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:
165 case RTE_FLOW_ITEM_TYPE_GTP:
169 case RTE_FLOW_ITEM_TYPE_IPV4:
172 if (item->mask != NULL &&
173 ((const struct rte_flow_item_ipv4 *)
174 item->mask)->hdr.next_proto_id)
176 ((const struct rte_flow_item_ipv4 *)
177 (item->spec))->hdr.next_proto_id &
178 ((const struct rte_flow_item_ipv4 *)
179 (item->mask))->hdr.next_proto_id;
180 if ((next_protocol == IPPROTO_IPIP ||
181 next_protocol == IPPROTO_IPV6) && tunnel_decap)
184 case RTE_FLOW_ITEM_TYPE_IPV6:
187 if (item->mask != NULL &&
188 ((const struct rte_flow_item_ipv6 *)
189 item->mask)->hdr.proto)
191 ((const struct rte_flow_item_ipv6 *)
192 (item->spec))->hdr.proto &
193 ((const struct rte_flow_item_ipv6 *)
194 (item->mask))->hdr.proto;
195 if ((next_protocol == IPPROTO_IPIP ||
196 next_protocol == IPPROTO_IPV6) && tunnel_decap)
199 case RTE_FLOW_ITEM_TYPE_UDP:
203 case RTE_FLOW_ITEM_TYPE_TCP:
215 * Convert rte_mtr_color to mlx5 color.
224 rte_col_2_mlx5_col(enum rte_color rcol)
227 case RTE_COLOR_GREEN:
228 return MLX5_FLOW_COLOR_GREEN;
229 case RTE_COLOR_YELLOW:
230 return MLX5_FLOW_COLOR_YELLOW;
232 return MLX5_FLOW_COLOR_RED;
236 return MLX5_FLOW_COLOR_UNDEFINED;
239 struct field_modify_info {
240 uint32_t size; /* Size of field in protocol header, in bytes. */
241 uint32_t offset; /* Offset of field in protocol header, in bytes. */
242 enum mlx5_modification_field id;
245 struct field_modify_info modify_eth[] = {
246 {4, 0, MLX5_MODI_OUT_DMAC_47_16},
247 {2, 4, MLX5_MODI_OUT_DMAC_15_0},
248 {4, 6, MLX5_MODI_OUT_SMAC_47_16},
249 {2, 10, MLX5_MODI_OUT_SMAC_15_0},
253 struct field_modify_info modify_vlan_out_first_vid[] = {
254 /* Size in bits !!! */
255 {12, 0, MLX5_MODI_OUT_FIRST_VID},
259 struct field_modify_info modify_ipv4[] = {
260 {1, 1, MLX5_MODI_OUT_IP_DSCP},
261 {1, 8, MLX5_MODI_OUT_IPV4_TTL},
262 {4, 12, MLX5_MODI_OUT_SIPV4},
263 {4, 16, MLX5_MODI_OUT_DIPV4},
267 struct field_modify_info modify_ipv6[] = {
268 {1, 0, MLX5_MODI_OUT_IP_DSCP},
269 {1, 7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
270 {4, 8, MLX5_MODI_OUT_SIPV6_127_96},
271 {4, 12, MLX5_MODI_OUT_SIPV6_95_64},
272 {4, 16, MLX5_MODI_OUT_SIPV6_63_32},
273 {4, 20, MLX5_MODI_OUT_SIPV6_31_0},
274 {4, 24, MLX5_MODI_OUT_DIPV6_127_96},
275 {4, 28, MLX5_MODI_OUT_DIPV6_95_64},
276 {4, 32, MLX5_MODI_OUT_DIPV6_63_32},
277 {4, 36, MLX5_MODI_OUT_DIPV6_31_0},
281 struct field_modify_info modify_udp[] = {
282 {2, 0, MLX5_MODI_OUT_UDP_SPORT},
283 {2, 2, MLX5_MODI_OUT_UDP_DPORT},
287 struct field_modify_info modify_tcp[] = {
288 {2, 0, MLX5_MODI_OUT_TCP_SPORT},
289 {2, 2, MLX5_MODI_OUT_TCP_DPORT},
290 {4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
291 {4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
296 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
297 uint8_t next_protocol, uint64_t *item_flags,
300 MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
301 item->type == RTE_FLOW_ITEM_TYPE_IPV6);
302 if (next_protocol == IPPROTO_IPIP) {
303 *item_flags |= MLX5_FLOW_LAYER_IPIP;
306 if (next_protocol == IPPROTO_IPV6) {
307 *item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
312 static inline struct mlx5_hlist *
313 flow_dv_hlist_prepare(struct mlx5_dev_ctx_shared *sh, struct mlx5_hlist **phl,
314 const char *name, uint32_t size, bool direct_key,
315 bool lcores_share, void *ctx,
316 mlx5_list_create_cb cb_create,
317 mlx5_list_match_cb cb_match,
318 mlx5_list_remove_cb cb_remove,
319 mlx5_list_clone_cb cb_clone,
320 mlx5_list_clone_free_cb cb_clone_free,
321 struct rte_flow_error *error)
323 struct mlx5_hlist *hl;
324 struct mlx5_hlist *expected = NULL;
325 char s[MLX5_NAME_SIZE];
327 hl = __atomic_load_n(phl, __ATOMIC_SEQ_CST);
330 snprintf(s, sizeof(s), "%s_%s", sh->ibdev_name, name);
331 hl = mlx5_hlist_create(s, size, direct_key, lcores_share,
332 ctx, cb_create, cb_match, cb_remove, cb_clone,
335 DRV_LOG(ERR, "%s hash creation failed", name);
336 rte_flow_error_set(error, ENOMEM,
337 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
338 "cannot allocate resource memory");
341 if (!__atomic_compare_exchange_n(phl, &expected, hl, false,
344 mlx5_hlist_destroy(hl);
345 hl = __atomic_load_n(phl, __ATOMIC_SEQ_CST);
350 /* Update VLAN's VID/PCP based on input rte_flow_action.
353 * Pointer to struct rte_flow_action.
355 * Pointer to struct rte_vlan_hdr.
358 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
359 struct rte_vlan_hdr *vlan)
362 if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
364 ((const struct rte_flow_action_of_set_vlan_pcp *)
365 action->conf)->vlan_pcp;
366 vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
367 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
368 vlan->vlan_tci |= vlan_tci;
369 } else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
370 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
371 vlan->vlan_tci |= rte_be_to_cpu_16
372 (((const struct rte_flow_action_of_set_vlan_vid *)
373 action->conf)->vlan_vid);
378 * Fetch 1, 2, 3 or 4 byte field from the byte array
379 * and return as unsigned integer in host-endian format.
382 * Pointer to data array.
384 * Size of field to extract.
387 * converted field in host endian format.
389 static inline uint32_t
390 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
399 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
402 ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
403 ret = (ret << 8) | *(data + sizeof(uint16_t));
406 ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
417 * Convert modify-header action to DV specification.
419 * Data length of each action is determined by provided field description
420 * and the item mask. Data bit offset and width of each action is determined
421 * by provided item mask.
424 * Pointer to item specification.
426 * Pointer to field modification information.
427 * For MLX5_MODIFICATION_TYPE_SET specifies destination field.
428 * For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
429 * For MLX5_MODIFICATION_TYPE_COPY specifies source field.
431 * Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
432 * Negative offset value sets the same offset as source offset.
433 * size field is ignored, value is taken from source field.
434 * @param[in,out] resource
435 * Pointer to the modify-header resource.
437 * Type of modification.
439 * Pointer to the error structure.
442 * 0 on success, a negative errno value otherwise and rte_errno is set.
445 flow_dv_convert_modify_action(struct rte_flow_item *item,
446 struct field_modify_info *field,
447 struct field_modify_info *dcopy,
448 struct mlx5_flow_dv_modify_hdr_resource *resource,
449 uint32_t type, struct rte_flow_error *error)
451 uint32_t i = resource->actions_num;
452 struct mlx5_modification_cmd *actions = resource->actions;
453 uint32_t carry_b = 0;
456 * The item and mask are provided in big-endian format.
457 * The fields should be presented as in big-endian format either.
458 * Mask must be always present, it defines the actual field width.
460 MLX5_ASSERT(item->mask);
461 MLX5_ASSERT(field->size);
467 bool next_field = true;
468 bool next_dcopy = true;
470 if (i >= MLX5_MAX_MODIFY_NUM)
471 return rte_flow_error_set(error, EINVAL,
472 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
473 "too many items to modify");
474 /* Fetch variable byte size mask from the array. */
475 mask = flow_dv_fetch_field((const uint8_t *)item->mask +
476 field->offset, field->size);
481 /* Deduce actual data width in bits from mask value. */
482 off_b = rte_bsf32(mask) + carry_b;
483 size_b = sizeof(uint32_t) * CHAR_BIT -
484 off_b - __builtin_clz(mask);
486 actions[i] = (struct mlx5_modification_cmd) {
490 .length = (size_b == sizeof(uint32_t) * CHAR_BIT) ?
493 if (type == MLX5_MODIFICATION_TYPE_COPY) {
495 actions[i].dst_field = dcopy->id;
496 actions[i].dst_offset =
497 (int)dcopy->offset < 0 ? off_b : dcopy->offset;
498 /* Convert entire record to big-endian format. */
499 actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
501 * Destination field overflow. Copy leftovers of
502 * a source field to the next destination field.
505 if ((size_b > dcopy->size * CHAR_BIT - dcopy->offset) &&
508 dcopy->size * CHAR_BIT - dcopy->offset;
509 carry_b = actions[i].length;
513 * Not enough bits in a source filed to fill a
514 * destination field. Switch to the next source.
516 if ((size_b < dcopy->size * CHAR_BIT - dcopy->offset) &&
517 (size_b == field->size * CHAR_BIT - off_b)) {
519 field->size * CHAR_BIT - off_b;
520 dcopy->offset += actions[i].length;
526 MLX5_ASSERT(item->spec);
527 data = flow_dv_fetch_field((const uint8_t *)item->spec +
528 field->offset, field->size);
529 /* Shift out the trailing masked bits from data. */
530 data = (data & mask) >> off_b;
531 actions[i].data1 = rte_cpu_to_be_32(data);
533 /* Convert entire record to expected big-endian format. */
534 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
538 } while (field->size);
539 if (resource->actions_num == i)
540 return rte_flow_error_set(error, EINVAL,
541 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
542 "invalid modification flow item");
543 resource->actions_num = i;
548 * Convert modify-header set IPv4 address action to DV specification.
550 * @param[in,out] resource
551 * Pointer to the modify-header resource.
553 * Pointer to action specification.
555 * Pointer to the error structure.
558 * 0 on success, a negative errno value otherwise and rte_errno is set.
561 flow_dv_convert_action_modify_ipv4
562 (struct mlx5_flow_dv_modify_hdr_resource *resource,
563 const struct rte_flow_action *action,
564 struct rte_flow_error *error)
566 const struct rte_flow_action_set_ipv4 *conf =
567 (const struct rte_flow_action_set_ipv4 *)(action->conf);
568 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
569 struct rte_flow_item_ipv4 ipv4;
570 struct rte_flow_item_ipv4 ipv4_mask;
572 memset(&ipv4, 0, sizeof(ipv4));
573 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
574 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
575 ipv4.hdr.src_addr = conf->ipv4_addr;
576 ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
578 ipv4.hdr.dst_addr = conf->ipv4_addr;
579 ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
582 item.mask = &ipv4_mask;
583 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
584 MLX5_MODIFICATION_TYPE_SET, error);
588 * Convert modify-header set IPv6 address action to DV specification.
590 * @param[in,out] resource
591 * Pointer to the modify-header resource.
593 * Pointer to action specification.
595 * Pointer to the error structure.
598 * 0 on success, a negative errno value otherwise and rte_errno is set.
601 flow_dv_convert_action_modify_ipv6
602 (struct mlx5_flow_dv_modify_hdr_resource *resource,
603 const struct rte_flow_action *action,
604 struct rte_flow_error *error)
606 const struct rte_flow_action_set_ipv6 *conf =
607 (const struct rte_flow_action_set_ipv6 *)(action->conf);
608 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
609 struct rte_flow_item_ipv6 ipv6;
610 struct rte_flow_item_ipv6 ipv6_mask;
612 memset(&ipv6, 0, sizeof(ipv6));
613 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
614 if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
615 memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
616 sizeof(ipv6.hdr.src_addr));
617 memcpy(&ipv6_mask.hdr.src_addr,
618 &rte_flow_item_ipv6_mask.hdr.src_addr,
619 sizeof(ipv6.hdr.src_addr));
621 memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
622 sizeof(ipv6.hdr.dst_addr));
623 memcpy(&ipv6_mask.hdr.dst_addr,
624 &rte_flow_item_ipv6_mask.hdr.dst_addr,
625 sizeof(ipv6.hdr.dst_addr));
628 item.mask = &ipv6_mask;
629 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
630 MLX5_MODIFICATION_TYPE_SET, error);
634 * Convert modify-header set MAC address action to DV specification.
636 * @param[in,out] resource
637 * Pointer to the modify-header resource.
639 * Pointer to action specification.
641 * Pointer to the error structure.
644 * 0 on success, a negative errno value otherwise and rte_errno is set.
647 flow_dv_convert_action_modify_mac
648 (struct mlx5_flow_dv_modify_hdr_resource *resource,
649 const struct rte_flow_action *action,
650 struct rte_flow_error *error)
652 const struct rte_flow_action_set_mac *conf =
653 (const struct rte_flow_action_set_mac *)(action->conf);
654 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
655 struct rte_flow_item_eth eth;
656 struct rte_flow_item_eth eth_mask;
658 memset(ð, 0, sizeof(eth));
659 memset(ð_mask, 0, sizeof(eth_mask));
660 if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
661 memcpy(ð.src.addr_bytes, &conf->mac_addr,
662 sizeof(eth.src.addr_bytes));
663 memcpy(ð_mask.src.addr_bytes,
664 &rte_flow_item_eth_mask.src.addr_bytes,
665 sizeof(eth_mask.src.addr_bytes));
667 memcpy(ð.dst.addr_bytes, &conf->mac_addr,
668 sizeof(eth.dst.addr_bytes));
669 memcpy(ð_mask.dst.addr_bytes,
670 &rte_flow_item_eth_mask.dst.addr_bytes,
671 sizeof(eth_mask.dst.addr_bytes));
674 item.mask = ð_mask;
675 return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
676 MLX5_MODIFICATION_TYPE_SET, error);
680 * Convert modify-header set VLAN VID action to DV specification.
682 * @param[in,out] resource
683 * Pointer to the modify-header resource.
685 * Pointer to action specification.
687 * Pointer to the error structure.
690 * 0 on success, a negative errno value otherwise and rte_errno is set.
693 flow_dv_convert_action_modify_vlan_vid
694 (struct mlx5_flow_dv_modify_hdr_resource *resource,
695 const struct rte_flow_action *action,
696 struct rte_flow_error *error)
698 const struct rte_flow_action_of_set_vlan_vid *conf =
699 (const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
700 int i = resource->actions_num;
701 struct mlx5_modification_cmd *actions = resource->actions;
702 struct field_modify_info *field = modify_vlan_out_first_vid;
704 if (i >= MLX5_MAX_MODIFY_NUM)
705 return rte_flow_error_set(error, EINVAL,
706 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
707 "too many items to modify");
708 actions[i] = (struct mlx5_modification_cmd) {
709 .action_type = MLX5_MODIFICATION_TYPE_SET,
711 .length = field->size,
712 .offset = field->offset,
714 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
715 actions[i].data1 = conf->vlan_vid;
716 actions[i].data1 = actions[i].data1 << 16;
717 resource->actions_num = ++i;
722 * Convert modify-header set TP action to DV specification.
724 * @param[in,out] resource
725 * Pointer to the modify-header resource.
727 * Pointer to action specification.
729 * Pointer to rte_flow_item objects list.
731 * Pointer to flow attributes structure.
732 * @param[in] dev_flow
733 * Pointer to the sub flow.
734 * @param[in] tunnel_decap
735 * Whether action is after tunnel decapsulation.
737 * Pointer to the error structure.
740 * 0 on success, a negative errno value otherwise and rte_errno is set.
743 flow_dv_convert_action_modify_tp
744 (struct mlx5_flow_dv_modify_hdr_resource *resource,
745 const struct rte_flow_action *action,
746 const struct rte_flow_item *items,
747 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
748 bool tunnel_decap, struct rte_flow_error *error)
750 const struct rte_flow_action_set_tp *conf =
751 (const struct rte_flow_action_set_tp *)(action->conf);
752 struct rte_flow_item item;
753 struct rte_flow_item_udp udp;
754 struct rte_flow_item_udp udp_mask;
755 struct rte_flow_item_tcp tcp;
756 struct rte_flow_item_tcp tcp_mask;
757 struct field_modify_info *field;
760 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
762 memset(&udp, 0, sizeof(udp));
763 memset(&udp_mask, 0, sizeof(udp_mask));
764 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
765 udp.hdr.src_port = conf->port;
766 udp_mask.hdr.src_port =
767 rte_flow_item_udp_mask.hdr.src_port;
769 udp.hdr.dst_port = conf->port;
770 udp_mask.hdr.dst_port =
771 rte_flow_item_udp_mask.hdr.dst_port;
773 item.type = RTE_FLOW_ITEM_TYPE_UDP;
775 item.mask = &udp_mask;
778 MLX5_ASSERT(attr->tcp);
779 memset(&tcp, 0, sizeof(tcp));
780 memset(&tcp_mask, 0, sizeof(tcp_mask));
781 if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
782 tcp.hdr.src_port = conf->port;
783 tcp_mask.hdr.src_port =
784 rte_flow_item_tcp_mask.hdr.src_port;
786 tcp.hdr.dst_port = conf->port;
787 tcp_mask.hdr.dst_port =
788 rte_flow_item_tcp_mask.hdr.dst_port;
790 item.type = RTE_FLOW_ITEM_TYPE_TCP;
792 item.mask = &tcp_mask;
795 return flow_dv_convert_modify_action(&item, field, NULL, resource,
796 MLX5_MODIFICATION_TYPE_SET, error);
800 * Convert modify-header set TTL action to DV specification.
802 * @param[in,out] resource
803 * Pointer to the modify-header resource.
805 * Pointer to action specification.
807 * Pointer to rte_flow_item objects list.
809 * Pointer to flow attributes structure.
810 * @param[in] dev_flow
811 * Pointer to the sub flow.
812 * @param[in] tunnel_decap
813 * Whether action is after tunnel decapsulation.
815 * Pointer to the error structure.
818 * 0 on success, a negative errno value otherwise and rte_errno is set.
821 flow_dv_convert_action_modify_ttl
822 (struct mlx5_flow_dv_modify_hdr_resource *resource,
823 const struct rte_flow_action *action,
824 const struct rte_flow_item *items,
825 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
826 bool tunnel_decap, struct rte_flow_error *error)
828 const struct rte_flow_action_set_ttl *conf =
829 (const struct rte_flow_action_set_ttl *)(action->conf);
830 struct rte_flow_item item;
831 struct rte_flow_item_ipv4 ipv4;
832 struct rte_flow_item_ipv4 ipv4_mask;
833 struct rte_flow_item_ipv6 ipv6;
834 struct rte_flow_item_ipv6 ipv6_mask;
835 struct field_modify_info *field;
838 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
840 memset(&ipv4, 0, sizeof(ipv4));
841 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
842 ipv4.hdr.time_to_live = conf->ttl_value;
843 ipv4_mask.hdr.time_to_live = 0xFF;
844 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
846 item.mask = &ipv4_mask;
849 MLX5_ASSERT(attr->ipv6);
850 memset(&ipv6, 0, sizeof(ipv6));
851 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
852 ipv6.hdr.hop_limits = conf->ttl_value;
853 ipv6_mask.hdr.hop_limits = 0xFF;
854 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
856 item.mask = &ipv6_mask;
859 return flow_dv_convert_modify_action(&item, field, NULL, resource,
860 MLX5_MODIFICATION_TYPE_SET, error);
864 * Convert modify-header decrement TTL action to DV specification.
866 * @param[in,out] resource
867 * Pointer to the modify-header resource.
869 * Pointer to action specification.
871 * Pointer to rte_flow_item objects list.
873 * Pointer to flow attributes structure.
874 * @param[in] dev_flow
875 * Pointer to the sub flow.
876 * @param[in] tunnel_decap
877 * Whether action is after tunnel decapsulation.
879 * Pointer to the error structure.
882 * 0 on success, a negative errno value otherwise and rte_errno is set.
885 flow_dv_convert_action_modify_dec_ttl
886 (struct mlx5_flow_dv_modify_hdr_resource *resource,
887 const struct rte_flow_item *items,
888 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
889 bool tunnel_decap, struct rte_flow_error *error)
891 struct rte_flow_item item;
892 struct rte_flow_item_ipv4 ipv4;
893 struct rte_flow_item_ipv4 ipv4_mask;
894 struct rte_flow_item_ipv6 ipv6;
895 struct rte_flow_item_ipv6 ipv6_mask;
896 struct field_modify_info *field;
899 flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
901 memset(&ipv4, 0, sizeof(ipv4));
902 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
903 ipv4.hdr.time_to_live = 0xFF;
904 ipv4_mask.hdr.time_to_live = 0xFF;
905 item.type = RTE_FLOW_ITEM_TYPE_IPV4;
907 item.mask = &ipv4_mask;
910 MLX5_ASSERT(attr->ipv6);
911 memset(&ipv6, 0, sizeof(ipv6));
912 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
913 ipv6.hdr.hop_limits = 0xFF;
914 ipv6_mask.hdr.hop_limits = 0xFF;
915 item.type = RTE_FLOW_ITEM_TYPE_IPV6;
917 item.mask = &ipv6_mask;
920 return flow_dv_convert_modify_action(&item, field, NULL, resource,
921 MLX5_MODIFICATION_TYPE_ADD, error);
925 * Convert modify-header increment/decrement TCP Sequence number
926 * to DV specification.
928 * @param[in,out] resource
929 * Pointer to the modify-header resource.
931 * Pointer to action specification.
933 * Pointer to the error structure.
936 * 0 on success, a negative errno value otherwise and rte_errno is set.
939 flow_dv_convert_action_modify_tcp_seq
940 (struct mlx5_flow_dv_modify_hdr_resource *resource,
941 const struct rte_flow_action *action,
942 struct rte_flow_error *error)
944 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
945 uint64_t value = rte_be_to_cpu_32(*conf);
946 struct rte_flow_item item;
947 struct rte_flow_item_tcp tcp;
948 struct rte_flow_item_tcp tcp_mask;
950 memset(&tcp, 0, sizeof(tcp));
951 memset(&tcp_mask, 0, sizeof(tcp_mask));
952 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
954 * The HW has no decrement operation, only increment operation.
955 * To simulate decrement X from Y using increment operation
956 * we need to add UINT32_MAX X times to Y.
957 * Each adding of UINT32_MAX decrements Y by 1.
960 tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
961 tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
962 item.type = RTE_FLOW_ITEM_TYPE_TCP;
964 item.mask = &tcp_mask;
965 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
966 MLX5_MODIFICATION_TYPE_ADD, error);
970 * Convert modify-header increment/decrement TCP Acknowledgment number
971 * to DV specification.
973 * @param[in,out] resource
974 * Pointer to the modify-header resource.
976 * Pointer to action specification.
978 * Pointer to the error structure.
981 * 0 on success, a negative errno value otherwise and rte_errno is set.
984 flow_dv_convert_action_modify_tcp_ack
985 (struct mlx5_flow_dv_modify_hdr_resource *resource,
986 const struct rte_flow_action *action,
987 struct rte_flow_error *error)
989 const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
990 uint64_t value = rte_be_to_cpu_32(*conf);
991 struct rte_flow_item item;
992 struct rte_flow_item_tcp tcp;
993 struct rte_flow_item_tcp tcp_mask;
995 memset(&tcp, 0, sizeof(tcp));
996 memset(&tcp_mask, 0, sizeof(tcp_mask));
997 if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
999 * The HW has no decrement operation, only increment operation.
1000 * To simulate decrement X from Y using increment operation
1001 * we need to add UINT32_MAX X times to Y.
1002 * Each adding of UINT32_MAX decrements Y by 1.
1004 value *= UINT32_MAX;
1005 tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
1006 tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
1007 item.type = RTE_FLOW_ITEM_TYPE_TCP;
1009 item.mask = &tcp_mask;
1010 return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
1011 MLX5_MODIFICATION_TYPE_ADD, error);
1014 static enum mlx5_modification_field reg_to_field[] = {
1015 [REG_NON] = MLX5_MODI_OUT_NONE,
1016 [REG_A] = MLX5_MODI_META_DATA_REG_A,
1017 [REG_B] = MLX5_MODI_META_DATA_REG_B,
1018 [REG_C_0] = MLX5_MODI_META_REG_C_0,
1019 [REG_C_1] = MLX5_MODI_META_REG_C_1,
1020 [REG_C_2] = MLX5_MODI_META_REG_C_2,
1021 [REG_C_3] = MLX5_MODI_META_REG_C_3,
1022 [REG_C_4] = MLX5_MODI_META_REG_C_4,
1023 [REG_C_5] = MLX5_MODI_META_REG_C_5,
1024 [REG_C_6] = MLX5_MODI_META_REG_C_6,
1025 [REG_C_7] = MLX5_MODI_META_REG_C_7,
1029 * Convert register set to DV specification.
1031 * @param[in,out] resource
1032 * Pointer to the modify-header resource.
1034 * Pointer to action specification.
1036 * Pointer to the error structure.
1039 * 0 on success, a negative errno value otherwise and rte_errno is set.
1042 flow_dv_convert_action_set_reg
1043 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1044 const struct rte_flow_action *action,
1045 struct rte_flow_error *error)
1047 const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
1048 struct mlx5_modification_cmd *actions = resource->actions;
1049 uint32_t i = resource->actions_num;
1051 if (i >= MLX5_MAX_MODIFY_NUM)
1052 return rte_flow_error_set(error, EINVAL,
1053 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1054 "too many items to modify");
1055 MLX5_ASSERT(conf->id != REG_NON);
1056 MLX5_ASSERT(conf->id < (enum modify_reg)RTE_DIM(reg_to_field));
1057 actions[i] = (struct mlx5_modification_cmd) {
1058 .action_type = MLX5_MODIFICATION_TYPE_SET,
1059 .field = reg_to_field[conf->id],
1060 .offset = conf->offset,
1061 .length = conf->length,
1063 actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
1064 actions[i].data1 = rte_cpu_to_be_32(conf->data);
1066 resource->actions_num = i;
1071 * Convert SET_TAG action to DV specification.
1074 * Pointer to the rte_eth_dev structure.
1075 * @param[in,out] resource
1076 * Pointer to the modify-header resource.
1078 * Pointer to action specification.
1080 * Pointer to the error structure.
1083 * 0 on success, a negative errno value otherwise and rte_errno is set.
1086 flow_dv_convert_action_set_tag
1087 (struct rte_eth_dev *dev,
1088 struct mlx5_flow_dv_modify_hdr_resource *resource,
1089 const struct rte_flow_action_set_tag *conf,
1090 struct rte_flow_error *error)
1092 rte_be32_t data = rte_cpu_to_be_32(conf->data);
1093 rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1094 struct rte_flow_item item = {
1098 struct field_modify_info reg_c_x[] = {
1101 enum mlx5_modification_field reg_type;
1104 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1107 MLX5_ASSERT(ret != REG_NON);
1108 MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1109 reg_type = reg_to_field[ret];
1110 MLX5_ASSERT(reg_type > 0);
1111 reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1112 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1113 MLX5_MODIFICATION_TYPE_SET, error);
1117 * Convert internal COPY_REG action to DV specification.
1120 * Pointer to the rte_eth_dev structure.
1121 * @param[in,out] res
1122 * Pointer to the modify-header resource.
1124 * Pointer to action specification.
1126 * Pointer to the error structure.
1129 * 0 on success, a negative errno value otherwise and rte_errno is set.
1132 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1133 struct mlx5_flow_dv_modify_hdr_resource *res,
1134 const struct rte_flow_action *action,
1135 struct rte_flow_error *error)
1137 const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1138 rte_be32_t mask = RTE_BE32(UINT32_MAX);
1139 struct rte_flow_item item = {
1143 struct field_modify_info reg_src[] = {
1144 {4, 0, reg_to_field[conf->src]},
1147 struct field_modify_info reg_dst = {
1149 .id = reg_to_field[conf->dst],
1151 /* Adjust reg_c[0] usage according to reported mask. */
1152 if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1153 struct mlx5_priv *priv = dev->data->dev_private;
1154 uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1156 MLX5_ASSERT(reg_c0);
1157 MLX5_ASSERT(priv->sh->config.dv_xmeta_en !=
1158 MLX5_XMETA_MODE_LEGACY);
1159 if (conf->dst == REG_C_0) {
1160 /* Copy to reg_c[0], within mask only. */
1161 reg_dst.offset = rte_bsf32(reg_c0);
1162 mask = rte_cpu_to_be_32(reg_c0 >> reg_dst.offset);
1165 mask = rte_cpu_to_be_32(reg_c0);
1168 return flow_dv_convert_modify_action(&item,
1169 reg_src, ®_dst, res,
1170 MLX5_MODIFICATION_TYPE_COPY,
1175 * Convert MARK action to DV specification. This routine is used
1176 * in extensive metadata only and requires metadata register to be
1177 * handled. In legacy mode hardware tag resource is engaged.
1180 * Pointer to the rte_eth_dev structure.
1182 * Pointer to MARK action specification.
1183 * @param[in,out] resource
1184 * Pointer to the modify-header resource.
1186 * Pointer to the error structure.
1189 * 0 on success, a negative errno value otherwise and rte_errno is set.
1192 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1193 const struct rte_flow_action_mark *conf,
1194 struct mlx5_flow_dv_modify_hdr_resource *resource,
1195 struct rte_flow_error *error)
1197 struct mlx5_priv *priv = dev->data->dev_private;
1198 rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1199 priv->sh->dv_mark_mask);
1200 rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1201 struct rte_flow_item item = {
1205 struct field_modify_info reg_c_x[] = {
1211 return rte_flow_error_set(error, EINVAL,
1212 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1213 NULL, "zero mark action mask");
1214 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1217 MLX5_ASSERT(reg > 0);
1218 if (reg == REG_C_0) {
1219 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1220 uint32_t shl_c0 = rte_bsf32(msk_c0);
1222 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1223 mask = rte_cpu_to_be_32(mask) & msk_c0;
1224 mask = rte_cpu_to_be_32(mask << shl_c0);
1226 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1227 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1228 MLX5_MODIFICATION_TYPE_SET, error);
1232 * Get metadata register index for specified steering domain.
1235 * Pointer to the rte_eth_dev structure.
1237 * Attributes of flow to determine steering domain.
1239 * Pointer to the error structure.
1242 * positive index on success, a negative errno value otherwise
1243 * and rte_errno is set.
1245 static enum modify_reg
1246 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1247 const struct rte_flow_attr *attr,
1248 struct rte_flow_error *error)
1251 mlx5_flow_get_reg_id(dev, attr->transfer ?
1255 MLX5_METADATA_RX, 0, error);
1257 return rte_flow_error_set(error,
1258 ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1259 NULL, "unavailable "
1260 "metadata register");
1265 * Convert SET_META action to DV specification.
1268 * Pointer to the rte_eth_dev structure.
1269 * @param[in,out] resource
1270 * Pointer to the modify-header resource.
1272 * Attributes of flow that includes this item.
1274 * Pointer to action specification.
1276 * Pointer to the error structure.
1279 * 0 on success, a negative errno value otherwise and rte_errno is set.
1282 flow_dv_convert_action_set_meta
1283 (struct rte_eth_dev *dev,
1284 struct mlx5_flow_dv_modify_hdr_resource *resource,
1285 const struct rte_flow_attr *attr,
1286 const struct rte_flow_action_set_meta *conf,
1287 struct rte_flow_error *error)
1289 uint32_t mask = rte_cpu_to_be_32(conf->mask);
1290 uint32_t data = rte_cpu_to_be_32(conf->data) & mask;
1291 struct rte_flow_item item = {
1295 struct field_modify_info reg_c_x[] = {
1298 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1302 MLX5_ASSERT(reg != REG_NON);
1303 if (reg == REG_C_0) {
1304 struct mlx5_priv *priv = dev->data->dev_private;
1305 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1306 uint32_t shl_c0 = rte_bsf32(msk_c0);
1308 data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1309 mask = rte_cpu_to_be_32(mask) & msk_c0;
1310 mask = rte_cpu_to_be_32(mask << shl_c0);
1312 reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1313 /* The routine expects parameters in memory as big-endian ones. */
1314 return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1315 MLX5_MODIFICATION_TYPE_SET, error);
1319 * Convert modify-header set IPv4 DSCP action to DV specification.
1321 * @param[in,out] resource
1322 * Pointer to the modify-header resource.
1324 * Pointer to action specification.
1326 * Pointer to the error structure.
1329 * 0 on success, a negative errno value otherwise and rte_errno is set.
1332 flow_dv_convert_action_modify_ipv4_dscp
1333 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1334 const struct rte_flow_action *action,
1335 struct rte_flow_error *error)
1337 const struct rte_flow_action_set_dscp *conf =
1338 (const struct rte_flow_action_set_dscp *)(action->conf);
1339 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1340 struct rte_flow_item_ipv4 ipv4;
1341 struct rte_flow_item_ipv4 ipv4_mask;
1343 memset(&ipv4, 0, sizeof(ipv4));
1344 memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1345 ipv4.hdr.type_of_service = conf->dscp;
1346 ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1348 item.mask = &ipv4_mask;
1349 return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1350 MLX5_MODIFICATION_TYPE_SET, error);
1354 * Convert modify-header set IPv6 DSCP action to DV specification.
1356 * @param[in,out] resource
1357 * Pointer to the modify-header resource.
1359 * Pointer to action specification.
1361 * Pointer to the error structure.
1364 * 0 on success, a negative errno value otherwise and rte_errno is set.
1367 flow_dv_convert_action_modify_ipv6_dscp
1368 (struct mlx5_flow_dv_modify_hdr_resource *resource,
1369 const struct rte_flow_action *action,
1370 struct rte_flow_error *error)
1372 const struct rte_flow_action_set_dscp *conf =
1373 (const struct rte_flow_action_set_dscp *)(action->conf);
1374 struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1375 struct rte_flow_item_ipv6 ipv6;
1376 struct rte_flow_item_ipv6 ipv6_mask;
1378 memset(&ipv6, 0, sizeof(ipv6));
1379 memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1381 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1382 * rdma-core only accept the DSCP bits byte aligned start from
1383 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1384 * bits in IPv6 case as rdma-core requires byte aligned value.
1386 ipv6.hdr.vtc_flow = conf->dscp;
1387 ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1389 item.mask = &ipv6_mask;
1390 return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1391 MLX5_MODIFICATION_TYPE_SET, error);
1395 mlx5_flow_item_field_width(struct rte_eth_dev *dev,
1396 enum rte_flow_field_id field, int inherit,
1397 const struct rte_flow_attr *attr,
1398 struct rte_flow_error *error)
1400 struct mlx5_priv *priv = dev->data->dev_private;
1403 case RTE_FLOW_FIELD_START:
1405 case RTE_FLOW_FIELD_MAC_DST:
1406 case RTE_FLOW_FIELD_MAC_SRC:
1408 case RTE_FLOW_FIELD_VLAN_TYPE:
1410 case RTE_FLOW_FIELD_VLAN_ID:
1412 case RTE_FLOW_FIELD_MAC_TYPE:
1414 case RTE_FLOW_FIELD_IPV4_DSCP:
1416 case RTE_FLOW_FIELD_IPV4_TTL:
1418 case RTE_FLOW_FIELD_IPV4_SRC:
1419 case RTE_FLOW_FIELD_IPV4_DST:
1421 case RTE_FLOW_FIELD_IPV6_DSCP:
1423 case RTE_FLOW_FIELD_IPV6_HOPLIMIT:
1425 case RTE_FLOW_FIELD_IPV6_SRC:
1426 case RTE_FLOW_FIELD_IPV6_DST:
1428 case RTE_FLOW_FIELD_TCP_PORT_SRC:
1429 case RTE_FLOW_FIELD_TCP_PORT_DST:
1431 case RTE_FLOW_FIELD_TCP_SEQ_NUM:
1432 case RTE_FLOW_FIELD_TCP_ACK_NUM:
1434 case RTE_FLOW_FIELD_TCP_FLAGS:
1436 case RTE_FLOW_FIELD_UDP_PORT_SRC:
1437 case RTE_FLOW_FIELD_UDP_PORT_DST:
1439 case RTE_FLOW_FIELD_VXLAN_VNI:
1440 case RTE_FLOW_FIELD_GENEVE_VNI:
1442 case RTE_FLOW_FIELD_GTP_TEID:
1443 case RTE_FLOW_FIELD_TAG:
1445 case RTE_FLOW_FIELD_MARK:
1446 return __builtin_popcount(priv->sh->dv_mark_mask);
1447 case RTE_FLOW_FIELD_META:
1448 return (flow_dv_get_metadata_reg(dev, attr, error) == REG_C_0) ?
1449 __builtin_popcount(priv->sh->dv_meta_mask) : 32;
1450 case RTE_FLOW_FIELD_POINTER:
1451 case RTE_FLOW_FIELD_VALUE:
1452 return inherit < 0 ? 0 : inherit;
1460 mlx5_flow_field_id_to_modify_info
1461 (const struct rte_flow_action_modify_data *data,
1462 struct field_modify_info *info, uint32_t *mask,
1463 uint32_t width, struct rte_eth_dev *dev,
1464 const struct rte_flow_attr *attr, struct rte_flow_error *error)
1466 struct mlx5_priv *priv = dev->data->dev_private;
1470 switch (data->field) {
1471 case RTE_FLOW_FIELD_START:
1472 /* not supported yet */
1475 case RTE_FLOW_FIELD_MAC_DST:
1476 off = data->offset > 16 ? data->offset - 16 : 0;
1478 if (data->offset < 16) {
1479 info[idx] = (struct field_modify_info){2, 4,
1480 MLX5_MODI_OUT_DMAC_15_0};
1482 mask[1] = rte_cpu_to_be_16(0xffff >>
1486 mask[1] = RTE_BE16(0xffff);
1493 info[idx] = (struct field_modify_info){4, 0,
1494 MLX5_MODI_OUT_DMAC_47_16};
1495 mask[0] = rte_cpu_to_be_32((0xffffffff >>
1496 (32 - width)) << off);
1498 if (data->offset < 16)
1499 info[idx++] = (struct field_modify_info){2, 0,
1500 MLX5_MODI_OUT_DMAC_15_0};
1501 info[idx] = (struct field_modify_info){4, off,
1502 MLX5_MODI_OUT_DMAC_47_16};
1505 case RTE_FLOW_FIELD_MAC_SRC:
1506 off = data->offset > 16 ? data->offset - 16 : 0;
1508 if (data->offset < 16) {
1509 info[idx] = (struct field_modify_info){2, 4,
1510 MLX5_MODI_OUT_SMAC_15_0};
1512 mask[1] = rte_cpu_to_be_16(0xffff >>
1516 mask[1] = RTE_BE16(0xffff);
1523 info[idx] = (struct field_modify_info){4, 0,
1524 MLX5_MODI_OUT_SMAC_47_16};
1525 mask[0] = rte_cpu_to_be_32((0xffffffff >>
1526 (32 - width)) << off);
1528 if (data->offset < 16)
1529 info[idx++] = (struct field_modify_info){2, 0,
1530 MLX5_MODI_OUT_SMAC_15_0};
1531 info[idx] = (struct field_modify_info){4, off,
1532 MLX5_MODI_OUT_SMAC_47_16};
1535 case RTE_FLOW_FIELD_VLAN_TYPE:
1536 /* not supported yet */
1538 case RTE_FLOW_FIELD_VLAN_ID:
1539 info[idx] = (struct field_modify_info){2, 0,
1540 MLX5_MODI_OUT_FIRST_VID};
1542 mask[idx] = rte_cpu_to_be_16(0x0fff >> (12 - width));
1544 case RTE_FLOW_FIELD_MAC_TYPE:
1545 info[idx] = (struct field_modify_info){2, 0,
1546 MLX5_MODI_OUT_ETHERTYPE};
1548 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1550 case RTE_FLOW_FIELD_IPV4_DSCP:
1551 info[idx] = (struct field_modify_info){1, 0,
1552 MLX5_MODI_OUT_IP_DSCP};
1554 mask[idx] = 0x3f >> (6 - width);
1556 case RTE_FLOW_FIELD_IPV4_TTL:
1557 info[idx] = (struct field_modify_info){1, 0,
1558 MLX5_MODI_OUT_IPV4_TTL};
1560 mask[idx] = 0xff >> (8 - width);
1562 case RTE_FLOW_FIELD_IPV4_SRC:
1563 info[idx] = (struct field_modify_info){4, 0,
1564 MLX5_MODI_OUT_SIPV4};
1566 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1569 case RTE_FLOW_FIELD_IPV4_DST:
1570 info[idx] = (struct field_modify_info){4, 0,
1571 MLX5_MODI_OUT_DIPV4};
1573 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1576 case RTE_FLOW_FIELD_IPV6_DSCP:
1577 info[idx] = (struct field_modify_info){1, 0,
1578 MLX5_MODI_OUT_IP_DSCP};
1580 mask[idx] = 0x3f >> (6 - width);
1582 case RTE_FLOW_FIELD_IPV6_HOPLIMIT:
1583 info[idx] = (struct field_modify_info){1, 0,
1584 MLX5_MODI_OUT_IPV6_HOPLIMIT};
1586 mask[idx] = 0xff >> (8 - width);
1588 case RTE_FLOW_FIELD_IPV6_SRC:
1590 if (data->offset < 32) {
1591 info[idx] = (struct field_modify_info){4, 12,
1592 MLX5_MODI_OUT_SIPV6_31_0};
1595 rte_cpu_to_be_32(0xffffffff >>
1599 mask[3] = RTE_BE32(0xffffffff);
1606 if (data->offset < 64) {
1607 info[idx] = (struct field_modify_info){4, 8,
1608 MLX5_MODI_OUT_SIPV6_63_32};
1611 rte_cpu_to_be_32(0xffffffff >>
1615 mask[2] = RTE_BE32(0xffffffff);
1622 if (data->offset < 96) {
1623 info[idx] = (struct field_modify_info){4, 4,
1624 MLX5_MODI_OUT_SIPV6_95_64};
1627 rte_cpu_to_be_32(0xffffffff >>
1631 mask[1] = RTE_BE32(0xffffffff);
1638 info[idx] = (struct field_modify_info){4, 0,
1639 MLX5_MODI_OUT_SIPV6_127_96};
1640 mask[0] = rte_cpu_to_be_32(0xffffffff >> (32 - width));
1642 if (data->offset < 32)
1643 info[idx++] = (struct field_modify_info){4, 0,
1644 MLX5_MODI_OUT_SIPV6_31_0};
1645 if (data->offset < 64)
1646 info[idx++] = (struct field_modify_info){4, 0,
1647 MLX5_MODI_OUT_SIPV6_63_32};
1648 if (data->offset < 96)
1649 info[idx++] = (struct field_modify_info){4, 0,
1650 MLX5_MODI_OUT_SIPV6_95_64};
1651 if (data->offset < 128)
1652 info[idx++] = (struct field_modify_info){4, 0,
1653 MLX5_MODI_OUT_SIPV6_127_96};
1656 case RTE_FLOW_FIELD_IPV6_DST:
1658 if (data->offset < 32) {
1659 info[idx] = (struct field_modify_info){4, 12,
1660 MLX5_MODI_OUT_DIPV6_31_0};
1663 rte_cpu_to_be_32(0xffffffff >>
1667 mask[3] = RTE_BE32(0xffffffff);
1674 if (data->offset < 64) {
1675 info[idx] = (struct field_modify_info){4, 8,
1676 MLX5_MODI_OUT_DIPV6_63_32};
1679 rte_cpu_to_be_32(0xffffffff >>
1683 mask[2] = RTE_BE32(0xffffffff);
1690 if (data->offset < 96) {
1691 info[idx] = (struct field_modify_info){4, 4,
1692 MLX5_MODI_OUT_DIPV6_95_64};
1695 rte_cpu_to_be_32(0xffffffff >>
1699 mask[1] = RTE_BE32(0xffffffff);
1706 info[idx] = (struct field_modify_info){4, 0,
1707 MLX5_MODI_OUT_DIPV6_127_96};
1708 mask[0] = rte_cpu_to_be_32(0xffffffff >> (32 - width));
1710 if (data->offset < 32)
1711 info[idx++] = (struct field_modify_info){4, 0,
1712 MLX5_MODI_OUT_DIPV6_31_0};
1713 if (data->offset < 64)
1714 info[idx++] = (struct field_modify_info){4, 0,
1715 MLX5_MODI_OUT_DIPV6_63_32};
1716 if (data->offset < 96)
1717 info[idx++] = (struct field_modify_info){4, 0,
1718 MLX5_MODI_OUT_DIPV6_95_64};
1719 if (data->offset < 128)
1720 info[idx++] = (struct field_modify_info){4, 0,
1721 MLX5_MODI_OUT_DIPV6_127_96};
1724 case RTE_FLOW_FIELD_TCP_PORT_SRC:
1725 info[idx] = (struct field_modify_info){2, 0,
1726 MLX5_MODI_OUT_TCP_SPORT};
1728 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1730 case RTE_FLOW_FIELD_TCP_PORT_DST:
1731 info[idx] = (struct field_modify_info){2, 0,
1732 MLX5_MODI_OUT_TCP_DPORT};
1734 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1736 case RTE_FLOW_FIELD_TCP_SEQ_NUM:
1737 info[idx] = (struct field_modify_info){4, 0,
1738 MLX5_MODI_OUT_TCP_SEQ_NUM};
1740 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1743 case RTE_FLOW_FIELD_TCP_ACK_NUM:
1744 info[idx] = (struct field_modify_info){4, 0,
1745 MLX5_MODI_OUT_TCP_ACK_NUM};
1747 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1750 case RTE_FLOW_FIELD_TCP_FLAGS:
1751 info[idx] = (struct field_modify_info){2, 0,
1752 MLX5_MODI_OUT_TCP_FLAGS};
1754 mask[idx] = rte_cpu_to_be_16(0x1ff >> (9 - width));
1756 case RTE_FLOW_FIELD_UDP_PORT_SRC:
1757 info[idx] = (struct field_modify_info){2, 0,
1758 MLX5_MODI_OUT_UDP_SPORT};
1760 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1762 case RTE_FLOW_FIELD_UDP_PORT_DST:
1763 info[idx] = (struct field_modify_info){2, 0,
1764 MLX5_MODI_OUT_UDP_DPORT};
1766 mask[idx] = rte_cpu_to_be_16(0xffff >> (16 - width));
1768 case RTE_FLOW_FIELD_VXLAN_VNI:
1769 /* not supported yet */
1771 case RTE_FLOW_FIELD_GENEVE_VNI:
1772 /* not supported yet*/
1774 case RTE_FLOW_FIELD_GTP_TEID:
1775 info[idx] = (struct field_modify_info){4, 0,
1776 MLX5_MODI_GTP_TEID};
1778 mask[idx] = rte_cpu_to_be_32(0xffffffff >>
1781 case RTE_FLOW_FIELD_TAG:
1783 int reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG,
1784 data->level, error);
1787 MLX5_ASSERT(reg != REG_NON);
1788 MLX5_ASSERT((unsigned int)reg < RTE_DIM(reg_to_field));
1789 info[idx] = (struct field_modify_info){4, 0,
1793 rte_cpu_to_be_32(0xffffffff >>
1797 case RTE_FLOW_FIELD_MARK:
1799 uint32_t mark_mask = priv->sh->dv_mark_mask;
1800 uint32_t mark_count = __builtin_popcount(mark_mask);
1801 int reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK,
1805 MLX5_ASSERT(reg != REG_NON);
1806 MLX5_ASSERT((unsigned int)reg < RTE_DIM(reg_to_field));
1807 info[idx] = (struct field_modify_info){4, 0,
1810 mask[idx] = rte_cpu_to_be_32((mark_mask >>
1811 (mark_count - width)) & mark_mask);
1814 case RTE_FLOW_FIELD_META:
1816 uint32_t meta_mask = priv->sh->dv_meta_mask;
1817 uint32_t meta_count = __builtin_popcount(meta_mask);
1818 int reg = flow_dv_get_metadata_reg(dev, attr, error);
1821 MLX5_ASSERT(reg != REG_NON);
1822 MLX5_ASSERT((unsigned int)reg < RTE_DIM(reg_to_field));
1823 info[idx] = (struct field_modify_info){4, 0,
1826 mask[idx] = rte_cpu_to_be_32((meta_mask >>
1827 (meta_count - width)) & meta_mask);
1830 case RTE_FLOW_FIELD_POINTER:
1831 case RTE_FLOW_FIELD_VALUE:
1839 * Convert modify_field action to DV specification.
1842 * Pointer to the rte_eth_dev structure.
1843 * @param[in,out] resource
1844 * Pointer to the modify-header resource.
1846 * Pointer to action specification.
1848 * Attributes of flow that includes this item.
1850 * Pointer to the error structure.
1853 * 0 on success, a negative errno value otherwise and rte_errno is set.
1856 flow_dv_convert_action_modify_field
1857 (struct rte_eth_dev *dev,
1858 struct mlx5_flow_dv_modify_hdr_resource *resource,
1859 const struct rte_flow_action *action,
1860 const struct rte_flow_attr *attr,
1861 struct rte_flow_error *error)
1863 const struct rte_flow_action_modify_field *conf =
1864 (const struct rte_flow_action_modify_field *)(action->conf);
1865 struct rte_flow_item item = {
1869 struct field_modify_info field[MLX5_ACT_MAX_MOD_FIELDS] = {
1871 struct field_modify_info dcopy[MLX5_ACT_MAX_MOD_FIELDS] = {
1873 uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS] = {0, 0, 0, 0, 0};
1874 uint32_t type, meta = 0;
1876 if (conf->src.field == RTE_FLOW_FIELD_POINTER ||
1877 conf->src.field == RTE_FLOW_FIELD_VALUE) {
1878 type = MLX5_MODIFICATION_TYPE_SET;
1879 /** For SET fill the destination field (field) first. */
1880 mlx5_flow_field_id_to_modify_info(&conf->dst, field, mask,
1883 item.spec = conf->src.field == RTE_FLOW_FIELD_POINTER ?
1884 (void *)(uintptr_t)conf->src.pvalue :
1885 (void *)(uintptr_t)&conf->src.value;
1886 if (conf->dst.field == RTE_FLOW_FIELD_META) {
1887 meta = *(const unaligned_uint32_t *)item.spec;
1888 meta = rte_cpu_to_be_32(meta);
1892 type = MLX5_MODIFICATION_TYPE_COPY;
1893 /** For COPY fill the destination field (dcopy) without mask. */
1894 mlx5_flow_field_id_to_modify_info(&conf->dst, dcopy, NULL,
1897 /** Then construct the source field (field) with mask. */
1898 mlx5_flow_field_id_to_modify_info(&conf->src, field, mask,
1903 return flow_dv_convert_modify_action(&item,
1904 field, dcopy, resource, type, error);
1908 * Validate MARK item.
1911 * Pointer to the rte_eth_dev structure.
1913 * Item specification.
1915 * Attributes of flow that includes this item.
1917 * Pointer to error structure.
1920 * 0 on success, a negative errno value otherwise and rte_errno is set.
1923 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1924 const struct rte_flow_item *item,
1925 const struct rte_flow_attr *attr __rte_unused,
1926 struct rte_flow_error *error)
1928 struct mlx5_priv *priv = dev->data->dev_private;
1929 struct mlx5_sh_config *config = &priv->sh->config;
1930 const struct rte_flow_item_mark *spec = item->spec;
1931 const struct rte_flow_item_mark *mask = item->mask;
1932 const struct rte_flow_item_mark nic_mask = {
1933 .id = priv->sh->dv_mark_mask,
1937 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1938 return rte_flow_error_set(error, ENOTSUP,
1939 RTE_FLOW_ERROR_TYPE_ITEM, item,
1940 "extended metadata feature"
1942 if (!mlx5_flow_ext_mreg_supported(dev))
1943 return rte_flow_error_set(error, ENOTSUP,
1944 RTE_FLOW_ERROR_TYPE_ITEM, item,
1945 "extended metadata register"
1946 " isn't supported");
1948 return rte_flow_error_set(error, ENOTSUP,
1949 RTE_FLOW_ERROR_TYPE_ITEM, item,
1950 "extended metadata register"
1951 " isn't available");
1952 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1956 return rte_flow_error_set(error, EINVAL,
1957 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1959 "data cannot be empty");
1960 if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1961 return rte_flow_error_set(error, EINVAL,
1962 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1964 "mark id exceeds the limit");
1968 return rte_flow_error_set(error, EINVAL,
1969 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1970 "mask cannot be zero");
1972 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1973 (const uint8_t *)&nic_mask,
1974 sizeof(struct rte_flow_item_mark),
1975 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1982 * Validate META item.
1985 * Pointer to the rte_eth_dev structure.
1987 * Item specification.
1989 * Attributes of flow that includes this item.
1991 * Pointer to error structure.
1994 * 0 on success, a negative errno value otherwise and rte_errno is set.
1997 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1998 const struct rte_flow_item *item,
1999 const struct rte_flow_attr *attr,
2000 struct rte_flow_error *error)
2002 struct mlx5_priv *priv = dev->data->dev_private;
2003 struct mlx5_sh_config *config = &priv->sh->config;
2004 const struct rte_flow_item_meta *spec = item->spec;
2005 const struct rte_flow_item_meta *mask = item->mask;
2006 struct rte_flow_item_meta nic_mask = {
2013 return rte_flow_error_set(error, EINVAL,
2014 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2016 "data cannot be empty");
2017 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
2018 if (!mlx5_flow_ext_mreg_supported(dev))
2019 return rte_flow_error_set(error, ENOTSUP,
2020 RTE_FLOW_ERROR_TYPE_ITEM, item,
2021 "extended metadata register"
2022 " isn't supported");
2023 reg = flow_dv_get_metadata_reg(dev, attr, error);
2027 return rte_flow_error_set(error, ENOTSUP,
2028 RTE_FLOW_ERROR_TYPE_ITEM, item,
2029 "unavailable extended metadata register");
2031 return rte_flow_error_set(error, ENOTSUP,
2032 RTE_FLOW_ERROR_TYPE_ITEM, item,
2036 nic_mask.data = priv->sh->dv_meta_mask;
2039 return rte_flow_error_set(error, ENOTSUP,
2040 RTE_FLOW_ERROR_TYPE_ITEM, item,
2041 "extended metadata feature "
2042 "should be enabled when "
2043 "meta item is requested "
2044 "with e-switch mode ");
2046 return rte_flow_error_set(error, ENOTSUP,
2047 RTE_FLOW_ERROR_TYPE_ITEM, item,
2048 "match on metadata for ingress "
2049 "is not supported in legacy "
2053 mask = &rte_flow_item_meta_mask;
2055 return rte_flow_error_set(error, EINVAL,
2056 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
2057 "mask cannot be zero");
2059 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2060 (const uint8_t *)&nic_mask,
2061 sizeof(struct rte_flow_item_meta),
2062 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2067 * Validate TAG item.
2070 * Pointer to the rte_eth_dev structure.
2072 * Item specification.
2074 * Attributes of flow that includes this item.
2076 * Pointer to error structure.
2079 * 0 on success, a negative errno value otherwise and rte_errno is set.
2082 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
2083 const struct rte_flow_item *item,
2084 const struct rte_flow_attr *attr __rte_unused,
2085 struct rte_flow_error *error)
2087 const struct rte_flow_item_tag *spec = item->spec;
2088 const struct rte_flow_item_tag *mask = item->mask;
2089 const struct rte_flow_item_tag nic_mask = {
2090 .data = RTE_BE32(UINT32_MAX),
2095 if (!mlx5_flow_ext_mreg_supported(dev))
2096 return rte_flow_error_set(error, ENOTSUP,
2097 RTE_FLOW_ERROR_TYPE_ITEM, item,
2098 "extensive metadata register"
2099 " isn't supported");
2101 return rte_flow_error_set(error, EINVAL,
2102 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2104 "data cannot be empty");
2106 mask = &rte_flow_item_tag_mask;
2108 return rte_flow_error_set(error, EINVAL,
2109 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
2110 "mask cannot be zero");
2112 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2113 (const uint8_t *)&nic_mask,
2114 sizeof(struct rte_flow_item_tag),
2115 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2118 if (mask->index != 0xff)
2119 return rte_flow_error_set(error, EINVAL,
2120 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
2121 "partial mask for tag index"
2122 " is not supported");
2123 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
2126 MLX5_ASSERT(ret != REG_NON);
2131 * Validate vport item.
2134 * Pointer to the rte_eth_dev structure.
2136 * Item specification.
2138 * Attributes of flow that includes this item.
2139 * @param[in] item_flags
2140 * Bit-fields that holds the items detected until now.
2142 * Pointer to error structure.
2145 * 0 on success, a negative errno value otherwise and rte_errno is set.
2148 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
2149 const struct rte_flow_item *item,
2150 const struct rte_flow_attr *attr,
2151 uint64_t item_flags,
2152 struct rte_flow_error *error)
2154 const struct rte_flow_item_port_id *spec = item->spec;
2155 const struct rte_flow_item_port_id *mask = item->mask;
2156 const struct rte_flow_item_port_id switch_mask = {
2159 struct mlx5_priv *esw_priv;
2160 struct mlx5_priv *dev_priv;
2163 if (!attr->transfer)
2164 return rte_flow_error_set(error, EINVAL,
2165 RTE_FLOW_ERROR_TYPE_ITEM,
2167 "match on port id is valid only"
2168 " when transfer flag is enabled");
2169 if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
2170 return rte_flow_error_set(error, ENOTSUP,
2171 RTE_FLOW_ERROR_TYPE_ITEM, item,
2172 "multiple source ports are not"
2175 mask = &switch_mask;
2176 if (mask->id != 0xffffffff)
2177 return rte_flow_error_set(error, ENOTSUP,
2178 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2180 "no support for partial mask on"
2182 ret = mlx5_flow_item_acceptable
2183 (item, (const uint8_t *)mask,
2184 (const uint8_t *)&rte_flow_item_port_id_mask,
2185 sizeof(struct rte_flow_item_port_id),
2186 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2191 if (spec->id == MLX5_PORT_ESW_MGR)
2193 esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
2195 return rte_flow_error_set(error, rte_errno,
2196 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
2197 "failed to obtain E-Switch info for"
2199 dev_priv = mlx5_dev_to_eswitch_info(dev);
2201 return rte_flow_error_set(error, rte_errno,
2202 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2204 "failed to obtain E-Switch info");
2205 if (esw_priv->domain_id != dev_priv->domain_id)
2206 return rte_flow_error_set(error, EINVAL,
2207 RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
2208 "cannot match on a port from a"
2209 " different E-Switch");
2214 * Validate VLAN item.
2217 * Item specification.
2218 * @param[in] item_flags
2219 * Bit-fields that holds the items detected until now.
2221 * Ethernet device flow is being created on.
2223 * Pointer to error structure.
2226 * 0 on success, a negative errno value otherwise and rte_errno is set.
2229 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
2230 uint64_t item_flags,
2231 struct rte_eth_dev *dev,
2232 struct rte_flow_error *error)
2234 const struct rte_flow_item_vlan *mask = item->mask;
2235 const struct rte_flow_item_vlan nic_mask = {
2236 .tci = RTE_BE16(UINT16_MAX),
2237 .inner_type = RTE_BE16(UINT16_MAX),
2240 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2242 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2243 MLX5_FLOW_LAYER_INNER_L4) :
2244 (MLX5_FLOW_LAYER_OUTER_L3 |
2245 MLX5_FLOW_LAYER_OUTER_L4);
2246 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2247 MLX5_FLOW_LAYER_OUTER_VLAN;
2249 if (item_flags & vlanm)
2250 return rte_flow_error_set(error, EINVAL,
2251 RTE_FLOW_ERROR_TYPE_ITEM, item,
2252 "multiple VLAN layers not supported");
2253 else if ((item_flags & l34m) != 0)
2254 return rte_flow_error_set(error, EINVAL,
2255 RTE_FLOW_ERROR_TYPE_ITEM, item,
2256 "VLAN cannot follow L3/L4 layer");
2258 mask = &rte_flow_item_vlan_mask;
2259 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2260 (const uint8_t *)&nic_mask,
2261 sizeof(struct rte_flow_item_vlan),
2262 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2265 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2266 struct mlx5_priv *priv = dev->data->dev_private;
2268 if (priv->vmwa_context) {
2270 * Non-NULL context means we have a virtual machine
2271 * and SR-IOV enabled, we have to create VLAN interface
2272 * to make hypervisor to setup E-Switch vport
2273 * context correctly. We avoid creating the multiple
2274 * VLAN interfaces, so we cannot support VLAN tag mask.
2276 return rte_flow_error_set(error, EINVAL,
2277 RTE_FLOW_ERROR_TYPE_ITEM,
2279 "VLAN tag mask is not"
2280 " supported in virtual"
2288 * GTP flags are contained in 1 byte of the format:
2289 * -------------------------------------------
2290 * | bit | 0 - 2 | 3 | 4 | 5 | 6 | 7 |
2291 * |-----------------------------------------|
2292 * | value | Version | PT | Res | E | S | PN |
2293 * -------------------------------------------
2295 * Matching is supported only for GTP flags E, S, PN.
2297 #define MLX5_GTP_FLAGS_MASK 0x07
2300 * Validate GTP item.
2303 * Pointer to the rte_eth_dev structure.
2305 * Item specification.
2306 * @param[in] item_flags
2307 * Bit-fields that holds the items detected until now.
2309 * Pointer to error structure.
2312 * 0 on success, a negative errno value otherwise and rte_errno is set.
2315 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
2316 const struct rte_flow_item *item,
2317 uint64_t item_flags,
2318 struct rte_flow_error *error)
2320 struct mlx5_priv *priv = dev->data->dev_private;
2321 const struct rte_flow_item_gtp *spec = item->spec;
2322 const struct rte_flow_item_gtp *mask = item->mask;
2323 const struct rte_flow_item_gtp nic_mask = {
2324 .v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
2326 .teid = RTE_BE32(0xffffffff),
2329 if (!priv->sh->cdev->config.hca_attr.tunnel_stateless_gtp)
2330 return rte_flow_error_set(error, ENOTSUP,
2331 RTE_FLOW_ERROR_TYPE_ITEM, item,
2332 "GTP support is not enabled");
2333 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2334 return rte_flow_error_set(error, ENOTSUP,
2335 RTE_FLOW_ERROR_TYPE_ITEM, item,
2336 "multiple tunnel layers not"
2338 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2339 return rte_flow_error_set(error, EINVAL,
2340 RTE_FLOW_ERROR_TYPE_ITEM, item,
2341 "no outer UDP layer found");
2343 mask = &rte_flow_item_gtp_mask;
2344 if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
2345 return rte_flow_error_set(error, ENOTSUP,
2346 RTE_FLOW_ERROR_TYPE_ITEM, item,
2347 "Match is supported for GTP"
2349 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2350 (const uint8_t *)&nic_mask,
2351 sizeof(struct rte_flow_item_gtp),
2352 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2356 * Validate GTP PSC item.
2359 * Item specification.
2360 * @param[in] last_item
2361 * Previous validated item in the pattern items.
2362 * @param[in] gtp_item
2363 * Previous GTP item specification.
2365 * Pointer to flow attributes.
2367 * Pointer to error structure.
2370 * 0 on success, a negative errno value otherwise and rte_errno is set.
2373 flow_dv_validate_item_gtp_psc(const struct rte_flow_item *item,
2375 const struct rte_flow_item *gtp_item,
2376 const struct rte_flow_attr *attr,
2377 struct rte_flow_error *error)
2379 const struct rte_flow_item_gtp *gtp_spec;
2380 const struct rte_flow_item_gtp *gtp_mask;
2381 const struct rte_flow_item_gtp_psc *mask;
2382 const struct rte_flow_item_gtp_psc nic_mask = {
2387 if (!gtp_item || !(last_item & MLX5_FLOW_LAYER_GTP))
2388 return rte_flow_error_set
2389 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2390 "GTP PSC item must be preceded with GTP item");
2391 gtp_spec = gtp_item->spec;
2392 gtp_mask = gtp_item->mask ? gtp_item->mask : &rte_flow_item_gtp_mask;
2393 /* GTP spec and E flag is requested to match zero. */
2395 (gtp_mask->v_pt_rsv_flags &
2396 ~gtp_spec->v_pt_rsv_flags & MLX5_GTP_EXT_HEADER_FLAG))
2397 return rte_flow_error_set
2398 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2399 "GTP E flag must be 1 to match GTP PSC");
2400 /* Check the flow is not created in group zero. */
2401 if (!attr->transfer && !attr->group)
2402 return rte_flow_error_set
2403 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2404 "GTP PSC is not supported for group 0");
2405 /* GTP spec is here and E flag is requested to match zero. */
2408 mask = item->mask ? item->mask : &rte_flow_item_gtp_psc_mask;
2409 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2410 (const uint8_t *)&nic_mask,
2411 sizeof(struct rte_flow_item_gtp_psc),
2412 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2416 * Validate IPV4 item.
2417 * Use existing validation function mlx5_flow_validate_item_ipv4(), and
2418 * add specific validation of fragment_offset field,
2421 * Item specification.
2422 * @param[in] item_flags
2423 * Bit-fields that holds the items detected until now.
2425 * Pointer to error structure.
2428 * 0 on success, a negative errno value otherwise and rte_errno is set.
2431 flow_dv_validate_item_ipv4(struct rte_eth_dev *dev,
2432 const struct rte_flow_item *item,
2433 uint64_t item_flags, uint64_t last_item,
2434 uint16_t ether_type, struct rte_flow_error *error)
2437 struct mlx5_priv *priv = dev->data->dev_private;
2438 struct mlx5_hca_attr *attr = &priv->sh->cdev->config.hca_attr;
2439 const struct rte_flow_item_ipv4 *spec = item->spec;
2440 const struct rte_flow_item_ipv4 *last = item->last;
2441 const struct rte_flow_item_ipv4 *mask = item->mask;
2442 rte_be16_t fragment_offset_spec = 0;
2443 rte_be16_t fragment_offset_last = 0;
2444 struct rte_flow_item_ipv4 nic_ipv4_mask = {
2446 .src_addr = RTE_BE32(0xffffffff),
2447 .dst_addr = RTE_BE32(0xffffffff),
2448 .type_of_service = 0xff,
2449 .fragment_offset = RTE_BE16(0xffff),
2450 .next_proto_id = 0xff,
2451 .time_to_live = 0xff,
2455 if (mask && (mask->hdr.version_ihl & RTE_IPV4_HDR_IHL_MASK)) {
2456 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2457 bool ihl_cap = !tunnel ?
2458 attr->outer_ipv4_ihl : attr->inner_ipv4_ihl;
2460 return rte_flow_error_set(error, ENOTSUP,
2461 RTE_FLOW_ERROR_TYPE_ITEM,
2463 "IPV4 ihl offload not supported");
2464 nic_ipv4_mask.hdr.version_ihl = mask->hdr.version_ihl;
2466 ret = mlx5_flow_validate_item_ipv4(item, item_flags, last_item,
2467 ether_type, &nic_ipv4_mask,
2468 MLX5_ITEM_RANGE_ACCEPTED, error);
2472 fragment_offset_spec = spec->hdr.fragment_offset &
2473 mask->hdr.fragment_offset;
2474 if (!fragment_offset_spec)
2477 * spec and mask are valid, enforce using full mask to make sure the
2478 * complete value is used correctly.
2480 if ((mask->hdr.fragment_offset & RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
2481 != RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
2482 return rte_flow_error_set(error, EINVAL,
2483 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2484 item, "must use full mask for"
2485 " fragment_offset");
2487 * Match on fragment_offset 0x2000 means MF is 1 and frag-offset is 0,
2488 * indicating this is 1st fragment of fragmented packet.
2489 * This is not yet supported in MLX5, return appropriate error message.
2491 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG))
2492 return rte_flow_error_set(error, ENOTSUP,
2493 RTE_FLOW_ERROR_TYPE_ITEM, item,
2494 "match on first fragment not "
2496 if (fragment_offset_spec && !last)
2497 return rte_flow_error_set(error, ENOTSUP,
2498 RTE_FLOW_ERROR_TYPE_ITEM, item,
2499 "specified value not supported");
2500 /* spec and last are valid, validate the specified range. */
2501 fragment_offset_last = last->hdr.fragment_offset &
2502 mask->hdr.fragment_offset;
2504 * Match on fragment_offset spec 0x2001 and last 0x3fff
2505 * means MF is 1 and frag-offset is > 0.
2506 * This packet is fragment 2nd and onward, excluding last.
2507 * This is not yet supported in MLX5, return appropriate
2510 if (fragment_offset_spec == RTE_BE16(RTE_IPV4_HDR_MF_FLAG + 1) &&
2511 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK))
2512 return rte_flow_error_set(error, ENOTSUP,
2513 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2514 last, "match on following "
2515 "fragments not supported");
2517 * Match on fragment_offset spec 0x0001 and last 0x1fff
2518 * means MF is 0 and frag-offset is > 0.
2519 * This packet is last fragment of fragmented packet.
2520 * This is not yet supported in MLX5, return appropriate
2523 if (fragment_offset_spec == RTE_BE16(1) &&
2524 fragment_offset_last == RTE_BE16(RTE_IPV4_HDR_OFFSET_MASK))
2525 return rte_flow_error_set(error, ENOTSUP,
2526 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2527 last, "match on last "
2528 "fragment not supported");
2530 * Match on fragment_offset spec 0x0001 and last 0x3fff
2531 * means MF and/or frag-offset is not 0.
2532 * This is a fragmented packet.
2533 * Other range values are invalid and rejected.
2535 if (!(fragment_offset_spec == RTE_BE16(1) &&
2536 fragment_offset_last == RTE_BE16(MLX5_IPV4_FRAG_OFFSET_MASK)))
2537 return rte_flow_error_set(error, ENOTSUP,
2538 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
2539 "specified range not supported");
2544 * Validate IPV6 fragment extension item.
2547 * Item specification.
2548 * @param[in] item_flags
2549 * Bit-fields that holds the items detected until now.
2551 * Pointer to error structure.
2554 * 0 on success, a negative errno value otherwise and rte_errno is set.
2557 flow_dv_validate_item_ipv6_frag_ext(const struct rte_flow_item *item,
2558 uint64_t item_flags,
2559 struct rte_flow_error *error)
2561 const struct rte_flow_item_ipv6_frag_ext *spec = item->spec;
2562 const struct rte_flow_item_ipv6_frag_ext *last = item->last;
2563 const struct rte_flow_item_ipv6_frag_ext *mask = item->mask;
2564 rte_be16_t frag_data_spec = 0;
2565 rte_be16_t frag_data_last = 0;
2566 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2567 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2568 MLX5_FLOW_LAYER_OUTER_L4;
2570 struct rte_flow_item_ipv6_frag_ext nic_mask = {
2572 .next_header = 0xff,
2573 .frag_data = RTE_BE16(0xffff),
2577 if (item_flags & l4m)
2578 return rte_flow_error_set(error, EINVAL,
2579 RTE_FLOW_ERROR_TYPE_ITEM, item,
2580 "ipv6 fragment extension item cannot "
2582 if ((tunnel && !(item_flags & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
2583 (!tunnel && !(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
2584 return rte_flow_error_set(error, EINVAL,
2585 RTE_FLOW_ERROR_TYPE_ITEM, item,
2586 "ipv6 fragment extension item must "
2587 "follow ipv6 item");
2589 frag_data_spec = spec->hdr.frag_data & mask->hdr.frag_data;
2590 if (!frag_data_spec)
2593 * spec and mask are valid, enforce using full mask to make sure the
2594 * complete value is used correctly.
2596 if ((mask->hdr.frag_data & RTE_BE16(RTE_IPV6_FRAG_USED_MASK)) !=
2597 RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
2598 return rte_flow_error_set(error, EINVAL,
2599 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2600 item, "must use full mask for"
2603 * Match on frag_data 0x00001 means M is 1 and frag-offset is 0.
2604 * This is 1st fragment of fragmented packet.
2606 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_MF_MASK))
2607 return rte_flow_error_set(error, ENOTSUP,
2608 RTE_FLOW_ERROR_TYPE_ITEM, item,
2609 "match on first fragment not "
2611 if (frag_data_spec && !last)
2612 return rte_flow_error_set(error, EINVAL,
2613 RTE_FLOW_ERROR_TYPE_ITEM, item,
2614 "specified value not supported");
2615 ret = mlx5_flow_item_acceptable
2616 (item, (const uint8_t *)mask,
2617 (const uint8_t *)&nic_mask,
2618 sizeof(struct rte_flow_item_ipv6_frag_ext),
2619 MLX5_ITEM_RANGE_ACCEPTED, error);
2622 /* spec and last are valid, validate the specified range. */
2623 frag_data_last = last->hdr.frag_data & mask->hdr.frag_data;
2625 * Match on frag_data spec 0x0009 and last 0xfff9
2626 * means M is 1 and frag-offset is > 0.
2627 * This packet is fragment 2nd and onward, excluding last.
2628 * This is not yet supported in MLX5, return appropriate
2631 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN |
2632 RTE_IPV6_EHDR_MF_MASK) &&
2633 frag_data_last == RTE_BE16(RTE_IPV6_FRAG_USED_MASK))
2634 return rte_flow_error_set(error, ENOTSUP,
2635 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2636 last, "match on following "
2637 "fragments not supported");
2639 * Match on frag_data spec 0x0008 and last 0xfff8
2640 * means M is 0 and frag-offset is > 0.
2641 * This packet is last fragment of fragmented packet.
2642 * This is not yet supported in MLX5, return appropriate
2645 if (frag_data_spec == RTE_BE16(RTE_IPV6_EHDR_FO_ALIGN) &&
2646 frag_data_last == RTE_BE16(RTE_IPV6_EHDR_FO_MASK))
2647 return rte_flow_error_set(error, ENOTSUP,
2648 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
2649 last, "match on last "
2650 "fragment not supported");
2651 /* Other range values are invalid and rejected. */
2652 return rte_flow_error_set(error, EINVAL,
2653 RTE_FLOW_ERROR_TYPE_ITEM_LAST, last,
2654 "specified range not supported");
2658 * Validate ASO CT item.
2661 * Pointer to the rte_eth_dev structure.
2663 * Item specification.
2664 * @param[in] item_flags
2665 * Pointer to bit-fields that holds the items detected until now.
2667 * Pointer to error structure.
2670 * 0 on success, a negative errno value otherwise and rte_errno is set.
2673 flow_dv_validate_item_aso_ct(struct rte_eth_dev *dev,
2674 const struct rte_flow_item *item,
2675 uint64_t *item_flags,
2676 struct rte_flow_error *error)
2678 const struct rte_flow_item_conntrack *spec = item->spec;
2679 const struct rte_flow_item_conntrack *mask = item->mask;
2683 if (*item_flags & MLX5_FLOW_LAYER_ASO_CT)
2684 return rte_flow_error_set(error, EINVAL,
2685 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
2686 "Only one CT is supported");
2688 mask = &rte_flow_item_conntrack_mask;
2689 flags = spec->flags & mask->flags;
2690 if ((flags & RTE_FLOW_CONNTRACK_PKT_STATE_VALID) &&
2691 ((flags & RTE_FLOW_CONNTRACK_PKT_STATE_INVALID) ||
2692 (flags & RTE_FLOW_CONNTRACK_PKT_STATE_BAD) ||
2693 (flags & RTE_FLOW_CONNTRACK_PKT_STATE_DISABLED)))
2694 return rte_flow_error_set(error, EINVAL,
2695 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
2696 "Conflict status bits");
2697 /* State change also needs to be considered. */
2698 *item_flags |= MLX5_FLOW_LAYER_ASO_CT;
2703 * Validate the pop VLAN action.
2706 * Pointer to the rte_eth_dev structure.
2707 * @param[in] action_flags
2708 * Holds the actions detected until now.
2710 * Pointer to the pop vlan action.
2711 * @param[in] item_flags
2712 * The items found in this flow rule.
2714 * Pointer to flow attributes.
2716 * Pointer to error structure.
2719 * 0 on success, a negative errno value otherwise and rte_errno is set.
2722 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
2723 uint64_t action_flags,
2724 const struct rte_flow_action *action,
2725 uint64_t item_flags,
2726 const struct rte_flow_attr *attr,
2727 struct rte_flow_error *error)
2729 const struct mlx5_priv *priv = dev->data->dev_private;
2730 struct mlx5_dev_ctx_shared *sh = priv->sh;
2731 bool direction_error = false;
2733 if (!priv->sh->pop_vlan_action)
2734 return rte_flow_error_set(error, ENOTSUP,
2735 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2737 "pop vlan action is not supported");
2738 /* Pop VLAN is not supported in egress except for CX6 FDB mode. */
2739 if (attr->transfer) {
2740 bool fdb_tx = priv->representor_id != UINT16_MAX;
2741 bool is_cx5 = sh->steering_format_version ==
2742 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5;
2744 if (fdb_tx && is_cx5)
2745 direction_error = true;
2746 } else if (attr->egress) {
2747 direction_error = true;
2749 if (direction_error)
2750 return rte_flow_error_set(error, ENOTSUP,
2751 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2753 "pop vlan action not supported for egress");
2754 if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
2755 return rte_flow_error_set(error, ENOTSUP,
2756 RTE_FLOW_ERROR_TYPE_ACTION, action,
2757 "no support for multiple VLAN "
2759 /* Pop VLAN with preceding Decap requires inner header with VLAN. */
2760 if ((action_flags & MLX5_FLOW_ACTION_DECAP) &&
2761 !(item_flags & MLX5_FLOW_LAYER_INNER_VLAN))
2762 return rte_flow_error_set(error, ENOTSUP,
2763 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2765 "cannot pop vlan after decap without "
2766 "match on inner vlan in the flow");
2767 /* Pop VLAN without preceding Decap requires outer header with VLAN. */
2768 if (!(action_flags & MLX5_FLOW_ACTION_DECAP) &&
2769 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2770 return rte_flow_error_set(error, ENOTSUP,
2771 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2773 "cannot pop vlan without a "
2774 "match on (outer) vlan in the flow");
2775 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2776 return rte_flow_error_set(error, EINVAL,
2777 RTE_FLOW_ERROR_TYPE_ACTION, action,
2778 "wrong action order, port_id should "
2779 "be after pop VLAN action");
2780 if (!attr->transfer && priv->representor)
2781 return rte_flow_error_set(error, ENOTSUP,
2782 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2783 "pop vlan action for VF representor "
2784 "not supported on NIC table");
2789 * Get VLAN default info from vlan match info.
2792 * the list of item specifications.
2794 * pointer VLAN info to fill to.
2797 * 0 on success, a negative errno value otherwise and rte_errno is set.
2800 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
2801 struct rte_vlan_hdr *vlan)
2803 const struct rte_flow_item_vlan nic_mask = {
2804 .tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
2805 MLX5DV_FLOW_VLAN_VID_MASK),
2806 .inner_type = RTE_BE16(0xffff),
2811 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2812 int type = items->type;
2814 if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
2815 type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
2818 if (items->type != RTE_FLOW_ITEM_TYPE_END) {
2819 const struct rte_flow_item_vlan *vlan_m = items->mask;
2820 const struct rte_flow_item_vlan *vlan_v = items->spec;
2822 /* If VLAN item in pattern doesn't contain data, return here. */
2827 /* Only full match values are accepted */
2828 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
2829 MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
2830 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
2832 rte_be_to_cpu_16(vlan_v->tci &
2833 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
2835 if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
2836 MLX5DV_FLOW_VLAN_VID_MASK_BE) {
2837 vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
2839 rte_be_to_cpu_16(vlan_v->tci &
2840 MLX5DV_FLOW_VLAN_VID_MASK_BE);
2842 if (vlan_m->inner_type == nic_mask.inner_type)
2843 vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
2844 vlan_m->inner_type);
2849 * Validate the push VLAN action.
2852 * Pointer to the rte_eth_dev structure.
2853 * @param[in] action_flags
2854 * Holds the actions detected until now.
2855 * @param[in] item_flags
2856 * The items found in this flow rule.
2858 * Pointer to the action structure.
2860 * Pointer to flow attributes
2862 * Pointer to error structure.
2865 * 0 on success, a negative errno value otherwise and rte_errno is set.
2868 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
2869 uint64_t action_flags,
2870 const struct rte_flow_item_vlan *vlan_m,
2871 const struct rte_flow_action *action,
2872 const struct rte_flow_attr *attr,
2873 struct rte_flow_error *error)
2875 const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
2876 const struct mlx5_priv *priv = dev->data->dev_private;
2878 if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
2879 push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
2880 return rte_flow_error_set(error, EINVAL,
2881 RTE_FLOW_ERROR_TYPE_ACTION, action,
2882 "invalid vlan ethertype");
2883 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2884 return rte_flow_error_set(error, EINVAL,
2885 RTE_FLOW_ERROR_TYPE_ACTION, action,
2886 "wrong action order, port_id should "
2887 "be after push VLAN");
2888 if (!attr->transfer && priv->representor)
2889 return rte_flow_error_set(error, ENOTSUP,
2890 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2891 "push vlan action for VF representor "
2892 "not supported on NIC table");
2894 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
2895 (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
2896 MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
2897 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
2898 !(mlx5_flow_find_action
2899 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
2900 return rte_flow_error_set(error, EINVAL,
2901 RTE_FLOW_ERROR_TYPE_ACTION, action,
2902 "not full match mask on VLAN PCP and "
2903 "there is no of_set_vlan_pcp action, "
2904 "push VLAN action cannot figure out "
2907 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
2908 (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
2909 MLX5DV_FLOW_VLAN_VID_MASK_BE &&
2910 !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
2911 !(mlx5_flow_find_action
2912 (action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
2913 return rte_flow_error_set(error, EINVAL,
2914 RTE_FLOW_ERROR_TYPE_ACTION, action,
2915 "not full match mask on VLAN VID and "
2916 "there is no of_set_vlan_vid action, "
2917 "push VLAN action cannot figure out "
2924 * Validate the set VLAN PCP.
2926 * @param[in] action_flags
2927 * Holds the actions detected until now.
2928 * @param[in] actions
2929 * Pointer to the list of actions remaining in the flow rule.
2931 * Pointer to error structure.
2934 * 0 on success, a negative errno value otherwise and rte_errno is set.
2937 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
2938 const struct rte_flow_action actions[],
2939 struct rte_flow_error *error)
2941 const struct rte_flow_action *action = actions;
2942 const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2944 if (conf->vlan_pcp > 7)
2945 return rte_flow_error_set(error, EINVAL,
2946 RTE_FLOW_ERROR_TYPE_ACTION, action,
2947 "VLAN PCP value is too big");
2948 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2949 return rte_flow_error_set(error, ENOTSUP,
2950 RTE_FLOW_ERROR_TYPE_ACTION, action,
2951 "set VLAN PCP action must follow "
2952 "the push VLAN action");
2953 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2954 return rte_flow_error_set(error, ENOTSUP,
2955 RTE_FLOW_ERROR_TYPE_ACTION, action,
2956 "Multiple VLAN PCP modification are "
2958 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2959 return rte_flow_error_set(error, EINVAL,
2960 RTE_FLOW_ERROR_TYPE_ACTION, action,
2961 "wrong action order, port_id should "
2962 "be after set VLAN PCP");
2967 * Validate the set VLAN VID.
2969 * @param[in] item_flags
2970 * Holds the items detected in this rule.
2971 * @param[in] action_flags
2972 * Holds the actions detected until now.
2973 * @param[in] actions
2974 * Pointer to the list of actions remaining in the flow rule.
2976 * Pointer to error structure.
2979 * 0 on success, a negative errno value otherwise and rte_errno is set.
2982 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
2983 uint64_t action_flags,
2984 const struct rte_flow_action actions[],
2985 struct rte_flow_error *error)
2987 const struct rte_flow_action *action = actions;
2988 const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
2990 if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
2991 return rte_flow_error_set(error, EINVAL,
2992 RTE_FLOW_ERROR_TYPE_ACTION, action,
2993 "VLAN VID value is too big");
2994 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2995 !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2996 return rte_flow_error_set(error, ENOTSUP,
2997 RTE_FLOW_ERROR_TYPE_ACTION, action,
2998 "set VLAN VID action must follow push"
2999 " VLAN action or match on VLAN item");
3000 if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
3001 return rte_flow_error_set(error, ENOTSUP,
3002 RTE_FLOW_ERROR_TYPE_ACTION, action,
3003 "Multiple VLAN VID modifications are "
3005 if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
3006 return rte_flow_error_set(error, EINVAL,
3007 RTE_FLOW_ERROR_TYPE_ACTION, action,
3008 "wrong action order, port_id should "
3009 "be after set VLAN VID");
3014 * Validate the FLAG action.
3017 * Pointer to the rte_eth_dev structure.
3018 * @param[in] action_flags
3019 * Holds the actions detected until now.
3021 * Pointer to flow attributes
3023 * Pointer to error structure.
3026 * 0 on success, a negative errno value otherwise and rte_errno is set.
3029 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
3030 uint64_t action_flags,
3031 const struct rte_flow_attr *attr,
3032 struct rte_flow_error *error)
3034 struct mlx5_priv *priv = dev->data->dev_private;
3035 struct mlx5_sh_config *config = &priv->sh->config;
3038 /* Fall back if no extended metadata register support. */
3039 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
3040 return mlx5_flow_validate_action_flag(action_flags, attr,
3042 /* Extensive metadata mode requires registers. */
3043 if (!mlx5_flow_ext_mreg_supported(dev))
3044 return rte_flow_error_set(error, ENOTSUP,
3045 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3046 "no metadata registers "
3047 "to support flag action");
3048 if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
3049 return rte_flow_error_set(error, ENOTSUP,
3050 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3051 "extended metadata register"
3052 " isn't available");
3053 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3056 MLX5_ASSERT(ret > 0);
3057 if (action_flags & MLX5_FLOW_ACTION_MARK)
3058 return rte_flow_error_set(error, EINVAL,
3059 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3060 "can't mark and flag in same flow");
3061 if (action_flags & MLX5_FLOW_ACTION_FLAG)
3062 return rte_flow_error_set(error, EINVAL,
3063 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3065 " actions in same flow");
3070 * Validate MARK action.
3073 * Pointer to the rte_eth_dev structure.
3075 * Pointer to action.
3076 * @param[in] action_flags
3077 * Holds the actions detected until now.
3079 * Pointer to flow attributes
3081 * Pointer to error structure.
3084 * 0 on success, a negative errno value otherwise and rte_errno is set.
3087 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
3088 const struct rte_flow_action *action,
3089 uint64_t action_flags,
3090 const struct rte_flow_attr *attr,
3091 struct rte_flow_error *error)
3093 struct mlx5_priv *priv = dev->data->dev_private;
3094 struct mlx5_sh_config *config = &priv->sh->config;
3095 const struct rte_flow_action_mark *mark = action->conf;
3098 if (is_tunnel_offload_active(dev))
3099 return rte_flow_error_set(error, ENOTSUP,
3100 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3102 "if tunnel offload active");
3103 /* Fall back if no extended metadata register support. */
3104 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
3105 return mlx5_flow_validate_action_mark(action, action_flags,
3107 /* Extensive metadata mode requires registers. */
3108 if (!mlx5_flow_ext_mreg_supported(dev))
3109 return rte_flow_error_set(error, ENOTSUP,
3110 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3111 "no metadata registers "
3112 "to support mark action");
3113 if (!priv->sh->dv_mark_mask)
3114 return rte_flow_error_set(error, ENOTSUP,
3115 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3116 "extended metadata register"
3117 " isn't available");
3118 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3121 MLX5_ASSERT(ret > 0);
3123 return rte_flow_error_set(error, EINVAL,
3124 RTE_FLOW_ERROR_TYPE_ACTION, action,
3125 "configuration cannot be null");
3126 if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
3127 return rte_flow_error_set(error, EINVAL,
3128 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3130 "mark id exceeds the limit");
3131 if (action_flags & MLX5_FLOW_ACTION_FLAG)
3132 return rte_flow_error_set(error, EINVAL,
3133 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3134 "can't flag and mark in same flow");
3135 if (action_flags & MLX5_FLOW_ACTION_MARK)
3136 return rte_flow_error_set(error, EINVAL,
3137 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3138 "can't have 2 mark actions in same"
3144 * Validate SET_META action.
3147 * Pointer to the rte_eth_dev structure.
3149 * Pointer to the action structure.
3150 * @param[in] action_flags
3151 * Holds the actions detected until now.
3153 * Pointer to flow attributes
3155 * Pointer to error structure.
3158 * 0 on success, a negative errno value otherwise and rte_errno is set.
3161 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
3162 const struct rte_flow_action *action,
3163 uint64_t action_flags __rte_unused,
3164 const struct rte_flow_attr *attr,
3165 struct rte_flow_error *error)
3167 struct mlx5_priv *priv = dev->data->dev_private;
3168 struct mlx5_sh_config *config = &priv->sh->config;
3169 const struct rte_flow_action_set_meta *conf;
3170 uint32_t nic_mask = UINT32_MAX;
3173 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
3174 !mlx5_flow_ext_mreg_supported(dev))
3175 return rte_flow_error_set(error, ENOTSUP,
3176 RTE_FLOW_ERROR_TYPE_ACTION, action,
3177 "extended metadata register"
3178 " isn't supported");
3179 reg = flow_dv_get_metadata_reg(dev, attr, error);
3183 return rte_flow_error_set(error, ENOTSUP,
3184 RTE_FLOW_ERROR_TYPE_ACTION, action,
3185 "unavailable extended metadata register");
3186 if (reg != REG_A && reg != REG_B) {
3187 struct mlx5_priv *priv = dev->data->dev_private;
3189 nic_mask = priv->sh->dv_meta_mask;
3191 if (!(action->conf))
3192 return rte_flow_error_set(error, EINVAL,
3193 RTE_FLOW_ERROR_TYPE_ACTION, action,
3194 "configuration cannot be null");
3195 conf = (const struct rte_flow_action_set_meta *)action->conf;
3197 return rte_flow_error_set(error, EINVAL,
3198 RTE_FLOW_ERROR_TYPE_ACTION, action,
3199 "zero mask doesn't have any effect");
3200 if (conf->mask & ~nic_mask)
3201 return rte_flow_error_set(error, EINVAL,
3202 RTE_FLOW_ERROR_TYPE_ACTION, action,
3203 "meta data must be within reg C0");
3208 * Validate SET_TAG action.
3211 * Pointer to the rte_eth_dev structure.
3213 * Pointer to the action structure.
3214 * @param[in] action_flags
3215 * Holds the actions detected until now.
3217 * Pointer to flow attributes
3219 * Pointer to error structure.
3222 * 0 on success, a negative errno value otherwise and rte_errno is set.
3225 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
3226 const struct rte_flow_action *action,
3227 uint64_t action_flags,
3228 const struct rte_flow_attr *attr,
3229 struct rte_flow_error *error)
3231 const struct rte_flow_action_set_tag *conf;
3232 const uint64_t terminal_action_flags =
3233 MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
3234 MLX5_FLOW_ACTION_RSS;
3237 if (!mlx5_flow_ext_mreg_supported(dev))
3238 return rte_flow_error_set(error, ENOTSUP,
3239 RTE_FLOW_ERROR_TYPE_ACTION, action,
3240 "extensive metadata register"
3241 " isn't supported");
3242 if (!(action->conf))
3243 return rte_flow_error_set(error, EINVAL,
3244 RTE_FLOW_ERROR_TYPE_ACTION, action,
3245 "configuration cannot be null");
3246 conf = (const struct rte_flow_action_set_tag *)action->conf;
3248 return rte_flow_error_set(error, EINVAL,
3249 RTE_FLOW_ERROR_TYPE_ACTION, action,
3250 "zero mask doesn't have any effect");
3251 ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
3254 if (!attr->transfer && attr->ingress &&
3255 (action_flags & terminal_action_flags))
3256 return rte_flow_error_set(error, EINVAL,
3257 RTE_FLOW_ERROR_TYPE_ACTION, action,
3258 "set_tag has no effect"
3259 " with terminal actions");
3264 * Indicates whether ASO aging is supported.
3267 * Pointer to shared device context structure.
3269 * Attributes of flow that includes AGE action.
3272 * True when ASO aging is supported, false otherwise.
3275 flow_hit_aso_supported(const struct mlx5_dev_ctx_shared *sh,
3276 const struct rte_flow_attr *attr)
3278 MLX5_ASSERT(sh && attr);
3279 return (sh->flow_hit_aso_en && (attr->transfer || attr->group));
3283 * Validate count action.
3286 * Pointer to rte_eth_dev structure.
3288 * Indicator if action is shared.
3289 * @param[in] action_flags
3290 * Holds the actions detected until now.
3292 * Attributes of flow that includes this action.
3294 * Pointer to error structure.
3297 * 0 on success, a negative errno value otherwise and rte_errno is set.
3300 flow_dv_validate_action_count(struct rte_eth_dev *dev, bool shared,
3301 uint64_t action_flags,
3302 const struct rte_flow_attr *attr,
3303 struct rte_flow_error *error)
3305 struct mlx5_priv *priv = dev->data->dev_private;
3307 if (!priv->sh->cdev->config.devx)
3309 if (action_flags & MLX5_FLOW_ACTION_COUNT)
3310 return rte_flow_error_set(error, EINVAL,
3311 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3312 "duplicate count actions set");
3313 if (shared && (action_flags & MLX5_FLOW_ACTION_AGE) &&
3314 !flow_hit_aso_supported(priv->sh, attr))
3315 return rte_flow_error_set(error, EINVAL,
3316 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3317 "old age and indirect count combination is not supported");
3318 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
3322 return rte_flow_error_set
3324 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3326 "count action not supported");
3330 * Validate the L2 encap action.
3333 * Pointer to the rte_eth_dev structure.
3334 * @param[in] action_flags
3335 * Holds the actions detected until now.
3337 * Pointer to the action structure.
3339 * Pointer to flow attributes.
3341 * Pointer to error structure.
3344 * 0 on success, a negative errno value otherwise and rte_errno is set.
3347 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
3348 uint64_t action_flags,
3349 const struct rte_flow_action *action,
3350 const struct rte_flow_attr *attr,
3351 struct rte_flow_error *error)
3353 const struct mlx5_priv *priv = dev->data->dev_private;
3355 if (!(action->conf))
3356 return rte_flow_error_set(error, EINVAL,
3357 RTE_FLOW_ERROR_TYPE_ACTION, action,
3358 "configuration cannot be null");
3359 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3360 return rte_flow_error_set(error, EINVAL,
3361 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3362 "can only have a single encap action "
3364 if (!attr->transfer && priv->representor)
3365 return rte_flow_error_set(error, ENOTSUP,
3366 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3367 "encap action for VF representor "
3368 "not supported on NIC table");
3373 * Validate a decap action.
3376 * Pointer to the rte_eth_dev structure.
3377 * @param[in] action_flags
3378 * Holds the actions detected until now.
3380 * Pointer to the action structure.
3381 * @param[in] item_flags
3382 * Holds the items detected.
3384 * Pointer to flow attributes
3386 * Pointer to error structure.
3389 * 0 on success, a negative errno value otherwise and rte_errno is set.
3392 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
3393 uint64_t action_flags,
3394 const struct rte_flow_action *action,
3395 const uint64_t item_flags,
3396 const struct rte_flow_attr *attr,
3397 struct rte_flow_error *error)
3399 const struct mlx5_priv *priv = dev->data->dev_private;
3401 if (priv->sh->cdev->config.hca_attr.scatter_fcs_w_decap_disable &&
3402 !priv->sh->config.decap_en)
3403 return rte_flow_error_set(error, ENOTSUP,
3404 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3405 "decap is not enabled");
3406 if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
3407 return rte_flow_error_set(error, ENOTSUP,
3408 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3410 MLX5_FLOW_ACTION_DECAP ? "can only "
3411 "have a single decap action" : "decap "
3412 "after encap is not supported");
3413 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
3414 return rte_flow_error_set(error, EINVAL,
3415 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3416 "can't have decap action after"
3419 return rte_flow_error_set(error, ENOTSUP,
3420 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
3422 "decap action not supported for "
3424 if (!attr->transfer && priv->representor)
3425 return rte_flow_error_set(error, ENOTSUP,
3426 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3427 "decap action for VF representor "
3428 "not supported on NIC table");
3429 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_DECAP &&
3430 !(item_flags & MLX5_FLOW_LAYER_VXLAN))
3431 return rte_flow_error_set(error, ENOTSUP,
3432 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3433 "VXLAN item should be present for VXLAN decap");
3437 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
3440 * Validate the raw encap and decap actions.
3443 * Pointer to the rte_eth_dev structure.
3445 * Pointer to the decap action.
3447 * Pointer to the encap action.
3449 * Pointer to flow attributes
3450 * @param[in/out] action_flags
3451 * Holds the actions detected until now.
3452 * @param[out] actions_n
3453 * pointer to the number of actions counter.
3455 * Pointer to the action structure.
3456 * @param[in] item_flags
3457 * Holds the items detected.
3459 * Pointer to error structure.
3462 * 0 on success, a negative errno value otherwise and rte_errno is set.
3465 flow_dv_validate_action_raw_encap_decap
3466 (struct rte_eth_dev *dev,
3467 const struct rte_flow_action_raw_decap *decap,
3468 const struct rte_flow_action_raw_encap *encap,
3469 const struct rte_flow_attr *attr, uint64_t *action_flags,
3470 int *actions_n, const struct rte_flow_action *action,
3471 uint64_t item_flags, struct rte_flow_error *error)
3473 const struct mlx5_priv *priv = dev->data->dev_private;
3476 if (encap && (!encap->size || !encap->data))
3477 return rte_flow_error_set(error, EINVAL,
3478 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3479 "raw encap data cannot be empty");
3480 if (decap && encap) {
3481 if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
3482 encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3485 else if (encap->size <=
3486 MLX5_ENCAPSULATION_DECISION_SIZE &&
3488 MLX5_ENCAPSULATION_DECISION_SIZE)
3491 else if (encap->size >
3492 MLX5_ENCAPSULATION_DECISION_SIZE &&
3494 MLX5_ENCAPSULATION_DECISION_SIZE)
3495 /* 2 L2 actions: encap and decap. */
3498 return rte_flow_error_set(error,
3500 RTE_FLOW_ERROR_TYPE_ACTION,
3501 NULL, "unsupported too small "
3502 "raw decap and too small raw "
3503 "encap combination");
3506 ret = flow_dv_validate_action_decap(dev, *action_flags, action,
3507 item_flags, attr, error);
3510 *action_flags |= MLX5_FLOW_ACTION_DECAP;
3514 if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
3515 return rte_flow_error_set(error, ENOTSUP,
3516 RTE_FLOW_ERROR_TYPE_ACTION,
3518 "small raw encap size");
3519 if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
3520 return rte_flow_error_set(error, EINVAL,
3521 RTE_FLOW_ERROR_TYPE_ACTION,
3523 "more than one encap action");
3524 if (!attr->transfer && priv->representor)
3525 return rte_flow_error_set
3527 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3528 "encap action for VF representor "
3529 "not supported on NIC table");
3530 *action_flags |= MLX5_FLOW_ACTION_ENCAP;
3537 * Validate the ASO CT action.
3540 * Pointer to the rte_eth_dev structure.
3541 * @param[in] action_flags
3542 * Holds the actions detected until now.
3543 * @param[in] item_flags
3544 * The items found in this flow rule.
3546 * Pointer to flow attributes.
3548 * Pointer to error structure.
3551 * 0 on success, a negative errno value otherwise and rte_errno is set.
3554 flow_dv_validate_action_aso_ct(struct rte_eth_dev *dev,
3555 uint64_t action_flags,
3556 uint64_t item_flags,
3557 const struct rte_flow_attr *attr,
3558 struct rte_flow_error *error)
3562 if (attr->group == 0 && !attr->transfer)
3563 return rte_flow_error_set(error, ENOTSUP,
3564 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3566 "Only support non-root table");
3567 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
3568 return rte_flow_error_set(error, ENOTSUP,
3569 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3570 "CT cannot follow a fate action");
3571 if ((action_flags & MLX5_FLOW_ACTION_METER) ||
3572 (action_flags & MLX5_FLOW_ACTION_AGE))
3573 return rte_flow_error_set(error, EINVAL,
3574 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3575 "Only one ASO action is supported");
3576 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3577 return rte_flow_error_set(error, EINVAL,
3578 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3579 "Encap cannot exist before CT");
3580 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP))
3581 return rte_flow_error_set(error, EINVAL,
3582 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3583 "Not a outer TCP packet");
3588 flow_dv_encap_decap_match_cb(void *tool_ctx __rte_unused,
3589 struct mlx5_list_entry *entry, void *cb_ctx)
3591 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3592 struct mlx5_flow_dv_encap_decap_resource *ctx_resource = ctx->data;
3593 struct mlx5_flow_dv_encap_decap_resource *resource;
3595 resource = container_of(entry, struct mlx5_flow_dv_encap_decap_resource,
3597 if (resource->reformat_type == ctx_resource->reformat_type &&
3598 resource->ft_type == ctx_resource->ft_type &&
3599 resource->flags == ctx_resource->flags &&
3600 resource->size == ctx_resource->size &&
3601 !memcmp((const void *)resource->buf,
3602 (const void *)ctx_resource->buf,
3608 struct mlx5_list_entry *
3609 flow_dv_encap_decap_create_cb(void *tool_ctx, void *cb_ctx)
3611 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3612 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3613 struct mlx5dv_dr_domain *domain;
3614 struct mlx5_flow_dv_encap_decap_resource *ctx_resource = ctx->data;
3615 struct mlx5_flow_dv_encap_decap_resource *resource;
3619 if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3620 domain = sh->fdb_domain;
3621 else if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
3622 domain = sh->rx_domain;
3624 domain = sh->tx_domain;
3625 /* Register new encap/decap resource. */
3626 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], &idx);
3628 rte_flow_error_set(ctx->error, ENOMEM,
3629 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3630 "cannot allocate resource memory");
3633 *resource = *ctx_resource;
3634 resource->idx = idx;
3635 ret = mlx5_flow_os_create_flow_action_packet_reformat(sh->cdev->ctx,
3639 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
3640 rte_flow_error_set(ctx->error, ENOMEM,
3641 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3642 NULL, "cannot create action");
3646 return &resource->entry;
3649 struct mlx5_list_entry *
3650 flow_dv_encap_decap_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
3653 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3654 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3655 struct mlx5_flow_dv_encap_decap_resource *cache_resource;
3658 cache_resource = mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
3660 if (!cache_resource) {
3661 rte_flow_error_set(ctx->error, ENOMEM,
3662 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3663 "cannot allocate resource memory");
3666 memcpy(cache_resource, oentry, sizeof(*cache_resource));
3667 cache_resource->idx = idx;
3668 return &cache_resource->entry;
3672 flow_dv_encap_decap_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
3674 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3675 struct mlx5_flow_dv_encap_decap_resource *res =
3676 container_of(entry, typeof(*res), entry);
3678 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], res->idx);
3682 * Find existing encap/decap resource or create and register a new one.
3684 * @param[in, out] dev
3685 * Pointer to rte_eth_dev structure.
3686 * @param[in, out] resource
3687 * Pointer to encap/decap resource.
3688 * @parm[in, out] dev_flow
3689 * Pointer to the dev_flow.
3691 * pointer to error structure.
3694 * 0 on success otherwise -errno and errno is set.
3697 flow_dv_encap_decap_resource_register
3698 (struct rte_eth_dev *dev,
3699 struct mlx5_flow_dv_encap_decap_resource *resource,
3700 struct mlx5_flow *dev_flow,
3701 struct rte_flow_error *error)
3703 struct mlx5_priv *priv = dev->data->dev_private;
3704 struct mlx5_dev_ctx_shared *sh = priv->sh;
3705 struct mlx5_list_entry *entry;
3709 uint32_t refmt_type:8;
3711 * Header reformat actions can be shared between
3712 * non-root tables. One bit to indicate non-root
3716 uint32_t reserve:15;
3719 } encap_decap_key = {
3721 .ft_type = resource->ft_type,
3722 .refmt_type = resource->reformat_type,
3723 .is_root = !!dev_flow->dv.group,
3727 struct mlx5_flow_cb_ctx ctx = {
3731 struct mlx5_hlist *encaps_decaps;
3734 encaps_decaps = flow_dv_hlist_prepare(sh, &sh->encaps_decaps,
3736 MLX5_FLOW_ENCAP_DECAP_HTABLE_SZ,
3738 flow_dv_encap_decap_create_cb,
3739 flow_dv_encap_decap_match_cb,
3740 flow_dv_encap_decap_remove_cb,
3741 flow_dv_encap_decap_clone_cb,
3742 flow_dv_encap_decap_clone_free_cb,
3744 if (unlikely(!encaps_decaps))
3746 resource->flags = dev_flow->dv.group ? 0 : 1;
3747 key64 = __rte_raw_cksum(&encap_decap_key.v32,
3748 sizeof(encap_decap_key.v32), 0);
3749 if (resource->reformat_type !=
3750 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2 &&
3752 key64 = __rte_raw_cksum(resource->buf, resource->size, key64);
3753 entry = mlx5_hlist_register(encaps_decaps, key64, &ctx);
3756 resource = container_of(entry, typeof(*resource), entry);
3757 dev_flow->dv.encap_decap = resource;
3758 dev_flow->handle->dvh.rix_encap_decap = resource->idx;
3763 * Find existing table jump resource or create and register a new one.
3765 * @param[in, out] dev
3766 * Pointer to rte_eth_dev structure.
3767 * @param[in, out] tbl
3768 * Pointer to flow table resource.
3769 * @parm[in, out] dev_flow
3770 * Pointer to the dev_flow.
3772 * pointer to error structure.
3775 * 0 on success otherwise -errno and errno is set.
3778 flow_dv_jump_tbl_resource_register
3779 (struct rte_eth_dev *dev __rte_unused,
3780 struct mlx5_flow_tbl_resource *tbl,
3781 struct mlx5_flow *dev_flow,
3782 struct rte_flow_error *error __rte_unused)
3784 struct mlx5_flow_tbl_data_entry *tbl_data =
3785 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
3788 MLX5_ASSERT(tbl_data->jump.action);
3789 dev_flow->handle->rix_jump = tbl_data->idx;
3790 dev_flow->dv.jump = &tbl_data->jump;
3795 flow_dv_port_id_match_cb(void *tool_ctx __rte_unused,
3796 struct mlx5_list_entry *entry, void *cb_ctx)
3798 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3799 struct mlx5_flow_dv_port_id_action_resource *ref = ctx->data;
3800 struct mlx5_flow_dv_port_id_action_resource *res =
3801 container_of(entry, typeof(*res), entry);
3803 return ref->port_id != res->port_id;
3806 struct mlx5_list_entry *
3807 flow_dv_port_id_create_cb(void *tool_ctx, void *cb_ctx)
3809 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3810 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3811 struct mlx5_flow_dv_port_id_action_resource *ref = ctx->data;
3812 struct mlx5_flow_dv_port_id_action_resource *resource;
3816 /* Register new port id action resource. */
3817 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID], &idx);
3819 rte_flow_error_set(ctx->error, ENOMEM,
3820 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3821 "cannot allocate port_id action memory");
3825 ret = mlx5_flow_os_create_flow_action_dest_port(sh->fdb_domain,
3829 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PORT_ID], idx);
3830 rte_flow_error_set(ctx->error, ENOMEM,
3831 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3832 "cannot create action");
3835 resource->idx = idx;
3836 return &resource->entry;
3839 struct mlx5_list_entry *
3840 flow_dv_port_id_clone_cb(void *tool_ctx,
3841 struct mlx5_list_entry *entry __rte_unused,
3844 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3845 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3846 struct mlx5_flow_dv_port_id_action_resource *resource;
3849 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID], &idx);
3851 rte_flow_error_set(ctx->error, ENOMEM,
3852 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3853 "cannot allocate port_id action memory");
3856 memcpy(resource, entry, sizeof(*resource));
3857 resource->idx = idx;
3858 return &resource->entry;
3862 flow_dv_port_id_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
3864 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3865 struct mlx5_flow_dv_port_id_action_resource *resource =
3866 container_of(entry, typeof(*resource), entry);
3868 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PORT_ID], resource->idx);
3872 * Find existing table port ID resource or create and register a new one.
3874 * @param[in, out] dev
3875 * Pointer to rte_eth_dev structure.
3876 * @param[in, out] ref
3877 * Pointer to port ID action resource reference.
3878 * @parm[in, out] dev_flow
3879 * Pointer to the dev_flow.
3881 * pointer to error structure.
3884 * 0 on success otherwise -errno and errno is set.
3887 flow_dv_port_id_action_resource_register
3888 (struct rte_eth_dev *dev,
3889 struct mlx5_flow_dv_port_id_action_resource *ref,
3890 struct mlx5_flow *dev_flow,
3891 struct rte_flow_error *error)
3893 struct mlx5_priv *priv = dev->data->dev_private;
3894 struct mlx5_list_entry *entry;
3895 struct mlx5_flow_dv_port_id_action_resource *resource;
3896 struct mlx5_flow_cb_ctx ctx = {
3901 entry = mlx5_list_register(priv->sh->port_id_action_list, &ctx);
3904 resource = container_of(entry, typeof(*resource), entry);
3905 dev_flow->dv.port_id_action = resource;
3906 dev_flow->handle->rix_port_id_action = resource->idx;
3911 flow_dv_push_vlan_match_cb(void *tool_ctx __rte_unused,
3912 struct mlx5_list_entry *entry, void *cb_ctx)
3914 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3915 struct mlx5_flow_dv_push_vlan_action_resource *ref = ctx->data;
3916 struct mlx5_flow_dv_push_vlan_action_resource *res =
3917 container_of(entry, typeof(*res), entry);
3919 return ref->vlan_tag != res->vlan_tag || ref->ft_type != res->ft_type;
3922 struct mlx5_list_entry *
3923 flow_dv_push_vlan_create_cb(void *tool_ctx, void *cb_ctx)
3925 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3926 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3927 struct mlx5_flow_dv_push_vlan_action_resource *ref = ctx->data;
3928 struct mlx5_flow_dv_push_vlan_action_resource *resource;
3929 struct mlx5dv_dr_domain *domain;
3933 /* Register new port id action resource. */
3934 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN], &idx);
3936 rte_flow_error_set(ctx->error, ENOMEM,
3937 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3938 "cannot allocate push_vlan action memory");
3942 if (ref->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3943 domain = sh->fdb_domain;
3944 else if (ref->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
3945 domain = sh->rx_domain;
3947 domain = sh->tx_domain;
3948 ret = mlx5_flow_os_create_flow_action_push_vlan(domain, ref->vlan_tag,
3951 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
3952 rte_flow_error_set(ctx->error, ENOMEM,
3953 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3954 "cannot create push vlan action");
3957 resource->idx = idx;
3958 return &resource->entry;
3961 struct mlx5_list_entry *
3962 flow_dv_push_vlan_clone_cb(void *tool_ctx,
3963 struct mlx5_list_entry *entry __rte_unused,
3966 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3967 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3968 struct mlx5_flow_dv_push_vlan_action_resource *resource;
3971 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN], &idx);
3973 rte_flow_error_set(ctx->error, ENOMEM,
3974 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
3975 "cannot allocate push_vlan action memory");
3978 memcpy(resource, entry, sizeof(*resource));
3979 resource->idx = idx;
3980 return &resource->entry;
3984 flow_dv_push_vlan_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
3986 struct mlx5_dev_ctx_shared *sh = tool_ctx;
3987 struct mlx5_flow_dv_push_vlan_action_resource *resource =
3988 container_of(entry, typeof(*resource), entry);
3990 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PUSH_VLAN], resource->idx);
3994 * Find existing push vlan resource or create and register a new one.
3996 * @param [in, out] dev
3997 * Pointer to rte_eth_dev structure.
3998 * @param[in, out] ref
3999 * Pointer to port ID action resource reference.
4000 * @parm[in, out] dev_flow
4001 * Pointer to the dev_flow.
4003 * pointer to error structure.
4006 * 0 on success otherwise -errno and errno is set.
4009 flow_dv_push_vlan_action_resource_register
4010 (struct rte_eth_dev *dev,
4011 struct mlx5_flow_dv_push_vlan_action_resource *ref,
4012 struct mlx5_flow *dev_flow,
4013 struct rte_flow_error *error)
4015 struct mlx5_priv *priv = dev->data->dev_private;
4016 struct mlx5_flow_dv_push_vlan_action_resource *resource;
4017 struct mlx5_list_entry *entry;
4018 struct mlx5_flow_cb_ctx ctx = {
4023 entry = mlx5_list_register(priv->sh->push_vlan_action_list, &ctx);
4026 resource = container_of(entry, typeof(*resource), entry);
4028 dev_flow->handle->dvh.rix_push_vlan = resource->idx;
4029 dev_flow->dv.push_vlan_res = resource;
4034 * Get the size of specific rte_flow_item_type hdr size
4036 * @param[in] item_type
4037 * Tested rte_flow_item_type.
4040 * sizeof struct item_type, 0 if void or irrelevant.
4043 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
4047 switch (item_type) {
4048 case RTE_FLOW_ITEM_TYPE_ETH:
4049 retval = sizeof(struct rte_ether_hdr);
4051 case RTE_FLOW_ITEM_TYPE_VLAN:
4052 retval = sizeof(struct rte_vlan_hdr);
4054 case RTE_FLOW_ITEM_TYPE_IPV4:
4055 retval = sizeof(struct rte_ipv4_hdr);
4057 case RTE_FLOW_ITEM_TYPE_IPV6:
4058 retval = sizeof(struct rte_ipv6_hdr);
4060 case RTE_FLOW_ITEM_TYPE_UDP:
4061 retval = sizeof(struct rte_udp_hdr);
4063 case RTE_FLOW_ITEM_TYPE_TCP:
4064 retval = sizeof(struct rte_tcp_hdr);
4066 case RTE_FLOW_ITEM_TYPE_VXLAN:
4067 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
4068 retval = sizeof(struct rte_vxlan_hdr);
4070 case RTE_FLOW_ITEM_TYPE_GRE:
4071 case RTE_FLOW_ITEM_TYPE_NVGRE:
4072 retval = sizeof(struct rte_gre_hdr);
4074 case RTE_FLOW_ITEM_TYPE_MPLS:
4075 retval = sizeof(struct rte_mpls_hdr);
4077 case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
4085 #define MLX5_ENCAP_IPV4_VERSION 0x40
4086 #define MLX5_ENCAP_IPV4_IHL_MIN 0x05
4087 #define MLX5_ENCAP_IPV4_TTL_DEF 0x40
4088 #define MLX5_ENCAP_IPV6_VTC_FLOW 0x60000000
4089 #define MLX5_ENCAP_IPV6_HOP_LIMIT 0xff
4090 #define MLX5_ENCAP_VXLAN_FLAGS 0x08000000
4091 #define MLX5_ENCAP_VXLAN_GPE_FLAGS 0x04
4094 * Convert the encap action data from list of rte_flow_item to raw buffer
4097 * Pointer to rte_flow_item objects list.
4099 * Pointer to the output buffer.
4101 * Pointer to the output buffer size.
4103 * Pointer to the error structure.
4106 * 0 on success, a negative errno value otherwise and rte_errno is set.
4109 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
4110 size_t *size, struct rte_flow_error *error)
4112 struct rte_ether_hdr *eth = NULL;
4113 struct rte_vlan_hdr *vlan = NULL;
4114 struct rte_ipv4_hdr *ipv4 = NULL;
4115 struct rte_ipv6_hdr *ipv6 = NULL;
4116 struct rte_udp_hdr *udp = NULL;
4117 struct rte_vxlan_hdr *vxlan = NULL;
4118 struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
4119 struct rte_gre_hdr *gre = NULL;
4121 size_t temp_size = 0;
4124 return rte_flow_error_set(error, EINVAL,
4125 RTE_FLOW_ERROR_TYPE_ACTION,
4126 NULL, "invalid empty data");
4127 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4128 len = flow_dv_get_item_hdr_len(items->type);
4129 if (len + temp_size > MLX5_ENCAP_MAX_LEN)
4130 return rte_flow_error_set(error, EINVAL,
4131 RTE_FLOW_ERROR_TYPE_ACTION,
4132 (void *)items->type,
4133 "items total size is too big"
4134 " for encap action");
4135 rte_memcpy((void *)&buf[temp_size], items->spec, len);
4136 switch (items->type) {
4137 case RTE_FLOW_ITEM_TYPE_ETH:
4138 eth = (struct rte_ether_hdr *)&buf[temp_size];
4140 case RTE_FLOW_ITEM_TYPE_VLAN:
4141 vlan = (struct rte_vlan_hdr *)&buf[temp_size];
4143 return rte_flow_error_set(error, EINVAL,
4144 RTE_FLOW_ERROR_TYPE_ACTION,
4145 (void *)items->type,
4146 "eth header not found");
4147 if (!eth->ether_type)
4148 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
4150 case RTE_FLOW_ITEM_TYPE_IPV4:
4151 ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
4153 return rte_flow_error_set(error, EINVAL,
4154 RTE_FLOW_ERROR_TYPE_ACTION,
4155 (void *)items->type,
4156 "neither eth nor vlan"
4158 if (vlan && !vlan->eth_proto)
4159 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
4160 else if (eth && !eth->ether_type)
4161 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
4162 if (!ipv4->version_ihl)
4163 ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
4164 MLX5_ENCAP_IPV4_IHL_MIN;
4165 if (!ipv4->time_to_live)
4166 ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
4168 case RTE_FLOW_ITEM_TYPE_IPV6:
4169 ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
4171 return rte_flow_error_set(error, EINVAL,
4172 RTE_FLOW_ERROR_TYPE_ACTION,
4173 (void *)items->type,
4174 "neither eth nor vlan"
4176 if (vlan && !vlan->eth_proto)
4177 vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
4178 else if (eth && !eth->ether_type)
4179 eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
4180 if (!ipv6->vtc_flow)
4182 RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
4183 if (!ipv6->hop_limits)
4184 ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
4186 case RTE_FLOW_ITEM_TYPE_UDP:
4187 udp = (struct rte_udp_hdr *)&buf[temp_size];
4189 return rte_flow_error_set(error, EINVAL,
4190 RTE_FLOW_ERROR_TYPE_ACTION,
4191 (void *)items->type,
4192 "ip header not found");
4193 if (ipv4 && !ipv4->next_proto_id)
4194 ipv4->next_proto_id = IPPROTO_UDP;
4195 else if (ipv6 && !ipv6->proto)
4196 ipv6->proto = IPPROTO_UDP;
4198 case RTE_FLOW_ITEM_TYPE_VXLAN:
4199 vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
4201 return rte_flow_error_set(error, EINVAL,
4202 RTE_FLOW_ERROR_TYPE_ACTION,
4203 (void *)items->type,
4204 "udp header not found");
4206 udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
4207 if (!vxlan->vx_flags)
4209 RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
4211 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
4212 vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
4214 return rte_flow_error_set(error, EINVAL,
4215 RTE_FLOW_ERROR_TYPE_ACTION,
4216 (void *)items->type,
4217 "udp header not found");
4218 if (!vxlan_gpe->proto)
4219 return rte_flow_error_set(error, EINVAL,
4220 RTE_FLOW_ERROR_TYPE_ACTION,
4221 (void *)items->type,
4222 "next protocol not found");
4225 RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
4226 if (!vxlan_gpe->vx_flags)
4227 vxlan_gpe->vx_flags =
4228 MLX5_ENCAP_VXLAN_GPE_FLAGS;
4230 case RTE_FLOW_ITEM_TYPE_GRE:
4231 case RTE_FLOW_ITEM_TYPE_NVGRE:
4232 gre = (struct rte_gre_hdr *)&buf[temp_size];
4234 return rte_flow_error_set(error, EINVAL,
4235 RTE_FLOW_ERROR_TYPE_ACTION,
4236 (void *)items->type,
4237 "next protocol not found");
4239 return rte_flow_error_set(error, EINVAL,
4240 RTE_FLOW_ERROR_TYPE_ACTION,
4241 (void *)items->type,
4242 "ip header not found");
4243 if (ipv4 && !ipv4->next_proto_id)
4244 ipv4->next_proto_id = IPPROTO_GRE;
4245 else if (ipv6 && !ipv6->proto)
4246 ipv6->proto = IPPROTO_GRE;
4248 case RTE_FLOW_ITEM_TYPE_VOID:
4251 return rte_flow_error_set(error, EINVAL,
4252 RTE_FLOW_ERROR_TYPE_ACTION,
4253 (void *)items->type,
4254 "unsupported item type");
4264 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
4266 struct rte_ether_hdr *eth = NULL;
4267 struct rte_vlan_hdr *vlan = NULL;
4268 struct rte_ipv6_hdr *ipv6 = NULL;
4269 struct rte_udp_hdr *udp = NULL;
4273 eth = (struct rte_ether_hdr *)data;
4274 next_hdr = (char *)(eth + 1);
4275 proto = RTE_BE16(eth->ether_type);
4278 while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
4279 vlan = (struct rte_vlan_hdr *)next_hdr;
4280 proto = RTE_BE16(vlan->eth_proto);
4281 next_hdr += sizeof(struct rte_vlan_hdr);
4284 /* HW calculates IPv4 csum. no need to proceed */
4285 if (proto == RTE_ETHER_TYPE_IPV4)
4288 /* non IPv4/IPv6 header. not supported */
4289 if (proto != RTE_ETHER_TYPE_IPV6) {
4290 return rte_flow_error_set(error, ENOTSUP,
4291 RTE_FLOW_ERROR_TYPE_ACTION,
4292 NULL, "Cannot offload non IPv4/IPv6");
4295 ipv6 = (struct rte_ipv6_hdr *)next_hdr;
4297 /* ignore non UDP */
4298 if (ipv6->proto != IPPROTO_UDP)
4301 udp = (struct rte_udp_hdr *)(ipv6 + 1);
4302 udp->dgram_cksum = 0;
4308 * Convert L2 encap action to DV specification.
4311 * Pointer to rte_eth_dev structure.
4313 * Pointer to action structure.
4314 * @param[in, out] dev_flow
4315 * Pointer to the mlx5_flow.
4316 * @param[in] transfer
4317 * Mark if the flow is E-Switch flow.
4319 * Pointer to the error structure.
4322 * 0 on success, a negative errno value otherwise and rte_errno is set.
4325 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
4326 const struct rte_flow_action *action,
4327 struct mlx5_flow *dev_flow,
4329 struct rte_flow_error *error)
4331 const struct rte_flow_item *encap_data;
4332 const struct rte_flow_action_raw_encap *raw_encap_data;
4333 struct mlx5_flow_dv_encap_decap_resource res = {
4335 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
4336 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
4337 MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
4340 if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
4342 (const struct rte_flow_action_raw_encap *)action->conf;
4343 res.size = raw_encap_data->size;
4344 memcpy(res.buf, raw_encap_data->data, res.size);
4346 if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
4348 ((const struct rte_flow_action_vxlan_encap *)
4349 action->conf)->definition;
4352 ((const struct rte_flow_action_nvgre_encap *)
4353 action->conf)->definition;
4354 if (flow_dv_convert_encap_data(encap_data, res.buf,
4358 if (flow_dv_zero_encap_udp_csum(res.buf, error))
4360 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
4361 return rte_flow_error_set(error, EINVAL,
4362 RTE_FLOW_ERROR_TYPE_ACTION,
4363 NULL, "can't create L2 encap action");
4368 * Convert L2 decap action to DV specification.
4371 * Pointer to rte_eth_dev structure.
4372 * @param[in, out] dev_flow
4373 * Pointer to the mlx5_flow.
4374 * @param[in] transfer
4375 * Mark if the flow is E-Switch flow.
4377 * Pointer to the error structure.
4380 * 0 on success, a negative errno value otherwise and rte_errno is set.
4383 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
4384 struct mlx5_flow *dev_flow,
4386 struct rte_flow_error *error)
4388 struct mlx5_flow_dv_encap_decap_resource res = {
4391 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
4392 .ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
4393 MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
4396 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
4397 return rte_flow_error_set(error, EINVAL,
4398 RTE_FLOW_ERROR_TYPE_ACTION,
4399 NULL, "can't create L2 decap action");
4404 * Convert raw decap/encap (L3 tunnel) action to DV specification.
4407 * Pointer to rte_eth_dev structure.
4409 * Pointer to action structure.
4410 * @param[in, out] dev_flow
4411 * Pointer to the mlx5_flow.
4413 * Pointer to the flow attributes.
4415 * Pointer to the error structure.
4418 * 0 on success, a negative errno value otherwise and rte_errno is set.
4421 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
4422 const struct rte_flow_action *action,
4423 struct mlx5_flow *dev_flow,
4424 const struct rte_flow_attr *attr,
4425 struct rte_flow_error *error)
4427 const struct rte_flow_action_raw_encap *encap_data;
4428 struct mlx5_flow_dv_encap_decap_resource res;
4430 memset(&res, 0, sizeof(res));
4431 encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
4432 res.size = encap_data->size;
4433 memcpy(res.buf, encap_data->data, res.size);
4434 res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
4435 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
4436 MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
4438 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
4440 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
4441 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
4442 if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
4443 return rte_flow_error_set(error, EINVAL,
4444 RTE_FLOW_ERROR_TYPE_ACTION,
4445 NULL, "can't create encap action");
4450 * Create action push VLAN.
4453 * Pointer to rte_eth_dev structure.
4455 * Pointer to the flow attributes.
4457 * Pointer to the vlan to push to the Ethernet header.
4458 * @param[in, out] dev_flow
4459 * Pointer to the mlx5_flow.
4461 * Pointer to the error structure.
4464 * 0 on success, a negative errno value otherwise and rte_errno is set.
4467 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
4468 const struct rte_flow_attr *attr,
4469 const struct rte_vlan_hdr *vlan,
4470 struct mlx5_flow *dev_flow,
4471 struct rte_flow_error *error)
4473 struct mlx5_flow_dv_push_vlan_action_resource res;
4475 memset(&res, 0, sizeof(res));
4477 rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
4480 res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
4482 res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
4483 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
4484 return flow_dv_push_vlan_action_resource_register
4485 (dev, &res, dev_flow, error);
4489 * Validate the modify-header actions.
4491 * @param[in] action_flags
4492 * Holds the actions detected until now.
4494 * Pointer to the modify action.
4496 * Pointer to error structure.
4499 * 0 on success, a negative errno value otherwise and rte_errno is set.
4502 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
4503 const struct rte_flow_action *action,
4504 struct rte_flow_error *error)
4506 if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
4507 return rte_flow_error_set(error, EINVAL,
4508 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4509 NULL, "action configuration not set");
4510 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
4511 return rte_flow_error_set(error, EINVAL,
4512 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4513 "can't have encap action before"
4519 * Validate the modify-header MAC address actions.
4521 * @param[in] action_flags
4522 * Holds the actions detected until now.
4524 * Pointer to the modify action.
4525 * @param[in] item_flags
4526 * Holds the items detected.
4528 * Pointer to error structure.
4531 * 0 on success, a negative errno value otherwise and rte_errno is set.
4534 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
4535 const struct rte_flow_action *action,
4536 const uint64_t item_flags,
4537 struct rte_flow_error *error)
4541 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4543 if (!(item_flags & MLX5_FLOW_LAYER_L2))
4544 return rte_flow_error_set(error, EINVAL,
4545 RTE_FLOW_ERROR_TYPE_ACTION,
4547 "no L2 item in pattern");
4553 * Validate the modify-header IPv4 address actions.
4555 * @param[in] action_flags
4556 * Holds the actions detected until now.
4558 * Pointer to the modify action.
4559 * @param[in] item_flags
4560 * Holds the items detected.
4562 * Pointer to error structure.
4565 * 0 on success, a negative errno value otherwise and rte_errno is set.
4568 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
4569 const struct rte_flow_action *action,
4570 const uint64_t item_flags,
4571 struct rte_flow_error *error)
4576 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4578 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4579 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
4580 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4581 if (!(item_flags & layer))
4582 return rte_flow_error_set(error, EINVAL,
4583 RTE_FLOW_ERROR_TYPE_ACTION,
4585 "no ipv4 item in pattern");
4591 * Validate the modify-header IPv6 address actions.
4593 * @param[in] action_flags
4594 * Holds the actions detected until now.
4596 * Pointer to the modify action.
4597 * @param[in] item_flags
4598 * Holds the items detected.
4600 * Pointer to error structure.
4603 * 0 on success, a negative errno value otherwise and rte_errno is set.
4606 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
4607 const struct rte_flow_action *action,
4608 const uint64_t item_flags,
4609 struct rte_flow_error *error)
4614 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4616 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4617 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
4618 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4619 if (!(item_flags & layer))
4620 return rte_flow_error_set(error, EINVAL,
4621 RTE_FLOW_ERROR_TYPE_ACTION,
4623 "no ipv6 item in pattern");
4629 * Validate the modify-header TP actions.
4631 * @param[in] action_flags
4632 * Holds the actions detected until now.
4634 * Pointer to the modify action.
4635 * @param[in] item_flags
4636 * Holds the items detected.
4638 * Pointer to error structure.
4641 * 0 on success, a negative errno value otherwise and rte_errno is set.
4644 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
4645 const struct rte_flow_action *action,
4646 const uint64_t item_flags,
4647 struct rte_flow_error *error)
4652 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4654 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4655 MLX5_FLOW_LAYER_INNER_L4 :
4656 MLX5_FLOW_LAYER_OUTER_L4;
4657 if (!(item_flags & layer))
4658 return rte_flow_error_set(error, EINVAL,
4659 RTE_FLOW_ERROR_TYPE_ACTION,
4660 NULL, "no transport layer "
4667 * Validate the modify-header actions of increment/decrement
4668 * TCP Sequence-number.
4670 * @param[in] action_flags
4671 * Holds the actions detected until now.
4673 * Pointer to the modify action.
4674 * @param[in] item_flags
4675 * Holds the items detected.
4677 * Pointer to error structure.
4680 * 0 on success, a negative errno value otherwise and rte_errno is set.
4683 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
4684 const struct rte_flow_action *action,
4685 const uint64_t item_flags,
4686 struct rte_flow_error *error)
4691 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4693 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4694 MLX5_FLOW_LAYER_INNER_L4_TCP :
4695 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4696 if (!(item_flags & layer))
4697 return rte_flow_error_set(error, EINVAL,
4698 RTE_FLOW_ERROR_TYPE_ACTION,
4699 NULL, "no TCP item in"
4701 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
4702 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
4703 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
4704 (action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
4705 return rte_flow_error_set(error, EINVAL,
4706 RTE_FLOW_ERROR_TYPE_ACTION,
4708 "cannot decrease and increase"
4709 " TCP sequence number"
4710 " at the same time");
4716 * Validate the modify-header actions of increment/decrement
4717 * TCP Acknowledgment number.
4719 * @param[in] action_flags
4720 * Holds the actions detected until now.
4722 * Pointer to the modify action.
4723 * @param[in] item_flags
4724 * Holds the items detected.
4726 * Pointer to error structure.
4729 * 0 on success, a negative errno value otherwise and rte_errno is set.
4732 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
4733 const struct rte_flow_action *action,
4734 const uint64_t item_flags,
4735 struct rte_flow_error *error)
4740 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4742 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4743 MLX5_FLOW_LAYER_INNER_L4_TCP :
4744 MLX5_FLOW_LAYER_OUTER_L4_TCP;
4745 if (!(item_flags & layer))
4746 return rte_flow_error_set(error, EINVAL,
4747 RTE_FLOW_ERROR_TYPE_ACTION,
4748 NULL, "no TCP item in"
4750 if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
4751 (action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
4752 (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
4753 (action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
4754 return rte_flow_error_set(error, EINVAL,
4755 RTE_FLOW_ERROR_TYPE_ACTION,
4757 "cannot decrease and increase"
4758 " TCP acknowledgment number"
4759 " at the same time");
4765 * Validate the modify-header TTL actions.
4767 * @param[in] action_flags
4768 * Holds the actions detected until now.
4770 * Pointer to the modify action.
4771 * @param[in] item_flags
4772 * Holds the items detected.
4774 * Pointer to error structure.
4777 * 0 on success, a negative errno value otherwise and rte_errno is set.
4780 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
4781 const struct rte_flow_action *action,
4782 const uint64_t item_flags,
4783 struct rte_flow_error *error)
4788 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4790 layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
4791 MLX5_FLOW_LAYER_INNER_L3 :
4792 MLX5_FLOW_LAYER_OUTER_L3;
4793 if (!(item_flags & layer))
4794 return rte_flow_error_set(error, EINVAL,
4795 RTE_FLOW_ERROR_TYPE_ACTION,
4797 "no IP protocol in pattern");
4803 * Validate the generic modify field actions.
4805 * Pointer to the rte_eth_dev structure.
4806 * @param[in] action_flags
4807 * Holds the actions detected until now.
4809 * Pointer to the modify action.
4811 * Pointer to the flow attributes.
4813 * Pointer to error structure.
4816 * Number of header fields to modify (0 or more) on success,
4817 * a negative errno value otherwise and rte_errno is set.
4820 flow_dv_validate_action_modify_field(struct rte_eth_dev *dev,
4821 const uint64_t action_flags,
4822 const struct rte_flow_action *action,
4823 const struct rte_flow_attr *attr,
4824 struct rte_flow_error *error)
4827 struct mlx5_priv *priv = dev->data->dev_private;
4828 struct mlx5_sh_config *config = &priv->sh->config;
4829 const struct rte_flow_action_modify_field *action_modify_field =
4831 uint32_t dst_width = mlx5_flow_item_field_width(dev,
4832 action_modify_field->dst.field,
4834 uint32_t src_width = mlx5_flow_item_field_width(dev,
4835 action_modify_field->src.field,
4836 dst_width, attr, error);
4838 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4842 if (action_modify_field->width == 0)
4843 return rte_flow_error_set(error, EINVAL,
4844 RTE_FLOW_ERROR_TYPE_ACTION, action,
4845 "no bits are requested to be modified");
4846 else if (action_modify_field->width > dst_width ||
4847 action_modify_field->width > src_width)
4848 return rte_flow_error_set(error, EINVAL,
4849 RTE_FLOW_ERROR_TYPE_ACTION, action,
4850 "cannot modify more bits than"
4851 " the width of a field");
4852 if (action_modify_field->dst.field != RTE_FLOW_FIELD_VALUE &&
4853 action_modify_field->dst.field != RTE_FLOW_FIELD_POINTER) {
4854 if ((action_modify_field->dst.offset +
4855 action_modify_field->width > dst_width) ||
4856 (action_modify_field->dst.offset % 32))
4857 return rte_flow_error_set(error, EINVAL,
4858 RTE_FLOW_ERROR_TYPE_ACTION, action,
4859 "destination offset is too big"
4860 " or not aligned to 4 bytes");
4861 if (action_modify_field->dst.level &&
4862 action_modify_field->dst.field != RTE_FLOW_FIELD_TAG)
4863 return rte_flow_error_set(error, ENOTSUP,
4864 RTE_FLOW_ERROR_TYPE_ACTION, action,
4865 "inner header fields modification"
4866 " is not supported");
4868 if (action_modify_field->src.field != RTE_FLOW_FIELD_VALUE &&
4869 action_modify_field->src.field != RTE_FLOW_FIELD_POINTER) {
4870 if (!attr->transfer && !attr->group)
4871 return rte_flow_error_set(error, ENOTSUP,
4872 RTE_FLOW_ERROR_TYPE_ACTION, action,
4873 "modify field action is not"
4874 " supported for group 0");
4875 if ((action_modify_field->src.offset +
4876 action_modify_field->width > src_width) ||
4877 (action_modify_field->src.offset % 32))
4878 return rte_flow_error_set(error, EINVAL,
4879 RTE_FLOW_ERROR_TYPE_ACTION, action,
4880 "source offset is too big"
4881 " or not aligned to 4 bytes");
4882 if (action_modify_field->src.level &&
4883 action_modify_field->src.field != RTE_FLOW_FIELD_TAG)
4884 return rte_flow_error_set(error, ENOTSUP,
4885 RTE_FLOW_ERROR_TYPE_ACTION, action,
4886 "inner header fields modification"
4887 " is not supported");
4889 if ((action_modify_field->dst.field ==
4890 action_modify_field->src.field) &&
4891 (action_modify_field->dst.level ==
4892 action_modify_field->src.level))
4893 return rte_flow_error_set(error, EINVAL,
4894 RTE_FLOW_ERROR_TYPE_ACTION, action,
4895 "source and destination fields"
4896 " cannot be the same");
4897 if (action_modify_field->dst.field == RTE_FLOW_FIELD_VALUE ||
4898 action_modify_field->dst.field == RTE_FLOW_FIELD_POINTER ||
4899 action_modify_field->dst.field == RTE_FLOW_FIELD_MARK)
4900 return rte_flow_error_set(error, EINVAL,
4901 RTE_FLOW_ERROR_TYPE_ACTION, action,
4902 "mark, immediate value or a pointer to it"
4903 " cannot be used as a destination");
4904 if (action_modify_field->dst.field == RTE_FLOW_FIELD_START ||
4905 action_modify_field->src.field == RTE_FLOW_FIELD_START)
4906 return rte_flow_error_set(error, ENOTSUP,
4907 RTE_FLOW_ERROR_TYPE_ACTION, action,
4908 "modifications of an arbitrary"
4909 " place in a packet is not supported");
4910 if (action_modify_field->dst.field == RTE_FLOW_FIELD_VLAN_TYPE ||
4911 action_modify_field->src.field == RTE_FLOW_FIELD_VLAN_TYPE)
4912 return rte_flow_error_set(error, ENOTSUP,
4913 RTE_FLOW_ERROR_TYPE_ACTION, action,
4914 "modifications of the 802.1Q Tag"
4915 " Identifier is not supported");
4916 if (action_modify_field->dst.field == RTE_FLOW_FIELD_VXLAN_VNI ||
4917 action_modify_field->src.field == RTE_FLOW_FIELD_VXLAN_VNI)
4918 return rte_flow_error_set(error, ENOTSUP,
4919 RTE_FLOW_ERROR_TYPE_ACTION, action,
4920 "modifications of the VXLAN Network"
4921 " Identifier is not supported");
4922 if (action_modify_field->dst.field == RTE_FLOW_FIELD_GENEVE_VNI ||
4923 action_modify_field->src.field == RTE_FLOW_FIELD_GENEVE_VNI)
4924 return rte_flow_error_set(error, ENOTSUP,
4925 RTE_FLOW_ERROR_TYPE_ACTION, action,
4926 "modifications of the GENEVE Network"
4927 " Identifier is not supported");
4928 if (action_modify_field->dst.field == RTE_FLOW_FIELD_MARK ||
4929 action_modify_field->src.field == RTE_FLOW_FIELD_MARK)
4930 if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4931 !mlx5_flow_ext_mreg_supported(dev))
4932 return rte_flow_error_set(error, ENOTSUP,
4933 RTE_FLOW_ERROR_TYPE_ACTION, action,
4934 "cannot modify mark in legacy mode"
4935 " or without extensive registers");
4936 if (action_modify_field->dst.field == RTE_FLOW_FIELD_META ||
4937 action_modify_field->src.field == RTE_FLOW_FIELD_META) {
4938 if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
4939 !mlx5_flow_ext_mreg_supported(dev))
4940 return rte_flow_error_set(error, ENOTSUP,
4941 RTE_FLOW_ERROR_TYPE_ACTION, action,
4942 "cannot modify meta without"
4943 " extensive registers support");
4944 ret = flow_dv_get_metadata_reg(dev, attr, error);
4945 if (ret < 0 || ret == REG_NON)
4946 return rte_flow_error_set(error, ENOTSUP,
4947 RTE_FLOW_ERROR_TYPE_ACTION, action,
4948 "cannot modify meta without"
4949 " extensive registers available");
4951 if (action_modify_field->operation != RTE_FLOW_MODIFY_SET)
4952 return rte_flow_error_set(error, ENOTSUP,
4953 RTE_FLOW_ERROR_TYPE_ACTION, action,
4954 "add and sub operations"
4955 " are not supported");
4956 return (action_modify_field->width / 32) +
4957 !!(action_modify_field->width % 32);
4961 * Validate jump action.
4964 * Pointer to the jump action.
4965 * @param[in] action_flags
4966 * Holds the actions detected until now.
4967 * @param[in] attributes
4968 * Pointer to flow attributes
4969 * @param[in] external
4970 * Action belongs to flow rule created by request external to PMD.
4972 * Pointer to error structure.
4975 * 0 on success, a negative errno value otherwise and rte_errno is set.
4978 flow_dv_validate_action_jump(struct rte_eth_dev *dev,
4979 const struct mlx5_flow_tunnel *tunnel,
4980 const struct rte_flow_action *action,
4981 uint64_t action_flags,
4982 const struct rte_flow_attr *attributes,
4983 bool external, struct rte_flow_error *error)
4985 uint32_t target_group, table = 0;
4987 struct flow_grp_info grp_info = {
4988 .external = !!external,
4989 .transfer = !!attributes->transfer,
4993 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
4994 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
4995 return rte_flow_error_set(error, EINVAL,
4996 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4997 "can't have 2 fate actions in"
5000 return rte_flow_error_set(error, EINVAL,
5001 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
5002 NULL, "action configuration not set");
5004 ((const struct rte_flow_action_jump *)action->conf)->group;
5005 ret = mlx5_flow_group_to_table(dev, tunnel, target_group, &table,
5009 if (attributes->group == target_group &&
5010 !(action_flags & (MLX5_FLOW_ACTION_TUNNEL_SET |
5011 MLX5_FLOW_ACTION_TUNNEL_MATCH)))
5012 return rte_flow_error_set(error, EINVAL,
5013 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5014 "target group must be other than"
5015 " the current flow group");
5017 return rte_flow_error_set(error, EINVAL,
5018 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
5019 NULL, "root table shouldn't be destination");
5024 * Validate action PORT_ID / REPRESENTED_PORT.
5027 * Pointer to rte_eth_dev structure.
5028 * @param[in] action_flags
5029 * Bit-fields that holds the actions detected until now.
5031 * PORT_ID / REPRESENTED_PORT action structure.
5033 * Attributes of flow that includes this action.
5035 * Pointer to error structure.
5038 * 0 on success, a negative errno value otherwise and rte_errno is set.
5041 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
5042 uint64_t action_flags,
5043 const struct rte_flow_action *action,
5044 const struct rte_flow_attr *attr,
5045 struct rte_flow_error *error)
5047 const struct rte_flow_action_port_id *port_id;
5048 const struct rte_flow_action_ethdev *ethdev;
5049 struct mlx5_priv *act_priv;
5050 struct mlx5_priv *dev_priv;
5053 if (!attr->transfer)
5054 return rte_flow_error_set(error, ENOTSUP,
5055 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5057 "port action is valid in transfer"
5059 if (!action || !action->conf)
5060 return rte_flow_error_set(error, ENOTSUP,
5061 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
5063 "port action parameters must be"
5065 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
5066 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
5067 return rte_flow_error_set(error, EINVAL,
5068 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5069 "can have only one fate actions in"
5071 dev_priv = mlx5_dev_to_eswitch_info(dev);
5073 return rte_flow_error_set(error, rte_errno,
5074 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5076 "failed to obtain E-Switch info");
5077 switch (action->type) {
5078 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5079 port_id = action->conf;
5080 port = port_id->original ? dev->data->port_id : port_id->id;
5082 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
5083 ethdev = action->conf;
5084 port = ethdev->port_id;
5088 return rte_flow_error_set
5090 RTE_FLOW_ERROR_TYPE_ACTION, action,
5091 "unknown E-Switch action");
5093 act_priv = mlx5_port_to_eswitch_info(port, false);
5095 return rte_flow_error_set
5097 RTE_FLOW_ERROR_TYPE_ACTION_CONF, action->conf,
5098 "failed to obtain E-Switch port id for port");
5099 if (act_priv->domain_id != dev_priv->domain_id)
5100 return rte_flow_error_set
5102 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5103 "port does not belong to"
5104 " E-Switch being configured");
5109 * Get the maximum number of modify header actions.
5112 * Pointer to rte_eth_dev structure.
5114 * Whether action is on root table.
5117 * Max number of modify header actions device can support.
5119 static inline unsigned int
5120 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
5124 * There's no way to directly query the max capacity from FW.
5125 * The maximal value on root table should be assumed to be supported.
5128 return MLX5_MAX_MODIFY_NUM;
5130 return MLX5_ROOT_TBL_MODIFY_NUM;
5134 * Validate the meter action.
5137 * Pointer to rte_eth_dev structure.
5138 * @param[in] action_flags
5139 * Bit-fields that holds the actions detected until now.
5140 * @param[in] item_flags
5141 * Holds the items detected.
5143 * Pointer to the meter action.
5145 * Attributes of flow that includes this action.
5146 * @param[in] port_id_item
5147 * Pointer to item indicating port id.
5149 * Pointer to error structure.
5152 * 0 on success, a negative errno value otherwise and rte_errno is set.
5155 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
5156 uint64_t action_flags, uint64_t item_flags,
5157 const struct rte_flow_action *action,
5158 const struct rte_flow_attr *attr,
5159 const struct rte_flow_item *port_id_item,
5161 struct rte_flow_error *error)
5163 struct mlx5_priv *priv = dev->data->dev_private;
5164 const struct rte_flow_action_meter *am = action->conf;
5165 struct mlx5_flow_meter_info *fm;
5166 struct mlx5_flow_meter_policy *mtr_policy;
5167 struct mlx5_flow_mtr_mng *mtrmng = priv->sh->mtrmng;
5170 return rte_flow_error_set(error, EINVAL,
5171 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5172 "meter action conf is NULL");
5174 if (action_flags & MLX5_FLOW_ACTION_METER)
5175 return rte_flow_error_set(error, ENOTSUP,
5176 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5177 "meter chaining not support");
5178 if (action_flags & MLX5_FLOW_ACTION_JUMP)
5179 return rte_flow_error_set(error, ENOTSUP,
5180 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5181 "meter with jump not support");
5183 return rte_flow_error_set(error, ENOTSUP,
5184 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5186 "meter action not supported");
5187 fm = mlx5_flow_meter_find(priv, am->mtr_id, NULL);
5189 return rte_flow_error_set(error, EINVAL,
5190 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5192 /* aso meter can always be shared by different domains */
5193 if (fm->ref_cnt && !priv->sh->meter_aso_en &&
5194 !(fm->transfer == attr->transfer ||
5195 (!fm->ingress && !attr->ingress && attr->egress) ||
5196 (!fm->egress && !attr->egress && attr->ingress)))
5197 return rte_flow_error_set(error, EINVAL,
5198 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5199 "Flow attributes domain are either invalid "
5200 "or have a domain conflict with current "
5201 "meter attributes");
5202 if (fm->def_policy) {
5203 if (!((attr->transfer &&
5204 mtrmng->def_policy[MLX5_MTR_DOMAIN_TRANSFER]) ||
5206 mtrmng->def_policy[MLX5_MTR_DOMAIN_EGRESS]) ||
5208 mtrmng->def_policy[MLX5_MTR_DOMAIN_INGRESS])))
5209 return rte_flow_error_set(error, EINVAL,
5210 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5211 "Flow attributes domain "
5212 "have a conflict with current "
5213 "meter domain attributes");
5216 mtr_policy = mlx5_flow_meter_policy_find(dev,
5217 fm->policy_id, NULL);
5219 return rte_flow_error_set(error, EINVAL,
5220 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5221 "Invalid policy id for meter ");
5222 if (!((attr->transfer && mtr_policy->transfer) ||
5223 (attr->egress && mtr_policy->egress) ||
5224 (attr->ingress && mtr_policy->ingress)))
5225 return rte_flow_error_set(error, EINVAL,
5226 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5227 "Flow attributes domain "
5228 "have a conflict with current "
5229 "meter domain attributes");
5230 if (attr->transfer && mtr_policy->dev) {
5232 * When policy has fate action of port_id,
5233 * the flow should have the same src port as policy.
5235 struct mlx5_priv *policy_port_priv =
5236 mtr_policy->dev->data->dev_private;
5237 int32_t flow_src_port = priv->representor_id;
5240 const struct rte_flow_item_port_id *spec =
5242 struct mlx5_priv *port_priv =
5243 mlx5_port_to_eswitch_info(spec->id,
5246 return rte_flow_error_set(error,
5248 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5250 "Failed to get port info.");
5251 flow_src_port = port_priv->representor_id;
5253 if (flow_src_port != policy_port_priv->representor_id)
5254 return rte_flow_error_set(error,
5256 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5258 "Flow and meter policy "
5259 "have different src port.");
5260 } else if (mtr_policy->is_rss) {
5261 struct mlx5_flow_meter_policy *fp;
5262 struct mlx5_meter_policy_action_container *acg;
5263 struct mlx5_meter_policy_action_container *acy;
5264 const struct rte_flow_action *rss_act;
5267 fp = mlx5_flow_meter_hierarchy_get_final_policy(dev,
5270 return rte_flow_error_set(error, EINVAL,
5271 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5272 "Unable to get the final "
5273 "policy in the hierarchy");
5274 acg = &fp->act_cnt[RTE_COLOR_GREEN];
5275 acy = &fp->act_cnt[RTE_COLOR_YELLOW];
5276 MLX5_ASSERT(acg->fate_action ==
5277 MLX5_FLOW_FATE_SHARED_RSS ||
5279 MLX5_FLOW_FATE_SHARED_RSS);
5280 if (acg->fate_action == MLX5_FLOW_FATE_SHARED_RSS)
5284 ret = mlx5_flow_validate_action_rss(rss_act,
5285 action_flags, dev, attr,
5290 *def_policy = false;
5296 * Validate the age action.
5298 * @param[in] action_flags
5299 * Holds the actions detected until now.
5301 * Pointer to the age action.
5303 * Pointer to the Ethernet device structure.
5305 * Pointer to error structure.
5308 * 0 on success, a negative errno value otherwise and rte_errno is set.
5311 flow_dv_validate_action_age(uint64_t action_flags,
5312 const struct rte_flow_action *action,
5313 struct rte_eth_dev *dev,
5314 struct rte_flow_error *error)
5316 struct mlx5_priv *priv = dev->data->dev_private;
5317 const struct rte_flow_action_age *age = action->conf;
5319 if (!priv->sh->cdev->config.devx ||
5320 (priv->sh->cmng.counter_fallback && !priv->sh->aso_age_mng))
5321 return rte_flow_error_set(error, ENOTSUP,
5322 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5324 "age action not supported");
5325 if (!(action->conf))
5326 return rte_flow_error_set(error, EINVAL,
5327 RTE_FLOW_ERROR_TYPE_ACTION, action,
5328 "configuration cannot be null");
5329 if (!(age->timeout))
5330 return rte_flow_error_set(error, EINVAL,
5331 RTE_FLOW_ERROR_TYPE_ACTION, action,
5332 "invalid timeout value 0");
5333 if (action_flags & MLX5_FLOW_ACTION_AGE)
5334 return rte_flow_error_set(error, EINVAL,
5335 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5336 "duplicate age actions set");
5341 * Validate the modify-header IPv4 DSCP actions.
5343 * @param[in] action_flags
5344 * Holds the actions detected until now.
5346 * Pointer to the modify action.
5347 * @param[in] item_flags
5348 * Holds the items detected.
5350 * Pointer to error structure.
5353 * 0 on success, a negative errno value otherwise and rte_errno is set.
5356 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
5357 const struct rte_flow_action *action,
5358 const uint64_t item_flags,
5359 struct rte_flow_error *error)
5363 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
5365 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
5366 return rte_flow_error_set(error, EINVAL,
5367 RTE_FLOW_ERROR_TYPE_ACTION,
5369 "no ipv4 item in pattern");
5375 * Validate the modify-header IPv6 DSCP actions.
5377 * @param[in] action_flags
5378 * Holds the actions detected until now.
5380 * Pointer to the modify action.
5381 * @param[in] item_flags
5382 * Holds the items detected.
5384 * Pointer to error structure.
5387 * 0 on success, a negative errno value otherwise and rte_errno is set.
5390 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
5391 const struct rte_flow_action *action,
5392 const uint64_t item_flags,
5393 struct rte_flow_error *error)
5397 ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
5399 if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
5400 return rte_flow_error_set(error, EINVAL,
5401 RTE_FLOW_ERROR_TYPE_ACTION,
5403 "no ipv6 item in pattern");
5409 flow_dv_modify_match_cb(void *tool_ctx __rte_unused,
5410 struct mlx5_list_entry *entry, void *cb_ctx)
5412 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
5413 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
5414 struct mlx5_flow_dv_modify_hdr_resource *resource =
5415 container_of(entry, typeof(*resource), entry);
5416 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
5418 key_len += ref->actions_num * sizeof(ref->actions[0]);
5419 return ref->actions_num != resource->actions_num ||
5420 memcmp(&ref->ft_type, &resource->ft_type, key_len);
5423 static struct mlx5_indexed_pool *
5424 flow_dv_modify_ipool_get(struct mlx5_dev_ctx_shared *sh, uint8_t index)
5426 struct mlx5_indexed_pool *ipool = __atomic_load_n
5427 (&sh->mdh_ipools[index], __ATOMIC_SEQ_CST);
5430 struct mlx5_indexed_pool *expected = NULL;
5431 struct mlx5_indexed_pool_config cfg =
5432 (struct mlx5_indexed_pool_config) {
5433 .size = sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
5435 sizeof(struct mlx5_modification_cmd),
5440 .release_mem_en = !!sh->config.reclaim_mode,
5442 sh->config.reclaim_mode ? 0 : (1 << 16),
5443 .malloc = mlx5_malloc,
5445 .type = "mlx5_modify_action_resource",
5448 cfg.size = RTE_ALIGN(cfg.size, sizeof(ipool));
5449 ipool = mlx5_ipool_create(&cfg);
5452 if (!__atomic_compare_exchange_n(&sh->mdh_ipools[index],
5453 &expected, ipool, false,
5455 __ATOMIC_SEQ_CST)) {
5456 mlx5_ipool_destroy(ipool);
5457 ipool = __atomic_load_n(&sh->mdh_ipools[index],
5464 struct mlx5_list_entry *
5465 flow_dv_modify_create_cb(void *tool_ctx, void *cb_ctx)
5467 struct mlx5_dev_ctx_shared *sh = tool_ctx;
5468 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
5469 struct mlx5dv_dr_domain *ns;
5470 struct mlx5_flow_dv_modify_hdr_resource *entry;
5471 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
5472 struct mlx5_indexed_pool *ipool = flow_dv_modify_ipool_get(sh,
5473 ref->actions_num - 1);
5475 uint32_t data_len = ref->actions_num * sizeof(ref->actions[0]);
5476 uint32_t key_len = sizeof(*ref) - offsetof(typeof(*ref), ft_type);
5479 if (unlikely(!ipool)) {
5480 rte_flow_error_set(ctx->error, ENOMEM,
5481 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5482 NULL, "cannot allocate modify ipool");
5485 entry = mlx5_ipool_zmalloc(ipool, &idx);
5487 rte_flow_error_set(ctx->error, ENOMEM,
5488 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5489 "cannot allocate resource memory");
5492 rte_memcpy(&entry->ft_type,
5493 RTE_PTR_ADD(ref, offsetof(typeof(*ref), ft_type)),
5494 key_len + data_len);
5495 if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
5496 ns = sh->fdb_domain;
5497 else if (entry->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
5501 ret = mlx5_flow_os_create_flow_action_modify_header
5502 (sh->cdev->ctx, ns, entry,
5503 data_len, &entry->action);
5505 mlx5_ipool_free(sh->mdh_ipools[ref->actions_num - 1], idx);
5506 rte_flow_error_set(ctx->error, ENOMEM,
5507 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5508 NULL, "cannot create modification action");
5512 return &entry->entry;
5515 struct mlx5_list_entry *
5516 flow_dv_modify_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
5519 struct mlx5_dev_ctx_shared *sh = tool_ctx;
5520 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
5521 struct mlx5_flow_dv_modify_hdr_resource *entry;
5522 struct mlx5_flow_dv_modify_hdr_resource *ref = ctx->data;
5523 uint32_t data_len = ref->actions_num * sizeof(ref->actions[0]);
5526 entry = mlx5_ipool_malloc(sh->mdh_ipools[ref->actions_num - 1],
5529 rte_flow_error_set(ctx->error, ENOMEM,
5530 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5531 "cannot allocate resource memory");
5534 memcpy(entry, oentry, sizeof(*entry) + data_len);
5536 return &entry->entry;
5540 flow_dv_modify_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
5542 struct mlx5_dev_ctx_shared *sh = tool_ctx;
5543 struct mlx5_flow_dv_modify_hdr_resource *res =
5544 container_of(entry, typeof(*res), entry);
5546 mlx5_ipool_free(sh->mdh_ipools[res->actions_num - 1], res->idx);
5550 * Validate the sample action.
5552 * @param[in, out] action_flags
5553 * Holds the actions detected until now.
5555 * Pointer to the sample action.
5557 * Pointer to the Ethernet device structure.
5559 * Attributes of flow that includes this action.
5560 * @param[in] item_flags
5561 * Holds the items detected.
5563 * Pointer to the RSS action.
5564 * @param[out] sample_rss
5565 * Pointer to the RSS action in sample action list.
5567 * Pointer to the COUNT action in sample action list.
5568 * @param[out] fdb_mirror_limit
5569 * Pointer to the FDB mirror limitation flag.
5571 * Pointer to error structure.
5574 * 0 on success, a negative errno value otherwise and rte_errno is set.
5577 flow_dv_validate_action_sample(uint64_t *action_flags,
5578 const struct rte_flow_action *action,
5579 struct rte_eth_dev *dev,
5580 const struct rte_flow_attr *attr,
5581 uint64_t item_flags,
5582 const struct rte_flow_action_rss *rss,
5583 const struct rte_flow_action_rss **sample_rss,
5584 const struct rte_flow_action_count **count,
5585 int *fdb_mirror_limit,
5586 struct rte_flow_error *error)
5588 struct mlx5_priv *priv = dev->data->dev_private;
5589 struct mlx5_sh_config *dev_conf = &priv->sh->config;
5590 const struct rte_flow_action_sample *sample = action->conf;
5591 const struct rte_flow_action *act;
5592 uint64_t sub_action_flags = 0;
5593 uint16_t queue_index = 0xFFFF;
5598 return rte_flow_error_set(error, EINVAL,
5599 RTE_FLOW_ERROR_TYPE_ACTION, action,
5600 "configuration cannot be NULL");
5601 if (sample->ratio == 0)
5602 return rte_flow_error_set(error, EINVAL,
5603 RTE_FLOW_ERROR_TYPE_ACTION, action,
5604 "ratio value starts from 1");
5605 if (!priv->sh->cdev->config.devx ||
5606 (sample->ratio > 0 && !priv->sampler_en))
5607 return rte_flow_error_set(error, ENOTSUP,
5608 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5610 "sample action not supported");
5611 if (*action_flags & MLX5_FLOW_ACTION_SAMPLE)
5612 return rte_flow_error_set(error, EINVAL,
5613 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5614 "Multiple sample actions not "
5616 if (*action_flags & MLX5_FLOW_ACTION_METER)
5617 return rte_flow_error_set(error, EINVAL,
5618 RTE_FLOW_ERROR_TYPE_ACTION, action,
5619 "wrong action order, meter should "
5620 "be after sample action");
5621 if (*action_flags & MLX5_FLOW_ACTION_JUMP)
5622 return rte_flow_error_set(error, EINVAL,
5623 RTE_FLOW_ERROR_TYPE_ACTION, action,
5624 "wrong action order, jump should "
5625 "be after sample action");
5626 if (*action_flags & MLX5_FLOW_ACTION_CT)
5627 return rte_flow_error_set(error, EINVAL,
5628 RTE_FLOW_ERROR_TYPE_ACTION, action,
5629 "Sample after CT not supported");
5630 act = sample->actions;
5631 for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
5632 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5633 return rte_flow_error_set(error, ENOTSUP,
5634 RTE_FLOW_ERROR_TYPE_ACTION,
5635 act, "too many actions");
5636 switch (act->type) {
5637 case RTE_FLOW_ACTION_TYPE_QUEUE:
5638 ret = mlx5_flow_validate_action_queue(act,
5644 queue_index = ((const struct rte_flow_action_queue *)
5645 (act->conf))->index;
5646 sub_action_flags |= MLX5_FLOW_ACTION_QUEUE;
5649 case RTE_FLOW_ACTION_TYPE_RSS:
5650 *sample_rss = act->conf;
5651 ret = mlx5_flow_validate_action_rss(act,
5658 if (rss && *sample_rss &&
5659 ((*sample_rss)->level != rss->level ||
5660 (*sample_rss)->types != rss->types))
5661 return rte_flow_error_set(error, ENOTSUP,
5662 RTE_FLOW_ERROR_TYPE_ACTION,
5664 "Can't use the different RSS types "
5665 "or level in the same flow");
5666 if (*sample_rss != NULL && (*sample_rss)->queue_num)
5667 queue_index = (*sample_rss)->queue[0];
5668 sub_action_flags |= MLX5_FLOW_ACTION_RSS;
5671 case RTE_FLOW_ACTION_TYPE_MARK:
5672 ret = flow_dv_validate_action_mark(dev, act,
5677 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
5678 sub_action_flags |= MLX5_FLOW_ACTION_MARK |
5679 MLX5_FLOW_ACTION_MARK_EXT;
5681 sub_action_flags |= MLX5_FLOW_ACTION_MARK;
5684 case RTE_FLOW_ACTION_TYPE_COUNT:
5685 ret = flow_dv_validate_action_count
5686 (dev, false, *action_flags | sub_action_flags,
5691 sub_action_flags |= MLX5_FLOW_ACTION_COUNT;
5692 *action_flags |= MLX5_FLOW_ACTION_COUNT;
5695 case RTE_FLOW_ACTION_TYPE_PORT_ID:
5696 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
5697 ret = flow_dv_validate_action_port_id(dev,
5704 sub_action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5707 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5708 ret = flow_dv_validate_action_raw_encap_decap
5709 (dev, NULL, act->conf, attr, &sub_action_flags,
5710 &actions_n, action, item_flags, error);
5715 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5716 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5717 ret = flow_dv_validate_action_l2_encap(dev,
5723 sub_action_flags |= MLX5_FLOW_ACTION_ENCAP;
5727 return rte_flow_error_set(error, ENOTSUP,
5728 RTE_FLOW_ERROR_TYPE_ACTION,
5730 "Doesn't support optional "
5734 if (attr->ingress && !attr->transfer) {
5735 if (!(sub_action_flags & (MLX5_FLOW_ACTION_QUEUE |
5736 MLX5_FLOW_ACTION_RSS)))
5737 return rte_flow_error_set(error, EINVAL,
5738 RTE_FLOW_ERROR_TYPE_ACTION,
5740 "Ingress must has a dest "
5741 "QUEUE for Sample");
5742 } else if (attr->egress && !attr->transfer) {
5743 return rte_flow_error_set(error, ENOTSUP,
5744 RTE_FLOW_ERROR_TYPE_ACTION,
5746 "Sample Only support Ingress "
5748 } else if (sample->actions->type != RTE_FLOW_ACTION_TYPE_END) {
5749 MLX5_ASSERT(attr->transfer);
5750 if (sample->ratio > 1)
5751 return rte_flow_error_set(error, ENOTSUP,
5752 RTE_FLOW_ERROR_TYPE_ACTION,
5754 "E-Switch doesn't support "
5755 "any optional action "
5757 if (sub_action_flags & MLX5_FLOW_ACTION_QUEUE)
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_RSS)
5763 return rte_flow_error_set(error, ENOTSUP,
5764 RTE_FLOW_ERROR_TYPE_ACTION,
5766 "unsupported action QUEUE");
5767 if (!(sub_action_flags & MLX5_FLOW_ACTION_PORT_ID))
5768 return rte_flow_error_set(error, EINVAL,
5769 RTE_FLOW_ERROR_TYPE_ACTION,
5771 "E-Switch must has a dest "
5772 "port for mirroring");
5773 if (!priv->sh->cdev->config.hca_attr.reg_c_preserve &&
5774 priv->representor_id != UINT16_MAX)
5775 *fdb_mirror_limit = 1;
5777 /* Continue validation for Xcap actions.*/
5778 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) &&
5779 (queue_index == 0xFFFF || !mlx5_rxq_is_hairpin(dev, queue_index))) {
5780 if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
5781 MLX5_FLOW_XCAP_ACTIONS)
5782 return rte_flow_error_set(error, ENOTSUP,
5783 RTE_FLOW_ERROR_TYPE_ACTION,
5784 NULL, "encap and decap "
5785 "combination aren't "
5787 if (!attr->transfer && attr->ingress && (sub_action_flags &
5788 MLX5_FLOW_ACTION_ENCAP))
5789 return rte_flow_error_set(error, ENOTSUP,
5790 RTE_FLOW_ERROR_TYPE_ACTION,
5791 NULL, "encap is not supported"
5792 " for ingress traffic");
5798 * Find existing modify-header resource or create and register a new one.
5800 * @param dev[in, out]
5801 * Pointer to rte_eth_dev structure.
5802 * @param[in, out] resource
5803 * Pointer to modify-header resource.
5804 * @parm[in, out] dev_flow
5805 * Pointer to the dev_flow.
5807 * pointer to error structure.
5810 * 0 on success otherwise -errno and errno is set.
5813 flow_dv_modify_hdr_resource_register
5814 (struct rte_eth_dev *dev,
5815 struct mlx5_flow_dv_modify_hdr_resource *resource,
5816 struct mlx5_flow *dev_flow,
5817 struct rte_flow_error *error)
5819 struct mlx5_priv *priv = dev->data->dev_private;
5820 struct mlx5_dev_ctx_shared *sh = priv->sh;
5821 uint32_t key_len = sizeof(*resource) -
5822 offsetof(typeof(*resource), ft_type) +
5823 resource->actions_num * sizeof(resource->actions[0]);
5824 struct mlx5_list_entry *entry;
5825 struct mlx5_flow_cb_ctx ctx = {
5829 struct mlx5_hlist *modify_cmds;
5832 modify_cmds = flow_dv_hlist_prepare(sh, &sh->modify_cmds,
5834 MLX5_FLOW_HDR_MODIFY_HTABLE_SZ,
5836 flow_dv_modify_create_cb,
5837 flow_dv_modify_match_cb,
5838 flow_dv_modify_remove_cb,
5839 flow_dv_modify_clone_cb,
5840 flow_dv_modify_clone_free_cb,
5842 if (unlikely(!modify_cmds))
5844 resource->root = !dev_flow->dv.group;
5845 if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
5847 return rte_flow_error_set(error, EOVERFLOW,
5848 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5849 "too many modify header items");
5850 key64 = __rte_raw_cksum(&resource->ft_type, key_len, 0);
5851 entry = mlx5_hlist_register(modify_cmds, key64, &ctx);
5854 resource = container_of(entry, typeof(*resource), entry);
5855 dev_flow->handle->dvh.modify_hdr = resource;
5860 * Get DV flow counter by index.
5863 * Pointer to the Ethernet device structure.
5865 * mlx5 flow counter index in the container.
5867 * mlx5 flow counter pool in the container.
5870 * Pointer to the counter, NULL otherwise.
5872 static struct mlx5_flow_counter *
5873 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
5875 struct mlx5_flow_counter_pool **ppool)
5877 struct mlx5_priv *priv = dev->data->dev_private;
5878 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
5879 struct mlx5_flow_counter_pool *pool;
5881 /* Decrease to original index and clear shared bit. */
5882 idx = (idx - 1) & (MLX5_CNT_SHARED_OFFSET - 1);
5883 MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cmng->n);
5884 pool = cmng->pools[idx / MLX5_COUNTERS_PER_POOL];
5888 return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
5892 * Check the devx counter belongs to the pool.
5895 * Pointer to the counter pool.
5897 * The counter devx ID.
5900 * True if counter belongs to the pool, false otherwise.
5903 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
5905 int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
5906 MLX5_COUNTERS_PER_POOL;
5908 if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
5914 * Get a pool by devx counter ID.
5917 * Pointer to the counter management.
5919 * The counter devx ID.
5922 * The counter pool pointer if exists, NULL otherwise,
5924 static struct mlx5_flow_counter_pool *
5925 flow_dv_find_pool_by_id(struct mlx5_flow_counter_mng *cmng, int id)
5928 struct mlx5_flow_counter_pool *pool = NULL;
5930 rte_spinlock_lock(&cmng->pool_update_sl);
5931 /* Check last used pool. */
5932 if (cmng->last_pool_idx != POOL_IDX_INVALID &&
5933 flow_dv_is_counter_in_pool(cmng->pools[cmng->last_pool_idx], id)) {
5934 pool = cmng->pools[cmng->last_pool_idx];
5937 /* ID out of range means no suitable pool in the container. */
5938 if (id > cmng->max_id || id < cmng->min_id)
5941 * Find the pool from the end of the container, since mostly counter
5942 * ID is sequence increasing, and the last pool should be the needed
5947 struct mlx5_flow_counter_pool *pool_tmp = cmng->pools[i];
5949 if (flow_dv_is_counter_in_pool(pool_tmp, id)) {
5955 rte_spinlock_unlock(&cmng->pool_update_sl);
5960 * Resize a counter container.
5963 * Pointer to the Ethernet device structure.
5966 * 0 on success, otherwise negative errno value and rte_errno is set.
5969 flow_dv_container_resize(struct rte_eth_dev *dev)
5971 struct mlx5_priv *priv = dev->data->dev_private;
5972 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
5973 void *old_pools = cmng->pools;
5974 uint32_t resize = cmng->n + MLX5_CNT_CONTAINER_RESIZE;
5975 uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
5976 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
5983 memcpy(pools, old_pools, cmng->n *
5984 sizeof(struct mlx5_flow_counter_pool *));
5986 cmng->pools = pools;
5988 mlx5_free(old_pools);
5993 * Query a devx flow counter.
5996 * Pointer to the Ethernet device structure.
5997 * @param[in] counter
5998 * Index to the flow counter.
6000 * The statistics value of packets.
6002 * The statistics value of bytes.
6005 * 0 on success, otherwise a negative errno value and rte_errno is set.
6008 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
6011 struct mlx5_priv *priv = dev->data->dev_private;
6012 struct mlx5_flow_counter_pool *pool = NULL;
6013 struct mlx5_flow_counter *cnt;
6016 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
6018 if (priv->sh->cmng.counter_fallback)
6019 return mlx5_devx_cmd_flow_counter_query(cnt->dcs_when_active, 0,
6020 0, pkts, bytes, 0, NULL, NULL, 0);
6021 rte_spinlock_lock(&pool->sl);
6026 offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
6027 *pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
6028 *bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
6030 rte_spinlock_unlock(&pool->sl);
6035 * Create and initialize a new counter pool.
6038 * Pointer to the Ethernet device structure.
6040 * The devX counter handle.
6042 * Whether the pool is for counter that was allocated for aging.
6043 * @param[in/out] cont_cur
6044 * Pointer to the container pointer, it will be update in pool resize.
6047 * The pool container pointer on success, NULL otherwise and rte_errno is set.
6049 static struct mlx5_flow_counter_pool *
6050 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
6053 struct mlx5_priv *priv = dev->data->dev_private;
6054 struct mlx5_flow_counter_pool *pool;
6055 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
6056 bool fallback = priv->sh->cmng.counter_fallback;
6057 uint32_t size = sizeof(*pool);
6059 size += MLX5_COUNTERS_PER_POOL * MLX5_CNT_SIZE;
6060 size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * MLX5_AGE_SIZE);
6061 pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
6067 pool->is_aged = !!age;
6068 pool->query_gen = 0;
6069 pool->min_dcs = dcs;
6070 rte_spinlock_init(&pool->sl);
6071 rte_spinlock_init(&pool->csl);
6072 TAILQ_INIT(&pool->counters[0]);
6073 TAILQ_INIT(&pool->counters[1]);
6074 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
6075 rte_spinlock_lock(&cmng->pool_update_sl);
6076 pool->index = cmng->n_valid;
6077 if (pool->index == cmng->n && flow_dv_container_resize(dev)) {
6079 rte_spinlock_unlock(&cmng->pool_update_sl);
6082 cmng->pools[pool->index] = pool;
6084 if (unlikely(fallback)) {
6085 int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
6087 if (base < cmng->min_id)
6088 cmng->min_id = base;
6089 if (base > cmng->max_id)
6090 cmng->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
6091 cmng->last_pool_idx = pool->index;
6093 rte_spinlock_unlock(&cmng->pool_update_sl);
6098 * Prepare a new counter and/or a new counter pool.
6101 * Pointer to the Ethernet device structure.
6102 * @param[out] cnt_free
6103 * Where to put the pointer of a new counter.
6105 * Whether the pool is for counter that was allocated for aging.
6108 * The counter pool pointer and @p cnt_free is set on success,
6109 * NULL otherwise and rte_errno is set.
6111 static struct mlx5_flow_counter_pool *
6112 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
6113 struct mlx5_flow_counter **cnt_free,
6116 struct mlx5_priv *priv = dev->data->dev_private;
6117 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
6118 struct mlx5_flow_counter_pool *pool;
6119 struct mlx5_counters tmp_tq;
6120 struct mlx5_devx_obj *dcs = NULL;
6121 struct mlx5_flow_counter *cnt;
6122 enum mlx5_counter_type cnt_type =
6123 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
6124 bool fallback = priv->sh->cmng.counter_fallback;
6128 /* bulk_bitmap must be 0 for single counter allocation. */
6129 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->cdev->ctx, 0);
6132 pool = flow_dv_find_pool_by_id(cmng, dcs->id);
6134 pool = flow_dv_pool_create(dev, dcs, age);
6136 mlx5_devx_cmd_destroy(dcs);
6140 i = dcs->id % MLX5_COUNTERS_PER_POOL;
6141 cnt = MLX5_POOL_GET_CNT(pool, i);
6143 cnt->dcs_when_free = dcs;
6147 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->cdev->ctx, 0x4);
6149 rte_errno = ENODATA;
6152 pool = flow_dv_pool_create(dev, dcs, age);
6154 mlx5_devx_cmd_destroy(dcs);
6157 TAILQ_INIT(&tmp_tq);
6158 for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
6159 cnt = MLX5_POOL_GET_CNT(pool, i);
6161 TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
6163 rte_spinlock_lock(&cmng->csl[cnt_type]);
6164 TAILQ_CONCAT(&cmng->counters[cnt_type], &tmp_tq, next);
6165 rte_spinlock_unlock(&cmng->csl[cnt_type]);
6166 *cnt_free = MLX5_POOL_GET_CNT(pool, 0);
6167 (*cnt_free)->pool = pool;
6172 * Allocate a flow counter.
6175 * Pointer to the Ethernet device structure.
6177 * Whether the counter was allocated for aging.
6180 * Index to flow counter on success, 0 otherwise and rte_errno is set.
6183 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t age)
6185 struct mlx5_priv *priv = dev->data->dev_private;
6186 struct mlx5_flow_counter_pool *pool = NULL;
6187 struct mlx5_flow_counter *cnt_free = NULL;
6188 bool fallback = priv->sh->cmng.counter_fallback;
6189 struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
6190 enum mlx5_counter_type cnt_type =
6191 age ? MLX5_COUNTER_TYPE_AGE : MLX5_COUNTER_TYPE_ORIGIN;
6194 if (!priv->sh->cdev->config.devx) {
6195 rte_errno = ENOTSUP;
6198 /* Get free counters from container. */
6199 rte_spinlock_lock(&cmng->csl[cnt_type]);
6200 cnt_free = TAILQ_FIRST(&cmng->counters[cnt_type]);
6202 TAILQ_REMOVE(&cmng->counters[cnt_type], cnt_free, next);
6203 rte_spinlock_unlock(&cmng->csl[cnt_type]);
6204 if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free, age))
6206 pool = cnt_free->pool;
6208 cnt_free->dcs_when_active = cnt_free->dcs_when_free;
6209 /* Create a DV counter action only in the first time usage. */
6210 if (!cnt_free->action) {
6212 struct mlx5_devx_obj *dcs;
6216 offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
6217 dcs = pool->min_dcs;
6220 dcs = cnt_free->dcs_when_free;
6222 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
6229 cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
6230 MLX5_CNT_ARRAY_IDX(pool, cnt_free));
6231 /* Update the counter reset values. */
6232 if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
6235 if (!fallback && !priv->sh->cmng.query_thread_on)
6236 /* Start the asynchronous batch query by the host thread. */
6237 mlx5_set_query_alarm(priv->sh);
6239 * When the count action isn't shared (by ID), shared_info field is
6240 * used for indirect action API's refcnt.
6241 * When the counter action is not shared neither by ID nor by indirect
6242 * action API, shared info must be 1.
6244 cnt_free->shared_info.refcnt = 1;
6248 cnt_free->pool = pool;
6250 cnt_free->dcs_when_free = cnt_free->dcs_when_active;
6251 rte_spinlock_lock(&cmng->csl[cnt_type]);
6252 TAILQ_INSERT_TAIL(&cmng->counters[cnt_type], cnt_free, next);
6253 rte_spinlock_unlock(&cmng->csl[cnt_type]);
6259 * Get age param from counter index.
6262 * Pointer to the Ethernet device structure.
6263 * @param[in] counter
6264 * Index to the counter handler.
6267 * The aging parameter specified for the counter index.
6269 static struct mlx5_age_param*
6270 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
6273 struct mlx5_flow_counter *cnt;
6274 struct mlx5_flow_counter_pool *pool = NULL;
6276 flow_dv_counter_get_by_idx(dev, counter, &pool);
6277 counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
6278 cnt = MLX5_POOL_GET_CNT(pool, counter);
6279 return MLX5_CNT_TO_AGE(cnt);
6283 * Remove a flow counter from aged counter list.
6286 * Pointer to the Ethernet device structure.
6287 * @param[in] counter
6288 * Index to the counter handler.
6290 * Pointer to the counter handler.
6293 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
6294 uint32_t counter, struct mlx5_flow_counter *cnt)
6296 struct mlx5_age_info *age_info;
6297 struct mlx5_age_param *age_param;
6298 struct mlx5_priv *priv = dev->data->dev_private;
6299 uint16_t expected = AGE_CANDIDATE;
6301 age_info = GET_PORT_AGE_INFO(priv);
6302 age_param = flow_dv_counter_idx_get_age(dev, counter);
6303 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
6304 AGE_FREE, false, __ATOMIC_RELAXED,
6305 __ATOMIC_RELAXED)) {
6307 * We need the lock even it is age timeout,
6308 * since counter may still in process.
6310 rte_spinlock_lock(&age_info->aged_sl);
6311 TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
6312 rte_spinlock_unlock(&age_info->aged_sl);
6313 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
6318 * Release a flow counter.
6321 * Pointer to the Ethernet device structure.
6322 * @param[in] counter
6323 * Index to the counter handler.
6326 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t counter)
6328 struct mlx5_priv *priv = dev->data->dev_private;
6329 struct mlx5_flow_counter_pool *pool = NULL;
6330 struct mlx5_flow_counter *cnt;
6331 enum mlx5_counter_type cnt_type;
6335 cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
6337 if (pool->is_aged) {
6338 flow_dv_counter_remove_from_age(dev, counter, cnt);
6341 * If the counter action is shared by indirect action API,
6342 * the atomic function reduces its references counter.
6343 * If after the reduction the action is still referenced, the
6344 * function returns here and does not release it.
6345 * When the counter action is not shared by
6346 * indirect action API, shared info is 1 before the reduction,
6347 * so this condition is failed and function doesn't return here.
6349 if (__atomic_sub_fetch(&cnt->shared_info.refcnt, 1,
6355 * Put the counter back to list to be updated in none fallback mode.
6356 * Currently, we are using two list alternately, while one is in query,
6357 * add the freed counter to the other list based on the pool query_gen
6358 * value. After query finishes, add counter the list to the global
6359 * container counter list. The list changes while query starts. In
6360 * this case, lock will not be needed as query callback and release
6361 * function both operate with the different list.
6363 if (!priv->sh->cmng.counter_fallback) {
6364 rte_spinlock_lock(&pool->csl);
6365 TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
6366 rte_spinlock_unlock(&pool->csl);
6368 cnt->dcs_when_free = cnt->dcs_when_active;
6369 cnt_type = pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6370 MLX5_COUNTER_TYPE_ORIGIN;
6371 rte_spinlock_lock(&priv->sh->cmng.csl[cnt_type]);
6372 TAILQ_INSERT_TAIL(&priv->sh->cmng.counters[cnt_type],
6374 rte_spinlock_unlock(&priv->sh->cmng.csl[cnt_type]);
6379 * Resize a meter id container.
6382 * Pointer to the Ethernet device structure.
6385 * 0 on success, otherwise negative errno value and rte_errno is set.
6388 flow_dv_mtr_container_resize(struct rte_eth_dev *dev)
6390 struct mlx5_priv *priv = dev->data->dev_private;
6391 struct mlx5_aso_mtr_pools_mng *pools_mng =
6392 &priv->sh->mtrmng->pools_mng;
6393 void *old_pools = pools_mng->pools;
6394 uint32_t resize = pools_mng->n + MLX5_MTRS_CONTAINER_RESIZE;
6395 uint32_t mem_size = sizeof(struct mlx5_aso_mtr_pool *) * resize;
6396 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
6403 if (mlx5_aso_queue_init(priv->sh, ASO_OPC_MOD_POLICER)) {
6408 memcpy(pools, old_pools, pools_mng->n *
6409 sizeof(struct mlx5_aso_mtr_pool *));
6410 pools_mng->n = resize;
6411 pools_mng->pools = pools;
6413 mlx5_free(old_pools);
6418 * Prepare a new meter and/or a new meter pool.
6421 * Pointer to the Ethernet device structure.
6422 * @param[out] mtr_free
6423 * Where to put the pointer of a new meter.g.
6426 * The meter pool pointer and @mtr_free is set on success,
6427 * NULL otherwise and rte_errno is set.
6429 static struct mlx5_aso_mtr_pool *
6430 flow_dv_mtr_pool_create(struct rte_eth_dev *dev, struct mlx5_aso_mtr **mtr_free)
6432 struct mlx5_priv *priv = dev->data->dev_private;
6433 struct mlx5_aso_mtr_pools_mng *pools_mng = &priv->sh->mtrmng->pools_mng;
6434 struct mlx5_aso_mtr_pool *pool = NULL;
6435 struct mlx5_devx_obj *dcs = NULL;
6437 uint32_t log_obj_size;
6439 log_obj_size = rte_log2_u32(MLX5_ASO_MTRS_PER_POOL >> 1);
6440 dcs = mlx5_devx_cmd_create_flow_meter_aso_obj(priv->sh->cdev->ctx,
6441 priv->sh->cdev->pdn,
6444 rte_errno = ENODATA;
6447 pool = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*pool), 0, SOCKET_ID_ANY);
6450 claim_zero(mlx5_devx_cmd_destroy(dcs));
6453 pool->devx_obj = dcs;
6454 rte_rwlock_write_lock(&pools_mng->resize_mtrwl);
6455 pool->index = pools_mng->n_valid;
6456 if (pool->index == pools_mng->n && flow_dv_mtr_container_resize(dev)) {
6458 claim_zero(mlx5_devx_cmd_destroy(dcs));
6459 rte_rwlock_write_unlock(&pools_mng->resize_mtrwl);
6462 pools_mng->pools[pool->index] = pool;
6463 pools_mng->n_valid++;
6464 rte_rwlock_write_unlock(&pools_mng->resize_mtrwl);
6465 for (i = 1; i < MLX5_ASO_MTRS_PER_POOL; ++i) {
6466 pool->mtrs[i].offset = i;
6467 LIST_INSERT_HEAD(&pools_mng->meters, &pool->mtrs[i], next);
6469 pool->mtrs[0].offset = 0;
6470 *mtr_free = &pool->mtrs[0];
6475 * Release a flow meter into pool.
6478 * Pointer to the Ethernet device structure.
6479 * @param[in] mtr_idx
6480 * Index to aso flow meter.
6483 flow_dv_aso_mtr_release_to_pool(struct rte_eth_dev *dev, uint32_t mtr_idx)
6485 struct mlx5_priv *priv = dev->data->dev_private;
6486 struct mlx5_aso_mtr_pools_mng *pools_mng =
6487 &priv->sh->mtrmng->pools_mng;
6488 struct mlx5_aso_mtr *aso_mtr = mlx5_aso_meter_by_idx(priv, mtr_idx);
6490 MLX5_ASSERT(aso_mtr);
6491 rte_spinlock_lock(&pools_mng->mtrsl);
6492 memset(&aso_mtr->fm, 0, sizeof(struct mlx5_flow_meter_info));
6493 aso_mtr->state = ASO_METER_FREE;
6494 LIST_INSERT_HEAD(&pools_mng->meters, aso_mtr, next);
6495 rte_spinlock_unlock(&pools_mng->mtrsl);
6499 * Allocate a aso flow meter.
6502 * Pointer to the Ethernet device structure.
6505 * Index to aso flow meter on success, 0 otherwise and rte_errno is set.
6508 flow_dv_mtr_alloc(struct rte_eth_dev *dev)
6510 struct mlx5_priv *priv = dev->data->dev_private;
6511 struct mlx5_aso_mtr *mtr_free = NULL;
6512 struct mlx5_aso_mtr_pools_mng *pools_mng =
6513 &priv->sh->mtrmng->pools_mng;
6514 struct mlx5_aso_mtr_pool *pool;
6515 uint32_t mtr_idx = 0;
6517 if (!priv->sh->cdev->config.devx) {
6518 rte_errno = ENOTSUP;
6521 /* Allocate the flow meter memory. */
6522 /* Get free meters from management. */
6523 rte_spinlock_lock(&pools_mng->mtrsl);
6524 mtr_free = LIST_FIRST(&pools_mng->meters);
6526 LIST_REMOVE(mtr_free, next);
6527 if (!mtr_free && !flow_dv_mtr_pool_create(dev, &mtr_free)) {
6528 rte_spinlock_unlock(&pools_mng->mtrsl);
6531 mtr_free->state = ASO_METER_WAIT;
6532 rte_spinlock_unlock(&pools_mng->mtrsl);
6533 pool = container_of(mtr_free,
6534 struct mlx5_aso_mtr_pool,
6535 mtrs[mtr_free->offset]);
6536 mtr_idx = MLX5_MAKE_MTR_IDX(pool->index, mtr_free->offset);
6537 if (!mtr_free->fm.meter_action) {
6538 #ifdef HAVE_MLX5_DR_CREATE_ACTION_ASO
6539 struct rte_flow_error error;
6542 reg_id = mlx5_flow_get_reg_id(dev, MLX5_MTR_COLOR, 0, &error);
6543 mtr_free->fm.meter_action =
6544 mlx5_glue->dv_create_flow_action_aso
6545 (priv->sh->rx_domain,
6546 pool->devx_obj->obj,
6548 (1 << MLX5_FLOW_COLOR_GREEN),
6550 #endif /* HAVE_MLX5_DR_CREATE_ACTION_ASO */
6551 if (!mtr_free->fm.meter_action) {
6552 flow_dv_aso_mtr_release_to_pool(dev, mtr_idx);
6560 * Verify the @p attributes will be correctly understood by the NIC and store
6561 * them in the @p flow if everything is correct.
6564 * Pointer to dev struct.
6565 * @param[in] attributes
6566 * Pointer to flow attributes
6567 * @param[in] external
6568 * This flow rule is created by request external to PMD.
6570 * Pointer to error structure.
6573 * - 0 on success and non root table.
6574 * - 1 on success and root table.
6575 * - a negative errno value otherwise and rte_errno is set.
6578 flow_dv_validate_attributes(struct rte_eth_dev *dev,
6579 const struct mlx5_flow_tunnel *tunnel,
6580 const struct rte_flow_attr *attributes,
6581 const struct flow_grp_info *grp_info,
6582 struct rte_flow_error *error)
6584 struct mlx5_priv *priv = dev->data->dev_private;
6585 uint32_t lowest_priority = mlx5_get_lowest_priority(dev, attributes);
6588 #ifndef HAVE_MLX5DV_DR
6589 RTE_SET_USED(tunnel);
6590 RTE_SET_USED(grp_info);
6591 if (attributes->group)
6592 return rte_flow_error_set(error, ENOTSUP,
6593 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6595 "groups are not supported");
6599 ret = mlx5_flow_group_to_table(dev, tunnel, attributes->group, &table,
6604 ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
6606 if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
6607 attributes->priority > lowest_priority)
6608 return rte_flow_error_set(error, ENOTSUP,
6609 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
6611 "priority out of range");
6612 if (attributes->transfer) {
6613 if (!priv->sh->config.dv_esw_en)
6614 return rte_flow_error_set
6616 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6617 "E-Switch dr is not supported");
6618 if (attributes->egress)
6619 return rte_flow_error_set
6621 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
6622 "egress is not supported");
6624 if (!(attributes->egress ^ attributes->ingress))
6625 return rte_flow_error_set(error, ENOTSUP,
6626 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
6627 "must specify exactly one of "
6628 "ingress or egress");
6633 validate_integrity_bits(const struct rte_flow_item_integrity *mask,
6634 int64_t pattern_flags, uint64_t l3_flags,
6635 uint64_t l4_flags, uint64_t ip4_flag,
6636 struct rte_flow_error *error)
6638 if (mask->l3_ok && !(pattern_flags & l3_flags))
6639 return rte_flow_error_set(error, EINVAL,
6640 RTE_FLOW_ERROR_TYPE_ITEM,
6641 NULL, "missing L3 protocol");
6643 if (mask->ipv4_csum_ok && !(pattern_flags & ip4_flag))
6644 return rte_flow_error_set(error, EINVAL,
6645 RTE_FLOW_ERROR_TYPE_ITEM,
6646 NULL, "missing IPv4 protocol");
6648 if ((mask->l4_ok || mask->l4_csum_ok) && !(pattern_flags & l4_flags))
6649 return rte_flow_error_set(error, EINVAL,
6650 RTE_FLOW_ERROR_TYPE_ITEM,
6651 NULL, "missing L4 protocol");
6657 flow_dv_validate_item_integrity_post(const struct
6658 rte_flow_item *integrity_items[2],
6659 int64_t pattern_flags,
6660 struct rte_flow_error *error)
6662 const struct rte_flow_item_integrity *mask;
6665 if (pattern_flags & MLX5_FLOW_ITEM_OUTER_INTEGRITY) {
6666 mask = (typeof(mask))integrity_items[0]->mask;
6667 ret = validate_integrity_bits(mask, pattern_flags,
6668 MLX5_FLOW_LAYER_OUTER_L3,
6669 MLX5_FLOW_LAYER_OUTER_L4,
6670 MLX5_FLOW_LAYER_OUTER_L3_IPV4,
6675 if (pattern_flags & MLX5_FLOW_ITEM_INNER_INTEGRITY) {
6676 mask = (typeof(mask))integrity_items[1]->mask;
6677 ret = validate_integrity_bits(mask, pattern_flags,
6678 MLX5_FLOW_LAYER_INNER_L3,
6679 MLX5_FLOW_LAYER_INNER_L4,
6680 MLX5_FLOW_LAYER_INNER_L3_IPV4,
6689 flow_dv_validate_item_integrity(struct rte_eth_dev *dev,
6690 const struct rte_flow_item *integrity_item,
6691 uint64_t pattern_flags, uint64_t *last_item,
6692 const struct rte_flow_item *integrity_items[2],
6693 struct rte_flow_error *error)
6695 struct mlx5_priv *priv = dev->data->dev_private;
6696 const struct rte_flow_item_integrity *mask = (typeof(mask))
6697 integrity_item->mask;
6698 const struct rte_flow_item_integrity *spec = (typeof(spec))
6699 integrity_item->spec;
6701 if (!priv->sh->cdev->config.hca_attr.pkt_integrity_match)
6702 return rte_flow_error_set(error, ENOTSUP,
6703 RTE_FLOW_ERROR_TYPE_ITEM,
6705 "packet integrity integrity_item not supported");
6707 return rte_flow_error_set(error, ENOTSUP,
6708 RTE_FLOW_ERROR_TYPE_ITEM,
6710 "no spec for integrity item");
6712 mask = &rte_flow_item_integrity_mask;
6713 if (!mlx5_validate_integrity_item(mask))
6714 return rte_flow_error_set(error, ENOTSUP,
6715 RTE_FLOW_ERROR_TYPE_ITEM,
6717 "unsupported integrity filter");
6718 if (spec->level > 1) {
6719 if (pattern_flags & MLX5_FLOW_ITEM_INNER_INTEGRITY)
6720 return rte_flow_error_set
6722 RTE_FLOW_ERROR_TYPE_ITEM,
6723 NULL, "multiple inner integrity items not supported");
6724 integrity_items[1] = integrity_item;
6725 *last_item |= MLX5_FLOW_ITEM_INNER_INTEGRITY;
6727 if (pattern_flags & MLX5_FLOW_ITEM_OUTER_INTEGRITY)
6728 return rte_flow_error_set
6730 RTE_FLOW_ERROR_TYPE_ITEM,
6731 NULL, "multiple outer integrity items not supported");
6732 integrity_items[0] = integrity_item;
6733 *last_item |= MLX5_FLOW_ITEM_OUTER_INTEGRITY;
6739 flow_dv_validate_item_flex(struct rte_eth_dev *dev,
6740 const struct rte_flow_item *item,
6741 uint64_t item_flags,
6742 uint64_t *last_item,
6744 struct rte_flow_error *error)
6746 const struct rte_flow_item_flex *flow_spec = item->spec;
6747 const struct rte_flow_item_flex *flow_mask = item->mask;
6748 struct mlx5_flex_item *flex;
6751 return rte_flow_error_set(error, EINVAL,
6752 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
6753 "flex flow item spec cannot be NULL");
6755 return rte_flow_error_set(error, EINVAL,
6756 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
6757 "flex flow item mask cannot be NULL");
6759 return rte_flow_error_set(error, ENOTSUP,
6760 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
6761 "flex flow item last not supported");
6762 if (mlx5_flex_acquire_index(dev, flow_spec->handle, false) < 0)
6763 return rte_flow_error_set(error, EINVAL,
6764 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
6765 "invalid flex flow item handle");
6766 flex = (struct mlx5_flex_item *)flow_spec->handle;
6767 switch (flex->tunnel_mode) {
6768 case FLEX_TUNNEL_MODE_SINGLE:
6770 (MLX5_FLOW_ITEM_OUTER_FLEX | MLX5_FLOW_ITEM_INNER_FLEX))
6771 rte_flow_error_set(error, EINVAL,
6772 RTE_FLOW_ERROR_TYPE_ITEM,
6773 NULL, "multiple flex items not supported");
6775 case FLEX_TUNNEL_MODE_OUTER:
6777 rte_flow_error_set(error, EINVAL,
6778 RTE_FLOW_ERROR_TYPE_ITEM,
6779 NULL, "inner flex item was not configured");
6780 if (item_flags & MLX5_FLOW_ITEM_OUTER_FLEX)
6781 rte_flow_error_set(error, ENOTSUP,
6782 RTE_FLOW_ERROR_TYPE_ITEM,
6783 NULL, "multiple flex items not supported");
6785 case FLEX_TUNNEL_MODE_INNER:
6787 rte_flow_error_set(error, EINVAL,
6788 RTE_FLOW_ERROR_TYPE_ITEM,
6789 NULL, "outer flex item was not configured");
6790 if (item_flags & MLX5_FLOW_ITEM_INNER_FLEX)
6791 rte_flow_error_set(error, EINVAL,
6792 RTE_FLOW_ERROR_TYPE_ITEM,
6793 NULL, "multiple flex items not supported");
6795 case FLEX_TUNNEL_MODE_MULTI:
6796 if ((is_inner && (item_flags & MLX5_FLOW_ITEM_INNER_FLEX)) ||
6797 (!is_inner && (item_flags & MLX5_FLOW_ITEM_OUTER_FLEX))) {
6798 rte_flow_error_set(error, EINVAL,
6799 RTE_FLOW_ERROR_TYPE_ITEM,
6800 NULL, "multiple flex items not supported");
6803 case FLEX_TUNNEL_MODE_TUNNEL:
6804 if (is_inner || (item_flags & MLX5_FLOW_ITEM_FLEX_TUNNEL))
6805 rte_flow_error_set(error, EINVAL,
6806 RTE_FLOW_ERROR_TYPE_ITEM,
6807 NULL, "multiple flex tunnel items not supported");
6810 rte_flow_error_set(error, EINVAL,
6811 RTE_FLOW_ERROR_TYPE_ITEM,
6812 NULL, "invalid flex item configuration");
6814 *last_item = flex->tunnel_mode == FLEX_TUNNEL_MODE_TUNNEL ?
6815 MLX5_FLOW_ITEM_FLEX_TUNNEL : is_inner ?
6816 MLX5_FLOW_ITEM_INNER_FLEX : MLX5_FLOW_ITEM_OUTER_FLEX;
6821 * Internal validation function. For validating both actions and items.
6824 * Pointer to the rte_eth_dev structure.
6826 * Pointer to the flow attributes.
6828 * Pointer to the list of items.
6829 * @param[in] actions
6830 * Pointer to the list of actions.
6831 * @param[in] external
6832 * This flow rule is created by request external to PMD.
6833 * @param[in] hairpin
6834 * Number of hairpin TX actions, 0 means classic flow.
6836 * Pointer to the error structure.
6839 * 0 on success, a negative errno value otherwise and rte_errno is set.
6842 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
6843 const struct rte_flow_item items[],
6844 const struct rte_flow_action actions[],
6845 bool external, int hairpin, struct rte_flow_error *error)
6848 uint64_t aso_mask, action_flags = 0;
6849 uint64_t item_flags = 0;
6850 uint64_t last_item = 0;
6851 uint8_t next_protocol = 0xff;
6852 uint16_t ether_type = 0;
6854 uint8_t item_ipv6_proto = 0;
6855 int fdb_mirror_limit = 0;
6856 int modify_after_mirror = 0;
6857 const struct rte_flow_item *geneve_item = NULL;
6858 const struct rte_flow_item *gre_item = NULL;
6859 const struct rte_flow_item *gtp_item = NULL;
6860 const struct rte_flow_action_raw_decap *decap;
6861 const struct rte_flow_action_raw_encap *encap;
6862 const struct rte_flow_action_rss *rss = NULL;
6863 const struct rte_flow_action_rss *sample_rss = NULL;
6864 const struct rte_flow_action_count *sample_count = NULL;
6865 const struct rte_flow_item_tcp nic_tcp_mask = {
6868 .src_port = RTE_BE16(UINT16_MAX),
6869 .dst_port = RTE_BE16(UINT16_MAX),
6872 const struct rte_flow_item_ipv6 nic_ipv6_mask = {
6875 "\xff\xff\xff\xff\xff\xff\xff\xff"
6876 "\xff\xff\xff\xff\xff\xff\xff\xff",
6878 "\xff\xff\xff\xff\xff\xff\xff\xff"
6879 "\xff\xff\xff\xff\xff\xff\xff\xff",
6880 .vtc_flow = RTE_BE32(0xffffffff),
6886 const struct rte_flow_item_ecpri nic_ecpri_mask = {
6890 RTE_BE32(((const struct rte_ecpri_common_hdr) {
6894 .dummy[0] = 0xffffffff,
6897 struct mlx5_priv *priv = dev->data->dev_private;
6898 struct mlx5_sh_config *dev_conf = &priv->sh->config;
6899 uint16_t queue_index = 0xFFFF;
6900 const struct rte_flow_item_vlan *vlan_m = NULL;
6901 uint32_t rw_act_num = 0;
6903 const struct mlx5_flow_tunnel *tunnel;
6904 enum mlx5_tof_rule_type tof_rule_type;
6905 struct flow_grp_info grp_info = {
6906 .external = !!external,
6907 .transfer = !!attr->transfer,
6908 .fdb_def_rule = !!priv->fdb_def_rule,
6909 .std_tbl_fix = true,
6911 const struct rte_eth_hairpin_conf *conf;
6912 const struct rte_flow_item *integrity_items[2] = {NULL, NULL};
6913 const struct rte_flow_item *port_id_item = NULL;
6914 bool def_policy = false;
6915 bool shared_count = false;
6916 uint16_t udp_dport = 0;
6917 uint32_t tag_id = 0;
6918 const struct rte_flow_action_age *non_shared_age = NULL;
6919 const struct rte_flow_action_count *count = NULL;
6923 tunnel = is_tunnel_offload_active(dev) ?
6924 mlx5_get_tof(items, actions, &tof_rule_type) : NULL;
6926 if (!dev_conf->dv_flow_en)
6927 return rte_flow_error_set
6929 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6930 NULL, "tunnel offload requires DV flow interface");
6931 if (priv->representor)
6932 return rte_flow_error_set
6934 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6935 NULL, "decap not supported for VF representor");
6936 if (tof_rule_type == MLX5_TUNNEL_OFFLOAD_SET_RULE)
6937 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
6938 else if (tof_rule_type == MLX5_TUNNEL_OFFLOAD_MATCH_RULE)
6939 action_flags |= MLX5_FLOW_ACTION_TUNNEL_MATCH |
6940 MLX5_FLOW_ACTION_DECAP;
6941 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
6942 (dev, attr, tunnel, tof_rule_type);
6944 ret = flow_dv_validate_attributes(dev, tunnel, attr, &grp_info, error);
6947 is_root = (uint64_t)ret;
6948 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
6949 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
6950 int type = items->type;
6952 if (!mlx5_flow_os_item_supported(type))
6953 return rte_flow_error_set(error, ENOTSUP,
6954 RTE_FLOW_ERROR_TYPE_ITEM,
6955 NULL, "item not supported");
6957 case RTE_FLOW_ITEM_TYPE_VOID:
6959 case RTE_FLOW_ITEM_TYPE_PORT_ID:
6960 ret = flow_dv_validate_item_port_id
6961 (dev, items, attr, item_flags, error);
6964 last_item = MLX5_FLOW_ITEM_PORT_ID;
6965 port_id_item = items;
6967 case RTE_FLOW_ITEM_TYPE_ETH:
6968 ret = mlx5_flow_validate_item_eth(items, item_flags,
6972 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
6973 MLX5_FLOW_LAYER_OUTER_L2;
6974 if (items->mask != NULL && items->spec != NULL) {
6976 ((const struct rte_flow_item_eth *)
6979 ((const struct rte_flow_item_eth *)
6981 ether_type = rte_be_to_cpu_16(ether_type);
6986 case RTE_FLOW_ITEM_TYPE_VLAN:
6987 ret = flow_dv_validate_item_vlan(items, item_flags,
6991 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
6992 MLX5_FLOW_LAYER_OUTER_VLAN;
6993 if (items->mask != NULL && items->spec != NULL) {
6995 ((const struct rte_flow_item_vlan *)
6996 items->spec)->inner_type;
6998 ((const struct rte_flow_item_vlan *)
6999 items->mask)->inner_type;
7000 ether_type = rte_be_to_cpu_16(ether_type);
7004 /* Store outer VLAN mask for of_push_vlan action. */
7006 vlan_m = items->mask;
7008 case RTE_FLOW_ITEM_TYPE_IPV4:
7009 mlx5_flow_tunnel_ip_check(items, next_protocol,
7010 &item_flags, &tunnel);
7011 ret = flow_dv_validate_item_ipv4(dev, items, item_flags,
7012 last_item, ether_type,
7016 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
7017 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
7018 if (items->mask != NULL &&
7019 ((const struct rte_flow_item_ipv4 *)
7020 items->mask)->hdr.next_proto_id) {
7022 ((const struct rte_flow_item_ipv4 *)
7023 (items->spec))->hdr.next_proto_id;
7025 ((const struct rte_flow_item_ipv4 *)
7026 (items->mask))->hdr.next_proto_id;
7028 /* Reset for inner layer. */
7029 next_protocol = 0xff;
7032 case RTE_FLOW_ITEM_TYPE_IPV6:
7033 mlx5_flow_tunnel_ip_check(items, next_protocol,
7034 &item_flags, &tunnel);
7035 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
7042 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
7043 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
7044 if (items->mask != NULL &&
7045 ((const struct rte_flow_item_ipv6 *)
7046 items->mask)->hdr.proto) {
7048 ((const struct rte_flow_item_ipv6 *)
7049 items->spec)->hdr.proto;
7051 ((const struct rte_flow_item_ipv6 *)
7052 items->spec)->hdr.proto;
7054 ((const struct rte_flow_item_ipv6 *)
7055 items->mask)->hdr.proto;
7057 /* Reset for inner layer. */
7058 next_protocol = 0xff;
7061 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
7062 ret = flow_dv_validate_item_ipv6_frag_ext(items,
7067 last_item = tunnel ?
7068 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
7069 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
7070 if (items->mask != NULL &&
7071 ((const struct rte_flow_item_ipv6_frag_ext *)
7072 items->mask)->hdr.next_header) {
7074 ((const struct rte_flow_item_ipv6_frag_ext *)
7075 items->spec)->hdr.next_header;
7077 ((const struct rte_flow_item_ipv6_frag_ext *)
7078 items->mask)->hdr.next_header;
7080 /* Reset for inner layer. */
7081 next_protocol = 0xff;
7084 case RTE_FLOW_ITEM_TYPE_TCP:
7085 ret = mlx5_flow_validate_item_tcp
7092 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
7093 MLX5_FLOW_LAYER_OUTER_L4_TCP;
7095 case RTE_FLOW_ITEM_TYPE_UDP:
7096 ret = mlx5_flow_validate_item_udp(items, item_flags,
7099 const struct rte_flow_item_udp *spec = items->spec;
7100 const struct rte_flow_item_udp *mask = items->mask;
7102 mask = &rte_flow_item_udp_mask;
7104 udp_dport = rte_be_to_cpu_16
7105 (spec->hdr.dst_port &
7106 mask->hdr.dst_port);
7109 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
7110 MLX5_FLOW_LAYER_OUTER_L4_UDP;
7112 case RTE_FLOW_ITEM_TYPE_GRE:
7113 ret = mlx5_flow_validate_item_gre(items, item_flags,
7114 next_protocol, error);
7118 last_item = MLX5_FLOW_LAYER_GRE;
7120 case RTE_FLOW_ITEM_TYPE_GRE_OPTION:
7121 ret = mlx5_flow_validate_item_gre_option(dev, items, item_flags,
7122 attr, gre_item, error);
7125 last_item = MLX5_FLOW_LAYER_GRE;
7127 case RTE_FLOW_ITEM_TYPE_NVGRE:
7128 ret = mlx5_flow_validate_item_nvgre(items, item_flags,
7133 last_item = MLX5_FLOW_LAYER_NVGRE;
7135 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
7136 ret = mlx5_flow_validate_item_gre_key
7137 (items, item_flags, gre_item, error);
7140 last_item = MLX5_FLOW_LAYER_GRE_KEY;
7142 case RTE_FLOW_ITEM_TYPE_VXLAN:
7143 ret = mlx5_flow_validate_item_vxlan(dev, udp_dport,
7148 last_item = MLX5_FLOW_LAYER_VXLAN;
7150 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
7151 ret = mlx5_flow_validate_item_vxlan_gpe(items,
7156 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
7158 case RTE_FLOW_ITEM_TYPE_GENEVE:
7159 ret = mlx5_flow_validate_item_geneve(items,
7164 geneve_item = items;
7165 last_item = MLX5_FLOW_LAYER_GENEVE;
7167 case RTE_FLOW_ITEM_TYPE_GENEVE_OPT:
7168 ret = mlx5_flow_validate_item_geneve_opt(items,
7175 last_item = MLX5_FLOW_LAYER_GENEVE_OPT;
7177 case RTE_FLOW_ITEM_TYPE_MPLS:
7178 ret = mlx5_flow_validate_item_mpls(dev, items,
7183 last_item = MLX5_FLOW_LAYER_MPLS;
7186 case RTE_FLOW_ITEM_TYPE_MARK:
7187 ret = flow_dv_validate_item_mark(dev, items, attr,
7191 last_item = MLX5_FLOW_ITEM_MARK;
7193 case RTE_FLOW_ITEM_TYPE_META:
7194 ret = flow_dv_validate_item_meta(dev, items, attr,
7198 last_item = MLX5_FLOW_ITEM_METADATA;
7200 case RTE_FLOW_ITEM_TYPE_ICMP:
7201 ret = mlx5_flow_validate_item_icmp(items, item_flags,
7206 last_item = MLX5_FLOW_LAYER_ICMP;
7208 case RTE_FLOW_ITEM_TYPE_ICMP6:
7209 ret = mlx5_flow_validate_item_icmp6(items, item_flags,
7214 item_ipv6_proto = IPPROTO_ICMPV6;
7215 last_item = MLX5_FLOW_LAYER_ICMP6;
7217 case RTE_FLOW_ITEM_TYPE_TAG:
7218 ret = flow_dv_validate_item_tag(dev, items,
7222 last_item = MLX5_FLOW_ITEM_TAG;
7224 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
7225 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
7227 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
7229 case RTE_FLOW_ITEM_TYPE_GTP:
7230 ret = flow_dv_validate_item_gtp(dev, items, item_flags,
7235 last_item = MLX5_FLOW_LAYER_GTP;
7237 case RTE_FLOW_ITEM_TYPE_GTP_PSC:
7238 ret = flow_dv_validate_item_gtp_psc(items, last_item,
7243 last_item = MLX5_FLOW_LAYER_GTP_PSC;
7245 case RTE_FLOW_ITEM_TYPE_ECPRI:
7246 /* Capacity will be checked in the translate stage. */
7247 ret = mlx5_flow_validate_item_ecpri(items, item_flags,
7254 last_item = MLX5_FLOW_LAYER_ECPRI;
7256 case RTE_FLOW_ITEM_TYPE_INTEGRITY:
7257 ret = flow_dv_validate_item_integrity(dev, items,
7265 case RTE_FLOW_ITEM_TYPE_CONNTRACK:
7266 ret = flow_dv_validate_item_aso_ct(dev, items,
7267 &item_flags, error);
7271 case MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL:
7272 /* tunnel offload item was processed before
7273 * list it here as a supported type
7276 case RTE_FLOW_ITEM_TYPE_FLEX:
7277 ret = flow_dv_validate_item_flex(dev, items, item_flags,
7279 tunnel != 0, error);
7284 return rte_flow_error_set(error, ENOTSUP,
7285 RTE_FLOW_ERROR_TYPE_ITEM,
7286 NULL, "item not supported");
7288 item_flags |= last_item;
7290 if (item_flags & MLX5_FLOW_ITEM_INTEGRITY) {
7291 ret = flow_dv_validate_item_integrity_post(integrity_items,
7296 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
7297 int type = actions->type;
7299 if (!mlx5_flow_os_action_supported(type))
7300 return rte_flow_error_set(error, ENOTSUP,
7301 RTE_FLOW_ERROR_TYPE_ACTION,
7303 "action not supported");
7304 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
7305 return rte_flow_error_set(error, ENOTSUP,
7306 RTE_FLOW_ERROR_TYPE_ACTION,
7307 actions, "too many actions");
7309 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY)
7310 return rte_flow_error_set(error, ENOTSUP,
7311 RTE_FLOW_ERROR_TYPE_ACTION,
7312 NULL, "meter action with policy "
7313 "must be the last action");
7315 case RTE_FLOW_ACTION_TYPE_VOID:
7317 case RTE_FLOW_ACTION_TYPE_PORT_ID:
7318 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
7319 ret = flow_dv_validate_action_port_id(dev,
7326 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
7329 case RTE_FLOW_ACTION_TYPE_FLAG:
7330 ret = flow_dv_validate_action_flag(dev, action_flags,
7334 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7335 /* Count all modify-header actions as one. */
7336 if (!(action_flags &
7337 MLX5_FLOW_MODIFY_HDR_ACTIONS))
7339 action_flags |= MLX5_FLOW_ACTION_FLAG |
7340 MLX5_FLOW_ACTION_MARK_EXT;
7341 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7342 modify_after_mirror = 1;
7345 action_flags |= MLX5_FLOW_ACTION_FLAG;
7348 rw_act_num += MLX5_ACT_NUM_SET_MARK;
7350 case RTE_FLOW_ACTION_TYPE_MARK:
7351 ret = flow_dv_validate_action_mark(dev, actions,
7356 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
7357 /* Count all modify-header actions as one. */
7358 if (!(action_flags &
7359 MLX5_FLOW_MODIFY_HDR_ACTIONS))
7361 action_flags |= MLX5_FLOW_ACTION_MARK |
7362 MLX5_FLOW_ACTION_MARK_EXT;
7363 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7364 modify_after_mirror = 1;
7366 action_flags |= MLX5_FLOW_ACTION_MARK;
7369 rw_act_num += MLX5_ACT_NUM_SET_MARK;
7371 case RTE_FLOW_ACTION_TYPE_SET_META:
7372 ret = flow_dv_validate_action_set_meta(dev, actions,
7377 /* Count all modify-header actions as one action. */
7378 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7380 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7381 modify_after_mirror = 1;
7382 action_flags |= MLX5_FLOW_ACTION_SET_META;
7383 rw_act_num += MLX5_ACT_NUM_SET_META;
7385 case RTE_FLOW_ACTION_TYPE_SET_TAG:
7386 ret = flow_dv_validate_action_set_tag(dev, actions,
7391 /* Count all modify-header actions as one action. */
7392 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7394 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7395 modify_after_mirror = 1;
7396 tag_id = ((const struct rte_flow_action_set_tag *)
7397 actions->conf)->index;
7398 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
7399 rw_act_num += MLX5_ACT_NUM_SET_TAG;
7401 case RTE_FLOW_ACTION_TYPE_DROP:
7402 ret = mlx5_flow_validate_action_drop(action_flags,
7406 action_flags |= MLX5_FLOW_ACTION_DROP;
7409 case RTE_FLOW_ACTION_TYPE_QUEUE:
7410 ret = mlx5_flow_validate_action_queue(actions,
7415 queue_index = ((const struct rte_flow_action_queue *)
7416 (actions->conf))->index;
7417 action_flags |= MLX5_FLOW_ACTION_QUEUE;
7420 case RTE_FLOW_ACTION_TYPE_RSS:
7421 rss = actions->conf;
7422 ret = mlx5_flow_validate_action_rss(actions,
7428 if (rss && sample_rss &&
7429 (sample_rss->level != rss->level ||
7430 sample_rss->types != rss->types))
7431 return rte_flow_error_set(error, ENOTSUP,
7432 RTE_FLOW_ERROR_TYPE_ACTION,
7434 "Can't use the different RSS types "
7435 "or level in the same flow");
7436 if (rss != NULL && rss->queue_num)
7437 queue_index = rss->queue[0];
7438 action_flags |= MLX5_FLOW_ACTION_RSS;
7441 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
7443 mlx5_flow_validate_action_default_miss(action_flags,
7447 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
7450 case MLX5_RTE_FLOW_ACTION_TYPE_COUNT:
7451 shared_count = true;
7453 case RTE_FLOW_ACTION_TYPE_COUNT:
7454 ret = flow_dv_validate_action_count(dev, shared_count,
7459 count = actions->conf;
7460 action_flags |= MLX5_FLOW_ACTION_COUNT;
7463 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
7464 if (flow_dv_validate_action_pop_vlan(dev,
7470 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7471 modify_after_mirror = 1;
7472 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
7475 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
7476 ret = flow_dv_validate_action_push_vlan(dev,
7483 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7484 modify_after_mirror = 1;
7485 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
7488 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
7489 ret = flow_dv_validate_action_set_vlan_pcp
7490 (action_flags, actions, error);
7493 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7494 modify_after_mirror = 1;
7495 /* Count PCP with push_vlan command. */
7496 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
7498 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
7499 ret = flow_dv_validate_action_set_vlan_vid
7500 (item_flags, action_flags,
7504 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7505 modify_after_mirror = 1;
7506 /* Count VID with push_vlan command. */
7507 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
7508 rw_act_num += MLX5_ACT_NUM_MDF_VID;
7510 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
7511 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
7512 ret = flow_dv_validate_action_l2_encap(dev,
7518 action_flags |= MLX5_FLOW_ACTION_ENCAP;
7521 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
7522 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
7523 ret = flow_dv_validate_action_decap(dev, action_flags,
7524 actions, item_flags,
7528 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7529 modify_after_mirror = 1;
7530 action_flags |= MLX5_FLOW_ACTION_DECAP;
7533 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
7534 ret = flow_dv_validate_action_raw_encap_decap
7535 (dev, NULL, actions->conf, attr, &action_flags,
7536 &actions_n, actions, item_flags, error);
7540 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
7541 decap = actions->conf;
7542 while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
7544 if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
7548 encap = actions->conf;
7550 ret = flow_dv_validate_action_raw_encap_decap
7552 decap ? decap : &empty_decap, encap,
7553 attr, &action_flags, &actions_n,
7554 actions, item_flags, error);
7557 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) &&
7558 (action_flags & MLX5_FLOW_ACTION_DECAP))
7559 modify_after_mirror = 1;
7561 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
7562 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
7563 ret = flow_dv_validate_action_modify_mac(action_flags,
7569 /* Count all modify-header actions as one action. */
7570 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7572 action_flags |= actions->type ==
7573 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
7574 MLX5_FLOW_ACTION_SET_MAC_SRC :
7575 MLX5_FLOW_ACTION_SET_MAC_DST;
7576 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7577 modify_after_mirror = 1;
7579 * Even if the source and destination MAC addresses have
7580 * overlap in the header with 4B alignment, the convert
7581 * function will handle them separately and 4 SW actions
7582 * will be created. And 2 actions will be added each
7583 * time no matter how many bytes of address will be set.
7585 rw_act_num += MLX5_ACT_NUM_MDF_MAC;
7587 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
7588 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
7589 ret = flow_dv_validate_action_modify_ipv4(action_flags,
7595 /* Count all modify-header actions as one action. */
7596 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7598 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7599 modify_after_mirror = 1;
7600 action_flags |= actions->type ==
7601 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
7602 MLX5_FLOW_ACTION_SET_IPV4_SRC :
7603 MLX5_FLOW_ACTION_SET_IPV4_DST;
7604 rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
7606 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
7607 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
7608 ret = flow_dv_validate_action_modify_ipv6(action_flags,
7614 if (item_ipv6_proto == IPPROTO_ICMPV6)
7615 return rte_flow_error_set(error, ENOTSUP,
7616 RTE_FLOW_ERROR_TYPE_ACTION,
7618 "Can't change header "
7619 "with ICMPv6 proto");
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_IPV6_SRC ?
7627 MLX5_FLOW_ACTION_SET_IPV6_SRC :
7628 MLX5_FLOW_ACTION_SET_IPV6_DST;
7629 rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
7631 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
7632 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
7633 ret = flow_dv_validate_action_modify_tp(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_TP_SRC ?
7646 MLX5_FLOW_ACTION_SET_TP_SRC :
7647 MLX5_FLOW_ACTION_SET_TP_DST;
7648 rw_act_num += MLX5_ACT_NUM_MDF_PORT;
7650 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
7651 case RTE_FLOW_ACTION_TYPE_SET_TTL:
7652 ret = flow_dv_validate_action_modify_ttl(action_flags,
7658 /* Count all modify-header actions as one action. */
7659 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7661 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7662 modify_after_mirror = 1;
7663 action_flags |= actions->type ==
7664 RTE_FLOW_ACTION_TYPE_SET_TTL ?
7665 MLX5_FLOW_ACTION_SET_TTL :
7666 MLX5_FLOW_ACTION_DEC_TTL;
7667 rw_act_num += MLX5_ACT_NUM_MDF_TTL;
7669 case RTE_FLOW_ACTION_TYPE_JUMP:
7670 ret = flow_dv_validate_action_jump(dev, tunnel, actions,
7676 if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) &&
7678 return rte_flow_error_set(error, EINVAL,
7679 RTE_FLOW_ERROR_TYPE_ACTION,
7681 "sample and jump action combination is not supported");
7683 action_flags |= MLX5_FLOW_ACTION_JUMP;
7685 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
7686 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
7687 ret = flow_dv_validate_action_modify_tcp_seq
7694 /* Count all modify-header actions as one action. */
7695 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7697 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7698 modify_after_mirror = 1;
7699 action_flags |= actions->type ==
7700 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
7701 MLX5_FLOW_ACTION_INC_TCP_SEQ :
7702 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
7703 rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
7705 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
7706 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
7707 ret = flow_dv_validate_action_modify_tcp_ack
7714 /* Count all modify-header actions as one action. */
7715 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7717 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7718 modify_after_mirror = 1;
7719 action_flags |= actions->type ==
7720 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
7721 MLX5_FLOW_ACTION_INC_TCP_ACK :
7722 MLX5_FLOW_ACTION_DEC_TCP_ACK;
7723 rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
7725 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
7727 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
7728 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
7729 rw_act_num += MLX5_ACT_NUM_SET_TAG;
7731 case RTE_FLOW_ACTION_TYPE_METER:
7732 ret = mlx5_flow_validate_action_meter(dev,
7741 action_flags |= MLX5_FLOW_ACTION_METER;
7744 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY;
7746 /* Meter action will add one more TAG action. */
7747 rw_act_num += MLX5_ACT_NUM_SET_TAG;
7749 case MLX5_RTE_FLOW_ACTION_TYPE_AGE:
7750 if (!attr->transfer && !attr->group)
7751 return rte_flow_error_set(error, ENOTSUP,
7752 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7754 "Shared ASO age action is not supported for group 0");
7755 if (action_flags & MLX5_FLOW_ACTION_AGE)
7756 return rte_flow_error_set
7758 RTE_FLOW_ERROR_TYPE_ACTION,
7760 "duplicate age actions set");
7761 action_flags |= MLX5_FLOW_ACTION_AGE;
7764 case RTE_FLOW_ACTION_TYPE_AGE:
7765 non_shared_age = actions->conf;
7766 ret = flow_dv_validate_action_age(action_flags,
7772 * Validate the regular AGE action (using counter)
7773 * mutual exclusion with indirect counter actions.
7775 if (!flow_hit_aso_supported(priv->sh, attr)) {
7777 return rte_flow_error_set
7779 RTE_FLOW_ERROR_TYPE_ACTION,
7781 "old age and indirect count combination is not supported");
7783 return rte_flow_error_set
7785 RTE_FLOW_ERROR_TYPE_ACTION,
7787 "old age action and count must be in the same sub flow");
7789 action_flags |= MLX5_FLOW_ACTION_AGE;
7792 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
7793 ret = flow_dv_validate_action_modify_ipv4_dscp
7800 /* Count all modify-header actions as one action. */
7801 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7803 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7804 modify_after_mirror = 1;
7805 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
7806 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
7808 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
7809 ret = flow_dv_validate_action_modify_ipv6_dscp
7816 /* Count all modify-header actions as one action. */
7817 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7819 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7820 modify_after_mirror = 1;
7821 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
7822 rw_act_num += MLX5_ACT_NUM_SET_DSCP;
7824 case RTE_FLOW_ACTION_TYPE_SAMPLE:
7825 ret = flow_dv_validate_action_sample(&action_flags,
7834 if ((action_flags & MLX5_FLOW_ACTION_SET_TAG) &&
7835 tag_id == 0 && priv->mtr_color_reg == REG_NON)
7836 return rte_flow_error_set(error, EINVAL,
7837 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7838 "sample after tag action causes metadata tag index 0 corruption");
7839 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
7842 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
7843 ret = flow_dv_validate_action_modify_field(dev,
7850 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
7851 modify_after_mirror = 1;
7852 /* Count all modify-header actions as one action. */
7853 if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
7855 action_flags |= MLX5_FLOW_ACTION_MODIFY_FIELD;
7858 case RTE_FLOW_ACTION_TYPE_CONNTRACK:
7859 ret = flow_dv_validate_action_aso_ct(dev, action_flags,
7864 action_flags |= MLX5_FLOW_ACTION_CT;
7866 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
7867 /* tunnel offload action was processed before
7868 * list it here as a supported type
7872 return rte_flow_error_set(error, ENOTSUP,
7873 RTE_FLOW_ERROR_TYPE_ACTION,
7875 "action not supported");
7879 * Validate actions in flow rules
7880 * - Explicit decap action is prohibited by the tunnel offload API.
7881 * - Drop action in tunnel steer rule is prohibited by the API.
7882 * - Application cannot use MARK action because it's value can mask
7883 * tunnel default miss notification.
7884 * - JUMP in tunnel match rule has no support in current PMD
7886 * - TAG & META are reserved for future uses.
7888 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_SET) {
7889 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_DECAP |
7890 MLX5_FLOW_ACTION_MARK |
7891 MLX5_FLOW_ACTION_SET_TAG |
7892 MLX5_FLOW_ACTION_SET_META |
7893 MLX5_FLOW_ACTION_DROP;
7895 if (action_flags & bad_actions_mask)
7896 return rte_flow_error_set
7898 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7899 "Invalid RTE action in tunnel "
7901 if (!(action_flags & MLX5_FLOW_ACTION_JUMP))
7902 return rte_flow_error_set
7904 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7905 "tunnel set decap rule must terminate "
7908 return rte_flow_error_set
7910 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7911 "tunnel flows for ingress traffic only");
7913 if (action_flags & MLX5_FLOW_ACTION_TUNNEL_MATCH) {
7914 uint64_t bad_actions_mask = MLX5_FLOW_ACTION_JUMP |
7915 MLX5_FLOW_ACTION_MARK |
7916 MLX5_FLOW_ACTION_SET_TAG |
7917 MLX5_FLOW_ACTION_SET_META;
7919 if (action_flags & bad_actions_mask)
7920 return rte_flow_error_set
7922 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7923 "Invalid RTE action in tunnel "
7927 * Validate the drop action mutual exclusion with other actions.
7928 * Drop action is mutually-exclusive with any other action, except for
7930 * Drop action compatibility with tunnel offload was already validated.
7932 if (action_flags & (MLX5_FLOW_ACTION_TUNNEL_MATCH |
7933 MLX5_FLOW_ACTION_TUNNEL_MATCH));
7934 else if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
7935 (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
7936 return rte_flow_error_set(error, EINVAL,
7937 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
7938 "Drop action is mutually-exclusive "
7939 "with any other action, except for "
7941 /* Eswitch has few restrictions on using items and actions */
7942 if (attr->transfer) {
7943 if (!mlx5_flow_ext_mreg_supported(dev) &&
7944 action_flags & MLX5_FLOW_ACTION_FLAG)
7945 return rte_flow_error_set(error, ENOTSUP,
7946 RTE_FLOW_ERROR_TYPE_ACTION,
7948 "unsupported action FLAG");
7949 if (!mlx5_flow_ext_mreg_supported(dev) &&
7950 action_flags & MLX5_FLOW_ACTION_MARK)
7951 return rte_flow_error_set(error, ENOTSUP,
7952 RTE_FLOW_ERROR_TYPE_ACTION,
7954 "unsupported action MARK");
7955 if (action_flags & MLX5_FLOW_ACTION_QUEUE)
7956 return rte_flow_error_set(error, ENOTSUP,
7957 RTE_FLOW_ERROR_TYPE_ACTION,
7959 "unsupported action QUEUE");
7960 if (action_flags & MLX5_FLOW_ACTION_RSS)
7961 return rte_flow_error_set(error, ENOTSUP,
7962 RTE_FLOW_ERROR_TYPE_ACTION,
7964 "unsupported action RSS");
7965 if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
7966 return rte_flow_error_set(error, EINVAL,
7967 RTE_FLOW_ERROR_TYPE_ACTION,
7969 "no fate action is found");
7971 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
7972 return rte_flow_error_set(error, EINVAL,
7973 RTE_FLOW_ERROR_TYPE_ACTION,
7975 "no fate action is found");
7978 * Continue validation for Xcap and VLAN actions.
7979 * If hairpin is working in explicit TX rule mode, there is no actions
7980 * splitting and the validation of hairpin ingress flow should be the
7981 * same as other standard flows.
7983 if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
7984 MLX5_FLOW_VLAN_ACTIONS)) &&
7985 (queue_index == 0xFFFF || !mlx5_rxq_is_hairpin(dev, queue_index) ||
7986 ((conf = mlx5_rxq_get_hairpin_conf(dev, queue_index)) != NULL &&
7987 conf->tx_explicit != 0))) {
7988 if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
7989 MLX5_FLOW_XCAP_ACTIONS)
7990 return rte_flow_error_set(error, ENOTSUP,
7991 RTE_FLOW_ERROR_TYPE_ACTION,
7992 NULL, "encap and decap "
7993 "combination aren't supported");
7994 /* Push VLAN is not supported in ingress except for NICs newer than CX5. */
7995 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) {
7996 struct mlx5_dev_ctx_shared *sh = priv->sh;
7997 bool direction_error = false;
7999 if (attr->transfer) {
8000 bool fdb_tx = priv->representor_id != UINT16_MAX;
8001 bool is_cx5 = sh->steering_format_version ==
8002 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5;
8004 if (!fdb_tx && is_cx5)
8005 direction_error = true;
8006 } else if (attr->ingress) {
8007 direction_error = true;
8009 if (direction_error)
8010 return rte_flow_error_set(error, ENOTSUP,
8011 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
8013 "push VLAN action not supported "
8016 if (!attr->transfer && attr->ingress) {
8017 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
8018 return rte_flow_error_set
8020 RTE_FLOW_ERROR_TYPE_ACTION,
8021 NULL, "encap is not supported"
8022 " for ingress traffic");
8023 else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
8024 MLX5_FLOW_VLAN_ACTIONS)
8025 return rte_flow_error_set
8027 RTE_FLOW_ERROR_TYPE_ACTION,
8028 NULL, "no support for "
8029 "multiple VLAN actions");
8032 if (action_flags & MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY) {
8033 if ((action_flags & (MLX5_FLOW_FATE_ACTIONS &
8034 ~MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY)) &&
8036 return rte_flow_error_set
8038 RTE_FLOW_ERROR_TYPE_ACTION,
8039 NULL, "fate action not supported for "
8040 "meter with policy");
8042 if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
8043 return rte_flow_error_set
8045 RTE_FLOW_ERROR_TYPE_ACTION,
8046 NULL, "modify header action in egress "
8047 "cannot be done before meter action");
8048 if (action_flags & MLX5_FLOW_ACTION_ENCAP)
8049 return rte_flow_error_set
8051 RTE_FLOW_ERROR_TYPE_ACTION,
8052 NULL, "encap action in egress "
8053 "cannot be done before meter action");
8054 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
8055 return rte_flow_error_set
8057 RTE_FLOW_ERROR_TYPE_ACTION,
8058 NULL, "push vlan action in egress "
8059 "cannot be done before meter action");
8063 * Only support one ASO action in a single flow rule.
8064 * non-shared AGE + counter will fallback to use HW counter, no ASO hit object.
8065 * Group 0 uses HW counter for AGE too even if no counter action.
8067 aso_mask = (action_flags & MLX5_FLOW_ACTION_METER && priv->sh->meter_aso_en) << 2 |
8068 (action_flags & MLX5_FLOW_ACTION_CT && priv->sh->ct_aso_en) << 1 |
8069 (action_flags & MLX5_FLOW_ACTION_AGE &&
8070 !(non_shared_age && count) &&
8071 (attr->group || (attr->transfer && priv->fdb_def_rule)) &&
8072 priv->sh->flow_hit_aso_en);
8073 if (__builtin_popcountl(aso_mask) > 1)
8074 return rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8075 NULL, "unsupported combining AGE, METER, CT ASO actions in a single rule");
8077 * Hairpin flow will add one more TAG action in TX implicit mode.
8078 * In TX explicit mode, there will be no hairpin flow ID.
8081 rw_act_num += MLX5_ACT_NUM_SET_TAG;
8082 /* extra metadata enabled: one more TAG action will be add. */
8083 if (dev_conf->dv_flow_en &&
8084 dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
8085 mlx5_flow_ext_mreg_supported(dev))
8086 rw_act_num += MLX5_ACT_NUM_SET_TAG;
8088 flow_dv_modify_hdr_action_max(dev, is_root)) {
8089 return rte_flow_error_set(error, ENOTSUP,
8090 RTE_FLOW_ERROR_TYPE_ACTION,
8091 NULL, "too many header modify"
8092 " actions to support");
8094 /* Eswitch egress mirror and modify flow has limitation on CX5 */
8095 if (fdb_mirror_limit && modify_after_mirror)
8096 return rte_flow_error_set(error, EINVAL,
8097 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
8098 "sample before modify action is not supported");
8100 * Validation the NIC Egress flow on representor, except implicit
8101 * hairpin default egress flow with TX_QUEUE item, other flows not
8102 * work due to metadata regC0 mismatch.
8104 if ((!attr->transfer && attr->egress) && priv->representor &&
8105 !(item_flags & MLX5_FLOW_ITEM_TX_QUEUE))
8106 return rte_flow_error_set(error, EINVAL,
8107 RTE_FLOW_ERROR_TYPE_ITEM,
8109 "NIC egress rules on representors"
8110 " is not supported");
8115 * Internal preparation function. Allocates the DV flow size,
8116 * this size is constant.
8119 * Pointer to the rte_eth_dev structure.
8121 * Pointer to the flow attributes.
8123 * Pointer to the list of items.
8124 * @param[in] actions
8125 * Pointer to the list of actions.
8127 * Pointer to the error structure.
8130 * Pointer to mlx5_flow object on success,
8131 * otherwise NULL and rte_errno is set.
8133 static struct mlx5_flow *
8134 flow_dv_prepare(struct rte_eth_dev *dev,
8135 const struct rte_flow_attr *attr __rte_unused,
8136 const struct rte_flow_item items[] __rte_unused,
8137 const struct rte_flow_action actions[] __rte_unused,
8138 struct rte_flow_error *error)
8140 uint32_t handle_idx = 0;
8141 struct mlx5_flow *dev_flow;
8142 struct mlx5_flow_handle *dev_handle;
8143 struct mlx5_priv *priv = dev->data->dev_private;
8144 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
8147 wks->skip_matcher_reg = 0;
8149 wks->final_policy = NULL;
8150 /* In case of corrupting the memory. */
8151 if (wks->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
8152 rte_flow_error_set(error, ENOSPC,
8153 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8154 "not free temporary device flow");
8157 dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8160 rte_flow_error_set(error, ENOMEM,
8161 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8162 "not enough memory to create flow handle");
8165 MLX5_ASSERT(wks->flow_idx < RTE_DIM(wks->flows));
8166 dev_flow = &wks->flows[wks->flow_idx++];
8167 memset(dev_flow, 0, sizeof(*dev_flow));
8168 dev_flow->handle = dev_handle;
8169 dev_flow->handle_idx = handle_idx;
8170 dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
8171 dev_flow->ingress = attr->ingress;
8172 dev_flow->dv.transfer = attr->transfer;
8176 #ifdef RTE_LIBRTE_MLX5_DEBUG
8178 * Sanity check for match mask and value. Similar to check_valid_spec() in
8179 * kernel driver. If unmasked bit is present in value, it returns failure.
8182 * pointer to match mask buffer.
8183 * @param match_value
8184 * pointer to match value buffer.
8187 * 0 if valid, -EINVAL otherwise.
8190 flow_dv_check_valid_spec(void *match_mask, void *match_value)
8192 uint8_t *m = match_mask;
8193 uint8_t *v = match_value;
8196 for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
8199 "match_value differs from match_criteria"
8200 " %p[%u] != %p[%u]",
8201 match_value, i, match_mask, i);
8210 * Add match of ip_version.
8214 * @param[in] headers_v
8215 * Values header pointer.
8216 * @param[in] headers_m
8217 * Masks header pointer.
8218 * @param[in] ip_version
8219 * The IP version to set.
8222 flow_dv_set_match_ip_version(uint32_t group,
8228 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
8230 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
8232 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
8233 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
8234 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
8238 * Add Ethernet item to matcher and to the value.
8240 * @param[in, out] matcher
8242 * @param[in, out] key
8243 * Flow matcher value.
8245 * Flow pattern to translate.
8247 * Item is inner pattern.
8250 flow_dv_translate_item_eth(void *matcher, void *key,
8251 const struct rte_flow_item *item, int inner,
8254 const struct rte_flow_item_eth *eth_m = item->mask;
8255 const struct rte_flow_item_eth *eth_v = item->spec;
8256 const struct rte_flow_item_eth nic_mask = {
8257 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
8258 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
8259 .type = RTE_BE16(0xffff),
8272 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
8274 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8276 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
8278 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8280 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, dmac_47_16),
8281 ð_m->dst, sizeof(eth_m->dst));
8282 /* The value must be in the range of the mask. */
8283 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, dmac_47_16);
8284 for (i = 0; i < sizeof(eth_m->dst); ++i)
8285 l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
8286 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, smac_47_16),
8287 ð_m->src, sizeof(eth_m->src));
8288 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, smac_47_16);
8289 /* The value must be in the range of the mask. */
8290 for (i = 0; i < sizeof(eth_m->dst); ++i)
8291 l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
8293 * HW supports match on one Ethertype, the Ethertype following the last
8294 * VLAN tag of the packet (see PRM).
8295 * Set match on ethertype only if ETH header is not followed by VLAN.
8296 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
8297 * ethertype, and use ip_version field instead.
8298 * eCPRI over Ether layer will use type value 0xAEFE.
8300 if (eth_m->type == 0xFFFF) {
8301 /* Set cvlan_tag mask for any single\multi\un-tagged case. */
8302 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
8303 switch (eth_v->type) {
8304 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
8305 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
8307 case RTE_BE16(RTE_ETHER_TYPE_QINQ):
8308 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
8309 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
8311 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
8312 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
8314 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
8315 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
8321 if (eth_m->has_vlan) {
8322 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
8323 if (eth_v->has_vlan) {
8325 * Here, when also has_more_vlan field in VLAN item is
8326 * not set, only single-tagged packets will be matched.
8328 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
8332 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
8333 rte_be_to_cpu_16(eth_m->type));
8334 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, ethertype);
8335 *(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
8339 * Add VLAN item to matcher and to the value.
8341 * @param[in, out] dev_flow
8343 * @param[in, out] matcher
8345 * @param[in, out] key
8346 * Flow matcher value.
8348 * Flow pattern to translate.
8350 * Item is inner pattern.
8353 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
8354 void *matcher, void *key,
8355 const struct rte_flow_item *item,
8356 int inner, uint32_t group)
8358 const struct rte_flow_item_vlan *vlan_m = item->mask;
8359 const struct rte_flow_item_vlan *vlan_v = item->spec;
8366 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
8368 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8370 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher,
8372 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8374 * This is workaround, masks are not supported,
8375 * and pre-validated.
8378 dev_flow->handle->vf_vlan.tag =
8379 rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
8382 * When VLAN item exists in flow, mark packet as tagged,
8383 * even if TCI is not specified.
8385 if (!MLX5_GET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag)) {
8386 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, cvlan_tag, 1);
8387 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 1);
8392 vlan_m = &rte_flow_item_vlan_mask;
8393 tci_m = rte_be_to_cpu_16(vlan_m->tci);
8394 tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
8395 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_vid, tci_m);
8396 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_vid, tci_v);
8397 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_cfi, tci_m >> 12);
8398 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_cfi, tci_v >> 12);
8399 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, first_prio, tci_m >> 13);
8400 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, first_prio, tci_v >> 13);
8402 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
8403 * ethertype, and use ip_version field instead.
8405 if (vlan_m->inner_type == 0xFFFF) {
8406 switch (vlan_v->inner_type) {
8407 case RTE_BE16(RTE_ETHER_TYPE_VLAN):
8408 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
8409 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
8410 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
8412 case RTE_BE16(RTE_ETHER_TYPE_IPV4):
8413 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 4);
8415 case RTE_BE16(RTE_ETHER_TYPE_IPV6):
8416 flow_dv_set_match_ip_version(group, hdrs_v, hdrs_m, 6);
8422 if (vlan_m->has_more_vlan && vlan_v->has_more_vlan) {
8423 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, svlan_tag, 1);
8424 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, svlan_tag, 1);
8425 /* Only one vlan_tag bit can be set. */
8426 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, cvlan_tag, 0);
8429 MLX5_SET(fte_match_set_lyr_2_4, hdrs_m, ethertype,
8430 rte_be_to_cpu_16(vlan_m->inner_type));
8431 MLX5_SET(fte_match_set_lyr_2_4, hdrs_v, ethertype,
8432 rte_be_to_cpu_16(vlan_m->inner_type & vlan_v->inner_type));
8436 * Add IPV4 item to matcher and to the value.
8438 * @param[in, out] matcher
8440 * @param[in, out] key
8441 * Flow matcher value.
8443 * Flow pattern to translate.
8445 * Item is inner pattern.
8447 * The group to insert the rule.
8450 flow_dv_translate_item_ipv4(void *matcher, void *key,
8451 const struct rte_flow_item *item,
8452 int inner, uint32_t group)
8454 const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
8455 const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
8456 const struct rte_flow_item_ipv4 nic_mask = {
8458 .src_addr = RTE_BE32(0xffffffff),
8459 .dst_addr = RTE_BE32(0xffffffff),
8460 .type_of_service = 0xff,
8461 .next_proto_id = 0xff,
8462 .time_to_live = 0xff,
8469 uint8_t tos, ihl_m, ihl_v;
8472 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8474 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8476 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8478 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8480 flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
8485 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
8486 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
8487 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
8488 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
8489 *(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
8490 *(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
8491 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
8492 src_ipv4_src_ipv6.ipv4_layout.ipv4);
8493 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
8494 src_ipv4_src_ipv6.ipv4_layout.ipv4);
8495 *(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
8496 *(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
8497 tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
8498 ihl_m = ipv4_m->hdr.version_ihl & RTE_IPV4_HDR_IHL_MASK;
8499 ihl_v = ipv4_v->hdr.version_ihl & RTE_IPV4_HDR_IHL_MASK;
8500 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ipv4_ihl, ihl_m);
8501 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ipv4_ihl, ihl_m & ihl_v);
8502 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
8503 ipv4_m->hdr.type_of_service);
8504 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
8505 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
8506 ipv4_m->hdr.type_of_service >> 2);
8507 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
8508 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
8509 ipv4_m->hdr.next_proto_id);
8510 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
8511 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
8512 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
8513 ipv4_m->hdr.time_to_live);
8514 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
8515 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
8516 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
8517 !!(ipv4_m->hdr.fragment_offset));
8518 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
8519 !!(ipv4_v->hdr.fragment_offset & ipv4_m->hdr.fragment_offset));
8523 * Add IPV6 item to matcher and to the value.
8525 * @param[in, out] matcher
8527 * @param[in, out] key
8528 * Flow matcher value.
8530 * Flow pattern to translate.
8532 * Item is inner pattern.
8534 * The group to insert the rule.
8537 flow_dv_translate_item_ipv6(void *matcher, void *key,
8538 const struct rte_flow_item *item,
8539 int inner, uint32_t group)
8541 const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
8542 const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
8543 const struct rte_flow_item_ipv6 nic_mask = {
8546 "\xff\xff\xff\xff\xff\xff\xff\xff"
8547 "\xff\xff\xff\xff\xff\xff\xff\xff",
8549 "\xff\xff\xff\xff\xff\xff\xff\xff"
8550 "\xff\xff\xff\xff\xff\xff\xff\xff",
8551 .vtc_flow = RTE_BE32(0xffffffff),
8558 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8559 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8568 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8570 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8572 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8574 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8576 flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
8581 size = sizeof(ipv6_m->hdr.dst_addr);
8582 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
8583 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
8584 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
8585 dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
8586 memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
8587 for (i = 0; i < size; ++i)
8588 l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
8589 l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
8590 src_ipv4_src_ipv6.ipv6_layout.ipv6);
8591 l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
8592 src_ipv4_src_ipv6.ipv6_layout.ipv6);
8593 memcpy(l24_m, ipv6_m->hdr.src_addr, size);
8594 for (i = 0; i < size; ++i)
8595 l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
8597 vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
8598 vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
8599 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
8600 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
8601 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
8602 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
8605 MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
8607 MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
8610 MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
8612 MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
8616 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
8618 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
8619 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
8621 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
8622 ipv6_m->hdr.hop_limits);
8623 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
8624 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
8625 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag,
8626 !!(ipv6_m->has_frag_ext));
8627 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
8628 !!(ipv6_v->has_frag_ext & ipv6_m->has_frag_ext));
8632 * Add IPV6 fragment extension item to matcher and to the value.
8634 * @param[in, out] matcher
8636 * @param[in, out] key
8637 * Flow matcher value.
8639 * Flow pattern to translate.
8641 * Item is inner pattern.
8644 flow_dv_translate_item_ipv6_frag_ext(void *matcher, void *key,
8645 const struct rte_flow_item *item,
8648 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_m = item->mask;
8649 const struct rte_flow_item_ipv6_frag_ext *ipv6_frag_ext_v = item->spec;
8650 const struct rte_flow_item_ipv6_frag_ext nic_mask = {
8652 .next_header = 0xff,
8653 .frag_data = RTE_BE16(0xffff),
8660 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8662 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8664 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8666 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8668 /* IPv6 fragment extension item exists, so packet is IP fragment. */
8669 MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
8670 MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 1);
8671 if (!ipv6_frag_ext_v)
8673 if (!ipv6_frag_ext_m)
8674 ipv6_frag_ext_m = &nic_mask;
8675 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
8676 ipv6_frag_ext_m->hdr.next_header);
8677 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
8678 ipv6_frag_ext_v->hdr.next_header &
8679 ipv6_frag_ext_m->hdr.next_header);
8683 * Add TCP item to matcher and to the value.
8685 * @param[in, out] matcher
8687 * @param[in, out] key
8688 * Flow matcher value.
8690 * Flow pattern to translate.
8692 * Item is inner pattern.
8695 flow_dv_translate_item_tcp(void *matcher, void *key,
8696 const struct rte_flow_item *item,
8699 const struct rte_flow_item_tcp *tcp_m = item->mask;
8700 const struct rte_flow_item_tcp *tcp_v = item->spec;
8705 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8707 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8709 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8711 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8713 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
8714 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
8718 tcp_m = &rte_flow_item_tcp_mask;
8719 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
8720 rte_be_to_cpu_16(tcp_m->hdr.src_port));
8721 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
8722 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
8723 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
8724 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
8725 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
8726 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
8727 MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
8728 tcp_m->hdr.tcp_flags);
8729 MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
8730 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
8734 * Add UDP item to matcher and to the value.
8736 * @param[in, out] matcher
8738 * @param[in, out] key
8739 * Flow matcher value.
8741 * Flow pattern to translate.
8743 * Item is inner pattern.
8746 flow_dv_translate_item_udp(void *matcher, void *key,
8747 const struct rte_flow_item *item,
8750 const struct rte_flow_item_udp *udp_m = item->mask;
8751 const struct rte_flow_item_udp *udp_v = item->spec;
8756 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8758 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
8760 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
8762 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8764 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
8765 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
8769 udp_m = &rte_flow_item_udp_mask;
8770 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
8771 rte_be_to_cpu_16(udp_m->hdr.src_port));
8772 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
8773 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
8774 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
8775 rte_be_to_cpu_16(udp_m->hdr.dst_port));
8776 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
8777 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
8781 * Add GRE optional Key item to matcher and to the value.
8783 * @param[in, out] matcher
8785 * @param[in, out] key
8786 * Flow matcher value.
8788 * Flow pattern to translate.
8790 * Item is inner pattern.
8793 flow_dv_translate_item_gre_key(void *matcher, void *key,
8794 const struct rte_flow_item *item)
8796 const rte_be32_t *key_m = item->mask;
8797 const rte_be32_t *key_v = item->spec;
8798 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8799 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8800 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
8802 /* GRE K bit must be on and should already be validated */
8803 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
8804 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
8808 key_m = &gre_key_default_mask;
8809 MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
8810 rte_be_to_cpu_32(*key_m) >> 8);
8811 MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
8812 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
8813 MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
8814 rte_be_to_cpu_32(*key_m) & 0xFF);
8815 MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
8816 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
8820 * Add GRE item to matcher and to the value.
8822 * @param[in, out] matcher
8824 * @param[in, out] key
8825 * Flow matcher value.
8827 * Flow pattern to translate.
8828 * @param[in] pattern_flags
8829 * Accumulated pattern flags.
8832 flow_dv_translate_item_gre(void *matcher, void *key,
8833 const struct rte_flow_item *item,
8834 uint64_t pattern_flags)
8836 static const struct rte_flow_item_gre empty_gre = {0,};
8837 const struct rte_flow_item_gre *gre_m = item->mask;
8838 const struct rte_flow_item_gre *gre_v = item->spec;
8839 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
8840 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
8841 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8842 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8849 uint16_t s_present:1;
8850 uint16_t k_present:1;
8851 uint16_t rsvd_bit1:1;
8852 uint16_t c_present:1;
8856 } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
8857 uint16_t protocol_m, protocol_v;
8859 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
8860 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
8866 gre_m = &rte_flow_item_gre_mask;
8868 gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
8869 gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
8870 MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
8871 gre_crks_rsvd0_ver_m.c_present);
8872 MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
8873 gre_crks_rsvd0_ver_v.c_present &
8874 gre_crks_rsvd0_ver_m.c_present);
8875 MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
8876 gre_crks_rsvd0_ver_m.k_present);
8877 MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
8878 gre_crks_rsvd0_ver_v.k_present &
8879 gre_crks_rsvd0_ver_m.k_present);
8880 MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
8881 gre_crks_rsvd0_ver_m.s_present);
8882 MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
8883 gre_crks_rsvd0_ver_v.s_present &
8884 gre_crks_rsvd0_ver_m.s_present);
8885 protocol_m = rte_be_to_cpu_16(gre_m->protocol);
8886 protocol_v = rte_be_to_cpu_16(gre_v->protocol);
8888 /* Force next protocol to prevent matchers duplication */
8889 protocol_v = mlx5_translate_tunnel_etypes(pattern_flags);
8891 protocol_m = 0xFFFF;
8893 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, protocol_m);
8894 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
8895 protocol_m & protocol_v);
8899 * Add GRE optional items to matcher and to the value.
8901 * @param[in, out] matcher
8903 * @param[in, out] key
8904 * Flow matcher value.
8906 * Flow pattern to translate.
8907 * @param[in] gre_item
8908 * Pointer to gre_item.
8909 * @param[in] pattern_flags
8910 * Accumulated pattern flags.
8913 flow_dv_translate_item_gre_option(void *matcher, void *key,
8914 const struct rte_flow_item *item,
8915 const struct rte_flow_item *gre_item,
8916 uint64_t pattern_flags)
8918 const struct rte_flow_item_gre_opt *option_m = item->mask;
8919 const struct rte_flow_item_gre_opt *option_v = item->spec;
8920 const struct rte_flow_item_gre *gre_m = gre_item->mask;
8921 const struct rte_flow_item_gre *gre_v = gre_item->spec;
8922 static const struct rte_flow_item_gre empty_gre = {0};
8923 struct rte_flow_item gre_key_item;
8924 uint16_t c_rsvd0_ver_m, c_rsvd0_ver_v;
8925 uint16_t protocol_m, protocol_v;
8930 * If only match key field, keep using misc for matching.
8931 * If need to match checksum or sequence, using misc5 and do
8932 * not need using misc.
8934 if (!(option_m->sequence.sequence ||
8935 option_m->checksum_rsvd.checksum)) {
8936 flow_dv_translate_item_gre(matcher, key, gre_item,
8938 gre_key_item.spec = &option_v->key.key;
8939 gre_key_item.mask = &option_m->key.key;
8940 flow_dv_translate_item_gre_key(matcher, key, &gre_key_item);
8948 gre_m = &rte_flow_item_gre_mask;
8950 protocol_v = gre_v->protocol;
8951 protocol_m = gre_m->protocol;
8953 /* Force next protocol to prevent matchers duplication */
8954 uint16_t ether_type =
8955 mlx5_translate_tunnel_etypes(pattern_flags);
8957 protocol_v = rte_be_to_cpu_16(ether_type);
8958 protocol_m = UINT16_MAX;
8961 c_rsvd0_ver_v = gre_v->c_rsvd0_ver;
8962 c_rsvd0_ver_m = gre_m->c_rsvd0_ver;
8963 if (option_m->sequence.sequence) {
8964 c_rsvd0_ver_v |= RTE_BE16(0x1000);
8965 c_rsvd0_ver_m |= RTE_BE16(0x1000);
8967 if (option_m->key.key) {
8968 c_rsvd0_ver_v |= RTE_BE16(0x2000);
8969 c_rsvd0_ver_m |= RTE_BE16(0x2000);
8971 if (option_m->checksum_rsvd.checksum) {
8972 c_rsvd0_ver_v |= RTE_BE16(0x8000);
8973 c_rsvd0_ver_m |= RTE_BE16(0x8000);
8976 * Hardware parses GRE optional field into the fixed location,
8977 * do not need to adjust the tunnel dword indices.
8979 misc5_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_5);
8980 misc5_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_5);
8981 MLX5_SET(fte_match_set_misc5, misc5_v, tunnel_header_0,
8982 rte_be_to_cpu_32((c_rsvd0_ver_v | protocol_v << 16) &
8983 (c_rsvd0_ver_m | protocol_m << 16)));
8984 MLX5_SET(fte_match_set_misc5, misc5_m, tunnel_header_0,
8985 rte_be_to_cpu_32(c_rsvd0_ver_m | protocol_m << 16));
8986 MLX5_SET(fte_match_set_misc5, misc5_v, tunnel_header_1,
8987 rte_be_to_cpu_32(option_v->checksum_rsvd.checksum &
8988 option_m->checksum_rsvd.checksum));
8989 MLX5_SET(fte_match_set_misc5, misc5_m, tunnel_header_1,
8990 rte_be_to_cpu_32(option_m->checksum_rsvd.checksum));
8991 MLX5_SET(fte_match_set_misc5, misc5_v, tunnel_header_2,
8992 rte_be_to_cpu_32(option_v->key.key & option_m->key.key));
8993 MLX5_SET(fte_match_set_misc5, misc5_m, tunnel_header_2,
8994 rte_be_to_cpu_32(option_m->key.key));
8995 MLX5_SET(fte_match_set_misc5, misc5_v, tunnel_header_3,
8996 rte_be_to_cpu_32(option_v->sequence.sequence &
8997 option_m->sequence.sequence));
8998 MLX5_SET(fte_match_set_misc5, misc5_m, tunnel_header_3,
8999 rte_be_to_cpu_32(option_m->sequence.sequence));
9003 * Add NVGRE item to matcher and to the value.
9005 * @param[in, out] matcher
9007 * @param[in, out] key
9008 * Flow matcher value.
9010 * Flow pattern to translate.
9011 * @param[in] pattern_flags
9012 * Accumulated pattern flags.
9015 flow_dv_translate_item_nvgre(void *matcher, void *key,
9016 const struct rte_flow_item *item,
9017 unsigned long pattern_flags)
9019 const struct rte_flow_item_nvgre *nvgre_m = item->mask;
9020 const struct rte_flow_item_nvgre *nvgre_v = item->spec;
9021 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9022 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9023 const char *tni_flow_id_m;
9024 const char *tni_flow_id_v;
9030 /* For NVGRE, GRE header fields must be set with defined values. */
9031 const struct rte_flow_item_gre gre_spec = {
9032 .c_rsvd0_ver = RTE_BE16(0x2000),
9033 .protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
9035 const struct rte_flow_item_gre gre_mask = {
9036 .c_rsvd0_ver = RTE_BE16(0xB000),
9037 .protocol = RTE_BE16(UINT16_MAX),
9039 const struct rte_flow_item gre_item = {
9044 flow_dv_translate_item_gre(matcher, key, &gre_item, pattern_flags);
9048 nvgre_m = &rte_flow_item_nvgre_mask;
9049 tni_flow_id_m = (const char *)nvgre_m->tni;
9050 tni_flow_id_v = (const char *)nvgre_v->tni;
9051 size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
9052 gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
9053 gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
9054 memcpy(gre_key_m, tni_flow_id_m, size);
9055 for (i = 0; i < size; ++i)
9056 gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
9060 * Add VXLAN item to matcher and to the value.
9063 * Pointer to the Ethernet device structure.
9065 * Flow rule attributes.
9066 * @param[in, out] matcher
9068 * @param[in, out] key
9069 * Flow matcher value.
9071 * Flow pattern to translate.
9073 * Item is inner pattern.
9076 flow_dv_translate_item_vxlan(struct rte_eth_dev *dev,
9077 const struct rte_flow_attr *attr,
9078 void *matcher, void *key,
9079 const struct rte_flow_item *item,
9082 const struct rte_flow_item_vxlan *vxlan_m = item->mask;
9083 const struct rte_flow_item_vxlan *vxlan_v = item->spec;
9088 uint32_t *tunnel_header_v;
9089 uint32_t *tunnel_header_m;
9091 struct mlx5_priv *priv = dev->data->dev_private;
9092 const struct rte_flow_item_vxlan nic_mask = {
9093 .vni = "\xff\xff\xff",
9098 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9100 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
9102 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9104 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9106 dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
9107 MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
9108 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
9109 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
9110 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
9112 dport = MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport);
9116 if ((!attr->group && !priv->sh->tunnel_header_0_1) ||
9117 (attr->group && !priv->sh->misc5_cap))
9118 vxlan_m = &rte_flow_item_vxlan_mask;
9120 vxlan_m = &nic_mask;
9122 if ((priv->sh->steering_format_version ==
9123 MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 &&
9124 dport != MLX5_UDP_PORT_VXLAN) ||
9125 (!attr->group && !attr->transfer && !priv->sh->tunnel_header_0_1) ||
9126 ((attr->group || attr->transfer) && !priv->sh->misc5_cap)) {
9133 misc_m = MLX5_ADDR_OF(fte_match_param,
9134 matcher, misc_parameters);
9135 misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9136 size = sizeof(vxlan_m->vni);
9137 vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
9138 vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
9139 memcpy(vni_m, vxlan_m->vni, size);
9140 for (i = 0; i < size; ++i)
9141 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
9144 misc5_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_5);
9145 misc5_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_5);
9146 tunnel_header_v = (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc5,
9149 tunnel_header_m = (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc5,
9152 *tunnel_header_v = (vxlan_v->vni[0] & vxlan_m->vni[0]) |
9153 (vxlan_v->vni[1] & vxlan_m->vni[1]) << 8 |
9154 (vxlan_v->vni[2] & vxlan_m->vni[2]) << 16;
9155 if (*tunnel_header_v)
9156 *tunnel_header_m = vxlan_m->vni[0] |
9157 vxlan_m->vni[1] << 8 |
9158 vxlan_m->vni[2] << 16;
9160 *tunnel_header_m = 0x0;
9161 *tunnel_header_v |= (vxlan_v->rsvd1 & vxlan_m->rsvd1) << 24;
9162 if (vxlan_v->rsvd1 & vxlan_m->rsvd1)
9163 *tunnel_header_m |= vxlan_m->rsvd1 << 24;
9167 * Add VXLAN-GPE item to matcher and to the value.
9169 * @param[in, out] matcher
9171 * @param[in, out] key
9172 * Flow matcher value.
9174 * Flow pattern to translate.
9176 * Item is inner pattern.
9180 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
9181 const struct rte_flow_item *item,
9182 const uint64_t pattern_flags)
9184 static const struct rte_flow_item_vxlan_gpe dummy_vxlan_gpe_hdr = {0, };
9185 const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
9186 const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
9187 /* The item was validated to be on the outer side */
9188 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
9189 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9191 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
9193 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9195 MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
9197 MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
9198 int i, size = sizeof(vxlan_m->vni);
9199 uint8_t flags_m = 0xff;
9200 uint8_t flags_v = 0xc;
9201 uint8_t m_protocol, v_protocol;
9203 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
9204 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
9205 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
9206 MLX5_UDP_PORT_VXLAN_GPE);
9209 vxlan_v = &dummy_vxlan_gpe_hdr;
9210 vxlan_m = &dummy_vxlan_gpe_hdr;
9213 vxlan_m = &rte_flow_item_vxlan_gpe_mask;
9215 memcpy(vni_m, vxlan_m->vni, size);
9216 for (i = 0; i < size; ++i)
9217 vni_v[i] = vni_m[i] & vxlan_v->vni[i];
9218 if (vxlan_m->flags) {
9219 flags_m = vxlan_m->flags;
9220 flags_v = vxlan_v->flags;
9222 MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
9223 MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
9224 m_protocol = vxlan_m->protocol;
9225 v_protocol = vxlan_v->protocol;
9227 /* Force next protocol to ensure next headers parsing. */
9228 if (pattern_flags & MLX5_FLOW_LAYER_INNER_L2)
9229 v_protocol = RTE_VXLAN_GPE_TYPE_ETH;
9230 else if (pattern_flags & MLX5_FLOW_LAYER_INNER_L3_IPV4)
9231 v_protocol = RTE_VXLAN_GPE_TYPE_IPV4;
9232 else if (pattern_flags & MLX5_FLOW_LAYER_INNER_L3_IPV6)
9233 v_protocol = RTE_VXLAN_GPE_TYPE_IPV6;
9237 MLX5_SET(fte_match_set_misc3, misc_m,
9238 outer_vxlan_gpe_next_protocol, m_protocol);
9239 MLX5_SET(fte_match_set_misc3, misc_v,
9240 outer_vxlan_gpe_next_protocol, m_protocol & v_protocol);
9244 * Add Geneve item to matcher and to the value.
9246 * @param[in, out] matcher
9248 * @param[in, out] key
9249 * Flow matcher value.
9251 * Flow pattern to translate.
9253 * Item is inner pattern.
9257 flow_dv_translate_item_geneve(void *matcher, void *key,
9258 const struct rte_flow_item *item,
9259 uint64_t pattern_flags)
9261 static const struct rte_flow_item_geneve empty_geneve = {0,};
9262 const struct rte_flow_item_geneve *geneve_m = item->mask;
9263 const struct rte_flow_item_geneve *geneve_v = item->spec;
9264 /* GENEVE flow item validation allows single tunnel item */
9265 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
9266 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9267 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9268 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9271 char *vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
9272 char *vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
9273 size_t size = sizeof(geneve_m->vni), i;
9274 uint16_t protocol_m, protocol_v;
9276 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
9277 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
9278 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
9279 MLX5_UDP_PORT_GENEVE);
9282 geneve_v = &empty_geneve;
9283 geneve_m = &empty_geneve;
9286 geneve_m = &rte_flow_item_geneve_mask;
9288 memcpy(vni_m, geneve_m->vni, size);
9289 for (i = 0; i < size; ++i)
9290 vni_v[i] = vni_m[i] & geneve_v->vni[i];
9291 gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
9292 gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
9293 MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
9294 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
9295 MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
9296 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
9297 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
9298 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
9299 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
9300 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
9301 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
9302 protocol_m = rte_be_to_cpu_16(geneve_m->protocol);
9303 protocol_v = rte_be_to_cpu_16(geneve_v->protocol);
9305 /* Force next protocol to prevent matchers duplication */
9306 protocol_v = mlx5_translate_tunnel_etypes(pattern_flags);
9308 protocol_m = 0xFFFF;
9310 MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type, protocol_m);
9311 MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
9312 protocol_m & protocol_v);
9316 * Create Geneve TLV option resource.
9318 * @param dev[in, out]
9319 * Pointer to rte_eth_dev structure.
9320 * @param[in, out] tag_be24
9321 * Tag value in big endian then R-shift 8.
9322 * @parm[in, out] dev_flow
9323 * Pointer to the dev_flow.
9325 * pointer to error structure.
9328 * 0 on success otherwise -errno and errno is set.
9332 flow_dev_geneve_tlv_option_resource_register(struct rte_eth_dev *dev,
9333 const struct rte_flow_item *item,
9334 struct rte_flow_error *error)
9336 struct mlx5_priv *priv = dev->data->dev_private;
9337 struct mlx5_dev_ctx_shared *sh = priv->sh;
9338 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource =
9339 sh->geneve_tlv_option_resource;
9340 struct mlx5_devx_obj *obj;
9341 const struct rte_flow_item_geneve_opt *geneve_opt_v = item->spec;
9346 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
9347 if (geneve_opt_resource != NULL) {
9348 if (geneve_opt_resource->option_class ==
9349 geneve_opt_v->option_class &&
9350 geneve_opt_resource->option_type ==
9351 geneve_opt_v->option_type &&
9352 geneve_opt_resource->length ==
9353 geneve_opt_v->option_len) {
9354 /* We already have GENEVE TLV option obj allocated. */
9355 __atomic_fetch_add(&geneve_opt_resource->refcnt, 1,
9358 ret = rte_flow_error_set(error, ENOMEM,
9359 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9360 "Only one GENEVE TLV option supported");
9364 /* Create a GENEVE TLV object and resource. */
9365 obj = mlx5_devx_cmd_create_geneve_tlv_option(sh->cdev->ctx,
9366 geneve_opt_v->option_class,
9367 geneve_opt_v->option_type,
9368 geneve_opt_v->option_len);
9370 ret = rte_flow_error_set(error, ENODATA,
9371 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9372 "Failed to create GENEVE TLV Devx object");
9375 sh->geneve_tlv_option_resource =
9376 mlx5_malloc(MLX5_MEM_ZERO,
9377 sizeof(*geneve_opt_resource),
9379 if (!sh->geneve_tlv_option_resource) {
9380 claim_zero(mlx5_devx_cmd_destroy(obj));
9381 ret = rte_flow_error_set(error, ENOMEM,
9382 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9383 "GENEVE TLV object memory allocation failed");
9386 geneve_opt_resource = sh->geneve_tlv_option_resource;
9387 geneve_opt_resource->obj = obj;
9388 geneve_opt_resource->option_class = geneve_opt_v->option_class;
9389 geneve_opt_resource->option_type = geneve_opt_v->option_type;
9390 geneve_opt_resource->length = geneve_opt_v->option_len;
9391 __atomic_store_n(&geneve_opt_resource->refcnt, 1,
9395 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
9400 * Add Geneve TLV option item to matcher.
9402 * @param[in, out] dev
9403 * Pointer to rte_eth_dev structure.
9404 * @param[in, out] matcher
9406 * @param[in, out] key
9407 * Flow matcher value.
9409 * Flow pattern to translate.
9411 * Pointer to error structure.
9414 flow_dv_translate_item_geneve_opt(struct rte_eth_dev *dev, void *matcher,
9415 void *key, const struct rte_flow_item *item,
9416 struct rte_flow_error *error)
9418 const struct rte_flow_item_geneve_opt *geneve_opt_m = item->mask;
9419 const struct rte_flow_item_geneve_opt *geneve_opt_v = item->spec;
9420 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9421 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9422 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
9424 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9425 rte_be32_t opt_data_key = 0, opt_data_mask = 0;
9431 geneve_opt_m = &rte_flow_item_geneve_opt_mask;
9432 ret = flow_dev_geneve_tlv_option_resource_register(dev, item,
9435 DRV_LOG(ERR, "Failed to create geneve_tlv_obj");
9439 * Set the option length in GENEVE header if not requested.
9440 * The GENEVE TLV option length is expressed by the option length field
9441 * in the GENEVE header.
9442 * If the option length was not requested but the GENEVE TLV option item
9443 * is present we set the option length field implicitly.
9445 if (!MLX5_GET16(fte_match_set_misc, misc_m, geneve_opt_len)) {
9446 MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
9447 MLX5_GENEVE_OPTLEN_MASK);
9448 MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
9449 geneve_opt_v->option_len + 1);
9451 MLX5_SET(fte_match_set_misc, misc_m, geneve_tlv_option_0_exist, 1);
9452 MLX5_SET(fte_match_set_misc, misc_v, geneve_tlv_option_0_exist, 1);
9454 if (geneve_opt_v->data) {
9455 memcpy(&opt_data_key, geneve_opt_v->data,
9456 RTE_MIN((uint32_t)(geneve_opt_v->option_len * 4),
9457 sizeof(opt_data_key)));
9458 MLX5_ASSERT((uint32_t)(geneve_opt_v->option_len * 4) <=
9459 sizeof(opt_data_key));
9460 memcpy(&opt_data_mask, geneve_opt_m->data,
9461 RTE_MIN((uint32_t)(geneve_opt_v->option_len * 4),
9462 sizeof(opt_data_mask)));
9463 MLX5_ASSERT((uint32_t)(geneve_opt_v->option_len * 4) <=
9464 sizeof(opt_data_mask));
9465 MLX5_SET(fte_match_set_misc3, misc3_m,
9466 geneve_tlv_option_0_data,
9467 rte_be_to_cpu_32(opt_data_mask));
9468 MLX5_SET(fte_match_set_misc3, misc3_v,
9469 geneve_tlv_option_0_data,
9470 rte_be_to_cpu_32(opt_data_key & opt_data_mask));
9476 * Add MPLS item to matcher and to the value.
9478 * @param[in, out] matcher
9480 * @param[in, out] key
9481 * Flow matcher value.
9483 * Flow pattern to translate.
9484 * @param[in] prev_layer
9485 * The protocol layer indicated in previous item.
9487 * Item is inner pattern.
9490 flow_dv_translate_item_mpls(void *matcher, void *key,
9491 const struct rte_flow_item *item,
9492 uint64_t prev_layer,
9495 const uint32_t *in_mpls_m = item->mask;
9496 const uint32_t *in_mpls_v = item->spec;
9497 uint32_t *out_mpls_m = 0;
9498 uint32_t *out_mpls_v = 0;
9499 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9500 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9501 void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
9503 void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
9504 void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
9505 void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9507 switch (prev_layer) {
9508 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
9509 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
9510 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
9512 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
9513 MLX5_UDP_PORT_MPLS);
9516 case MLX5_FLOW_LAYER_GRE:
9518 case MLX5_FLOW_LAYER_GRE_KEY:
9519 if (!MLX5_GET16(fte_match_set_misc, misc_v, gre_protocol)) {
9520 MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
9522 MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
9523 RTE_ETHER_TYPE_MPLS);
9532 in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
9533 switch (prev_layer) {
9534 case MLX5_FLOW_LAYER_OUTER_L4_UDP:
9536 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
9537 outer_first_mpls_over_udp);
9539 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
9540 outer_first_mpls_over_udp);
9542 case MLX5_FLOW_LAYER_GRE:
9544 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
9545 outer_first_mpls_over_gre);
9547 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
9548 outer_first_mpls_over_gre);
9551 /* Inner MPLS not over GRE is not supported. */
9554 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
9558 (uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
9564 if (out_mpls_m && out_mpls_v) {
9565 *out_mpls_m = *in_mpls_m;
9566 *out_mpls_v = *in_mpls_v & *in_mpls_m;
9571 * Add metadata register item to matcher
9573 * @param[in, out] matcher
9575 * @param[in, out] key
9576 * Flow matcher value.
9577 * @param[in] reg_type
9578 * Type of device metadata register
9585 flow_dv_match_meta_reg(void *matcher, void *key,
9586 enum modify_reg reg_type,
9587 uint32_t data, uint32_t mask)
9590 MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
9592 MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
9598 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
9599 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
9602 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
9603 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
9607 * The metadata register C0 field might be divided into
9608 * source vport index and META item value, we should set
9609 * this field according to specified mask, not as whole one.
9611 temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
9613 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
9614 temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
9617 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
9620 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
9621 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
9624 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
9625 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
9628 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
9629 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
9632 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
9633 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
9636 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
9637 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
9640 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
9641 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
9644 MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
9645 MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
9654 * Add MARK item to matcher
9657 * The device to configure through.
9658 * @param[in, out] matcher
9660 * @param[in, out] key
9661 * Flow matcher value.
9663 * Flow pattern to translate.
9666 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
9667 void *matcher, void *key,
9668 const struct rte_flow_item *item)
9670 struct mlx5_priv *priv = dev->data->dev_private;
9671 const struct rte_flow_item_mark *mark;
9675 mark = item->mask ? (const void *)item->mask :
9676 &rte_flow_item_mark_mask;
9677 mask = mark->id & priv->sh->dv_mark_mask;
9678 mark = (const void *)item->spec;
9680 value = mark->id & priv->sh->dv_mark_mask & mask;
9682 enum modify_reg reg;
9684 /* Get the metadata register index for the mark. */
9685 reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
9686 MLX5_ASSERT(reg > 0);
9687 if (reg == REG_C_0) {
9688 struct mlx5_priv *priv = dev->data->dev_private;
9689 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
9690 uint32_t shl_c0 = rte_bsf32(msk_c0);
9696 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
9701 * Add META item to matcher
9704 * The devich to configure through.
9705 * @param[in, out] matcher
9707 * @param[in, out] key
9708 * Flow matcher value.
9710 * Attributes of flow that includes this item.
9712 * Flow pattern to translate.
9715 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
9716 void *matcher, void *key,
9717 const struct rte_flow_attr *attr,
9718 const struct rte_flow_item *item)
9720 const struct rte_flow_item_meta *meta_m;
9721 const struct rte_flow_item_meta *meta_v;
9723 meta_m = (const void *)item->mask;
9725 meta_m = &rte_flow_item_meta_mask;
9726 meta_v = (const void *)item->spec;
9729 uint32_t value = meta_v->data;
9730 uint32_t mask = meta_m->data;
9732 reg = flow_dv_get_metadata_reg(dev, attr, NULL);
9735 MLX5_ASSERT(reg != REG_NON);
9736 if (reg == REG_C_0) {
9737 struct mlx5_priv *priv = dev->data->dev_private;
9738 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
9739 uint32_t shl_c0 = rte_bsf32(msk_c0);
9745 flow_dv_match_meta_reg(matcher, key, reg, value, mask);
9750 * Add vport metadata Reg C0 item to matcher
9752 * @param[in, out] matcher
9754 * @param[in, out] key
9755 * Flow matcher value.
9757 * Flow pattern to translate.
9760 flow_dv_translate_item_meta_vport(void *matcher, void *key,
9761 uint32_t value, uint32_t mask)
9763 flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
9767 * Add tag item to matcher
9770 * The devich to configure through.
9771 * @param[in, out] matcher
9773 * @param[in, out] key
9774 * Flow matcher value.
9776 * Flow pattern to translate.
9779 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
9780 void *matcher, void *key,
9781 const struct rte_flow_item *item)
9783 const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
9784 const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
9785 uint32_t mask, value;
9788 value = tag_v->data;
9789 mask = tag_m ? tag_m->data : UINT32_MAX;
9790 if (tag_v->id == REG_C_0) {
9791 struct mlx5_priv *priv = dev->data->dev_private;
9792 uint32_t msk_c0 = priv->sh->dv_regc0_mask;
9793 uint32_t shl_c0 = rte_bsf32(msk_c0);
9799 flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
9803 * Add TAG item to matcher
9806 * The devich to configure through.
9807 * @param[in, out] matcher
9809 * @param[in, out] key
9810 * Flow matcher value.
9812 * Flow pattern to translate.
9815 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
9816 void *matcher, void *key,
9817 const struct rte_flow_item *item)
9819 const struct rte_flow_item_tag *tag_v = item->spec;
9820 const struct rte_flow_item_tag *tag_m = item->mask;
9821 enum modify_reg reg;
9824 tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
9825 /* Get the metadata register index for the tag. */
9826 reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
9827 MLX5_ASSERT(reg > 0);
9828 flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
9832 * Add source vport match to the specified matcher.
9834 * @param[in, out] matcher
9836 * @param[in, out] key
9837 * Flow matcher value.
9839 * Source vport value to match
9844 flow_dv_translate_item_source_vport(void *matcher, void *key,
9845 int16_t port, uint16_t mask)
9847 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
9848 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
9850 MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
9851 MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
9855 * Translate port-id item to eswitch match on port-id.
9858 * The devich to configure through.
9859 * @param[in, out] matcher
9861 * @param[in, out] key
9862 * Flow matcher value.
9864 * Flow pattern to translate.
9869 * 0 on success, a negative errno value otherwise.
9872 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
9873 void *key, const struct rte_flow_item *item,
9874 const struct rte_flow_attr *attr)
9876 const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
9877 const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
9878 struct mlx5_priv *priv;
9881 if (pid_v && pid_v->id == MLX5_PORT_ESW_MGR) {
9882 flow_dv_translate_item_source_vport(matcher, key,
9883 flow_dv_get_esw_manager_vport_id(dev), 0xffff);
9886 mask = pid_m ? pid_m->id : 0xffff;
9887 id = pid_v ? pid_v->id : dev->data->port_id;
9888 priv = mlx5_port_to_eswitch_info(id, item == NULL);
9892 * Translate to vport field or to metadata, depending on mode.
9893 * Kernel can use either misc.source_port or half of C0 metadata
9896 if (priv->vport_meta_mask) {
9898 * Provide the hint for SW steering library
9899 * to insert the flow into ingress domain and
9900 * save the extra vport match.
9902 if (mask == 0xffff && priv->vport_id == 0xffff &&
9903 priv->pf_bond < 0 && attr->transfer)
9904 flow_dv_translate_item_source_vport
9905 (matcher, key, priv->vport_id, mask);
9907 * We should always set the vport metadata register,
9908 * otherwise the SW steering library can drop
9909 * the rule if wire vport metadata value is not zero,
9910 * it depends on kernel configuration.
9912 flow_dv_translate_item_meta_vport(matcher, key,
9913 priv->vport_meta_tag,
9914 priv->vport_meta_mask);
9916 flow_dv_translate_item_source_vport(matcher, key,
9917 priv->vport_id, mask);
9923 * Add ICMP6 item to matcher and to the value.
9925 * @param[in, out] matcher
9927 * @param[in, out] key
9928 * Flow matcher value.
9930 * Flow pattern to translate.
9932 * Item is inner pattern.
9935 flow_dv_translate_item_icmp6(void *matcher, void *key,
9936 const struct rte_flow_item *item,
9939 const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
9940 const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
9943 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
9945 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9947 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9949 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
9951 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9953 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
9955 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
9956 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
9960 icmp6_m = &rte_flow_item_icmp6_mask;
9961 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
9962 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
9963 icmp6_v->type & icmp6_m->type);
9964 MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
9965 MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
9966 icmp6_v->code & icmp6_m->code);
9970 * Add ICMP item to matcher and to the value.
9972 * @param[in, out] matcher
9974 * @param[in, out] key
9975 * Flow matcher value.
9977 * Flow pattern to translate.
9979 * Item is inner pattern.
9982 flow_dv_translate_item_icmp(void *matcher, void *key,
9983 const struct rte_flow_item *item,
9986 const struct rte_flow_item_icmp *icmp_m = item->mask;
9987 const struct rte_flow_item_icmp *icmp_v = item->spec;
9988 uint32_t icmp_header_data_m = 0;
9989 uint32_t icmp_header_data_v = 0;
9992 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
9994 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
9996 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
9998 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
10000 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
10002 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
10004 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
10005 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
10009 icmp_m = &rte_flow_item_icmp_mask;
10010 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
10011 icmp_m->hdr.icmp_type);
10012 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
10013 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
10014 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
10015 icmp_m->hdr.icmp_code);
10016 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
10017 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
10018 icmp_header_data_m = rte_be_to_cpu_16(icmp_m->hdr.icmp_seq_nb);
10019 icmp_header_data_m |= rte_be_to_cpu_16(icmp_m->hdr.icmp_ident) << 16;
10020 if (icmp_header_data_m) {
10021 icmp_header_data_v = rte_be_to_cpu_16(icmp_v->hdr.icmp_seq_nb);
10022 icmp_header_data_v |=
10023 rte_be_to_cpu_16(icmp_v->hdr.icmp_ident) << 16;
10024 MLX5_SET(fte_match_set_misc3, misc3_m, icmp_header_data,
10025 icmp_header_data_m);
10026 MLX5_SET(fte_match_set_misc3, misc3_v, icmp_header_data,
10027 icmp_header_data_v & icmp_header_data_m);
10032 * Add GTP item to matcher and to the value.
10034 * @param[in, out] matcher
10036 * @param[in, out] key
10037 * Flow matcher value.
10039 * Flow pattern to translate.
10041 * Item is inner pattern.
10044 flow_dv_translate_item_gtp(void *matcher, void *key,
10045 const struct rte_flow_item *item, int inner)
10047 const struct rte_flow_item_gtp *gtp_m = item->mask;
10048 const struct rte_flow_item_gtp *gtp_v = item->spec;
10051 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
10052 misc_parameters_3);
10053 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
10054 uint16_t dport = RTE_GTPU_UDP_PORT;
10057 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
10059 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
10061 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
10063 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
10065 if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
10066 MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
10067 MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
10072 gtp_m = &rte_flow_item_gtp_mask;
10073 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
10074 gtp_m->v_pt_rsv_flags);
10075 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
10076 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
10077 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
10078 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
10079 gtp_v->msg_type & gtp_m->msg_type);
10080 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
10081 rte_be_to_cpu_32(gtp_m->teid));
10082 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
10083 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
10087 * Add GTP PSC item to matcher.
10089 * @param[in, out] matcher
10091 * @param[in, out] key
10092 * Flow matcher value.
10094 * Flow pattern to translate.
10097 flow_dv_translate_item_gtp_psc(void *matcher, void *key,
10098 const struct rte_flow_item *item)
10100 const struct rte_flow_item_gtp_psc *gtp_psc_m = item->mask;
10101 const struct rte_flow_item_gtp_psc *gtp_psc_v = item->spec;
10102 void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
10103 misc_parameters_3);
10104 void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
10110 uint8_t next_ext_header_type;
10115 /* Always set E-flag match on one, regardless of GTP item settings. */
10116 gtp_flags = MLX5_GET(fte_match_set_misc3, misc3_m, gtpu_msg_flags);
10117 gtp_flags |= MLX5_GTP_EXT_HEADER_FLAG;
10118 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags, gtp_flags);
10119 gtp_flags = MLX5_GET(fte_match_set_misc3, misc3_v, gtpu_msg_flags);
10120 gtp_flags |= MLX5_GTP_EXT_HEADER_FLAG;
10121 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags, gtp_flags);
10122 /*Set next extension header type. */
10125 dw_2.next_ext_header_type = 0xff;
10126 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_dw_2,
10127 rte_cpu_to_be_32(dw_2.w32));
10130 dw_2.next_ext_header_type = 0x85;
10131 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_dw_2,
10132 rte_cpu_to_be_32(dw_2.w32));
10138 uint8_t type_flags;
10144 /*Set extension header PDU type and Qos. */
10146 gtp_psc_m = &rte_flow_item_gtp_psc_mask;
10148 dw_0.type_flags = MLX5_GTP_PDU_TYPE_SHIFT(gtp_psc_m->hdr.type);
10149 dw_0.qfi = gtp_psc_m->hdr.qfi;
10150 MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_first_ext_dw_0,
10151 rte_cpu_to_be_32(dw_0.w32));
10153 dw_0.type_flags = MLX5_GTP_PDU_TYPE_SHIFT(gtp_psc_v->hdr.type &
10154 gtp_psc_m->hdr.type);
10155 dw_0.qfi = gtp_psc_v->hdr.qfi & gtp_psc_m->hdr.qfi;
10156 MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_first_ext_dw_0,
10157 rte_cpu_to_be_32(dw_0.w32));
10163 * Add eCPRI item to matcher and to the value.
10166 * The devich to configure through.
10167 * @param[in, out] matcher
10169 * @param[in, out] key
10170 * Flow matcher value.
10172 * Flow pattern to translate.
10173 * @param[in] last_item
10177 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
10178 void *key, const struct rte_flow_item *item,
10179 uint64_t last_item)
10181 struct mlx5_priv *priv = dev->data->dev_private;
10182 const struct rte_flow_item_ecpri *ecpri_m = item->mask;
10183 const struct rte_flow_item_ecpri *ecpri_v = item->spec;
10184 struct rte_ecpri_common_hdr common;
10185 void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
10186 misc_parameters_4);
10187 void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
10193 * In case of eCPRI over Ethernet, if EtherType is not specified,
10194 * match on eCPRI EtherType implicitly.
10196 if (last_item & MLX5_FLOW_LAYER_OUTER_L2) {
10197 void *hdrs_m, *hdrs_v, *l2m, *l2v;
10199 hdrs_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
10200 hdrs_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
10201 l2m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_m, ethertype);
10202 l2v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, hdrs_v, ethertype);
10203 if (*(uint16_t *)l2m == 0 && *(uint16_t *)l2v == 0) {
10204 *(uint16_t *)l2m = UINT16_MAX;
10205 *(uint16_t *)l2v = RTE_BE16(RTE_ETHER_TYPE_ECPRI);
10211 ecpri_m = &rte_flow_item_ecpri_mask;
10213 * Maximal four DW samples are supported in a single matching now.
10214 * Two are used now for a eCPRI matching:
10215 * 1. Type: one byte, mask should be 0x00ff0000 in network order
10216 * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
10219 if (!ecpri_m->hdr.common.u32)
10221 samples = priv->sh->ecpri_parser.ids;
10222 /* Need to take the whole DW as the mask to fill the entry. */
10223 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
10224 prog_sample_field_value_0);
10225 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
10226 prog_sample_field_value_0);
10227 /* Already big endian (network order) in the header. */
10228 *(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
10229 *(uint32_t *)dw_v = ecpri_v->hdr.common.u32 & ecpri_m->hdr.common.u32;
10230 /* Sample#0, used for matching type, offset 0. */
10231 MLX5_SET(fte_match_set_misc4, misc4_m,
10232 prog_sample_field_id_0, samples[0]);
10233 /* It makes no sense to set the sample ID in the mask field. */
10234 MLX5_SET(fte_match_set_misc4, misc4_v,
10235 prog_sample_field_id_0, samples[0]);
10237 * Checking if message body part needs to be matched.
10238 * Some wildcard rules only matching type field should be supported.
10240 if (ecpri_m->hdr.dummy[0]) {
10241 common.u32 = rte_be_to_cpu_32(ecpri_v->hdr.common.u32);
10242 switch (common.type) {
10243 case RTE_ECPRI_MSG_TYPE_IQ_DATA:
10244 case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
10245 case RTE_ECPRI_MSG_TYPE_DLY_MSR:
10246 dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
10247 prog_sample_field_value_1);
10248 dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
10249 prog_sample_field_value_1);
10250 *(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
10251 *(uint32_t *)dw_v = ecpri_v->hdr.dummy[0] &
10252 ecpri_m->hdr.dummy[0];
10253 /* Sample#1, to match message body, offset 4. */
10254 MLX5_SET(fte_match_set_misc4, misc4_m,
10255 prog_sample_field_id_1, samples[1]);
10256 MLX5_SET(fte_match_set_misc4, misc4_v,
10257 prog_sample_field_id_1, samples[1]);
10260 /* Others, do not match any sample ID. */
10267 * Add connection tracking status item to matcher
10270 * The devich to configure through.
10271 * @param[in, out] matcher
10273 * @param[in, out] key
10274 * Flow matcher value.
10276 * Flow pattern to translate.
10279 flow_dv_translate_item_aso_ct(struct rte_eth_dev *dev,
10280 void *matcher, void *key,
10281 const struct rte_flow_item *item)
10283 uint32_t reg_value = 0;
10285 /* 8LSB 0b 11/0000/11, middle 4 bits are reserved. */
10286 uint32_t reg_mask = 0;
10287 const struct rte_flow_item_conntrack *spec = item->spec;
10288 const struct rte_flow_item_conntrack *mask = item->mask;
10290 struct rte_flow_error error;
10293 mask = &rte_flow_item_conntrack_mask;
10294 if (!spec || !mask->flags)
10296 flags = spec->flags & mask->flags;
10297 /* The conflict should be checked in the validation. */
10298 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_VALID)
10299 reg_value |= MLX5_CT_SYNDROME_VALID;
10300 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_CHANGED)
10301 reg_value |= MLX5_CT_SYNDROME_STATE_CHANGE;
10302 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_INVALID)
10303 reg_value |= MLX5_CT_SYNDROME_INVALID;
10304 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_DISABLED)
10305 reg_value |= MLX5_CT_SYNDROME_TRAP;
10306 if (flags & RTE_FLOW_CONNTRACK_PKT_STATE_BAD)
10307 reg_value |= MLX5_CT_SYNDROME_BAD_PACKET;
10308 if (mask->flags & (RTE_FLOW_CONNTRACK_PKT_STATE_VALID |
10309 RTE_FLOW_CONNTRACK_PKT_STATE_INVALID |
10310 RTE_FLOW_CONNTRACK_PKT_STATE_DISABLED))
10312 if (mask->flags & RTE_FLOW_CONNTRACK_PKT_STATE_CHANGED)
10313 reg_mask |= MLX5_CT_SYNDROME_STATE_CHANGE;
10314 if (mask->flags & RTE_FLOW_CONNTRACK_PKT_STATE_BAD)
10315 reg_mask |= MLX5_CT_SYNDROME_BAD_PACKET;
10316 /* The REG_C_x value could be saved during startup. */
10317 reg_id = mlx5_flow_get_reg_id(dev, MLX5_ASO_CONNTRACK, 0, &error);
10318 if (reg_id == REG_NON)
10320 flow_dv_match_meta_reg(matcher, key, (enum modify_reg)reg_id,
10321 reg_value, reg_mask);
10325 flow_dv_translate_item_flex(struct rte_eth_dev *dev, void *matcher, void *key,
10326 const struct rte_flow_item *item,
10327 struct mlx5_flow *dev_flow, bool is_inner)
10329 const struct rte_flow_item_flex *spec =
10330 (const struct rte_flow_item_flex *)item->spec;
10331 int index = mlx5_flex_acquire_index(dev, spec->handle, false);
10333 MLX5_ASSERT(index >= 0 && index <= (int)(sizeof(uint32_t) * CHAR_BIT));
10336 if (!(dev_flow->handle->flex_item & RTE_BIT32(index))) {
10337 /* Don't count both inner and outer flex items in one rule. */
10338 if (mlx5_flex_acquire_index(dev, spec->handle, true) != index)
10339 MLX5_ASSERT(false);
10340 dev_flow->handle->flex_item |= (uint8_t)RTE_BIT32(index);
10342 mlx5_flex_flow_translate_item(dev, matcher, key, item, is_inner);
10345 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
10347 #define HEADER_IS_ZERO(match_criteria, headers) \
10348 !(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
10349 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
10352 * Calculate flow matcher enable bitmap.
10354 * @param match_criteria
10355 * Pointer to flow matcher criteria.
10358 * Bitmap of enabled fields.
10361 flow_dv_matcher_enable(uint32_t *match_criteria)
10363 uint8_t match_criteria_enable;
10365 match_criteria_enable =
10366 (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
10367 MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
10368 match_criteria_enable |=
10369 (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
10370 MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
10371 match_criteria_enable |=
10372 (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
10373 MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
10374 match_criteria_enable |=
10375 (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
10376 MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
10377 match_criteria_enable |=
10378 (!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
10379 MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
10380 match_criteria_enable |=
10381 (!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
10382 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
10383 match_criteria_enable |=
10384 (!HEADER_IS_ZERO(match_criteria, misc_parameters_5)) <<
10385 MLX5_MATCH_CRITERIA_ENABLE_MISC5_BIT;
10386 return match_criteria_enable;
10390 __flow_dv_adjust_buf_size(size_t *size, uint8_t match_criteria)
10393 * Check flow matching criteria first, subtract misc5/4 length if flow
10394 * doesn't own misc5/4 parameters. In some old rdma-core releases,
10395 * misc5/4 are not supported, and matcher creation failure is expected
10396 * w/o subtraction. If misc5 is provided, misc4 must be counted in since
10397 * misc5 is right after misc4.
10399 if (!(match_criteria & (1 << MLX5_MATCH_CRITERIA_ENABLE_MISC5_BIT))) {
10400 *size = MLX5_ST_SZ_BYTES(fte_match_param) -
10401 MLX5_ST_SZ_BYTES(fte_match_set_misc5);
10402 if (!(match_criteria & (1 <<
10403 MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT))) {
10404 *size -= MLX5_ST_SZ_BYTES(fte_match_set_misc4);
10409 static struct mlx5_list_entry *
10410 flow_dv_matcher_clone_cb(void *tool_ctx __rte_unused,
10411 struct mlx5_list_entry *entry, void *cb_ctx)
10413 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10414 struct mlx5_flow_dv_matcher *ref = ctx->data;
10415 struct mlx5_flow_tbl_data_entry *tbl = container_of(ref->tbl,
10416 typeof(*tbl), tbl);
10417 struct mlx5_flow_dv_matcher *resource = mlx5_malloc(MLX5_MEM_ANY,
10422 rte_flow_error_set(ctx->error, ENOMEM,
10423 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10424 "cannot create matcher");
10427 memcpy(resource, entry, sizeof(*resource));
10428 resource->tbl = &tbl->tbl;
10429 return &resource->entry;
10433 flow_dv_matcher_clone_free_cb(void *tool_ctx __rte_unused,
10434 struct mlx5_list_entry *entry)
10439 struct mlx5_list_entry *
10440 flow_dv_tbl_create_cb(void *tool_ctx, void *cb_ctx)
10442 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10443 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10444 struct rte_eth_dev *dev = ctx->dev;
10445 struct mlx5_flow_tbl_data_entry *tbl_data;
10446 struct mlx5_flow_tbl_tunnel_prm *tt_prm = ctx->data2;
10447 struct rte_flow_error *error = ctx->error;
10448 union mlx5_flow_tbl_key key = { .v64 = *(uint64_t *)(ctx->data) };
10449 struct mlx5_flow_tbl_resource *tbl;
10454 tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
10456 rte_flow_error_set(error, ENOMEM,
10457 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10459 "cannot allocate flow table data entry");
10462 tbl_data->idx = idx;
10463 tbl_data->tunnel = tt_prm->tunnel;
10464 tbl_data->group_id = tt_prm->group_id;
10465 tbl_data->external = !!tt_prm->external;
10466 tbl_data->tunnel_offload = is_tunnel_offload_active(dev);
10467 tbl_data->is_egress = !!key.is_egress;
10468 tbl_data->is_transfer = !!key.is_fdb;
10469 tbl_data->dummy = !!key.dummy;
10470 tbl_data->level = key.level;
10471 tbl_data->id = key.id;
10472 tbl = &tbl_data->tbl;
10474 return &tbl_data->entry;
10476 domain = sh->fdb_domain;
10477 else if (key.is_egress)
10478 domain = sh->tx_domain;
10480 domain = sh->rx_domain;
10481 ret = mlx5_flow_os_create_flow_tbl(domain, key.level, &tbl->obj);
10483 rte_flow_error_set(error, ENOMEM,
10484 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10485 NULL, "cannot create flow table object");
10486 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
10489 if (key.level != 0) {
10490 ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
10491 (tbl->obj, &tbl_data->jump.action);
10493 rte_flow_error_set(error, ENOMEM,
10494 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10496 "cannot create flow jump action");
10497 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
10498 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
10502 MKSTR(matcher_name, "%s_%s_%u_%u_matcher_list",
10503 key.is_fdb ? "FDB" : "NIC", key.is_egress ? "egress" : "ingress",
10504 key.level, key.id);
10505 tbl_data->matchers = mlx5_list_create(matcher_name, sh, true,
10506 flow_dv_matcher_create_cb,
10507 flow_dv_matcher_match_cb,
10508 flow_dv_matcher_remove_cb,
10509 flow_dv_matcher_clone_cb,
10510 flow_dv_matcher_clone_free_cb);
10511 if (!tbl_data->matchers) {
10512 rte_flow_error_set(error, ENOMEM,
10513 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10515 "cannot create tbl matcher list");
10516 mlx5_flow_os_destroy_flow_action(tbl_data->jump.action);
10517 mlx5_flow_os_destroy_flow_tbl(tbl->obj);
10518 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
10521 return &tbl_data->entry;
10525 flow_dv_tbl_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
10528 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10529 struct mlx5_flow_tbl_data_entry *tbl_data =
10530 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
10531 union mlx5_flow_tbl_key key = { .v64 = *(uint64_t *)(ctx->data) };
10533 return tbl_data->level != key.level ||
10534 tbl_data->id != key.id ||
10535 tbl_data->dummy != key.dummy ||
10536 tbl_data->is_transfer != !!key.is_fdb ||
10537 tbl_data->is_egress != !!key.is_egress;
10540 struct mlx5_list_entry *
10541 flow_dv_tbl_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
10544 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10545 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10546 struct mlx5_flow_tbl_data_entry *tbl_data;
10547 struct rte_flow_error *error = ctx->error;
10550 tbl_data = mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
10552 rte_flow_error_set(error, ENOMEM,
10553 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10555 "cannot allocate flow table data entry");
10558 memcpy(tbl_data, oentry, sizeof(*tbl_data));
10559 tbl_data->idx = idx;
10560 return &tbl_data->entry;
10564 flow_dv_tbl_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
10566 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10567 struct mlx5_flow_tbl_data_entry *tbl_data =
10568 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
10570 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], tbl_data->idx);
10574 * Get a flow table.
10576 * @param[in, out] dev
10577 * Pointer to rte_eth_dev structure.
10578 * @param[in] table_level
10579 * Table level to use.
10580 * @param[in] egress
10581 * Direction of the table.
10582 * @param[in] transfer
10583 * E-Switch or NIC flow.
10585 * Dummy entry for dv API.
10586 * @param[in] table_id
10588 * @param[out] error
10589 * pointer to error structure.
10592 * Returns tables resource based on the index, NULL in case of failed.
10594 struct mlx5_flow_tbl_resource *
10595 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
10596 uint32_t table_level, uint8_t egress,
10599 const struct mlx5_flow_tunnel *tunnel,
10600 uint32_t group_id, uint8_t dummy,
10602 struct rte_flow_error *error)
10604 struct mlx5_priv *priv = dev->data->dev_private;
10605 union mlx5_flow_tbl_key table_key = {
10607 .level = table_level,
10611 .is_fdb = !!transfer,
10612 .is_egress = !!egress,
10615 struct mlx5_flow_tbl_tunnel_prm tt_prm = {
10617 .group_id = group_id,
10618 .external = external,
10620 struct mlx5_flow_cb_ctx ctx = {
10623 .data = &table_key.v64,
10626 struct mlx5_list_entry *entry;
10627 struct mlx5_flow_tbl_data_entry *tbl_data;
10629 entry = mlx5_hlist_register(priv->sh->flow_tbls, table_key.v64, &ctx);
10631 rte_flow_error_set(error, ENOMEM,
10632 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10633 "cannot get table");
10636 DRV_LOG(DEBUG, "table_level %u table_id %u "
10637 "tunnel %u group %u registered.",
10638 table_level, table_id,
10639 tunnel ? tunnel->tunnel_id : 0, group_id);
10640 tbl_data = container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
10641 return &tbl_data->tbl;
10645 flow_dv_tbl_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
10647 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10648 struct mlx5_flow_tbl_data_entry *tbl_data =
10649 container_of(entry, struct mlx5_flow_tbl_data_entry, entry);
10651 MLX5_ASSERT(entry && sh);
10652 if (tbl_data->jump.action)
10653 mlx5_flow_os_destroy_flow_action(tbl_data->jump.action);
10654 if (tbl_data->tbl.obj)
10655 mlx5_flow_os_destroy_flow_tbl(tbl_data->tbl.obj);
10656 if (tbl_data->tunnel_offload && tbl_data->external) {
10657 struct mlx5_list_entry *he;
10658 struct mlx5_hlist *tunnel_grp_hash;
10659 struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
10660 union tunnel_tbl_key tunnel_key = {
10661 .tunnel_id = tbl_data->tunnel ?
10662 tbl_data->tunnel->tunnel_id : 0,
10663 .group = tbl_data->group_id
10665 uint32_t table_level = tbl_data->level;
10666 struct mlx5_flow_cb_ctx ctx = {
10667 .data = (void *)&tunnel_key.val,
10670 tunnel_grp_hash = tbl_data->tunnel ?
10671 tbl_data->tunnel->groups :
10673 he = mlx5_hlist_lookup(tunnel_grp_hash, tunnel_key.val, &ctx);
10675 mlx5_hlist_unregister(tunnel_grp_hash, he);
10677 "table_level %u id %u tunnel %u group %u released.",
10681 tbl_data->tunnel->tunnel_id : 0,
10682 tbl_data->group_id);
10684 if (tbl_data->matchers)
10685 mlx5_list_destroy(tbl_data->matchers);
10686 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], tbl_data->idx);
10690 * Release a flow table.
10693 * Pointer to device shared structure.
10695 * Table resource to be released.
10698 * Returns 0 if table was released, else return 1;
10701 flow_dv_tbl_resource_release(struct mlx5_dev_ctx_shared *sh,
10702 struct mlx5_flow_tbl_resource *tbl)
10704 struct mlx5_flow_tbl_data_entry *tbl_data =
10705 container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
10709 return mlx5_hlist_unregister(sh->flow_tbls, &tbl_data->entry);
10713 flow_dv_matcher_match_cb(void *tool_ctx __rte_unused,
10714 struct mlx5_list_entry *entry, void *cb_ctx)
10716 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10717 struct mlx5_flow_dv_matcher *ref = ctx->data;
10718 struct mlx5_flow_dv_matcher *cur = container_of(entry, typeof(*cur),
10721 return cur->crc != ref->crc ||
10722 cur->priority != ref->priority ||
10723 memcmp((const void *)cur->mask.buf,
10724 (const void *)ref->mask.buf, ref->mask.size);
10727 struct mlx5_list_entry *
10728 flow_dv_matcher_create_cb(void *tool_ctx, void *cb_ctx)
10730 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10731 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10732 struct mlx5_flow_dv_matcher *ref = ctx->data;
10733 struct mlx5_flow_dv_matcher *resource;
10734 struct mlx5dv_flow_matcher_attr dv_attr = {
10735 .type = IBV_FLOW_ATTR_NORMAL,
10736 .match_mask = (void *)&ref->mask,
10738 struct mlx5_flow_tbl_data_entry *tbl = container_of(ref->tbl,
10739 typeof(*tbl), tbl);
10742 resource = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*resource), 0,
10745 rte_flow_error_set(ctx->error, ENOMEM,
10746 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10747 "cannot create matcher");
10751 dv_attr.match_criteria_enable =
10752 flow_dv_matcher_enable(resource->mask.buf);
10753 __flow_dv_adjust_buf_size(&ref->mask.size,
10754 dv_attr.match_criteria_enable);
10755 dv_attr.priority = ref->priority;
10756 if (tbl->is_egress)
10757 dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
10758 ret = mlx5_flow_os_create_flow_matcher(sh->cdev->ctx, &dv_attr,
10760 &resource->matcher_object);
10762 mlx5_free(resource);
10763 rte_flow_error_set(ctx->error, ENOMEM,
10764 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10765 "cannot create matcher");
10768 return &resource->entry;
10772 * Register the flow matcher.
10774 * @param[in, out] dev
10775 * Pointer to rte_eth_dev structure.
10776 * @param[in, out] matcher
10777 * Pointer to flow matcher.
10778 * @param[in, out] key
10779 * Pointer to flow table key.
10780 * @parm[in, out] dev_flow
10781 * Pointer to the dev_flow.
10782 * @param[out] error
10783 * pointer to error structure.
10786 * 0 on success otherwise -errno and errno is set.
10789 flow_dv_matcher_register(struct rte_eth_dev *dev,
10790 struct mlx5_flow_dv_matcher *ref,
10791 union mlx5_flow_tbl_key *key,
10792 struct mlx5_flow *dev_flow,
10793 const struct mlx5_flow_tunnel *tunnel,
10795 struct rte_flow_error *error)
10797 struct mlx5_list_entry *entry;
10798 struct mlx5_flow_dv_matcher *resource;
10799 struct mlx5_flow_tbl_resource *tbl;
10800 struct mlx5_flow_tbl_data_entry *tbl_data;
10801 struct mlx5_flow_cb_ctx ctx = {
10806 * tunnel offload API requires this registration for cases when
10807 * tunnel match rule was inserted before tunnel set rule.
10809 tbl = flow_dv_tbl_resource_get(dev, key->level,
10810 key->is_egress, key->is_fdb,
10811 dev_flow->external, tunnel,
10812 group_id, 0, key->id, error);
10814 return -rte_errno; /* No need to refill the error info */
10815 tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
10817 entry = mlx5_list_register(tbl_data->matchers, &ctx);
10819 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
10820 return rte_flow_error_set(error, ENOMEM,
10821 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10822 "cannot allocate ref memory");
10824 resource = container_of(entry, typeof(*resource), entry);
10825 dev_flow->handle->dvh.matcher = resource;
10829 struct mlx5_list_entry *
10830 flow_dv_tag_create_cb(void *tool_ctx, void *cb_ctx)
10832 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10833 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10834 struct mlx5_flow_dv_tag_resource *entry;
10838 entry = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG], &idx);
10840 rte_flow_error_set(ctx->error, ENOMEM,
10841 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10842 "cannot allocate resource memory");
10846 entry->tag_id = *(uint32_t *)(ctx->data);
10847 ret = mlx5_flow_os_create_flow_action_tag(entry->tag_id,
10850 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], idx);
10851 rte_flow_error_set(ctx->error, ENOMEM,
10852 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10853 NULL, "cannot create action");
10856 return &entry->entry;
10860 flow_dv_tag_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
10863 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10864 struct mlx5_flow_dv_tag_resource *tag =
10865 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
10867 return *(uint32_t *)(ctx->data) != tag->tag_id;
10870 struct mlx5_list_entry *
10871 flow_dv_tag_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
10874 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10875 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10876 struct mlx5_flow_dv_tag_resource *entry;
10879 entry = mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TAG], &idx);
10881 rte_flow_error_set(ctx->error, ENOMEM,
10882 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10883 "cannot allocate tag resource memory");
10886 memcpy(entry, oentry, sizeof(*entry));
10888 return &entry->entry;
10892 flow_dv_tag_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
10894 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10895 struct mlx5_flow_dv_tag_resource *tag =
10896 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
10898 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], tag->idx);
10902 * Find existing tag resource or create and register a new one.
10904 * @param dev[in, out]
10905 * Pointer to rte_eth_dev structure.
10906 * @param[in, out] tag_be24
10907 * Tag value in big endian then R-shift 8.
10908 * @parm[in, out] dev_flow
10909 * Pointer to the dev_flow.
10910 * @param[out] error
10911 * pointer to error structure.
10914 * 0 on success otherwise -errno and errno is set.
10917 flow_dv_tag_resource_register
10918 (struct rte_eth_dev *dev,
10920 struct mlx5_flow *dev_flow,
10921 struct rte_flow_error *error)
10923 struct mlx5_priv *priv = dev->data->dev_private;
10924 struct mlx5_flow_dv_tag_resource *resource;
10925 struct mlx5_list_entry *entry;
10926 struct mlx5_flow_cb_ctx ctx = {
10930 struct mlx5_hlist *tag_table;
10932 tag_table = flow_dv_hlist_prepare(priv->sh, &priv->sh->tag_table,
10934 MLX5_TAGS_HLIST_ARRAY_SIZE,
10935 false, false, priv->sh,
10936 flow_dv_tag_create_cb,
10937 flow_dv_tag_match_cb,
10938 flow_dv_tag_remove_cb,
10939 flow_dv_tag_clone_cb,
10940 flow_dv_tag_clone_free_cb,
10942 if (unlikely(!tag_table))
10944 entry = mlx5_hlist_register(tag_table, tag_be24, &ctx);
10946 resource = container_of(entry, struct mlx5_flow_dv_tag_resource,
10948 dev_flow->handle->dvh.rix_tag = resource->idx;
10949 dev_flow->dv.tag_resource = resource;
10956 flow_dv_tag_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
10958 struct mlx5_dev_ctx_shared *sh = tool_ctx;
10959 struct mlx5_flow_dv_tag_resource *tag =
10960 container_of(entry, struct mlx5_flow_dv_tag_resource, entry);
10962 MLX5_ASSERT(tag && sh && tag->action);
10963 claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
10964 DRV_LOG(DEBUG, "Tag %p: removed.", (void *)tag);
10965 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TAG], tag->idx);
10972 * Pointer to Ethernet device.
10977 * 1 while a reference on it exists, 0 when freed.
10980 flow_dv_tag_release(struct rte_eth_dev *dev,
10983 struct mlx5_priv *priv = dev->data->dev_private;
10984 struct mlx5_flow_dv_tag_resource *tag;
10986 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
10989 DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
10990 dev->data->port_id, (void *)tag, tag->entry.ref_cnt);
10991 return mlx5_hlist_unregister(priv->sh->tag_table, &tag->entry);
10995 * Translate action PORT_ID / REPRESENTED_PORT to vport.
10998 * Pointer to rte_eth_dev structure.
10999 * @param[in] action
11000 * Pointer to action PORT_ID / REPRESENTED_PORT.
11001 * @param[out] dst_port_id
11002 * The target port ID.
11003 * @param[out] error
11004 * Pointer to the error structure.
11007 * 0 on success, a negative errno value otherwise and rte_errno is set.
11010 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
11011 const struct rte_flow_action *action,
11012 uint32_t *dst_port_id,
11013 struct rte_flow_error *error)
11016 struct mlx5_priv *priv;
11018 switch (action->type) {
11019 case RTE_FLOW_ACTION_TYPE_PORT_ID: {
11020 const struct rte_flow_action_port_id *conf;
11022 conf = (const struct rte_flow_action_port_id *)action->conf;
11023 port = conf->original ? dev->data->port_id : conf->id;
11026 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT: {
11027 const struct rte_flow_action_ethdev *ethdev;
11029 ethdev = (const struct rte_flow_action_ethdev *)action->conf;
11030 port = ethdev->port_id;
11034 MLX5_ASSERT(false);
11035 return rte_flow_error_set(error, EINVAL,
11036 RTE_FLOW_ERROR_TYPE_ACTION, action,
11037 "unknown E-Switch action");
11040 priv = mlx5_port_to_eswitch_info(port, false);
11042 return rte_flow_error_set(error, -rte_errno,
11043 RTE_FLOW_ERROR_TYPE_ACTION,
11045 "No eswitch info was found for port");
11046 #ifdef HAVE_MLX5DV_DR_CREATE_DEST_IB_PORT
11048 * This parameter is transferred to
11049 * mlx5dv_dr_action_create_dest_ib_port().
11051 *dst_port_id = priv->dev_port;
11054 * Legacy mode, no LAG configurations is supported.
11055 * This parameter is transferred to
11056 * mlx5dv_dr_action_create_dest_vport().
11058 *dst_port_id = priv->vport_id;
11064 * Create a counter with aging configuration.
11067 * Pointer to rte_eth_dev structure.
11068 * @param[in] dev_flow
11069 * Pointer to the mlx5_flow.
11070 * @param[out] count
11071 * Pointer to the counter action configuration.
11073 * Pointer to the aging action configuration.
11076 * Index to flow counter on success, 0 otherwise.
11079 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
11080 struct mlx5_flow *dev_flow,
11081 const struct rte_flow_action_count *count
11083 const struct rte_flow_action_age *age)
11086 struct mlx5_age_param *age_param;
11088 counter = flow_dv_counter_alloc(dev, !!age);
11089 if (!counter || age == NULL)
11091 age_param = flow_dv_counter_idx_get_age(dev, counter);
11092 age_param->context = age->context ? age->context :
11093 (void *)(uintptr_t)(dev_flow->flow_idx);
11094 age_param->timeout = age->timeout;
11095 age_param->port_id = dev->data->port_id;
11096 __atomic_store_n(&age_param->sec_since_last_hit, 0, __ATOMIC_RELAXED);
11097 __atomic_store_n(&age_param->state, AGE_CANDIDATE, __ATOMIC_RELAXED);
11102 * Add Tx queue matcher
11105 * Pointer to the dev struct.
11106 * @param[in, out] matcher
11108 * @param[in, out] key
11109 * Flow matcher value.
11111 * Flow pattern to translate.
11113 * Item is inner pattern.
11116 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
11117 void *matcher, void *key,
11118 const struct rte_flow_item *item)
11120 const struct mlx5_rte_flow_item_tx_queue *queue_m;
11121 const struct mlx5_rte_flow_item_tx_queue *queue_v;
11122 void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
11123 void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
11124 struct mlx5_txq_ctrl *txq;
11125 uint32_t queue, mask;
11127 queue_m = (const void *)item->mask;
11128 queue_v = (const void *)item->spec;
11131 txq = mlx5_txq_get(dev, queue_v->queue);
11134 if (txq->is_hairpin)
11135 queue = txq->obj->sq->id;
11137 queue = txq->obj->sq_obj.sq->id;
11138 mask = queue_m == NULL ? UINT32_MAX : queue_m->queue;
11139 MLX5_SET(fte_match_set_misc, misc_m, source_sqn, mask);
11140 MLX5_SET(fte_match_set_misc, misc_v, source_sqn, queue & mask);
11141 mlx5_txq_release(dev, queue_v->queue);
11145 * Set the hash fields according to the @p flow information.
11147 * @param[in] item_flags
11148 * The match pattern item flags.
11149 * @param[in] rss_desc
11150 * Pointer to the mlx5_flow_rss_desc.
11151 * @param[out] hash_fields
11152 * Pointer to the RSS hash fields.
11155 flow_dv_hashfields_set(uint64_t item_flags,
11156 struct mlx5_flow_rss_desc *rss_desc,
11157 uint64_t *hash_fields)
11159 uint64_t items = item_flags;
11160 uint64_t fields = 0;
11162 uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
11165 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
11166 if (rss_desc->level >= 2)
11169 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
11170 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4)) ||
11172 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
11173 if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
11174 fields |= IBV_RX_HASH_SRC_IPV4;
11175 else if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
11176 fields |= IBV_RX_HASH_DST_IPV4;
11178 fields |= MLX5_IPV4_IBV_RX_HASH;
11180 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
11181 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6)) ||
11183 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
11184 if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
11185 fields |= IBV_RX_HASH_SRC_IPV6;
11186 else if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
11187 fields |= IBV_RX_HASH_DST_IPV6;
11189 fields |= MLX5_IPV6_IBV_RX_HASH;
11194 * There is no match between the RSS types and the
11195 * L3 protocol (IPv4/IPv6) defined in the flow rule.
11198 if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
11199 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP)) ||
11201 if (rss_types & RTE_ETH_RSS_UDP) {
11202 if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
11203 fields |= IBV_RX_HASH_SRC_PORT_UDP;
11204 else if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
11205 fields |= IBV_RX_HASH_DST_PORT_UDP;
11207 fields |= MLX5_UDP_IBV_RX_HASH;
11209 } else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
11210 (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP)) ||
11212 if (rss_types & RTE_ETH_RSS_TCP) {
11213 if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
11214 fields |= IBV_RX_HASH_SRC_PORT_TCP;
11215 else if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
11216 fields |= IBV_RX_HASH_DST_PORT_TCP;
11218 fields |= MLX5_TCP_IBV_RX_HASH;
11222 fields |= IBV_RX_HASH_INNER;
11223 *hash_fields = fields;
11227 * Prepare an Rx Hash queue.
11230 * Pointer to Ethernet device.
11231 * @param[in] dev_flow
11232 * Pointer to the mlx5_flow.
11233 * @param[in] rss_desc
11234 * Pointer to the mlx5_flow_rss_desc.
11235 * @param[out] hrxq_idx
11236 * Hash Rx queue index.
11239 * The Verbs/DevX object initialised, NULL otherwise and rte_errno is set.
11241 static struct mlx5_hrxq *
11242 flow_dv_hrxq_prepare(struct rte_eth_dev *dev,
11243 struct mlx5_flow *dev_flow,
11244 struct mlx5_flow_rss_desc *rss_desc,
11245 uint32_t *hrxq_idx)
11247 struct mlx5_flow_handle *dh = dev_flow->handle;
11248 uint32_t shared_rss = rss_desc->shared_rss;
11249 struct mlx5_hrxq *hrxq;
11251 MLX5_ASSERT(rss_desc->queue_num);
11252 rss_desc->key_len = MLX5_RSS_HASH_KEY_LEN;
11253 rss_desc->hash_fields = dev_flow->hash_fields;
11254 rss_desc->tunnel = !!(dh->layers & MLX5_FLOW_LAYER_TUNNEL);
11255 rss_desc->shared_rss = 0;
11256 if (rss_desc->hash_fields == 0)
11257 rss_desc->queue_num = 1;
11258 hrxq = mlx5_hrxq_get(dev, rss_desc);
11259 *hrxq_idx = hrxq ? hrxq->idx : 0;
11260 rss_desc->shared_rss = shared_rss;
11265 * Release sample sub action resource.
11267 * @param[in, out] dev
11268 * Pointer to rte_eth_dev structure.
11269 * @param[in] act_res
11270 * Pointer to sample sub action resource.
11273 flow_dv_sample_sub_actions_release(struct rte_eth_dev *dev,
11274 struct mlx5_flow_sub_actions_idx *act_res)
11276 if (act_res->rix_hrxq) {
11277 mlx5_hrxq_release(dev, act_res->rix_hrxq);
11278 act_res->rix_hrxq = 0;
11280 if (act_res->rix_encap_decap) {
11281 flow_dv_encap_decap_resource_release(dev,
11282 act_res->rix_encap_decap);
11283 act_res->rix_encap_decap = 0;
11285 if (act_res->rix_port_id_action) {
11286 flow_dv_port_id_action_resource_release(dev,
11287 act_res->rix_port_id_action);
11288 act_res->rix_port_id_action = 0;
11290 if (act_res->rix_tag) {
11291 flow_dv_tag_release(dev, act_res->rix_tag);
11292 act_res->rix_tag = 0;
11294 if (act_res->rix_jump) {
11295 flow_dv_jump_tbl_resource_release(dev, act_res->rix_jump);
11296 act_res->rix_jump = 0;
11301 flow_dv_sample_match_cb(void *tool_ctx __rte_unused,
11302 struct mlx5_list_entry *entry, void *cb_ctx)
11304 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11305 struct rte_eth_dev *dev = ctx->dev;
11306 struct mlx5_flow_dv_sample_resource *ctx_resource = ctx->data;
11307 struct mlx5_flow_dv_sample_resource *resource = container_of(entry,
11311 if (ctx_resource->ratio == resource->ratio &&
11312 ctx_resource->ft_type == resource->ft_type &&
11313 ctx_resource->ft_id == resource->ft_id &&
11314 ctx_resource->set_action == resource->set_action &&
11315 !memcmp((void *)&ctx_resource->sample_act,
11316 (void *)&resource->sample_act,
11317 sizeof(struct mlx5_flow_sub_actions_list))) {
11319 * Existing sample action should release the prepared
11320 * sub-actions reference counter.
11322 flow_dv_sample_sub_actions_release(dev,
11323 &ctx_resource->sample_idx);
11329 struct mlx5_list_entry *
11330 flow_dv_sample_create_cb(void *tool_ctx __rte_unused, void *cb_ctx)
11332 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11333 struct rte_eth_dev *dev = ctx->dev;
11334 struct mlx5_flow_dv_sample_resource *ctx_resource = ctx->data;
11335 void **sample_dv_actions = ctx_resource->sub_actions;
11336 struct mlx5_flow_dv_sample_resource *resource;
11337 struct mlx5dv_dr_flow_sampler_attr sampler_attr;
11338 struct mlx5_priv *priv = dev->data->dev_private;
11339 struct mlx5_dev_ctx_shared *sh = priv->sh;
11340 struct mlx5_flow_tbl_resource *tbl;
11342 const uint32_t next_ft_step = 1;
11343 uint32_t next_ft_id = ctx_resource->ft_id + next_ft_step;
11344 uint8_t is_egress = 0;
11345 uint8_t is_transfer = 0;
11346 struct rte_flow_error *error = ctx->error;
11348 /* Register new sample resource. */
11349 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE], &idx);
11351 rte_flow_error_set(error, ENOMEM,
11352 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11354 "cannot allocate resource memory");
11357 *resource = *ctx_resource;
11358 /* Create normal path table level */
11359 if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
11361 else if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
11363 tbl = flow_dv_tbl_resource_get(dev, next_ft_id,
11364 is_egress, is_transfer,
11365 true, NULL, 0, 0, 0, error);
11367 rte_flow_error_set(error, ENOMEM,
11368 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11370 "fail to create normal path table "
11374 resource->normal_path_tbl = tbl;
11375 if (ctx_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
11376 if (!sh->default_miss_action) {
11377 rte_flow_error_set(error, ENOMEM,
11378 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11380 "default miss action was not "
11384 sample_dv_actions[ctx_resource->sample_act.actions_num++] =
11385 sh->default_miss_action;
11387 /* Create a DR sample action */
11388 sampler_attr.sample_ratio = resource->ratio;
11389 sampler_attr.default_next_table = tbl->obj;
11390 sampler_attr.num_sample_actions = ctx_resource->sample_act.actions_num;
11391 sampler_attr.sample_actions = (struct mlx5dv_dr_action **)
11392 &sample_dv_actions[0];
11393 sampler_attr.action = resource->set_action;
11394 if (mlx5_os_flow_dr_create_flow_action_sampler
11395 (&sampler_attr, &resource->verbs_action)) {
11396 rte_flow_error_set(error, ENOMEM,
11397 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11398 NULL, "cannot create sample action");
11401 resource->idx = idx;
11402 resource->dev = dev;
11403 return &resource->entry;
11405 if (resource->ft_type != MLX5DV_FLOW_TABLE_TYPE_FDB)
11406 flow_dv_sample_sub_actions_release(dev,
11407 &resource->sample_idx);
11408 if (resource->normal_path_tbl)
11409 flow_dv_tbl_resource_release(MLX5_SH(dev),
11410 resource->normal_path_tbl);
11411 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_SAMPLE], idx);
11416 struct mlx5_list_entry *
11417 flow_dv_sample_clone_cb(void *tool_ctx __rte_unused,
11418 struct mlx5_list_entry *entry __rte_unused,
11421 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11422 struct rte_eth_dev *dev = ctx->dev;
11423 struct mlx5_flow_dv_sample_resource *resource;
11424 struct mlx5_priv *priv = dev->data->dev_private;
11425 struct mlx5_dev_ctx_shared *sh = priv->sh;
11428 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE], &idx);
11430 rte_flow_error_set(ctx->error, ENOMEM,
11431 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11433 "cannot allocate resource memory");
11436 memcpy(resource, entry, sizeof(*resource));
11437 resource->idx = idx;
11438 resource->dev = dev;
11439 return &resource->entry;
11443 flow_dv_sample_clone_free_cb(void *tool_ctx __rte_unused,
11444 struct mlx5_list_entry *entry)
11446 struct mlx5_flow_dv_sample_resource *resource =
11447 container_of(entry, typeof(*resource), entry);
11448 struct rte_eth_dev *dev = resource->dev;
11449 struct mlx5_priv *priv = dev->data->dev_private;
11451 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE], resource->idx);
11455 * Find existing sample resource or create and register a new one.
11457 * @param[in, out] dev
11458 * Pointer to rte_eth_dev structure.
11460 * Pointer to sample resource reference.
11461 * @parm[in, out] dev_flow
11462 * Pointer to the dev_flow.
11463 * @param[out] error
11464 * pointer to error structure.
11467 * 0 on success otherwise -errno and errno is set.
11470 flow_dv_sample_resource_register(struct rte_eth_dev *dev,
11471 struct mlx5_flow_dv_sample_resource *ref,
11472 struct mlx5_flow *dev_flow,
11473 struct rte_flow_error *error)
11475 struct mlx5_flow_dv_sample_resource *resource;
11476 struct mlx5_list_entry *entry;
11477 struct mlx5_priv *priv = dev->data->dev_private;
11478 struct mlx5_flow_cb_ctx ctx = {
11484 entry = mlx5_list_register(priv->sh->sample_action_list, &ctx);
11487 resource = container_of(entry, typeof(*resource), entry);
11488 dev_flow->handle->dvh.rix_sample = resource->idx;
11489 dev_flow->dv.sample_res = resource;
11494 flow_dv_dest_array_match_cb(void *tool_ctx __rte_unused,
11495 struct mlx5_list_entry *entry, void *cb_ctx)
11497 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11498 struct mlx5_flow_dv_dest_array_resource *ctx_resource = ctx->data;
11499 struct rte_eth_dev *dev = ctx->dev;
11500 struct mlx5_flow_dv_dest_array_resource *resource =
11501 container_of(entry, typeof(*resource), entry);
11504 if (ctx_resource->num_of_dest == resource->num_of_dest &&
11505 ctx_resource->ft_type == resource->ft_type &&
11506 !memcmp((void *)resource->sample_act,
11507 (void *)ctx_resource->sample_act,
11508 (ctx_resource->num_of_dest *
11509 sizeof(struct mlx5_flow_sub_actions_list)))) {
11511 * Existing sample action should release the prepared
11512 * sub-actions reference counter.
11514 for (idx = 0; idx < ctx_resource->num_of_dest; idx++)
11515 flow_dv_sample_sub_actions_release(dev,
11516 &ctx_resource->sample_idx[idx]);
11522 struct mlx5_list_entry *
11523 flow_dv_dest_array_create_cb(void *tool_ctx __rte_unused, void *cb_ctx)
11525 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11526 struct rte_eth_dev *dev = ctx->dev;
11527 struct mlx5_flow_dv_dest_array_resource *resource;
11528 struct mlx5_flow_dv_dest_array_resource *ctx_resource = ctx->data;
11529 struct mlx5dv_dr_action_dest_attr *dest_attr[MLX5_MAX_DEST_NUM] = { 0 };
11530 struct mlx5dv_dr_action_dest_reformat dest_reformat[MLX5_MAX_DEST_NUM];
11531 struct mlx5_priv *priv = dev->data->dev_private;
11532 struct mlx5_dev_ctx_shared *sh = priv->sh;
11533 struct mlx5_flow_sub_actions_list *sample_act;
11534 struct mlx5dv_dr_domain *domain;
11535 uint32_t idx = 0, res_idx = 0;
11536 struct rte_flow_error *error = ctx->error;
11537 uint64_t action_flags;
11540 /* Register new destination array resource. */
11541 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
11544 rte_flow_error_set(error, ENOMEM,
11545 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11547 "cannot allocate resource memory");
11550 *resource = *ctx_resource;
11551 if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
11552 domain = sh->fdb_domain;
11553 else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
11554 domain = sh->rx_domain;
11556 domain = sh->tx_domain;
11557 for (idx = 0; idx < ctx_resource->num_of_dest; idx++) {
11558 dest_attr[idx] = (struct mlx5dv_dr_action_dest_attr *)
11559 mlx5_malloc(MLX5_MEM_ZERO,
11560 sizeof(struct mlx5dv_dr_action_dest_attr),
11562 if (!dest_attr[idx]) {
11563 rte_flow_error_set(error, ENOMEM,
11564 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11566 "cannot allocate resource memory");
11569 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST;
11570 sample_act = &ctx_resource->sample_act[idx];
11571 action_flags = sample_act->action_flags;
11572 switch (action_flags) {
11573 case MLX5_FLOW_ACTION_QUEUE:
11574 dest_attr[idx]->dest = sample_act->dr_queue_action;
11576 case (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_ENCAP):
11577 dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST_REFORMAT;
11578 dest_attr[idx]->dest_reformat = &dest_reformat[idx];
11579 dest_attr[idx]->dest_reformat->reformat =
11580 sample_act->dr_encap_action;
11581 dest_attr[idx]->dest_reformat->dest =
11582 sample_act->dr_port_id_action;
11584 case MLX5_FLOW_ACTION_PORT_ID:
11585 dest_attr[idx]->dest = sample_act->dr_port_id_action;
11587 case MLX5_FLOW_ACTION_JUMP:
11588 dest_attr[idx]->dest = sample_act->dr_jump_action;
11591 rte_flow_error_set(error, EINVAL,
11592 RTE_FLOW_ERROR_TYPE_ACTION,
11594 "unsupported actions type");
11598 /* create a dest array action */
11599 ret = mlx5_os_flow_dr_create_flow_action_dest_array
11601 resource->num_of_dest,
11603 &resource->action);
11605 rte_flow_error_set(error, ENOMEM,
11606 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11608 "cannot create destination array action");
11611 resource->idx = res_idx;
11612 resource->dev = dev;
11613 for (idx = 0; idx < ctx_resource->num_of_dest; idx++)
11614 mlx5_free(dest_attr[idx]);
11615 return &resource->entry;
11617 for (idx = 0; idx < ctx_resource->num_of_dest; idx++) {
11618 flow_dv_sample_sub_actions_release(dev,
11619 &resource->sample_idx[idx]);
11620 if (dest_attr[idx])
11621 mlx5_free(dest_attr[idx]);
11623 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DEST_ARRAY], res_idx);
11627 struct mlx5_list_entry *
11628 flow_dv_dest_array_clone_cb(void *tool_ctx __rte_unused,
11629 struct mlx5_list_entry *entry __rte_unused,
11632 struct mlx5_flow_cb_ctx *ctx = cb_ctx;
11633 struct rte_eth_dev *dev = ctx->dev;
11634 struct mlx5_flow_dv_dest_array_resource *resource;
11635 struct mlx5_priv *priv = dev->data->dev_private;
11636 struct mlx5_dev_ctx_shared *sh = priv->sh;
11637 uint32_t res_idx = 0;
11638 struct rte_flow_error *error = ctx->error;
11640 resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
11643 rte_flow_error_set(error, ENOMEM,
11644 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11646 "cannot allocate dest-array memory");
11649 memcpy(resource, entry, sizeof(*resource));
11650 resource->idx = res_idx;
11651 resource->dev = dev;
11652 return &resource->entry;
11656 flow_dv_dest_array_clone_free_cb(void *tool_ctx __rte_unused,
11657 struct mlx5_list_entry *entry)
11659 struct mlx5_flow_dv_dest_array_resource *resource =
11660 container_of(entry, typeof(*resource), entry);
11661 struct rte_eth_dev *dev = resource->dev;
11662 struct mlx5_priv *priv = dev->data->dev_private;
11664 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY], resource->idx);
11668 * Find existing destination array resource or create and register a new one.
11670 * @param[in, out] dev
11671 * Pointer to rte_eth_dev structure.
11673 * Pointer to destination array resource reference.
11674 * @parm[in, out] dev_flow
11675 * Pointer to the dev_flow.
11676 * @param[out] error
11677 * pointer to error structure.
11680 * 0 on success otherwise -errno and errno is set.
11683 flow_dv_dest_array_resource_register(struct rte_eth_dev *dev,
11684 struct mlx5_flow_dv_dest_array_resource *ref,
11685 struct mlx5_flow *dev_flow,
11686 struct rte_flow_error *error)
11688 struct mlx5_flow_dv_dest_array_resource *resource;
11689 struct mlx5_priv *priv = dev->data->dev_private;
11690 struct mlx5_list_entry *entry;
11691 struct mlx5_flow_cb_ctx ctx = {
11697 entry = mlx5_list_register(priv->sh->dest_array_list, &ctx);
11700 resource = container_of(entry, typeof(*resource), entry);
11701 dev_flow->handle->dvh.rix_dest_array = resource->idx;
11702 dev_flow->dv.dest_array_res = resource;
11707 * Convert Sample action to DV specification.
11710 * Pointer to rte_eth_dev structure.
11711 * @param[in] action
11712 * Pointer to sample action structure.
11713 * @param[in, out] dev_flow
11714 * Pointer to the mlx5_flow.
11716 * Pointer to the flow attributes.
11717 * @param[in, out] num_of_dest
11718 * Pointer to the num of destination.
11719 * @param[in, out] sample_actions
11720 * Pointer to sample actions list.
11721 * @param[in, out] res
11722 * Pointer to sample resource.
11723 * @param[out] error
11724 * Pointer to the error structure.
11727 * 0 on success, a negative errno value otherwise and rte_errno is set.
11730 flow_dv_translate_action_sample(struct rte_eth_dev *dev,
11731 const struct rte_flow_action_sample *action,
11732 struct mlx5_flow *dev_flow,
11733 const struct rte_flow_attr *attr,
11734 uint32_t *num_of_dest,
11735 void **sample_actions,
11736 struct mlx5_flow_dv_sample_resource *res,
11737 struct rte_flow_error *error)
11739 struct mlx5_priv *priv = dev->data->dev_private;
11740 const struct rte_flow_action *sub_actions;
11741 struct mlx5_flow_sub_actions_list *sample_act;
11742 struct mlx5_flow_sub_actions_idx *sample_idx;
11743 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
11744 struct rte_flow *flow = dev_flow->flow;
11745 struct mlx5_flow_rss_desc *rss_desc;
11746 uint64_t action_flags = 0;
11749 rss_desc = &wks->rss_desc;
11750 sample_act = &res->sample_act;
11751 sample_idx = &res->sample_idx;
11752 res->ratio = action->ratio;
11753 sub_actions = action->actions;
11754 for (; sub_actions->type != RTE_FLOW_ACTION_TYPE_END; sub_actions++) {
11755 int type = sub_actions->type;
11756 uint32_t pre_rix = 0;
11759 case RTE_FLOW_ACTION_TYPE_QUEUE:
11761 const struct rte_flow_action_queue *queue;
11762 struct mlx5_hrxq *hrxq;
11765 queue = sub_actions->conf;
11766 rss_desc->queue_num = 1;
11767 rss_desc->queue[0] = queue->index;
11768 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
11769 rss_desc, &hrxq_idx);
11771 return rte_flow_error_set
11773 RTE_FLOW_ERROR_TYPE_ACTION,
11775 "cannot create fate queue");
11776 sample_act->dr_queue_action = hrxq->action;
11777 sample_idx->rix_hrxq = hrxq_idx;
11778 sample_actions[sample_act->actions_num++] =
11781 action_flags |= MLX5_FLOW_ACTION_QUEUE;
11782 if (action_flags & MLX5_FLOW_ACTION_MARK)
11783 dev_flow->handle->rix_hrxq = hrxq_idx;
11784 dev_flow->handle->fate_action =
11785 MLX5_FLOW_FATE_QUEUE;
11788 case RTE_FLOW_ACTION_TYPE_RSS:
11790 struct mlx5_hrxq *hrxq;
11792 const struct rte_flow_action_rss *rss;
11793 const uint8_t *rss_key;
11795 rss = sub_actions->conf;
11796 memcpy(rss_desc->queue, rss->queue,
11797 rss->queue_num * sizeof(uint16_t));
11798 rss_desc->queue_num = rss->queue_num;
11799 /* NULL RSS key indicates default RSS key. */
11800 rss_key = !rss->key ? rss_hash_default_key : rss->key;
11801 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
11803 * rss->level and rss.types should be set in advance
11804 * when expanding items for RSS.
11806 flow_dv_hashfields_set(dev_flow->handle->layers,
11808 &dev_flow->hash_fields);
11809 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
11810 rss_desc, &hrxq_idx);
11812 return rte_flow_error_set
11814 RTE_FLOW_ERROR_TYPE_ACTION,
11816 "cannot create fate queue");
11817 sample_act->dr_queue_action = hrxq->action;
11818 sample_idx->rix_hrxq = hrxq_idx;
11819 sample_actions[sample_act->actions_num++] =
11822 action_flags |= MLX5_FLOW_ACTION_RSS;
11823 if (action_flags & MLX5_FLOW_ACTION_MARK)
11824 dev_flow->handle->rix_hrxq = hrxq_idx;
11825 dev_flow->handle->fate_action =
11826 MLX5_FLOW_FATE_QUEUE;
11829 case RTE_FLOW_ACTION_TYPE_MARK:
11831 uint32_t tag_be = mlx5_flow_mark_set
11832 (((const struct rte_flow_action_mark *)
11833 (sub_actions->conf))->id);
11836 pre_rix = dev_flow->handle->dvh.rix_tag;
11837 /* Save the mark resource before sample */
11838 pre_r = dev_flow->dv.tag_resource;
11839 if (flow_dv_tag_resource_register(dev, tag_be,
11842 MLX5_ASSERT(dev_flow->dv.tag_resource);
11843 sample_act->dr_tag_action =
11844 dev_flow->dv.tag_resource->action;
11845 sample_idx->rix_tag =
11846 dev_flow->handle->dvh.rix_tag;
11847 sample_actions[sample_act->actions_num++] =
11848 sample_act->dr_tag_action;
11849 /* Recover the mark resource after sample */
11850 dev_flow->dv.tag_resource = pre_r;
11851 dev_flow->handle->dvh.rix_tag = pre_rix;
11852 action_flags |= MLX5_FLOW_ACTION_MARK;
11855 case RTE_FLOW_ACTION_TYPE_COUNT:
11857 if (!flow->counter) {
11859 flow_dv_translate_create_counter(dev,
11860 dev_flow, sub_actions->conf,
11862 if (!flow->counter)
11863 return rte_flow_error_set
11865 RTE_FLOW_ERROR_TYPE_ACTION,
11867 "cannot create counter"
11870 sample_act->dr_cnt_action =
11871 (flow_dv_counter_get_by_idx(dev,
11872 flow->counter, NULL))->action;
11873 sample_actions[sample_act->actions_num++] =
11874 sample_act->dr_cnt_action;
11875 action_flags |= MLX5_FLOW_ACTION_COUNT;
11878 case RTE_FLOW_ACTION_TYPE_PORT_ID:
11879 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
11881 struct mlx5_flow_dv_port_id_action_resource
11883 uint32_t port_id = 0;
11885 memset(&port_id_resource, 0, sizeof(port_id_resource));
11886 /* Save the port id resource before sample */
11887 pre_rix = dev_flow->handle->rix_port_id_action;
11888 pre_r = dev_flow->dv.port_id_action;
11889 if (flow_dv_translate_action_port_id(dev, sub_actions,
11892 port_id_resource.port_id = port_id;
11893 if (flow_dv_port_id_action_resource_register
11894 (dev, &port_id_resource, dev_flow, error))
11896 sample_act->dr_port_id_action =
11897 dev_flow->dv.port_id_action->action;
11898 sample_idx->rix_port_id_action =
11899 dev_flow->handle->rix_port_id_action;
11900 sample_actions[sample_act->actions_num++] =
11901 sample_act->dr_port_id_action;
11902 /* Recover the port id resource after sample */
11903 dev_flow->dv.port_id_action = pre_r;
11904 dev_flow->handle->rix_port_id_action = pre_rix;
11906 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
11909 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
11910 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
11911 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
11912 /* Save the encap resource before sample */
11913 pre_rix = dev_flow->handle->dvh.rix_encap_decap;
11914 pre_r = dev_flow->dv.encap_decap;
11915 if (flow_dv_create_action_l2_encap(dev, sub_actions,
11920 sample_act->dr_encap_action =
11921 dev_flow->dv.encap_decap->action;
11922 sample_idx->rix_encap_decap =
11923 dev_flow->handle->dvh.rix_encap_decap;
11924 sample_actions[sample_act->actions_num++] =
11925 sample_act->dr_encap_action;
11926 /* Recover the encap resource after sample */
11927 dev_flow->dv.encap_decap = pre_r;
11928 dev_flow->handle->dvh.rix_encap_decap = pre_rix;
11929 action_flags |= MLX5_FLOW_ACTION_ENCAP;
11932 return rte_flow_error_set(error, EINVAL,
11933 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11935 "Not support for sampler action");
11938 sample_act->action_flags = action_flags;
11939 res->ft_id = dev_flow->dv.group;
11940 if (attr->transfer) {
11942 uint32_t action_in[MLX5_ST_SZ_DW(set_action_in)];
11943 uint64_t set_action;
11944 } action_ctx = { .set_action = 0 };
11946 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
11947 MLX5_SET(set_action_in, action_ctx.action_in, action_type,
11948 MLX5_MODIFICATION_TYPE_SET);
11949 MLX5_SET(set_action_in, action_ctx.action_in, field,
11950 MLX5_MODI_META_REG_C_0);
11951 MLX5_SET(set_action_in, action_ctx.action_in, data,
11952 priv->vport_meta_tag);
11953 res->set_action = action_ctx.set_action;
11954 } else if (attr->ingress) {
11955 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
11957 res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_TX;
11963 * Convert Sample action to DV specification.
11966 * Pointer to rte_eth_dev structure.
11967 * @param[in, out] dev_flow
11968 * Pointer to the mlx5_flow.
11969 * @param[in] num_of_dest
11970 * The num of destination.
11971 * @param[in, out] res
11972 * Pointer to sample resource.
11973 * @param[in, out] mdest_res
11974 * Pointer to destination array resource.
11975 * @param[in] sample_actions
11976 * Pointer to sample path actions list.
11977 * @param[in] action_flags
11978 * Holds the actions detected until now.
11979 * @param[out] error
11980 * Pointer to the error structure.
11983 * 0 on success, a negative errno value otherwise and rte_errno is set.
11986 flow_dv_create_action_sample(struct rte_eth_dev *dev,
11987 struct mlx5_flow *dev_flow,
11988 uint32_t num_of_dest,
11989 struct mlx5_flow_dv_sample_resource *res,
11990 struct mlx5_flow_dv_dest_array_resource *mdest_res,
11991 void **sample_actions,
11992 uint64_t action_flags,
11993 struct rte_flow_error *error)
11995 /* update normal path action resource into last index of array */
11996 uint32_t dest_index = MLX5_MAX_DEST_NUM - 1;
11997 struct mlx5_flow_sub_actions_list *sample_act =
11998 &mdest_res->sample_act[dest_index];
11999 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
12000 struct mlx5_flow_rss_desc *rss_desc;
12001 uint32_t normal_idx = 0;
12002 struct mlx5_hrxq *hrxq;
12006 rss_desc = &wks->rss_desc;
12007 if (num_of_dest > 1) {
12008 if (sample_act->action_flags & MLX5_FLOW_ACTION_QUEUE) {
12009 /* Handle QP action for mirroring */
12010 hrxq = flow_dv_hrxq_prepare(dev, dev_flow,
12011 rss_desc, &hrxq_idx);
12013 return rte_flow_error_set
12015 RTE_FLOW_ERROR_TYPE_ACTION,
12017 "cannot create rx queue");
12019 mdest_res->sample_idx[dest_index].rix_hrxq = hrxq_idx;
12020 sample_act->dr_queue_action = hrxq->action;
12021 if (action_flags & MLX5_FLOW_ACTION_MARK)
12022 dev_flow->handle->rix_hrxq = hrxq_idx;
12023 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
12025 if (sample_act->action_flags & MLX5_FLOW_ACTION_ENCAP) {
12027 mdest_res->sample_idx[dest_index].rix_encap_decap =
12028 dev_flow->handle->dvh.rix_encap_decap;
12029 sample_act->dr_encap_action =
12030 dev_flow->dv.encap_decap->action;
12031 dev_flow->handle->dvh.rix_encap_decap = 0;
12033 if (sample_act->action_flags & MLX5_FLOW_ACTION_PORT_ID) {
12035 mdest_res->sample_idx[dest_index].rix_port_id_action =
12036 dev_flow->handle->rix_port_id_action;
12037 sample_act->dr_port_id_action =
12038 dev_flow->dv.port_id_action->action;
12039 dev_flow->handle->rix_port_id_action = 0;
12041 if (sample_act->action_flags & MLX5_FLOW_ACTION_JUMP) {
12043 mdest_res->sample_idx[dest_index].rix_jump =
12044 dev_flow->handle->rix_jump;
12045 sample_act->dr_jump_action =
12046 dev_flow->dv.jump->action;
12047 dev_flow->handle->rix_jump = 0;
12049 sample_act->actions_num = normal_idx;
12050 /* update sample action resource into first index of array */
12051 mdest_res->ft_type = res->ft_type;
12052 memcpy(&mdest_res->sample_idx[0], &res->sample_idx,
12053 sizeof(struct mlx5_flow_sub_actions_idx));
12054 memcpy(&mdest_res->sample_act[0], &res->sample_act,
12055 sizeof(struct mlx5_flow_sub_actions_list));
12056 mdest_res->num_of_dest = num_of_dest;
12057 if (flow_dv_dest_array_resource_register(dev, mdest_res,
12059 return rte_flow_error_set(error, EINVAL,
12060 RTE_FLOW_ERROR_TYPE_ACTION,
12061 NULL, "can't create sample "
12064 res->sub_actions = sample_actions;
12065 if (flow_dv_sample_resource_register(dev, res, dev_flow, error))
12066 return rte_flow_error_set(error, EINVAL,
12067 RTE_FLOW_ERROR_TYPE_ACTION,
12069 "can't create sample action");
12075 * Remove an ASO age action from age actions list.
12078 * Pointer to the Ethernet device structure.
12080 * Pointer to the aso age action handler.
12083 flow_dv_aso_age_remove_from_age(struct rte_eth_dev *dev,
12084 struct mlx5_aso_age_action *age)
12086 struct mlx5_age_info *age_info;
12087 struct mlx5_age_param *age_param = &age->age_params;
12088 struct mlx5_priv *priv = dev->data->dev_private;
12089 uint16_t expected = AGE_CANDIDATE;
12091 age_info = GET_PORT_AGE_INFO(priv);
12092 if (!__atomic_compare_exchange_n(&age_param->state, &expected,
12093 AGE_FREE, false, __ATOMIC_RELAXED,
12094 __ATOMIC_RELAXED)) {
12096 * We need the lock even it is age timeout,
12097 * since age action may still in process.
12099 rte_spinlock_lock(&age_info->aged_sl);
12100 LIST_REMOVE(age, next);
12101 rte_spinlock_unlock(&age_info->aged_sl);
12102 __atomic_store_n(&age_param->state, AGE_FREE, __ATOMIC_RELAXED);
12107 * Release an ASO age action.
12110 * Pointer to the Ethernet device structure.
12111 * @param[in] age_idx
12112 * Index of ASO age action to release.
12114 * True if the release operation is during flow destroy operation.
12115 * False if the release operation is during action destroy operation.
12118 * 0 when age action was removed, otherwise the number of references.
12121 flow_dv_aso_age_release(struct rte_eth_dev *dev, uint32_t age_idx)
12123 struct mlx5_priv *priv = dev->data->dev_private;
12124 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
12125 struct mlx5_aso_age_action *age = flow_aso_age_get_by_idx(dev, age_idx);
12126 uint32_t ret = __atomic_sub_fetch(&age->refcnt, 1, __ATOMIC_RELAXED);
12129 flow_dv_aso_age_remove_from_age(dev, age);
12130 rte_spinlock_lock(&mng->free_sl);
12131 LIST_INSERT_HEAD(&mng->free, age, next);
12132 rte_spinlock_unlock(&mng->free_sl);
12138 * Resize the ASO age pools array by MLX5_CNT_CONTAINER_RESIZE pools.
12141 * Pointer to the Ethernet device structure.
12144 * 0 on success, otherwise negative errno value and rte_errno is set.
12147 flow_dv_aso_age_pools_resize(struct rte_eth_dev *dev)
12149 struct mlx5_priv *priv = dev->data->dev_private;
12150 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
12151 void *old_pools = mng->pools;
12152 uint32_t resize = mng->n + MLX5_CNT_CONTAINER_RESIZE;
12153 uint32_t mem_size = sizeof(struct mlx5_aso_age_pool *) * resize;
12154 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
12157 rte_errno = ENOMEM;
12161 memcpy(pools, old_pools,
12162 mng->n * sizeof(struct mlx5_flow_counter_pool *));
12163 mlx5_free(old_pools);
12165 /* First ASO flow hit allocation - starting ASO data-path. */
12166 int ret = mlx5_aso_flow_hit_queue_poll_start(priv->sh);
12174 mng->pools = pools;
12179 * Create and initialize a new ASO aging pool.
12182 * Pointer to the Ethernet device structure.
12183 * @param[out] age_free
12184 * Where to put the pointer of a new age action.
12187 * The age actions pool pointer and @p age_free is set on success,
12188 * NULL otherwise and rte_errno is set.
12190 static struct mlx5_aso_age_pool *
12191 flow_dv_age_pool_create(struct rte_eth_dev *dev,
12192 struct mlx5_aso_age_action **age_free)
12194 struct mlx5_priv *priv = dev->data->dev_private;
12195 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
12196 struct mlx5_aso_age_pool *pool = NULL;
12197 struct mlx5_devx_obj *obj = NULL;
12200 obj = mlx5_devx_cmd_create_flow_hit_aso_obj(priv->sh->cdev->ctx,
12201 priv->sh->cdev->pdn);
12203 rte_errno = ENODATA;
12204 DRV_LOG(ERR, "Failed to create flow_hit_aso_obj using DevX.");
12207 pool = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*pool), 0, SOCKET_ID_ANY);
12209 claim_zero(mlx5_devx_cmd_destroy(obj));
12210 rte_errno = ENOMEM;
12213 pool->flow_hit_aso_obj = obj;
12214 pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
12215 rte_rwlock_write_lock(&mng->resize_rwl);
12216 pool->index = mng->next;
12217 /* Resize pools array if there is no room for the new pool in it. */
12218 if (pool->index == mng->n && flow_dv_aso_age_pools_resize(dev)) {
12219 claim_zero(mlx5_devx_cmd_destroy(obj));
12221 rte_rwlock_write_unlock(&mng->resize_rwl);
12224 mng->pools[pool->index] = pool;
12226 rte_rwlock_write_unlock(&mng->resize_rwl);
12227 /* Assign the first action in the new pool, the rest go to free list. */
12228 *age_free = &pool->actions[0];
12229 for (i = 1; i < MLX5_ASO_AGE_ACTIONS_PER_POOL; i++) {
12230 pool->actions[i].offset = i;
12231 LIST_INSERT_HEAD(&mng->free, &pool->actions[i], next);
12237 * Allocate a ASO aging bit.
12240 * Pointer to the Ethernet device structure.
12241 * @param[out] error
12242 * Pointer to the error structure.
12245 * Index to ASO age action on success, 0 otherwise and rte_errno is set.
12248 flow_dv_aso_age_alloc(struct rte_eth_dev *dev, struct rte_flow_error *error)
12250 struct mlx5_priv *priv = dev->data->dev_private;
12251 const struct mlx5_aso_age_pool *pool;
12252 struct mlx5_aso_age_action *age_free = NULL;
12253 struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
12256 /* Try to get the next free age action bit. */
12257 rte_spinlock_lock(&mng->free_sl);
12258 age_free = LIST_FIRST(&mng->free);
12260 LIST_REMOVE(age_free, next);
12261 } else if (!flow_dv_age_pool_create(dev, &age_free)) {
12262 rte_spinlock_unlock(&mng->free_sl);
12263 rte_flow_error_set(error, rte_errno, RTE_FLOW_ERROR_TYPE_ACTION,
12264 NULL, "failed to create ASO age pool");
12265 return 0; /* 0 is an error. */
12267 rte_spinlock_unlock(&mng->free_sl);
12268 pool = container_of
12269 ((const struct mlx5_aso_age_action (*)[MLX5_ASO_AGE_ACTIONS_PER_POOL])
12270 (age_free - age_free->offset), const struct mlx5_aso_age_pool,
12272 if (!age_free->dr_action) {
12273 int reg_c = mlx5_flow_get_reg_id(dev, MLX5_ASO_FLOW_HIT, 0,
12277 rte_flow_error_set(error, rte_errno,
12278 RTE_FLOW_ERROR_TYPE_ACTION,
12279 NULL, "failed to get reg_c "
12280 "for ASO flow hit");
12281 return 0; /* 0 is an error. */
12283 #ifdef HAVE_MLX5_DR_CREATE_ACTION_ASO
12284 age_free->dr_action = mlx5_glue->dv_create_flow_action_aso
12285 (priv->sh->rx_domain,
12286 pool->flow_hit_aso_obj->obj, age_free->offset,
12287 MLX5DV_DR_ACTION_FLAGS_ASO_FIRST_HIT_SET,
12288 (reg_c - REG_C_0));
12289 #endif /* HAVE_MLX5_DR_CREATE_ACTION_ASO */
12290 if (!age_free->dr_action) {
12292 rte_spinlock_lock(&mng->free_sl);
12293 LIST_INSERT_HEAD(&mng->free, age_free, next);
12294 rte_spinlock_unlock(&mng->free_sl);
12295 rte_flow_error_set(error, rte_errno,
12296 RTE_FLOW_ERROR_TYPE_ACTION,
12297 NULL, "failed to create ASO "
12298 "flow hit action");
12299 return 0; /* 0 is an error. */
12302 __atomic_store_n(&age_free->refcnt, 1, __ATOMIC_RELAXED);
12303 return pool->index | ((age_free->offset + 1) << 16);
12307 * Initialize flow ASO age parameters.
12310 * Pointer to rte_eth_dev structure.
12311 * @param[in] age_idx
12312 * Index of ASO age action.
12313 * @param[in] context
12314 * Pointer to flow counter age context.
12315 * @param[in] timeout
12316 * Aging timeout in seconds.
12320 flow_dv_aso_age_params_init(struct rte_eth_dev *dev,
12325 struct mlx5_aso_age_action *aso_age;
12327 aso_age = flow_aso_age_get_by_idx(dev, age_idx);
12328 MLX5_ASSERT(aso_age);
12329 aso_age->age_params.context = context;
12330 aso_age->age_params.timeout = timeout;
12331 aso_age->age_params.port_id = dev->data->port_id;
12332 __atomic_store_n(&aso_age->age_params.sec_since_last_hit, 0,
12334 __atomic_store_n(&aso_age->age_params.state, AGE_CANDIDATE,
12339 flow_dv_translate_integrity_l4(const struct rte_flow_item_integrity *mask,
12340 const struct rte_flow_item_integrity *value,
12341 void *headers_m, void *headers_v)
12344 /* RTE l4_ok filter aggregates hardware l4_ok and
12345 * l4_checksum_ok filters.
12346 * Positive RTE l4_ok match requires hardware match on both L4
12347 * hardware integrity bits.
12348 * For negative match, check hardware l4_checksum_ok bit only,
12349 * because hardware sets that bit to 0 for all packets
12352 if (value->l4_ok) {
12353 MLX5_SET(fte_match_set_lyr_2_4, headers_m, l4_ok, 1);
12354 MLX5_SET(fte_match_set_lyr_2_4, headers_v, l4_ok, 1);
12356 MLX5_SET(fte_match_set_lyr_2_4, headers_m, l4_checksum_ok, 1);
12357 MLX5_SET(fte_match_set_lyr_2_4, headers_v, l4_checksum_ok,
12360 if (mask->l4_csum_ok) {
12361 MLX5_SET(fte_match_set_lyr_2_4, headers_m, l4_checksum_ok, 1);
12362 MLX5_SET(fte_match_set_lyr_2_4, headers_v, l4_checksum_ok,
12363 value->l4_csum_ok);
12368 flow_dv_translate_integrity_l3(const struct rte_flow_item_integrity *mask,
12369 const struct rte_flow_item_integrity *value,
12370 void *headers_m, void *headers_v, bool is_ipv4)
12373 /* RTE l3_ok filter aggregates for IPv4 hardware l3_ok and
12374 * ipv4_csum_ok filters.
12375 * Positive RTE l3_ok match requires hardware match on both L3
12376 * hardware integrity bits.
12377 * For negative match, check hardware l3_csum_ok bit only,
12378 * because hardware sets that bit to 0 for all packets
12382 if (value->l3_ok) {
12383 MLX5_SET(fte_match_set_lyr_2_4, headers_m,
12385 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
12388 MLX5_SET(fte_match_set_lyr_2_4, headers_m,
12389 ipv4_checksum_ok, 1);
12390 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
12391 ipv4_checksum_ok, !!value->l3_ok);
12393 MLX5_SET(fte_match_set_lyr_2_4, headers_m, l3_ok, 1);
12394 MLX5_SET(fte_match_set_lyr_2_4, headers_v, l3_ok,
12398 if (mask->ipv4_csum_ok) {
12399 MLX5_SET(fte_match_set_lyr_2_4, headers_m, ipv4_checksum_ok, 1);
12400 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ipv4_checksum_ok,
12401 value->ipv4_csum_ok);
12406 set_integrity_bits(void *headers_m, void *headers_v,
12407 const struct rte_flow_item *integrity_item, bool is_l3_ip4)
12409 const struct rte_flow_item_integrity *spec = integrity_item->spec;
12410 const struct rte_flow_item_integrity *mask = integrity_item->mask;
12412 /* Integrity bits validation cleared spec pointer */
12413 MLX5_ASSERT(spec != NULL);
12415 mask = &rte_flow_item_integrity_mask;
12416 flow_dv_translate_integrity_l3(mask, spec, headers_m, headers_v,
12418 flow_dv_translate_integrity_l4(mask, spec, headers_m, headers_v);
12422 flow_dv_translate_item_integrity_post(void *matcher, void *key,
12424 struct rte_flow_item *integrity_items[2],
12425 uint64_t pattern_flags)
12427 void *headers_m, *headers_v;
12430 if (pattern_flags & MLX5_FLOW_ITEM_INNER_INTEGRITY) {
12431 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
12433 headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
12434 is_l3_ip4 = (pattern_flags & MLX5_FLOW_LAYER_INNER_L3_IPV4) !=
12436 set_integrity_bits(headers_m, headers_v,
12437 integrity_items[1], is_l3_ip4);
12439 if (pattern_flags & MLX5_FLOW_ITEM_OUTER_INTEGRITY) {
12440 headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
12442 headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
12443 is_l3_ip4 = (pattern_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4) !=
12445 set_integrity_bits(headers_m, headers_v,
12446 integrity_items[0], is_l3_ip4);
12451 flow_dv_translate_item_integrity(const struct rte_flow_item *item,
12452 const struct rte_flow_item *integrity_items[2],
12453 uint64_t *last_item)
12455 const struct rte_flow_item_integrity *spec = (typeof(spec))item->spec;
12457 /* integrity bits validation cleared spec pointer */
12458 MLX5_ASSERT(spec != NULL);
12459 if (spec->level > 1) {
12460 integrity_items[1] = item;
12461 *last_item |= MLX5_FLOW_ITEM_INNER_INTEGRITY;
12463 integrity_items[0] = item;
12464 *last_item |= MLX5_FLOW_ITEM_OUTER_INTEGRITY;
12469 * Prepares DV flow counter with aging configuration.
12470 * Gets it by index when exists, creates a new one when doesn't.
12473 * Pointer to rte_eth_dev structure.
12474 * @param[in] dev_flow
12475 * Pointer to the mlx5_flow.
12476 * @param[in, out] flow
12477 * Pointer to the sub flow.
12479 * Pointer to the counter action configuration.
12481 * Pointer to the aging action configuration.
12482 * @param[out] error
12483 * Pointer to the error structure.
12486 * Pointer to the counter, NULL otherwise.
12488 static struct mlx5_flow_counter *
12489 flow_dv_prepare_counter(struct rte_eth_dev *dev,
12490 struct mlx5_flow *dev_flow,
12491 struct rte_flow *flow,
12492 const struct rte_flow_action_count *count,
12493 const struct rte_flow_action_age *age,
12494 struct rte_flow_error *error)
12496 if (!flow->counter) {
12497 flow->counter = flow_dv_translate_create_counter(dev, dev_flow,
12499 if (!flow->counter) {
12500 rte_flow_error_set(error, rte_errno,
12501 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12502 "cannot create counter object.");
12506 return flow_dv_counter_get_by_idx(dev, flow->counter, NULL);
12510 * Release an ASO CT action by its own device.
12513 * Pointer to the Ethernet device structure.
12515 * Index of ASO CT action to release.
12518 * 0 when CT action was removed, otherwise the number of references.
12521 flow_dv_aso_ct_dev_release(struct rte_eth_dev *dev, uint32_t idx)
12523 struct mlx5_priv *priv = dev->data->dev_private;
12524 struct mlx5_aso_ct_pools_mng *mng = priv->sh->ct_mng;
12526 struct mlx5_aso_ct_action *ct = flow_aso_ct_get_by_dev_idx(dev, idx);
12527 enum mlx5_aso_ct_state state =
12528 __atomic_load_n(&ct->state, __ATOMIC_RELAXED);
12530 /* Cannot release when CT is in the ASO SQ. */
12531 if (state == ASO_CONNTRACK_WAIT || state == ASO_CONNTRACK_QUERY)
12533 ret = __atomic_sub_fetch(&ct->refcnt, 1, __ATOMIC_RELAXED);
12535 if (ct->dr_action_orig) {
12536 #ifdef HAVE_MLX5_DR_ACTION_ASO_CT
12537 claim_zero(mlx5_glue->destroy_flow_action
12538 (ct->dr_action_orig));
12540 ct->dr_action_orig = NULL;
12542 if (ct->dr_action_rply) {
12543 #ifdef HAVE_MLX5_DR_ACTION_ASO_CT
12544 claim_zero(mlx5_glue->destroy_flow_action
12545 (ct->dr_action_rply));
12547 ct->dr_action_rply = NULL;
12549 /* Clear the state to free, no need in 1st allocation. */
12550 MLX5_ASO_CT_UPDATE_STATE(ct, ASO_CONNTRACK_FREE);
12551 rte_spinlock_lock(&mng->ct_sl);
12552 LIST_INSERT_HEAD(&mng->free_cts, ct, next);
12553 rte_spinlock_unlock(&mng->ct_sl);
12559 flow_dv_aso_ct_release(struct rte_eth_dev *dev, uint32_t own_idx,
12560 struct rte_flow_error *error)
12562 uint16_t owner = (uint16_t)MLX5_INDIRECT_ACT_CT_GET_OWNER(own_idx);
12563 uint32_t idx = MLX5_INDIRECT_ACT_CT_GET_IDX(own_idx);
12564 struct rte_eth_dev *owndev = &rte_eth_devices[owner];
12567 MLX5_ASSERT(owner < RTE_MAX_ETHPORTS);
12568 if (dev->data->dev_started != 1)
12569 return rte_flow_error_set(error, EAGAIN,
12570 RTE_FLOW_ERROR_TYPE_ACTION,
12572 "Indirect CT action cannot be destroyed when the port is stopped");
12573 ret = flow_dv_aso_ct_dev_release(owndev, idx);
12575 return rte_flow_error_set(error, EAGAIN,
12576 RTE_FLOW_ERROR_TYPE_ACTION,
12578 "Current state prevents indirect CT action from being destroyed");
12583 * Resize the ASO CT pools array by 64 pools.
12586 * Pointer to the Ethernet device structure.
12589 * 0 on success, otherwise negative errno value and rte_errno is set.
12592 flow_dv_aso_ct_pools_resize(struct rte_eth_dev *dev)
12594 struct mlx5_priv *priv = dev->data->dev_private;
12595 struct mlx5_aso_ct_pools_mng *mng = priv->sh->ct_mng;
12596 void *old_pools = mng->pools;
12597 /* Magic number now, need a macro. */
12598 uint32_t resize = mng->n + 64;
12599 uint32_t mem_size = sizeof(struct mlx5_aso_ct_pool *) * resize;
12600 void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
12603 rte_errno = ENOMEM;
12606 rte_rwlock_write_lock(&mng->resize_rwl);
12607 /* ASO SQ/QP was already initialized in the startup. */
12609 /* Realloc could be an alternative choice. */
12610 rte_memcpy(pools, old_pools,
12611 mng->n * sizeof(struct mlx5_aso_ct_pool *));
12612 mlx5_free(old_pools);
12615 mng->pools = pools;
12616 rte_rwlock_write_unlock(&mng->resize_rwl);
12621 * Create and initialize a new ASO CT pool.
12624 * Pointer to the Ethernet device structure.
12625 * @param[out] ct_free
12626 * Where to put the pointer of a new CT action.
12629 * The CT actions pool pointer and @p ct_free is set on success,
12630 * NULL otherwise and rte_errno is set.
12632 static struct mlx5_aso_ct_pool *
12633 flow_dv_ct_pool_create(struct rte_eth_dev *dev,
12634 struct mlx5_aso_ct_action **ct_free)
12636 struct mlx5_priv *priv = dev->data->dev_private;
12637 struct mlx5_aso_ct_pools_mng *mng = priv->sh->ct_mng;
12638 struct mlx5_aso_ct_pool *pool = NULL;
12639 struct mlx5_devx_obj *obj = NULL;
12641 uint32_t log_obj_size = rte_log2_u32(MLX5_ASO_CT_ACTIONS_PER_POOL);
12643 obj = mlx5_devx_cmd_create_conn_track_offload_obj(priv->sh->cdev->ctx,
12644 priv->sh->cdev->pdn,
12647 rte_errno = ENODATA;
12648 DRV_LOG(ERR, "Failed to create conn_track_offload_obj using DevX.");
12651 pool = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*pool), 0, SOCKET_ID_ANY);
12653 rte_errno = ENOMEM;
12654 claim_zero(mlx5_devx_cmd_destroy(obj));
12657 pool->devx_obj = obj;
12658 pool->index = mng->next;
12659 /* Resize pools array if there is no room for the new pool in it. */
12660 if (pool->index == mng->n && flow_dv_aso_ct_pools_resize(dev)) {
12661 claim_zero(mlx5_devx_cmd_destroy(obj));
12665 mng->pools[pool->index] = pool;
12667 /* Assign the first action in the new pool, the rest go to free list. */
12668 *ct_free = &pool->actions[0];
12669 /* Lock outside, the list operation is safe here. */
12670 for (i = 1; i < MLX5_ASO_CT_ACTIONS_PER_POOL; i++) {
12671 /* refcnt is 0 when allocating the memory. */
12672 pool->actions[i].offset = i;
12673 LIST_INSERT_HEAD(&mng->free_cts, &pool->actions[i], next);
12679 * Allocate a ASO CT action from free list.
12682 * Pointer to the Ethernet device structure.
12683 * @param[out] error
12684 * Pointer to the error structure.
12687 * Index to ASO CT action on success, 0 otherwise and rte_errno is set.
12690 flow_dv_aso_ct_alloc(struct rte_eth_dev *dev, struct rte_flow_error *error)
12692 struct mlx5_priv *priv = dev->data->dev_private;
12693 struct mlx5_aso_ct_pools_mng *mng = priv->sh->ct_mng;
12694 struct mlx5_aso_ct_action *ct = NULL;
12695 struct mlx5_aso_ct_pool *pool;
12700 if (!priv->sh->cdev->config.devx) {
12701 rte_errno = ENOTSUP;
12704 /* Get a free CT action, if no, a new pool will be created. */
12705 rte_spinlock_lock(&mng->ct_sl);
12706 ct = LIST_FIRST(&mng->free_cts);
12708 LIST_REMOVE(ct, next);
12709 } else if (!flow_dv_ct_pool_create(dev, &ct)) {
12710 rte_spinlock_unlock(&mng->ct_sl);
12711 rte_flow_error_set(error, rte_errno, RTE_FLOW_ERROR_TYPE_ACTION,
12712 NULL, "failed to create ASO CT pool");
12715 rte_spinlock_unlock(&mng->ct_sl);
12716 pool = container_of(ct, struct mlx5_aso_ct_pool, actions[ct->offset]);
12717 ct_idx = MLX5_MAKE_CT_IDX(pool->index, ct->offset);
12718 /* 0: inactive, 1: created, 2+: used by flows. */
12719 __atomic_store_n(&ct->refcnt, 1, __ATOMIC_RELAXED);
12720 reg_c = mlx5_flow_get_reg_id(dev, MLX5_ASO_CONNTRACK, 0, error);
12721 if (!ct->dr_action_orig) {
12722 #ifdef HAVE_MLX5_DR_ACTION_ASO_CT
12723 ct->dr_action_orig = mlx5_glue->dv_create_flow_action_aso
12724 (priv->sh->rx_domain, pool->devx_obj->obj,
12726 MLX5DV_DR_ACTION_FLAGS_ASO_CT_DIRECTION_INITIATOR,
12729 RTE_SET_USED(reg_c);
12731 if (!ct->dr_action_orig) {
12732 flow_dv_aso_ct_dev_release(dev, ct_idx);
12733 rte_flow_error_set(error, rte_errno,
12734 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12735 "failed to create ASO CT action");
12739 if (!ct->dr_action_rply) {
12740 #ifdef HAVE_MLX5_DR_ACTION_ASO_CT
12741 ct->dr_action_rply = mlx5_glue->dv_create_flow_action_aso
12742 (priv->sh->rx_domain, pool->devx_obj->obj,
12744 MLX5DV_DR_ACTION_FLAGS_ASO_CT_DIRECTION_RESPONDER,
12747 if (!ct->dr_action_rply) {
12748 flow_dv_aso_ct_dev_release(dev, ct_idx);
12749 rte_flow_error_set(error, rte_errno,
12750 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12751 "failed to create ASO CT action");
12759 * Create a conntrack object with context and actions by using ASO mechanism.
12762 * Pointer to rte_eth_dev structure.
12764 * Pointer to conntrack information profile.
12765 * @param[out] error
12766 * Pointer to the error structure.
12769 * Index to conntrack object on success, 0 otherwise.
12772 flow_dv_translate_create_conntrack(struct rte_eth_dev *dev,
12773 const struct rte_flow_action_conntrack *pro,
12774 struct rte_flow_error *error)
12776 struct mlx5_priv *priv = dev->data->dev_private;
12777 struct mlx5_dev_ctx_shared *sh = priv->sh;
12778 struct mlx5_aso_ct_action *ct;
12781 if (!sh->ct_aso_en)
12782 return rte_flow_error_set(error, ENOTSUP,
12783 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12784 "Connection is not supported");
12785 idx = flow_dv_aso_ct_alloc(dev, error);
12787 return rte_flow_error_set(error, rte_errno,
12788 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12789 "Failed to allocate CT object");
12790 ct = flow_aso_ct_get_by_dev_idx(dev, idx);
12791 if (mlx5_aso_ct_update_by_wqe(sh, ct, pro))
12792 return rte_flow_error_set(error, EBUSY,
12793 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
12794 "Failed to update CT");
12795 ct->is_original = !!pro->is_original_dir;
12796 ct->peer = pro->peer_port;
12801 * Fill the flow with DV spec, lock free
12802 * (mutex should be acquired by caller).
12805 * Pointer to rte_eth_dev structure.
12806 * @param[in, out] dev_flow
12807 * Pointer to the sub flow.
12809 * Pointer to the flow attributes.
12811 * Pointer to the list of items.
12812 * @param[in] actions
12813 * Pointer to the list of actions.
12814 * @param[out] error
12815 * Pointer to the error structure.
12818 * 0 on success, a negative errno value otherwise and rte_errno is set.
12821 flow_dv_translate(struct rte_eth_dev *dev,
12822 struct mlx5_flow *dev_flow,
12823 const struct rte_flow_attr *attr,
12824 const struct rte_flow_item items[],
12825 const struct rte_flow_action actions[],
12826 struct rte_flow_error *error)
12828 struct mlx5_priv *priv = dev->data->dev_private;
12829 struct mlx5_sh_config *dev_conf = &priv->sh->config;
12830 struct rte_flow *flow = dev_flow->flow;
12831 struct mlx5_flow_handle *handle = dev_flow->handle;
12832 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
12833 struct mlx5_flow_rss_desc *rss_desc;
12834 uint64_t item_flags = 0;
12835 uint64_t last_item = 0;
12836 uint64_t action_flags = 0;
12837 struct mlx5_flow_dv_matcher matcher = {
12839 .size = sizeof(matcher.mask.buf),
12843 bool actions_end = false;
12845 struct mlx5_flow_dv_modify_hdr_resource res;
12846 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
12847 sizeof(struct mlx5_modification_cmd) *
12848 (MLX5_MAX_MODIFY_NUM + 1)];
12850 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
12851 const struct rte_flow_action_count *count = NULL;
12852 const struct rte_flow_action_age *non_shared_age = NULL;
12853 union flow_dv_attr flow_attr = { .attr = 0 };
12855 union mlx5_flow_tbl_key tbl_key;
12856 uint32_t modify_action_position = UINT32_MAX;
12857 void *match_mask = matcher.mask.buf;
12858 void *match_value = dev_flow->dv.value.buf;
12859 uint8_t next_protocol = 0xff;
12860 struct rte_vlan_hdr vlan = { 0 };
12861 struct mlx5_flow_dv_dest_array_resource mdest_res;
12862 struct mlx5_flow_dv_sample_resource sample_res;
12863 void *sample_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
12864 const struct rte_flow_action_sample *sample = NULL;
12865 struct mlx5_flow_sub_actions_list *sample_act;
12866 uint32_t sample_act_pos = UINT32_MAX;
12867 uint32_t age_act_pos = UINT32_MAX;
12868 uint32_t num_of_dest = 0;
12869 int tmp_actions_n = 0;
12872 const struct mlx5_flow_tunnel *tunnel = NULL;
12873 struct flow_grp_info grp_info = {
12874 .external = !!dev_flow->external,
12875 .transfer = !!attr->transfer,
12876 .fdb_def_rule = !!priv->fdb_def_rule,
12877 .skip_scale = dev_flow->skip_scale &
12878 (1 << MLX5_SCALE_FLOW_GROUP_BIT),
12879 .std_tbl_fix = true,
12881 const struct rte_flow_item *integrity_items[2] = {NULL, NULL};
12882 const struct rte_flow_item *tunnel_item = NULL;
12883 const struct rte_flow_item *gre_item = NULL;
12886 return rte_flow_error_set(error, ENOMEM,
12887 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
12889 "failed to push flow workspace");
12890 rss_desc = &wks->rss_desc;
12891 memset(&mdest_res, 0, sizeof(struct mlx5_flow_dv_dest_array_resource));
12892 memset(&sample_res, 0, sizeof(struct mlx5_flow_dv_sample_resource));
12893 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
12894 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
12895 /* update normal path action resource into last index of array */
12896 sample_act = &mdest_res.sample_act[MLX5_MAX_DEST_NUM - 1];
12897 if (is_tunnel_offload_active(dev)) {
12898 if (dev_flow->tunnel) {
12899 RTE_VERIFY(dev_flow->tof_type ==
12900 MLX5_TUNNEL_OFFLOAD_MISS_RULE);
12901 tunnel = dev_flow->tunnel;
12903 tunnel = mlx5_get_tof(items, actions,
12904 &dev_flow->tof_type);
12905 dev_flow->tunnel = tunnel;
12907 grp_info.std_tbl_fix = tunnel_use_standard_attr_group_translate
12908 (dev, attr, tunnel, dev_flow->tof_type);
12910 mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
12911 MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
12912 ret = mlx5_flow_group_to_table(dev, tunnel, attr->group, &table,
12916 dev_flow->dv.group = table;
12917 if (attr->transfer)
12918 mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
12919 /* number of actions must be set to 0 in case of dirty stack. */
12920 mhdr_res->actions_num = 0;
12921 if (is_flow_tunnel_match_rule(dev_flow->tof_type)) {
12923 * do not add decap action if match rule drops packet
12924 * HW rejects rules with decap & drop
12926 * if tunnel match rule was inserted before matching tunnel set
12927 * rule flow table used in the match rule must be registered.
12928 * current implementation handles that in the
12929 * flow_dv_match_register() at the function end.
12931 bool add_decap = true;
12932 const struct rte_flow_action *ptr = actions;
12934 for (; ptr->type != RTE_FLOW_ACTION_TYPE_END; ptr++) {
12935 if (ptr->type == RTE_FLOW_ACTION_TYPE_DROP) {
12941 if (flow_dv_create_action_l2_decap(dev, dev_flow,
12945 dev_flow->dv.actions[actions_n++] =
12946 dev_flow->dv.encap_decap->action;
12947 action_flags |= MLX5_FLOW_ACTION_DECAP;
12950 for (; !actions_end ; actions++) {
12951 const struct rte_flow_action_queue *queue;
12952 const struct rte_flow_action_rss *rss;
12953 const struct rte_flow_action *action = actions;
12954 const uint8_t *rss_key;
12955 struct mlx5_flow_tbl_resource *tbl;
12956 struct mlx5_aso_age_action *age_act;
12957 struct mlx5_flow_counter *cnt_act;
12958 uint32_t port_id = 0;
12959 struct mlx5_flow_dv_port_id_action_resource port_id_resource;
12960 int action_type = actions->type;
12961 const struct rte_flow_action *found_action = NULL;
12962 uint32_t jump_group = 0;
12963 uint32_t owner_idx;
12964 struct mlx5_aso_ct_action *ct;
12966 if (!mlx5_flow_os_action_supported(action_type))
12967 return rte_flow_error_set(error, ENOTSUP,
12968 RTE_FLOW_ERROR_TYPE_ACTION,
12970 "action not supported");
12971 switch (action_type) {
12972 case MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET:
12973 action_flags |= MLX5_FLOW_ACTION_TUNNEL_SET;
12975 case RTE_FLOW_ACTION_TYPE_VOID:
12977 case RTE_FLOW_ACTION_TYPE_PORT_ID:
12978 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
12979 if (flow_dv_translate_action_port_id(dev, action,
12982 port_id_resource.port_id = port_id;
12983 MLX5_ASSERT(!handle->rix_port_id_action);
12984 if (flow_dv_port_id_action_resource_register
12985 (dev, &port_id_resource, dev_flow, error))
12987 dev_flow->dv.actions[actions_n++] =
12988 dev_flow->dv.port_id_action->action;
12989 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
12990 dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
12991 sample_act->action_flags |= MLX5_FLOW_ACTION_PORT_ID;
12994 case RTE_FLOW_ACTION_TYPE_FLAG:
12995 action_flags |= MLX5_FLOW_ACTION_FLAG;
12997 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
12998 struct rte_flow_action_mark mark = {
12999 .id = MLX5_FLOW_MARK_DEFAULT,
13002 if (flow_dv_convert_action_mark(dev, &mark,
13006 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
13009 tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
13011 * Only one FLAG or MARK is supported per device flow
13012 * right now. So the pointer to the tag resource must be
13013 * zero before the register process.
13015 MLX5_ASSERT(!handle->dvh.rix_tag);
13016 if (flow_dv_tag_resource_register(dev, tag_be,
13019 MLX5_ASSERT(dev_flow->dv.tag_resource);
13020 dev_flow->dv.actions[actions_n++] =
13021 dev_flow->dv.tag_resource->action;
13023 case RTE_FLOW_ACTION_TYPE_MARK:
13024 action_flags |= MLX5_FLOW_ACTION_MARK;
13026 if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
13027 const struct rte_flow_action_mark *mark =
13028 (const struct rte_flow_action_mark *)
13031 if (flow_dv_convert_action_mark(dev, mark,
13035 action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
13039 case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
13040 /* Legacy (non-extensive) MARK action. */
13041 tag_be = mlx5_flow_mark_set
13042 (((const struct rte_flow_action_mark *)
13043 (actions->conf))->id);
13044 MLX5_ASSERT(!handle->dvh.rix_tag);
13045 if (flow_dv_tag_resource_register(dev, tag_be,
13048 MLX5_ASSERT(dev_flow->dv.tag_resource);
13049 dev_flow->dv.actions[actions_n++] =
13050 dev_flow->dv.tag_resource->action;
13052 case RTE_FLOW_ACTION_TYPE_SET_META:
13053 if (flow_dv_convert_action_set_meta
13054 (dev, mhdr_res, attr,
13055 (const struct rte_flow_action_set_meta *)
13056 actions->conf, error))
13058 action_flags |= MLX5_FLOW_ACTION_SET_META;
13060 case RTE_FLOW_ACTION_TYPE_SET_TAG:
13061 if (flow_dv_convert_action_set_tag
13063 (const struct rte_flow_action_set_tag *)
13064 actions->conf, error))
13066 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
13068 case RTE_FLOW_ACTION_TYPE_DROP:
13069 action_flags |= MLX5_FLOW_ACTION_DROP;
13070 dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
13072 case RTE_FLOW_ACTION_TYPE_QUEUE:
13073 queue = actions->conf;
13074 rss_desc->queue_num = 1;
13075 rss_desc->queue[0] = queue->index;
13076 action_flags |= MLX5_FLOW_ACTION_QUEUE;
13077 dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
13078 sample_act->action_flags |= MLX5_FLOW_ACTION_QUEUE;
13081 case RTE_FLOW_ACTION_TYPE_RSS:
13082 rss = actions->conf;
13083 memcpy(rss_desc->queue, rss->queue,
13084 rss->queue_num * sizeof(uint16_t));
13085 rss_desc->queue_num = rss->queue_num;
13086 /* NULL RSS key indicates default RSS key. */
13087 rss_key = !rss->key ? rss_hash_default_key : rss->key;
13088 memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
13090 * rss->level and rss.types should be set in advance
13091 * when expanding items for RSS.
13093 action_flags |= MLX5_FLOW_ACTION_RSS;
13094 dev_flow->handle->fate_action = rss_desc->shared_rss ?
13095 MLX5_FLOW_FATE_SHARED_RSS :
13096 MLX5_FLOW_FATE_QUEUE;
13098 case MLX5_RTE_FLOW_ACTION_TYPE_AGE:
13099 owner_idx = (uint32_t)(uintptr_t)action->conf;
13100 age_act = flow_aso_age_get_by_idx(dev, owner_idx);
13101 if (flow->age == 0) {
13102 flow->age = owner_idx;
13103 __atomic_fetch_add(&age_act->refcnt, 1,
13106 age_act_pos = actions_n++;
13107 action_flags |= MLX5_FLOW_ACTION_AGE;
13109 case RTE_FLOW_ACTION_TYPE_AGE:
13110 non_shared_age = action->conf;
13111 age_act_pos = actions_n++;
13112 action_flags |= MLX5_FLOW_ACTION_AGE;
13114 case MLX5_RTE_FLOW_ACTION_TYPE_COUNT:
13115 owner_idx = (uint32_t)(uintptr_t)action->conf;
13116 cnt_act = flow_dv_counter_get_by_idx(dev, owner_idx,
13118 MLX5_ASSERT(cnt_act != NULL);
13120 * When creating meter drop flow in drop table, the
13121 * counter should not overwrite the rte flow counter.
13123 if (attr->group == MLX5_FLOW_TABLE_LEVEL_METER &&
13124 dev_flow->dv.table_id == MLX5_MTR_TABLE_ID_DROP) {
13125 dev_flow->dv.actions[actions_n++] =
13128 if (flow->counter == 0) {
13129 flow->counter = owner_idx;
13131 (&cnt_act->shared_info.refcnt,
13132 1, __ATOMIC_RELAXED);
13134 /* Save information first, will apply later. */
13135 action_flags |= MLX5_FLOW_ACTION_COUNT;
13138 case RTE_FLOW_ACTION_TYPE_COUNT:
13139 if (!priv->sh->cdev->config.devx) {
13140 return rte_flow_error_set
13142 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
13144 "count action not supported");
13146 /* Save information first, will apply later. */
13147 count = action->conf;
13148 action_flags |= MLX5_FLOW_ACTION_COUNT;
13150 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
13151 dev_flow->dv.actions[actions_n++] =
13152 priv->sh->pop_vlan_action;
13153 action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
13155 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
13156 if (!(action_flags &
13157 MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
13158 flow_dev_get_vlan_info_from_items(items, &vlan);
13159 vlan.eth_proto = rte_be_to_cpu_16
13160 ((((const struct rte_flow_action_of_push_vlan *)
13161 actions->conf)->ethertype));
13162 found_action = mlx5_flow_find_action
13164 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
13166 mlx5_update_vlan_vid_pcp(found_action, &vlan);
13167 found_action = mlx5_flow_find_action
13169 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
13171 mlx5_update_vlan_vid_pcp(found_action, &vlan);
13172 if (flow_dv_create_action_push_vlan
13173 (dev, attr, &vlan, dev_flow, error))
13175 dev_flow->dv.actions[actions_n++] =
13176 dev_flow->dv.push_vlan_res->action;
13177 action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
13179 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
13180 /* of_vlan_push action handled this action */
13181 MLX5_ASSERT(action_flags &
13182 MLX5_FLOW_ACTION_OF_PUSH_VLAN);
13184 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
13185 if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
13187 flow_dev_get_vlan_info_from_items(items, &vlan);
13188 mlx5_update_vlan_vid_pcp(actions, &vlan);
13189 /* If no VLAN push - this is a modify header action */
13190 if (flow_dv_convert_action_modify_vlan_vid
13191 (mhdr_res, actions, error))
13193 action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
13195 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
13196 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
13197 if (flow_dv_create_action_l2_encap(dev, actions,
13202 dev_flow->dv.actions[actions_n++] =
13203 dev_flow->dv.encap_decap->action;
13204 action_flags |= MLX5_FLOW_ACTION_ENCAP;
13205 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
13206 sample_act->action_flags |=
13207 MLX5_FLOW_ACTION_ENCAP;
13209 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
13210 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
13211 if (flow_dv_create_action_l2_decap(dev, dev_flow,
13215 dev_flow->dv.actions[actions_n++] =
13216 dev_flow->dv.encap_decap->action;
13217 action_flags |= MLX5_FLOW_ACTION_DECAP;
13219 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
13220 /* Handle encap with preceding decap. */
13221 if (action_flags & MLX5_FLOW_ACTION_DECAP) {
13222 if (flow_dv_create_action_raw_encap
13223 (dev, actions, dev_flow, attr, error))
13225 dev_flow->dv.actions[actions_n++] =
13226 dev_flow->dv.encap_decap->action;
13228 /* Handle encap without preceding decap. */
13229 if (flow_dv_create_action_l2_encap
13230 (dev, actions, dev_flow, attr->transfer,
13233 dev_flow->dv.actions[actions_n++] =
13234 dev_flow->dv.encap_decap->action;
13236 action_flags |= MLX5_FLOW_ACTION_ENCAP;
13237 if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
13238 sample_act->action_flags |=
13239 MLX5_FLOW_ACTION_ENCAP;
13241 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
13242 while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
13244 if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
13245 if (flow_dv_create_action_l2_decap
13246 (dev, dev_flow, attr->transfer, error))
13248 dev_flow->dv.actions[actions_n++] =
13249 dev_flow->dv.encap_decap->action;
13251 /* If decap is followed by encap, handle it at encap. */
13252 action_flags |= MLX5_FLOW_ACTION_DECAP;
13254 case MLX5_RTE_FLOW_ACTION_TYPE_JUMP:
13255 dev_flow->dv.actions[actions_n++] =
13256 (void *)(uintptr_t)action->conf;
13257 action_flags |= MLX5_FLOW_ACTION_JUMP;
13259 case RTE_FLOW_ACTION_TYPE_JUMP:
13260 jump_group = ((const struct rte_flow_action_jump *)
13261 action->conf)->group;
13262 grp_info.std_tbl_fix = 0;
13263 if (dev_flow->skip_scale &
13264 (1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT))
13265 grp_info.skip_scale = 1;
13267 grp_info.skip_scale = 0;
13268 ret = mlx5_flow_group_to_table(dev, tunnel,
13274 tbl = flow_dv_tbl_resource_get(dev, table, attr->egress,
13276 !!dev_flow->external,
13277 tunnel, jump_group, 0,
13280 return rte_flow_error_set
13282 RTE_FLOW_ERROR_TYPE_ACTION,
13284 "cannot create jump action.");
13285 if (flow_dv_jump_tbl_resource_register
13286 (dev, tbl, dev_flow, error)) {
13287 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
13288 return rte_flow_error_set
13290 RTE_FLOW_ERROR_TYPE_ACTION,
13292 "cannot create jump action.");
13294 dev_flow->dv.actions[actions_n++] =
13295 dev_flow->dv.jump->action;
13296 action_flags |= MLX5_FLOW_ACTION_JUMP;
13297 dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
13298 sample_act->action_flags |= MLX5_FLOW_ACTION_JUMP;
13301 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
13302 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
13303 if (flow_dv_convert_action_modify_mac
13304 (mhdr_res, actions, error))
13306 action_flags |= actions->type ==
13307 RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
13308 MLX5_FLOW_ACTION_SET_MAC_SRC :
13309 MLX5_FLOW_ACTION_SET_MAC_DST;
13311 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
13312 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
13313 if (flow_dv_convert_action_modify_ipv4
13314 (mhdr_res, actions, error))
13316 action_flags |= actions->type ==
13317 RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
13318 MLX5_FLOW_ACTION_SET_IPV4_SRC :
13319 MLX5_FLOW_ACTION_SET_IPV4_DST;
13321 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
13322 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
13323 if (flow_dv_convert_action_modify_ipv6
13324 (mhdr_res, actions, error))
13326 action_flags |= actions->type ==
13327 RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
13328 MLX5_FLOW_ACTION_SET_IPV6_SRC :
13329 MLX5_FLOW_ACTION_SET_IPV6_DST;
13331 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
13332 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
13333 if (flow_dv_convert_action_modify_tp
13334 (mhdr_res, actions, items,
13335 &flow_attr, dev_flow, !!(action_flags &
13336 MLX5_FLOW_ACTION_DECAP), error))
13338 action_flags |= actions->type ==
13339 RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
13340 MLX5_FLOW_ACTION_SET_TP_SRC :
13341 MLX5_FLOW_ACTION_SET_TP_DST;
13343 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
13344 if (flow_dv_convert_action_modify_dec_ttl
13345 (mhdr_res, items, &flow_attr, dev_flow,
13347 MLX5_FLOW_ACTION_DECAP), error))
13349 action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
13351 case RTE_FLOW_ACTION_TYPE_SET_TTL:
13352 if (flow_dv_convert_action_modify_ttl
13353 (mhdr_res, actions, items, &flow_attr,
13354 dev_flow, !!(action_flags &
13355 MLX5_FLOW_ACTION_DECAP), error))
13357 action_flags |= MLX5_FLOW_ACTION_SET_TTL;
13359 case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
13360 case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
13361 if (flow_dv_convert_action_modify_tcp_seq
13362 (mhdr_res, actions, error))
13364 action_flags |= actions->type ==
13365 RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
13366 MLX5_FLOW_ACTION_INC_TCP_SEQ :
13367 MLX5_FLOW_ACTION_DEC_TCP_SEQ;
13370 case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
13371 case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
13372 if (flow_dv_convert_action_modify_tcp_ack
13373 (mhdr_res, actions, error))
13375 action_flags |= actions->type ==
13376 RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
13377 MLX5_FLOW_ACTION_INC_TCP_ACK :
13378 MLX5_FLOW_ACTION_DEC_TCP_ACK;
13380 case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
13381 if (flow_dv_convert_action_set_reg
13382 (mhdr_res, actions, error))
13384 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
13386 case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
13387 if (flow_dv_convert_action_copy_mreg
13388 (dev, mhdr_res, actions, error))
13390 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
13392 case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
13393 action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
13394 dev_flow->handle->fate_action =
13395 MLX5_FLOW_FATE_DEFAULT_MISS;
13397 case RTE_FLOW_ACTION_TYPE_METER:
13399 return rte_flow_error_set(error, rte_errno,
13400 RTE_FLOW_ERROR_TYPE_ACTION,
13401 NULL, "Failed to get meter in flow.");
13402 /* Set the meter action. */
13403 dev_flow->dv.actions[actions_n++] =
13404 wks->fm->meter_action;
13405 action_flags |= MLX5_FLOW_ACTION_METER;
13407 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
13408 if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
13411 action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
13413 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
13414 if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
13417 action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
13419 case RTE_FLOW_ACTION_TYPE_SAMPLE:
13420 sample_act_pos = actions_n;
13421 sample = (const struct rte_flow_action_sample *)
13424 action_flags |= MLX5_FLOW_ACTION_SAMPLE;
13425 /* put encap action into group if work with port id */
13426 if ((action_flags & MLX5_FLOW_ACTION_ENCAP) &&
13427 (action_flags & MLX5_FLOW_ACTION_PORT_ID))
13428 sample_act->action_flags |=
13429 MLX5_FLOW_ACTION_ENCAP;
13431 case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
13432 if (flow_dv_convert_action_modify_field
13433 (dev, mhdr_res, actions, attr, error))
13435 action_flags |= MLX5_FLOW_ACTION_MODIFY_FIELD;
13437 case RTE_FLOW_ACTION_TYPE_CONNTRACK:
13438 owner_idx = (uint32_t)(uintptr_t)action->conf;
13439 ct = flow_aso_ct_get_by_idx(dev, owner_idx);
13441 return rte_flow_error_set(error, EINVAL,
13442 RTE_FLOW_ERROR_TYPE_ACTION,
13444 "Failed to get CT object.");
13445 if (mlx5_aso_ct_available(priv->sh, ct))
13446 return rte_flow_error_set(error, rte_errno,
13447 RTE_FLOW_ERROR_TYPE_ACTION,
13449 "CT is unavailable.");
13450 if (ct->is_original)
13451 dev_flow->dv.actions[actions_n] =
13452 ct->dr_action_orig;
13454 dev_flow->dv.actions[actions_n] =
13455 ct->dr_action_rply;
13456 if (flow->ct == 0) {
13457 flow->indirect_type =
13458 MLX5_INDIRECT_ACTION_TYPE_CT;
13459 flow->ct = owner_idx;
13460 __atomic_fetch_add(&ct->refcnt, 1,
13464 action_flags |= MLX5_FLOW_ACTION_CT;
13466 case RTE_FLOW_ACTION_TYPE_END:
13467 actions_end = true;
13468 if (mhdr_res->actions_num) {
13469 /* create modify action if needed. */
13470 if (flow_dv_modify_hdr_resource_register
13471 (dev, mhdr_res, dev_flow, error))
13473 dev_flow->dv.actions[modify_action_position] =
13474 handle->dvh.modify_hdr->action;
13477 * Handle AGE and COUNT action by single HW counter
13478 * when they are not shared.
13480 if (action_flags & MLX5_FLOW_ACTION_AGE) {
13481 if ((non_shared_age && count) ||
13482 !flow_hit_aso_supported(priv->sh, attr)) {
13483 /* Creates age by counters. */
13484 cnt_act = flow_dv_prepare_counter
13491 dev_flow->dv.actions[age_act_pos] =
13495 if (!flow->age && non_shared_age) {
13496 flow->age = flow_dv_aso_age_alloc
13500 flow_dv_aso_age_params_init
13502 non_shared_age->context ?
13503 non_shared_age->context :
13504 (void *)(uintptr_t)
13505 (dev_flow->flow_idx),
13506 non_shared_age->timeout);
13508 age_act = flow_aso_age_get_by_idx(dev,
13510 dev_flow->dv.actions[age_act_pos] =
13511 age_act->dr_action;
13513 if (action_flags & MLX5_FLOW_ACTION_COUNT) {
13515 * Create one count action, to be used
13516 * by all sub-flows.
13518 cnt_act = flow_dv_prepare_counter(dev, dev_flow,
13523 dev_flow->dv.actions[actions_n++] =
13529 if (mhdr_res->actions_num &&
13530 modify_action_position == UINT32_MAX)
13531 modify_action_position = actions_n++;
13533 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
13534 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
13535 int item_type = items->type;
13537 if (!mlx5_flow_os_item_supported(item_type))
13538 return rte_flow_error_set(error, ENOTSUP,
13539 RTE_FLOW_ERROR_TYPE_ITEM,
13540 NULL, "item not supported");
13541 switch (item_type) {
13542 case RTE_FLOW_ITEM_TYPE_PORT_ID:
13543 flow_dv_translate_item_port_id
13544 (dev, match_mask, match_value, items, attr);
13545 last_item = MLX5_FLOW_ITEM_PORT_ID;
13547 case RTE_FLOW_ITEM_TYPE_ETH:
13548 flow_dv_translate_item_eth(match_mask, match_value,
13550 dev_flow->dv.group);
13551 matcher.priority = action_flags &
13552 MLX5_FLOW_ACTION_DEFAULT_MISS &&
13553 !dev_flow->external ?
13554 MLX5_PRIORITY_MAP_L3 :
13555 MLX5_PRIORITY_MAP_L2;
13556 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
13557 MLX5_FLOW_LAYER_OUTER_L2;
13559 case RTE_FLOW_ITEM_TYPE_VLAN:
13560 flow_dv_translate_item_vlan(dev_flow,
13561 match_mask, match_value,
13563 dev_flow->dv.group);
13564 matcher.priority = MLX5_PRIORITY_MAP_L2;
13565 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
13566 MLX5_FLOW_LAYER_INNER_VLAN) :
13567 (MLX5_FLOW_LAYER_OUTER_L2 |
13568 MLX5_FLOW_LAYER_OUTER_VLAN);
13570 case RTE_FLOW_ITEM_TYPE_IPV4:
13571 mlx5_flow_tunnel_ip_check(items, next_protocol,
13572 &item_flags, &tunnel);
13573 flow_dv_translate_item_ipv4(match_mask, match_value,
13575 dev_flow->dv.group);
13576 matcher.priority = MLX5_PRIORITY_MAP_L3;
13577 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
13578 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
13579 if (items->mask != NULL &&
13580 ((const struct rte_flow_item_ipv4 *)
13581 items->mask)->hdr.next_proto_id) {
13583 ((const struct rte_flow_item_ipv4 *)
13584 (items->spec))->hdr.next_proto_id;
13586 ((const struct rte_flow_item_ipv4 *)
13587 (items->mask))->hdr.next_proto_id;
13589 /* Reset for inner layer. */
13590 next_protocol = 0xff;
13593 case RTE_FLOW_ITEM_TYPE_IPV6:
13594 mlx5_flow_tunnel_ip_check(items, next_protocol,
13595 &item_flags, &tunnel);
13596 flow_dv_translate_item_ipv6(match_mask, match_value,
13598 dev_flow->dv.group);
13599 matcher.priority = MLX5_PRIORITY_MAP_L3;
13600 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
13601 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
13602 if (items->mask != NULL &&
13603 ((const struct rte_flow_item_ipv6 *)
13604 items->mask)->hdr.proto) {
13606 ((const struct rte_flow_item_ipv6 *)
13607 items->spec)->hdr.proto;
13609 ((const struct rte_flow_item_ipv6 *)
13610 items->mask)->hdr.proto;
13612 /* Reset for inner layer. */
13613 next_protocol = 0xff;
13616 case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
13617 flow_dv_translate_item_ipv6_frag_ext(match_mask,
13620 last_item = tunnel ?
13621 MLX5_FLOW_LAYER_INNER_L3_IPV6_FRAG_EXT :
13622 MLX5_FLOW_LAYER_OUTER_L3_IPV6_FRAG_EXT;
13623 if (items->mask != NULL &&
13624 ((const struct rte_flow_item_ipv6_frag_ext *)
13625 items->mask)->hdr.next_header) {
13627 ((const struct rte_flow_item_ipv6_frag_ext *)
13628 items->spec)->hdr.next_header;
13630 ((const struct rte_flow_item_ipv6_frag_ext *)
13631 items->mask)->hdr.next_header;
13633 /* Reset for inner layer. */
13634 next_protocol = 0xff;
13637 case RTE_FLOW_ITEM_TYPE_TCP:
13638 flow_dv_translate_item_tcp(match_mask, match_value,
13640 matcher.priority = MLX5_PRIORITY_MAP_L4;
13641 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
13642 MLX5_FLOW_LAYER_OUTER_L4_TCP;
13644 case RTE_FLOW_ITEM_TYPE_UDP:
13645 flow_dv_translate_item_udp(match_mask, match_value,
13647 matcher.priority = MLX5_PRIORITY_MAP_L4;
13648 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
13649 MLX5_FLOW_LAYER_OUTER_L4_UDP;
13651 case RTE_FLOW_ITEM_TYPE_GRE:
13652 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13653 last_item = MLX5_FLOW_LAYER_GRE;
13654 tunnel_item = items;
13657 case RTE_FLOW_ITEM_TYPE_GRE_KEY:
13658 flow_dv_translate_item_gre_key(match_mask,
13659 match_value, items);
13660 last_item = MLX5_FLOW_LAYER_GRE_KEY;
13662 case RTE_FLOW_ITEM_TYPE_GRE_OPTION:
13663 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13664 last_item = MLX5_FLOW_LAYER_GRE;
13665 tunnel_item = items;
13667 case RTE_FLOW_ITEM_TYPE_NVGRE:
13668 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13669 last_item = MLX5_FLOW_LAYER_GRE;
13670 tunnel_item = items;
13672 case RTE_FLOW_ITEM_TYPE_VXLAN:
13673 flow_dv_translate_item_vxlan(dev, attr,
13674 match_mask, match_value,
13676 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13677 last_item = MLX5_FLOW_LAYER_VXLAN;
13679 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
13680 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13681 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
13682 tunnel_item = items;
13684 case RTE_FLOW_ITEM_TYPE_GENEVE:
13685 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13686 last_item = MLX5_FLOW_LAYER_GENEVE;
13687 tunnel_item = items;
13689 case RTE_FLOW_ITEM_TYPE_GENEVE_OPT:
13690 ret = flow_dv_translate_item_geneve_opt(dev, match_mask,
13694 return rte_flow_error_set(error, -ret,
13695 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
13696 "cannot create GENEVE TLV option");
13697 flow->geneve_tlv_option = 1;
13698 last_item = MLX5_FLOW_LAYER_GENEVE_OPT;
13700 case RTE_FLOW_ITEM_TYPE_MPLS:
13701 flow_dv_translate_item_mpls(match_mask, match_value,
13702 items, last_item, tunnel);
13703 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13704 last_item = MLX5_FLOW_LAYER_MPLS;
13706 case RTE_FLOW_ITEM_TYPE_MARK:
13707 flow_dv_translate_item_mark(dev, match_mask,
13708 match_value, items);
13709 last_item = MLX5_FLOW_ITEM_MARK;
13711 case RTE_FLOW_ITEM_TYPE_META:
13712 flow_dv_translate_item_meta(dev, match_mask,
13713 match_value, attr, items);
13714 last_item = MLX5_FLOW_ITEM_METADATA;
13716 case RTE_FLOW_ITEM_TYPE_ICMP:
13717 flow_dv_translate_item_icmp(match_mask, match_value,
13719 matcher.priority = MLX5_PRIORITY_MAP_L4;
13720 last_item = MLX5_FLOW_LAYER_ICMP;
13722 case RTE_FLOW_ITEM_TYPE_ICMP6:
13723 flow_dv_translate_item_icmp6(match_mask, match_value,
13725 matcher.priority = MLX5_PRIORITY_MAP_L4;
13726 last_item = MLX5_FLOW_LAYER_ICMP6;
13728 case RTE_FLOW_ITEM_TYPE_TAG:
13729 flow_dv_translate_item_tag(dev, match_mask,
13730 match_value, items);
13731 last_item = MLX5_FLOW_ITEM_TAG;
13733 case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
13734 flow_dv_translate_mlx5_item_tag(dev, match_mask,
13735 match_value, items);
13736 last_item = MLX5_FLOW_ITEM_TAG;
13738 case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
13739 flow_dv_translate_item_tx_queue(dev, match_mask,
13742 last_item = MLX5_FLOW_ITEM_TX_QUEUE;
13744 case RTE_FLOW_ITEM_TYPE_GTP:
13745 flow_dv_translate_item_gtp(match_mask, match_value,
13747 matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
13748 last_item = MLX5_FLOW_LAYER_GTP;
13750 case RTE_FLOW_ITEM_TYPE_GTP_PSC:
13751 ret = flow_dv_translate_item_gtp_psc(match_mask,
13755 return rte_flow_error_set(error, -ret,
13756 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
13757 "cannot create GTP PSC item");
13758 last_item = MLX5_FLOW_LAYER_GTP_PSC;
13760 case RTE_FLOW_ITEM_TYPE_ECPRI:
13761 if (!mlx5_flex_parser_ecpri_exist(dev)) {
13762 /* Create it only the first time to be used. */
13763 ret = mlx5_flex_parser_ecpri_alloc(dev);
13765 return rte_flow_error_set
13767 RTE_FLOW_ERROR_TYPE_ITEM,
13769 "cannot create eCPRI parser");
13771 flow_dv_translate_item_ecpri(dev, match_mask,
13772 match_value, items,
13774 /* No other protocol should follow eCPRI layer. */
13775 last_item = MLX5_FLOW_LAYER_ECPRI;
13777 case RTE_FLOW_ITEM_TYPE_INTEGRITY:
13778 flow_dv_translate_item_integrity(items, integrity_items,
13781 case RTE_FLOW_ITEM_TYPE_CONNTRACK:
13782 flow_dv_translate_item_aso_ct(dev, match_mask,
13783 match_value, items);
13785 case RTE_FLOW_ITEM_TYPE_FLEX:
13786 flow_dv_translate_item_flex(dev, match_mask,
13787 match_value, items,
13788 dev_flow, tunnel != 0);
13789 last_item = tunnel ? MLX5_FLOW_ITEM_INNER_FLEX :
13790 MLX5_FLOW_ITEM_OUTER_FLEX;
13795 item_flags |= last_item;
13798 * When E-Switch mode is enabled, we have two cases where we need to
13799 * set the source port manually.
13800 * The first one, is in case of NIC ingress steering rule, and the
13801 * second is E-Switch rule where no port_id item was found.
13802 * In both cases the source port is set according the current port
13805 if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) && priv->sh->esw_mode &&
13806 !(attr->egress && !attr->transfer)) {
13807 if (flow_dv_translate_item_port_id(dev, match_mask,
13808 match_value, NULL, attr))
13811 if (item_flags & MLX5_FLOW_ITEM_INTEGRITY) {
13812 flow_dv_translate_item_integrity_post(match_mask, match_value,
13816 if (item_flags & MLX5_FLOW_LAYER_VXLAN_GPE)
13817 flow_dv_translate_item_vxlan_gpe(match_mask, match_value,
13818 tunnel_item, item_flags);
13819 else if (item_flags & MLX5_FLOW_LAYER_GENEVE)
13820 flow_dv_translate_item_geneve(match_mask, match_value,
13821 tunnel_item, item_flags);
13822 else if (item_flags & MLX5_FLOW_LAYER_GRE) {
13823 if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_GRE)
13824 flow_dv_translate_item_gre(match_mask, match_value,
13825 tunnel_item, item_flags);
13826 else if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_NVGRE)
13827 flow_dv_translate_item_nvgre(match_mask, match_value,
13828 tunnel_item, item_flags);
13829 else if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_GRE_OPTION)
13830 flow_dv_translate_item_gre_option(match_mask, match_value,
13831 tunnel_item, gre_item, item_flags);
13833 MLX5_ASSERT(false);
13835 #ifdef RTE_LIBRTE_MLX5_DEBUG
13836 MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
13837 dev_flow->dv.value.buf));
13840 * Layers may be already initialized from prefix flow if this dev_flow
13841 * is the suffix flow.
13843 handle->layers |= item_flags;
13844 if (action_flags & MLX5_FLOW_ACTION_RSS)
13845 flow_dv_hashfields_set(dev_flow->handle->layers,
13847 &dev_flow->hash_fields);
13848 /* If has RSS action in the sample action, the Sample/Mirror resource
13849 * should be registered after the hash filed be update.
13851 if (action_flags & MLX5_FLOW_ACTION_SAMPLE) {
13852 ret = flow_dv_translate_action_sample(dev,
13861 ret = flow_dv_create_action_sample(dev,
13870 return rte_flow_error_set
13872 RTE_FLOW_ERROR_TYPE_ACTION,
13874 "cannot create sample action");
13875 if (num_of_dest > 1) {
13876 dev_flow->dv.actions[sample_act_pos] =
13877 dev_flow->dv.dest_array_res->action;
13879 dev_flow->dv.actions[sample_act_pos] =
13880 dev_flow->dv.sample_res->verbs_action;
13884 * For multiple destination (sample action with ratio=1), the encap
13885 * action and port id action will be combined into group action.
13886 * So need remove the original these actions in the flow and only
13887 * use the sample action instead of.
13889 if (num_of_dest > 1 &&
13890 (sample_act->dr_port_id_action || sample_act->dr_jump_action)) {
13892 void *temp_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
13894 for (i = 0; i < actions_n; i++) {
13895 if ((sample_act->dr_encap_action &&
13896 sample_act->dr_encap_action ==
13897 dev_flow->dv.actions[i]) ||
13898 (sample_act->dr_port_id_action &&
13899 sample_act->dr_port_id_action ==
13900 dev_flow->dv.actions[i]) ||
13901 (sample_act->dr_jump_action &&
13902 sample_act->dr_jump_action ==
13903 dev_flow->dv.actions[i]))
13905 temp_actions[tmp_actions_n++] = dev_flow->dv.actions[i];
13907 memcpy((void *)dev_flow->dv.actions,
13908 (void *)temp_actions,
13909 tmp_actions_n * sizeof(void *));
13910 actions_n = tmp_actions_n;
13912 dev_flow->dv.actions_n = actions_n;
13913 dev_flow->act_flags = action_flags;
13914 if (wks->skip_matcher_reg)
13916 /* Register matcher. */
13917 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
13918 matcher.mask.size);
13919 matcher.priority = mlx5_get_matcher_priority(dev, attr,
13921 dev_flow->external);
13923 * When creating meter drop flow in drop table, using original
13924 * 5-tuple match, the matcher priority should be lower than
13927 if (attr->group == MLX5_FLOW_TABLE_LEVEL_METER &&
13928 dev_flow->dv.table_id == MLX5_MTR_TABLE_ID_DROP &&
13929 matcher.priority <= MLX5_REG_BITS)
13930 matcher.priority += MLX5_REG_BITS;
13931 /* reserved field no needs to be set to 0 here. */
13932 tbl_key.is_fdb = attr->transfer;
13933 tbl_key.is_egress = attr->egress;
13934 tbl_key.level = dev_flow->dv.group;
13935 tbl_key.id = dev_flow->dv.table_id;
13936 if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow,
13937 tunnel, attr->group, error))
13943 * Set hash RX queue by hash fields (see enum ibv_rx_hash_fields)
13946 * @param[in, out] action
13947 * Shred RSS action holding hash RX queue objects.
13948 * @param[in] hash_fields
13949 * Defines combination of packet fields to participate in RX hash.
13950 * @param[in] tunnel
13952 * @param[in] hrxq_idx
13953 * Hash RX queue index to set.
13956 * 0 on success, otherwise negative errno value.
13959 __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
13960 const uint64_t hash_fields,
13963 uint32_t *hrxqs = action->hrxq;
13965 switch (hash_fields & ~IBV_RX_HASH_INNER) {
13966 case MLX5_RSS_HASH_IPV4:
13967 /* fall-through. */
13968 case MLX5_RSS_HASH_IPV4_DST_ONLY:
13969 /* fall-through. */
13970 case MLX5_RSS_HASH_IPV4_SRC_ONLY:
13971 hrxqs[0] = hrxq_idx;
13973 case MLX5_RSS_HASH_IPV4_TCP:
13974 /* fall-through. */
13975 case MLX5_RSS_HASH_IPV4_TCP_DST_ONLY:
13976 /* fall-through. */
13977 case MLX5_RSS_HASH_IPV4_TCP_SRC_ONLY:
13978 hrxqs[1] = hrxq_idx;
13980 case MLX5_RSS_HASH_IPV4_UDP:
13981 /* fall-through. */
13982 case MLX5_RSS_HASH_IPV4_UDP_DST_ONLY:
13983 /* fall-through. */
13984 case MLX5_RSS_HASH_IPV4_UDP_SRC_ONLY:
13985 hrxqs[2] = hrxq_idx;
13987 case MLX5_RSS_HASH_IPV6:
13988 /* fall-through. */
13989 case MLX5_RSS_HASH_IPV6_DST_ONLY:
13990 /* fall-through. */
13991 case MLX5_RSS_HASH_IPV6_SRC_ONLY:
13992 hrxqs[3] = hrxq_idx;
13994 case MLX5_RSS_HASH_IPV6_TCP:
13995 /* fall-through. */
13996 case MLX5_RSS_HASH_IPV6_TCP_DST_ONLY:
13997 /* fall-through. */
13998 case MLX5_RSS_HASH_IPV6_TCP_SRC_ONLY:
13999 hrxqs[4] = hrxq_idx;
14001 case MLX5_RSS_HASH_IPV6_UDP:
14002 /* fall-through. */
14003 case MLX5_RSS_HASH_IPV6_UDP_DST_ONLY:
14004 /* fall-through. */
14005 case MLX5_RSS_HASH_IPV6_UDP_SRC_ONLY:
14006 hrxqs[5] = hrxq_idx;
14008 case MLX5_RSS_HASH_NONE:
14009 hrxqs[6] = hrxq_idx;
14017 * Look up for hash RX queue by hash fields (see enum ibv_rx_hash_fields)
14021 * Pointer to the Ethernet device structure.
14023 * Shared RSS action ID holding hash RX queue objects.
14024 * @param[in] hash_fields
14025 * Defines combination of packet fields to participate in RX hash.
14026 * @param[in] tunnel
14030 * Valid hash RX queue index, otherwise 0.
14033 flow_dv_action_rss_hrxq_lookup(struct rte_eth_dev *dev, uint32_t idx,
14034 const uint64_t hash_fields)
14036 struct mlx5_priv *priv = dev->data->dev_private;
14037 struct mlx5_shared_action_rss *shared_rss =
14038 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
14039 const uint32_t *hrxqs = shared_rss->hrxq;
14041 switch (hash_fields & ~IBV_RX_HASH_INNER) {
14042 case MLX5_RSS_HASH_IPV4:
14043 /* fall-through. */
14044 case MLX5_RSS_HASH_IPV4_DST_ONLY:
14045 /* fall-through. */
14046 case MLX5_RSS_HASH_IPV4_SRC_ONLY:
14048 case MLX5_RSS_HASH_IPV4_TCP:
14049 /* fall-through. */
14050 case MLX5_RSS_HASH_IPV4_TCP_DST_ONLY:
14051 /* fall-through. */
14052 case MLX5_RSS_HASH_IPV4_TCP_SRC_ONLY:
14054 case MLX5_RSS_HASH_IPV4_UDP:
14055 /* fall-through. */
14056 case MLX5_RSS_HASH_IPV4_UDP_DST_ONLY:
14057 /* fall-through. */
14058 case MLX5_RSS_HASH_IPV4_UDP_SRC_ONLY:
14060 case MLX5_RSS_HASH_IPV6:
14061 /* fall-through. */
14062 case MLX5_RSS_HASH_IPV6_DST_ONLY:
14063 /* fall-through. */
14064 case MLX5_RSS_HASH_IPV6_SRC_ONLY:
14066 case MLX5_RSS_HASH_IPV6_TCP:
14067 /* fall-through. */
14068 case MLX5_RSS_HASH_IPV6_TCP_DST_ONLY:
14069 /* fall-through. */
14070 case MLX5_RSS_HASH_IPV6_TCP_SRC_ONLY:
14072 case MLX5_RSS_HASH_IPV6_UDP:
14073 /* fall-through. */
14074 case MLX5_RSS_HASH_IPV6_UDP_DST_ONLY:
14075 /* fall-through. */
14076 case MLX5_RSS_HASH_IPV6_UDP_SRC_ONLY:
14078 case MLX5_RSS_HASH_NONE:
14087 * Apply the flow to the NIC, lock free,
14088 * (mutex should be acquired by caller).
14091 * Pointer to the Ethernet device structure.
14092 * @param[in, out] flow
14093 * Pointer to flow structure.
14094 * @param[out] error
14095 * Pointer to error structure.
14098 * 0 on success, a negative errno value otherwise and rte_errno is set.
14101 flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
14102 struct rte_flow_error *error)
14104 struct mlx5_flow_dv_workspace *dv;
14105 struct mlx5_flow_handle *dh;
14106 struct mlx5_flow_handle_dv *dv_h;
14107 struct mlx5_flow *dev_flow;
14108 struct mlx5_priv *priv = dev->data->dev_private;
14109 uint32_t handle_idx;
14113 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
14114 struct mlx5_flow_rss_desc *rss_desc = &wks->rss_desc;
14118 for (idx = wks->flow_idx - 1; idx >= 0; idx--) {
14119 dev_flow = &wks->flows[idx];
14120 dv = &dev_flow->dv;
14121 dh = dev_flow->handle;
14124 if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
14125 if (dv->transfer) {
14126 MLX5_ASSERT(priv->sh->dr_drop_action);
14127 dv->actions[n++] = priv->sh->dr_drop_action;
14129 #ifdef HAVE_MLX5DV_DR
14130 /* DR supports drop action placeholder. */
14131 MLX5_ASSERT(priv->sh->dr_drop_action);
14132 dv->actions[n++] = dv->group ?
14133 priv->sh->dr_drop_action :
14134 priv->root_drop_action;
14136 /* For DV we use the explicit drop queue. */
14137 MLX5_ASSERT(priv->drop_queue.hrxq);
14139 priv->drop_queue.hrxq->action;
14142 } else if ((dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
14143 !dv_h->rix_sample && !dv_h->rix_dest_array)) {
14144 struct mlx5_hrxq *hrxq;
14147 hrxq = flow_dv_hrxq_prepare(dev, dev_flow, rss_desc,
14152 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14153 "cannot get hash queue");
14156 dh->rix_hrxq = hrxq_idx;
14157 dv->actions[n++] = hrxq->action;
14158 } else if (dh->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
14159 struct mlx5_hrxq *hrxq = NULL;
14162 hrxq_idx = flow_dv_action_rss_hrxq_lookup(dev,
14163 rss_desc->shared_rss,
14164 dev_flow->hash_fields);
14166 hrxq = mlx5_ipool_get
14167 (priv->sh->ipool[MLX5_IPOOL_HRXQ],
14172 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14173 "cannot get hash queue");
14176 dh->rix_srss = rss_desc->shared_rss;
14177 dv->actions[n++] = hrxq->action;
14178 } else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
14179 if (!priv->sh->default_miss_action) {
14182 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14183 "default miss action not be created.");
14186 dv->actions[n++] = priv->sh->default_miss_action;
14188 misc_mask = flow_dv_matcher_enable(dv->value.buf);
14189 __flow_dv_adjust_buf_size(&dv->value.size, misc_mask);
14190 err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
14191 (void *)&dv->value, n,
14192 dv->actions, &dh->drv_flow);
14196 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
14198 (!priv->sh->config.allow_duplicate_pattern &&
14200 "duplicating pattern is not allowed" :
14201 "hardware refuses to create flow");
14204 if (priv->vmwa_context &&
14205 dh->vf_vlan.tag && !dh->vf_vlan.created) {
14207 * The rule contains the VLAN pattern.
14208 * For VF we are going to create VLAN
14209 * interface to make hypervisor set correct
14210 * e-Switch vport context.
14212 mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
14217 err = rte_errno; /* Save rte_errno before cleanup. */
14218 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
14219 handle_idx, dh, next) {
14220 /* hrxq is union, don't clear it if the flag is not set. */
14221 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE && dh->rix_hrxq) {
14222 mlx5_hrxq_release(dev, dh->rix_hrxq);
14224 } else if (dh->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
14227 if (dh->vf_vlan.tag && dh->vf_vlan.created)
14228 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
14230 rte_errno = err; /* Restore rte_errno. */
14235 flow_dv_matcher_remove_cb(void *tool_ctx __rte_unused,
14236 struct mlx5_list_entry *entry)
14238 struct mlx5_flow_dv_matcher *resource = container_of(entry,
14242 claim_zero(mlx5_flow_os_destroy_flow_matcher(resource->matcher_object));
14243 mlx5_free(resource);
14247 * Release the flow matcher.
14250 * Pointer to Ethernet device.
14252 * Index to port ID action resource.
14255 * 1 while a reference on it exists, 0 when freed.
14258 flow_dv_matcher_release(struct rte_eth_dev *dev,
14259 struct mlx5_flow_handle *handle)
14261 struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
14262 struct mlx5_flow_tbl_data_entry *tbl = container_of(matcher->tbl,
14263 typeof(*tbl), tbl);
14266 MLX5_ASSERT(matcher->matcher_object);
14267 ret = mlx5_list_unregister(tbl->matchers, &matcher->entry);
14268 flow_dv_tbl_resource_release(MLX5_SH(dev), &tbl->tbl);
14273 flow_dv_encap_decap_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
14275 struct mlx5_dev_ctx_shared *sh = tool_ctx;
14276 struct mlx5_flow_dv_encap_decap_resource *res =
14277 container_of(entry, typeof(*res), entry);
14279 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
14280 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], res->idx);
14284 * Release an encap/decap resource.
14287 * Pointer to Ethernet device.
14288 * @param encap_decap_idx
14289 * Index of encap decap resource.
14292 * 1 while a reference on it exists, 0 when freed.
14295 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
14296 uint32_t encap_decap_idx)
14298 struct mlx5_priv *priv = dev->data->dev_private;
14299 struct mlx5_flow_dv_encap_decap_resource *resource;
14301 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
14305 MLX5_ASSERT(resource->action);
14306 return mlx5_hlist_unregister(priv->sh->encaps_decaps, &resource->entry);
14310 * Release an jump to table action resource.
14313 * Pointer to Ethernet device.
14315 * Index to the jump action resource.
14318 * 1 while a reference on it exists, 0 when freed.
14321 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
14324 struct mlx5_priv *priv = dev->data->dev_private;
14325 struct mlx5_flow_tbl_data_entry *tbl_data;
14327 tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
14331 return flow_dv_tbl_resource_release(MLX5_SH(dev), &tbl_data->tbl);
14335 flow_dv_modify_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
14337 struct mlx5_flow_dv_modify_hdr_resource *res =
14338 container_of(entry, typeof(*res), entry);
14339 struct mlx5_dev_ctx_shared *sh = tool_ctx;
14341 claim_zero(mlx5_flow_os_destroy_flow_action(res->action));
14342 mlx5_ipool_free(sh->mdh_ipools[res->actions_num - 1], res->idx);
14346 * Release a modify-header resource.
14349 * Pointer to Ethernet device.
14351 * Pointer to mlx5_flow_handle.
14354 * 1 while a reference on it exists, 0 when freed.
14357 flow_dv_modify_hdr_resource_release(struct rte_eth_dev *dev,
14358 struct mlx5_flow_handle *handle)
14360 struct mlx5_priv *priv = dev->data->dev_private;
14361 struct mlx5_flow_dv_modify_hdr_resource *entry = handle->dvh.modify_hdr;
14363 MLX5_ASSERT(entry->action);
14364 return mlx5_hlist_unregister(priv->sh->modify_cmds, &entry->entry);
14368 flow_dv_port_id_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
14370 struct mlx5_dev_ctx_shared *sh = tool_ctx;
14371 struct mlx5_flow_dv_port_id_action_resource *resource =
14372 container_of(entry, typeof(*resource), entry);
14374 claim_zero(mlx5_flow_os_destroy_flow_action(resource->action));
14375 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PORT_ID], resource->idx);
14379 * Release port ID action resource.
14382 * Pointer to Ethernet device.
14384 * Pointer to mlx5_flow_handle.
14387 * 1 while a reference on it exists, 0 when freed.
14390 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
14393 struct mlx5_priv *priv = dev->data->dev_private;
14394 struct mlx5_flow_dv_port_id_action_resource *resource;
14396 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID], port_id);
14399 MLX5_ASSERT(resource->action);
14400 return mlx5_list_unregister(priv->sh->port_id_action_list,
14405 * Release shared RSS action resource.
14408 * Pointer to Ethernet device.
14410 * Shared RSS action index.
14413 flow_dv_shared_rss_action_release(struct rte_eth_dev *dev, uint32_t srss)
14415 struct mlx5_priv *priv = dev->data->dev_private;
14416 struct mlx5_shared_action_rss *shared_rss;
14418 shared_rss = mlx5_ipool_get
14419 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], srss);
14420 __atomic_sub_fetch(&shared_rss->refcnt, 1, __ATOMIC_RELAXED);
14424 flow_dv_push_vlan_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
14426 struct mlx5_dev_ctx_shared *sh = tool_ctx;
14427 struct mlx5_flow_dv_push_vlan_action_resource *resource =
14428 container_of(entry, typeof(*resource), entry);
14430 claim_zero(mlx5_flow_os_destroy_flow_action(resource->action));
14431 mlx5_ipool_free(sh->ipool[MLX5_IPOOL_PUSH_VLAN], resource->idx);
14435 * Release push vlan action resource.
14438 * Pointer to Ethernet device.
14440 * Pointer to mlx5_flow_handle.
14443 * 1 while a reference on it exists, 0 when freed.
14446 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
14447 struct mlx5_flow_handle *handle)
14449 struct mlx5_priv *priv = dev->data->dev_private;
14450 struct mlx5_flow_dv_push_vlan_action_resource *resource;
14451 uint32_t idx = handle->dvh.rix_push_vlan;
14453 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
14456 MLX5_ASSERT(resource->action);
14457 return mlx5_list_unregister(priv->sh->push_vlan_action_list,
14462 * Release the fate resource.
14465 * Pointer to Ethernet device.
14467 * Pointer to mlx5_flow_handle.
14470 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
14471 struct mlx5_flow_handle *handle)
14473 if (!handle->rix_fate)
14475 switch (handle->fate_action) {
14476 case MLX5_FLOW_FATE_QUEUE:
14477 if (!handle->dvh.rix_sample && !handle->dvh.rix_dest_array)
14478 mlx5_hrxq_release(dev, handle->rix_hrxq);
14480 case MLX5_FLOW_FATE_JUMP:
14481 flow_dv_jump_tbl_resource_release(dev, handle->rix_jump);
14483 case MLX5_FLOW_FATE_PORT_ID:
14484 flow_dv_port_id_action_resource_release(dev,
14485 handle->rix_port_id_action);
14488 DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
14491 handle->rix_fate = 0;
14495 flow_dv_sample_remove_cb(void *tool_ctx __rte_unused,
14496 struct mlx5_list_entry *entry)
14498 struct mlx5_flow_dv_sample_resource *resource = container_of(entry,
14501 struct rte_eth_dev *dev = resource->dev;
14502 struct mlx5_priv *priv = dev->data->dev_private;
14504 if (resource->verbs_action)
14505 claim_zero(mlx5_flow_os_destroy_flow_action
14506 (resource->verbs_action));
14507 if (resource->normal_path_tbl)
14508 flow_dv_tbl_resource_release(MLX5_SH(dev),
14509 resource->normal_path_tbl);
14510 flow_dv_sample_sub_actions_release(dev, &resource->sample_idx);
14511 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE], resource->idx);
14512 DRV_LOG(DEBUG, "sample resource %p: removed", (void *)resource);
14516 * Release an sample resource.
14519 * Pointer to Ethernet device.
14521 * Pointer to mlx5_flow_handle.
14524 * 1 while a reference on it exists, 0 when freed.
14527 flow_dv_sample_resource_release(struct rte_eth_dev *dev,
14528 struct mlx5_flow_handle *handle)
14530 struct mlx5_priv *priv = dev->data->dev_private;
14531 struct mlx5_flow_dv_sample_resource *resource;
14533 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
14534 handle->dvh.rix_sample);
14537 MLX5_ASSERT(resource->verbs_action);
14538 return mlx5_list_unregister(priv->sh->sample_action_list,
14543 flow_dv_dest_array_remove_cb(void *tool_ctx __rte_unused,
14544 struct mlx5_list_entry *entry)
14546 struct mlx5_flow_dv_dest_array_resource *resource =
14547 container_of(entry, typeof(*resource), entry);
14548 struct rte_eth_dev *dev = resource->dev;
14549 struct mlx5_priv *priv = dev->data->dev_private;
14552 MLX5_ASSERT(resource->action);
14553 if (resource->action)
14554 claim_zero(mlx5_flow_os_destroy_flow_action(resource->action));
14555 for (; i < resource->num_of_dest; i++)
14556 flow_dv_sample_sub_actions_release(dev,
14557 &resource->sample_idx[i]);
14558 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY], resource->idx);
14559 DRV_LOG(DEBUG, "destination array resource %p: removed",
14564 * Release an destination array resource.
14567 * Pointer to Ethernet device.
14569 * Pointer to mlx5_flow_handle.
14572 * 1 while a reference on it exists, 0 when freed.
14575 flow_dv_dest_array_resource_release(struct rte_eth_dev *dev,
14576 struct mlx5_flow_handle *handle)
14578 struct mlx5_priv *priv = dev->data->dev_private;
14579 struct mlx5_flow_dv_dest_array_resource *resource;
14581 resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
14582 handle->dvh.rix_dest_array);
14585 MLX5_ASSERT(resource->action);
14586 return mlx5_list_unregister(priv->sh->dest_array_list,
14591 flow_dv_geneve_tlv_option_resource_release(struct rte_eth_dev *dev)
14593 struct mlx5_priv *priv = dev->data->dev_private;
14594 struct mlx5_dev_ctx_shared *sh = priv->sh;
14595 struct mlx5_geneve_tlv_option_resource *geneve_opt_resource =
14596 sh->geneve_tlv_option_resource;
14597 rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
14598 if (geneve_opt_resource) {
14599 if (!(__atomic_sub_fetch(&geneve_opt_resource->refcnt, 1,
14600 __ATOMIC_RELAXED))) {
14601 claim_zero(mlx5_devx_cmd_destroy
14602 (geneve_opt_resource->obj));
14603 mlx5_free(sh->geneve_tlv_option_resource);
14604 sh->geneve_tlv_option_resource = NULL;
14607 rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
14611 * Remove the flow from the NIC but keeps it in memory.
14612 * Lock free, (mutex should be acquired by caller).
14615 * Pointer to Ethernet device.
14616 * @param[in, out] flow
14617 * Pointer to flow structure.
14620 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
14622 struct mlx5_flow_handle *dh;
14623 uint32_t handle_idx;
14624 struct mlx5_priv *priv = dev->data->dev_private;
14628 handle_idx = flow->dev_handles;
14629 while (handle_idx) {
14630 dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
14634 if (dh->drv_flow) {
14635 claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
14636 dh->drv_flow = NULL;
14638 if (dh->fate_action == MLX5_FLOW_FATE_QUEUE)
14639 flow_dv_fate_resource_release(dev, dh);
14640 if (dh->vf_vlan.tag && dh->vf_vlan.created)
14641 mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
14642 handle_idx = dh->next.next;
14647 * Remove the flow from the NIC and the memory.
14648 * Lock free, (mutex should be acquired by caller).
14651 * Pointer to the Ethernet device structure.
14652 * @param[in, out] flow
14653 * Pointer to flow structure.
14656 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
14658 struct mlx5_flow_handle *dev_handle;
14659 struct mlx5_priv *priv = dev->data->dev_private;
14660 struct mlx5_flow_meter_info *fm = NULL;
14665 flow_dv_remove(dev, flow);
14666 if (flow->counter) {
14667 flow_dv_counter_free(dev, flow->counter);
14671 fm = flow_dv_meter_find_by_idx(priv, flow->meter);
14673 mlx5_flow_meter_detach(priv, fm);
14676 /* Keep the current age handling by default. */
14677 if (flow->indirect_type == MLX5_INDIRECT_ACTION_TYPE_CT && flow->ct)
14678 flow_dv_aso_ct_release(dev, flow->ct, NULL);
14679 else if (flow->age)
14680 flow_dv_aso_age_release(dev, flow->age);
14681 if (flow->geneve_tlv_option) {
14682 flow_dv_geneve_tlv_option_resource_release(dev);
14683 flow->geneve_tlv_option = 0;
14685 while (flow->dev_handles) {
14686 uint32_t tmp_idx = flow->dev_handles;
14688 dev_handle = mlx5_ipool_get(priv->sh->ipool
14689 [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
14692 flow->dev_handles = dev_handle->next.next;
14693 while (dev_handle->flex_item) {
14694 int index = rte_bsf32(dev_handle->flex_item);
14696 mlx5_flex_release_index(dev, index);
14697 dev_handle->flex_item &= ~(uint8_t)RTE_BIT32(index);
14699 if (dev_handle->dvh.matcher)
14700 flow_dv_matcher_release(dev, dev_handle);
14701 if (dev_handle->dvh.rix_sample)
14702 flow_dv_sample_resource_release(dev, dev_handle);
14703 if (dev_handle->dvh.rix_dest_array)
14704 flow_dv_dest_array_resource_release(dev, dev_handle);
14705 if (dev_handle->dvh.rix_encap_decap)
14706 flow_dv_encap_decap_resource_release(dev,
14707 dev_handle->dvh.rix_encap_decap);
14708 if (dev_handle->dvh.modify_hdr)
14709 flow_dv_modify_hdr_resource_release(dev, dev_handle);
14710 if (dev_handle->dvh.rix_push_vlan)
14711 flow_dv_push_vlan_action_resource_release(dev,
14713 if (dev_handle->dvh.rix_tag)
14714 flow_dv_tag_release(dev,
14715 dev_handle->dvh.rix_tag);
14716 if (dev_handle->fate_action != MLX5_FLOW_FATE_SHARED_RSS)
14717 flow_dv_fate_resource_release(dev, dev_handle);
14719 srss = dev_handle->rix_srss;
14720 if (fm && dev_handle->is_meter_flow_id &&
14721 dev_handle->split_flow_id)
14722 mlx5_ipool_free(fm->flow_ipool,
14723 dev_handle->split_flow_id);
14724 else if (dev_handle->split_flow_id &&
14725 !dev_handle->is_meter_flow_id)
14726 mlx5_ipool_free(priv->sh->ipool
14727 [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
14728 dev_handle->split_flow_id);
14729 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
14733 flow_dv_shared_rss_action_release(dev, srss);
14737 * Release array of hash RX queue objects.
14741 * Pointer to the Ethernet device structure.
14742 * @param[in, out] hrxqs
14743 * Array of hash RX queue objects.
14746 * Total number of references to hash RX queue objects in *hrxqs* array
14747 * after this operation.
14750 __flow_dv_hrxqs_release(struct rte_eth_dev *dev,
14751 uint32_t (*hrxqs)[MLX5_RSS_HASH_FIELDS_LEN])
14756 for (i = 0; i < RTE_DIM(*hrxqs); i++) {
14757 int ret = mlx5_hrxq_release(dev, (*hrxqs)[i]);
14767 * Release all hash RX queue objects representing shared RSS action.
14770 * Pointer to the Ethernet device structure.
14771 * @param[in, out] action
14772 * Shared RSS action to remove hash RX queue objects from.
14775 * Total number of references to hash RX queue objects stored in *action*
14776 * after this operation.
14777 * Expected to be 0 if no external references held.
14780 __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
14781 struct mlx5_shared_action_rss *shared_rss)
14783 return __flow_dv_hrxqs_release(dev, &shared_rss->hrxq);
14787 * Adjust L3/L4 hash value of pre-created shared RSS hrxq according to
14790 * Only one hash value is available for one L3+L4 combination:
14792 * MLX5_RSS_HASH_IPV4, MLX5_RSS_HASH_IPV4_SRC_ONLY, and
14793 * MLX5_RSS_HASH_IPV4_DST_ONLY are mutually exclusive so they can share
14794 * same slot in mlx5_rss_hash_fields.
14796 * @param[in] orig_rss_types
14797 * RSS type as provided in shared RSS action.
14798 * @param[in, out] hash_field
14799 * hash_field variable needed to be adjusted.
14805 flow_dv_action_rss_l34_hash_adjust(uint64_t orig_rss_types,
14806 uint64_t *hash_field)
14808 uint64_t rss_types = rte_eth_rss_hf_refine(orig_rss_types);
14810 switch (*hash_field & ~IBV_RX_HASH_INNER) {
14811 case MLX5_RSS_HASH_IPV4:
14812 if (rss_types & MLX5_IPV4_LAYER_TYPES) {
14813 *hash_field &= ~MLX5_RSS_HASH_IPV4;
14814 if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
14815 *hash_field |= IBV_RX_HASH_DST_IPV4;
14816 else if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
14817 *hash_field |= IBV_RX_HASH_SRC_IPV4;
14819 *hash_field |= MLX5_RSS_HASH_IPV4;
14822 case MLX5_RSS_HASH_IPV6:
14823 if (rss_types & MLX5_IPV6_LAYER_TYPES) {
14824 *hash_field &= ~MLX5_RSS_HASH_IPV6;
14825 if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
14826 *hash_field |= IBV_RX_HASH_DST_IPV6;
14827 else if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
14828 *hash_field |= IBV_RX_HASH_SRC_IPV6;
14830 *hash_field |= MLX5_RSS_HASH_IPV6;
14833 case MLX5_RSS_HASH_IPV4_UDP:
14834 /* fall-through. */
14835 case MLX5_RSS_HASH_IPV6_UDP:
14836 if (rss_types & RTE_ETH_RSS_UDP) {
14837 *hash_field &= ~MLX5_UDP_IBV_RX_HASH;
14838 if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
14839 *hash_field |= IBV_RX_HASH_DST_PORT_UDP;
14840 else if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
14841 *hash_field |= IBV_RX_HASH_SRC_PORT_UDP;
14843 *hash_field |= MLX5_UDP_IBV_RX_HASH;
14846 case MLX5_RSS_HASH_IPV4_TCP:
14847 /* fall-through. */
14848 case MLX5_RSS_HASH_IPV6_TCP:
14849 if (rss_types & RTE_ETH_RSS_TCP) {
14850 *hash_field &= ~MLX5_TCP_IBV_RX_HASH;
14851 if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
14852 *hash_field |= IBV_RX_HASH_DST_PORT_TCP;
14853 else if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
14854 *hash_field |= IBV_RX_HASH_SRC_PORT_TCP;
14856 *hash_field |= MLX5_TCP_IBV_RX_HASH;
14865 * Setup shared RSS action.
14866 * Prepare set of hash RX queue objects sufficient to handle all valid
14867 * hash_fields combinations (see enum ibv_rx_hash_fields).
14870 * Pointer to the Ethernet device structure.
14871 * @param[in] action_idx
14872 * Shared RSS action ipool index.
14873 * @param[in, out] action
14874 * Partially initialized shared RSS action.
14875 * @param[out] error
14876 * Perform verbose error reporting if not NULL. Initialized in case of
14880 * 0 on success, otherwise negative errno value.
14883 __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
14884 uint32_t action_idx,
14885 struct mlx5_shared_action_rss *shared_rss,
14886 struct rte_flow_error *error)
14888 struct mlx5_priv *priv = dev->data->dev_private;
14889 struct mlx5_flow_rss_desc rss_desc = { 0 };
14893 shared_rss->ind_tbl = mlx5_ind_table_obj_new
14894 (dev, shared_rss->origin.queue,
14895 shared_rss->origin.queue_num,
14897 !!dev->data->dev_started);
14898 if (!shared_rss->ind_tbl)
14899 return rte_flow_error_set(error, rte_errno,
14900 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14901 "cannot setup indirection table");
14902 memcpy(rss_desc.key, shared_rss->origin.key, MLX5_RSS_HASH_KEY_LEN);
14903 rss_desc.key_len = MLX5_RSS_HASH_KEY_LEN;
14904 rss_desc.const_q = shared_rss->origin.queue;
14905 rss_desc.queue_num = shared_rss->origin.queue_num;
14906 /* Set non-zero value to indicate a shared RSS. */
14907 rss_desc.shared_rss = action_idx;
14908 rss_desc.ind_tbl = shared_rss->ind_tbl;
14909 if (priv->sh->config.dv_flow_en == 2)
14910 rss_desc.hws_flags = MLX5DR_ACTION_FLAG_HWS_RX;
14911 for (i = 0; i < MLX5_RSS_HASH_FIELDS_LEN; i++) {
14912 struct mlx5_hrxq *hrxq;
14913 uint64_t hash_fields = mlx5_rss_hash_fields[i];
14916 flow_dv_action_rss_l34_hash_adjust(shared_rss->origin.types,
14918 if (shared_rss->origin.level > 1) {
14919 hash_fields |= IBV_RX_HASH_INNER;
14922 rss_desc.tunnel = tunnel;
14923 rss_desc.hash_fields = hash_fields;
14924 hrxq = mlx5_hrxq_get(dev, &rss_desc);
14928 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14929 "cannot get hash queue");
14930 goto error_hrxq_new;
14932 err = __flow_dv_action_rss_hrxq_set
14933 (shared_rss, hash_fields, hrxq->idx);
14939 __flow_dv_action_rss_hrxqs_release(dev, shared_rss);
14940 if (!mlx5_ind_table_obj_release(dev, shared_rss->ind_tbl, true))
14941 shared_rss->ind_tbl = NULL;
14947 * Create shared RSS action.
14950 * Pointer to the Ethernet device structure.
14952 * Shared action configuration.
14954 * RSS action specification used to create shared action.
14955 * @param[out] error
14956 * Perform verbose error reporting if not NULL. Initialized in case of
14960 * A valid shared action ID in case of success, 0 otherwise and
14961 * rte_errno is set.
14964 __flow_dv_action_rss_create(struct rte_eth_dev *dev,
14965 const struct rte_flow_indir_action_conf *conf,
14966 const struct rte_flow_action_rss *rss,
14967 struct rte_flow_error *error)
14969 struct mlx5_priv *priv = dev->data->dev_private;
14970 struct mlx5_shared_action_rss *shared_rss = NULL;
14971 struct rte_flow_action_rss *origin;
14972 const uint8_t *rss_key;
14975 RTE_SET_USED(conf);
14976 shared_rss = mlx5_ipool_zmalloc
14977 (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], &idx);
14979 rte_flow_error_set(error, ENOMEM,
14980 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14981 "cannot allocate resource memory");
14982 goto error_rss_init;
14984 if (idx > (1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET)) {
14985 rte_flow_error_set(error, E2BIG,
14986 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
14987 "rss action number out of range");
14988 goto error_rss_init;
14990 origin = &shared_rss->origin;
14991 origin->func = rss->func;
14992 origin->level = rss->level;
14993 /* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
14994 origin->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
14995 /* NULL RSS key indicates default RSS key. */
14996 rss_key = !rss->key ? rss_hash_default_key : rss->key;
14997 memcpy(shared_rss->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
14998 origin->key = &shared_rss->key[0];
14999 origin->key_len = MLX5_RSS_HASH_KEY_LEN;
15000 origin->queue = rss->queue;
15001 origin->queue_num = rss->queue_num;
15002 if (__flow_dv_action_rss_setup(dev, idx, shared_rss, error))
15003 goto error_rss_init;
15004 /* Update queue with indirect table queue memoyr. */
15005 origin->queue = shared_rss->ind_tbl->queues;
15006 rte_spinlock_init(&shared_rss->action_rss_sl);
15007 __atomic_add_fetch(&shared_rss->refcnt, 1, __ATOMIC_RELAXED);
15008 rte_spinlock_lock(&priv->shared_act_sl);
15009 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
15010 &priv->rss_shared_actions, idx, shared_rss, next);
15011 rte_spinlock_unlock(&priv->shared_act_sl);
15015 if (shared_rss->ind_tbl)
15016 mlx5_ind_table_obj_release(dev, shared_rss->ind_tbl,
15017 !!dev->data->dev_started);
15018 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
15025 * Destroy the shared RSS action.
15026 * Release related hash RX queue objects.
15029 * Pointer to the Ethernet device structure.
15031 * The shared RSS action object ID to be removed.
15032 * @param[out] error
15033 * Perform verbose error reporting if not NULL. Initialized in case of
15037 * 0 on success, otherwise negative errno value.
15040 __flow_dv_action_rss_release(struct rte_eth_dev *dev, uint32_t idx,
15041 struct rte_flow_error *error)
15043 struct mlx5_priv *priv = dev->data->dev_private;
15044 struct mlx5_shared_action_rss *shared_rss =
15045 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
15046 uint32_t old_refcnt = 1;
15050 return rte_flow_error_set(error, EINVAL,
15051 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
15052 "invalid shared action");
15053 if (!__atomic_compare_exchange_n(&shared_rss->refcnt, &old_refcnt,
15054 0, 0, __ATOMIC_ACQUIRE,
15056 return rte_flow_error_set(error, EBUSY,
15057 RTE_FLOW_ERROR_TYPE_ACTION,
15059 "shared rss has references");
15060 remaining = __flow_dv_action_rss_hrxqs_release(dev, shared_rss);
15062 return rte_flow_error_set(error, EBUSY,
15063 RTE_FLOW_ERROR_TYPE_ACTION,
15065 "shared rss hrxq has references");
15066 remaining = mlx5_ind_table_obj_release(dev, shared_rss->ind_tbl,
15067 !!dev->data->dev_started);
15069 return rte_flow_error_set(error, EBUSY,
15070 RTE_FLOW_ERROR_TYPE_ACTION,
15072 "shared rss indirection table has"
15074 rte_spinlock_lock(&priv->shared_act_sl);
15075 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
15076 &priv->rss_shared_actions, idx, shared_rss, next);
15077 rte_spinlock_unlock(&priv->shared_act_sl);
15078 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
15084 * Create indirect action, lock free,
15085 * (mutex should be acquired by caller).
15086 * Dispatcher for action type specific call.
15089 * Pointer to the Ethernet device structure.
15091 * Shared action configuration.
15092 * @param[in] action
15093 * Action specification used to create indirect action.
15094 * @param[out] error
15095 * Perform verbose error reporting if not NULL. Initialized in case of
15099 * A valid shared action handle in case of success, NULL otherwise and
15100 * rte_errno is set.
15102 struct rte_flow_action_handle *
15103 flow_dv_action_create(struct rte_eth_dev *dev,
15104 const struct rte_flow_indir_action_conf *conf,
15105 const struct rte_flow_action *action,
15106 struct rte_flow_error *err)
15108 struct mlx5_priv *priv = dev->data->dev_private;
15109 uint32_t age_idx = 0;
15113 switch (action->type) {
15114 case RTE_FLOW_ACTION_TYPE_RSS:
15115 ret = __flow_dv_action_rss_create(dev, conf, action->conf, err);
15116 idx = (MLX5_INDIRECT_ACTION_TYPE_RSS <<
15117 MLX5_INDIRECT_ACTION_TYPE_OFFSET) | ret;
15119 case RTE_FLOW_ACTION_TYPE_AGE:
15120 age_idx = flow_dv_aso_age_alloc(dev, err);
15125 idx = (MLX5_INDIRECT_ACTION_TYPE_AGE <<
15126 MLX5_INDIRECT_ACTION_TYPE_OFFSET) | age_idx;
15127 flow_dv_aso_age_params_init(dev, age_idx,
15128 ((const struct rte_flow_action_age *)
15129 action->conf)->context ?
15130 ((const struct rte_flow_action_age *)
15131 action->conf)->context :
15132 (void *)(uintptr_t)idx,
15133 ((const struct rte_flow_action_age *)
15134 action->conf)->timeout);
15137 case RTE_FLOW_ACTION_TYPE_COUNT:
15138 ret = flow_dv_translate_create_counter(dev, NULL, NULL, NULL);
15139 idx = (MLX5_INDIRECT_ACTION_TYPE_COUNT <<
15140 MLX5_INDIRECT_ACTION_TYPE_OFFSET) | ret;
15142 case RTE_FLOW_ACTION_TYPE_CONNTRACK:
15143 ret = flow_dv_translate_create_conntrack(dev, action->conf,
15145 idx = MLX5_INDIRECT_ACT_CT_GEN_IDX(PORT_ID(priv), ret);
15148 rte_flow_error_set(err, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
15149 NULL, "action type not supported");
15152 return ret ? (struct rte_flow_action_handle *)(uintptr_t)idx : NULL;
15156 * Destroy the indirect action.
15157 * Release action related resources on the NIC and the memory.
15158 * Lock free, (mutex should be acquired by caller).
15159 * Dispatcher for action type specific call.
15162 * Pointer to the Ethernet device structure.
15163 * @param[in] handle
15164 * The indirect action object handle to be removed.
15165 * @param[out] error
15166 * Perform verbose error reporting if not NULL. Initialized in case of
15170 * 0 on success, otherwise negative errno value.
15173 flow_dv_action_destroy(struct rte_eth_dev *dev,
15174 struct rte_flow_action_handle *handle,
15175 struct rte_flow_error *error)
15177 uint32_t act_idx = (uint32_t)(uintptr_t)handle;
15178 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
15179 uint32_t idx = act_idx & ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
15180 struct mlx5_flow_counter *cnt;
15181 uint32_t no_flow_refcnt = 1;
15185 case MLX5_INDIRECT_ACTION_TYPE_RSS:
15186 return __flow_dv_action_rss_release(dev, idx, error);
15187 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
15188 cnt = flow_dv_counter_get_by_idx(dev, idx, NULL);
15189 if (!__atomic_compare_exchange_n(&cnt->shared_info.refcnt,
15190 &no_flow_refcnt, 1, false,
15193 return rte_flow_error_set(error, EBUSY,
15194 RTE_FLOW_ERROR_TYPE_ACTION,
15196 "Indirect count action has references");
15197 flow_dv_counter_free(dev, idx);
15199 case MLX5_INDIRECT_ACTION_TYPE_AGE:
15200 ret = flow_dv_aso_age_release(dev, idx);
15203 * In this case, the last flow has a reference will
15204 * actually release the age action.
15206 DRV_LOG(DEBUG, "Indirect age action %" PRIu32 " was"
15207 " released with references %d.", idx, ret);
15209 case MLX5_INDIRECT_ACTION_TYPE_CT:
15210 ret = flow_dv_aso_ct_release(dev, idx, error);
15214 DRV_LOG(DEBUG, "Connection tracking object %u still "
15215 "has references %d.", idx, ret);
15218 return rte_flow_error_set(error, ENOTSUP,
15219 RTE_FLOW_ERROR_TYPE_ACTION,
15221 "action type not supported");
15226 * Updates in place shared RSS action configuration.
15229 * Pointer to the Ethernet device structure.
15231 * The shared RSS action object ID to be updated.
15232 * @param[in] action_conf
15233 * RSS action specification used to modify *shared_rss*.
15234 * @param[out] error
15235 * Perform verbose error reporting if not NULL. Initialized in case of
15239 * 0 on success, otherwise negative errno value.
15240 * @note: currently only support update of RSS queues.
15243 __flow_dv_action_rss_update(struct rte_eth_dev *dev, uint32_t idx,
15244 const struct rte_flow_action_rss *action_conf,
15245 struct rte_flow_error *error)
15247 struct mlx5_priv *priv = dev->data->dev_private;
15248 struct mlx5_shared_action_rss *shared_rss =
15249 mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
15251 void *queue = NULL;
15252 void *queue_i = NULL;
15253 uint32_t queue_size = action_conf->queue_num * sizeof(uint16_t);
15254 bool dev_started = !!dev->data->dev_started;
15257 return rte_flow_error_set(error, EINVAL,
15258 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
15259 "invalid shared action to update");
15260 if (priv->obj_ops.ind_table_modify == NULL)
15261 return rte_flow_error_set(error, ENOTSUP,
15262 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
15263 "cannot modify indirection table");
15264 queue = mlx5_malloc(MLX5_MEM_ZERO,
15265 RTE_ALIGN_CEIL(queue_size, sizeof(void *)),
15268 return rte_flow_error_set(error, ENOMEM,
15269 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15271 "cannot allocate resource memory");
15272 memcpy(queue, action_conf->queue, queue_size);
15273 MLX5_ASSERT(shared_rss->ind_tbl);
15274 rte_spinlock_lock(&shared_rss->action_rss_sl);
15275 queue_i = shared_rss->ind_tbl->queues;
15276 ret = mlx5_ind_table_obj_modify(dev, shared_rss->ind_tbl,
15277 queue, action_conf->queue_num,
15278 true /* standalone */,
15279 dev_started /* ref_new_qs */,
15280 dev_started /* deref_old_qs */);
15282 ret = rte_flow_error_set(error, rte_errno,
15283 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
15284 "cannot update indirection table");
15286 /* Restore the queue to indirect table internal queue. */
15287 memcpy(queue_i, queue, queue_size);
15288 shared_rss->ind_tbl->queues = queue_i;
15289 shared_rss->origin.queue_num = action_conf->queue_num;
15292 rte_spinlock_unlock(&shared_rss->action_rss_sl);
15297 * Updates in place conntrack context or direction.
15298 * Context update should be synchronized.
15301 * Pointer to the Ethernet device structure.
15303 * The conntrack object ID to be updated.
15304 * @param[in] update
15305 * Pointer to the structure of information to update.
15306 * @param[out] error
15307 * Perform verbose error reporting if not NULL. Initialized in case of
15311 * 0 on success, otherwise negative errno value.
15314 __flow_dv_action_ct_update(struct rte_eth_dev *dev, uint32_t idx,
15315 const struct rte_flow_modify_conntrack *update,
15316 struct rte_flow_error *error)
15318 struct mlx5_priv *priv = dev->data->dev_private;
15319 struct mlx5_aso_ct_action *ct;
15320 const struct rte_flow_action_conntrack *new_prf;
15322 uint16_t owner = (uint16_t)MLX5_INDIRECT_ACT_CT_GET_OWNER(idx);
15325 if (PORT_ID(priv) != owner)
15326 return rte_flow_error_set(error, EACCES,
15327 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15329 "CT object owned by another port");
15330 dev_idx = MLX5_INDIRECT_ACT_CT_GET_IDX(idx);
15331 ct = flow_aso_ct_get_by_dev_idx(dev, dev_idx);
15333 return rte_flow_error_set(error, ENOMEM,
15334 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15336 "CT object is inactive");
15337 new_prf = &update->new_ct;
15338 if (update->direction)
15339 ct->is_original = !!new_prf->is_original_dir;
15340 if (update->state) {
15341 /* Only validate the profile when it needs to be updated. */
15342 ret = mlx5_validate_action_ct(dev, new_prf, error);
15345 ret = mlx5_aso_ct_update_by_wqe(priv->sh, ct, new_prf);
15347 return rte_flow_error_set(error, EIO,
15348 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15350 "Failed to send CT context update WQE");
15351 /* Block until ready or a failure. */
15352 ret = mlx5_aso_ct_available(priv->sh, ct);
15354 rte_flow_error_set(error, rte_errno,
15355 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
15357 "Timeout to get the CT update");
15363 * Updates in place shared action configuration, lock free,
15364 * (mutex should be acquired by caller).
15367 * Pointer to the Ethernet device structure.
15368 * @param[in] handle
15369 * The indirect action object handle to be updated.
15370 * @param[in] update
15371 * Action specification used to modify the action pointed by *handle*.
15372 * *update* could be of same type with the action pointed by the *handle*
15373 * handle argument, or some other structures like a wrapper, depending on
15374 * the indirect action type.
15375 * @param[out] error
15376 * Perform verbose error reporting if not NULL. Initialized in case of
15380 * 0 on success, otherwise negative errno value.
15383 flow_dv_action_update(struct rte_eth_dev *dev,
15384 struct rte_flow_action_handle *handle,
15385 const void *update,
15386 struct rte_flow_error *err)
15388 uint32_t act_idx = (uint32_t)(uintptr_t)handle;
15389 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
15390 uint32_t idx = act_idx & ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
15391 const void *action_conf;
15394 case MLX5_INDIRECT_ACTION_TYPE_RSS:
15395 action_conf = ((const struct rte_flow_action *)update)->conf;
15396 return __flow_dv_action_rss_update(dev, idx, action_conf, err);
15397 case MLX5_INDIRECT_ACTION_TYPE_CT:
15398 return __flow_dv_action_ct_update(dev, idx, update, err);
15400 return rte_flow_error_set(err, ENOTSUP,
15401 RTE_FLOW_ERROR_TYPE_ACTION,
15403 "action type update not supported");
15408 * Destroy the meter sub policy table rules.
15409 * Lock free, (mutex should be acquired by caller).
15412 * Pointer to Ethernet device.
15413 * @param[in] sub_policy
15414 * Pointer to meter sub policy table.
15417 __flow_dv_destroy_sub_policy_rules(struct rte_eth_dev *dev,
15418 struct mlx5_flow_meter_sub_policy *sub_policy)
15420 struct mlx5_priv *priv = dev->data->dev_private;
15421 struct mlx5_flow_tbl_data_entry *tbl;
15422 struct mlx5_flow_meter_policy *policy = sub_policy->main_policy;
15423 struct mlx5_flow_meter_info *next_fm;
15424 struct mlx5_sub_policy_color_rule *color_rule;
15428 for (i = 0; i < RTE_COLORS; i++) {
15430 if (i == RTE_COLOR_GREEN && policy &&
15431 policy->act_cnt[i].fate_action == MLX5_FLOW_FATE_MTR)
15432 next_fm = mlx5_flow_meter_find(priv,
15433 policy->act_cnt[i].next_mtr_id, NULL);
15434 RTE_TAILQ_FOREACH_SAFE(color_rule, &sub_policy->color_rules[i],
15436 claim_zero(mlx5_flow_os_destroy_flow(color_rule->rule));
15437 tbl = container_of(color_rule->matcher->tbl,
15438 typeof(*tbl), tbl);
15439 mlx5_list_unregister(tbl->matchers,
15440 &color_rule->matcher->entry);
15441 TAILQ_REMOVE(&sub_policy->color_rules[i],
15442 color_rule, next_port);
15443 mlx5_free(color_rule);
15445 mlx5_flow_meter_detach(priv, next_fm);
15448 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
15449 if (sub_policy->rix_hrxq[i]) {
15450 if (policy && !policy->is_hierarchy)
15451 mlx5_hrxq_release(dev, sub_policy->rix_hrxq[i]);
15452 sub_policy->rix_hrxq[i] = 0;
15454 if (sub_policy->jump_tbl[i]) {
15455 flow_dv_tbl_resource_release(MLX5_SH(dev),
15456 sub_policy->jump_tbl[i]);
15457 sub_policy->jump_tbl[i] = NULL;
15460 if (sub_policy->tbl_rsc) {
15461 flow_dv_tbl_resource_release(MLX5_SH(dev),
15462 sub_policy->tbl_rsc);
15463 sub_policy->tbl_rsc = NULL;
15468 * Destroy policy rules, lock free,
15469 * (mutex should be acquired by caller).
15470 * Dispatcher for action type specific call.
15473 * Pointer to the Ethernet device structure.
15474 * @param[in] mtr_policy
15475 * Meter policy struct.
15478 flow_dv_destroy_policy_rules(struct rte_eth_dev *dev,
15479 struct mlx5_flow_meter_policy *mtr_policy)
15482 struct mlx5_flow_meter_sub_policy *sub_policy;
15483 uint16_t sub_policy_num;
15485 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
15486 sub_policy_num = (mtr_policy->sub_policy_num >>
15487 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i)) &
15488 MLX5_MTR_SUB_POLICY_NUM_MASK;
15489 for (j = 0; j < sub_policy_num; j++) {
15490 sub_policy = mtr_policy->sub_policys[i][j];
15492 __flow_dv_destroy_sub_policy_rules(dev,
15499 * Destroy policy action, lock free,
15500 * (mutex should be acquired by caller).
15501 * Dispatcher for action type specific call.
15504 * Pointer to the Ethernet device structure.
15505 * @param[in] mtr_policy
15506 * Meter policy struct.
15509 flow_dv_destroy_mtr_policy_acts(struct rte_eth_dev *dev,
15510 struct mlx5_flow_meter_policy *mtr_policy)
15512 struct rte_flow_action *rss_action;
15513 struct mlx5_flow_handle dev_handle;
15516 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
15517 if (mtr_policy->act_cnt[i].rix_mark) {
15518 flow_dv_tag_release(dev,
15519 mtr_policy->act_cnt[i].rix_mark);
15520 mtr_policy->act_cnt[i].rix_mark = 0;
15522 if (mtr_policy->act_cnt[i].modify_hdr) {
15523 dev_handle.dvh.modify_hdr =
15524 mtr_policy->act_cnt[i].modify_hdr;
15525 flow_dv_modify_hdr_resource_release(dev, &dev_handle);
15527 switch (mtr_policy->act_cnt[i].fate_action) {
15528 case MLX5_FLOW_FATE_SHARED_RSS:
15529 rss_action = mtr_policy->act_cnt[i].rss;
15530 mlx5_free(rss_action);
15532 case MLX5_FLOW_FATE_PORT_ID:
15533 if (mtr_policy->act_cnt[i].rix_port_id_action) {
15534 flow_dv_port_id_action_resource_release(dev,
15535 mtr_policy->act_cnt[i].rix_port_id_action);
15536 mtr_policy->act_cnt[i].rix_port_id_action = 0;
15539 case MLX5_FLOW_FATE_DROP:
15540 case MLX5_FLOW_FATE_JUMP:
15541 for (j = 0; j < MLX5_MTR_DOMAIN_MAX; j++)
15542 mtr_policy->act_cnt[i].dr_jump_action[j] =
15546 /*Queue action do nothing*/
15550 for (j = 0; j < MLX5_MTR_DOMAIN_MAX; j++)
15551 mtr_policy->dr_drop_action[j] = NULL;
15555 * Create policy action per domain, lock free,
15556 * (mutex should be acquired by caller).
15557 * Dispatcher for action type specific call.
15560 * Pointer to the Ethernet device structure.
15561 * @param[in] mtr_policy
15562 * Meter policy struct.
15563 * @param[in] action
15564 * Action specification used to create meter actions.
15565 * @param[out] error
15566 * Perform verbose error reporting if not NULL. Initialized in case of
15570 * 0 on success, otherwise negative errno value.
15573 __flow_dv_create_domain_policy_acts(struct rte_eth_dev *dev,
15574 struct mlx5_flow_meter_policy *mtr_policy,
15575 const struct rte_flow_action *actions[RTE_COLORS],
15576 enum mlx5_meter_domain domain,
15577 struct rte_mtr_error *error)
15579 struct mlx5_priv *priv = dev->data->dev_private;
15580 struct rte_flow_error flow_err;
15581 const struct rte_flow_action *act;
15582 uint64_t action_flags;
15583 struct mlx5_flow_handle dh;
15584 struct mlx5_flow dev_flow;
15585 struct mlx5_flow_dv_port_id_action_resource port_id_action;
15587 uint8_t egress, transfer;
15588 struct mlx5_meter_policy_action_container *act_cnt = NULL;
15590 struct mlx5_flow_dv_modify_hdr_resource res;
15591 uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
15592 sizeof(struct mlx5_modification_cmd) *
15593 (MLX5_MAX_MODIFY_NUM + 1)];
15595 struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
15597 egress = (domain == MLX5_MTR_DOMAIN_EGRESS) ? 1 : 0;
15598 transfer = (domain == MLX5_MTR_DOMAIN_TRANSFER) ? 1 : 0;
15599 memset(&dh, 0, sizeof(struct mlx5_flow_handle));
15600 memset(&dev_flow, 0, sizeof(struct mlx5_flow));
15601 memset(&port_id_action, 0,
15602 sizeof(struct mlx5_flow_dv_port_id_action_resource));
15603 memset(mhdr_res, 0, sizeof(*mhdr_res));
15604 mhdr_res->ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
15605 (egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
15606 MLX5DV_FLOW_TABLE_TYPE_NIC_RX);
15607 dev_flow.handle = &dh;
15608 dev_flow.dv.port_id_action = &port_id_action;
15609 dev_flow.external = true;
15610 for (i = 0; i < RTE_COLORS; i++) {
15611 if (i < MLX5_MTR_RTE_COLORS)
15612 act_cnt = &mtr_policy->act_cnt[i];
15613 /* Skip the color policy actions creation. */
15614 if ((i == RTE_COLOR_YELLOW && mtr_policy->skip_y) ||
15615 (i == RTE_COLOR_GREEN && mtr_policy->skip_g))
15618 for (act = actions[i];
15619 act && act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
15620 switch (act->type) {
15621 case RTE_FLOW_ACTION_TYPE_MARK:
15623 uint32_t tag_be = mlx5_flow_mark_set
15624 (((const struct rte_flow_action_mark *)
15627 if (i >= MLX5_MTR_RTE_COLORS)
15628 return -rte_mtr_error_set(error,
15630 RTE_MTR_ERROR_TYPE_METER_POLICY,
15632 "cannot create policy "
15633 "mark action for this color");
15634 if (flow_dv_tag_resource_register(dev, tag_be,
15635 &dev_flow, &flow_err))
15636 return -rte_mtr_error_set(error,
15638 RTE_MTR_ERROR_TYPE_METER_POLICY,
15640 "cannot setup policy mark action");
15641 MLX5_ASSERT(dev_flow.dv.tag_resource);
15642 act_cnt->rix_mark =
15643 dev_flow.handle->dvh.rix_tag;
15644 action_flags |= MLX5_FLOW_ACTION_MARK;
15645 mtr_policy->mark = 1;
15648 case RTE_FLOW_ACTION_TYPE_SET_TAG:
15649 if (i >= MLX5_MTR_RTE_COLORS)
15650 return -rte_mtr_error_set(error,
15652 RTE_MTR_ERROR_TYPE_METER_POLICY,
15654 "cannot create policy "
15655 "set tag action for this color");
15656 if (flow_dv_convert_action_set_tag
15658 (const struct rte_flow_action_set_tag *)
15659 act->conf, &flow_err))
15660 return -rte_mtr_error_set(error,
15662 RTE_MTR_ERROR_TYPE_METER_POLICY,
15663 NULL, "cannot convert policy "
15665 if (!mhdr_res->actions_num)
15666 return -rte_mtr_error_set(error,
15668 RTE_MTR_ERROR_TYPE_METER_POLICY,
15669 NULL, "cannot find policy "
15671 action_flags |= MLX5_FLOW_ACTION_SET_TAG;
15673 case RTE_FLOW_ACTION_TYPE_DROP:
15675 struct mlx5_flow_mtr_mng *mtrmng =
15677 struct mlx5_flow_tbl_data_entry *tbl_data;
15680 * Create the drop table with
15681 * METER DROP level.
15683 if (!mtrmng->drop_tbl[domain]) {
15684 mtrmng->drop_tbl[domain] =
15685 flow_dv_tbl_resource_get(dev,
15686 MLX5_FLOW_TABLE_LEVEL_METER,
15687 egress, transfer, false, NULL, 0,
15688 0, MLX5_MTR_TABLE_ID_DROP, &flow_err);
15689 if (!mtrmng->drop_tbl[domain])
15690 return -rte_mtr_error_set
15692 RTE_MTR_ERROR_TYPE_METER_POLICY,
15694 "Failed to create meter drop table");
15696 tbl_data = container_of
15697 (mtrmng->drop_tbl[domain],
15698 struct mlx5_flow_tbl_data_entry, tbl);
15699 if (i < MLX5_MTR_RTE_COLORS) {
15700 act_cnt->dr_jump_action[domain] =
15701 tbl_data->jump.action;
15702 act_cnt->fate_action =
15703 MLX5_FLOW_FATE_DROP;
15705 if (i == RTE_COLOR_RED)
15706 mtr_policy->dr_drop_action[domain] =
15707 tbl_data->jump.action;
15708 action_flags |= MLX5_FLOW_ACTION_DROP;
15711 case RTE_FLOW_ACTION_TYPE_QUEUE:
15713 if (i >= MLX5_MTR_RTE_COLORS)
15714 return -rte_mtr_error_set(error,
15716 RTE_MTR_ERROR_TYPE_METER_POLICY,
15717 NULL, "cannot create policy "
15718 "fate queue for this color");
15720 ((const struct rte_flow_action_queue *)
15721 (act->conf))->index;
15722 act_cnt->fate_action =
15723 MLX5_FLOW_FATE_QUEUE;
15724 dev_flow.handle->fate_action =
15725 MLX5_FLOW_FATE_QUEUE;
15726 mtr_policy->is_queue = 1;
15727 action_flags |= MLX5_FLOW_ACTION_QUEUE;
15730 case RTE_FLOW_ACTION_TYPE_RSS:
15734 if (i >= MLX5_MTR_RTE_COLORS)
15735 return -rte_mtr_error_set(error,
15737 RTE_MTR_ERROR_TYPE_METER_POLICY,
15739 "cannot create policy "
15740 "rss action for this color");
15742 * Save RSS conf into policy struct
15743 * for translate stage.
15745 rss_size = (int)rte_flow_conv
15746 (RTE_FLOW_CONV_OP_ACTION,
15747 NULL, 0, act, &flow_err);
15749 return -rte_mtr_error_set(error,
15751 RTE_MTR_ERROR_TYPE_METER_POLICY,
15752 NULL, "Get the wrong "
15753 "rss action struct size");
15754 act_cnt->rss = mlx5_malloc(MLX5_MEM_ZERO,
15755 rss_size, 0, SOCKET_ID_ANY);
15757 return -rte_mtr_error_set(error,
15759 RTE_MTR_ERROR_TYPE_METER_POLICY,
15761 "Fail to malloc rss action memory");
15762 ret = rte_flow_conv(RTE_FLOW_CONV_OP_ACTION,
15763 act_cnt->rss, rss_size,
15766 return -rte_mtr_error_set(error,
15768 RTE_MTR_ERROR_TYPE_METER_POLICY,
15769 NULL, "Fail to save "
15770 "rss action into policy struct");
15771 act_cnt->fate_action =
15772 MLX5_FLOW_FATE_SHARED_RSS;
15773 action_flags |= MLX5_FLOW_ACTION_RSS;
15776 case RTE_FLOW_ACTION_TYPE_PORT_ID:
15777 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
15779 struct mlx5_flow_dv_port_id_action_resource
15781 uint32_t port_id = 0;
15783 if (i >= MLX5_MTR_RTE_COLORS)
15784 return -rte_mtr_error_set(error,
15786 RTE_MTR_ERROR_TYPE_METER_POLICY,
15787 NULL, "cannot create policy "
15788 "port action for this color");
15789 memset(&port_id_resource, 0,
15790 sizeof(port_id_resource));
15791 if (flow_dv_translate_action_port_id(dev, act,
15792 &port_id, &flow_err))
15793 return -rte_mtr_error_set(error,
15795 RTE_MTR_ERROR_TYPE_METER_POLICY,
15796 NULL, "cannot translate "
15797 "policy port action");
15798 port_id_resource.port_id = port_id;
15799 if (flow_dv_port_id_action_resource_register
15800 (dev, &port_id_resource,
15801 &dev_flow, &flow_err))
15802 return -rte_mtr_error_set(error,
15804 RTE_MTR_ERROR_TYPE_METER_POLICY,
15805 NULL, "cannot setup "
15806 "policy port action");
15807 act_cnt->rix_port_id_action =
15808 dev_flow.handle->rix_port_id_action;
15809 act_cnt->fate_action =
15810 MLX5_FLOW_FATE_PORT_ID;
15811 action_flags |= MLX5_FLOW_ACTION_PORT_ID;
15814 case RTE_FLOW_ACTION_TYPE_JUMP:
15816 uint32_t jump_group = 0;
15817 uint32_t table = 0;
15818 struct mlx5_flow_tbl_data_entry *tbl_data;
15819 struct flow_grp_info grp_info = {
15820 .external = !!dev_flow.external,
15821 .transfer = !!transfer,
15822 .fdb_def_rule = !!priv->fdb_def_rule,
15824 .skip_scale = dev_flow.skip_scale &
15825 (1 << MLX5_SCALE_FLOW_GROUP_BIT),
15827 struct mlx5_flow_meter_sub_policy *sub_policy =
15828 mtr_policy->sub_policys[domain][0];
15830 if (i >= MLX5_MTR_RTE_COLORS)
15831 return -rte_mtr_error_set(error,
15833 RTE_MTR_ERROR_TYPE_METER_POLICY,
15835 "cannot create policy "
15836 "jump action for this color");
15838 ((const struct rte_flow_action_jump *)
15840 if (mlx5_flow_group_to_table(dev, NULL,
15843 &grp_info, &flow_err))
15844 return -rte_mtr_error_set(error,
15846 RTE_MTR_ERROR_TYPE_METER_POLICY,
15847 NULL, "cannot setup "
15848 "policy jump action");
15849 sub_policy->jump_tbl[i] =
15850 flow_dv_tbl_resource_get(dev,
15853 !!dev_flow.external,
15854 NULL, jump_group, 0,
15857 (!sub_policy->jump_tbl[i])
15858 return -rte_mtr_error_set(error,
15860 RTE_MTR_ERROR_TYPE_METER_POLICY,
15861 NULL, "cannot create jump action.");
15862 tbl_data = container_of
15863 (sub_policy->jump_tbl[i],
15864 struct mlx5_flow_tbl_data_entry, tbl);
15865 act_cnt->dr_jump_action[domain] =
15866 tbl_data->jump.action;
15867 act_cnt->fate_action =
15868 MLX5_FLOW_FATE_JUMP;
15869 action_flags |= MLX5_FLOW_ACTION_JUMP;
15873 * No need to check meter hierarchy for Y or R colors
15874 * here since it is done in the validation stage.
15876 case RTE_FLOW_ACTION_TYPE_METER:
15878 const struct rte_flow_action_meter *mtr;
15879 struct mlx5_flow_meter_info *next_fm;
15880 struct mlx5_flow_meter_policy *next_policy;
15881 struct rte_flow_action tag_action;
15882 struct mlx5_rte_flow_action_set_tag set_tag;
15883 uint32_t next_mtr_idx = 0;
15886 next_fm = mlx5_flow_meter_find(priv,
15890 return -rte_mtr_error_set(error, EINVAL,
15891 RTE_MTR_ERROR_TYPE_MTR_ID, NULL,
15892 "Fail to find next meter.");
15893 if (next_fm->def_policy)
15894 return -rte_mtr_error_set(error, EINVAL,
15895 RTE_MTR_ERROR_TYPE_MTR_ID, NULL,
15896 "Hierarchy only supports termination meter.");
15897 next_policy = mlx5_flow_meter_policy_find(dev,
15898 next_fm->policy_id, NULL);
15899 MLX5_ASSERT(next_policy);
15900 if (next_fm->drop_cnt) {
15903 mlx5_flow_get_reg_id(dev,
15906 (struct rte_flow_error *)error);
15907 set_tag.offset = (priv->mtr_reg_share ?
15908 MLX5_MTR_COLOR_BITS : 0);
15909 set_tag.length = (priv->mtr_reg_share ?
15910 MLX5_MTR_IDLE_BITS_IN_COLOR_REG :
15912 set_tag.data = next_mtr_idx;
15914 (enum rte_flow_action_type)
15915 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
15916 tag_action.conf = &set_tag;
15917 if (flow_dv_convert_action_set_reg
15918 (mhdr_res, &tag_action,
15919 (struct rte_flow_error *)error))
15922 MLX5_FLOW_ACTION_SET_TAG;
15924 act_cnt->fate_action = MLX5_FLOW_FATE_MTR;
15925 act_cnt->next_mtr_id = next_fm->meter_id;
15926 act_cnt->next_sub_policy = NULL;
15927 mtr_policy->is_hierarchy = 1;
15928 mtr_policy->dev = next_policy->dev;
15929 if (next_policy->mark)
15930 mtr_policy->mark = 1;
15932 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY;
15936 return -rte_mtr_error_set(error, ENOTSUP,
15937 RTE_MTR_ERROR_TYPE_METER_POLICY,
15938 NULL, "action type not supported");
15940 if (action_flags & MLX5_FLOW_ACTION_SET_TAG) {
15941 /* create modify action if needed. */
15942 dev_flow.dv.group = 1;
15943 if (flow_dv_modify_hdr_resource_register
15944 (dev, mhdr_res, &dev_flow, &flow_err))
15945 return -rte_mtr_error_set(error,
15947 RTE_MTR_ERROR_TYPE_METER_POLICY,
15948 NULL, "cannot register policy "
15950 act_cnt->modify_hdr =
15951 dev_flow.handle->dvh.modify_hdr;
15959 * Create policy action per domain, lock free,
15960 * (mutex should be acquired by caller).
15961 * Dispatcher for action type specific call.
15964 * Pointer to the Ethernet device structure.
15965 * @param[in] mtr_policy
15966 * Meter policy struct.
15967 * @param[in] action
15968 * Action specification used to create meter actions.
15969 * @param[out] error
15970 * Perform verbose error reporting if not NULL. Initialized in case of
15974 * 0 on success, otherwise negative errno value.
15977 flow_dv_create_mtr_policy_acts(struct rte_eth_dev *dev,
15978 struct mlx5_flow_meter_policy *mtr_policy,
15979 const struct rte_flow_action *actions[RTE_COLORS],
15980 struct rte_mtr_error *error)
15983 uint16_t sub_policy_num;
15985 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
15986 sub_policy_num = (mtr_policy->sub_policy_num >>
15987 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i)) &
15988 MLX5_MTR_SUB_POLICY_NUM_MASK;
15989 if (sub_policy_num) {
15990 ret = __flow_dv_create_domain_policy_acts(dev,
15991 mtr_policy, actions,
15992 (enum mlx5_meter_domain)i, error);
15993 /* Cleaning resource is done in the caller level. */
16002 * Query a DV flow rule for its statistics via DevX.
16005 * Pointer to Ethernet device.
16006 * @param[in] cnt_idx
16007 * Index to the flow counter.
16009 * Data retrieved by the query.
16010 * @param[out] error
16011 * Perform verbose error reporting if not NULL.
16014 * 0 on success, a negative errno value otherwise and rte_errno is set.
16017 flow_dv_query_count(struct rte_eth_dev *dev, uint32_t cnt_idx, void *data,
16018 struct rte_flow_error *error)
16020 struct mlx5_priv *priv = dev->data->dev_private;
16021 struct rte_flow_query_count *qc = data;
16023 if (!priv->sh->cdev->config.devx)
16024 return rte_flow_error_set(error, ENOTSUP,
16025 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
16027 "counters are not supported");
16029 uint64_t pkts, bytes;
16030 struct mlx5_flow_counter *cnt;
16031 int err = _flow_dv_query_count(dev, cnt_idx, &pkts, &bytes);
16034 return rte_flow_error_set(error, -err,
16035 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
16036 NULL, "cannot read counters");
16037 cnt = flow_dv_counter_get_by_idx(dev, cnt_idx, NULL);
16040 qc->hits = pkts - cnt->hits;
16041 qc->bytes = bytes - cnt->bytes;
16044 cnt->bytes = bytes;
16048 return rte_flow_error_set(error, EINVAL,
16049 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
16051 "counters are not available");
16055 flow_dv_action_query(struct rte_eth_dev *dev,
16056 const struct rte_flow_action_handle *handle, void *data,
16057 struct rte_flow_error *error)
16059 struct mlx5_age_param *age_param;
16060 struct rte_flow_query_age *resp;
16061 uint32_t act_idx = (uint32_t)(uintptr_t)handle;
16062 uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
16063 uint32_t idx = act_idx & ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
16064 struct mlx5_priv *priv = dev->data->dev_private;
16065 struct mlx5_aso_ct_action *ct;
16070 case MLX5_INDIRECT_ACTION_TYPE_AGE:
16071 age_param = &flow_aso_age_get_by_idx(dev, idx)->age_params;
16073 resp->aged = __atomic_load_n(&age_param->state,
16074 __ATOMIC_RELAXED) == AGE_TMOUT ?
16076 resp->sec_since_last_hit_valid = !resp->aged;
16077 if (resp->sec_since_last_hit_valid)
16078 resp->sec_since_last_hit = __atomic_load_n
16079 (&age_param->sec_since_last_hit, __ATOMIC_RELAXED);
16081 case MLX5_INDIRECT_ACTION_TYPE_COUNT:
16082 return flow_dv_query_count(dev, idx, data, error);
16083 case MLX5_INDIRECT_ACTION_TYPE_CT:
16084 owner = (uint16_t)MLX5_INDIRECT_ACT_CT_GET_OWNER(idx);
16085 if (owner != PORT_ID(priv))
16086 return rte_flow_error_set(error, EACCES,
16087 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
16089 "CT object owned by another port");
16090 dev_idx = MLX5_INDIRECT_ACT_CT_GET_IDX(idx);
16091 ct = flow_aso_ct_get_by_dev_idx(dev, dev_idx);
16094 return rte_flow_error_set(error, EFAULT,
16095 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
16097 "CT object is inactive");
16098 ((struct rte_flow_action_conntrack *)data)->peer_port =
16100 ((struct rte_flow_action_conntrack *)data)->is_original_dir =
16102 if (mlx5_aso_ct_query_by_wqe(priv->sh, ct, data))
16103 return rte_flow_error_set(error, EIO,
16104 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
16106 "Failed to query CT context");
16109 return rte_flow_error_set(error, ENOTSUP,
16110 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
16111 "action type query not supported");
16116 * Query a flow rule AGE action for aging information.
16119 * Pointer to Ethernet device.
16121 * Pointer to the sub flow.
16123 * data retrieved by the query.
16124 * @param[out] error
16125 * Perform verbose error reporting if not NULL.
16128 * 0 on success, a negative errno value otherwise and rte_errno is set.
16131 flow_dv_query_age(struct rte_eth_dev *dev, struct rte_flow *flow,
16132 void *data, struct rte_flow_error *error)
16134 struct rte_flow_query_age *resp = data;
16135 struct mlx5_age_param *age_param;
16138 struct mlx5_aso_age_action *act =
16139 flow_aso_age_get_by_idx(dev, flow->age);
16141 age_param = &act->age_params;
16142 } else if (flow->counter) {
16143 age_param = flow_dv_counter_idx_get_age(dev, flow->counter);
16145 if (!age_param || !age_param->timeout)
16146 return rte_flow_error_set
16148 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
16149 NULL, "cannot read age data");
16151 return rte_flow_error_set(error, EINVAL,
16152 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
16153 NULL, "age data not available");
16155 resp->aged = __atomic_load_n(&age_param->state, __ATOMIC_RELAXED) ==
16157 resp->sec_since_last_hit_valid = !resp->aged;
16158 if (resp->sec_since_last_hit_valid)
16159 resp->sec_since_last_hit = __atomic_load_n
16160 (&age_param->sec_since_last_hit, __ATOMIC_RELAXED);
16167 * @see rte_flow_query()
16168 * @see rte_flow_ops
16171 flow_dv_query(struct rte_eth_dev *dev,
16172 struct rte_flow *flow __rte_unused,
16173 const struct rte_flow_action *actions __rte_unused,
16174 void *data __rte_unused,
16175 struct rte_flow_error *error __rte_unused)
16179 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
16180 switch (actions->type) {
16181 case RTE_FLOW_ACTION_TYPE_VOID:
16183 case RTE_FLOW_ACTION_TYPE_COUNT:
16184 ret = flow_dv_query_count(dev, flow->counter, data,
16187 case RTE_FLOW_ACTION_TYPE_AGE:
16188 ret = flow_dv_query_age(dev, flow, data, error);
16191 return rte_flow_error_set(error, ENOTSUP,
16192 RTE_FLOW_ERROR_TYPE_ACTION,
16194 "action not supported");
16201 * Destroy the meter table set.
16202 * Lock free, (mutex should be acquired by caller).
16205 * Pointer to Ethernet device.
16207 * Meter information table.
16210 flow_dv_destroy_mtr_tbls(struct rte_eth_dev *dev,
16211 struct mlx5_flow_meter_info *fm)
16213 struct mlx5_priv *priv = dev->data->dev_private;
16216 if (!fm || !priv->sh->config.dv_flow_en)
16218 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16219 if (fm->drop_rule[i]) {
16220 claim_zero(mlx5_flow_os_destroy_flow(fm->drop_rule[i]));
16221 fm->drop_rule[i] = NULL;
16227 flow_dv_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
16229 struct mlx5_priv *priv = dev->data->dev_private;
16230 struct mlx5_flow_mtr_mng *mtrmng = priv->sh->mtrmng;
16231 struct mlx5_flow_tbl_data_entry *tbl;
16234 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16235 if (mtrmng->def_rule[i]) {
16236 claim_zero(mlx5_flow_os_destroy_flow
16237 (mtrmng->def_rule[i]));
16238 mtrmng->def_rule[i] = NULL;
16240 if (mtrmng->def_matcher[i]) {
16241 tbl = container_of(mtrmng->def_matcher[i]->tbl,
16242 struct mlx5_flow_tbl_data_entry, tbl);
16243 mlx5_list_unregister(tbl->matchers,
16244 &mtrmng->def_matcher[i]->entry);
16245 mtrmng->def_matcher[i] = NULL;
16247 for (j = 0; j < MLX5_REG_BITS; j++) {
16248 if (mtrmng->drop_matcher[i][j]) {
16250 container_of(mtrmng->drop_matcher[i][j]->tbl,
16251 struct mlx5_flow_tbl_data_entry,
16253 mlx5_list_unregister(tbl->matchers,
16254 &mtrmng->drop_matcher[i][j]->entry);
16255 mtrmng->drop_matcher[i][j] = NULL;
16258 if (mtrmng->drop_tbl[i]) {
16259 flow_dv_tbl_resource_release(MLX5_SH(dev),
16260 mtrmng->drop_tbl[i]);
16261 mtrmng->drop_tbl[i] = NULL;
16266 /* Number of meter flow actions, count and jump or count and drop. */
16267 #define METER_ACTIONS 2
16270 __flow_dv_destroy_domain_def_policy(struct rte_eth_dev *dev,
16271 enum mlx5_meter_domain domain)
16273 struct mlx5_priv *priv = dev->data->dev_private;
16274 struct mlx5_flow_meter_def_policy *def_policy =
16275 priv->sh->mtrmng->def_policy[domain];
16277 __flow_dv_destroy_sub_policy_rules(dev, &def_policy->sub_policy);
16278 mlx5_free(def_policy);
16279 priv->sh->mtrmng->def_policy[domain] = NULL;
16283 * Destroy the default policy table set.
16286 * Pointer to Ethernet device.
16289 flow_dv_destroy_def_policy(struct rte_eth_dev *dev)
16291 struct mlx5_priv *priv = dev->data->dev_private;
16294 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++)
16295 if (priv->sh->mtrmng->def_policy[i])
16296 __flow_dv_destroy_domain_def_policy(dev,
16297 (enum mlx5_meter_domain)i);
16298 priv->sh->mtrmng->def_policy_id = MLX5_INVALID_POLICY_ID;
16302 __flow_dv_create_policy_flow(struct rte_eth_dev *dev,
16303 uint32_t color_reg_c_idx,
16304 enum rte_color color, void *matcher_object,
16305 int actions_n, void *actions,
16306 bool match_src_port, const struct rte_flow_item *item,
16307 void **rule, const struct rte_flow_attr *attr)
16310 struct mlx5_flow_dv_match_params value = {
16311 .size = sizeof(value.buf),
16313 struct mlx5_flow_dv_match_params matcher = {
16314 .size = sizeof(matcher.buf),
16316 struct mlx5_priv *priv = dev->data->dev_private;
16319 if (match_src_port && priv->sh->esw_mode) {
16320 if (flow_dv_translate_item_port_id(dev, matcher.buf,
16321 value.buf, item, attr)) {
16322 DRV_LOG(ERR, "Failed to create meter policy%d flow's"
16323 " value with port.", color);
16327 flow_dv_match_meta_reg(matcher.buf, value.buf,
16328 (enum modify_reg)color_reg_c_idx,
16329 rte_col_2_mlx5_col(color), UINT32_MAX);
16330 misc_mask = flow_dv_matcher_enable(value.buf);
16331 __flow_dv_adjust_buf_size(&value.size, misc_mask);
16332 ret = mlx5_flow_os_create_flow(matcher_object, (void *)&value,
16333 actions_n, actions, rule);
16335 DRV_LOG(ERR, "Failed to create meter policy%d flow.", color);
16342 __flow_dv_create_policy_matcher(struct rte_eth_dev *dev,
16343 uint32_t color_reg_c_idx,
16345 struct mlx5_flow_meter_sub_policy *sub_policy,
16346 const struct rte_flow_attr *attr,
16347 bool match_src_port,
16348 const struct rte_flow_item *item,
16349 struct mlx5_flow_dv_matcher **policy_matcher,
16350 struct rte_flow_error *error)
16352 struct mlx5_list_entry *entry;
16353 struct mlx5_flow_tbl_resource *tbl_rsc = sub_policy->tbl_rsc;
16354 struct mlx5_flow_dv_matcher matcher = {
16356 .size = sizeof(matcher.mask.buf),
16360 struct mlx5_flow_dv_match_params value = {
16361 .size = sizeof(value.buf),
16363 struct mlx5_flow_cb_ctx ctx = {
16367 struct mlx5_flow_tbl_data_entry *tbl_data;
16368 struct mlx5_priv *priv = dev->data->dev_private;
16369 const uint32_t color_mask = (UINT32_C(1) << MLX5_MTR_COLOR_BITS) - 1;
16371 if (match_src_port && priv->sh->esw_mode) {
16372 if (flow_dv_translate_item_port_id(dev, matcher.mask.buf,
16373 value.buf, item, attr)) {
16374 DRV_LOG(ERR, "Failed to register meter policy%d matcher"
16375 " with port.", priority);
16379 tbl_data = container_of(tbl_rsc, struct mlx5_flow_tbl_data_entry, tbl);
16380 if (priority < RTE_COLOR_RED)
16381 flow_dv_match_meta_reg(matcher.mask.buf, value.buf,
16382 (enum modify_reg)color_reg_c_idx, 0, color_mask);
16383 matcher.priority = priority;
16384 matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
16385 matcher.mask.size);
16386 entry = mlx5_list_register(tbl_data->matchers, &ctx);
16388 DRV_LOG(ERR, "Failed to register meter drop matcher.");
16392 container_of(entry, struct mlx5_flow_dv_matcher, entry);
16397 * Create the policy rules per domain.
16400 * Pointer to Ethernet device.
16401 * @param[in] sub_policy
16402 * Pointer to sub policy table..
16403 * @param[in] egress
16404 * Direction of the table.
16405 * @param[in] transfer
16406 * E-Switch or NIC flow.
16408 * Pointer to policy action list per color.
16411 * 0 on success, -1 otherwise.
16414 __flow_dv_create_domain_policy_rules(struct rte_eth_dev *dev,
16415 struct mlx5_flow_meter_sub_policy *sub_policy,
16416 uint8_t egress, uint8_t transfer, bool match_src_port,
16417 struct mlx5_meter_policy_acts acts[RTE_COLORS])
16419 struct mlx5_priv *priv = dev->data->dev_private;
16420 struct rte_flow_error flow_err;
16421 uint32_t color_reg_c_idx;
16422 struct rte_flow_attr attr = {
16423 .group = MLX5_FLOW_TABLE_LEVEL_POLICY,
16426 .egress = !!egress,
16427 .transfer = !!transfer,
16431 int ret = mlx5_flow_get_reg_id(dev, MLX5_MTR_COLOR, 0, &flow_err);
16432 struct mlx5_sub_policy_color_rule *color_rule;
16434 struct mlx5_sub_policy_color_rule *tmp_rules[RTE_COLORS] = {NULL};
16438 /* Create policy table with POLICY level. */
16439 if (!sub_policy->tbl_rsc)
16440 sub_policy->tbl_rsc = flow_dv_tbl_resource_get(dev,
16441 MLX5_FLOW_TABLE_LEVEL_POLICY,
16442 egress, transfer, false, NULL, 0, 0,
16443 sub_policy->idx, &flow_err);
16444 if (!sub_policy->tbl_rsc) {
16446 "Failed to create meter sub policy table.");
16449 /* Prepare matchers. */
16450 color_reg_c_idx = ret;
16451 for (i = 0; i < RTE_COLORS; i++) {
16452 TAILQ_INIT(&sub_policy->color_rules[i]);
16453 if (!acts[i].actions_n)
16455 color_rule = mlx5_malloc(MLX5_MEM_ZERO,
16456 sizeof(struct mlx5_sub_policy_color_rule),
16459 DRV_LOG(ERR, "No memory to create color rule.");
16462 tmp_rules[i] = color_rule;
16463 TAILQ_INSERT_TAIL(&sub_policy->color_rules[i],
16464 color_rule, next_port);
16465 color_rule->src_port = priv->representor_id;
16468 /* Create matchers for colors. */
16469 svport_match = (i != RTE_COLOR_RED) ? match_src_port : false;
16470 if (__flow_dv_create_policy_matcher(dev, color_reg_c_idx,
16471 MLX5_MTR_POLICY_MATCHER_PRIO, sub_policy,
16472 &attr, svport_match, NULL,
16473 &color_rule->matcher, &flow_err)) {
16474 DRV_LOG(ERR, "Failed to create color%u matcher.", i);
16477 /* Create flow, matching color. */
16478 if (__flow_dv_create_policy_flow(dev,
16479 color_reg_c_idx, (enum rte_color)i,
16480 color_rule->matcher->matcher_object,
16481 acts[i].actions_n, acts[i].dv_actions,
16482 svport_match, NULL, &color_rule->rule,
16484 DRV_LOG(ERR, "Failed to create color%u rule.", i);
16490 /* All the policy rules will be cleared. */
16492 color_rule = tmp_rules[i];
16494 if (color_rule->rule)
16495 mlx5_flow_os_destroy_flow(color_rule->rule);
16496 if (color_rule->matcher) {
16497 struct mlx5_flow_tbl_data_entry *tbl =
16498 container_of(color_rule->matcher->tbl,
16499 typeof(*tbl), tbl);
16500 mlx5_list_unregister(tbl->matchers,
16501 &color_rule->matcher->entry);
16503 TAILQ_REMOVE(&sub_policy->color_rules[i],
16504 color_rule, next_port);
16505 mlx5_free(color_rule);
16512 __flow_dv_create_policy_acts_rules(struct rte_eth_dev *dev,
16513 struct mlx5_flow_meter_policy *mtr_policy,
16514 struct mlx5_flow_meter_sub_policy *sub_policy,
16517 struct mlx5_priv *priv = dev->data->dev_private;
16518 struct mlx5_meter_policy_acts acts[RTE_COLORS];
16519 struct mlx5_flow_dv_tag_resource *tag;
16520 struct mlx5_flow_dv_port_id_action_resource *port_action;
16521 struct mlx5_hrxq *hrxq;
16522 struct mlx5_flow_meter_info *next_fm = NULL;
16523 struct mlx5_flow_meter_policy *next_policy;
16524 struct mlx5_flow_meter_sub_policy *next_sub_policy;
16525 struct mlx5_flow_tbl_data_entry *tbl_data;
16526 struct rte_flow_error error;
16527 uint8_t egress = (domain == MLX5_MTR_DOMAIN_EGRESS) ? 1 : 0;
16528 uint8_t transfer = (domain == MLX5_MTR_DOMAIN_TRANSFER) ? 1 : 0;
16529 bool mtr_first = egress || (transfer && priv->representor_id != UINT16_MAX);
16530 bool match_src_port = false;
16533 /* If RSS or Queue, no previous actions / rules is created. */
16534 for (i = 0; i < RTE_COLORS; i++) {
16535 acts[i].actions_n = 0;
16536 if (i == RTE_COLOR_RED) {
16537 /* Only support drop on red. */
16538 acts[i].dv_actions[0] =
16539 mtr_policy->dr_drop_action[domain];
16540 acts[i].actions_n = 1;
16543 if (i == RTE_COLOR_GREEN &&
16544 mtr_policy->act_cnt[i].fate_action == MLX5_FLOW_FATE_MTR) {
16545 struct rte_flow_attr attr = {
16546 .transfer = transfer
16549 next_fm = mlx5_flow_meter_find(priv,
16550 mtr_policy->act_cnt[i].next_mtr_id,
16554 "Failed to get next hierarchy meter.");
16557 if (mlx5_flow_meter_attach(priv, next_fm,
16559 DRV_LOG(ERR, "%s", error.message);
16563 /* Meter action must be the first for TX. */
16565 acts[i].dv_actions[acts[i].actions_n] =
16566 next_fm->meter_action;
16567 acts[i].actions_n++;
16570 if (mtr_policy->act_cnt[i].rix_mark) {
16571 tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG],
16572 mtr_policy->act_cnt[i].rix_mark);
16574 DRV_LOG(ERR, "Failed to find "
16575 "mark action for policy.");
16578 acts[i].dv_actions[acts[i].actions_n] = tag->action;
16579 acts[i].actions_n++;
16581 if (mtr_policy->act_cnt[i].modify_hdr) {
16582 acts[i].dv_actions[acts[i].actions_n] =
16583 mtr_policy->act_cnt[i].modify_hdr->action;
16584 acts[i].actions_n++;
16586 if (mtr_policy->act_cnt[i].fate_action) {
16587 switch (mtr_policy->act_cnt[i].fate_action) {
16588 case MLX5_FLOW_FATE_PORT_ID:
16589 port_action = mlx5_ipool_get
16590 (priv->sh->ipool[MLX5_IPOOL_PORT_ID],
16591 mtr_policy->act_cnt[i].rix_port_id_action);
16592 if (!port_action) {
16593 DRV_LOG(ERR, "Failed to find "
16594 "port action for policy.");
16597 acts[i].dv_actions[acts[i].actions_n] =
16598 port_action->action;
16599 acts[i].actions_n++;
16600 mtr_policy->dev = dev;
16601 match_src_port = true;
16603 case MLX5_FLOW_FATE_DROP:
16604 case MLX5_FLOW_FATE_JUMP:
16605 acts[i].dv_actions[acts[i].actions_n] =
16606 mtr_policy->act_cnt[i].dr_jump_action[domain];
16607 acts[i].actions_n++;
16609 case MLX5_FLOW_FATE_SHARED_RSS:
16610 case MLX5_FLOW_FATE_QUEUE:
16611 hrxq = mlx5_ipool_get
16612 (priv->sh->ipool[MLX5_IPOOL_HRXQ],
16613 sub_policy->rix_hrxq[i]);
16615 DRV_LOG(ERR, "Failed to find "
16616 "queue action for policy.");
16619 acts[i].dv_actions[acts[i].actions_n] =
16621 acts[i].actions_n++;
16623 case MLX5_FLOW_FATE_MTR:
16626 "No next hierarchy meter.");
16630 acts[i].dv_actions[acts[i].actions_n] =
16631 next_fm->meter_action;
16632 acts[i].actions_n++;
16634 if (mtr_policy->act_cnt[i].next_sub_policy) {
16636 mtr_policy->act_cnt[i].next_sub_policy;
16639 mlx5_flow_meter_policy_find(dev,
16640 next_fm->policy_id, NULL);
16641 MLX5_ASSERT(next_policy);
16643 next_policy->sub_policys[domain][0];
16646 container_of(next_sub_policy->tbl_rsc,
16647 struct mlx5_flow_tbl_data_entry, tbl);
16648 acts[i].dv_actions[acts[i].actions_n++] =
16649 tbl_data->jump.action;
16650 if (mtr_policy->act_cnt[i].modify_hdr)
16651 match_src_port = !!transfer;
16654 /*Queue action do nothing*/
16659 if (__flow_dv_create_domain_policy_rules(dev, sub_policy,
16660 egress, transfer, match_src_port, acts)) {
16662 "Failed to create policy rules per domain.");
16668 mlx5_flow_meter_detach(priv, next_fm);
16673 * Create the policy rules.
16676 * Pointer to Ethernet device.
16677 * @param[in,out] mtr_policy
16678 * Pointer to meter policy table.
16681 * 0 on success, -1 otherwise.
16684 flow_dv_create_policy_rules(struct rte_eth_dev *dev,
16685 struct mlx5_flow_meter_policy *mtr_policy)
16688 uint16_t sub_policy_num;
16690 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16691 sub_policy_num = (mtr_policy->sub_policy_num >>
16692 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i)) &
16693 MLX5_MTR_SUB_POLICY_NUM_MASK;
16694 if (!sub_policy_num)
16696 /* Prepare actions list and create policy rules. */
16697 if (__flow_dv_create_policy_acts_rules(dev, mtr_policy,
16698 mtr_policy->sub_policys[i][0], i)) {
16699 DRV_LOG(ERR, "Failed to create policy action "
16700 "list per domain.");
16708 __flow_dv_create_domain_def_policy(struct rte_eth_dev *dev, uint32_t domain)
16710 struct mlx5_priv *priv = dev->data->dev_private;
16711 struct mlx5_flow_mtr_mng *mtrmng = priv->sh->mtrmng;
16712 struct mlx5_flow_meter_def_policy *def_policy;
16713 struct mlx5_flow_tbl_resource *jump_tbl;
16714 struct mlx5_flow_tbl_data_entry *tbl_data;
16715 uint8_t egress, transfer;
16716 struct rte_flow_error error;
16717 struct mlx5_meter_policy_acts acts[RTE_COLORS];
16720 egress = (domain == MLX5_MTR_DOMAIN_EGRESS) ? 1 : 0;
16721 transfer = (domain == MLX5_MTR_DOMAIN_TRANSFER) ? 1 : 0;
16722 def_policy = mtrmng->def_policy[domain];
16724 def_policy = mlx5_malloc(MLX5_MEM_ZERO,
16725 sizeof(struct mlx5_flow_meter_def_policy),
16726 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
16728 DRV_LOG(ERR, "Failed to alloc default policy table.");
16729 goto def_policy_error;
16731 mtrmng->def_policy[domain] = def_policy;
16732 /* Create the meter suffix table with SUFFIX level. */
16733 jump_tbl = flow_dv_tbl_resource_get(dev,
16734 MLX5_FLOW_TABLE_LEVEL_METER,
16735 egress, transfer, false, NULL, 0,
16736 0, MLX5_MTR_TABLE_ID_SUFFIX, &error);
16739 "Failed to create meter suffix table.");
16740 goto def_policy_error;
16742 def_policy->sub_policy.jump_tbl[RTE_COLOR_GREEN] = jump_tbl;
16743 tbl_data = container_of(jump_tbl,
16744 struct mlx5_flow_tbl_data_entry, tbl);
16745 def_policy->dr_jump_action[RTE_COLOR_GREEN] =
16746 tbl_data->jump.action;
16747 acts[RTE_COLOR_GREEN].dv_actions[0] = tbl_data->jump.action;
16748 acts[RTE_COLOR_GREEN].actions_n = 1;
16750 * YELLOW has the same default policy as GREEN does.
16751 * G & Y share the same table and action. The 2nd time of table
16752 * resource getting is just to update the reference count for
16753 * the releasing stage.
16755 jump_tbl = flow_dv_tbl_resource_get(dev,
16756 MLX5_FLOW_TABLE_LEVEL_METER,
16757 egress, transfer, false, NULL, 0,
16758 0, MLX5_MTR_TABLE_ID_SUFFIX, &error);
16761 "Failed to get meter suffix table.");
16762 goto def_policy_error;
16764 def_policy->sub_policy.jump_tbl[RTE_COLOR_YELLOW] = jump_tbl;
16765 tbl_data = container_of(jump_tbl,
16766 struct mlx5_flow_tbl_data_entry, tbl);
16767 def_policy->dr_jump_action[RTE_COLOR_YELLOW] =
16768 tbl_data->jump.action;
16769 acts[RTE_COLOR_YELLOW].dv_actions[0] = tbl_data->jump.action;
16770 acts[RTE_COLOR_YELLOW].actions_n = 1;
16771 /* Create jump action to the drop table. */
16772 if (!mtrmng->drop_tbl[domain]) {
16773 mtrmng->drop_tbl[domain] = flow_dv_tbl_resource_get
16774 (dev, MLX5_FLOW_TABLE_LEVEL_METER,
16775 egress, transfer, false, NULL, 0,
16776 0, MLX5_MTR_TABLE_ID_DROP, &error);
16777 if (!mtrmng->drop_tbl[domain]) {
16778 DRV_LOG(ERR, "Failed to create meter "
16779 "drop table for default policy.");
16780 goto def_policy_error;
16783 /* all RED: unique Drop table for jump action. */
16784 tbl_data = container_of(mtrmng->drop_tbl[domain],
16785 struct mlx5_flow_tbl_data_entry, tbl);
16786 def_policy->dr_jump_action[RTE_COLOR_RED] =
16787 tbl_data->jump.action;
16788 acts[RTE_COLOR_RED].dv_actions[0] = tbl_data->jump.action;
16789 acts[RTE_COLOR_RED].actions_n = 1;
16790 /* Create default policy rules. */
16791 ret = __flow_dv_create_domain_policy_rules(dev,
16792 &def_policy->sub_policy,
16793 egress, transfer, false, acts);
16795 DRV_LOG(ERR, "Failed to create default policy rules.");
16796 goto def_policy_error;
16801 __flow_dv_destroy_domain_def_policy(dev,
16802 (enum mlx5_meter_domain)domain);
16807 * Create the default policy table set.
16810 * Pointer to Ethernet device.
16812 * 0 on success, -1 otherwise.
16815 flow_dv_create_def_policy(struct rte_eth_dev *dev)
16817 struct mlx5_priv *priv = dev->data->dev_private;
16820 /* Non-termination policy table. */
16821 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16822 if (!priv->sh->config.dv_esw_en &&
16823 i == MLX5_MTR_DOMAIN_TRANSFER)
16825 if (__flow_dv_create_domain_def_policy(dev, i)) {
16826 DRV_LOG(ERR, "Failed to create default policy");
16827 /* Rollback the created default policies for others. */
16828 flow_dv_destroy_def_policy(dev);
16836 * Create the needed meter tables.
16837 * Lock free, (mutex should be acquired by caller).
16840 * Pointer to Ethernet device.
16842 * Meter information table.
16843 * @param[in] mtr_idx
16845 * @param[in] domain_bitmap
16848 * 0 on success, -1 otherwise.
16851 flow_dv_create_mtr_tbls(struct rte_eth_dev *dev,
16852 struct mlx5_flow_meter_info *fm,
16854 uint8_t domain_bitmap)
16856 struct mlx5_priv *priv = dev->data->dev_private;
16857 struct mlx5_flow_mtr_mng *mtrmng = priv->sh->mtrmng;
16858 struct rte_flow_error error;
16859 struct mlx5_flow_tbl_data_entry *tbl_data;
16860 uint8_t egress, transfer;
16861 void *actions[METER_ACTIONS];
16862 int domain, ret, i;
16863 struct mlx5_flow_counter *cnt;
16864 struct mlx5_flow_dv_match_params value = {
16865 .size = sizeof(value.buf),
16867 struct mlx5_flow_dv_match_params matcher_para = {
16868 .size = sizeof(matcher_para.buf),
16870 int mtr_id_reg_c = mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
16872 uint32_t mtr_id_mask = (UINT32_C(1) << mtrmng->max_mtr_bits) - 1;
16873 uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
16874 struct mlx5_list_entry *entry;
16875 struct mlx5_flow_dv_matcher matcher = {
16877 .size = sizeof(matcher.mask.buf),
16880 struct mlx5_flow_dv_matcher *drop_matcher;
16881 struct mlx5_flow_cb_ctx ctx = {
16887 if (!priv->mtr_en || mtr_id_reg_c < 0) {
16888 rte_errno = ENOTSUP;
16891 for (domain = 0; domain < MLX5_MTR_DOMAIN_MAX; domain++) {
16892 if (!(domain_bitmap & (1 << domain)) ||
16893 (mtrmng->def_rule[domain] && !fm->drop_cnt))
16895 egress = (domain == MLX5_MTR_DOMAIN_EGRESS) ? 1 : 0;
16896 transfer = (domain == MLX5_MTR_DOMAIN_TRANSFER) ? 1 : 0;
16897 /* Create the drop table with METER DROP level. */
16898 if (!mtrmng->drop_tbl[domain]) {
16899 mtrmng->drop_tbl[domain] = flow_dv_tbl_resource_get(dev,
16900 MLX5_FLOW_TABLE_LEVEL_METER,
16901 egress, transfer, false, NULL, 0,
16902 0, MLX5_MTR_TABLE_ID_DROP, &error);
16903 if (!mtrmng->drop_tbl[domain]) {
16904 DRV_LOG(ERR, "Failed to create meter drop table.");
16908 /* Create default matcher in drop table. */
16909 matcher.tbl = mtrmng->drop_tbl[domain],
16910 tbl_data = container_of(mtrmng->drop_tbl[domain],
16911 struct mlx5_flow_tbl_data_entry, tbl);
16912 if (!mtrmng->def_matcher[domain]) {
16913 flow_dv_match_meta_reg(matcher.mask.buf, value.buf,
16914 (enum modify_reg)mtr_id_reg_c,
16916 matcher.priority = MLX5_MTRS_DEFAULT_RULE_PRIORITY;
16917 matcher.crc = rte_raw_cksum
16918 ((const void *)matcher.mask.buf,
16919 matcher.mask.size);
16920 entry = mlx5_list_register(tbl_data->matchers, &ctx);
16922 DRV_LOG(ERR, "Failed to register meter "
16923 "drop default matcher.");
16926 mtrmng->def_matcher[domain] = container_of(entry,
16927 struct mlx5_flow_dv_matcher, entry);
16929 /* Create default rule in drop table. */
16930 if (!mtrmng->def_rule[domain]) {
16932 actions[i++] = priv->sh->dr_drop_action;
16933 flow_dv_match_meta_reg(matcher_para.buf, value.buf,
16934 (enum modify_reg)mtr_id_reg_c, 0, 0);
16935 misc_mask = flow_dv_matcher_enable(value.buf);
16936 __flow_dv_adjust_buf_size(&value.size, misc_mask);
16937 ret = mlx5_flow_os_create_flow
16938 (mtrmng->def_matcher[domain]->matcher_object,
16939 (void *)&value, i, actions,
16940 &mtrmng->def_rule[domain]);
16942 DRV_LOG(ERR, "Failed to create meter "
16943 "default drop rule for drop table.");
16949 MLX5_ASSERT(mtrmng->max_mtr_bits);
16950 if (!mtrmng->drop_matcher[domain][mtrmng->max_mtr_bits - 1]) {
16951 /* Create matchers for Drop. */
16952 flow_dv_match_meta_reg(matcher.mask.buf, value.buf,
16953 (enum modify_reg)mtr_id_reg_c, 0,
16954 (mtr_id_mask << mtr_id_offset));
16955 matcher.priority = MLX5_REG_BITS - mtrmng->max_mtr_bits;
16956 matcher.crc = rte_raw_cksum
16957 ((const void *)matcher.mask.buf,
16958 matcher.mask.size);
16959 entry = mlx5_list_register(tbl_data->matchers, &ctx);
16962 "Failed to register meter drop matcher.");
16965 mtrmng->drop_matcher[domain][mtrmng->max_mtr_bits - 1] =
16966 container_of(entry, struct mlx5_flow_dv_matcher,
16970 mtrmng->drop_matcher[domain][mtrmng->max_mtr_bits - 1];
16971 /* Create drop rule, matching meter_id only. */
16972 flow_dv_match_meta_reg(matcher_para.buf, value.buf,
16973 (enum modify_reg)mtr_id_reg_c,
16974 (mtr_idx << mtr_id_offset), UINT32_MAX);
16976 cnt = flow_dv_counter_get_by_idx(dev,
16977 fm->drop_cnt, NULL);
16978 actions[i++] = cnt->action;
16979 actions[i++] = priv->sh->dr_drop_action;
16980 misc_mask = flow_dv_matcher_enable(value.buf);
16981 __flow_dv_adjust_buf_size(&value.size, misc_mask);
16982 ret = mlx5_flow_os_create_flow(drop_matcher->matcher_object,
16983 (void *)&value, i, actions,
16984 &fm->drop_rule[domain]);
16986 DRV_LOG(ERR, "Failed to create meter "
16987 "drop rule for drop table.");
16993 for (i = 0; i < MLX5_MTR_DOMAIN_MAX; i++) {
16994 if (fm->drop_rule[i]) {
16995 claim_zero(mlx5_flow_os_destroy_flow
16996 (fm->drop_rule[i]));
16997 fm->drop_rule[i] = NULL;
17003 static struct mlx5_flow_meter_sub_policy *
17004 __flow_dv_meter_get_rss_sub_policy(struct rte_eth_dev *dev,
17005 struct mlx5_flow_meter_policy *mtr_policy,
17006 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS],
17007 struct mlx5_flow_meter_sub_policy *next_sub_policy,
17010 struct mlx5_priv *priv = dev->data->dev_private;
17011 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
17012 uint32_t sub_policy_idx = 0;
17013 uint32_t hrxq_idx[MLX5_MTR_RTE_COLORS] = {0};
17015 struct mlx5_hrxq *hrxq;
17016 struct mlx5_flow_handle dh;
17017 struct mlx5_meter_policy_action_container *act_cnt;
17018 uint32_t domain = MLX5_MTR_DOMAIN_INGRESS;
17019 uint16_t sub_policy_num;
17020 struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
17023 rte_spinlock_lock(&mtr_policy->sl);
17024 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
17027 hrxq = mlx5_hrxq_get(dev, rss_desc[i]);
17029 rte_spinlock_unlock(&mtr_policy->sl);
17032 hrxq_idx[i] = hrxq->idx;
17034 sub_policy_num = (mtr_policy->sub_policy_num >>
17035 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
17036 MLX5_MTR_SUB_POLICY_NUM_MASK;
17037 for (j = 0; j < sub_policy_num; j++) {
17038 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
17041 mtr_policy->sub_policys[domain][j]->rix_hrxq[i])
17044 if (i >= MLX5_MTR_RTE_COLORS) {
17046 * Found the sub policy table with
17047 * the same queue per color.
17049 rte_spinlock_unlock(&mtr_policy->sl);
17050 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++)
17051 mlx5_hrxq_release(dev, hrxq_idx[i]);
17053 return mtr_policy->sub_policys[domain][j];
17056 /* Create sub policy. */
17057 if (!mtr_policy->sub_policys[domain][0]->rix_hrxq[0]) {
17058 /* Reuse the first pre-allocated sub_policy. */
17059 sub_policy = mtr_policy->sub_policys[domain][0];
17060 sub_policy_idx = sub_policy->idx;
17062 sub_policy = mlx5_ipool_zmalloc
17063 (priv->sh->ipool[MLX5_IPOOL_MTR_POLICY],
17066 sub_policy_idx > MLX5_MAX_SUB_POLICY_TBL_NUM) {
17067 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++)
17068 mlx5_hrxq_release(dev, hrxq_idx[i]);
17069 goto rss_sub_policy_error;
17071 sub_policy->idx = sub_policy_idx;
17072 sub_policy->main_policy = mtr_policy;
17074 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
17077 sub_policy->rix_hrxq[i] = hrxq_idx[i];
17078 if (mtr_policy->is_hierarchy) {
17079 act_cnt = &mtr_policy->act_cnt[i];
17080 act_cnt->next_sub_policy = next_sub_policy;
17081 mlx5_hrxq_release(dev, hrxq_idx[i]);
17084 * Overwrite the last action from
17085 * RSS action to Queue action.
17087 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
17090 DRV_LOG(ERR, "Failed to get policy hrxq");
17091 goto rss_sub_policy_error;
17093 act_cnt = &mtr_policy->act_cnt[i];
17094 if (act_cnt->rix_mark || act_cnt->modify_hdr) {
17095 memset(&dh, 0, sizeof(struct mlx5_flow_handle));
17096 if (act_cnt->rix_mark)
17098 dh.fate_action = MLX5_FLOW_FATE_QUEUE;
17099 dh.rix_hrxq = hrxq_idx[i];
17100 flow_drv_rxq_flags_set(dev, &dh);
17104 if (__flow_dv_create_policy_acts_rules(dev, mtr_policy,
17105 sub_policy, domain)) {
17106 DRV_LOG(ERR, "Failed to create policy "
17107 "rules for ingress domain.");
17108 goto rss_sub_policy_error;
17110 if (sub_policy != mtr_policy->sub_policys[domain][0]) {
17111 i = (mtr_policy->sub_policy_num >>
17112 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
17113 MLX5_MTR_SUB_POLICY_NUM_MASK;
17114 if (i >= MLX5_MTR_RSS_MAX_SUB_POLICY) {
17115 DRV_LOG(ERR, "No free sub-policy slot.");
17116 goto rss_sub_policy_error;
17118 mtr_policy->sub_policys[domain][i] = sub_policy;
17120 mtr_policy->sub_policy_num &= ~(MLX5_MTR_SUB_POLICY_NUM_MASK <<
17121 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain));
17122 mtr_policy->sub_policy_num |=
17123 (i & MLX5_MTR_SUB_POLICY_NUM_MASK) <<
17124 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain);
17126 rte_spinlock_unlock(&mtr_policy->sl);
17129 rss_sub_policy_error:
17131 __flow_dv_destroy_sub_policy_rules(dev, sub_policy);
17132 if (sub_policy != mtr_policy->sub_policys[domain][0]) {
17133 i = (mtr_policy->sub_policy_num >>
17134 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
17135 MLX5_MTR_SUB_POLICY_NUM_MASK;
17136 mtr_policy->sub_policys[domain][i] = NULL;
17137 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MTR_POLICY],
17141 rte_spinlock_unlock(&mtr_policy->sl);
17146 * Find the policy table for prefix table with RSS.
17149 * Pointer to Ethernet device.
17150 * @param[in] mtr_policy
17151 * Pointer to meter policy table.
17152 * @param[in] rss_desc
17153 * Pointer to rss_desc
17155 * Pointer to table set on success, NULL otherwise and rte_errno is set.
17157 static struct mlx5_flow_meter_sub_policy *
17158 flow_dv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
17159 struct mlx5_flow_meter_policy *mtr_policy,
17160 struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
17162 struct mlx5_priv *priv = dev->data->dev_private;
17163 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
17164 struct mlx5_flow_meter_info *next_fm;
17165 struct mlx5_flow_meter_policy *next_policy;
17166 struct mlx5_flow_meter_sub_policy *next_sub_policy = NULL;
17167 struct mlx5_flow_meter_policy *policies[MLX5_MTR_CHAIN_MAX_NUM];
17168 struct mlx5_flow_meter_sub_policy *sub_policies[MLX5_MTR_CHAIN_MAX_NUM];
17169 uint32_t domain = MLX5_MTR_DOMAIN_INGRESS;
17170 bool reuse_sub_policy;
17175 /* Iterate hierarchy to get all policies in this hierarchy. */
17176 policies[i++] = mtr_policy;
17177 if (!mtr_policy->is_hierarchy)
17179 if (i >= MLX5_MTR_CHAIN_MAX_NUM) {
17180 DRV_LOG(ERR, "Exceed max meter number in hierarchy.");
17183 next_fm = mlx5_flow_meter_find(priv,
17184 mtr_policy->act_cnt[RTE_COLOR_GREEN].next_mtr_id, NULL);
17186 DRV_LOG(ERR, "Failed to get next meter in hierarchy.");
17190 mlx5_flow_meter_policy_find(dev, next_fm->policy_id,
17192 MLX5_ASSERT(next_policy);
17193 mtr_policy = next_policy;
17197 * From last policy to the first one in hierarchy,
17198 * create / get the sub policy for each of them.
17200 sub_policy = __flow_dv_meter_get_rss_sub_policy(dev,
17204 &reuse_sub_policy);
17206 DRV_LOG(ERR, "Failed to get the sub policy.");
17209 if (!reuse_sub_policy)
17210 sub_policies[j++] = sub_policy;
17211 next_sub_policy = sub_policy;
17216 uint16_t sub_policy_num;
17218 sub_policy = sub_policies[--j];
17219 mtr_policy = sub_policy->main_policy;
17220 __flow_dv_destroy_sub_policy_rules(dev, sub_policy);
17221 if (sub_policy != mtr_policy->sub_policys[domain][0]) {
17222 sub_policy_num = (mtr_policy->sub_policy_num >>
17223 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
17224 MLX5_MTR_SUB_POLICY_NUM_MASK;
17225 mtr_policy->sub_policys[domain][sub_policy_num - 1] =
17228 mtr_policy->sub_policy_num &=
17229 ~(MLX5_MTR_SUB_POLICY_NUM_MASK <<
17230 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i));
17231 mtr_policy->sub_policy_num |=
17232 (sub_policy_num & MLX5_MTR_SUB_POLICY_NUM_MASK) <<
17233 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * i);
17234 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MTR_POLICY],
17242 * Create the sub policy tag rule for all meters in hierarchy.
17245 * Pointer to Ethernet device.
17247 * Meter information table.
17248 * @param[in] src_port
17249 * The src port this extra rule should use.
17251 * The src port match item.
17252 * @param[out] error
17253 * Perform verbose error reporting if not NULL.
17255 * 0 on success, a negative errno value otherwise and rte_errno is set.
17258 flow_dv_meter_hierarchy_rule_create(struct rte_eth_dev *dev,
17259 struct mlx5_flow_meter_info *fm,
17261 const struct rte_flow_item *item,
17262 struct rte_flow_error *error)
17264 struct mlx5_priv *priv = dev->data->dev_private;
17265 struct mlx5_flow_meter_policy *mtr_policy;
17266 struct mlx5_flow_meter_sub_policy *sub_policy;
17267 struct mlx5_flow_meter_info *next_fm = NULL;
17268 struct mlx5_flow_meter_policy *next_policy;
17269 struct mlx5_flow_meter_sub_policy *next_sub_policy;
17270 struct mlx5_flow_tbl_data_entry *tbl_data;
17271 struct mlx5_sub_policy_color_rule *color_rule;
17272 struct mlx5_meter_policy_acts acts;
17273 uint32_t color_reg_c_idx;
17274 bool mtr_first = (src_port != UINT16_MAX) ? true : false;
17275 struct rte_flow_attr attr = {
17276 .group = MLX5_FLOW_TABLE_LEVEL_POLICY,
17283 uint32_t domain = MLX5_MTR_DOMAIN_TRANSFER;
17286 mtr_policy = mlx5_flow_meter_policy_find(dev, fm->policy_id, NULL);
17287 MLX5_ASSERT(mtr_policy);
17288 if (!mtr_policy->is_hierarchy)
17290 next_fm = mlx5_flow_meter_find(priv,
17291 mtr_policy->act_cnt[RTE_COLOR_GREEN].next_mtr_id, NULL);
17293 return rte_flow_error_set(error, EINVAL,
17294 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
17295 "Failed to find next meter in hierarchy.");
17297 if (!next_fm->drop_cnt)
17299 color_reg_c_idx = mlx5_flow_get_reg_id(dev, MLX5_MTR_COLOR, 0, error);
17300 sub_policy = mtr_policy->sub_policys[domain][0];
17301 for (i = 0; i < RTE_COLORS; i++) {
17302 bool rule_exist = false;
17303 struct mlx5_meter_policy_action_container *act_cnt;
17305 if (i >= RTE_COLOR_YELLOW)
17307 TAILQ_FOREACH(color_rule,
17308 &sub_policy->color_rules[i], next_port)
17309 if (color_rule->src_port == src_port) {
17315 color_rule = mlx5_malloc(MLX5_MEM_ZERO,
17316 sizeof(struct mlx5_sub_policy_color_rule),
17319 return rte_flow_error_set(error, ENOMEM,
17320 RTE_FLOW_ERROR_TYPE_ACTION,
17321 NULL, "No memory to create tag color rule.");
17322 color_rule->src_port = src_port;
17324 next_policy = mlx5_flow_meter_policy_find(dev,
17325 next_fm->policy_id, NULL);
17326 MLX5_ASSERT(next_policy);
17327 next_sub_policy = next_policy->sub_policys[domain][0];
17328 tbl_data = container_of(next_sub_policy->tbl_rsc,
17329 struct mlx5_flow_tbl_data_entry, tbl);
17330 act_cnt = &mtr_policy->act_cnt[i];
17332 acts.dv_actions[0] = next_fm->meter_action;
17333 acts.dv_actions[1] = act_cnt->modify_hdr->action;
17335 acts.dv_actions[0] = act_cnt->modify_hdr->action;
17336 acts.dv_actions[1] = next_fm->meter_action;
17338 acts.dv_actions[2] = tbl_data->jump.action;
17339 acts.actions_n = 3;
17340 if (mlx5_flow_meter_attach(priv, next_fm, &attr, error)) {
17344 if (__flow_dv_create_policy_matcher(dev, color_reg_c_idx,
17345 MLX5_MTR_POLICY_MATCHER_PRIO, sub_policy,
17347 &color_rule->matcher, error)) {
17348 rte_flow_error_set(error, errno,
17349 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
17350 "Failed to create hierarchy meter matcher.");
17353 if (__flow_dv_create_policy_flow(dev, color_reg_c_idx,
17355 color_rule->matcher->matcher_object,
17356 acts.actions_n, acts.dv_actions,
17358 &color_rule->rule, &attr)) {
17359 rte_flow_error_set(error, errno,
17360 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
17361 "Failed to create hierarchy meter rule.");
17364 TAILQ_INSERT_TAIL(&sub_policy->color_rules[i],
17365 color_rule, next_port);
17369 * Recursive call to iterate all meters in hierarchy and
17370 * create needed rules.
17372 return flow_dv_meter_hierarchy_rule_create(dev, next_fm,
17373 src_port, item, error);
17376 if (color_rule->rule)
17377 mlx5_flow_os_destroy_flow(color_rule->rule);
17378 if (color_rule->matcher) {
17379 struct mlx5_flow_tbl_data_entry *tbl =
17380 container_of(color_rule->matcher->tbl,
17381 typeof(*tbl), tbl);
17382 mlx5_list_unregister(tbl->matchers,
17383 &color_rule->matcher->entry);
17385 mlx5_free(color_rule);
17388 mlx5_flow_meter_detach(priv, next_fm);
17393 * Destroy the sub policy table with RX queue.
17396 * Pointer to Ethernet device.
17397 * @param[in] mtr_policy
17398 * Pointer to meter policy table.
17401 flow_dv_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
17402 struct mlx5_flow_meter_policy *mtr_policy)
17404 struct mlx5_priv *priv = dev->data->dev_private;
17405 struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
17406 uint32_t domain = MLX5_MTR_DOMAIN_INGRESS;
17408 uint16_t sub_policy_num, new_policy_num;
17410 rte_spinlock_lock(&mtr_policy->sl);
17411 for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
17412 switch (mtr_policy->act_cnt[i].fate_action) {
17413 case MLX5_FLOW_FATE_SHARED_RSS:
17414 sub_policy_num = (mtr_policy->sub_policy_num >>
17415 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) &
17416 MLX5_MTR_SUB_POLICY_NUM_MASK;
17417 new_policy_num = sub_policy_num;
17418 for (j = 0; j < sub_policy_num; j++) {
17420 mtr_policy->sub_policys[domain][j];
17422 __flow_dv_destroy_sub_policy_rules(dev,
17425 mtr_policy->sub_policys[domain][0]) {
17426 mtr_policy->sub_policys[domain][j] =
17429 (priv->sh->ipool[MLX5_IPOOL_MTR_POLICY],
17435 if (new_policy_num != sub_policy_num) {
17436 mtr_policy->sub_policy_num &=
17437 ~(MLX5_MTR_SUB_POLICY_NUM_MASK <<
17438 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain));
17439 mtr_policy->sub_policy_num |=
17441 MLX5_MTR_SUB_POLICY_NUM_MASK) <<
17442 (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain);
17445 case MLX5_FLOW_FATE_QUEUE:
17446 sub_policy = mtr_policy->sub_policys[domain][0];
17447 __flow_dv_destroy_sub_policy_rules(dev,
17451 /*Other actions without queue and do nothing*/
17455 rte_spinlock_unlock(&mtr_policy->sl);
17458 * Check whether the DR drop action is supported on the root table or not.
17460 * Create a simple flow with DR drop action on root table to validate
17461 * if DR drop action on root table is supported or not.
17464 * Pointer to rte_eth_dev structure.
17467 * 0 on success, a negative errno value otherwise and rte_errno is set.
17470 mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev)
17472 struct mlx5_priv *priv = dev->data->dev_private;
17473 struct mlx5_dev_ctx_shared *sh = priv->sh;
17474 struct mlx5_flow_dv_match_params mask = {
17475 .size = sizeof(mask.buf),
17477 struct mlx5_flow_dv_match_params value = {
17478 .size = sizeof(value.buf),
17480 struct mlx5dv_flow_matcher_attr dv_attr = {
17481 .type = IBV_FLOW_ATTR_NORMAL,
17483 .match_criteria_enable = 0,
17484 .match_mask = (void *)&mask,
17486 struct mlx5_flow_tbl_resource *tbl = NULL;
17487 void *matcher = NULL;
17491 tbl = flow_dv_tbl_resource_get(dev, 0, 0, 0, false, NULL,
17495 dv_attr.match_criteria_enable = flow_dv_matcher_enable(mask.buf);
17496 __flow_dv_adjust_buf_size(&mask.size, dv_attr.match_criteria_enable);
17497 ret = mlx5_flow_os_create_flow_matcher(sh->cdev->ctx, &dv_attr,
17498 tbl->obj, &matcher);
17501 __flow_dv_adjust_buf_size(&value.size, dv_attr.match_criteria_enable);
17502 ret = mlx5_flow_os_create_flow(matcher, (void *)&value, 1,
17503 &sh->dr_drop_action, &flow);
17506 * If DR drop action is not supported on root table, flow create will
17507 * be failed with EOPNOTSUPP or EPROTONOSUPPORT.
17511 (errno == EPROTONOSUPPORT || errno == EOPNOTSUPP))
17512 DRV_LOG(INFO, "DR drop action is not supported in root table.");
17514 DRV_LOG(ERR, "Unexpected error in DR drop action support detection");
17517 claim_zero(mlx5_flow_os_destroy_flow(flow));
17520 claim_zero(mlx5_flow_os_destroy_flow_matcher(matcher));
17522 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
17527 * Validate the batch counter support in root table.
17529 * Create a simple flow with invalid counter and drop action on root table to
17530 * validate if batch counter with offset on root table is supported or not.
17533 * Pointer to rte_eth_dev structure.
17536 * 0 on success, a negative errno value otherwise and rte_errno is set.
17539 mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev)
17541 struct mlx5_priv *priv = dev->data->dev_private;
17542 struct mlx5_dev_ctx_shared *sh = priv->sh;
17543 struct mlx5_flow_dv_match_params mask = {
17544 .size = sizeof(mask.buf),
17546 struct mlx5_flow_dv_match_params value = {
17547 .size = sizeof(value.buf),
17549 struct mlx5dv_flow_matcher_attr dv_attr = {
17550 .type = IBV_FLOW_ATTR_NORMAL | IBV_FLOW_ATTR_FLAGS_EGRESS,
17552 .match_criteria_enable = 0,
17553 .match_mask = (void *)&mask,
17555 void *actions[2] = { 0 };
17556 struct mlx5_flow_tbl_resource *tbl = NULL;
17557 struct mlx5_devx_obj *dcs = NULL;
17558 void *matcher = NULL;
17562 tbl = flow_dv_tbl_resource_get(dev, 0, 1, 0, false, NULL,
17566 dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->cdev->ctx, 0x4);
17569 ret = mlx5_flow_os_create_flow_action_count(dcs->obj, UINT16_MAX,
17573 dv_attr.match_criteria_enable = flow_dv_matcher_enable(mask.buf);
17574 __flow_dv_adjust_buf_size(&mask.size, dv_attr.match_criteria_enable);
17575 ret = mlx5_flow_os_create_flow_matcher(sh->cdev->ctx, &dv_attr,
17576 tbl->obj, &matcher);
17579 __flow_dv_adjust_buf_size(&value.size, dv_attr.match_criteria_enable);
17580 ret = mlx5_flow_os_create_flow(matcher, (void *)&value, 1,
17584 * If batch counter with offset is not supported, the driver will not
17585 * validate the invalid offset value, flow create should success.
17586 * In this case, it means batch counter is not supported in root table.
17588 * Otherwise, if flow create is failed, counter offset is supported.
17591 DRV_LOG(INFO, "Batch counter is not supported in root "
17592 "table. Switch to fallback mode.");
17593 rte_errno = ENOTSUP;
17595 claim_zero(mlx5_flow_os_destroy_flow(flow));
17597 /* Check matcher to make sure validate fail at flow create. */
17598 if (!matcher || (matcher && errno != EINVAL))
17599 DRV_LOG(ERR, "Unexpected error in counter offset "
17600 "support detection");
17604 claim_zero(mlx5_flow_os_destroy_flow_action(actions[0]));
17606 claim_zero(mlx5_flow_os_destroy_flow_matcher(matcher));
17608 flow_dv_tbl_resource_release(MLX5_SH(dev), tbl);
17610 claim_zero(mlx5_devx_cmd_destroy(dcs));
17615 * Query a devx counter.
17618 * Pointer to the Ethernet device structure.
17620 * Index to the flow counter.
17622 * Set to clear the counter statistics.
17624 * The statistics value of packets.
17625 * @param[out] bytes
17626 * The statistics value of bytes.
17629 * 0 on success, otherwise return -1.
17632 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
17633 uint64_t *pkts, uint64_t *bytes, void **action)
17635 struct mlx5_priv *priv = dev->data->dev_private;
17636 struct mlx5_flow_counter *cnt;
17637 uint64_t inn_pkts, inn_bytes;
17640 if (!priv->sh->cdev->config.devx)
17643 ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
17646 cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
17648 *action = cnt->action;
17650 *pkts = inn_pkts - cnt->hits;
17651 *bytes = inn_bytes - cnt->bytes;
17653 cnt->hits = inn_pkts;
17654 cnt->bytes = inn_bytes;
17660 * Get aged-out flows.
17663 * Pointer to the Ethernet device structure.
17664 * @param[in] context
17665 * The address of an array of pointers to the aged-out flows contexts.
17666 * @param[in] nb_contexts
17667 * The length of context array pointers.
17668 * @param[out] error
17669 * Perform verbose error reporting if not NULL. Initialized in case of
17673 * how many contexts get in success, otherwise negative errno value.
17674 * if nb_contexts is 0, return the amount of all aged contexts.
17675 * if nb_contexts is not 0 , return the amount of aged flows reported
17676 * in the context array.
17677 * @note: only stub for now
17680 flow_dv_get_aged_flows(struct rte_eth_dev *dev,
17682 uint32_t nb_contexts,
17683 struct rte_flow_error *error)
17685 struct mlx5_priv *priv = dev->data->dev_private;
17686 struct mlx5_age_info *age_info;
17687 struct mlx5_age_param *age_param;
17688 struct mlx5_flow_counter *counter;
17689 struct mlx5_aso_age_action *act;
17692 if (nb_contexts && !context)
17693 return rte_flow_error_set(error, EINVAL,
17694 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
17695 NULL, "empty context");
17696 age_info = GET_PORT_AGE_INFO(priv);
17697 rte_spinlock_lock(&age_info->aged_sl);
17698 LIST_FOREACH(act, &age_info->aged_aso, next) {
17701 context[nb_flows - 1] =
17702 act->age_params.context;
17703 if (!(--nb_contexts))
17707 TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
17710 age_param = MLX5_CNT_TO_AGE(counter);
17711 context[nb_flows - 1] = age_param->context;
17712 if (!(--nb_contexts))
17716 rte_spinlock_unlock(&age_info->aged_sl);
17717 MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
17722 * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
17725 flow_dv_counter_allocate(struct rte_eth_dev *dev)
17727 return flow_dv_counter_alloc(dev, 0);
17731 * Validate indirect action.
17732 * Dispatcher for action type specific validation.
17735 * Pointer to the Ethernet device structure.
17737 * Indirect action configuration.
17738 * @param[in] action
17739 * The indirect action object to validate.
17740 * @param[out] error
17741 * Perform verbose error reporting if not NULL. Initialized in case of
17745 * 0 on success, otherwise negative errno value.
17748 flow_dv_action_validate(struct rte_eth_dev *dev,
17749 const struct rte_flow_indir_action_conf *conf,
17750 const struct rte_flow_action *action,
17751 struct rte_flow_error *err)
17753 struct mlx5_priv *priv = dev->data->dev_private;
17755 RTE_SET_USED(conf);
17756 switch (action->type) {
17757 case RTE_FLOW_ACTION_TYPE_RSS:
17759 * priv->obj_ops is set according to driver capabilities.
17760 * When DevX capabilities are
17761 * sufficient, it is set to devx_obj_ops.
17762 * Otherwise, it is set to ibv_obj_ops.
17763 * ibv_obj_ops doesn't support ind_table_modify operation.
17764 * In this case the indirect RSS action can't be used.
17766 if (priv->obj_ops.ind_table_modify == NULL)
17767 return rte_flow_error_set
17769 RTE_FLOW_ERROR_TYPE_ACTION,
17771 "Indirect RSS action not supported");
17772 return mlx5_validate_action_rss(dev, action, err);
17773 case RTE_FLOW_ACTION_TYPE_AGE:
17774 if (!priv->sh->aso_age_mng)
17775 return rte_flow_error_set(err, ENOTSUP,
17776 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
17778 "Indirect age action not supported");
17779 return flow_dv_validate_action_age(0, action, dev, err);
17780 case RTE_FLOW_ACTION_TYPE_COUNT:
17781 return flow_dv_validate_action_count(dev, true, 0, NULL, err);
17782 case RTE_FLOW_ACTION_TYPE_CONNTRACK:
17783 if (!priv->sh->ct_aso_en)
17784 return rte_flow_error_set(err, ENOTSUP,
17785 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
17786 "ASO CT is not supported");
17787 return mlx5_validate_action_ct(dev, action->conf, err);
17789 return rte_flow_error_set(err, ENOTSUP,
17790 RTE_FLOW_ERROR_TYPE_ACTION,
17792 "action type not supported");
17797 * Check if the RSS configurations for colors of a meter policy match
17798 * each other, except the queues.
17801 * Pointer to the first RSS flow action.
17803 * Pointer to the second RSS flow action.
17806 * 0 on match, 1 on conflict.
17809 flow_dv_mtr_policy_rss_compare(const struct rte_flow_action_rss *r1,
17810 const struct rte_flow_action_rss *r2)
17812 if (r1 == NULL || r2 == NULL)
17814 if (!(r1->level <= 1 && r2->level <= 1) &&
17815 !(r1->level > 1 && r2->level > 1))
17817 if (r1->types != r2->types &&
17818 !((r1->types == 0 || r1->types == RTE_ETH_RSS_IP) &&
17819 (r2->types == 0 || r2->types == RTE_ETH_RSS_IP)))
17821 if (r1->key || r2->key) {
17822 const void *key1 = r1->key ? r1->key : rss_hash_default_key;
17823 const void *key2 = r2->key ? r2->key : rss_hash_default_key;
17825 if (memcmp(key1, key2, MLX5_RSS_HASH_KEY_LEN))
17832 * Validate the meter hierarchy chain for meter policy.
17835 * Pointer to the Ethernet device structure.
17836 * @param[in] meter_id
17838 * @param[in] action_flags
17839 * Holds the actions detected until now.
17840 * @param[out] is_rss
17842 * @param[out] hierarchy_domain
17843 * The domain bitmap for hierarchy policy.
17844 * @param[out] error
17845 * Perform verbose error reporting if not NULL. Initialized in case of
17849 * 0 on success, otherwise negative errno value with error set.
17852 flow_dv_validate_policy_mtr_hierarchy(struct rte_eth_dev *dev,
17854 uint64_t action_flags,
17856 uint8_t *hierarchy_domain,
17857 struct rte_mtr_error *error)
17859 struct mlx5_priv *priv = dev->data->dev_private;
17860 struct mlx5_flow_meter_info *fm;
17861 struct mlx5_flow_meter_policy *policy;
17864 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
17865 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
17866 return -rte_mtr_error_set(error, EINVAL,
17867 RTE_MTR_ERROR_TYPE_POLICER_ACTION_GREEN,
17869 "Multiple fate actions not supported.");
17870 *hierarchy_domain = 0;
17872 fm = mlx5_flow_meter_find(priv, meter_id, NULL);
17874 return -rte_mtr_error_set(error, EINVAL,
17875 RTE_MTR_ERROR_TYPE_MTR_ID, NULL,
17876 "Meter not found in meter hierarchy.");
17877 if (fm->def_policy)
17878 return -rte_mtr_error_set(error, EINVAL,
17879 RTE_MTR_ERROR_TYPE_MTR_ID, NULL,
17880 "Non termination meter not supported in hierarchy.");
17881 policy = mlx5_flow_meter_policy_find(dev, fm->policy_id, NULL);
17882 MLX5_ASSERT(policy);
17884 * Only inherit the supported domains of the first meter in
17886 * One meter supports at least one domain.
17888 if (!*hierarchy_domain) {
17889 if (policy->transfer)
17890 *hierarchy_domain |=
17891 MLX5_MTR_DOMAIN_TRANSFER_BIT;
17892 if (policy->ingress)
17893 *hierarchy_domain |=
17894 MLX5_MTR_DOMAIN_INGRESS_BIT;
17895 if (policy->egress)
17896 *hierarchy_domain |= MLX5_MTR_DOMAIN_EGRESS_BIT;
17898 if (!policy->is_hierarchy) {
17899 *is_rss = policy->is_rss;
17902 meter_id = policy->act_cnt[RTE_COLOR_GREEN].next_mtr_id;
17903 if (++cnt >= MLX5_MTR_CHAIN_MAX_NUM)
17904 return -rte_mtr_error_set(error, EINVAL,
17905 RTE_MTR_ERROR_TYPE_METER_POLICY, NULL,
17906 "Exceed max hierarchy meter number.");
17912 * Validate meter policy actions.
17913 * Dispatcher for action type specific validation.
17916 * Pointer to the Ethernet device structure.
17917 * @param[in] action
17918 * The meter policy action object to validate.
17920 * Attributes of flow to determine steering domain.
17921 * @param[out] error
17922 * Perform verbose error reporting if not NULL. Initialized in case of
17926 * 0 on success, otherwise negative errno value.
17929 flow_dv_validate_mtr_policy_acts(struct rte_eth_dev *dev,
17930 const struct rte_flow_action *actions[RTE_COLORS],
17931 struct rte_flow_attr *attr,
17933 uint8_t *domain_bitmap,
17934 uint8_t *policy_mode,
17935 struct rte_mtr_error *error)
17937 struct mlx5_priv *priv = dev->data->dev_private;
17938 struct mlx5_sh_config *dev_conf = &priv->sh->config;
17939 const struct rte_flow_action *act;
17940 uint64_t action_flags[RTE_COLORS] = {0};
17943 struct rte_flow_error flow_err;
17944 uint8_t domain_color[RTE_COLORS] = {0};
17945 uint8_t def_domain = MLX5_MTR_ALL_DOMAIN_BIT;
17946 uint8_t hierarchy_domain = 0;
17947 const struct rte_flow_action_meter *mtr;
17948 bool def_green = false;
17949 bool def_yellow = false;
17950 const struct rte_flow_action_rss *rss_color[RTE_COLORS] = {NULL};
17952 if (!dev_conf->dv_esw_en)
17953 def_domain &= ~MLX5_MTR_DOMAIN_TRANSFER_BIT;
17954 *domain_bitmap = def_domain;
17955 /* Red color could only support DROP action. */
17956 if (!actions[RTE_COLOR_RED] ||
17957 actions[RTE_COLOR_RED]->type != RTE_FLOW_ACTION_TYPE_DROP)
17958 return -rte_mtr_error_set(error, ENOTSUP,
17959 RTE_MTR_ERROR_TYPE_METER_POLICY,
17960 NULL, "Red color only supports drop action.");
17962 * Check default policy actions:
17963 * Green / Yellow: no action, Red: drop action
17964 * Either G or Y will trigger default policy actions to be created.
17966 if (!actions[RTE_COLOR_GREEN] ||
17967 actions[RTE_COLOR_GREEN]->type == RTE_FLOW_ACTION_TYPE_END)
17969 if (!actions[RTE_COLOR_YELLOW] ||
17970 actions[RTE_COLOR_YELLOW]->type == RTE_FLOW_ACTION_TYPE_END)
17972 if (def_green && def_yellow) {
17973 *policy_mode = MLX5_MTR_POLICY_MODE_DEF;
17975 } else if (!def_green && def_yellow) {
17976 *policy_mode = MLX5_MTR_POLICY_MODE_OG;
17977 } else if (def_green && !def_yellow) {
17978 *policy_mode = MLX5_MTR_POLICY_MODE_OY;
17980 *policy_mode = MLX5_MTR_POLICY_MODE_ALL;
17982 /* Set to empty string in case of NULL pointer access by user. */
17983 flow_err.message = "";
17984 for (i = 0; i < RTE_COLORS; i++) {
17986 for (action_flags[i] = 0, actions_n = 0;
17987 act && act->type != RTE_FLOW_ACTION_TYPE_END;
17989 if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
17990 return -rte_mtr_error_set(error, ENOTSUP,
17991 RTE_MTR_ERROR_TYPE_METER_POLICY,
17992 NULL, "too many actions");
17993 switch (act->type) {
17994 case RTE_FLOW_ACTION_TYPE_PORT_ID:
17995 case RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT:
17996 if (!dev_conf->dv_esw_en)
17997 return -rte_mtr_error_set(error,
17999 RTE_MTR_ERROR_TYPE_METER_POLICY,
18000 NULL, "PORT action validate check"
18001 " fail for ESW disable");
18002 ret = flow_dv_validate_action_port_id(dev,
18004 act, attr, &flow_err);
18006 return -rte_mtr_error_set(error,
18008 RTE_MTR_ERROR_TYPE_METER_POLICY,
18009 NULL, flow_err.message ?
18011 "PORT action validate check fail");
18013 action_flags[i] |= MLX5_FLOW_ACTION_PORT_ID;
18015 case RTE_FLOW_ACTION_TYPE_MARK:
18016 ret = flow_dv_validate_action_mark(dev, act,
18020 return -rte_mtr_error_set(error,
18022 RTE_MTR_ERROR_TYPE_METER_POLICY,
18023 NULL, flow_err.message ?
18025 "Mark action validate check fail");
18026 if (dev_conf->dv_xmeta_en !=
18027 MLX5_XMETA_MODE_LEGACY)
18028 return -rte_mtr_error_set(error,
18030 RTE_MTR_ERROR_TYPE_METER_POLICY,
18031 NULL, "Extend MARK action is "
18032 "not supported. Please try use "
18033 "default policy for meter.");
18034 action_flags[i] |= MLX5_FLOW_ACTION_MARK;
18037 case RTE_FLOW_ACTION_TYPE_SET_TAG:
18038 ret = flow_dv_validate_action_set_tag(dev,
18039 act, action_flags[i],
18042 return -rte_mtr_error_set(error,
18044 RTE_MTR_ERROR_TYPE_METER_POLICY,
18045 NULL, flow_err.message ?
18047 "Set tag action validate check fail");
18048 action_flags[i] |= MLX5_FLOW_ACTION_SET_TAG;
18051 case RTE_FLOW_ACTION_TYPE_DROP:
18052 ret = mlx5_flow_validate_action_drop
18053 (action_flags[i], attr, &flow_err);
18055 return -rte_mtr_error_set(error,
18057 RTE_MTR_ERROR_TYPE_METER_POLICY,
18058 NULL, flow_err.message ?
18060 "Drop action validate check fail");
18061 action_flags[i] |= MLX5_FLOW_ACTION_DROP;
18064 case RTE_FLOW_ACTION_TYPE_QUEUE:
18066 * Check whether extensive
18067 * metadata feature is engaged.
18069 if (dev_conf->dv_flow_en &&
18070 (dev_conf->dv_xmeta_en !=
18071 MLX5_XMETA_MODE_LEGACY) &&
18072 mlx5_flow_ext_mreg_supported(dev))
18073 return -rte_mtr_error_set(error,
18075 RTE_MTR_ERROR_TYPE_METER_POLICY,
18076 NULL, "Queue action with meta "
18077 "is not supported. Please try use "
18078 "default policy for meter.");
18079 ret = mlx5_flow_validate_action_queue(act,
18080 action_flags[i], dev,
18083 return -rte_mtr_error_set(error,
18085 RTE_MTR_ERROR_TYPE_METER_POLICY,
18086 NULL, flow_err.message ?
18088 "Queue action validate check fail");
18089 action_flags[i] |= MLX5_FLOW_ACTION_QUEUE;
18092 case RTE_FLOW_ACTION_TYPE_RSS:
18093 if (dev_conf->dv_flow_en &&
18094 (dev_conf->dv_xmeta_en !=
18095 MLX5_XMETA_MODE_LEGACY) &&
18096 mlx5_flow_ext_mreg_supported(dev))
18097 return -rte_mtr_error_set(error,
18099 RTE_MTR_ERROR_TYPE_METER_POLICY,
18100 NULL, "RSS action with meta "
18101 "is not supported. Please try use "
18102 "default policy for meter.");
18103 ret = mlx5_validate_action_rss(dev, act,
18106 return -rte_mtr_error_set(error,
18108 RTE_MTR_ERROR_TYPE_METER_POLICY,
18109 NULL, flow_err.message ?
18111 "RSS action validate check fail");
18112 action_flags[i] |= MLX5_FLOW_ACTION_RSS;
18114 /* Either G or Y will set the RSS. */
18115 rss_color[i] = act->conf;
18117 case RTE_FLOW_ACTION_TYPE_JUMP:
18118 ret = flow_dv_validate_action_jump(dev,
18119 NULL, act, action_flags[i],
18120 attr, true, &flow_err);
18122 return -rte_mtr_error_set(error,
18124 RTE_MTR_ERROR_TYPE_METER_POLICY,
18125 NULL, flow_err.message ?
18127 "Jump action validate check fail");
18129 action_flags[i] |= MLX5_FLOW_ACTION_JUMP;
18132 * Only the last meter in the hierarchy will support
18133 * the YELLOW color steering. Then in the meter policy
18134 * actions list, there should be no other meter inside.
18136 case RTE_FLOW_ACTION_TYPE_METER:
18137 if (i != RTE_COLOR_GREEN)
18138 return -rte_mtr_error_set(error,
18140 RTE_MTR_ERROR_TYPE_METER_POLICY,
18142 "Meter hierarchy only supports GREEN color.");
18143 if (*policy_mode != MLX5_MTR_POLICY_MODE_OG)
18144 return -rte_mtr_error_set(error,
18146 RTE_MTR_ERROR_TYPE_METER_POLICY,
18148 "No yellow policy should be provided in meter hierarchy.");
18150 ret = flow_dv_validate_policy_mtr_hierarchy(dev,
18160 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY;
18163 return -rte_mtr_error_set(error, ENOTSUP,
18164 RTE_MTR_ERROR_TYPE_METER_POLICY,
18166 "Doesn't support optional action");
18169 if (action_flags[i] & MLX5_FLOW_ACTION_PORT_ID) {
18170 domain_color[i] = MLX5_MTR_DOMAIN_TRANSFER_BIT;
18171 } else if ((action_flags[i] &
18172 (MLX5_FLOW_ACTION_RSS | MLX5_FLOW_ACTION_QUEUE)) ||
18173 (action_flags[i] & MLX5_FLOW_ACTION_MARK)) {
18175 * Only support MLX5_XMETA_MODE_LEGACY
18176 * so MARK action is only in ingress domain.
18178 domain_color[i] = MLX5_MTR_DOMAIN_INGRESS_BIT;
18180 domain_color[i] = def_domain;
18181 if (action_flags[i] &&
18182 !(action_flags[i] & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
18184 ~MLX5_MTR_DOMAIN_TRANSFER_BIT;
18186 if (action_flags[i] &
18187 MLX5_FLOW_ACTION_METER_WITH_TERMINATED_POLICY)
18188 domain_color[i] &= hierarchy_domain;
18190 * Non-termination actions only support NIC Tx domain.
18191 * The adjustion should be skipped when there is no
18192 * action or only END is provided. The default domains
18193 * bit-mask is set to find the MIN intersection.
18194 * The action flags checking should also be skipped.
18196 if ((def_green && i == RTE_COLOR_GREEN) ||
18197 (def_yellow && i == RTE_COLOR_YELLOW))
18200 * Validate the drop action mutual exclusion
18201 * with other actions. Drop action is mutually-exclusive
18202 * with any other action, except for Count action.
18204 if ((action_flags[i] & MLX5_FLOW_ACTION_DROP) &&
18205 (action_flags[i] & ~MLX5_FLOW_ACTION_DROP)) {
18206 return -rte_mtr_error_set(error, ENOTSUP,
18207 RTE_MTR_ERROR_TYPE_METER_POLICY,
18208 NULL, "Drop action is mutually-exclusive "
18209 "with any other action");
18211 /* Eswitch has few restrictions on using items and actions */
18212 if (domain_color[i] & MLX5_MTR_DOMAIN_TRANSFER_BIT) {
18213 if (!mlx5_flow_ext_mreg_supported(dev) &&
18214 action_flags[i] & MLX5_FLOW_ACTION_MARK)
18215 return -rte_mtr_error_set(error, ENOTSUP,
18216 RTE_MTR_ERROR_TYPE_METER_POLICY,
18217 NULL, "unsupported action MARK");
18218 if (action_flags[i] & MLX5_FLOW_ACTION_QUEUE)
18219 return -rte_mtr_error_set(error, ENOTSUP,
18220 RTE_MTR_ERROR_TYPE_METER_POLICY,
18221 NULL, "unsupported action QUEUE");
18222 if (action_flags[i] & MLX5_FLOW_ACTION_RSS)
18223 return -rte_mtr_error_set(error, ENOTSUP,
18224 RTE_MTR_ERROR_TYPE_METER_POLICY,
18225 NULL, "unsupported action RSS");
18226 if (!(action_flags[i] & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
18227 return -rte_mtr_error_set(error, ENOTSUP,
18228 RTE_MTR_ERROR_TYPE_METER_POLICY,
18229 NULL, "no fate action is found");
18231 if (!(action_flags[i] & MLX5_FLOW_FATE_ACTIONS) &&
18232 (domain_color[i] & MLX5_MTR_DOMAIN_INGRESS_BIT)) {
18233 if ((domain_color[i] &
18234 MLX5_MTR_DOMAIN_EGRESS_BIT))
18236 MLX5_MTR_DOMAIN_EGRESS_BIT;
18238 return -rte_mtr_error_set(error,
18240 RTE_MTR_ERROR_TYPE_METER_POLICY,
18242 "no fate action is found");
18246 /* If both colors have RSS, the attributes should be the same. */
18247 if (flow_dv_mtr_policy_rss_compare(rss_color[RTE_COLOR_GREEN],
18248 rss_color[RTE_COLOR_YELLOW]))
18249 return -rte_mtr_error_set(error, EINVAL,
18250 RTE_MTR_ERROR_TYPE_METER_POLICY,
18251 NULL, "policy RSS attr conflict");
18252 if (rss_color[RTE_COLOR_GREEN] || rss_color[RTE_COLOR_YELLOW])
18254 /* "domain_color[C]" is non-zero for each color, default is ALL. */
18255 if (!def_green && !def_yellow &&
18256 domain_color[RTE_COLOR_GREEN] != domain_color[RTE_COLOR_YELLOW] &&
18257 !(action_flags[RTE_COLOR_GREEN] & MLX5_FLOW_ACTION_DROP) &&
18258 !(action_flags[RTE_COLOR_YELLOW] & MLX5_FLOW_ACTION_DROP))
18259 return -rte_mtr_error_set(error, EINVAL,
18260 RTE_MTR_ERROR_TYPE_METER_POLICY,
18261 NULL, "policy domains conflict");
18263 * At least one color policy is listed in the actions, the domains
18264 * to be supported should be the intersection.
18266 *domain_bitmap = domain_color[RTE_COLOR_GREEN] &
18267 domain_color[RTE_COLOR_YELLOW];
18272 flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
18274 struct mlx5_priv *priv = dev->data->dev_private;
18277 if ((domains & MLX5_DOMAIN_BIT_NIC_RX) && priv->sh->rx_domain != NULL) {
18278 ret = mlx5_os_flow_dr_sync_domain(priv->sh->rx_domain,
18283 if ((domains & MLX5_DOMAIN_BIT_NIC_TX) && priv->sh->tx_domain != NULL) {
18284 ret = mlx5_os_flow_dr_sync_domain(priv->sh->tx_domain, flags);
18288 if ((domains & MLX5_DOMAIN_BIT_FDB) && priv->sh->fdb_domain != NULL) {
18289 ret = mlx5_os_flow_dr_sync_domain(priv->sh->fdb_domain, flags);
18297 * Discover the number of available flow priorities
18298 * by trying to create a flow with the highest priority value
18299 * for each possible number.
18304 * List of possible number of available priorities.
18305 * @param[in] vprio_n
18306 * Size of @p vprio array.
18308 * On success, number of available flow priorities.
18309 * On failure, a negative errno-style code and rte_errno is set.
18312 flow_dv_discover_priorities(struct rte_eth_dev *dev,
18313 const uint16_t *vprio, int vprio_n)
18315 struct mlx5_priv *priv = dev->data->dev_private;
18316 struct mlx5_indexed_pool *pool = priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW];
18317 struct rte_flow_item_eth eth;
18318 struct rte_flow_item item = {
18319 .type = RTE_FLOW_ITEM_TYPE_ETH,
18323 struct mlx5_flow_dv_matcher matcher = {
18325 .size = sizeof(matcher.mask.buf),
18328 union mlx5_flow_tbl_key tbl_key;
18329 struct mlx5_flow flow;
18331 struct rte_flow_error error;
18333 int i, err, ret = -ENOTSUP;
18336 * Prepare a flow with a catch-all pattern and a drop action.
18337 * Use drop queue, because shared drop action may be unavailable.
18339 action = priv->drop_queue.hrxq->action;
18340 if (action == NULL) {
18341 DRV_LOG(ERR, "Priority discovery requires a drop action");
18342 rte_errno = ENOTSUP;
18345 memset(&flow, 0, sizeof(flow));
18346 flow.handle = mlx5_ipool_zmalloc(pool, &flow.handle_idx);
18347 if (flow.handle == NULL) {
18348 DRV_LOG(ERR, "Cannot create flow handle");
18349 rte_errno = ENOMEM;
18352 flow.ingress = true;
18353 flow.dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
18354 flow.dv.actions[0] = action;
18355 flow.dv.actions_n = 1;
18356 memset(ð, 0, sizeof(eth));
18357 flow_dv_translate_item_eth(matcher.mask.buf, flow.dv.value.buf,
18358 &item, /* inner */ false, /* group */ 0);
18359 matcher.crc = rte_raw_cksum(matcher.mask.buf, matcher.mask.size);
18360 for (i = 0; i < vprio_n; i++) {
18361 /* Configure the next proposed maximum priority. */
18362 matcher.priority = vprio[i] - 1;
18363 memset(&tbl_key, 0, sizeof(tbl_key));
18364 err = flow_dv_matcher_register(dev, &matcher, &tbl_key, &flow,
18369 /* This action is pure SW and must always succeed. */
18370 DRV_LOG(ERR, "Cannot register matcher");
18374 /* Try to apply the flow to HW. */
18375 misc_mask = flow_dv_matcher_enable(flow.dv.value.buf);
18376 __flow_dv_adjust_buf_size(&flow.dv.value.size, misc_mask);
18377 err = mlx5_flow_os_create_flow
18378 (flow.handle->dvh.matcher->matcher_object,
18379 (void *)&flow.dv.value, flow.dv.actions_n,
18380 flow.dv.actions, &flow.handle->drv_flow);
18382 claim_zero(mlx5_flow_os_destroy_flow
18383 (flow.handle->drv_flow));
18384 flow.handle->drv_flow = NULL;
18386 claim_zero(flow_dv_matcher_release(dev, flow.handle));
18391 mlx5_ipool_free(pool, flow.handle_idx);
18392 /* Set rte_errno if no expected priority value matched. */
18398 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
18399 .validate = flow_dv_validate,
18400 .prepare = flow_dv_prepare,
18401 .translate = flow_dv_translate,
18402 .apply = flow_dv_apply,
18403 .remove = flow_dv_remove,
18404 .destroy = flow_dv_destroy,
18405 .query = flow_dv_query,
18406 .create_mtr_tbls = flow_dv_create_mtr_tbls,
18407 .destroy_mtr_tbls = flow_dv_destroy_mtr_tbls,
18408 .destroy_mtr_drop_tbls = flow_dv_destroy_mtr_drop_tbls,
18409 .create_meter = flow_dv_mtr_alloc,
18410 .free_meter = flow_dv_aso_mtr_release_to_pool,
18411 .validate_mtr_acts = flow_dv_validate_mtr_policy_acts,
18412 .create_mtr_acts = flow_dv_create_mtr_policy_acts,
18413 .destroy_mtr_acts = flow_dv_destroy_mtr_policy_acts,
18414 .create_policy_rules = flow_dv_create_policy_rules,
18415 .destroy_policy_rules = flow_dv_destroy_policy_rules,
18416 .create_def_policy = flow_dv_create_def_policy,
18417 .destroy_def_policy = flow_dv_destroy_def_policy,
18418 .meter_sub_policy_rss_prepare = flow_dv_meter_sub_policy_rss_prepare,
18419 .meter_hierarchy_rule_create = flow_dv_meter_hierarchy_rule_create,
18420 .destroy_sub_policy_with_rxq = flow_dv_destroy_sub_policy_with_rxq,
18421 .counter_alloc = flow_dv_counter_allocate,
18422 .counter_free = flow_dv_counter_free,
18423 .counter_query = flow_dv_counter_query,
18424 .get_aged_flows = flow_dv_get_aged_flows,
18425 .action_validate = flow_dv_action_validate,
18426 .action_create = flow_dv_action_create,
18427 .action_destroy = flow_dv_action_destroy,
18428 .action_update = flow_dv_action_update,
18429 .action_query = flow_dv_action_query,
18430 .sync_domain = flow_dv_sync_domain,
18431 .discover_priorities = flow_dv_discover_priorities,
18432 .item_create = flow_dv_item_create,
18433 .item_release = flow_dv_item_release,
18436 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */