 if (!priv->sh->cdev->config.devx)
17457 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
17460 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
17462 *action = cnt->action;
17464 *pkts = inn_pkts - cnt->hits;
17465 *bytes = inn_bytes - cnt->bytes;
17467 cnt->hits = inn_pkts;
17468 cnt->bytes = inn_bytes;
17474 * Get aged-out flows.
17477 * Pointer to the Ethernet device structure.
17478 * @param[in] context
17479 * The address of an array of pointers to the aged-out flows contexts.
17480 * @param[in] nb_contexts
17481 * The length of context array pointers.
17482 * @param[out] error
17483 * Perform verbose error reporting if not NULL. Initialized in case of
17487 * how many contexts get in success, otherwise negative errno value.
17488 * if nb_contexts is 0, return the amount of all aged contexts.
17489 * if nb_contexts is not 0 , return the amount of aged flows reported
17490 * in the context array.
17491 * @note: only stub for now
17494 flow_dv_get_aged_flows(struct rte_eth_dev *dev,
17496 uint32_t nb_contexts,
17497 struct rte_flow_error *error)
17499 struct mlx5_priv *priv = dev->data->dev_private;
17500 struct mlx5_age_info *age_info;
17501 struct mlx5_age_param *age_param;
17502 struct mlx5_flow_counter *counter;
17503 struct mlx5_aso_age_action *act;
17506 if (nb_contexts && !context)
17507 return rte_flow_error_set(error, EINVAL,
17508 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
17509 NULL, "empty context");
17510 age_info = GET_PORT_AGE_INFO(priv);
17511 rte_spinlock_lock(&age_info->aged_sl);
17512 LIST_FOREACH(act, &age_info->aged_aso, next) {
17515 context[nb_flows - 1] =
17516 act->age_params.context;
17517 if (!(--nb_contexts))
17521 TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
17524 age_param = MLX5_CNT_TO_AGE(counter);
17525 context[nb_flows - 1] = age_param->context;
17526 if (!(--nb_contexts))
17530 rte_spinlock_unlock(&age_info->aged_sl);
17531 MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
17536 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
17539 flow_dv_counter_allocate(struct rte_eth_dev *dev)
17541 return flow_dv_counter_alloc(dev, 0);
17545 * Validate indirect action.
17546 * Dispatcher for action type specific validation.
17549 * Pointer to the Ethernet device structure.
17551 * Indirect action configuration.
17552 * @param[in] action
17553 * The indirect action object to validate.
17554 * @param[out] error
17555 * Perform verbose error reporting if not NULL. Initialized in case of
17559 * 0 on success, otherwise negative errno value.
17562 flow_dv_action_validate(struct rte_eth_dev *dev,
17563 const struct rte_flow_indir_action_conf *conf,
17564 const struct rte_flow_action *action,
17565 struct rte_flow_error *err)
17567 struct mlx5_priv *priv = dev->data->dev_private;
17569 RTE_SET_USED(conf);
17570 switch (action->type) {
17571 case RTE_FLOW_ACTION_TYPE_RSS:
17573 * priv->obj_ops is set according to driver capabilities.
17574 * When DevX capabilities are
17575 * sufficient, it is set to devx_obj_ops.
17576 * Otherwise, it is set to ibv_obj_ops.
17577 * ibv_obj_ops doesn't support ind_table_modify operation.
17578 * In this case the indirect RSS action can't be used.
17580 if (priv->obj_ops.ind_table_modify == NULL)
17581 return rte_flow_error_set
17583 RTE_FLOW_ERROR_TYPE_ACTION,
17585 "Indirect RSS action not supported");
17586 return mlx5_validate_action_rss(dev, action, err);
17587 case RTE_FLOW_ACTION_TYPE_AGE:
17588 if (!priv->sh->aso_age_mng)
17589 return rte_flow_error_set(err, ENOTSUP,
17590 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
17592 "Indirect age action not supported");
17593 return flow_dv_validate_action_age(0, action, dev, err);
17594 case RTE_FLOW_ACTION_TYPE_COUNT:
17595 return flow_dv_validate_action_count(dev, true, 0, err);
17596 case RTE_FLOW_ACTION_TYPE_CONNTRACK:
17597 if (!priv->sh->ct_aso_en)
17598 return rte_flow_error_set(err, ENOTSUP,
17599 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
17600 "ASO CT is not supported");
17601 return mlx5_validate_action_ct(dev, action->conf, err);
17603 return rte_flow_error_set(err, ENOTSUP,
17604 RTE_FLOW_ERROR_TYPE_ACTION,
17606 "action type not supported");
17611 * Check if the RSS configurations for colors of a meter policy match
17612 * each other, except the queues.
17615 * Pointer to the first RSS flow action.
17617 * Pointer to the second RSS flow action.
17620 * 0 on match, 1 on conflict.
17623 flow_dv_mtr_policy_rss_compare(const struct rte_flow_action_rss *r1,
17624 const struct rte_flow_action_rss *r2)
17626 if (r1 == NULL || r2 == NULL)
17628 if (!(r1->level <= 1 && r2->level <= 1) &&
17629 !(r1->level > 1 && r2->level > 1))
17631 if (r1->types != r2->types &&
17632 !((r1->types == 0 || r1->types == RTE_ETH_RSS_IP) &&
17633 (r2->types == 0 || r2->types == RTE_ETH_RSS_IP)))
17635 if (r1->key || r2->key) {
17636 const void *key1 = r1->key ? r1->key : rss_hash_default_key;
17637 const void *key2 = r2->key ? r2->key : rss_hash_default_key;
17639 if (memcmp(key1, key2, MLX5_RSS_HASH_KEY_LEN))
17646 * Validate the meter hierarchy chain for meter policy.
17649 * Pointer to the Ethernet device structure.
17650 * @param[in] meter_id
17652 * @param[in] action_flags
17653 * Holds the actions detected until now.
17654 * @param[out] is_rss
17656 * @param[out] hierarchy_domain
17657 * The domain bitmap for hierarchy policy.
17658 * @param[out] error
17659 * Perform verbose error reporting if not NULL. Initialized in case of
17663 * 0 on success, otherwise negative errno value with error set.
17666 flow_dv_validate_policy_mtr_hierarchy(struct rte_eth_dev *dev,
17668 uint64_t action_flags,
17670 uint8_t *hierarchy_domain,
17671 struct rte_mtr_error *error)
17673 struct mlx5_priv *priv = dev->data->dev_private;
17674 struct mlx5_flow_meter_info *fm;
17675 struct mlx5_flow_meter_policy *policy;
17678 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
17679 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
17680 return -rte_mtr_error_set(error, EINVAL,
17681 RTE_MTR_ERROR_TYPE_POLICER_ACTION_GREEN,
17683 "Multiple fate actions not supported.");
17684 *hierarchy_domain = 0;
17686 fm = mlx5_flow_meter_find(priv, meter_id, NULL);
17688 return -rte_mtr_error_set(error, EINVAL,
17689 RTE_MTR_ERROR_TYPE_MTR_ID, NULL,
17690 "Meter not found in meter hierarchy.");
17691 if (fm->def_policy)
17692 return -rte_mtr_error_set(error, EINVAL,
17693 RTE_MTR_ERROR_TYPE_MTR_ID, NULL,
17694 "Non termination meter not supported in hierarchy.");
17695 policy = mlx5_flow_meter_policy_find(dev, fm->policy_id, NULL);
17696 MLX5_ASSERT(policy);
17698 * Only inherit the supported domains of the first meter in
17700 * One meter supports at least one domain.
17702 if (!*hierarchy_domain) {
17703 if (policy->transfer)
17704 *hierarchy_domain |=
17705 MLX5_MTR_DOMAIN_TRANSFER_BIT;
17706 if (policy->ingress)
17707 *hierarchy_domain |=
17708 MLX5_MTR_DOMAIN_INGRESS_BIT;
17709 if (policy->egress)
17710 *hierarchy_domain |= MLX5_MTR_DOMAIN_EGRESS_BIT;
17712 if (!policy->is_hierarchy) {
17713 *is_rss = policy->is_rss;
17716 meter_id = policy->act_cnt[RTE_COLOR_GREEN].next_mtr_id;
17717 if (++cnt >= MLX5_MTR_CHAIN_MAX_NUM)
17718 return -rte_mtr_error_set(error, EINVAL,
17719 RTE_MTR_ERROR_TYPE_METER_POLICY, NULL,
17720 "Exceed max hierarchy meter number.");
17726 * Validate meter policy actions.
17727 * Dispatcher for action type specific validation.
17730 * Pointer to the Ethernet device structure.
17731 * @param[in] action
17732 * The meter policy action object to validate.
17734 * Attributes of flow to determine steering domain.
17735 * @param[out] error
17736 * Perform verbose error reporting if not NULL. Initialized in case of
17740 * 0 on success, otherwise negative errno value.
17743 flow_dv_validate_mtr_policy_acts(struct rte_eth_dev *dev,
17744 const struct rte_flow_action *actions[RTE_COLORS],
17745 struct rte_flow_attr *attr,
17747 uint8_t *domain_bitmap,
17748 uint8_t *policy_mode,
17749 struct rte_mtr_error *error)
17751 struct mlx5_priv *priv = dev->data->dev_private;
17752 struct mlx5_sh_config *dev_conf = &priv->sh->config;
17753 const struct rte_flow_action *act;
17754 uint64_t action_flags[RTE_COLORS] = {0};
17757 struct rte_flow_error flow_err;
17758 uint8_t domain_color[RTE_COLORS] = {0};
17759 uint8_t def_domain = MLX5_MTR_ALL_DOMAIN_BIT;
17760 uint8_t hierarchy_domain = 0;
17761 const struct rte_flow_action_meter *mtr;
17762 bool def_green = false;
17763 bool def_yellow = false;
17764 const struct rte_flow_action_rss *rss_color[RTE_COLORS] = {NULL};
17766 if (!dev_conf->dv_esw_en)
17767 def_domain &= ~MLX5_MTR_DOMAIN_TRANSFER_BIT;
17768 *domain_bitmap = def_domain;
17769 /* Red color could only support DROP action. */
17770 if (!actions[RTE_COLOR_RED] ||
17771 actions[RTE_COLOR_RED]->type != RTE_FLOW_ACTION_TYPE_DROP)
17772 return -rte_mtr_error_set(error, ENOTSUP,
17773 RTE_MTR_ERROR_TYPE_METER_POLICY,
17774 NULL, "Red color only supports drop action.");
17776 * Check default policy actions:
17777 * Green / Yellow: no action, Red: drop action
17778 * Either G or Y will trigger default policy actions to be created.
17780 if (!actions[RTE_COLOR_GREEN] ||
17781 actions[RTE_COLOR_GREEN]->type == RTE_FLOW_ACTION_TYPE_END)
17783 if (!actions[RTE_COLOR_YELLOW] ||
17784 actions[RTE_COLOR_YELLOW]->type == RTE_FLOW_ACTION_TYPE_END)
17786 if (def_green && def_yellow) {
17787 *policy_mode = MLX5_MTR_POLICY_MODE_DEF;
17789 } else if (!def_green && def_yellow) {
17790 *policy_mode = MLX5_MTR_POLICY_MODE_OG;
17791 } else if (def_green && !def_yellow) {
17792 *policy_mode = MLX5_MTR_POLICY_MODE_OY;
17794 *policy_mode = MLX5_MTR_POLICY_MODE_ALL;
17796 /* Set to empty string in case of NULL pointer access by user. */
17797 flow_err.message = "";
17798 for (i = 0; i < RTE_COLORS; i++) {
17800 for (action_flags[i] = 0, actions_n = 0;
17801 act && act->type != RTE_FLOW_ACTION_TYPE_END;
17803 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
17804 return -rte_mtr_error_set(error, ENOTSUP,
17805 RTE_MTR_ERROR_TYPE_METER_POLICY,
17806 NULL, "too many actions");
17807 switch (act->type) {
17808 case RTE_FLOW_ACTION_TYPE_PORT_ID:
17809 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
17810 if (!dev_conf->dv_esw_en)
17811 return -rte_mtr_error_set(error,
17813 RTE_MTR_ERROR_TYPE_METER_POLICY,
17814 NULL, "PORT action validate check"
17815 " fail for ESW disable");
17816 ret = flow_dv_validate_action_port_id(dev,
17818 act, attr, &flow_err);
17820 return -rte_mtr_error_set(error,
17822 RTE_MTR_ERROR_TYPE_METER_POLICY,
17823 NULL, flow_err.message ?
17825 "PORT action validate check fail");
17827 action_flags[i] |= MLX5_FLOW_ACTION_PORT_ID;
17829 case RTE_FLOW_ACTION_TYPE_MARK:
17830 ret = flow_dv_validate_action_mark(dev, act,
17834 return -rte_mtr_error_set(error,
17836 RTE_MTR_ERROR_TYPE_METER_POLICY,
17837 NULL, flow_err.message ?
17839 "Mark action validate check fail");
17840 if (dev_conf->dv_xmeta_en !=
17841 MLX5_XMETA_MODE_LEGACY)
17842 return -rte_mtr_error_set(error,
17844 RTE_MTR_ERROR_TYPE_METER_POLICY,
17845 NULL, "Extend MARK action is "
17846 "not supported. Please try use "
17847 "default policy for meter.");
17848 action_flags[i] |= MLX5_FLOW_ACTION_MARK;
17851 case RTE_FLOW_ACTION_TYPE_SET_TAG:
17852 ret = flow_dv_validate_action_set_tag(dev,
17853 act, action_flags[i],
17856 return -rte_mtr_error_set(error,
17858 RTE_MTR_ERROR_TYPE_METER_POLICY,
17859 NULL, flow_err.message ?
17861 "Set tag action validate check fail");
17862 action_flags[i] |= MLX5_FLOW_ACTION_SET_TAG;
17865 case RTE_FLOW_ACTION_TYPE_DROP:
17866 ret = mlx5_flow_validate_action_drop
17867 (action_flags[i], attr, &flow_err);
17869 return -rte_mtr_error_set(error,
17871 RTE_MTR_ERROR_TYPE_METER_POLICY,
17872 NULL, flow_err.message ?
17874 "Drop action validate check fail");
17875 action_flags[i] |= MLX5_FLOW_ACTION_DROP;
17878 case RTE_FLOW_ACTION_TYPE_QUEUE:
17880 * Check whether extensive
17881 * metadata feature is engaged.
17883 if (dev_conf->dv_flow_en &&
17884 (dev_conf->dv_xmeta_en !=
17885 MLX5_XMETA_MODE_LEGACY) &&
17886 mlx5_flow_ext_mreg_supported(dev))
17887 return -rte_mtr_error_set(error,
17889 RTE_MTR_ERROR_TYPE_METER_POLICY,
17890 NULL, "Queue action with meta "
17891 "is not supported. Please try use "
17892 "default policy for meter.");
17893 ret = mlx5_flow_validate_action_queue(act,
17894 action_flags[i], dev,
17897 return -rte_mtr_error_set(error,
17899 RTE_MTR_ERROR_TYPE_METER_POLICY,
17900 NULL, flow_err.message ?
17902 "Queue action validate check fail");
17903 action_flags[i] |= MLX5_FLOW_ACTION_QUEUE;
17906 case RTE_FLOW_ACTION_TYPE_RSS:
17907 if (dev_conf->dv_flow_en &&
17908 (dev_conf->dv_xmeta_en !=
17909 MLX5_XMETA_MODE_LEGACY) &&
17910 mlx5_flow_ext_mreg_supported(dev))
17911 return -rte_mtr_error_set(error,
17913 RTE_MTR_ERROR_TYPE_METER_POLICY,
17914 NULL, "RSS action with meta "
17915 "is not supported. Please try use "
17916 "default policy for meter.");
17917 ret = mlx5_validate_action_rss(dev, act,
17920 return -rte_mtr_error_set(error,
17922 RTE_MTR_ERROR_TYPE_METER_POLICY,
17923 NULL, flow_err.message ?
17925 "RSS action validate check fail");
17926 action_flags[i] |= MLX5_FLOW_ACTION_RSS;
17928 /* Either G or Y will set the RSS. */
17929 rss_color[i] = act->conf;
17931 case RTE_FLOW_ACTION_TYPE_JUMP:
17932 ret = flow_dv_validate_action_jump(dev,
17933 NULL, act, action_flags[i],
17934 attr, true, &flow_err);
17936 return -rte_mtr_error_set(error,
17938 RTE_MTR_ERROR_TYPE_METER_POLICY,
17939 NULL, flow_err.message ?
17941 "Jump action validate check fail");
17943 action_flags[i] |= MLX5_FLOW_ACTION_JUMP;
17946 * Only the last meter in the hierarchy will support
17947 * the YELLOW color steering. Then in the meter policy
17948 * actions list, there should be no other meter inside.
17950 case RTE_FLOW_ACTION_TYPE_METER:
17951 if (i != RTE_COLOR_GREEN)
17952 return -rte_mtr_error_set(error,
17954 RTE_MTR_ERROR_TYPE_METER_POLICY,
17956 "Meter hierarchy only supports GREEN color.");
17957 if (*policy_mode != MLX5_MTR_POLICY_MODE_OG)
17958 return -rte_mtr_error_set(error,
17960 RTE_MTR_ERROR_TYPE_METER_POLICY,
17962 "No yellow policy should be provided in meter hierarchy.");
17964 ret = flow_dv_validate_policy_mtr_hierarchy(dev,
17974 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY;
17977 return -rte_mtr_error_set(error, ENOTSUP,
17978 RTE_MTR_ERROR_TYPE_METER_POLICY,
17980 "Doesn't support optional action");
17983 if (action_flags[i] & MLX5_FLOW_ACTION_PORT_ID) {
17984 domain_color[i] = MLX5_MTR_DOMAIN_TRANSFER_BIT;
17985 } else if ((action_flags[i] &
17986 (MLX5_FLOW_ACTION_RSS | MLX5_FLOW_ACTION_QUEUE)) ||
17987 (action_flags[i] & MLX5_FLOW_ACTION_MARK)) {
17989 * Only support MLX5_XMETA_MODE_LEGACY
17990 * so MARK action is only in ingress domain.
17992 domain_color[i] = MLX5_MTR_DOMAIN_INGRESS_BIT;
17994 domain_color[i] = def_domain;
17995 if (action_flags[i] &&
17996 !(action_flags[i] & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
17998 ~MLX5_MTR_DOMAIN_TRANSFER_BIT;
18000 if (action_flags[i] &
18001 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY)
18002 domain_color[i] &= hierarchy_domain;
18004 * Non-termination actions only support NIC Tx domain.
18005 * The adjustion should be skipped when there is no
18006 * action or only END is provided. The default domains
18007 * bit-mask is set to find the MIN intersection.
18008 * The action flags checking should also be skipped.
18010 if ((def_green && i == RTE_COLOR_GREEN) ||
18011 (def_yellow && i == RTE_COLOR_YELLOW))
18014 * Validate the drop action mutual exclusion
18015 * with other actions. Drop action is mutually-exclusive
18016 * with any other action, except for Count action.
18018 if ((action_flags[i] & MLX5_FLOW_ACTION_DROP) &&
18019 (action_flags[i] & ~MLX5_FLOW_ACTION_DROP)) {
18020 return -rte_mtr_error_set(error, ENOTSUP,
18021 RTE_MTR_ERROR_TYPE_METER_POLICY,
18022 NULL, "Drop action is mutually-exclusive "
18023 "with any other action");
18025 /* Eswitch has few restrictions on using items and actions */
18026 if (domain_color[i] & MLX5_MTR_DOMAIN_TRANSFER_BIT) {
18027 if (!mlx5_flow_ext_mreg_supported(dev) &&
18028 action_flags[i] & MLX5_FLOW_ACTION_MARK)
18029 return -rte_mtr_error_set(error, ENOTSUP,
18030 RTE_MTR_ERROR_TYPE_METER_POLICY,
18031 NULL, "unsupported action MARK");
18032 if (action_flags[i] & MLX5_FLOW_ACTION_QUEUE)
18033 return -rte_mtr_error_set(error, ENOTSUP,
18034 RTE_MTR_ERROR_TYPE_METER_POLICY,
18035 NULL, "unsupported action QUEUE");
18036 if (action_flags[i] & MLX5_FLOW_ACTION_RSS)
18037 return -rte_mtr_error_set(error, ENOTSUP,
18038 RTE_MTR_ERROR_TYPE_METER_POLICY,
18039 NULL, "unsupported action RSS");
18040 if (!(action_flags[i] & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
18041 return -rte_mtr_error_set(error, ENOTSUP,
18042 RTE_MTR_ERROR_TYPE_METER_POLICY,
18043 NULL, "no fate action is found");
18045 if (!(action_flags[i] & MLX5_FLOW_FATE_ACTIONS) &&
18046 (domain_color[i] & MLX5_MTR_DOMAIN_INGRESS_BIT)) {
18047 if ((domain_color[i] &
18048 MLX5_MTR_DOMAIN_EGRESS_BIT))
18050 MLX5_MTR_DOMAIN_EGRESS_BIT;
18052 return -rte_mtr_error_set(error,
18054 RTE_MTR_ERROR_TYPE_METER_POLICY,
18056 "no fate action is found");
18060 /* If both colors have RSS, the attributes should be the same. */
18061 if (flow_dv_mtr_policy_rss_compare(rss_color[RTE_COLOR_GREEN],
18062 rss_color[RTE_COLOR_YELLOW]))
18063 return -rte_mtr_error_set(error, EINVAL,
18064 RTE_MTR_ERROR_TYPE_METER_POLICY,
18065 NULL, "policy RSS attr conflict");
18066 if (rss_color[RTE_COLOR_GREEN] || rss_color[RTE_COLOR_YELLOW])
18068 /* "domain_color[C]" is non-zero for each color, default is ALL. */
18069 if (!def_green && !def_yellow &&
18070 domain_color[RTE_COLOR_GREEN] != domain_color[RTE_COLOR_YELLOW] &&
18071 !(action_flags[RTE_COLOR_GREEN] & MLX5_FLOW_ACTION_DROP) &&
18072 !(action_flags[RTE_COLOR_YELLOW] & MLX5_FLOW_ACTION_DROP))
18073 return -rte_mtr_error_set(error, EINVAL,
18074 RTE_MTR_ERROR_TYPE_METER_POLICY,
18075 NULL, "policy domains conflict");
18077 * At least one color policy is listed in the actions, the domains
18078 * to be supported should be the intersection.
18080 *domain_bitmap = domain_color[RTE_COLOR_GREEN] &
18081 domain_color[RTE_COLOR_YELLOW];
18086 flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
18088 struct mlx5_priv *priv = dev->data->dev_private;
18091 if ((domains & MLX5_DOMAIN_BIT_NIC_RX) && priv->sh->rx_domain != NULL) {
18092 ret = mlx5_os_flow_dr_sync_domain(priv->sh->rx_domain,
18097 if ((domains & MLX5_DOMAIN_BIT_NIC_TX) && priv->sh->tx_domain != NULL) {
18098 ret = mlx5_os_flow_dr_sync_domain(priv->sh->tx_domain, flags);
18102 if ((domains & MLX5_DOMAIN_BIT_FDB) && priv->sh->fdb_domain != NULL) {
18103 ret = mlx5_os_flow_dr_sync_domain(priv->sh->fdb_domain, flags);
18111 * Discover the number of available flow priorities
18112 * by trying to create a flow with the highest priority value
18113 * for each possible number.
18118 * List of possible number of available priorities.
18119 * @param[in] vprio_n
18120 * Size of @p vprio array.
18122 * On success, number of available flow priorities.
18123 * On failure, a negative errno-style code and rte_errno is set.
18126 flow_dv_discover_priorities(struct rte_eth_dev *dev,
18127 const uint16_t *vprio, int vprio_n)
18129 struct mlx5_priv *priv = dev->data->dev_private;
18130 struct mlx5_indexed_pool *pool = priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW];
18131 struct rte_flow_item_eth eth;
18132 struct rte_flow_item item = {
18133 .type = RTE_FLOW_ITEM_TYPE_ETH,
18137 struct mlx5_flow_dv_matcher matcher = {
18139 .size = sizeof(matcher.mask.buf),
18142 union mlx5_flow_tbl_key tbl_key;
18143 struct mlx5_flow flow;
18145 struct rte_flow_error error;
18147 int i, err, ret = -ENOTSUP;
18150 * Prepare a flow with a catch-all pattern and a drop action.
18151 * Use drop queue, because shared drop action may be unavailable.
18153 action = priv->drop_queue.hrxq->action;
18154 if (action == NULL) {
18155 DRV_LOG(ERR, "Priority discovery requires a drop action");
18156 rte_errno = ENOTSUP;
18159 memset(&flow, 0, sizeof(flow));
18160 flow.handle = mlx5_ipool_zmalloc(pool, &flow.handle_idx);
18161 if (flow.handle == NULL) {
18162 DRV_LOG(ERR, "Cannot create flow handle");
18163 rte_errno = ENOMEM;
18166 flow.ingress = true;
18167 flow.dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
18168 flow.dv.actions[0] = action;
18169 flow.dv.actions_n = 1;
18170 memset(ð, 0, sizeof(eth));
18171 flow_dv_translate_item_eth(matcher.mask.buf, flow.dv.value.buf,
18172 &item, /* inner */ false, /* group */ 0);
18173 matcher.crc = rte_raw_cksum(matcher.mask.buf, matcher.mask.size);
18174 for (i = 0; i < vprio_n; i++) {
18175 /* Configure the next proposed maximum priority. */
18176 matcher.priority = vprio[i] - 1;
18177 memset(&tbl_key, 0, sizeof(tbl_key));
18178 err = flow_dv_matcher_register(dev, &matcher, &tbl_key, &flow,
18183 /* This action is pure SW and must always succeed. */
18184 DRV_LOG(ERR, "Cannot register matcher");
18188 /* Try to apply the flow to HW. */
18189 misc_mask = flow_dv_matcher_enable(flow.dv.value.buf);
18190 __flow_dv_adjust_buf_size(&flow.dv.value.size, misc_mask);
18191 err = mlx5_flow_os_create_flow
18192 (flow.handle->dvh.matcher->matcher_object,
18193 (void *)&flow.dv.value, flow.dv.actions_n,
18194 flow.dv.actions, &flow.handle->drv_flow);
18196 claim_zero(mlx5_flow_os_destroy_flow
18197 (flow.handle->drv_flow));
18198 flow.handle->drv_flow = NULL;
18200 claim_zero(flow_dv_matcher_release(dev, flow.handle));
18205 mlx5_ipool_free(pool, flow.handle_idx);
18206 /* Set rte_errno if no expected priority value matched. */
18212 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
18213 .validate = flow_dv_validate,
18214 .prepare = flow_dv_prepare,
18215 .translate = flow_dv_translate,
18216 .apply = flow_dv_apply,
18217 .remove = flow_dv_remove,
18218 .destroy = flow_dv_destroy,
18219 .query = flow_dv_query,
18220 .create_mtr_tbls = flow_dv_create_mtr_tbls,
18221 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbls,
18222 .destroy_mtr_drop_tbls = flow_dv_destroy_mtr_drop_tbls,
18223 .create_meter = flow_dv_mtr_alloc,
18224 .free_meter = flow_dv_aso_mtr_release_to_pool,
18225 .validate_mtr_acts = flow_dv_validate_mtr_policy_acts,
18226 .create_mtr_acts = flow_dv_create_mtr_policy_acts,
18227 .destroy_mtr_acts = flow_dv_destroy_mtr_policy_acts,
18228 .create_policy_rules = flow_dv_create_policy_rules,
18229 .destroy_policy_rules = flow_dv_destroy_policy_rules,
18230 .create_def_policy = flow_dv_create_def_policy,
18231 .destroy_def_policy = flow_dv_destroy_def_policy,
18232 .meter_sub_policy_rss_prepare = flow_dv_meter_sub_policy_rss_prepare,
18233 .meter_hierarchy_rule_create = flow_dv_meter_hierarchy_rule_create,
18234 .destroy_sub_policy_with_rxq = flow_dv_destroy_sub_policy_with_rxq,
18235 .counter_alloc = flow_dv_counter_allocate,
18236 .counter_free = flow_dv_counter_free,
18237 .counter_query = flow_dv_counter_query,
18238 .get_aged_flows = flow_dv_get_aged_flows,
18239 .action_validate = flow_dv_action_validate,
18240 .action_create = flow_dv_action_create,
18241 .action_destroy = flow_dv_action_destroy,
18242 .action_update = flow_dv_action_update,
18243 .action_query = flow_dv_action_query,
18244 .sync_domain = flow_dv_sync_domain,
18245 .discover_priorities = flow_dv_discover_priorities,
18246 .item_create = flow_dv_item_create,
18247 .item_release = flow_dv_item_release,
18250 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */