1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
6 #include <netinet/in.h>
13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
15 #pragma GCC diagnostic ignored "-Wpedantic"
17 #include <infiniband/verbs.h>
19 #pragma GCC diagnostic error "-Wpedantic"
22 #include <rte_common.h>
23 #include <rte_ether.h>
24 #include <rte_ethdev_driver.h>
26 #include <rte_flow_driver.h>
27 #include <rte_malloc.h>
31 #include "mlx5_defs.h"
32 #include "mlx5_flow.h"
33 #include "mlx5_glue.h"
35 #include "mlx5_rxtx.h"
37 /* Dev ops structure defined in mlx5.c */
38 extern const struct eth_dev_ops mlx5_dev_ops;
39 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
41 /** Device flow drivers. */
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
45 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
47 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
49 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
50 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
51 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
52 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
54 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
55 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
60 MLX5_EXPANSION_ROOT_OUTER,
61 MLX5_EXPANSION_ROOT_ETH_VLAN,
62 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
63 MLX5_EXPANSION_OUTER_ETH,
64 MLX5_EXPANSION_OUTER_ETH_VLAN,
65 MLX5_EXPANSION_OUTER_VLAN,
66 MLX5_EXPANSION_OUTER_IPV4,
67 MLX5_EXPANSION_OUTER_IPV4_UDP,
68 MLX5_EXPANSION_OUTER_IPV4_TCP,
69 MLX5_EXPANSION_OUTER_IPV6,
70 MLX5_EXPANSION_OUTER_IPV6_UDP,
71 MLX5_EXPANSION_OUTER_IPV6_TCP,
73 MLX5_EXPANSION_VXLAN_GPE,
77 MLX5_EXPANSION_ETH_VLAN,
80 MLX5_EXPANSION_IPV4_UDP,
81 MLX5_EXPANSION_IPV4_TCP,
83 MLX5_EXPANSION_IPV6_UDP,
84 MLX5_EXPANSION_IPV6_TCP,
87 /** Supported expansion of items. */
88 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
89 [MLX5_EXPANSION_ROOT] = {
90 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
93 .type = RTE_FLOW_ITEM_TYPE_END,
95 [MLX5_EXPANSION_ROOT_OUTER] = {
96 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
97 MLX5_EXPANSION_OUTER_IPV4,
98 MLX5_EXPANSION_OUTER_IPV6),
99 .type = RTE_FLOW_ITEM_TYPE_END,
101 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
102 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
103 .type = RTE_FLOW_ITEM_TYPE_END,
105 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
106 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
107 .type = RTE_FLOW_ITEM_TYPE_END,
109 [MLX5_EXPANSION_OUTER_ETH] = {
110 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
111 MLX5_EXPANSION_OUTER_IPV6,
112 MLX5_EXPANSION_MPLS),
113 .type = RTE_FLOW_ITEM_TYPE_ETH,
116 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
117 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
118 .type = RTE_FLOW_ITEM_TYPE_ETH,
121 [MLX5_EXPANSION_OUTER_VLAN] = {
122 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
123 MLX5_EXPANSION_OUTER_IPV6),
124 .type = RTE_FLOW_ITEM_TYPE_VLAN,
126 [MLX5_EXPANSION_OUTER_IPV4] = {
127 .next = RTE_FLOW_EXPAND_RSS_NEXT
128 (MLX5_EXPANSION_OUTER_IPV4_UDP,
129 MLX5_EXPANSION_OUTER_IPV4_TCP,
132 MLX5_EXPANSION_IPV6),
133 .type = RTE_FLOW_ITEM_TYPE_IPV4,
134 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
135 ETH_RSS_NONFRAG_IPV4_OTHER,
137 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
138 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
139 MLX5_EXPANSION_VXLAN_GPE),
140 .type = RTE_FLOW_ITEM_TYPE_UDP,
141 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
143 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
144 .type = RTE_FLOW_ITEM_TYPE_TCP,
145 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
147 [MLX5_EXPANSION_OUTER_IPV6] = {
148 .next = RTE_FLOW_EXPAND_RSS_NEXT
149 (MLX5_EXPANSION_OUTER_IPV6_UDP,
150 MLX5_EXPANSION_OUTER_IPV6_TCP,
152 MLX5_EXPANSION_IPV6),
153 .type = RTE_FLOW_ITEM_TYPE_IPV6,
154 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
155 ETH_RSS_NONFRAG_IPV6_OTHER,
157 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
158 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
159 MLX5_EXPANSION_VXLAN_GPE),
160 .type = RTE_FLOW_ITEM_TYPE_UDP,
161 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
163 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
164 .type = RTE_FLOW_ITEM_TYPE_TCP,
165 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
167 [MLX5_EXPANSION_VXLAN] = {
168 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
169 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
171 [MLX5_EXPANSION_VXLAN_GPE] = {
172 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
174 MLX5_EXPANSION_IPV6),
175 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
177 [MLX5_EXPANSION_GRE] = {
178 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
179 .type = RTE_FLOW_ITEM_TYPE_GRE,
181 [MLX5_EXPANSION_MPLS] = {
182 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
183 MLX5_EXPANSION_IPV6),
184 .type = RTE_FLOW_ITEM_TYPE_MPLS,
186 [MLX5_EXPANSION_ETH] = {
187 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 MLX5_EXPANSION_IPV6),
189 .type = RTE_FLOW_ITEM_TYPE_ETH,
191 [MLX5_EXPANSION_ETH_VLAN] = {
192 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
193 .type = RTE_FLOW_ITEM_TYPE_ETH,
195 [MLX5_EXPANSION_VLAN] = {
196 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
197 MLX5_EXPANSION_IPV6),
198 .type = RTE_FLOW_ITEM_TYPE_VLAN,
200 [MLX5_EXPANSION_IPV4] = {
201 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
202 MLX5_EXPANSION_IPV4_TCP),
203 .type = RTE_FLOW_ITEM_TYPE_IPV4,
204 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
205 ETH_RSS_NONFRAG_IPV4_OTHER,
207 [MLX5_EXPANSION_IPV4_UDP] = {
208 .type = RTE_FLOW_ITEM_TYPE_UDP,
209 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
211 [MLX5_EXPANSION_IPV4_TCP] = {
212 .type = RTE_FLOW_ITEM_TYPE_TCP,
213 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
215 [MLX5_EXPANSION_IPV6] = {
216 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
217 MLX5_EXPANSION_IPV6_TCP),
218 .type = RTE_FLOW_ITEM_TYPE_IPV6,
219 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
220 ETH_RSS_NONFRAG_IPV6_OTHER,
222 [MLX5_EXPANSION_IPV6_UDP] = {
223 .type = RTE_FLOW_ITEM_TYPE_UDP,
224 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
226 [MLX5_EXPANSION_IPV6_TCP] = {
227 .type = RTE_FLOW_ITEM_TYPE_TCP,
228 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
232 static const struct rte_flow_ops mlx5_flow_ops = {
233 .validate = mlx5_flow_validate,
234 .create = mlx5_flow_create,
235 .destroy = mlx5_flow_destroy,
236 .flush = mlx5_flow_flush,
237 .isolate = mlx5_flow_isolate,
238 .query = mlx5_flow_query,
241 /* Convert FDIR request to Generic flow. */
243 struct rte_flow_attr attr;
244 struct rte_flow_item items[4];
245 struct rte_flow_item_eth l2;
246 struct rte_flow_item_eth l2_mask;
248 struct rte_flow_item_ipv4 ipv4;
249 struct rte_flow_item_ipv6 ipv6;
252 struct rte_flow_item_ipv4 ipv4;
253 struct rte_flow_item_ipv6 ipv6;
256 struct rte_flow_item_udp udp;
257 struct rte_flow_item_tcp tcp;
260 struct rte_flow_item_udp udp;
261 struct rte_flow_item_tcp tcp;
263 struct rte_flow_action actions[2];
264 struct rte_flow_action_queue queue;
267 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
268 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
269 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
272 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
273 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
274 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
275 { 9, 10, 11 }, { 12, 13, 14 },
278 /* Tunnel information. */
279 struct mlx5_flow_tunnel_info {
280 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
281 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
284 static struct mlx5_flow_tunnel_info tunnels_info[] = {
286 .tunnel = MLX5_FLOW_LAYER_VXLAN,
287 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
290 .tunnel = MLX5_FLOW_LAYER_GENEVE,
291 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
294 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
295 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
298 .tunnel = MLX5_FLOW_LAYER_GRE,
299 .ptype = RTE_PTYPE_TUNNEL_GRE,
302 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
303 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
306 .tunnel = MLX5_FLOW_LAYER_MPLS,
307 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
310 .tunnel = MLX5_FLOW_LAYER_NVGRE,
311 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
314 .tunnel = MLX5_FLOW_LAYER_IPIP,
315 .ptype = RTE_PTYPE_TUNNEL_IP,
318 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
319 .ptype = RTE_PTYPE_TUNNEL_IP,
323 enum mlx5_feature_name {
330 * Translate tag ID to register.
333 * Pointer to the Ethernet device structure.
335 * The feature that request the register.
337 * The request register ID.
339 * Error description in case of any.
342 * The request register on success, a negative errno
343 * value otherwise and rte_errno is set.
346 static enum modify_reg flow_get_reg_id(struct rte_eth_dev *dev,
347 enum mlx5_feature_name feature,
349 struct rte_flow_error *error)
351 static enum modify_reg id2reg[] = {
360 case MLX5_HAIRPIN_RX:
362 case MLX5_HAIRPIN_TX:
364 case MLX5_APPLICATION:
366 return rte_flow_error_set(error, EINVAL,
367 RTE_FLOW_ERROR_TYPE_ITEM,
368 NULL, "invalid tag id");
371 return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM,
372 NULL, "invalid feature name");
376 * Discover the maximum number of priority available.
379 * Pointer to the Ethernet device structure.
382 * number of supported flow priority on success, a negative errno
383 * value otherwise and rte_errno is set.
386 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
388 struct mlx5_priv *priv = dev->data->dev_private;
390 struct ibv_flow_attr attr;
391 struct ibv_flow_spec_eth eth;
392 struct ibv_flow_spec_action_drop drop;
396 .port = (uint8_t)priv->ibv_port,
399 .type = IBV_FLOW_SPEC_ETH,
400 .size = sizeof(struct ibv_flow_spec_eth),
403 .size = sizeof(struct ibv_flow_spec_action_drop),
404 .type = IBV_FLOW_SPEC_ACTION_DROP,
407 struct ibv_flow *flow;
408 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
409 uint16_t vprio[] = { 8, 16 };
417 for (i = 0; i != RTE_DIM(vprio); i++) {
418 flow_attr.attr.priority = vprio[i] - 1;
419 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
422 claim_zero(mlx5_glue->destroy_flow(flow));
425 mlx5_hrxq_drop_release(dev);
428 priority = RTE_DIM(priority_map_3);
431 priority = RTE_DIM(priority_map_5);
436 "port %u verbs maximum priority: %d expected 8/16",
437 dev->data->port_id, priority);
440 DRV_LOG(INFO, "port %u flow maximum priority: %d",
441 dev->data->port_id, priority);
446 * Adjust flow priority based on the highest layer and the request priority.
449 * Pointer to the Ethernet device structure.
450 * @param[in] priority
451 * The rule base priority.
452 * @param[in] subpriority
453 * The priority based on the items.
458 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
459 uint32_t subpriority)
462 struct mlx5_priv *priv = dev->data->dev_private;
464 switch (priv->config.flow_prio) {
465 case RTE_DIM(priority_map_3):
466 res = priority_map_3[priority][subpriority];
468 case RTE_DIM(priority_map_5):
469 res = priority_map_5[priority][subpriority];
476 * Verify the @p item specifications (spec, last, mask) are compatible with the
480 * Item specification.
482 * @p item->mask or flow default bit-masks.
483 * @param[in] nic_mask
484 * Bit-masks covering supported fields by the NIC to compare with user mask.
486 * Bit-masks size in bytes.
488 * Pointer to error structure.
491 * 0 on success, a negative errno value otherwise and rte_errno is set.
494 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
496 const uint8_t *nic_mask,
498 struct rte_flow_error *error)
503 for (i = 0; i < size; ++i)
504 if ((nic_mask[i] | mask[i]) != nic_mask[i])
505 return rte_flow_error_set(error, ENOTSUP,
506 RTE_FLOW_ERROR_TYPE_ITEM,
508 "mask enables non supported"
510 if (!item->spec && (item->mask || item->last))
511 return rte_flow_error_set(error, EINVAL,
512 RTE_FLOW_ERROR_TYPE_ITEM, item,
513 "mask/last without a spec is not"
515 if (item->spec && item->last) {
521 for (i = 0; i < size; ++i) {
522 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
523 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
525 ret = memcmp(spec, last, size);
527 return rte_flow_error_set(error, EINVAL,
528 RTE_FLOW_ERROR_TYPE_ITEM,
530 "range is not valid");
536 * Adjust the hash fields according to the @p flow information.
538 * @param[in] dev_flow.
539 * Pointer to the mlx5_flow.
541 * 1 when the hash field is for a tunnel item.
542 * @param[in] layer_types
544 * @param[in] hash_fields
548 * The hash fields that should be used.
551 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
552 int tunnel __rte_unused, uint64_t layer_types,
553 uint64_t hash_fields)
555 struct rte_flow *flow = dev_flow->flow;
556 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
557 int rss_request_inner = flow->rss.level >= 2;
559 /* Check RSS hash level for tunnel. */
560 if (tunnel && rss_request_inner)
561 hash_fields |= IBV_RX_HASH_INNER;
562 else if (tunnel || rss_request_inner)
565 /* Check if requested layer matches RSS hash fields. */
566 if (!(flow->rss.types & layer_types))
572 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
573 * if several tunnel rules are used on this queue, the tunnel ptype will be
577 * Rx queue to update.
580 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
583 uint32_t tunnel_ptype = 0;
585 /* Look up for the ptype to use. */
586 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
587 if (!rxq_ctrl->flow_tunnels_n[i])
590 tunnel_ptype = tunnels_info[i].ptype;
596 rxq_ctrl->rxq.tunnel = tunnel_ptype;
600 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
604 * Pointer to the Ethernet device structure.
605 * @param[in] dev_flow
606 * Pointer to device flow structure.
609 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
611 struct mlx5_priv *priv = dev->data->dev_private;
612 struct rte_flow *flow = dev_flow->flow;
613 const int mark = !!(dev_flow->actions &
614 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
615 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
618 for (i = 0; i != flow->rss.queue_num; ++i) {
619 int idx = (*flow->rss.queue)[i];
620 struct mlx5_rxq_ctrl *rxq_ctrl =
621 container_of((*priv->rxqs)[idx],
622 struct mlx5_rxq_ctrl, rxq);
625 rxq_ctrl->rxq.mark = 1;
626 rxq_ctrl->flow_mark_n++;
631 /* Increase the counter matching the flow. */
632 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
633 if ((tunnels_info[j].tunnel &
635 tunnels_info[j].tunnel) {
636 rxq_ctrl->flow_tunnels_n[j]++;
640 flow_rxq_tunnel_ptype_update(rxq_ctrl);
646 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
649 * Pointer to the Ethernet device structure.
651 * Pointer to flow structure.
654 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
656 struct mlx5_flow *dev_flow;
658 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
659 flow_drv_rxq_flags_set(dev, dev_flow);
663 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
664 * device flow if no other flow uses it with the same kind of request.
667 * Pointer to Ethernet device.
668 * @param[in] dev_flow
669 * Pointer to the device flow.
672 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
674 struct mlx5_priv *priv = dev->data->dev_private;
675 struct rte_flow *flow = dev_flow->flow;
676 const int mark = !!(dev_flow->actions &
677 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
678 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
681 assert(dev->data->dev_started);
682 for (i = 0; i != flow->rss.queue_num; ++i) {
683 int idx = (*flow->rss.queue)[i];
684 struct mlx5_rxq_ctrl *rxq_ctrl =
685 container_of((*priv->rxqs)[idx],
686 struct mlx5_rxq_ctrl, rxq);
689 rxq_ctrl->flow_mark_n--;
690 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
695 /* Decrease the counter matching the flow. */
696 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
697 if ((tunnels_info[j].tunnel &
699 tunnels_info[j].tunnel) {
700 rxq_ctrl->flow_tunnels_n[j]--;
704 flow_rxq_tunnel_ptype_update(rxq_ctrl);
710 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
711 * @p flow if no other flow uses it with the same kind of request.
714 * Pointer to Ethernet device.
716 * Pointer to the flow.
719 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
721 struct mlx5_flow *dev_flow;
723 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
724 flow_drv_rxq_flags_trim(dev, dev_flow);
728 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
731 * Pointer to Ethernet device.
734 flow_rxq_flags_clear(struct rte_eth_dev *dev)
736 struct mlx5_priv *priv = dev->data->dev_private;
739 for (i = 0; i != priv->rxqs_n; ++i) {
740 struct mlx5_rxq_ctrl *rxq_ctrl;
743 if (!(*priv->rxqs)[i])
745 rxq_ctrl = container_of((*priv->rxqs)[i],
746 struct mlx5_rxq_ctrl, rxq);
747 rxq_ctrl->flow_mark_n = 0;
748 rxq_ctrl->rxq.mark = 0;
749 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
750 rxq_ctrl->flow_tunnels_n[j] = 0;
751 rxq_ctrl->rxq.tunnel = 0;
756 * return a pointer to the desired action in the list of actions.
759 * The list of actions to search the action in.
761 * The action to find.
764 * Pointer to the action in the list, if found. NULL otherwise.
766 const struct rte_flow_action *
767 mlx5_flow_find_action(const struct rte_flow_action *actions,
768 enum rte_flow_action_type action)
772 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
773 if (actions->type == action)
779 * Validate the flag action.
781 * @param[in] action_flags
782 * Bit-fields that holds the actions detected until now.
784 * Attributes of flow that includes this action.
786 * Pointer to error structure.
789 * 0 on success, a negative errno value otherwise and rte_errno is set.
792 mlx5_flow_validate_action_flag(uint64_t action_flags,
793 const struct rte_flow_attr *attr,
794 struct rte_flow_error *error)
797 if (action_flags & MLX5_FLOW_ACTION_DROP)
798 return rte_flow_error_set(error, EINVAL,
799 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
800 "can't drop and flag in same flow");
801 if (action_flags & MLX5_FLOW_ACTION_MARK)
802 return rte_flow_error_set(error, EINVAL,
803 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
804 "can't mark and flag in same flow");
805 if (action_flags & MLX5_FLOW_ACTION_FLAG)
806 return rte_flow_error_set(error, EINVAL,
807 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
809 " actions in same flow");
811 return rte_flow_error_set(error, ENOTSUP,
812 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
813 "flag action not supported for "
819 * Validate the mark action.
822 * Pointer to the queue action.
823 * @param[in] action_flags
824 * Bit-fields that holds the actions detected until now.
826 * Attributes of flow that includes this action.
828 * Pointer to error structure.
831 * 0 on success, a negative errno value otherwise and rte_errno is set.
834 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
835 uint64_t action_flags,
836 const struct rte_flow_attr *attr,
837 struct rte_flow_error *error)
839 const struct rte_flow_action_mark *mark = action->conf;
842 return rte_flow_error_set(error, EINVAL,
843 RTE_FLOW_ERROR_TYPE_ACTION,
845 "configuration cannot be null");
846 if (mark->id >= MLX5_FLOW_MARK_MAX)
847 return rte_flow_error_set(error, EINVAL,
848 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
850 "mark id must in 0 <= id < "
851 RTE_STR(MLX5_FLOW_MARK_MAX));
852 if (action_flags & MLX5_FLOW_ACTION_DROP)
853 return rte_flow_error_set(error, EINVAL,
854 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
855 "can't drop and mark in same flow");
856 if (action_flags & MLX5_FLOW_ACTION_FLAG)
857 return rte_flow_error_set(error, EINVAL,
858 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
859 "can't flag and mark in same flow");
860 if (action_flags & MLX5_FLOW_ACTION_MARK)
861 return rte_flow_error_set(error, EINVAL,
862 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
863 "can't have 2 mark actions in same"
866 return rte_flow_error_set(error, ENOTSUP,
867 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
868 "mark action not supported for "
874 * Validate the drop action.
876 * @param[in] action_flags
877 * Bit-fields that holds the actions detected until now.
879 * Attributes of flow that includes this action.
881 * Pointer to error structure.
884 * 0 on success, a negative errno value otherwise and rte_errno is set.
887 mlx5_flow_validate_action_drop(uint64_t action_flags,
888 const struct rte_flow_attr *attr,
889 struct rte_flow_error *error)
891 if (action_flags & MLX5_FLOW_ACTION_FLAG)
892 return rte_flow_error_set(error, EINVAL,
893 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
894 "can't drop and flag in same flow");
895 if (action_flags & MLX5_FLOW_ACTION_MARK)
896 return rte_flow_error_set(error, EINVAL,
897 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
898 "can't drop and mark in same flow");
899 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
900 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
901 return rte_flow_error_set(error, EINVAL,
902 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
903 "can't have 2 fate actions in"
906 return rte_flow_error_set(error, ENOTSUP,
907 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
908 "drop action not supported for "
914 * Validate the queue action.
917 * Pointer to the queue action.
918 * @param[in] action_flags
919 * Bit-fields that holds the actions detected until now.
921 * Pointer to the Ethernet device structure.
923 * Attributes of flow that includes this action.
925 * Pointer to error structure.
928 * 0 on success, a negative errno value otherwise and rte_errno is set.
931 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
932 uint64_t action_flags,
933 struct rte_eth_dev *dev,
934 const struct rte_flow_attr *attr,
935 struct rte_flow_error *error)
937 struct mlx5_priv *priv = dev->data->dev_private;
938 const struct rte_flow_action_queue *queue = action->conf;
940 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
941 return rte_flow_error_set(error, EINVAL,
942 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
943 "can't have 2 fate actions in"
946 return rte_flow_error_set(error, EINVAL,
947 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
948 NULL, "No Rx queues configured");
949 if (queue->index >= priv->rxqs_n)
950 return rte_flow_error_set(error, EINVAL,
951 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
953 "queue index out of range");
954 if (!(*priv->rxqs)[queue->index])
955 return rte_flow_error_set(error, EINVAL,
956 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
958 "queue is not configured");
960 return rte_flow_error_set(error, ENOTSUP,
961 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
962 "queue action not supported for "
968 * Validate the rss action.
971 * Pointer to the queue action.
972 * @param[in] action_flags
973 * Bit-fields that holds the actions detected until now.
975 * Pointer to the Ethernet device structure.
977 * Attributes of flow that includes this action.
978 * @param[in] item_flags
979 * Items that were detected.
981 * Pointer to error structure.
984 * 0 on success, a negative errno value otherwise and rte_errno is set.
987 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
988 uint64_t action_flags,
989 struct rte_eth_dev *dev,
990 const struct rte_flow_attr *attr,
992 struct rte_flow_error *error)
994 struct mlx5_priv *priv = dev->data->dev_private;
995 const struct rte_flow_action_rss *rss = action->conf;
996 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
999 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1000 return rte_flow_error_set(error, EINVAL,
1001 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1002 "can't have 2 fate actions"
1004 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1005 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1006 return rte_flow_error_set(error, ENOTSUP,
1007 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1009 "RSS hash function not supported");
1010 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1015 return rte_flow_error_set(error, ENOTSUP,
1016 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1018 "tunnel RSS is not supported");
1019 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1020 if (rss->key_len == 0 && rss->key != NULL)
1021 return rte_flow_error_set(error, ENOTSUP,
1022 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1024 "RSS hash key length 0");
1025 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1026 return rte_flow_error_set(error, ENOTSUP,
1027 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1029 "RSS hash key too small");
1030 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1031 return rte_flow_error_set(error, ENOTSUP,
1032 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1034 "RSS hash key too large");
1035 if (rss->queue_num > priv->config.ind_table_max_size)
1036 return rte_flow_error_set(error, ENOTSUP,
1037 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1039 "number of queues too large");
1040 if (rss->types & MLX5_RSS_HF_MASK)
1041 return rte_flow_error_set(error, ENOTSUP,
1042 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1044 "some RSS protocols are not"
1047 return rte_flow_error_set(error, EINVAL,
1048 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1049 NULL, "No Rx queues configured");
1050 if (!rss->queue_num)
1051 return rte_flow_error_set(error, EINVAL,
1052 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1053 NULL, "No queues configured");
1054 for (i = 0; i != rss->queue_num; ++i) {
1055 if (!(*priv->rxqs)[rss->queue[i]])
1056 return rte_flow_error_set
1057 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1058 &rss->queue[i], "queue is not configured");
1061 return rte_flow_error_set(error, ENOTSUP,
1062 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1063 "rss action not supported for "
1065 if (rss->level > 1 && !tunnel)
1066 return rte_flow_error_set(error, EINVAL,
1067 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1068 "inner RSS is not supported for "
1069 "non-tunnel flows");
1074 * Validate the count action.
1077 * Pointer to the Ethernet device structure.
1079 * Attributes of flow that includes this action.
1081 * Pointer to error structure.
1084 * 0 on success, a negative errno value otherwise and rte_errno is set.
1087 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1088 const struct rte_flow_attr *attr,
1089 struct rte_flow_error *error)
1092 return rte_flow_error_set(error, ENOTSUP,
1093 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1094 "count action not supported for "
1100 * Verify the @p attributes will be correctly understood by the NIC and store
1101 * them in the @p flow if everything is correct.
1104 * Pointer to the Ethernet device structure.
1105 * @param[in] attributes
1106 * Pointer to flow attributes
1108 * Pointer to error structure.
1111 * 0 on success, a negative errno value otherwise and rte_errno is set.
1114 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1115 const struct rte_flow_attr *attributes,
1116 struct rte_flow_error *error)
1118 struct mlx5_priv *priv = dev->data->dev_private;
1119 uint32_t priority_max = priv->config.flow_prio - 1;
1121 if (attributes->group)
1122 return rte_flow_error_set(error, ENOTSUP,
1123 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1124 NULL, "groups is not supported");
1125 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1126 attributes->priority >= priority_max)
1127 return rte_flow_error_set(error, ENOTSUP,
1128 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1129 NULL, "priority out of range");
1130 if (attributes->egress)
1131 return rte_flow_error_set(error, ENOTSUP,
1132 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1133 "egress is not supported");
1134 if (attributes->transfer && !priv->config.dv_esw_en)
1135 return rte_flow_error_set(error, ENOTSUP,
1136 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1137 NULL, "transfer is not supported");
1138 if (!attributes->ingress)
1139 return rte_flow_error_set(error, EINVAL,
1140 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1142 "ingress attribute is mandatory");
1147 * Validate ICMP6 item.
1150 * Item specification.
1151 * @param[in] item_flags
1152 * Bit-fields that holds the items detected until now.
1154 * Pointer to error structure.
1157 * 0 on success, a negative errno value otherwise and rte_errno is set.
1160 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1161 uint64_t item_flags,
1162 uint8_t target_protocol,
1163 struct rte_flow_error *error)
1165 const struct rte_flow_item_icmp6 *mask = item->mask;
1166 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1167 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1168 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1169 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1170 MLX5_FLOW_LAYER_OUTER_L4;
1173 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1174 return rte_flow_error_set(error, EINVAL,
1175 RTE_FLOW_ERROR_TYPE_ITEM, item,
1176 "protocol filtering not compatible"
1177 " with ICMP6 layer");
1178 if (!(item_flags & l3m))
1179 return rte_flow_error_set(error, EINVAL,
1180 RTE_FLOW_ERROR_TYPE_ITEM, item,
1181 "IPv6 is mandatory to filter on"
1183 if (item_flags & l4m)
1184 return rte_flow_error_set(error, EINVAL,
1185 RTE_FLOW_ERROR_TYPE_ITEM, item,
1186 "multiple L4 layers not supported");
1188 mask = &rte_flow_item_icmp6_mask;
1189 ret = mlx5_flow_item_acceptable
1190 (item, (const uint8_t *)mask,
1191 (const uint8_t *)&rte_flow_item_icmp6_mask,
1192 sizeof(struct rte_flow_item_icmp6), error);
1199 * Validate ICMP item.
1202 * Item specification.
1203 * @param[in] item_flags
1204 * Bit-fields that holds the items detected until now.
1206 * Pointer to error structure.
1209 * 0 on success, a negative errno value otherwise and rte_errno is set.
1212 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1213 uint64_t item_flags,
1214 uint8_t target_protocol,
1215 struct rte_flow_error *error)
1217 const struct rte_flow_item_icmp *mask = item->mask;
1218 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1219 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1220 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1221 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1222 MLX5_FLOW_LAYER_OUTER_L4;
1225 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1226 return rte_flow_error_set(error, EINVAL,
1227 RTE_FLOW_ERROR_TYPE_ITEM, item,
1228 "protocol filtering not compatible"
1229 " with ICMP layer");
1230 if (!(item_flags & l3m))
1231 return rte_flow_error_set(error, EINVAL,
1232 RTE_FLOW_ERROR_TYPE_ITEM, item,
1233 "IPv4 is mandatory to filter"
1235 if (item_flags & l4m)
1236 return rte_flow_error_set(error, EINVAL,
1237 RTE_FLOW_ERROR_TYPE_ITEM, item,
1238 "multiple L4 layers not supported");
1240 mask = &rte_flow_item_icmp_mask;
1241 ret = mlx5_flow_item_acceptable
1242 (item, (const uint8_t *)mask,
1243 (const uint8_t *)&rte_flow_item_icmp_mask,
1244 sizeof(struct rte_flow_item_icmp), error);
1251 * Validate Ethernet item.
1254 * Item specification.
1255 * @param[in] item_flags
1256 * Bit-fields that holds the items detected until now.
1258 * Pointer to error structure.
1261 * 0 on success, a negative errno value otherwise and rte_errno is set.
1264 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1265 uint64_t item_flags,
1266 struct rte_flow_error *error)
1268 const struct rte_flow_item_eth *mask = item->mask;
1269 const struct rte_flow_item_eth nic_mask = {
1270 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1271 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1272 .type = RTE_BE16(0xffff),
1275 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1276 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1277 MLX5_FLOW_LAYER_OUTER_L2;
1279 if (item_flags & ethm)
1280 return rte_flow_error_set(error, ENOTSUP,
1281 RTE_FLOW_ERROR_TYPE_ITEM, item,
1282 "multiple L2 layers not supported");
1283 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1284 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1285 return rte_flow_error_set(error, EINVAL,
1286 RTE_FLOW_ERROR_TYPE_ITEM, item,
1287 "L2 layer should not follow "
1289 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1290 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1291 return rte_flow_error_set(error, EINVAL,
1292 RTE_FLOW_ERROR_TYPE_ITEM, item,
1293 "L2 layer should not follow VLAN");
1295 mask = &rte_flow_item_eth_mask;
1296 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1297 (const uint8_t *)&nic_mask,
1298 sizeof(struct rte_flow_item_eth),
1304 * Validate VLAN item.
1307 * Item specification.
1308 * @param[in] item_flags
1309 * Bit-fields that holds the items detected until now.
1311 * Ethernet device flow is being created on.
1313 * Pointer to error structure.
1316 * 0 on success, a negative errno value otherwise and rte_errno is set.
1319 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1320 uint64_t item_flags,
1321 struct rte_eth_dev *dev,
1322 struct rte_flow_error *error)
1324 const struct rte_flow_item_vlan *spec = item->spec;
1325 const struct rte_flow_item_vlan *mask = item->mask;
1326 const struct rte_flow_item_vlan nic_mask = {
1327 .tci = RTE_BE16(UINT16_MAX),
1328 .inner_type = RTE_BE16(UINT16_MAX),
1330 uint16_t vlan_tag = 0;
1331 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1333 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1334 MLX5_FLOW_LAYER_INNER_L4) :
1335 (MLX5_FLOW_LAYER_OUTER_L3 |
1336 MLX5_FLOW_LAYER_OUTER_L4);
1337 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1338 MLX5_FLOW_LAYER_OUTER_VLAN;
1340 if (item_flags & vlanm)
1341 return rte_flow_error_set(error, EINVAL,
1342 RTE_FLOW_ERROR_TYPE_ITEM, item,
1343 "multiple VLAN layers not supported");
1344 else if ((item_flags & l34m) != 0)
1345 return rte_flow_error_set(error, EINVAL,
1346 RTE_FLOW_ERROR_TYPE_ITEM, item,
1347 "VLAN cannot follow L3/L4 layer");
1349 mask = &rte_flow_item_vlan_mask;
1350 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1351 (const uint8_t *)&nic_mask,
1352 sizeof(struct rte_flow_item_vlan),
1356 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1357 struct mlx5_priv *priv = dev->data->dev_private;
1359 if (priv->vmwa_context) {
1361 * Non-NULL context means we have a virtual machine
1362 * and SR-IOV enabled, we have to create VLAN interface
1363 * to make hypervisor to setup E-Switch vport
1364 * context correctly. We avoid creating the multiple
1365 * VLAN interfaces, so we cannot support VLAN tag mask.
1367 return rte_flow_error_set(error, EINVAL,
1368 RTE_FLOW_ERROR_TYPE_ITEM,
1370 "VLAN tag mask is not"
1371 " supported in virtual"
1376 vlan_tag = spec->tci;
1377 vlan_tag &= mask->tci;
1380 * From verbs perspective an empty VLAN is equivalent
1381 * to a packet without VLAN layer.
1384 return rte_flow_error_set(error, EINVAL,
1385 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1387 "VLAN cannot be empty");
1392 * Validate IPV4 item.
1395 * Item specification.
1396 * @param[in] item_flags
1397 * Bit-fields that holds the items detected until now.
1398 * @param[in] acc_mask
1399 * Acceptable mask, if NULL default internal default mask
1400 * will be used to check whether item fields are supported.
1402 * Pointer to error structure.
1405 * 0 on success, a negative errno value otherwise and rte_errno is set.
1408 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1409 uint64_t item_flags,
1411 uint16_t ether_type,
1412 const struct rte_flow_item_ipv4 *acc_mask,
1413 struct rte_flow_error *error)
1415 const struct rte_flow_item_ipv4 *mask = item->mask;
1416 const struct rte_flow_item_ipv4 *spec = item->spec;
1417 const struct rte_flow_item_ipv4 nic_mask = {
1419 .src_addr = RTE_BE32(0xffffffff),
1420 .dst_addr = RTE_BE32(0xffffffff),
1421 .type_of_service = 0xff,
1422 .next_proto_id = 0xff,
1425 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1426 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1427 MLX5_FLOW_LAYER_OUTER_L3;
1428 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1429 MLX5_FLOW_LAYER_OUTER_L4;
1431 uint8_t next_proto = 0xFF;
1432 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1433 MLX5_FLOW_LAYER_OUTER_VLAN |
1434 MLX5_FLOW_LAYER_INNER_VLAN);
1436 if ((last_item & l2_vlan) && ether_type &&
1437 ether_type != RTE_ETHER_TYPE_IPV4)
1438 return rte_flow_error_set(error, EINVAL,
1439 RTE_FLOW_ERROR_TYPE_ITEM, item,
1440 "IPv4 cannot follow L2/VLAN layer "
1441 "which ether type is not IPv4");
1442 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1444 next_proto = mask->hdr.next_proto_id &
1445 spec->hdr.next_proto_id;
1446 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1447 return rte_flow_error_set(error, EINVAL,
1448 RTE_FLOW_ERROR_TYPE_ITEM,
1453 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1454 return rte_flow_error_set(error, EINVAL,
1455 RTE_FLOW_ERROR_TYPE_ITEM, item,
1456 "wrong tunnel type - IPv6 specified "
1457 "but IPv4 item provided");
1458 if (item_flags & l3m)
1459 return rte_flow_error_set(error, ENOTSUP,
1460 RTE_FLOW_ERROR_TYPE_ITEM, item,
1461 "multiple L3 layers not supported");
1462 else if (item_flags & l4m)
1463 return rte_flow_error_set(error, EINVAL,
1464 RTE_FLOW_ERROR_TYPE_ITEM, item,
1465 "L3 cannot follow an L4 layer.");
1466 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1467 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1468 return rte_flow_error_set(error, EINVAL,
1469 RTE_FLOW_ERROR_TYPE_ITEM, item,
1470 "L3 cannot follow an NVGRE layer.");
1472 mask = &rte_flow_item_ipv4_mask;
1473 else if (mask->hdr.next_proto_id != 0 &&
1474 mask->hdr.next_proto_id != 0xff)
1475 return rte_flow_error_set(error, EINVAL,
1476 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1477 "partial mask is not supported"
1479 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1480 acc_mask ? (const uint8_t *)acc_mask
1481 : (const uint8_t *)&nic_mask,
1482 sizeof(struct rte_flow_item_ipv4),
1490 * Validate IPV6 item.
1493 * Item specification.
1494 * @param[in] item_flags
1495 * Bit-fields that holds the items detected until now.
1496 * @param[in] acc_mask
1497 * Acceptable mask, if NULL default internal default mask
1498 * will be used to check whether item fields are supported.
1500 * Pointer to error structure.
1503 * 0 on success, a negative errno value otherwise and rte_errno is set.
1506 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1507 uint64_t item_flags,
1509 uint16_t ether_type,
1510 const struct rte_flow_item_ipv6 *acc_mask,
1511 struct rte_flow_error *error)
1513 const struct rte_flow_item_ipv6 *mask = item->mask;
1514 const struct rte_flow_item_ipv6 *spec = item->spec;
1515 const struct rte_flow_item_ipv6 nic_mask = {
1518 "\xff\xff\xff\xff\xff\xff\xff\xff"
1519 "\xff\xff\xff\xff\xff\xff\xff\xff",
1521 "\xff\xff\xff\xff\xff\xff\xff\xff"
1522 "\xff\xff\xff\xff\xff\xff\xff\xff",
1523 .vtc_flow = RTE_BE32(0xffffffff),
1528 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1529 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1530 MLX5_FLOW_LAYER_OUTER_L3;
1531 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1532 MLX5_FLOW_LAYER_OUTER_L4;
1534 uint8_t next_proto = 0xFF;
1535 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1536 MLX5_FLOW_LAYER_OUTER_VLAN |
1537 MLX5_FLOW_LAYER_INNER_VLAN);
1539 if ((last_item & l2_vlan) && ether_type &&
1540 ether_type != RTE_ETHER_TYPE_IPV6)
1541 return rte_flow_error_set(error, EINVAL,
1542 RTE_FLOW_ERROR_TYPE_ITEM, item,
1543 "IPv6 cannot follow L2/VLAN layer "
1544 "which ether type is not IPv6");
1545 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1547 next_proto = mask->hdr.proto & spec->hdr.proto;
1548 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1549 return rte_flow_error_set(error, EINVAL,
1550 RTE_FLOW_ERROR_TYPE_ITEM,
1555 if (item_flags & MLX5_FLOW_LAYER_IPIP)
1556 return rte_flow_error_set(error, EINVAL,
1557 RTE_FLOW_ERROR_TYPE_ITEM, item,
1558 "wrong tunnel type - IPv4 specified "
1559 "but IPv6 item provided");
1560 if (item_flags & l3m)
1561 return rte_flow_error_set(error, ENOTSUP,
1562 RTE_FLOW_ERROR_TYPE_ITEM, item,
1563 "multiple L3 layers not supported");
1564 else if (item_flags & l4m)
1565 return rte_flow_error_set(error, EINVAL,
1566 RTE_FLOW_ERROR_TYPE_ITEM, item,
1567 "L3 cannot follow an L4 layer.");
1568 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1569 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1570 return rte_flow_error_set(error, EINVAL,
1571 RTE_FLOW_ERROR_TYPE_ITEM, item,
1572 "L3 cannot follow an NVGRE layer.");
1574 mask = &rte_flow_item_ipv6_mask;
1575 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1576 acc_mask ? (const uint8_t *)acc_mask
1577 : (const uint8_t *)&nic_mask,
1578 sizeof(struct rte_flow_item_ipv6),
1586 * Validate UDP item.
1589 * Item specification.
1590 * @param[in] item_flags
1591 * Bit-fields that holds the items detected until now.
1592 * @param[in] target_protocol
1593 * The next protocol in the previous item.
1594 * @param[in] flow_mask
1595 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1597 * Pointer to error structure.
1600 * 0 on success, a negative errno value otherwise and rte_errno is set.
1603 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1604 uint64_t item_flags,
1605 uint8_t target_protocol,
1606 struct rte_flow_error *error)
1608 const struct rte_flow_item_udp *mask = item->mask;
1609 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1610 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1611 MLX5_FLOW_LAYER_OUTER_L3;
1612 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1613 MLX5_FLOW_LAYER_OUTER_L4;
1616 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1617 return rte_flow_error_set(error, EINVAL,
1618 RTE_FLOW_ERROR_TYPE_ITEM, item,
1619 "protocol filtering not compatible"
1621 if (!(item_flags & l3m))
1622 return rte_flow_error_set(error, EINVAL,
1623 RTE_FLOW_ERROR_TYPE_ITEM, item,
1624 "L3 is mandatory to filter on L4");
1625 if (item_flags & l4m)
1626 return rte_flow_error_set(error, EINVAL,
1627 RTE_FLOW_ERROR_TYPE_ITEM, item,
1628 "multiple L4 layers not supported");
1630 mask = &rte_flow_item_udp_mask;
1631 ret = mlx5_flow_item_acceptable
1632 (item, (const uint8_t *)mask,
1633 (const uint8_t *)&rte_flow_item_udp_mask,
1634 sizeof(struct rte_flow_item_udp), error);
1641 * Validate TCP item.
1644 * Item specification.
1645 * @param[in] item_flags
1646 * Bit-fields that holds the items detected until now.
1647 * @param[in] target_protocol
1648 * The next protocol in the previous item.
1650 * Pointer to error structure.
1653 * 0 on success, a negative errno value otherwise and rte_errno is set.
1656 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1657 uint64_t item_flags,
1658 uint8_t target_protocol,
1659 const struct rte_flow_item_tcp *flow_mask,
1660 struct rte_flow_error *error)
1662 const struct rte_flow_item_tcp *mask = item->mask;
1663 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1664 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1665 MLX5_FLOW_LAYER_OUTER_L3;
1666 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1667 MLX5_FLOW_LAYER_OUTER_L4;
1671 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1672 return rte_flow_error_set(error, EINVAL,
1673 RTE_FLOW_ERROR_TYPE_ITEM, item,
1674 "protocol filtering not compatible"
1676 if (!(item_flags & l3m))
1677 return rte_flow_error_set(error, EINVAL,
1678 RTE_FLOW_ERROR_TYPE_ITEM, item,
1679 "L3 is mandatory to filter on L4");
1680 if (item_flags & l4m)
1681 return rte_flow_error_set(error, EINVAL,
1682 RTE_FLOW_ERROR_TYPE_ITEM, item,
1683 "multiple L4 layers not supported");
1685 mask = &rte_flow_item_tcp_mask;
1686 ret = mlx5_flow_item_acceptable
1687 (item, (const uint8_t *)mask,
1688 (const uint8_t *)flow_mask,
1689 sizeof(struct rte_flow_item_tcp), error);
1696 * Validate VXLAN item.
1699 * Item specification.
1700 * @param[in] item_flags
1701 * Bit-fields that holds the items detected until now.
1702 * @param[in] target_protocol
1703 * The next protocol in the previous item.
1705 * Pointer to error structure.
1708 * 0 on success, a negative errno value otherwise and rte_errno is set.
1711 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1712 uint64_t item_flags,
1713 struct rte_flow_error *error)
1715 const struct rte_flow_item_vxlan *spec = item->spec;
1716 const struct rte_flow_item_vxlan *mask = item->mask;
1721 } id = { .vlan_id = 0, };
1722 uint32_t vlan_id = 0;
1725 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1726 return rte_flow_error_set(error, ENOTSUP,
1727 RTE_FLOW_ERROR_TYPE_ITEM, item,
1728 "multiple tunnel layers not"
1731 * Verify only UDPv4 is present as defined in
1732 * https://tools.ietf.org/html/rfc7348
1734 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1735 return rte_flow_error_set(error, EINVAL,
1736 RTE_FLOW_ERROR_TYPE_ITEM, item,
1737 "no outer UDP layer found");
1739 mask = &rte_flow_item_vxlan_mask;
1740 ret = mlx5_flow_item_acceptable
1741 (item, (const uint8_t *)mask,
1742 (const uint8_t *)&rte_flow_item_vxlan_mask,
1743 sizeof(struct rte_flow_item_vxlan),
1748 memcpy(&id.vni[1], spec->vni, 3);
1749 vlan_id = id.vlan_id;
1750 memcpy(&id.vni[1], mask->vni, 3);
1751 vlan_id &= id.vlan_id;
1754 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1755 * only this layer is defined in the Verbs specification it is
1756 * interpreted as wildcard and all packets will match this
1757 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1758 * udp), all packets matching the layers before will also
1759 * match this rule. To avoid such situation, VNI 0 is
1760 * currently refused.
1763 return rte_flow_error_set(error, ENOTSUP,
1764 RTE_FLOW_ERROR_TYPE_ITEM, item,
1765 "VXLAN vni cannot be 0");
1766 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1767 return rte_flow_error_set(error, ENOTSUP,
1768 RTE_FLOW_ERROR_TYPE_ITEM, item,
1769 "VXLAN tunnel must be fully defined");
1774 * Validate VXLAN_GPE item.
1777 * Item specification.
1778 * @param[in] item_flags
1779 * Bit-fields that holds the items detected until now.
1781 * Pointer to the private data structure.
1782 * @param[in] target_protocol
1783 * The next protocol in the previous item.
1785 * Pointer to error structure.
1788 * 0 on success, a negative errno value otherwise and rte_errno is set.
1791 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1792 uint64_t item_flags,
1793 struct rte_eth_dev *dev,
1794 struct rte_flow_error *error)
1796 struct mlx5_priv *priv = dev->data->dev_private;
1797 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1798 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1803 } id = { .vlan_id = 0, };
1804 uint32_t vlan_id = 0;
1806 if (!priv->config.l3_vxlan_en)
1807 return rte_flow_error_set(error, ENOTSUP,
1808 RTE_FLOW_ERROR_TYPE_ITEM, item,
1809 "L3 VXLAN is not enabled by device"
1810 " parameter and/or not configured in"
1812 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1813 return rte_flow_error_set(error, ENOTSUP,
1814 RTE_FLOW_ERROR_TYPE_ITEM, item,
1815 "multiple tunnel layers not"
1818 * Verify only UDPv4 is present as defined in
1819 * https://tools.ietf.org/html/rfc7348
1821 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1822 return rte_flow_error_set(error, EINVAL,
1823 RTE_FLOW_ERROR_TYPE_ITEM, item,
1824 "no outer UDP layer found");
1826 mask = &rte_flow_item_vxlan_gpe_mask;
1827 ret = mlx5_flow_item_acceptable
1828 (item, (const uint8_t *)mask,
1829 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1830 sizeof(struct rte_flow_item_vxlan_gpe),
1836 return rte_flow_error_set(error, ENOTSUP,
1837 RTE_FLOW_ERROR_TYPE_ITEM,
1839 "VxLAN-GPE protocol"
1841 memcpy(&id.vni[1], spec->vni, 3);
1842 vlan_id = id.vlan_id;
1843 memcpy(&id.vni[1], mask->vni, 3);
1844 vlan_id &= id.vlan_id;
1847 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1848 * layer is defined in the Verbs specification it is interpreted as
1849 * wildcard and all packets will match this rule, if it follows a full
1850 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1851 * before will also match this rule. To avoid such situation, VNI 0
1852 * is currently refused.
1855 return rte_flow_error_set(error, ENOTSUP,
1856 RTE_FLOW_ERROR_TYPE_ITEM, item,
1857 "VXLAN-GPE vni cannot be 0");
1858 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1859 return rte_flow_error_set(error, ENOTSUP,
1860 RTE_FLOW_ERROR_TYPE_ITEM, item,
1861 "VXLAN-GPE tunnel must be fully"
1866 * Validate GRE Key item.
1869 * Item specification.
1870 * @param[in] item_flags
1871 * Bit flags to mark detected items.
1872 * @param[in] gre_item
1873 * Pointer to gre_item
1875 * Pointer to error structure.
1878 * 0 on success, a negative errno value otherwise and rte_errno is set.
1881 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1882 uint64_t item_flags,
1883 const struct rte_flow_item *gre_item,
1884 struct rte_flow_error *error)
1886 const rte_be32_t *mask = item->mask;
1888 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1889 const struct rte_flow_item_gre *gre_spec = gre_item->spec;
1890 const struct rte_flow_item_gre *gre_mask = gre_item->mask;
1892 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1893 return rte_flow_error_set(error, ENOTSUP,
1894 RTE_FLOW_ERROR_TYPE_ITEM, item,
1895 "Multiple GRE key not support");
1896 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1897 return rte_flow_error_set(error, ENOTSUP,
1898 RTE_FLOW_ERROR_TYPE_ITEM, item,
1899 "No preceding GRE header");
1900 if (item_flags & MLX5_FLOW_LAYER_INNER)
1901 return rte_flow_error_set(error, ENOTSUP,
1902 RTE_FLOW_ERROR_TYPE_ITEM, item,
1903 "GRE key following a wrong item");
1905 gre_mask = &rte_flow_item_gre_mask;
1906 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
1907 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
1908 return rte_flow_error_set(error, EINVAL,
1909 RTE_FLOW_ERROR_TYPE_ITEM, item,
1910 "Key bit must be on");
1913 mask = &gre_key_default_mask;
1914 ret = mlx5_flow_item_acceptable
1915 (item, (const uint8_t *)mask,
1916 (const uint8_t *)&gre_key_default_mask,
1917 sizeof(rte_be32_t), error);
1922 * Validate GRE item.
1925 * Item specification.
1926 * @param[in] item_flags
1927 * Bit flags to mark detected items.
1928 * @param[in] target_protocol
1929 * The next protocol in the previous item.
1931 * Pointer to error structure.
1934 * 0 on success, a negative errno value otherwise and rte_errno is set.
1937 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1938 uint64_t item_flags,
1939 uint8_t target_protocol,
1940 struct rte_flow_error *error)
1942 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1943 const struct rte_flow_item_gre *mask = item->mask;
1945 const struct rte_flow_item_gre nic_mask = {
1946 .c_rsvd0_ver = RTE_BE16(0xB000),
1947 .protocol = RTE_BE16(UINT16_MAX),
1950 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1951 return rte_flow_error_set(error, EINVAL,
1952 RTE_FLOW_ERROR_TYPE_ITEM, item,
1953 "protocol filtering not compatible"
1954 " with this GRE layer");
1955 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1956 return rte_flow_error_set(error, ENOTSUP,
1957 RTE_FLOW_ERROR_TYPE_ITEM, item,
1958 "multiple tunnel layers not"
1960 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1961 return rte_flow_error_set(error, ENOTSUP,
1962 RTE_FLOW_ERROR_TYPE_ITEM, item,
1963 "L3 Layer is missing");
1965 mask = &rte_flow_item_gre_mask;
1966 ret = mlx5_flow_item_acceptable
1967 (item, (const uint8_t *)mask,
1968 (const uint8_t *)&nic_mask,
1969 sizeof(struct rte_flow_item_gre), error);
1972 #ifndef HAVE_MLX5DV_DR
1973 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1974 if (spec && (spec->protocol & mask->protocol))
1975 return rte_flow_error_set(error, ENOTSUP,
1976 RTE_FLOW_ERROR_TYPE_ITEM, item,
1977 "without MPLS support the"
1978 " specification cannot be used for"
1986 * Validate Geneve item.
1989 * Item specification.
1990 * @param[in] itemFlags
1991 * Bit-fields that holds the items detected until now.
1993 * Pointer to the private data structure.
1995 * Pointer to error structure.
1998 * 0 on success, a negative errno value otherwise and rte_errno is set.
2002 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2003 uint64_t item_flags,
2004 struct rte_eth_dev *dev,
2005 struct rte_flow_error *error)
2007 struct mlx5_priv *priv = dev->data->dev_private;
2008 const struct rte_flow_item_geneve *spec = item->spec;
2009 const struct rte_flow_item_geneve *mask = item->mask;
2012 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2013 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2014 const struct rte_flow_item_geneve nic_mask = {
2015 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2016 .vni = "\xff\xff\xff",
2017 .protocol = RTE_BE16(UINT16_MAX),
2020 if (!(priv->config.hca_attr.flex_parser_protocols &
2021 MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2022 !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2023 return rte_flow_error_set(error, ENOTSUP,
2024 RTE_FLOW_ERROR_TYPE_ITEM, item,
2025 "L3 Geneve is not enabled by device"
2026 " parameter and/or not configured in"
2028 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2029 return rte_flow_error_set(error, ENOTSUP,
2030 RTE_FLOW_ERROR_TYPE_ITEM, item,
2031 "multiple tunnel layers not"
2034 * Verify only UDPv4 is present as defined in
2035 * https://tools.ietf.org/html/rfc7348
2037 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2038 return rte_flow_error_set(error, EINVAL,
2039 RTE_FLOW_ERROR_TYPE_ITEM, item,
2040 "no outer UDP layer found");
2042 mask = &rte_flow_item_geneve_mask;
2043 ret = mlx5_flow_item_acceptable
2044 (item, (const uint8_t *)mask,
2045 (const uint8_t *)&nic_mask,
2046 sizeof(struct rte_flow_item_geneve), error);
2050 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2051 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2052 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2053 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2054 return rte_flow_error_set(error, ENOTSUP,
2055 RTE_FLOW_ERROR_TYPE_ITEM,
2057 "Geneve protocol unsupported"
2058 " fields are being used");
2059 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2060 return rte_flow_error_set
2062 RTE_FLOW_ERROR_TYPE_ITEM,
2064 "Unsupported Geneve options length");
2066 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2067 return rte_flow_error_set
2069 RTE_FLOW_ERROR_TYPE_ITEM, item,
2070 "Geneve tunnel must be fully defined");
2075 * Validate MPLS item.
2078 * Pointer to the rte_eth_dev structure.
2080 * Item specification.
2081 * @param[in] item_flags
2082 * Bit-fields that holds the items detected until now.
2083 * @param[in] prev_layer
2084 * The protocol layer indicated in previous item.
2086 * Pointer to error structure.
2089 * 0 on success, a negative errno value otherwise and rte_errno is set.
2092 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2093 const struct rte_flow_item *item __rte_unused,
2094 uint64_t item_flags __rte_unused,
2095 uint64_t prev_layer __rte_unused,
2096 struct rte_flow_error *error)
2098 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2099 const struct rte_flow_item_mpls *mask = item->mask;
2100 struct mlx5_priv *priv = dev->data->dev_private;
2103 if (!priv->config.mpls_en)
2104 return rte_flow_error_set(error, ENOTSUP,
2105 RTE_FLOW_ERROR_TYPE_ITEM, item,
2106 "MPLS not supported or"
2107 " disabled in firmware"
2109 /* MPLS over IP, UDP, GRE is allowed */
2110 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2111 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2112 MLX5_FLOW_LAYER_GRE)))
2113 return rte_flow_error_set(error, EINVAL,
2114 RTE_FLOW_ERROR_TYPE_ITEM, item,
2115 "protocol filtering not compatible"
2116 " with MPLS layer");
2117 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2118 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2119 !(item_flags & MLX5_FLOW_LAYER_GRE))
2120 return rte_flow_error_set(error, ENOTSUP,
2121 RTE_FLOW_ERROR_TYPE_ITEM, item,
2122 "multiple tunnel layers not"
2125 mask = &rte_flow_item_mpls_mask;
2126 ret = mlx5_flow_item_acceptable
2127 (item, (const uint8_t *)mask,
2128 (const uint8_t *)&rte_flow_item_mpls_mask,
2129 sizeof(struct rte_flow_item_mpls), error);
2134 return rte_flow_error_set(error, ENOTSUP,
2135 RTE_FLOW_ERROR_TYPE_ITEM, item,
2136 "MPLS is not supported by Verbs, please"
2141 * Validate NVGRE item.
2144 * Item specification.
2145 * @param[in] item_flags
2146 * Bit flags to mark detected items.
2147 * @param[in] target_protocol
2148 * The next protocol in the previous item.
2150 * Pointer to error structure.
2153 * 0 on success, a negative errno value otherwise and rte_errno is set.
2156 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2157 uint64_t item_flags,
2158 uint8_t target_protocol,
2159 struct rte_flow_error *error)
2161 const struct rte_flow_item_nvgre *mask = item->mask;
2164 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2165 return rte_flow_error_set(error, EINVAL,
2166 RTE_FLOW_ERROR_TYPE_ITEM, item,
2167 "protocol filtering not compatible"
2168 " with this GRE layer");
2169 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2170 return rte_flow_error_set(error, ENOTSUP,
2171 RTE_FLOW_ERROR_TYPE_ITEM, item,
2172 "multiple tunnel layers not"
2174 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2175 return rte_flow_error_set(error, ENOTSUP,
2176 RTE_FLOW_ERROR_TYPE_ITEM, item,
2177 "L3 Layer is missing");
2179 mask = &rte_flow_item_nvgre_mask;
2180 ret = mlx5_flow_item_acceptable
2181 (item, (const uint8_t *)mask,
2182 (const uint8_t *)&rte_flow_item_nvgre_mask,
2183 sizeof(struct rte_flow_item_nvgre), error);
2190 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2191 const struct rte_flow_attr *attr __rte_unused,
2192 const struct rte_flow_item items[] __rte_unused,
2193 const struct rte_flow_action actions[] __rte_unused,
2194 bool external __rte_unused,
2195 struct rte_flow_error *error)
2197 return rte_flow_error_set(error, ENOTSUP,
2198 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2201 static struct mlx5_flow *
2202 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2203 const struct rte_flow_item items[] __rte_unused,
2204 const struct rte_flow_action actions[] __rte_unused,
2205 struct rte_flow_error *error)
2207 rte_flow_error_set(error, ENOTSUP,
2208 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2213 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2214 struct mlx5_flow *dev_flow __rte_unused,
2215 const struct rte_flow_attr *attr __rte_unused,
2216 const struct rte_flow_item items[] __rte_unused,
2217 const struct rte_flow_action actions[] __rte_unused,
2218 struct rte_flow_error *error)
2220 return rte_flow_error_set(error, ENOTSUP,
2221 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2225 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2226 struct rte_flow *flow __rte_unused,
2227 struct rte_flow_error *error)
2229 return rte_flow_error_set(error, ENOTSUP,
2230 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2234 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2235 struct rte_flow *flow __rte_unused)
2240 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2241 struct rte_flow *flow __rte_unused)
2246 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2247 struct rte_flow *flow __rte_unused,
2248 const struct rte_flow_action *actions __rte_unused,
2249 void *data __rte_unused,
2250 struct rte_flow_error *error)
2252 return rte_flow_error_set(error, ENOTSUP,
2253 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2256 /* Void driver to protect from null pointer reference. */
2257 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2258 .validate = flow_null_validate,
2259 .prepare = flow_null_prepare,
2260 .translate = flow_null_translate,
2261 .apply = flow_null_apply,
2262 .remove = flow_null_remove,
2263 .destroy = flow_null_destroy,
2264 .query = flow_null_query,
2268 * Select flow driver type according to flow attributes and device
2272 * Pointer to the dev structure.
2274 * Pointer to the flow attributes.
2277 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2279 static enum mlx5_flow_drv_type
2280 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2282 struct mlx5_priv *priv = dev->data->dev_private;
2283 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2285 if (attr->transfer && priv->config.dv_esw_en)
2286 type = MLX5_FLOW_TYPE_DV;
2287 if (!attr->transfer)
2288 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2289 MLX5_FLOW_TYPE_VERBS;
2293 #define flow_get_drv_ops(type) flow_drv_ops[type]
2296 * Flow driver validation API. This abstracts calling driver specific functions.
2297 * The type of flow driver is determined according to flow attributes.
2300 * Pointer to the dev structure.
2302 * Pointer to the flow attributes.
2304 * Pointer to the list of items.
2305 * @param[in] actions
2306 * Pointer to the list of actions.
2307 * @param[in] external
2308 * This flow rule is created by request external to PMD.
2310 * Pointer to the error structure.
2313 * 0 on success, a negative errno value otherwise and rte_errno is set.
2316 flow_drv_validate(struct rte_eth_dev *dev,
2317 const struct rte_flow_attr *attr,
2318 const struct rte_flow_item items[],
2319 const struct rte_flow_action actions[],
2320 bool external, struct rte_flow_error *error)
2322 const struct mlx5_flow_driver_ops *fops;
2323 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2325 fops = flow_get_drv_ops(type);
2326 return fops->validate(dev, attr, items, actions, external, error);
2330 * Flow driver preparation API. This abstracts calling driver specific
2331 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2332 * calculates the size of memory required for device flow, allocates the memory,
2333 * initializes the device flow and returns the pointer.
2336 * This function initializes device flow structure such as dv or verbs in
2337 * struct mlx5_flow. However, it is caller's responsibility to initialize the
2338 * rest. For example, adding returning device flow to flow->dev_flow list and
2339 * setting backward reference to the flow should be done out of this function.
2340 * layers field is not filled either.
2343 * Pointer to the flow attributes.
2345 * Pointer to the list of items.
2346 * @param[in] actions
2347 * Pointer to the list of actions.
2349 * Pointer to the error structure.
2352 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
2354 static inline struct mlx5_flow *
2355 flow_drv_prepare(const struct rte_flow *flow,
2356 const struct rte_flow_attr *attr,
2357 const struct rte_flow_item items[],
2358 const struct rte_flow_action actions[],
2359 struct rte_flow_error *error)
2361 const struct mlx5_flow_driver_ops *fops;
2362 enum mlx5_flow_drv_type type = flow->drv_type;
2364 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2365 fops = flow_get_drv_ops(type);
2366 return fops->prepare(attr, items, actions, error);
2370 * Flow driver translation API. This abstracts calling driver specific
2371 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2372 * translates a generic flow into a driver flow. flow_drv_prepare() must
2376 * dev_flow->layers could be filled as a result of parsing during translation
2377 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2378 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2379 * flow->actions could be overwritten even though all the expanded dev_flows
2380 * have the same actions.
2383 * Pointer to the rte dev structure.
2384 * @param[in, out] dev_flow
2385 * Pointer to the mlx5 flow.
2387 * Pointer to the flow attributes.
2389 * Pointer to the list of items.
2390 * @param[in] actions
2391 * Pointer to the list of actions.
2393 * Pointer to the error structure.
2396 * 0 on success, a negative errno value otherwise and rte_errno is set.
2399 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2400 const struct rte_flow_attr *attr,
2401 const struct rte_flow_item items[],
2402 const struct rte_flow_action actions[],
2403 struct rte_flow_error *error)
2405 const struct mlx5_flow_driver_ops *fops;
2406 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2408 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2409 fops = flow_get_drv_ops(type);
2410 return fops->translate(dev, dev_flow, attr, items, actions, error);
2414 * Flow driver apply API. This abstracts calling driver specific functions.
2415 * Parent flow (rte_flow) should have driver type (drv_type). It applies
2416 * translated driver flows on to device. flow_drv_translate() must precede.
2419 * Pointer to Ethernet device structure.
2420 * @param[in, out] flow
2421 * Pointer to flow structure.
2423 * Pointer to error structure.
2426 * 0 on success, a negative errno value otherwise and rte_errno is set.
2429 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2430 struct rte_flow_error *error)
2432 const struct mlx5_flow_driver_ops *fops;
2433 enum mlx5_flow_drv_type type = flow->drv_type;
2435 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2436 fops = flow_get_drv_ops(type);
2437 return fops->apply(dev, flow, error);
2441 * Flow driver remove API. This abstracts calling driver specific functions.
2442 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2443 * on device. All the resources of the flow should be freed by calling
2444 * flow_drv_destroy().
2447 * Pointer to Ethernet device.
2448 * @param[in, out] flow
2449 * Pointer to flow structure.
2452 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2454 const struct mlx5_flow_driver_ops *fops;
2455 enum mlx5_flow_drv_type type = flow->drv_type;
2457 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2458 fops = flow_get_drv_ops(type);
2459 fops->remove(dev, flow);
2463 * Flow driver destroy API. This abstracts calling driver specific functions.
2464 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2465 * on device and releases resources of the flow.
2468 * Pointer to Ethernet device.
2469 * @param[in, out] flow
2470 * Pointer to flow structure.
2473 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2475 const struct mlx5_flow_driver_ops *fops;
2476 enum mlx5_flow_drv_type type = flow->drv_type;
2478 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2479 fops = flow_get_drv_ops(type);
2480 fops->destroy(dev, flow);
2484 * Validate a flow supported by the NIC.
2486 * @see rte_flow_validate()
2490 mlx5_flow_validate(struct rte_eth_dev *dev,
2491 const struct rte_flow_attr *attr,
2492 const struct rte_flow_item items[],
2493 const struct rte_flow_action actions[],
2494 struct rte_flow_error *error)
2498 ret = flow_drv_validate(dev, attr, items, actions, true, error);
2505 * Get RSS action from the action list.
2507 * @param[in] actions
2508 * Pointer to the list of actions.
2511 * Pointer to the RSS action if exist, else return NULL.
2513 static const struct rte_flow_action_rss*
2514 flow_get_rss_action(const struct rte_flow_action actions[])
2516 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2517 switch (actions->type) {
2518 case RTE_FLOW_ACTION_TYPE_RSS:
2519 return (const struct rte_flow_action_rss *)
2529 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2531 const struct rte_flow_item *item;
2532 unsigned int has_vlan = 0;
2534 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2535 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2541 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2542 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2543 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2544 MLX5_EXPANSION_ROOT_OUTER;
2548 * Check if the flow should be splited due to hairpin.
2549 * The reason for the split is that in current HW we can't
2550 * support encap on Rx, so if a flow have encap we move it
2554 * Pointer to Ethernet device.
2556 * Flow rule attributes.
2557 * @param[in] actions
2558 * Associated actions (list terminated by the END action).
2561 * > 0 the number of actions and the flow should be split,
2562 * 0 when no split required.
2565 flow_check_hairpin_split(struct rte_eth_dev *dev,
2566 const struct rte_flow_attr *attr,
2567 const struct rte_flow_action actions[])
2569 int queue_action = 0;
2572 const struct rte_flow_action_queue *queue;
2573 const struct rte_flow_action_rss *rss;
2574 const struct rte_flow_action_raw_encap *raw_encap;
2578 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2579 switch (actions->type) {
2580 case RTE_FLOW_ACTION_TYPE_QUEUE:
2581 queue = actions->conf;
2582 if (mlx5_rxq_get_type(dev, queue->index) !=
2583 MLX5_RXQ_TYPE_HAIRPIN)
2588 case RTE_FLOW_ACTION_TYPE_RSS:
2589 rss = actions->conf;
2590 if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2591 MLX5_RXQ_TYPE_HAIRPIN)
2596 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2597 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2601 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2602 raw_encap = actions->conf;
2603 if (raw_encap->size >
2604 (sizeof(struct rte_flow_item_eth) +
2605 sizeof(struct rte_flow_item_ipv4)))
2614 if (encap == 1 && queue_action)
2619 #define MLX5_MAX_SPLIT_ACTIONS 24
2620 #define MLX5_MAX_SPLIT_ITEMS 24
2623 * Split the hairpin flow.
2624 * Since HW can't support encap on Rx we move the encap to Tx.
2625 * If the count action is after the encap then we also
2626 * move the count action. in this case the count will also measure
2630 * Pointer to Ethernet device.
2631 * @param[in] actions
2632 * Associated actions (list terminated by the END action).
2633 * @param[out] actions_rx
2635 * @param[out] actions_tx
2637 * @param[out] pattern_tx
2638 * The pattern items for the Tx flow.
2639 * @param[out] flow_id
2640 * The flow ID connected to this flow.
2646 flow_hairpin_split(struct rte_eth_dev *dev,
2647 const struct rte_flow_action actions[],
2648 struct rte_flow_action actions_rx[],
2649 struct rte_flow_action actions_tx[],
2650 struct rte_flow_item pattern_tx[],
2653 struct mlx5_priv *priv = dev->data->dev_private;
2654 const struct rte_flow_action_raw_encap *raw_encap;
2655 const struct rte_flow_action_raw_decap *raw_decap;
2656 struct mlx5_rte_flow_action_set_tag *set_tag;
2657 struct rte_flow_action *tag_action;
2658 struct mlx5_rte_flow_item_tag *tag_item;
2659 struct rte_flow_item *item;
2661 struct rte_flow_error error;
2664 mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
2665 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2666 switch (actions->type) {
2667 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2668 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2669 rte_memcpy(actions_tx, actions,
2670 sizeof(struct rte_flow_action));
2673 case RTE_FLOW_ACTION_TYPE_COUNT:
2675 rte_memcpy(actions_tx, actions,
2676 sizeof(struct rte_flow_action));
2679 rte_memcpy(actions_rx, actions,
2680 sizeof(struct rte_flow_action));
2684 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2685 raw_encap = actions->conf;
2686 if (raw_encap->size >
2687 (sizeof(struct rte_flow_item_eth) +
2688 sizeof(struct rte_flow_item_ipv4))) {
2689 memcpy(actions_tx, actions,
2690 sizeof(struct rte_flow_action));
2694 rte_memcpy(actions_rx, actions,
2695 sizeof(struct rte_flow_action));
2699 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2700 raw_decap = actions->conf;
2701 if (raw_decap->size <
2702 (sizeof(struct rte_flow_item_eth) +
2703 sizeof(struct rte_flow_item_ipv4))) {
2704 memcpy(actions_tx, actions,
2705 sizeof(struct rte_flow_action));
2708 rte_memcpy(actions_rx, actions,
2709 sizeof(struct rte_flow_action));
2714 rte_memcpy(actions_rx, actions,
2715 sizeof(struct rte_flow_action));
2720 /* Add set meta action and end action for the Rx flow. */
2721 tag_action = actions_rx;
2722 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
2724 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
2726 set_tag = (void *)actions_rx;
2727 set_tag->id = flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, &error);
2728 set_tag->data = *flow_id;
2729 tag_action->conf = set_tag;
2730 /* Create Tx item list. */
2731 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
2732 addr = (void *)&pattern_tx[2];
2734 item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
2735 tag_item = (void *)addr;
2736 tag_item->data = *flow_id;
2737 tag_item->id = flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
2738 item->spec = tag_item;
2739 addr += sizeof(struct mlx5_rte_flow_item_tag);
2740 tag_item = (void *)addr;
2741 tag_item->data = UINT32_MAX;
2742 tag_item->id = UINT16_MAX;
2743 item->mask = tag_item;
2744 addr += sizeof(struct mlx5_rte_flow_item_tag);
2747 item->type = RTE_FLOW_ITEM_TYPE_END;
2752 * Create a flow and add it to @p list.
2755 * Pointer to Ethernet device.
2757 * Pointer to a TAILQ flow list. If this parameter NULL,
2758 * no list insertion occurred, flow is just created,
2759 * this is caller's responsibility to track the
2762 * Flow rule attributes.
2764 * Pattern specification (list terminated by the END pattern item).
2765 * @param[in] actions
2766 * Associated actions (list terminated by the END action).
2767 * @param[in] external
2768 * This flow rule is created by request external to PMD.
2770 * Perform verbose error reporting if not NULL.
2773 * A flow on success, NULL otherwise and rte_errno is set.
2775 static struct rte_flow *
2776 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2777 const struct rte_flow_attr *attr,
2778 const struct rte_flow_item items[],
2779 const struct rte_flow_action actions[],
2780 bool external, struct rte_flow_error *error)
2782 struct mlx5_priv *priv = dev->data->dev_private;
2783 struct rte_flow *flow = NULL;
2784 struct mlx5_flow *dev_flow;
2785 const struct rte_flow_action_rss *rss;
2787 struct rte_flow_expand_rss buf;
2788 uint8_t buffer[2048];
2791 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
2792 uint8_t buffer[2048];
2795 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
2796 uint8_t buffer[2048];
2797 } actions_hairpin_tx;
2799 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
2800 uint8_t buffer[2048];
2802 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2803 const struct rte_flow_action *p_actions_rx = actions;
2807 int hairpin_flow = 0;
2808 uint32_t hairpin_id = 0;
2809 struct rte_flow_attr attr_tx = { .priority = 0 };
2811 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
2812 if (hairpin_flow > 0) {
2813 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
2817 flow_hairpin_split(dev, actions, actions_rx.actions,
2818 actions_hairpin_tx.actions, items_tx.items,
2820 p_actions_rx = actions_rx.actions;
2822 ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
2825 goto error_before_flow;
2826 flow_size = sizeof(struct rte_flow);
2827 rss = flow_get_rss_action(p_actions_rx);
2829 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2832 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2833 flow = rte_calloc(__func__, 1, flow_size, 0);
2836 goto error_before_flow;
2838 flow->drv_type = flow_get_drv_type(dev, attr);
2839 if (hairpin_id != 0)
2840 flow->hairpin_flow_id = hairpin_id;
2841 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2842 flow->drv_type < MLX5_FLOW_TYPE_MAX);
2843 flow->rss.queue = (void *)(flow + 1);
2846 * The following information is required by
2847 * mlx5_flow_hashfields_adjust() in advance.
2849 flow->rss.level = rss->level;
2850 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
2851 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
2853 LIST_INIT(&flow->dev_flows);
2854 if (rss && rss->types) {
2855 unsigned int graph_root;
2857 graph_root = find_graph_root(items, rss->level);
2858 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2860 mlx5_support_expansion,
2863 (unsigned int)ret < sizeof(expand_buffer.buffer));
2866 buf->entry[0].pattern = (void *)(uintptr_t)items;
2868 for (i = 0; i < buf->entries; ++i) {
2869 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2870 p_actions_rx, error);
2873 dev_flow->flow = flow;
2874 dev_flow->external = external;
2875 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2876 ret = flow_drv_translate(dev, dev_flow, attr,
2877 buf->entry[i].pattern,
2878 p_actions_rx, error);
2882 /* Create the tx flow. */
2884 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
2885 attr_tx.ingress = 0;
2887 dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
2888 actions_hairpin_tx.actions, error);
2891 dev_flow->flow = flow;
2892 dev_flow->external = 0;
2893 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2894 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
2896 actions_hairpin_tx.actions, error);
2900 if (dev->data->dev_started) {
2901 ret = flow_drv_apply(dev, flow, error);
2906 TAILQ_INSERT_TAIL(list, flow, next);
2907 flow_rxq_flags_set(dev, flow);
2911 mlx5_flow_id_release(priv->sh->flow_id_pool,
2915 ret = rte_errno; /* Save rte_errno before cleanup. */
2916 if (flow->hairpin_flow_id)
2917 mlx5_flow_id_release(priv->sh->flow_id_pool,
2918 flow->hairpin_flow_id);
2920 flow_drv_destroy(dev, flow);
2922 rte_errno = ret; /* Restore rte_errno. */
2927 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
2928 * incoming packets to table 1.
2930 * Other flow rules, requested for group n, will be created in
2931 * e-switch table n+1.
2932 * Jump action to e-switch group n will be created to group n+1.
2934 * Used when working in switchdev mode, to utilise advantages of table 1
2938 * Pointer to Ethernet device.
2941 * Pointer to flow on success, NULL otherwise and rte_errno is set.
2944 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
2946 const struct rte_flow_attr attr = {
2953 const struct rte_flow_item pattern = {
2954 .type = RTE_FLOW_ITEM_TYPE_END,
2956 struct rte_flow_action_jump jump = {
2959 const struct rte_flow_action actions[] = {
2961 .type = RTE_FLOW_ACTION_TYPE_JUMP,
2965 .type = RTE_FLOW_ACTION_TYPE_END,
2968 struct mlx5_priv *priv = dev->data->dev_private;
2969 struct rte_flow_error error;
2971 return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
2972 actions, false, &error);
2978 * @see rte_flow_create()
2982 mlx5_flow_create(struct rte_eth_dev *dev,
2983 const struct rte_flow_attr *attr,
2984 const struct rte_flow_item items[],
2985 const struct rte_flow_action actions[],
2986 struct rte_flow_error *error)
2988 struct mlx5_priv *priv = dev->data->dev_private;
2990 return flow_list_create(dev, &priv->flows,
2991 attr, items, actions, true, error);
2995 * Destroy a flow in a list.
2998 * Pointer to Ethernet device.
3000 * Pointer to a TAILQ flow list. If this parameter NULL,
3001 * there is no flow removal from the list.
3006 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
3007 struct rte_flow *flow)
3009 struct mlx5_priv *priv = dev->data->dev_private;
3012 * Update RX queue flags only if port is started, otherwise it is
3015 if (dev->data->dev_started)
3016 flow_rxq_flags_trim(dev, flow);
3017 if (flow->hairpin_flow_id)
3018 mlx5_flow_id_release(priv->sh->flow_id_pool,
3019 flow->hairpin_flow_id);
3020 flow_drv_destroy(dev, flow);
3022 TAILQ_REMOVE(list, flow, next);
3023 rte_free(flow->fdir);
3028 * Destroy all flows.
3031 * Pointer to Ethernet device.
3033 * Pointer to a TAILQ flow list.
3036 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
3038 while (!TAILQ_EMPTY(list)) {
3039 struct rte_flow *flow;
3041 flow = TAILQ_FIRST(list);
3042 flow_list_destroy(dev, list, flow);
3050 * Pointer to Ethernet device.
3052 * Pointer to a TAILQ flow list.
3055 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
3057 struct rte_flow *flow;
3059 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
3060 flow_drv_remove(dev, flow);
3061 flow_rxq_flags_clear(dev);
3068 * Pointer to Ethernet device.
3070 * Pointer to a TAILQ flow list.
3073 * 0 on success, a negative errno value otherwise and rte_errno is set.
3076 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
3078 struct rte_flow *flow;
3079 struct rte_flow_error error;
3082 TAILQ_FOREACH(flow, list, next) {
3083 ret = flow_drv_apply(dev, flow, &error);
3086 flow_rxq_flags_set(dev, flow);
3090 ret = rte_errno; /* Save rte_errno before cleanup. */
3091 mlx5_flow_stop(dev, list);
3092 rte_errno = ret; /* Restore rte_errno. */
3097 * Verify the flow list is empty
3100 * Pointer to Ethernet device.
3102 * @return the number of flows not released.
3105 mlx5_flow_verify(struct rte_eth_dev *dev)
3107 struct mlx5_priv *priv = dev->data->dev_private;
3108 struct rte_flow *flow;
3111 TAILQ_FOREACH(flow, &priv->flows, next) {
3112 DRV_LOG(DEBUG, "port %u flow %p still referenced",
3113 dev->data->port_id, (void *)flow);
3120 * Enable default hairpin egress flow.
3123 * Pointer to Ethernet device.
3128 * 0 on success, a negative errno value otherwise and rte_errno is set.
3131 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
3134 struct mlx5_priv *priv = dev->data->dev_private;
3135 const struct rte_flow_attr attr = {
3139 struct mlx5_rte_flow_item_tx_queue queue_spec = {
3142 struct mlx5_rte_flow_item_tx_queue queue_mask = {
3143 .queue = UINT32_MAX,
3145 struct rte_flow_item items[] = {
3147 .type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
3148 .spec = &queue_spec,
3150 .mask = &queue_mask,
3153 .type = RTE_FLOW_ITEM_TYPE_END,
3156 struct rte_flow_action_jump jump = {
3157 .group = MLX5_HAIRPIN_TX_TABLE,
3159 struct rte_flow_action actions[2];
3160 struct rte_flow *flow;
3161 struct rte_flow_error error;
3163 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
3164 actions[0].conf = &jump;
3165 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
3166 flow = flow_list_create(dev, &priv->ctrl_flows,
3167 &attr, items, actions, false, &error);
3170 "Failed to create ctrl flow: rte_errno(%d),"
3171 " type(%d), message(%s)",
3172 rte_errno, error.type,
3173 error.message ? error.message : " (no stated reason)");
3180 * Enable a control flow configured from the control plane.
3183 * Pointer to Ethernet device.
3185 * An Ethernet flow spec to apply.
3187 * An Ethernet flow mask to apply.
3189 * A VLAN flow spec to apply.
3191 * A VLAN flow mask to apply.
3194 * 0 on success, a negative errno value otherwise and rte_errno is set.
3197 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
3198 struct rte_flow_item_eth *eth_spec,
3199 struct rte_flow_item_eth *eth_mask,
3200 struct rte_flow_item_vlan *vlan_spec,
3201 struct rte_flow_item_vlan *vlan_mask)
3203 struct mlx5_priv *priv = dev->data->dev_private;
3204 const struct rte_flow_attr attr = {
3206 .priority = MLX5_FLOW_PRIO_RSVD,
3208 struct rte_flow_item items[] = {
3210 .type = RTE_FLOW_ITEM_TYPE_ETH,
3216 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
3217 RTE_FLOW_ITEM_TYPE_END,
3223 .type = RTE_FLOW_ITEM_TYPE_END,
3226 uint16_t queue[priv->reta_idx_n];
3227 struct rte_flow_action_rss action_rss = {
3228 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
3230 .types = priv->rss_conf.rss_hf,
3231 .key_len = priv->rss_conf.rss_key_len,
3232 .queue_num = priv->reta_idx_n,
3233 .key = priv->rss_conf.rss_key,
3236 struct rte_flow_action actions[] = {
3238 .type = RTE_FLOW_ACTION_TYPE_RSS,
3239 .conf = &action_rss,
3242 .type = RTE_FLOW_ACTION_TYPE_END,
3245 struct rte_flow *flow;
3246 struct rte_flow_error error;
3249 if (!priv->reta_idx_n || !priv->rxqs_n) {
3252 for (i = 0; i != priv->reta_idx_n; ++i)
3253 queue[i] = (*priv->reta_idx)[i];
3254 flow = flow_list_create(dev, &priv->ctrl_flows,
3255 &attr, items, actions, false, &error);
3262 * Enable a flow control configured from the control plane.
3265 * Pointer to Ethernet device.
3267 * An Ethernet flow spec to apply.
3269 * An Ethernet flow mask to apply.
3272 * 0 on success, a negative errno value otherwise and rte_errno is set.
3275 mlx5_ctrl_flow(struct rte_eth_dev *dev,
3276 struct rte_flow_item_eth *eth_spec,
3277 struct rte_flow_item_eth *eth_mask)
3279 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
3285 * @see rte_flow_destroy()
3289 mlx5_flow_destroy(struct rte_eth_dev *dev,
3290 struct rte_flow *flow,
3291 struct rte_flow_error *error __rte_unused)
3293 struct mlx5_priv *priv = dev->data->dev_private;
3295 flow_list_destroy(dev, &priv->flows, flow);
3300 * Destroy all flows.
3302 * @see rte_flow_flush()
3306 mlx5_flow_flush(struct rte_eth_dev *dev,
3307 struct rte_flow_error *error __rte_unused)
3309 struct mlx5_priv *priv = dev->data->dev_private;
3311 mlx5_flow_list_flush(dev, &priv->flows);
3318 * @see rte_flow_isolate()
3322 mlx5_flow_isolate(struct rte_eth_dev *dev,
3324 struct rte_flow_error *error)
3326 struct mlx5_priv *priv = dev->data->dev_private;
3328 if (dev->data->dev_started) {
3329 rte_flow_error_set(error, EBUSY,
3330 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3332 "port must be stopped first");
3335 priv->isolated = !!enable;
3337 dev->dev_ops = &mlx5_dev_ops_isolate;
3339 dev->dev_ops = &mlx5_dev_ops;
3346 * @see rte_flow_query()
3350 flow_drv_query(struct rte_eth_dev *dev,
3351 struct rte_flow *flow,
3352 const struct rte_flow_action *actions,
3354 struct rte_flow_error *error)
3356 const struct mlx5_flow_driver_ops *fops;
3357 enum mlx5_flow_drv_type ftype = flow->drv_type;
3359 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
3360 fops = flow_get_drv_ops(ftype);
3362 return fops->query(dev, flow, actions, data, error);
3368 * @see rte_flow_query()
3372 mlx5_flow_query(struct rte_eth_dev *dev,
3373 struct rte_flow *flow,
3374 const struct rte_flow_action *actions,
3376 struct rte_flow_error *error)
3380 ret = flow_drv_query(dev, flow, actions, data, error);
3387 * Convert a flow director filter to a generic flow.
3390 * Pointer to Ethernet device.
3391 * @param fdir_filter
3392 * Flow director filter to add.
3394 * Generic flow parameters structure.
3397 * 0 on success, a negative errno value otherwise and rte_errno is set.
3400 flow_fdir_filter_convert(struct rte_eth_dev *dev,
3401 const struct rte_eth_fdir_filter *fdir_filter,
3402 struct mlx5_fdir *attributes)
3404 struct mlx5_priv *priv = dev->data->dev_private;
3405 const struct rte_eth_fdir_input *input = &fdir_filter->input;
3406 const struct rte_eth_fdir_masks *mask =
3407 &dev->data->dev_conf.fdir_conf.mask;
3409 /* Validate queue number. */
3410 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
3411 DRV_LOG(ERR, "port %u invalid queue number %d",
3412 dev->data->port_id, fdir_filter->action.rx_queue);
3416 attributes->attr.ingress = 1;
3417 attributes->items[0] = (struct rte_flow_item) {
3418 .type = RTE_FLOW_ITEM_TYPE_ETH,
3419 .spec = &attributes->l2,
3420 .mask = &attributes->l2_mask,
3422 switch (fdir_filter->action.behavior) {
3423 case RTE_ETH_FDIR_ACCEPT:
3424 attributes->actions[0] = (struct rte_flow_action){
3425 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
3426 .conf = &attributes->queue,
3429 case RTE_ETH_FDIR_REJECT:
3430 attributes->actions[0] = (struct rte_flow_action){
3431 .type = RTE_FLOW_ACTION_TYPE_DROP,
3435 DRV_LOG(ERR, "port %u invalid behavior %d",
3437 fdir_filter->action.behavior);
3438 rte_errno = ENOTSUP;
3441 attributes->queue.index = fdir_filter->action.rx_queue;
3443 switch (fdir_filter->input.flow_type) {
3444 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
3445 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
3446 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
3447 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
3448 .src_addr = input->flow.ip4_flow.src_ip,
3449 .dst_addr = input->flow.ip4_flow.dst_ip,
3450 .time_to_live = input->flow.ip4_flow.ttl,
3451 .type_of_service = input->flow.ip4_flow.tos,
3453 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
3454 .src_addr = mask->ipv4_mask.src_ip,
3455 .dst_addr = mask->ipv4_mask.dst_ip,
3456 .time_to_live = mask->ipv4_mask.ttl,
3457 .type_of_service = mask->ipv4_mask.tos,
3458 .next_proto_id = mask->ipv4_mask.proto,
3460 attributes->items[1] = (struct rte_flow_item){
3461 .type = RTE_FLOW_ITEM_TYPE_IPV4,
3462 .spec = &attributes->l3,
3463 .mask = &attributes->l3_mask,
3466 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
3467 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
3468 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
3469 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
3470 .hop_limits = input->flow.ipv6_flow.hop_limits,
3471 .proto = input->flow.ipv6_flow.proto,
3474 memcpy(attributes->l3.ipv6.hdr.src_addr,
3475 input->flow.ipv6_flow.src_ip,
3476 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
3477 memcpy(attributes->l3.ipv6.hdr.dst_addr,
3478 input->flow.ipv6_flow.dst_ip,
3479 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
3480 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
3481 mask->ipv6_mask.src_ip,
3482 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
3483 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
3484 mask->ipv6_mask.dst_ip,
3485 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
3486 attributes->items[1] = (struct rte_flow_item){
3487 .type = RTE_FLOW_ITEM_TYPE_IPV6,
3488 .spec = &attributes->l3,
3489 .mask = &attributes->l3_mask,
3493 DRV_LOG(ERR, "port %u invalid flow type%d",
3494 dev->data->port_id, fdir_filter->input.flow_type);
3495 rte_errno = ENOTSUP;
3499 switch (fdir_filter->input.flow_type) {
3500 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
3501 attributes->l4.udp.hdr = (struct rte_udp_hdr){
3502 .src_port = input->flow.udp4_flow.src_port,
3503 .dst_port = input->flow.udp4_flow.dst_port,
3505 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
3506 .src_port = mask->src_port_mask,
3507 .dst_port = mask->dst_port_mask,
3509 attributes->items[2] = (struct rte_flow_item){
3510 .type = RTE_FLOW_ITEM_TYPE_UDP,
3511 .spec = &attributes->l4,
3512 .mask = &attributes->l4_mask,
3515 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
3516 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
3517 .src_port = input->flow.tcp4_flow.src_port,
3518 .dst_port = input->flow.tcp4_flow.dst_port,
3520 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
3521 .src_port = mask->src_port_mask,
3522 .dst_port = mask->dst_port_mask,
3524 attributes->items[2] = (struct rte_flow_item){
3525 .type = RTE_FLOW_ITEM_TYPE_TCP,
3526 .spec = &attributes->l4,
3527 .mask = &attributes->l4_mask,
3530 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
3531 attributes->l4.udp.hdr = (struct rte_udp_hdr){
3532 .src_port = input->flow.udp6_flow.src_port,
3533 .dst_port = input->flow.udp6_flow.dst_port,
3535 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
3536 .src_port = mask->src_port_mask,
3537 .dst_port = mask->dst_port_mask,
3539 attributes->items[2] = (struct rte_flow_item){
3540 .type = RTE_FLOW_ITEM_TYPE_UDP,
3541 .spec = &attributes->l4,
3542 .mask = &attributes->l4_mask,
3545 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
3546 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
3547 .src_port = input->flow.tcp6_flow.src_port,
3548 .dst_port = input->flow.tcp6_flow.dst_port,
3550 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
3551 .src_port = mask->src_port_mask,
3552 .dst_port = mask->dst_port_mask,
3554 attributes->items[2] = (struct rte_flow_item){
3555 .type = RTE_FLOW_ITEM_TYPE_TCP,
3556 .spec = &attributes->l4,
3557 .mask = &attributes->l4_mask,
3560 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
3561 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
3564 DRV_LOG(ERR, "port %u invalid flow type%d",
3565 dev->data->port_id, fdir_filter->input.flow_type);
3566 rte_errno = ENOTSUP;
3572 #define FLOW_FDIR_CMP(f1, f2, fld) \
3573 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
3576 * Compare two FDIR flows. If items and actions are identical, the two flows are
3580 * Pointer to Ethernet device.
3582 * FDIR flow to compare.
3584 * FDIR flow to compare.
3587 * Zero on match, 1 otherwise.
3590 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
3592 if (FLOW_FDIR_CMP(f1, f2, attr) ||
3593 FLOW_FDIR_CMP(f1, f2, l2) ||
3594 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
3595 FLOW_FDIR_CMP(f1, f2, l3) ||
3596 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
3597 FLOW_FDIR_CMP(f1, f2, l4) ||
3598 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
3599 FLOW_FDIR_CMP(f1, f2, actions[0].type))
3601 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
3602 FLOW_FDIR_CMP(f1, f2, queue))
3608 * Search device flow list to find out a matched FDIR flow.
3611 * Pointer to Ethernet device.
3613 * FDIR flow to lookup.
3616 * Pointer of flow if found, NULL otherwise.
3618 static struct rte_flow *
3619 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
3621 struct mlx5_priv *priv = dev->data->dev_private;
3622 struct rte_flow *flow = NULL;
3625 TAILQ_FOREACH(flow, &priv->flows, next) {
3626 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
3627 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
3628 dev->data->port_id, (void *)flow);
3636 * Add new flow director filter and store it in list.
3639 * Pointer to Ethernet device.
3640 * @param fdir_filter
3641 * Flow director filter to add.
3644 * 0 on success, a negative errno value otherwise and rte_errno is set.
3647 flow_fdir_filter_add(struct rte_eth_dev *dev,
3648 const struct rte_eth_fdir_filter *fdir_filter)
3650 struct mlx5_priv *priv = dev->data->dev_private;
3651 struct mlx5_fdir *fdir_flow;
3652 struct rte_flow *flow;
3655 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
3660 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
3663 flow = flow_fdir_filter_lookup(dev, fdir_flow);
3668 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
3669 fdir_flow->items, fdir_flow->actions, true,
3673 assert(!flow->fdir);
3674 flow->fdir = fdir_flow;
3675 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
3676 dev->data->port_id, (void *)flow);
3679 rte_free(fdir_flow);
3684 * Delete specific filter.
3687 * Pointer to Ethernet device.
3688 * @param fdir_filter
3689 * Filter to be deleted.
3692 * 0 on success, a negative errno value otherwise and rte_errno is set.
3695 flow_fdir_filter_delete(struct rte_eth_dev *dev,
3696 const struct rte_eth_fdir_filter *fdir_filter)
3698 struct mlx5_priv *priv = dev->data->dev_private;
3699 struct rte_flow *flow;
3700 struct mlx5_fdir fdir_flow = {
3705 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
3708 flow = flow_fdir_filter_lookup(dev, &fdir_flow);
3713 flow_list_destroy(dev, &priv->flows, flow);
3714 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
3715 dev->data->port_id, (void *)flow);
3720 * Update queue for specific filter.
3723 * Pointer to Ethernet device.
3724 * @param fdir_filter
3725 * Filter to be updated.
3728 * 0 on success, a negative errno value otherwise and rte_errno is set.
3731 flow_fdir_filter_update(struct rte_eth_dev *dev,
3732 const struct rte_eth_fdir_filter *fdir_filter)
3736 ret = flow_fdir_filter_delete(dev, fdir_filter);
3739 return flow_fdir_filter_add(dev, fdir_filter);
3743 * Flush all filters.
3746 * Pointer to Ethernet device.
3749 flow_fdir_filter_flush(struct rte_eth_dev *dev)
3751 struct mlx5_priv *priv = dev->data->dev_private;
3753 mlx5_flow_list_flush(dev, &priv->flows);
3757 * Get flow director information.
3760 * Pointer to Ethernet device.
3761 * @param[out] fdir_info
3762 * Resulting flow director information.
3765 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
3767 struct rte_eth_fdir_masks *mask =
3768 &dev->data->dev_conf.fdir_conf.mask;
3770 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
3771 fdir_info->guarant_spc = 0;
3772 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
3773 fdir_info->max_flexpayload = 0;
3774 fdir_info->flow_types_mask[0] = 0;
3775 fdir_info->flex_payload_unit = 0;
3776 fdir_info->max_flex_payload_segment_num = 0;
3777 fdir_info->flex_payload_limit = 0;
3778 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
3782 * Deal with flow director operations.
3785 * Pointer to Ethernet device.
3787 * Operation to perform.
3789 * Pointer to operation-specific structure.
3792 * 0 on success, a negative errno value otherwise and rte_errno is set.
3795 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
3798 enum rte_fdir_mode fdir_mode =
3799 dev->data->dev_conf.fdir_conf.mode;
3801 if (filter_op == RTE_ETH_FILTER_NOP)
3803 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
3804 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
3805 DRV_LOG(ERR, "port %u flow director mode %d not supported",
3806 dev->data->port_id, fdir_mode);
3810 switch (filter_op) {
3811 case RTE_ETH_FILTER_ADD:
3812 return flow_fdir_filter_add(dev, arg);
3813 case RTE_ETH_FILTER_UPDATE:
3814 return flow_fdir_filter_update(dev, arg);
3815 case RTE_ETH_FILTER_DELETE:
3816 return flow_fdir_filter_delete(dev, arg);
3817 case RTE_ETH_FILTER_FLUSH:
3818 flow_fdir_filter_flush(dev);
3820 case RTE_ETH_FILTER_INFO:
3821 flow_fdir_info_get(dev, arg);
3824 DRV_LOG(DEBUG, "port %u unknown operation %u",
3825 dev->data->port_id, filter_op);
3833 * Manage filter operations.
3836 * Pointer to Ethernet device structure.
3837 * @param filter_type
3840 * Operation to perform.
3842 * Pointer to operation-specific structure.
3845 * 0 on success, a negative errno value otherwise and rte_errno is set.
3848 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
3849 enum rte_filter_type filter_type,
3850 enum rte_filter_op filter_op,
3853 switch (filter_type) {
3854 case RTE_ETH_FILTER_GENERIC:
3855 if (filter_op != RTE_ETH_FILTER_GET) {
3859 *(const void **)arg = &mlx5_flow_ops;
3861 case RTE_ETH_FILTER_FDIR:
3862 return flow_fdir_ctrl_func(dev, filter_op, arg);
3864 DRV_LOG(ERR, "port %u filter type (%d) not supported",
3865 dev->data->port_id, filter_type);
3866 rte_errno = ENOTSUP;
3872 #define MLX5_POOL_QUERY_FREQ_US 1000000
3875 * Set the periodic procedure for triggering asynchronous batch queries for all
3876 * the counter pools.
3879 * Pointer to mlx5_ibv_shared object.
3882 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
3884 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
3885 uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
3888 cont = MLX5_CNT_CONTAINER(sh, 1, 0);
3889 pools_n += rte_atomic16_read(&cont->n_valid);
3890 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
3891 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
3892 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
3893 sh->cmng.query_thread_on = 0;
3894 DRV_LOG(ERR, "Cannot reinitialize query alarm");
3896 sh->cmng.query_thread_on = 1;
3901 * The periodic procedure for triggering asynchronous batch queries for all the
3902 * counter pools. This function is probably called by the host thread.
3905 * The parameter for the alarm process.
3908 mlx5_flow_query_alarm(void *arg)
3910 struct mlx5_ibv_shared *sh = arg;
3911 struct mlx5_devx_obj *dcs;
3914 uint8_t batch = sh->cmng.batch;
3915 uint16_t pool_index = sh->cmng.pool_index;
3916 struct mlx5_pools_container *cont;
3917 struct mlx5_pools_container *mcont;
3918 struct mlx5_flow_counter_pool *pool;
3920 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
3923 cont = MLX5_CNT_CONTAINER(sh, batch, 1);
3924 mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
3925 /* Check if resize was done and need to flip a container. */
3926 if (cont != mcont) {
3928 /* Clean the old container. */
3929 rte_free(cont->pools);
3930 memset(cont, 0, sizeof(*cont));
3933 /* Flip the host container. */
3934 sh->cmng.mhi[batch] ^= (uint8_t)2;
3938 /* 2 empty containers case is unexpected. */
3939 if (unlikely(batch != sh->cmng.batch))
3943 goto next_container;
3945 pool = cont->pools[pool_index];
3947 /* There is a pool query in progress. */
3950 LIST_FIRST(&sh->cmng.free_stat_raws);
3952 /* No free counter statistics raw memory. */
3954 dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
3956 offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
3957 ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
3959 pool->raw_hw->mem_mng->dm->id,
3961 (pool->raw_hw->data + offset),
3963 (uint64_t)(uintptr_t)pool);
3965 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
3966 " %d", pool->min_dcs->id);
3967 pool->raw_hw = NULL;
3970 pool->raw_hw->min_dcs_id = dcs->id;
3971 LIST_REMOVE(pool->raw_hw, next);
3972 sh->cmng.pending_queries++;
3974 if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
3979 sh->cmng.batch = batch;
3980 sh->cmng.pool_index = pool_index;
3981 mlx5_set_query_alarm(sh);
3985 * Handler for the HW respond about ready values from an asynchronous batch
3986 * query. This function is probably called by the host thread.
3989 * The pointer to the shared IB device context.
3990 * @param[in] async_id
3991 * The Devx async ID.
3993 * The status of the completion.
3996 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
3997 uint64_t async_id, int status)
3999 struct mlx5_flow_counter_pool *pool =
4000 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
4001 struct mlx5_counter_stats_raw *raw_to_free;
4003 if (unlikely(status)) {
4004 raw_to_free = pool->raw_hw;
4006 raw_to_free = pool->raw;
4007 rte_spinlock_lock(&pool->sl);
4008 pool->raw = pool->raw_hw;
4009 rte_spinlock_unlock(&pool->sl);
4010 rte_atomic64_add(&pool->query_gen, 1);
4011 /* Be sure the new raw counters data is updated in memory. */
4014 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
4015 pool->raw_hw = NULL;
4016 sh->cmng.pending_queries--;
4020 * Translate the rte_flow group index to HW table value.
4022 * @param[in] attributes
4023 * Pointer to flow attributes
4024 * @param[in] external
4025 * Value is part of flow rule created by request external to PMD.
4027 * rte_flow group index value.
4031 * Pointer to error structure.
4034 * 0 on success, a negative errno value otherwise and rte_errno is set.
4037 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
4038 uint32_t group, uint32_t *table,
4039 struct rte_flow_error *error)
4041 if (attributes->transfer && external) {
4042 if (group == UINT32_MAX)
4043 return rte_flow_error_set
4045 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
4047 "group index not supported");